{"archive_ref": "sim_att-technical-journal_1946-10_25_4", "canonical_url": "https://archive.org/details/sim_att-technical-journal_1946-10_25_4", "char_count": 79402, "collection": "bstj", "doc_id": 9, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc9", "record_count": 1, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": "pdftotext", "selected_extraction_score": 0.891634, "source_family": "archive_org", "source_url": "https://archive.org/details/sim_att-technical-journal_1946-10_25_4", "split": "test", "text": "The servo system is a special type of feedback amplifier, usually including a\nmixture of electrical, mechanical, thermal, or hydraulic circuits. With suitable\n\n21 Ele\n\ndesign, the behavior of these various circuits can be described in the universal\nlanguage of linear systems.\nFurther, if the servo system is treated in terms of\ncircuit response to sinusoidally varying signals, it then becomes possible to\n\ndraw upon the wealth of linear feedback amplifier design based on frequency\nanalysis.\n\nThis paper discusses a typical analogy between electrical and mechanical\nsystems and describes, in frequency response language, the behavior of such\ncommon servo components as motors, synchro circuits, potentiometers, and tach-\n\nometers. The elementary concepts of frequency analysis are reviewed briefly,\nand the familiar Nyquist stability criterion is applied to a typical motor-drive\nservo system. The factors to be considered in choosing stability margins are\nlisted\u2014system\n\nvariability,\n\nnoise\n\nenhancement,\n\nand\n\ntransient\n\nresponse.\n\nThe\n\nbasic gain-phase interrelations shown by Bode are summarized, and some of\ntheir design implications discussed. In addition to the classical methods,\nsimple approximate methods for calculating dynamic response of servo systems\nare presented and illustrated.\nNoise in the input signal is discussed as a compelling factor in the choice of\n\nservo loop characteristics. The need for tailoring the servo loop to match the\ninput signal is pointed out, and a performance comparison given for two simple\nservos designed to track an airplane over a straight line course.\nThe use of\nsubsidiary or local feedback to linearize motor-drive systems, and predistortion\n\nof the input signal to reduce overall dynamic error are described.\n1. INTRODUCTION\n\nSIMPLE servo system is one which controls an output quantity\naccording to some required function of an input quantity. This\ncontrol is of the \u2018\u2018report back\u201d type. That is, some property of the output\nis monitored and compared against the input quantity, producing a net\ninput or \u201cerror\u201d signal which is then amplified to form the output. The\nfirst statement defines the servo as a transmission system; the second, as a\nfeedback loop. The problem of servo design is then to fashion the desired\ntransmission properties while meeting the stability requirements of the\nfeedback loop. This is the familiar design problem of the feedback amplifier.\n2. THe\n\nSERVO\n\nCIRCUIT\n\nThe design of linear feedback amplifiers has been developed to a high\ndegree in terms of frequency response; that is, in terms of circuit response\nto sinusoidal signals.'. The servo system is a special type of feedback\namplifier, and usually can be made fairly linear. Thus, it is logical to\nanalyze and design the servo circuit on a frequency response basis. Also,\n1See \u201cNetwork Analysis and Feedback Amplifier Design,\u201d by H. W. Bode, D. Van\nNostrand Co., 1945.\n616\n\nCont\n\nwhich\nrotatio\n\nFig. 1:\npliane\n\nangule\nanalog\n\ndimen\nthe ra\n\nresist\ntorqu\nSor\nchani\nconv\u00e9\nd6/di\nbetw\n\nactin\ncircu\nload\n\nLINEAR\n\nSERVO\n\nTHEORY\n\n617\n\nservo systems usually are combinations of electrical, triechanical, thermal,\nor hydraulic circuits. In order to describe the behavior of these various\ncircuits in homogeneous terms, it is desirable to recognize the analogous\nrelationships established by similarity of the underlying differential equations. Before proceeding to a discussion of frequency analysis, a typical\nanalogy between electrical and mechanical systems will be described.\n\n2.1 Electrical-Mechanical Analogy\nConfining the discussion to rotating mechanical systems, the analogy\nwhich will be chosen here puts voltage equivalent to torque, and current to\nrotational speed. This choice leads to the array of equivalents shown in\nFig. 1; inductance, capacity, and resistance corresponding to inertia, compliance, and mechanical resistance, respectively. Charge is equivalent to\nangular displacement, and both kinetic and potential energy are selfanalogous. The ratio of voltage to current, or torque to speed has the\ndimensions of resistance.\nIn an interconnected electro-mechanical system,\nthe ratio of voltage to speed or torque to current may be called a transfer\nresistance. Similarly, the ratio of voltage to angular displacement, or of\ntorque to charge, is a transfer stiffness (reciprocal of capacity or compliance).\nSome commonly used devices for coupling between electrical and mechanical circuits are shown in Figs. 2 and 3. The motor, Fig. 2a, is used to\nconvert an electrical current 7 into a mechanical speed or \u201ccurrent\u201d @ (=\nd6/dt). The electrical control current 7 is produced by the voltage difference\nbetween an applied emf e and a counter-rotational emf (not indicated),\nacting upon the total electrical mesh resistance R,.* In the mechanical\ncircuit, a torque proportional to i forces a \u201c\u2018current\u201d 6 through the mechanical\nond Ky, J.\nAn equivalent mechanical mesh directly relating shaft speed to the applied\nemf is shown in Fig. 2b. A fictitious generated torque u,e acts upon the\nmechanical load through an apparent mechanical resistance Rm. uy\nis a transfer constant determined both by the motor properties and the\nelectrical mesh resistance R,. Ry is similarly governed and is inversely\nproportional to R, .\nThe motor may be compared to a vacuum tube having an amplification\nfactor yw, and a plate resistance R.,.. However, the motor is usually much\nmore a bilateral coupling element than the vacuum tube, due to the effect\nof the counter emf upon the electrical mesh.\nThe potentiometer, tachometer, and synchro circuit shown in Fig. 3 are\nall means for converting a mechanical quantity to an electrical one. All\nthree are substantially unilateral coupling elements. The potentiometer\n* R, includes both the source resistance and the motor winding resistance.\n\n618\n\nBELL SYSTEM\n\nTECHNICAL\n\nJOURNAL\n\ndelivers an output voltage e proportional to its shaft angle\n6. Thus, the\nratio of \u20ac to @ is a transfer stiffness constant S,. The synchro\ncircuit\ncon- |\nsists of a synchro generator connected to a control transformer,\nand delivers\nan output voltage e proportional to6; \u2014 6, the angular\ndifference between\n\nELECTRICAL\n\nMECHANICAL\n\nINDUC TANCE\nL\n\nJ\n\nO00 wo\nt\n\n@\n\n>\n\n6\n\n\u2014\u00bb\n\nCOMPLIANCE\n\nc\n\negg\n\n,\n\nc\n\n}\n\n8\nT=\n\nCc.\n\n,gyy-Y\nFee\n\nBrig.\n\ng\n\nSPRING\n\nSS\n\nie\n\nle\n\na.\ne=c\n\nFe\n/idt\n\nT= ae? /edt\n\nRESISTANCE\n\nRANE A,\n\nRESISTANCE\n\nPee\n\nR\n\nom\\/\\\\\u2014m0\n\nar\n\nom \\V\\\\\u2014o\n|\n\n\u2014\u2014\n\nSERIES L-R-C\n\nSERIES J-R-C\n\nB.\n<<\nVV bao\n\na\u2014\u2014\n\n@\u20ac \u2014__\u00bb\n\nsi\n\nJ\nom]\n\nL\n\ne=Lddepist\n\nte /igt\n/t0\nFig.\n\n\u2014=\n\nDASHPOT\nH\n\nR\n\n-\n\n\u00a2\n{|\n\nwe\n6\n\n\u2014\u2014\u2014_\u2014_\n\n2\n\ncf\nq\n\nT \u2014\n\nT=RO\n\nJt\nOm OO\n\n8\n\n6\n\n@=Ri\n\nat\n\n6\n\npean ee\n\n.\n\n~\n\n\\)\n\nFLYWHEEL\n\nde\ndt\n\nCAPACITY\n\na\n\n4\n\n|\n\nTey\n\ne=_\u2014\u2014\n\netal.\n\n~\n\n.\n\\\nSan)\n\nee\n\ndi\ndt\n\ntran\n\n:\n\nTO 0\n\n-\n\ntion\n\n-\n\nINERTIA\n\n_\u2014\n\n<\u2014\u2014\n\nTI\nto el\nchro\npro}\nis tk\n\n6\n\nJ\n\n(7\n{ /\n\n| Ts wn\nYY\n\n,\n\nCc\n\nRL\n\nfT\u2014__\n\n=) \u201cYap\n98 (Ore\nsp641 /odt\n/j\n\n2.\n\n1\u2014Electrical-mechanical analogy.\n\nthe two shafts. Thus the action of the synchro pair is that of a combined\ntransfer stiffness and differential. The tachometer is a\ngenerator which\nproduces an output voltage e proportional to 6, the angular\nspeed of its\nshaft. The ratio of e to 6 is a transfer resistance constant\nR, .\n\n|\n\nIt\nir\nq\nis\ns\n\nLINEAR\n\nSERVO\n\nTHEORY\n\n619\n\nThere are many other specific devices used to convert from mechanical\nto electrical quantities.\nMost are equivalent to the potentiometer or synchro circuit, one such being a lobing radar antenna, which delivers a voltage\nproportional to an angular difference. A different, less widely used device\nis the accelerometer, a generator which delivers an output voltage proportional to the angular acceleration of its shaft.\nIts characteristic is that of a\ntransfer inductance or inertia.\n'\n\nRm\n\nA A\n\nWN\n\nee.\n=\n\n<Rm\n\nTapeOy | \u00a9 4\n|\n\n=\n\n2,\nOo\n\nJ ef+(Rmm+Rin)O\n=pye\n(a)\n\n\u2018b)\n\nFig. 2\u2014Motor asa transfer device.\n\n(a) Motorand load.\n\n(b) Equivalent mechanical mesh.\n\nPOTENTIOMETER\n\na\n\nwea\n\n\u2014\n\ne=S;,8e\n\ne\n\n(\n\n<<\nSYNCHRO\n\nCIRCUIT\n\n\u2014\u2014\u2014\u2014\u2014\n\nae\n\n|\n\n=\n\n=\n\n$,+(0)-6\neee\n\n\"e\n\nTACHOMETER\nRea\neee\n\n:\ne\n\n.\ne=R,6\n\n\u2014\u2014\u2014\u2014\u2014\n\nFig.\n\n3\u2014Mechanical-electrical transfer devices.\n\n2.2 Frequency Analysis\nA brief review of the basic concepts of periodic analysis will be presented.\nIt is assumed that the driving force applied to a network may be analyzed\nin terms of a series of sinusoidal components of various amplitudes, frequencies, and phases. The network response to each sinusoidal component\nis then evaluated, and the over-all result obtained by a summation of all\nsuch elementary responses.\nThis is the formal procedure.\nActually\n\n620\n\nBELL SYSTEM\n\nTECHNICAL\n\nJOURNAL\n\nit is often unnecessary to perform these precise operations in order to obtain\na broad picture of the network behavior.\nThe method for determining the network response to a sinusoidal input is\ndeveloped as follows.\nIt is assumed that the circuit parameters are constant, independent of signal amplitude. Then, as indicated in Fig. 1, a\nsingle R-L-C or R-J-C mesh may be represented by a constant-coefficient\nlinear differential equation. Choosing the electrical mesh for illustration,\n\n(Lp + R+ 1/Cp)i) = etd),\nwhere p\u201d = d\u201d/dt\u201d, 1/p = [a and e(#), i(\u00a2) are the mesh voltage and current respectively.\n\nThis may be further abbreviated as\n\nZ(p)i(t) =e(?),\n\n(1)\n\nwhere Z(p) = Lp + R+ 1/Cp. For purposes of frequency analysis we\nare interested only in the forced or steady-state solution of (1), where e(?)\nis a sinusoidal voltage E sin wt. This steady-state solution is\n\ni(t) \u2014.\n\n|Z(jw)\n\nsin (wt + 9),\n\nwhere jw has replaced p in the function Z(p), and the phase shift \u00a2 is the\nnegative of the phase angle of the complex number Z(jw).? This result is\nconventionally abbreviated as\n\nE\nZ (jw)\n\nfm ge,\n\n(2)\n\nwhere J is a complex number whose magnitude equals the peak amplitude\nof the current, and whose phase angle gives the associated phase shift.\nThe function Z(jw) is called the impedance of the mesh.\nThe relationship between torque, angular speed, and mechanical impedance is of course the same as expressed by (2). That is,\n\n=\n\n\u2018g\n\n_\u2014.,\n\nZ(jw)\n\n(2.1)\n\nwhere 6 is the complex peak amplitude of the sinusoidally varying speed, T\nis the peak amplitude of the applied sinusoidal torque, and Z(jw) is the\nmechanical impedance obtained by substituting jw for p in the operator\nZ(p) = Jp+R+1/Cp.\nSince the angular displacement is the time integral\n2\n\nis used to represent frequency in radians/sec, or 2 times frequency in cycles/sec.\n\nLINEAR\n\nSERVO\n\nTHEORY\n\n621\n\nof the speed, the expression for @ may be obtained by dividing both sides of\n\nequation (2.1) by p or, for the periodic case, byjw.\n\nThus\n\nmi pie\njal (jw) \u201d\n\n(2.2)\n\nThe function jwZ(jw) is the complex stiffness of the mechanical mesh.\nThe phase shift of 6 relative to 6 is \u201490 degrees, as seen from a comparison\nof (2.1) and (2.2).\nFor an electro-mechanical network consisting of a number of interconnected meshes, a set of simultaneous differential equations of the type of\n(1) may be written. If jw is substituted for p in these equations, there\nresults a set of simultaneous algebraic equations which lead directly to the\nsteady-state periodic solution.\nIf a sinusoidal voltage or torque is applied\nat some mesh of the network, the resulting sinusoidal current or speed\nresponse in some other mesh is given by, using the notation of (2),\n(Response)\n\n= (Cause)\nZ.(jw)\n\nwhere Z,(jw) for the chosen pair of meshes is obtained from the solution\nof the algebraic equations. 7Z,(jw) is called a transfer impedance, and may\nexpress the ratio of a voltage to current or speed, or of a torque to current\nor speed. The above relation also may be written as\n(Response)\n\n=\n\nY,(jw)- (Cause),\n\nwhere Y,(jw) = 1/Z;(jw) is called the transfer admittance between the two\nchosen parts of the network.\nIn this form the response amplitude is obtained by multiplying the forcing sinusoid by | Y;(jw) |, while the phase\nshift is given directly by the angle of Y,(jw). The transfer ratio between\nlike or analogous quantities at two parts of the network is similarly a\ncomplex function of frequency, having the dimensions of a pure numeric.\nServo systems usually consist largely of elementary networks isolated\nby unilateral coupling devices (vacuum tubes, potentiometers, .etc.).\nThus, over-all transfer ratios often may be evaluated by taking the product\nof a number of simple \u2018\u2018stage\u2019\u2019 transfer ratios, rather than by solving a large\narray of simultaneous equations. If the absolute magnitudes of the transfer\nratios or \u2018\u201c\u2018transmissions\u201d are expressed in decibels* of logarithmic gain,\nboth the over-all gain and phase shift of a number of tandem stages may be\nobtained by simple addition of the individual stage gain and phase values.\nThe transfer ratios of the conversion devices shown in Figs. 2 and 3\n3 The gain in decibels for a given transfer ratio is taken to be 20 logio of the absolute\nvalue of the ratio.\n\n622\n\nBELL SYSTEM\n\nTECHNICAL\n\nJOURNAL\n\nmay be written by inspection.\nReferring to Fig. 2b, the transfer admittance\nof a motor with resistance and inertia load may be written as\n\nof the\nassem\nequal\n\n6 ss\n\nE\n\nthe fo\n\nMt\n\njw + Rn + Ri,\u2019\nMee\n\n1\n\nJ jo + wm\u2019\n\nwhere\nequal\npartic\nservo\n\nwhere\n\no, = Km t Rm\nm\n\nBi\n\nad\n\nwm is the reciprocal of the time-constant of the motor and control mesh,\nand is 2% times the \u2018\u2018corner\u201d frequency at which the inertial impedance just\nequals the apparent mechanical resistance.\nWriting the transfer characteristic in terms of shaft position, rather than speed,\n\na\n\nEJ\n\njw(jw + wm)\n\n(3.1)\n\nFor values of w small compared with w, , 6/E is proportional to 1/jw.\nThis factor has a phase shift of \u201490 degrees and approaches infinity as w\napproaches zero. This is merely a statement in frequency analysis language\nthat the motor shaft angle is the time integral of the applied voltage, for\nslowly changing voltage.\nor more rapidly varying voltage, such that w\nis large compared to w, ,0/E is proportional to 1/(jw)? or \u2014 1/w\", the angular\nvariation being shifted \u2014 180 degrees with respect to the voltage variation.\nThe transfer ratio of the potentiometer, Fig. 3a, may be written as\nE\n--9\nwhile\n\nfor the tachometer,\n\n=et\n\n(4.1)\n(4.1)\n\njoR: \u00b0\n\n(4.3)\n\nFig. 3c,\n\nE\n\n6\n\n=\n\nUsually at some point in the system a compensating or \u201cequalizing\u201d\nnetwork will be included to modify the transfer ratio of the basic components to the desired over-all transmission characteristic.\nFrequently\n\nthis equalizer is incorporated in the electrical section of the servo because\n\n23 S$\nTh\ning \u00a2\n\nLINEAR\n\nSERVO\n\nTHEORY\n\n623\n\nof the comparative ease with which electrical circuit components may be\nassembled in desired combinations.\nThe transfer characteristic of the\nequalizer may be simple or complicated, but in general may be written in\nthe form,\n\u2014, Go + o1)(jw + ws)\n\nGeo + 0a) (joo+ wa) -->\u201d\n\nee\n\nwhere the constants \u00ab1 , w2 , etc. may be real or complex.\nThe synthesis of\nequalizing networks is a well known art and will not be discussed here,\nparticularly since most of the equalization characteristics used in present\nservo\u2019 systems can be realized with simple networks.\n2.3 Simple Servo System (Single Loop)\nThe simple servo system may be divided into two basic parts, an amplifying circuit and a monitoring or comparison circuit. Such a division is\nf\n\n,(t)\n\nF, +BFo\n\nq\n\nBFo\n\nFo\n\nFo(t\n\n,\n\nFo\n\nBS\n\nLOOP\n\nTRANSMISSION= U6\n\nTRANSFER\n\nCHARACTERISTIC\n\ne\n\nae\ngs\n\n1-UB\n\nFy\n\nFig. 4\u2014Simple servo system.\n\nindicated in Fig. 4, where uw and @ are the transfer characteristics of the amplifying and monitoring parts, respectively.\n/; and Fy, represent typical\nsinusoidal components of the total input and output quantities /,(\u00a2) and\nfo(t),* while \u00bb and 8 are complex-valued functions of jw as described in the\nprevious section.\nThe return signal 8F2 from the monitoring circuit is added\ninput F; to form a net u circuit input /; + 8F2.\n\nto the servo\n\nThe servo transfer char-\n\nacteristic is found by setting\n\nPF; = w(Fi + Bo),\nfrom which\n\nae\n2\nee\n\n1\n\ni\n\n(6)\n\n4 That is, F; and F; are complex quantities employed in the same fashion as J in equation (2).\n\n624\n\nBELL SYSTEM\n\nTECHNICAL\n\nJOURNAL\n\nThe closed system formed by the two basic circuits in tandem is of course\nfeedback loop, the loop transmission characteristic being given by yf.\nAny desired form of servo transfer ratio may be obtained by an unlimited\nnumber\n\nof uw and @ circuit combinations.>\n\nHowever,\n\nthe 8 characteristic,\n\nwhich is usually determined by a passive network or an inherent propert)\nof a monitoring device, tends to be more stable with time and varying signa\namplitude\n\nthan\n\nthat of the uw circuit, which\n\nmay\n\ninclude vacuum\n\ntubes,\n\nmotors, and other variable components.\nConsequently, it is desirable to\nhave the servo transfer characteristic largely dependent upon the 8 circuit\nproperties alone. This may be accomplished by making the loop trans\n(a)\n\n8\n\nGEN\n\net.\n\nGIN\nSouT~=B =6iN\n\nFig.\n\n5\u2014Synchro follow-up system.\n\nmission w8 very large compared to unity over the essential frequency range\nof the servo input signal f;(\u00a2).. Under this condition, equation (6) becomes,\nF,\n\na\n\nup |\n\n(6.1)\n\nThus the external transfer characteristic is set by 8.* If, for instance, F;\nand fF\u00bb are similar or analogous quantities and it is desired to have the servo\noutput a replica of the input, 8 may be chosen as \u20141, yielding F2 ~ F,.\nIt is not always easy to determine the basic parts \u00bb and 8 of a servo by\ninspection of a schematic diagram of the system. An example is furnished\nby the synchro follow-up system shown in Fig. 5a. As previously discussed,\nthe characteristic of the synchro comparison circuit is that of a differential\n\u00ae Feedback stability requirements place certain restrictions on the permissible forms of\nu8.\n\nThis will be discussed\n\nin the next section.\n\n* The error arising from the approximate\nsection as one type of \u201cservo error.\u201d\n\nnature of (6.1) will be discussed in the next\n\ntransf\ngiven\n\nFig. 5\nin the\n\nrelati\n\nLINEAR\n\nSERVO\n\nTHEORY\n\n625\n\ntransfer stiffness S, , the voltage output of the control transformer being\ngiven by S;(@: \u2014 62). However, as seen from the modified diagram of\nFig. 5b, the 8 characteristic is simply \u20141, the transfer constant S, appearing\nin the w circuit. Thus if the loop transmission is kept large, the essential\nrelation demanded between 6; and 4 does not depend upon the value of S;\n\ny\n\nFig. 6\u2014Potentiometer loop.\nTACHOMETER.\n\u2014\u2014\n6\n+\n\n4\n\n+\n\n:\n\nEp ~ -R,e\n\noa\n\nFig.\n\nE\n\n7\u2014Tachometer loop.\n\n(as is obvious from physical considerations).\nThis result also applies to a\nradar angle-tracking loop, where the received deviation or error signal is\nproportional to the difference between the angular coordinate of the target\nand that of the antenna system.\nFigures 6 and 7 represent two servo systems where the input is electrical\nand the output mechanical.\nIn Fig. 6 a potentiometer is used as a monitor-\n\n626\n\nBELL SYSTEM\n\nTECHNICAL\n\nJOURNAL\n\ning device, the transfer stiffness in this case appearing in the 8 circ)\nIf Ais regarded as the output, then 8 = \u2014S;, ,* and the transfer characteris\nis, for high loop gain,\n\nAs us\nforman\nrestrict\n\nto be t\n\nA~\n\n=\n\n\u2014\u00a3\n\n=\n\ncompat\n\nwhere C, = 1/S;. Thus the over-all characteristic between input voltage\nIp\nand angular displacement is simply a transfer compliance constant.\nFig. 7 a tachometer monitor is used.\nRegarding angular speed 6 as the\noutput,\n\nthen\n\n8 =\n\n\u2014R,,\n\nthe transfer resistance of the tachometer.\n\nThe\n\ntransfer equation is thus\n6\n\naie\n\noe. me\n\n=\n\ng. E,\n\nwhere g;(= 1/R;) is a transfer conductance.\n\nis neitk\nthe loo\nsuch as\ndistrib\n\nThe ef\ncause |\nIt will\nthe lo\n\nfreque\nreduce\n\ntribut\n3. DESIGN OF SIMPLE LINEAR SERVO SYSTEMS\n\nThe majority of servo systems in use, while often greatly extended in\n\nspace and frequently including highly diversified transmission elements,\nmay be represented by one essential feedback loop. However, a well\ndesigned servo often will incorporate numerous subsidiary or local feedback\nloops around stages of the system, in order to obtain a desired degree of\nlinearity or performance with easily obtainable circuit components.\nCommon examples of such local feedback loops are electrical feedback around\nvacuum\ntube amplifiers, and tachometer (velocity) feedback around motordrive systems. These subsidiary feedback loops are almost always designed\nso that they are individually stable when the over-all feedback loop is\nopened (assuming the method employed for opening the over-all loop does\nnot disturb the impedance terminations of the local feedback stages).\nIf it is thus assumed that any subsidiary loops are individually stable, then\nthe primary servo loop design may be treated simply as that of a single loop,\nwhose over-all loop transmission is obtained by taking the product of the\nexternal transfer ratios of the various stages.\nThe design of a single loop servo may be divided into the design of the\nloop transmission 48, and one of the remaining parts, \u00bb or 8. As previously\ndescribed, it is usually desirable to fix 8 according to the required basic\ninput-output relationship of the servo, thus leaving uf as a single characteristic to be chosen.\n*It is assumed here that the transmission of the yu circuit is basically positive. The\nnegative signs associated with S;, of Fig. 6 and R; of Fig. 7 are then introduced (by poling)\nto make the loop transmission yf essentially negative. This stipulation ensures what is\ncommonly called \u201cnegative feedback,\u201d when the loop delay is zero.\n\nThe\n\nsignal\nInput\noverlo\n\nare re\nteristi\nby (6\nby rec\nOn\ntainir\ntrans\n\nnitud\nupon\nTh\nnoise\ntuati\nstrict\nSA.\n\nLINEAR\n\nSERVO\n\nTHEORY\n\nAs usual, specification of the form of u8(jw)\n\nis beset\n\n627\n\nby a series of per-\n\nformance objectives on the one hand and a set of physical limitations and\nrestrictions on the other. Assuming the relationship expressed by (6.1)\nto be the required one, it would seem desirable to make u8(jw) very large\ncompared to unity at all frequencies.\nHowever, there are reasons why this\nis neither possible nor actually desirable.\nAs the value of w is increased,\nthe loop transmission is eventually controlled by paras: ic circuit elements\nsuch as distributed capacity and inductance in the electrical circuits, and\ndistributed inertia, compliance, and backlash in the mechanical circuits.\nThe effect of these parasitic elements at the higher signal frequencies is to\ncause |u@ | to decrease as a very high order of 1/w with increasing frequency.\nIt will be shown, however, that feedback stability considerations require\nthe loop transmission to be decreasing comparatively slowly through the\nfrequency region where | u@ | is of the order of unity. Thus u& must be\nreduced below unity at a frequency sufficiently low to avoid excessive contribution from the parasitics.\nThe presence of \u201cnoise\u201d or undesired disturbances in the servo input\nsignal is another compelling factor in the design of the loop characteristic.\nInput noise is harmful both in causing spurious output fluctuations and in\noverloading the power stages of the servo system.\nBoth of these effects\nare reduced by narrowing the frequency band of the servo transfer charac-\n\nteristic.\n\nReferring to the expression for the transfer characteristic given\n\nby (6), it may be seen that a restricted transfer bandwidth may be obta\u2018ned\nby reducing u and uf well below unity at a small value of signal frequency.\u00b0\nOn the other side of the picture is the requirement of fidelity in maintaining the desired input-output relationship. Undue narrowing of the\ntransfer bandwidth of the servo results in large dynamic error, the magnitude of which depends both upon the character of the input signal and\nupon the chosen transfer characteristic.\nThe optimum design of a servo system, for a specified input signal and\nnoise, thus is a compromise between dynamic error and output noise fluctuations, with stability considerations and parasitic circuit elements restricting the possible choice of loop transmission characteristics.\n3.1 Stability of Single Loop Systems\nThe word stable as applied to a servo system is used here to imply a system whose transient response decreases with increasing time. It is possible\n6 When the 8 characteristic is under suitable design control, another method is available:\nThus if 8 is made to rise in the frequency region of the desired transfer cut-off, and if 48\nis maintained large beyond this region, (6.1) shows that the desired restriction is effected\n\nFor a given transfer characteristic, this cut-off method requires a wider frequency range\nfor uf and is thus more vulnerable to the effects of parasitic circuit elements.\n\nHowever,\n\nshaping of both the uw and @ circuits permits a more rapid cut-off of the servo transfer\ncharacteristic than is possible with \u00ab circuit shaping alone.\n\n628\n\nBELL SYSTEM\n\nTECHNICAL\n\nJOURNAL\n\nto determine the stability of a completed servo design by obtaining\n+e\ntransient solution of its differential equation. Though often very tediv:is,\nthis is fairly straightforward.\nHowever, this method of procedure often js\nof little help either in guiding the initial design or in predicting the necessary\nchanges, should the trial design be found unstable. The addition of even\none circuit element to a design will generally create an entirely new differential equation whose solution must be found.\nAn alternative method for determining servo system stability, based on\nfrequency analysis, furnishes the necessary information in a form which\ngreatly facilitates the design procedure. This method is relatively simple\nto apply, even when the system has a large number of meshes and a high\norder differential equation, and the additive effects of minor circuit modifications are easily evaluated.\nThe stability of a single loop servo system\u2014or of a primary loop, when\nthe subsidiary loops are individually stable\u2014may be investigated by plotting the negative of the loop transmission, \u2014u8(jw), on a complex plane\nfor real values of w ranging from minus infinity to plus infinity. (The\nnegative signis introduced because the loop transmission yf is generally\narranged to have an implicit negative sign, apart from network phase shifts.\nThus \u2014uyf is a positive real number when the network phase shift is zero.)\nThen the necessary and sufficient criterion for system stability is that the resulting closed curve must not encircle or intersect the point \u2014/,0.* This type of\nplot is commonly called a Nyquist diagram, and is widely used in the design\nof electrical feedback amplifiers. An added stipulation is necessary if uB(jw)\nbecomes infinite at a real value of w, say w\u2019. In this case an infinitesimal\npositive real quantity \u00ab must be added\nto jw; that is, the function\nto be\nplotted is 48(jw + \u20ac). This has no effect upon the plot except in the neighborhood of the singularity, where u@(jw + e\u20ac) is caused to traverse an arc of\n\ninfinite radius as w is varied through the value w\u2019.\nAs seen from (3.1), inclusion of a motor in a servo loop of the type shown\nin Figs. 5 and 6 will cause an infinite loop transmission at w = 0, assuming\nthere is transmission around the remainder of the loop at zero frequency.\nThe motor is the only commonly encountered circuit element capable of\nproducing an infinite loop transmission at' real frequencies.\nIn order to illustrate the use of the Nyquist diagram, a motor servo system of the type shown in Fig. 6 will be chosen. Again referring to equation\n(3.1), it may be seen that the transfer ratio of the motor and potentiometer\n1S,\n\n* This criterion is due to H. Nyquist, \u201cRegeneration Theory,\u201d\u2019. B. S. T. J., January 1932.\nDetailed descriptions of stability criteria for single and multiple loop systems are given by\nBode, loc. cit., and by L. A. MacColl, \u201cFundamental Theory of Servomechanisms,\u201d D.\nVan Nostrand Co., 1945.\n\nIt is as\nthe ove\n\nwhere .\ngiven 1\n\nwhere\n\nThe t\nunity\ngiven\nto pl\n\nin th\ntity,\ningl}\nima\naxis\n\nequ\n\nLINEAR\n\nSERVO\n\nTHEORY\n\n629\n\nig ed\npm\n\nJ\n\n= jw(jw + wn)\n\nIt is assumed that the amplifier includes an equalizing network such that\nthe over-all amplifier characteristic is\n\n4 (ete\\\n1\n\njo + we)\n\ny\n\n\u2019\n\nwhere A, w, ,and w, are positive real constants.\n\u2018The loop transmission is\ngiven by the product of these two transfer factors and thus may be written\nas\n\n3)\n\nuB(Gw)\n\n= \u201420( om\n\n2 (e\u2014\\\n\nfe tom) 4)\nw\n\nJ\\jwo\n+\n\n7\n\nar)\u2019\n\n@)\n\nwhere w is a positive real constant given by\njin\n\non\na\n\nAS: ut\n\nva\n\na\n\n\u2014\n\nOm J\n\nAS\n=\n\nui\n\n:\n\n.\n\nRn + Rm\n\nThe three factors in parenthesis have been so grouped that they all approach\nunity for small values of w. Thus the low-frequency behavior of \u2014y is\ngiven by wo/jw. This quantity has a pole at w = 0, so that it is necessary\nto plot the function\n\n\u2014uB(jo + \u00ab-)\n\n~\u2014\u201c-,\n\n(7.1)\n\njute\n\nin the neighborhood of w = 0. Aswis, in succession, a small negative quantity, zero, and a small positive quantity; the expression of (7.1) is correspondingly a large positive imaginary, a large positive real, and a large negative\nimaginary. Thus (7.1) traverses an infinite arc from the positive imaginary\naxis to the negative imaginary axis as w increases through the value zero.\nAssuming the numerical values\nwo = 200 sec\njimh\nwm\n\n=\n\n10\n\n\u201c*\n\nw\n\n=\n\n200\n\n\u201c,\n\nequation (7) may be rewritten as\n\ner\n\neee\nuB( Jo)\n\nee\n\njo +10\\/\n\njw (;,+. et\n\n200\nYs + 200)\n\n)\n\u00b0\n\nes\nia\n\nThe phase angle of \u2014y@ in degrees is, from (7.2),\n\nB = \u201490\u2014 an tan\u201d w+ tan\nta +=-10 \u2014\n\n3tan\u201d 00\u2019\n-\n\n(7.3)\n\n630\n\nBELL SYSTEM\n\nTECHNICAL\n\nwhile the absolute magnitude is given by\neer eee\nr\n\nJOURNAL\n\nthe syst\nto knov\n\neS Lae a Nr\n\nco ffom\n1a) = 2w V (K\nEE\nP.\n1+ w\n10?\n200? + w*?/ \u00b0\ngeneral\n\nPe\n\neT\n\ne\n\npee\na\nlhe Nyquist diagram of (7.2)\nis shown in Fig. 8. To emphasize the jim.\nportant features, radial magnitudes have been plotted on a logarithm)\n\ntions, \u00a2\nsign dt\nunallov\ninstabi\nand pe\nsignal.\nStal\nbetwee\ntwo n\n\nINFINITE\nNEGATIVE\n\nwill val\n\nw\n\nThese\nGay 2\nshort\n\ngain I\nincrea\ncrease\nfacto!\na\nnegat\n\nabou\nwere\ncoinc\nTh\nA\nPOSITIVE\n\n:\n\nw\n\nmarg\nserve\n\nthe \u00ab\nrewr\nFig. 8\u2014Nyquist diagram of \u2014y.\n\nscale.\u2019 The arrows indicate the direction of traversal as w is varied from\n\u2014* to-+\u00ab,.\nThe infinite arc traversed as w varies through zero is indicated symbolically by the dotted semicircle in the right half plane.\u2019 As\nis the case for any physical system, the plot for negative values of w is\nsimply the mirror image of the positive frequency plot.\nhe\n\n.\n\n.\n\na\nSince the polar plot does not encircle or intersect the \u2018critical\u2019\npoint \u20141,0,\n\n7 Except in the immediate neighborhood of the origin, where a linear scale must be employed to plot the value ug = 0\n* The exact shape of this arc is of no consequence.\n\nThe\nhoes\n\n2\n\ncircl\n\ntip \u00ab\n\nLINEAR\n\nSERVO\n\nTHEORY\n\n631\n\nthe system is seen to be stable.\u201d From a practical standpoint it is necessary\nto know not only that a design is stable, but that it has sufficient margin\nagainst instability. The need for proper stability margin arises from two\ngeneral considerations.\nFirst, the loop transmission of the physical system\nwill vary with time due to aging, temperature changes, line voltage fluctuations, etc. Also the physical embodiment will depart from the paper design due to errors of adjustment and measurement, and to the effects of\nunallowed-for parasitic elements.\nSecond, a design which is too near\ninstability will have an undesirable transient response\u2014large overshoots\nand persistent oscillations\u2014and will unduly enhance noise in the input\nsignal.\nStability margin is measured in a sense by the minimum displacement\nbetween the polar plot and the point \u20141,0.\nIn feedback amplifier design,\ntwo numbers often are taken as a measure of margin against instability.\nThese are called the gain margin and the phase margin.\nThe gain margin,\nG\u00bb , Measures\n\nthe amount,\n\nin decibels, by which\n\nthe magnitude of y@ fails\n\nshort of unity, at a phase angle of +180 degrees. The numerical value of\ngain margin for the system of Fig. 8 is about 18 db, which is the required\nincrease in amplifier gain to make the servo unstable.\u201d That is, this increase in amplifier gain would multiply the curve of Fig. 8 by a constant\nfactor such that it would intersect the point \u20141,0. The phase margin,\nB\u00bb, , is equal to the absolute magnitude of the angle between \u2014y8 and the\nnegative real axis, at |u8| = 1. Figure 8 illustrates a phase margin of\nabout\n\n50 degrees.\n\nThat\n\nis, if the points on the curve\n\nwhere\n\nywB\n\n=\n\n1.0\n\nwere swung toward the negative real axis by about 50 degrees they would\ncoincide with the point \u20141,0, and the servo would be unstable.\nThe type of transient response obtained with reasonable gain and phase\nmargins is indicated in Fig. 9, which shows the response of the illustrative\nservo system to an input step. The initial overshoot is about 25\u00b0), and\nthe oscillation damps out very quickly. For the general case, (6) may be\nrewritten\n\nin the form\nos\nF, =\n\nThe\n\nrelation\n\nF, =\n\nF,/\u20148\n\nmay\n\n|\n\n1\u2014\n\nF\nHB\n\nwp}\n\n|\n\nbe regarded\n\n: \u2018\n\n(8)\n\n\u2014B\n\nas the desired\n\nbracketed factor acting as the inevitable modifier.\n\none,\n\nwith\n\nthe\n\nThen if the quantity\n\nWith more complicated systems it may not be obvious whether or not the plot en\ncircles \u20141,0. A simple test employs a vector with its origin at the \u20141, 0 point and its\ntip on the curve. If the vector undergoes zero net rotation as it traces along the curve\nfrom w = Oto w = ~, the curve does not encircle the critical point.\n10 In some servo systems a decrease in amplifier gain also may cause instability.\nSuch\nsystems are still covered by the polar plot criterion of stability, and are commonly called\n\n\u2018Nyquist stable,\u201d or \u201cconditionally stable.\u201d\n\n632\n\nBELL SYSTEM\n\nTECHNICAL\n\nJOURNAL\n\n\u2014yf exhibits gain and phase margins of the order of 10 db and 50 deg:\nrespectively, the transient response of the modifying factor to a step func:\nwill be well-damped and generally not overshoot more than about 25\nIf the gain margin is sufficient, the phase margin usually will be the dominant\nfactor in determining the size of the initial overshoot.\nThe required phase\nmargin for critical damping depends upon the exact shape of uB(jw), but\nin general is about 60 degrees. The gain margin needed in a particular\ndesign will depend upon the expected variability of the loop transmission\nRadar tracking loops should usually have gain margins of the order of 15\n\nRESPONSE\nSERVO\n\n0.1\n\n0.2\n\n0.3\nTIME\n\nIN\n\n0.4\n0.5\nSECONDS ->\n\n0.6\n\n0.7\n\n0.8\n\nFig. 9\u2014Transient response of illustrative servo system.\n\ndb or more because of the large number of factors which may cause the Joop\ngain to vary.\nWhile the polar diagram gives a clear picture of stability considerations,\nit is usually more convenient for design purposes to plot the gain and phase\nof \u2014yf as separate curves on a logarithmic frequency scale, for positive\nvalues of w. This is illustrated in Fig. 10, for the sample servo system.\nUnder two commonly met conditions, the requirement for single loop\"\nstability on this type of plot is simply that the absolute value of phase shift\nbe less than 180 degrees at zero db gain (|u8| = 1). The conditions are\nthat the connective polarity be such as to make \u2014yf positive when the\n\" Again, multiple loop systems may be included if all subsidiary loops are individually\nstable.\n\nnetwork\none fre\nAn ac\nforms 0\nand thr\n\nbe seer\n\n2000,\nare dra\n\nLINEAR\n\nSERVO\n\nTHEORY\n\n633\n\nnetwork phase shifts vanish, and that the gain curve cross zero db at only\none frequency.\u201d\nAn advantage of this logarithmic diagram is that commonly encountered\nforms of |u8 |vary as w*\" for intermediate or asymptotic frequency regions,\nand thus plot as corresponding straight line segments.\nFrom (7.4) it may\nbe seen that the illustrative form of |u@| behaves, in turn, as 200w,\n200w~*, 20w~!, and 1.6 X 10%w~4, as w is increased. These asymptotic lines\nare drawn in lightly in Fig. 10, the actual gain describing smooth transi-\n\n- 100\n\n-120\n\u201d\n\na\nWw\n\n\u00a9\n\n-140\n\nWw\n\nte\n\u00ab\nC\no\n\no\n\nz\n\nz\n\n_\n\nz\n\n-160\n\n<\n\n4\n<q\n\n0\n\na\n\n8\n\n-180\n\na\n\n-200\n\n1000\nwo\n\nFig. 10\u2014Loop characteristic of illustrative servo system.\n\ntions between adjacent asymptotes.\nSince the logarithmic slope of w*\" is\n+6k db/octave,\" the successive asymptotic slopes in Fig. 10 are \u20146, \u201412,\n\u20146, and \u201424 db/octave. The junctures of adiacent asymptotes occur at\nvalues of w of 1, 10, and 200. These juncture frequencies are called \u2018\u2018corner\u2019\u2019 frequencies, and may be seen from (7.2) to coincide with the real constants or \u201croots\u201d added to jw in each of the three factors in parentheses.\nThe corner associated with the factor 200/jw is of course atw = 0. The\ncorner of the last, or cubed factor is a multiple one, joining two asymptotes\ndiffering in slope by 18 db/octave.\nFrom a knowledge of such corner\n12 This discussion assumes that uf has a low-pass configuration; that is, only the high\nfrequency cut-off is considered. If u8 has a ow-frequency cut-off also, then a corresponding requirement identical with the above must be added.\n13 An octave is taken to be a 2:1 span in frequency, and 6 db is of course very closely\nequivalent to a 2:1 increase in |uA |.\n\n634\n\nBELL\n\nSYSTEM\n\nTECHNICAL\n\nJOURNAL\n\nfrequencies, and the fact that the actual gain curve lies 3 db from an\nlated simple asymptotic corner, the gain curve can usually be drawn jn\nwitheut further computation.'* The phase curve also is easily constructed\nby adding up the elementary phase curves associated with the various corners. As may be seen from (7.3), these component phase curves all w)\\!\nhave the same shape on a logarithmic frequency plot, merely being shifte\nalong the frequency scale. The phase contributed by each simple corner\nwill be + 45 degrees at the corner frequency, the sign depending upon\nwhether the associated root appears in the numerator or in the denominator.\nIt is an extremely important fact that the very requirement of stability\nimposes an unambiguous interrelationship between the gain and the phase\nshift of most types of transfer characteristic!\nBy the general mathematical methods leading to the previously discussed stability criteria, Bode!\u2019\nhas shown that this is true for the broad class of network structures com\nmonly used in feedback loop design. \u2018That is, if either the transfer gain\nor phase shift is specified at all frequencies, the attendant phase or gain can\nbe computed without further information.\nThis class of networks is called\nminimum phase. Any stable structure composed of lumped circuit ele\nments will have a transfer characteristic of the minimum phase type, pro\nvided it does not include an all-pass section.!\u00ae All-pass characteristics are\nseldom used in the design of feedback loops, since their inclusion in the loop\nalways reduces the stability margins achievable with a given high-frequency)\ncut-off. Thus the unique interrelationship between phase and gain may\nbe assumed for the loop characteristic \u2014y@ in single-loop feedback systems.\nThe nature of this relationship is discussed in detail by Bode.\nBriefly, the\nphase shift at any frequency w, is proportional to a weighted average of\nthe gain slope in db/octave, over the entire logarithmic frequency scale.\nThe weighting factor sharply emphasizes gain slopes in the immediate vicinity of w, , while the contributions of gain slopes at remote frequencies are\nreduced in proportion to the logarithmic frequency span from the particular frequency w,.!? For transfer characteristics of the form w\u2019,\nhaving a constant gain slope of +6k db/octave,'* the associated phase shift\nis also constant and equal to +90 degrees. For transfer functions which\nbehave approximately as w*\u201d over a finite frequency span, the phase shift\n\u2018* The corner frequency concept is less useful if the roots are complex. However a\ngreat many servo systems are so constructed that uf has only real roots.\n* Loc. cit. Also see \u201cRelations between attenuation and phase in feedback amplifier\ndesign,\u201d by H. W. Bode, B. S. T. J., July 1940, p. 421.\n6 An all-pass section is one which has constant gain but varying phase shift versus\n\nof +9\nis incl\nThi\nFig. 1\nappro\n\n\u201410.\n\n\u201490\nshift |\nappre\nFo\n\nthe s\nable\ndesig\nterist\n\nzero\nshift:\n\ning 1\nthat\nchar\nable\n\n$.2 .\nA\ntion\nsub:\n\nque\nwhi\nable\nof t\nque\nTh\nnoi\n\ninf\n\nfrequency, and is usually composed of a lattice, bridged T, or other bridge type circuit.\n\n1 About 60% of the area under this weighting function lies between w = 0.5 w, and\nw = 2w., 80% between 0.25 w, and 4 a, .\n16 That is, for transfer characteristics whose\n\nabsolute magnitude is given by wt*---.\nsit!\n\nLINEAR\n\nSERVO\n\nTHEORY\n\n635\n\nof +90k degrees is approached more and more closely as the length of span\n\nis increased,\nThis may be observed qualitatively from the transfer characteristic of\nFig. 10. For w \u00ab1, the gain slope is \u20146 db/octave, and the phase shift\n\napproaches \u201490 degrees.\n\nFor 1 < w < 10, the average gain slope is about\n\n\u201410 db/octave, and the phase shift near w = 3 is \u2014148 degrees (instead of\n\u201490 X 10/6 = \u2014150 degrees). As w increases toward 200, the phase\nshift increases rapidly due to the asymptotic slope of \u2014 24 db/octave, finally\napproaching \u2014 360 degrees (\u201490 X 24/6) for w> 200.\nForeknowledge of the inevitable gain-phase relationship is of great value to\nthe servo designer, in making clear the comparatively small class of realizable gain-phase combinations and thus averting attempts at non-physical\ndesigns. For example the design use of too-rapidly falling loop gain characteristics in the region of the high-frequency gain cross-over (that is, near\nzero db loop gain) is not permissible because of the large negative phase\nshifts which must accompany the steep gain slopes. Another way of stating the advantage of an early realization of the gain-phase laws is to say\nthat the designer is assured in advance that any paired gain and phase\ncharacteristics which he chooses within the basic restrictions will be achievable with stable physical networks.\"\n3.2 Dynamic\n\nError\n\nA servo system is usually designed to transmit some class of ing.ut functions with a required degree of fidelity. Thisiclass of functions may reduce\nsubstantially to one specific input signal whose time variation or whose frequency spectrum is known, or it may include a great variety of signals\nwhich have certain properties in common.\nIn the latter case it is conceivable that definite limits may be placed upon the allowable amplitude ranges\nof the input signal and its various time derivatives, or certain limiting fre-\n\nquency spectrum characteristics may be specified for the input function.\nServomechanisms are subject to several types of transmission error.\nThe systematic error, or dynamic error, is predictable from knowledge of the\nnoise-free input signal and of the transfer frequency characteristic of the\nservo system.\nFor simplicity, the discussion of error will be limited to the\ncase where the output signal is desired to be a replica of the input, and\nwhere 8 = \u20141. Thus the loop transmission 48 becomes simply \u2014u.\nThe\ninput-output relationship as given by (6) is therefore\n\nP= 1+u_*_f,\n.\n\nd\n\nx\n\n(9)\n\n19 With some necessary reservations as to practicable dissipation constants and parasitic circuit constants.\n\n.\n\n636\n\nBELL SYSTEM\n\nTECHNICAL\n\nJOURNAL\n\nwhere F; and F2 are again typical sinusoidal components of the input an\noutput respectively. Thus the corresponding sinusoidal error componeni\nmay be written as\n\n(9.1),\nmand\nsigna\ncomp\n\ns=h-\n\nhot.\n\n(9.1)\n\nThe methods which may be used to determine the actual dynamic error\nA(t) from (9.1) depend both upon the nature off;(#) and the type of information available about f(#). If the input signal is a known periodic function,\nA(t) may be found by applying (9.1) for each sinusoidal component of the\ninput and summing the resulting terms. If the input is non-periodic in\ncharacter, then the error may be calculated from the Fourier integral expression\nA(t) =\n\nMs\n\n\u201d Fi(w)\nF\nBe\nae e** du,\n\n(10)\n\nwhere F;(w) represents the continuous seadaned spectrum of fi(#), as obtained from\nF,(w) = | file** dt.\n\n(10.1)\n\nThe problems of calculating /;(w) from fi(\u00a2) and A(\u00e9) from F;(w) often may\nbe avoided by consulting well-known tabular lists of paired time and frequency functions.\u201d\nEquation (9.1) may be used as a broad guide in selecting the type of u\ncharacteristic best suited to a particular input signal. It has been mentioned previously that because of input noise and parasitic circuit elements,\nthe servo transfer bandwidth usually should be kept as narrow as possible,\nconsistent with dynamic error requirements.\nThe transfer characteristic\nu/(1 + yu) will be closely equal to unity while |x| > 1, will rise slightly\u201d\nin the region where |\u00bb | = 1, and fall off as u when u is small compared\nwith unity. The \u201c\u2018cross-over\u201d frequency, for which |\u00bb |= 1, may be taken\nas a rough measure of the transfer bandwidth.\nThus, the requirement of\nminimizing the bandwidth may be restated as that of minimizing the crossover frequency, while holding the dynamic error within specified limits.\nReasoning in a general way, this requirement may be met by designing\nso that the amplitudes of the sinusoidal error components, as given by\n20 An excellent list is given by G. Aj Campbell and R. M. Foster in a Bell System monograph \u2018\u2018Fourier Integrals for Practical Application,\u201d ag ig 1931. \u201cA table of Laplace\nTransforms, which also may be used, is given by\nF. Gardner and J. L. Barnes in\n\u201cTransients in Linear Systems,\u201d John Wiley por /\nAred Inc., 1942.\n21 Assuming a phase margin of the order of 60 degrees.\n\nover\n\nsigna\neven\nreduc\nina |\n3.21\nFr\n\ntype\ning f\nof le\nener}\nversi\ninpu\n\n|w |\nredu\ncont\n\nlecte\nserie\n\nwhe\nB\ncrea\n\nfew\ntota\n\nwhe\n\npro]\nof t\nspec\nmis\nFor\nlarge\nsma\n\n)-\n\no\n\nLINEAR\n\nSERVO\n\nTHEORY\n\n637\n\n(9.1), are roughly constant with frequency over the servo band.\nThis demands that 4 have somewhat the same frequency distribution as the inpul\nsignal spectrum (for |u| >> 1). Because of stability requirements and\ncomplexity of the necessary apparatus, this rule can usually be followed\nover only a part of the servo frequency band, especially when the input\nsignal spectrum falls off very rapidly with increasing frequency.\nHowever,\neven a rough adherence to this desired relation is usually of real worth in\nreducing the noise errors of the servo.\nAn illustration of this will be given\nin a later section.\n3.21 Approximate Calculation of Dynamic Error\nFrequently the servo requirement is to transmit, with great accuracy, a\ntype of signal whose frequency spectrum falls off very rapidly with increasing frequency. As may be seen from (9.1) this demands very large values\nof loop transmission \u00bb at the lower frequencies where the input signal\nenergy is concentrated, but permits a rapidly dropping loop transmission\nversus frequency commensurate with the falling amplitude spectrum of the\ninput signal. Such a rapid reduction in loop gain is practicable while\n|u| > 1. However. stability considerations force a more gradual gain\nreduction as the region of gain cross-over is approached.\nAs a result,\ncontributions to the servo error from this frequency region may be neglected compared with those from the lower frequencies.\nThis suggests a\nseries expansion of (9.1) in the form,\nA =\n\n[ao a ay( jw) +\n\n@2(jw)? a a3( jw)? 4+ -- \u201c| F, P\n\n(11)\n\nwhere do , a; , etc. are real constants.\nBecause of the assumed rapid drop in component amplitude F; with increasing frequency it is often unnecessary to take account of mote than a\nfew terms of the expansion.\u201d\nIt is,\u00a2asy to show that (11) may be rewritten on a time basis to give the\n\ntotal #foamic error as\nA(t) = aofi(t) + ay fi(t) + aa} 1(t) 4 see,\n\n(12)\n\nd( )/dt. Thus the coefficient ao gives the error component\n\u2018nal to input displacement.\nSimilarly, a; and a are the coefficients\nror components due to input velocity and input acceleration, reiy. For a great many motor-drive servo systems the loop transu approaches infinity as 1/jw when w approaches zero. This enseries may be said to converge rapidly in a practical sense, for the following reason:\ni values of w the higher order terms are negligible.\n\nFor values of w sufficiently\n\nlarge ts (: the high order terms may no longer be neglected the coefficient F; has become so\nsmall w to make the contribution of the entire series negligible.\n\n638\n\nBELL SYSTEM\n\nTECHNICAL\n\nJOURNAL\n\nsures that a and thus the displacement error will be zero, leaving pri\ncipally the velocity and acceleration errors to be considered.\nThe coefficients ao , a; , a2 , etc. may be calculated easily for any partir\nlarcase.\nFor illustration, the three common forms ofu characteristic sho,\nin Fig. 11 will beexamined.\n(As previously discussed, the designated forn\nof uw need hold only for | u | > 1.)\n\nAgair\na cele\n\nType\nTh\nservo\n\nand t\n\nand \u00ab\ncom]\n\n3.3 1\n\nTt\nsigna\n|u|\nDECIBELS\nIN\n\nwhe!\nnoise\ncalet\n\never\nstan\nated\nnois\n\nWy\n\nLOG\n\nW\u2014e\n\neasil\nfreq\n\nFig. 11\u2014Elementary yu forms.\nType\n\n1.\n\n\u20146\n\ndb/octave,\n\na=\n\nwo/Jw\n\nThe error expansion becomes,\n\n1 :\nA(t) = \u2014f,\n\nL's\n\u2014-sfA@+---.\n\n3)\n\nWo\n\n(12.1)\n\nFor the combination of high accuracy and rapidly converging input spec\ntrum, the first term is the only one of importance.\nThus this type of system has essentially a velocity error.\n\nType 2. \u201412 db/octave, up = (wo/jw)?\nHere the error is\nis\n\npe\n\nA(t)\nf(t)\u201c \u2014 wi hil + Sad\n(\u00a2) = we!\n\nLINEAR\n\nSERVO\n\nTHEORY\n\n639\n\nAgain for the rapidly converging case, this system will have principally an\nacceleration error.\nType 3. \u20146, \u201412 db/octave, wp= wows/Jw(Jw + w1)\nThis is perhaps the most commonly encountered characteristic in simple\nservos. The corresponding error expansion is\n-\n\noe\n\nA(\u00e9)\n\n=\n\n\u2014\n\nf(t)\n\na\n-f-\n\nWo\n\n,\nfil\n\n\u2014\n\n,\n\n2\n\nWo W1\n\nAil)\n\n\u2014.\n\neee\n\n(wo\n\n>>\n\n@)).\n\n(12.3)\n\nWy W1\n\nand the principal error for this type system thus is a combination of velocily\nand acceleration components.\nEither the velocity or the acceleration error\ncomponent may be predominant, depending upon the various parameters.\n3.3 Noise\n\nErrors\n\nThe typical sinusoidal component of servo error due to noise (unwanted\nsignals or irregularities) in the input signal may be written as*\n\n4, =\n\n\u2014\u2014\u2014 N,\n\n(13)\n\n1+u\n\nwhere N represents the corresponding sinusoidal component\nnoise.\n\nof the input\n\nIf the noise signal #(/) is known, the total noise error A,(/) may be\n\ncalculated from\n\n(13) in the ways described\n\never, the noise input is seldom\n\nknown\n\nfor the dynamic\n\nin this sense, although\n\nerror.\n\nHow-\n\ncertain out-\n\nstanding components sometimes may he estimated and their effects evaluated. On the other hand the average disturbance due to random input\nnoise, of the kind described as \u2018\u2018thermal noise\u201d\u2019 in electrical circuits, may\neasily be calculated.\nThis type of noise has constant amplitude versus\nfrequency, and the total power in the output noise error may be found from\n@\n\nP.\n40\n\n|\n\nmv\n\nrotons\n1 of uM\n\n2\n\n1-day,\n\nos\n\n(14)\n\n.\n\nwhere K is a constant dependent upon the input noise power.\nInput noise also causes overloading of the power amplifier and overheating of the motor.\nThese effects are aggravated by the falling transfer\ncharacteristic versus frequency of the motor, as seen from the following discussion. The servo transfer characteristic is maintained approximately\nat unity out to the cross-over frequency.\nHowever the transfer ratio of\nthe motor, equation (3.1), will be falling at least at 6 db/octave, usually\nat 12 db/octave, at frequencies below this point. Thus the transfer\n* Again assuming 8B=\n\n\u20141.\n\n23 Assuming that the mechanical load impedance is a series combination of resistance\nand inertia.\n\n640\n\nBELL\n\nSYSTEM\n\nTECHNICAL\n\nJOURNAL\n\ncharacteristic (loop closed) from the servo input up to the motor and pow\namplifier must rise correspondingly with frequency, out to the cross-o.\npoint. Again assuming input noise of the uniform amplitude versus fr\nquency type, the total noise power at the motor input is therefore,\n\nand tk\nmecha\nthat tl\n\nmore |\n3.4 Ce\n\n| om\n\nnee\n\nK,\n\n|\n\n_\n\ni,\n\n(w\n\n+\n\nwn)w\n\ndw.\n\n(15\n\nIn \u00a2\n\nAgain, w\u00bb is the reciprocal time-constant of the motor and K, is a propor\ntionality constant.\nIf @\u00bb is less than about half the cross-over frequency,\nthen the noise power at the motor input increases as the fifth power of the\nbandwidth of the servo transfer characteristic.* Thus, if the input signal\nnoise ratio is small, this effect may be an important design consideration.\nSull other servo errors may result from local extraneous signals or from\ncoulomb and static frictional effects. These error sources are in a some\nwhat different class from those discussed previously, in that they are more\nnearly under the designer\u2019s control. That is, such extraneous signals\nand friction may be kept small by proper design and the residual friction\neffects further reduced by the use of local feedback.\nIn the absence of\nlocal feedback, the servo error resulting from frictional or other torque disturbances at the output shaft readily is found to be\na.\nT\n\nLee,\n\nS(jo)\n\n(16)\n\n1+ 4\n\nS(jw) is the actual! stiffness (loop opened) of the output mesh, and T is the\ndisturbing torque.\n7 conceivably may represent static or coulomb friction,\nload-torque irregularities due to fluctuating running-friction, or wind torque\nAgain assuming the mechanical impedance to be resistance and inertia in\nseries, the mechanical stiffness is, from (2.2), S(jw) = jw(R + jwJ).\n\nThus\n\nthe error is\nAr\n\nT\n\n1\n\n~ jo(R + joJ) 1+ p\u2019\n\n(16.1)\n\nand the apparent output stiffness (loop closed) is\nS\u2019 = jwo(R + go) (1 + py).\n\n(16.2)\n\nIf T is taken as the static load torque, the resulting static error is found\nby setting\n\nw =\n\nO in (16.1).\n\nAssuming\n\nthat \u00bb behaves\n\nas wo/jw when w\n\napproaches zero, the static error is\ni\nAr\n\n=\n\nw) R\u2019\n\n(16.3)\n\n* This assumes a constant functional form for the transfer characteristic. However,\nthe statement holds approximately, even with considerable variation in this form.\n\nfor a]\nan\n\nau\n\nlinear\n6, bei\n\nangle\n\nLINEAR\n\nSERVO\n\nTHEORY\n\n641\n\nand the apparent low-frequency stiffness is woR, being the ratio of the\nmechanical resistance to the velocity error coefficient.\nIt may be noted\nthat the static error will vanish if the loop transmission approaches infinity\nmore rapidly than 1/w as w approaches zero.\n3.4 Comparison ofu Characteristics for a Particular Input Signal\nIn order to illustrate the advantages of shaping the loop characteristic\nfor a particular input signal, a brief discussion will be given of the design of\nan automatic radar loop to track an airplane in azimuth over a constant\nlinear-velocity course.\nThe servo configuration is that given by Fig. 5b,\n6, being the azimuth\n\nangle.\n\nangle of the target and @ the corresponding antenna\n\nThe lobing radar antenna has been assumed to take the place of the\nT\n2\n\n\u201c4\n\n-6\n\n-4\n\n2\n\n6\n\nen\n\nTIME IN SECONDS \u2014=\n\nFig. 12\u2014Azimuth\n\n8\n\nangle for constant linear velocity course.\n\nsynchro pair. Thus 8 = \u20141, and an error signal proportional to 6; \u2014 4 is\n25\ndeveloped.\nAssuming a constant linear-velocity course having a maximum azimuth\nrate of 30 degrees/sec, the target azimuth angle is given by *\nA(t) = tan\n\n.5244\u00a2,\n\n(17)\n\nwhich is plotted in Fig. 12.\nThis course will develop a maximum azimuth acceleration 6 of +10.3\ndegrees/sec? and a maximum @, of \u201416.4 degrees/sec*. The continuous\nfrequency spectrum of 6;(\u00a2) may be found from (10.1) to be\nFie)\n25Assuming\n\na low elevation\n\ncourse.\n\n=\n\n\u00a2\n\n\u20141.9]a]\n\u2014\u2014-..\n\njw\n\n(18)\n\n% The azimuth angle has been so taken that zero azimuth is obtained at the point of\nnearest approach.\n\n642\n\nBELL SYSTEM\n\nTECHNICAL\n\nJOURNAL\n\nA logarithmic plot of |F,(w) | is shown in Fig. 13.*\n\nIt may be seen that tie\n\ninput signal spectrum falls at 6 db/octave for w K 0.5, at 12 db/octa\nforw = 0.524, and 30 db/octave at w = 2.1.\nAssuming that the permissible dynamic error is 0.3 degree, a comparison\nwill be made between the type 1 and type 3 loop characteristics of the pr\nvious section.\nFor the type 1 system, which will have essentially a pur\nvelocity error, (12.1) shows the required value of wo to be 30/0.3 or 100\n20\n\nBaal\n\n-12 DB/OCTAVE SLO\nPE\n\nlus\nDE\nin\n\n-20\n-30 0B/OCTAVE\n\nROR\n\n-40\n\ns\n\n(W)|\n|DECIBELS\nIN\nFy\n\n+\n\n+\n|\n\n|\n|\n\n+\n\n-80}-\u2014\u2014\u2014-\n\n~\u2014\u2014-\u2014\u2014+\n\n|\n|\n\n-10\n\n\u2018\n\n085)\n\n02\n\n0.4\n\n06\n\n08\n10\nWw\n\n4\n\n20\n\n40\n\n60\n\nFig. 13\u2014Target frequency spectrum for constant velocity course.\n\nThus \u00bb = 100/jw. Figure 14 shows a logarithmic plot of the corresponding\n|u|. This characteristic departs rapidly from the shape of input signal\nspectrum given by Fig. 13, as w is increased abeve 0.1.\nThe type 3 characteristic permits a considerably better match.\nChoosing\na compromise value for w of 0.1, (12.3) may be used to calculate the necessary value of wo as 415. Thus the loop transmission becomes php =\n41.5/jw(jw + 0.1). Figure 14 shows a plot of the corresponding | x |,\nmodified near the gain cross-over to satisfy .the stability requirements.\nThis curve is a considerably better average match for the target frequencyspectrum up tow = 1. The resulting type 3 system has a predominant\nacceleration error as judged from the maximum velocity and acceleration\nerrors of .072 degree and 0.25 degree respectively.\nThe total dynamic error curves for the constant-velocity course are given\n*| Fi(w) |= x has been taken as the zero db level.\n\n644\n\nBELL\n\nSYSTEM\n\nTECHNICAL\n\nJOURNAL\n\nin Fig. 15. The velocity error of the type 1 system is always a lagging err:\nand is maximum at the point of nearest approach. The type 3 composit.\nof velocity and acceleration errors is lagging over about the first half of th:\ncourse and leading for the second half, having lead and lag maxima at point\nclosely grouped about the point of nearest approach.\nAlthough the two loop characteristics develop the same maximum dy\nnamic error on the specified target course, their transient responses to an\ninput step differ widely, as may be seen from Fig. 16. The rise time for the\ntype 1 loop is about .03 second compared with an initial rise in 0.17 second\n\nthe ne\nto be\nwante\n\nRS\nTh\nservo\ntube\nties,\n\n1.4\n\nWw\n\u201d\nz\n9\n\n\u2014+-\u2014\n\n-\n\n|\n4+ \u2014\u2014_____+--___J\n\n+>}\n\n\u201d)\nud\n\n\u00ab\n\n\u00b0\n>\n\n+\n\n$$$\n\n\u2014_\u2014_\u2014\u2014_+\u2014_\u2014___+___\n\na\nad\n\n\u201d\n\n+\u2014\u2014\u2014\u2014_}+\u2014\u2014\u2014\u2014-++\n|\n\nt\n\na\n\n|\n\n|\n\n|\n\n0.3\n\ni\n0.4\n\n|\n0.5\n\n0.6\n\n07\n\n0.8\n\nTIME IN SECONDS \u2014>\n\nFig. 16\u2014Transient response of tracking servos.\n\nfor the type 3 system. Also, because of the overshoot the type 3 system\nrequires about 0.7 second to settle within 5% of the equilibrium value.\nFor a final comparison of the two systems the corresponding transfer\ncharacteristics, u/(1 + qm), are plotted in Fig. 17 on arithmetic amplitude\nand frequency scales. It may be seen that the type 1 system is vulnerable\nto noise and interfering signals over a far wider frequency range than the\ntype 3. Again assuming uniform input noise versus frequency, (14) may\nbe used to show that the ratio of output noise power for the two systems is\nabout 7.5:1.\nThus the luxury of crisp transient response as obtained with the type 1\nsystem may demand a heavy penalty in terms of output fluctuations due to\nnoise and other unwanted signal variations. This is a clear illustration of\n\n|\n\nsup\nfrec\n\u201c6\n\nsta\n\ntac\n\nfee\n\ning\ntac\n\nant\n\nle\n\nLINEAR\n\nSERVO\n\nTHEORY\n\n645\n\nthe necessity for designing the servo loop to match the type of input signal\nto be transmitted, particularly for radar tracking systems where the \u2018\u2018unwanted variations\u201d are ever present.\n3.5 Use of Local Feedback\nThere are many examples of the use of local or subsidiary feedback in\nservo systems. The more common of these include feedback around vacuum\ntube power amplifiers to obtain improved linearity and impedance properties, and over-all feedback around amplifier and motor-drive systems to\n\nRATIO\nTRANSFER\n\nte)\n\n40\n\n80\n\n120\n\n160\n\n200\n\n240\n\n280\n\n320\n\nW->\n\nFig. 17\u2014Frequency response of tracking servos.\n\nsuppress frictional effects, increase output stiffness, and modify the inherent\nfrequency characteristics of the basic components.\u201d\nThe tendency toward\n\u201c8 circuit dependency\u201d as previously discussed also produces greater constancy of the stage transfer characteristics with time, temperature, etc.\nPerhaps the simplest and most useful kind of local feedback is negative\ntachometer (velocity) feedback around motor-drive systems. This type of\nfeedback widens the transfer frequency band of the drive system by reducing its time-constant, and increases the linear speed range of the motor.\nThis may be illustrated by referring back to Fig. 7, which shows a typical\ntachometer loop. Assuming the transfer ratio of the amplifier to be a con27 In a slightly different class are the servo systems used to provide automatic frequency\n\nand gain control in radio systems.\n\n646\n\nBELL SYSTEM\n\nTECHNICAL\n\nJOURNAL\n\nstant A, the transfer ratio of the motor and amplifier without feedback js.\nfrom (3.1),\ntr\n\n=\n\n=\n\n6\n\n\u2014 (loop\n\n(loop\n\nopen)\n\n=\n\nopen)\n\nMo\n\n1\n\n\"\u2014 \u2014\u2014\u2014___\u2014.\n\n(19\n\nJ jw(jw + wm)\u2019\n\nwhere the constant wo = Apw;. (To avoid confusion with primary loo;\nquantities, the tachometer loop will be represented by the symbols ur and\n8,,rather than and.)\nThe quantity w,, was defined as the ratio (R,, 4\nR;,)/J (see Fig. 2b), and is the reciprocal of the motor time constant.\nReplacing (R,, + Ry) by R for convenience, (19) may be rewritten as\n@\n\n.\n\nMo\n\n. = \u2014 (loop open) = \u2014\u2014\u2014\u2014;. .\naie loop of\njwo(R + juJ)\n\n.\n\nEqu\n\nwhere\nThe\nshow!\n\nwitho\nsuch\na\nWm/\n\nWy\n\nlow-fi\n\n19.1)\n\nThe transfer ratio of the tachometer is\nBr\n\nEs\n6\n\n:\n\n=\u2014=\n\n\u2014jwR, \u2019\n\nand thus the loop transmission characteristic is\n2\nMr Br =\n\npo Ry\n\n\u2014-_-\u2014_,\nR+ joJ\n\n(20)\n\nFor values of w small compared with w,, this loop transmission is constant\nand closely given by wrBr(0) = \u2014poR:/R.\nWhen w > wm, urBr approaches the form \u2014yoR;/jwJ, and thus falls off at 6 db/octave. Consequently the maximum phase shift of the factor \u2014u787 is \u2014 90 degrees, and\nno stability problem arises for the local tachometer loop.\u201d\u00ae\nFrom (19.1) and (20), the over-all transfer ratio with feedback is\nH (loop closed) = yen\n(21)\n\u2014_\n\nHo\n\nSees\n\n~ jo(R + woR: + jot)\u201d\nComparing (21) with (19.1), it may be seen that the sole effect of the tachometer feedback upon the over-all transfer ratio has been to add an apparent\n\u201cohmic\u201d\n\nfriction\n\nor mechanical\n\nresistance\n\noR;\n\nto the original\n\nvalue\n\nR.\n\n(It will be shown that this increase in apparent mechanical resistance also\nis effective in increasing the mechanical output impedance, although no\npower is dissipated in the added component yoR;:.)\n8 Actually, the effects of parasitic elements always modify this situation somewhat,\nespecially if unusually high loop transmission is sought. However tachometer loops often\nrequire little or no stabilizing equalization.\n\nste\n\nLINEAR SERVO\n\nTHEORY\n\nEquation (21) also may be written as\n: (loop closed)\nwhere w,, =\n\n=\n\nLo\n\n1\n\nI jae\n+oh)\u2019\n\na\n\n(R + woR,)/J is the new corner frequency.\n\nThe change in over-all transfer ratio due to the tachometer feedback is\nshown in Fig. 18. The solid line diagrams A and B are the transfer gains\nwithout feedback and with feedback, respectively.2\u00ae At low frequencies\nsuch that w < w,,, the feedback reduces the transfer ratio by the factor\nwm/W@m, the ratio of the two corner frequencies.\u201d In order to restore this\nlow-frequency loss in transmission, it is necessary to provide an added\na)\n\nal\nw\n\n@\n\nUv\n\nwW\n\n{a}\n\nz\n\u00b0\n\u00e9\n\n<\nc\nx\n\nWwW\nu\n\u201d\nZz\nqt\nc\n\n|\n\n|\n|\n|\n|\n1\n\na\n\nWm\nLOG\n\nW\n\nFig. 18\u2014Effect of tachometer feedback on motor characteristic.\n\namplification wm/wm.\n\nIf this is accomplished by increasing yo and decreas-\n\ning R, so that the product yok; remains constant,\"\n\nthe resulting transfer\n\nratio will be that shown by the dotted lines C in Fig. 18. Comparing A and\nC, it may be seen that the net result of applying tachometer feedback and\nincreasing the amplifier gain is to widen the transfer bandwidth by the\nfactor wm/wm.\n\nThe required increase in amplification is the cost of widening\n\nthe transfer bandwidth either by tachometer feedback or by non-feedback\nmeans, such as the use of a \u2018\u201c\u2018forward-acting\u2019 \u2019 equalizer in the amplifier.\n(However, such forward acting equalization fails to provide the increased\nover-all linearity and mechanical impedance obtained by the feedback\nmethod.)\nAt frequencies sufficiently high that w >> w,, the change in\ntransfer ratio due to the feedback disappears, the mechanical inertia becoming the controlling element.\n29 The straight line asymptotes have been drawn instead of the actual gain curves.\n380 This is also the factor by which the feedback reduces the output speed obtained for a\nsteady input voltage, neglecting circuit non-linearites and coulomb\n\nfriction.\n\n31 This ensures a fixed loop transmission, and thus an unchanging value for ws .\n\n648\n\nBELL\n\nSYSTEM\n\nTECHNICAL\n\nJOURNAL\n\nbeen in\n\nFor w small compared with w,, , (21) becomes\n.\n\n.\n\n,\n\n,\n\nlocal fe\n6\n\nMo\n\nz - |(loopop closed)\nclosed)\n\na: jo(R\n\u2014\u2014\u2014\u2014\u2014\u2014\n+ wR)\u2019\n\nIf the tachometer feedback is substantial (wm >\n\napproximated\n\n(w\nK\n\nwm), this may\n\n,\n\nw,).\n\nbe further\n\nas\n\n0\n\n(loop closed)\n\n~\n\nene\n\n1\n\n-\u2014~\n\nJoR,\n\nw Kw\u00bb\n\n,\n\na\n\nWm >> Wm\n\nP\n\n21.2)\n\nah\n\nand the corner frequency becomes\n\nthe tw\neductic\ntorque\n\nmay b\n3.6 Er\nIns\nmay b\nsysten\n\niR\n\nWin ~ \u2014\n\n:\n\n(cm > Wm).\n\nThus for reasonably high feedback, the over-all transfer ratio (21.2)\ndepends only upon the tachometer characteristic, being substantially independent of changes in the original mechanical resistance R or the amplifiermotor factor uo. The corner frequency wm is similarly independent of\nchanges in R, although still a direct function of uo. Thus the principal\nnon-linearity of two-phase induction motors, namely variation in electrical\ndamping with speed, is effectively suppressed by this type of local feedback, and systems employing such motors up to 80\u00b0% of their synchronous\nspeed may be designed on a linear basis.\nThe increase in mechanical impedance due to the feedback may be shown\nby assuming a torque disturbance 7 applied at the output shaft. Without\nfeedback, the resulting speed disturbance is\n6 (loop\n\nopen)\n\n=\n\nF: =\n\nRE\n\n\u00b0\n\nTha\ninpu\nto r\n\nWith feedback, the corresponding shaft speed disturbance becomes\n\n6 (loop\n\nclosed) =\n\n( tipi\n\nT\n\n1\n\n3- - \u2014\u2014\u2014\u2014.\n\nZan\n\ni-\n\n,\n\nMrBr\n\nT\n\nR + wR: + jw] \u201d\nThus the apparent mechanical resistance, and therefore the protection against frictional\ntorques, has been multiplied by a factor\n(1+ ywoR:/R) = wm/wm.\nIf the motor-drive system with tachometer feedback is employed in a simple follow-up system of the type of Fig. 5,\nequation (16.3) shows that the resulting low-frequency output-shaft stiffness\nwill be wo(R + pwoR:) or (wm /tom oooR22 Therefore the output stiffness has\n*2 The low-frequency loop transmission of the follow-up loop is again taken to be wo/jw.\n\nacce\u00e9\nsim)\nintc\nsyn\nsys!\n\nthe\n\nHal\u2019\nsho\n\nLINEAR\n\nSERVO\n\nTHEOR}\n\n649\n\nbeen increased by the factor w,/w, over that obtained without the use of\nlocal feedback, assuming identical follow-up loop characteristics (u8) for\nthe two cases.\nThe ratio w,/w, thus directly measures the feedback\neduction of static and low-speed errors of the follow-up system due to\ntorque disturbances.\nIn practice the resulting increase in static accuracy\nmay be of the order of 10 to 100 times.\n3.6 Error Reduction by Non-Feedback\n\nMeans\n\nIn situations where the noise associated with the input signal is small, it\nmay be desirable to reduce the dynamic errors obtained with a given servo\nsystem by the use of forward-acting equalization external to the loop.\nADDED\n\n\u2018cn\nhr)\n\n|\n\n|\n\nTHoF\n\nfy (t | Fy\n\nFy-Fo|\n\n|\n\n1\n\n>\n\nFo\n\nHo\n\nfo(t\n\nSERVO\nLOOP\n\nFo\n\n.\n\nena\n\nFOR\n\nA\n\nZERO\n\nDYNAMIC\n\nCnr\n\n1\n\nMeri\n\nFig. 19\u2014Forward-acting error compensation.\n\nThat is, the dynamic error characteristic may be computed, and the servo\ninput or output modified by supplementary networks in such a fashion as\nto reduce the over-all error.\nAn illustrative arrangement, which is suitable when the input member is\naccessible,\u00ae is shown in Fig. 19.\n\nFor convenience the servo is taken to be a\n\nsimple follow-up system having 8 = \u20141. The yu circuit is shown divided\ninto two parts, uw: and we. Typically, u: may be the transfer stiffness of a\nsynchro pair (Fig. 3b), and we the transfer characteristic of a motor-drive\nsystem. The normal dynamic error component for such a loop, omitting\nthe dotted line, has been shown to be F;/(1 + uw).\n\nIf an additional signal\n\nual\u2019; is obtained from the input member and injected into the system as\nshown by the dotted line, then\n\nSt\n\noe\n\n\u2014\n\nten\n\ni\n\nHF\n\nHome\n\noe\n\ni\n7\n\nl+u\n38 This is not the case for a radar tracking loop, for instance.\n\n650\n\nBELL\n\nSYSTEM\n\nTECHNICAL\n\nJOURNAL\n\nus is gi\n\nThus the over-all error becomes\n\nfor sub\nF,\n\nai\n\nF,\n\n=\n\n(1\na\n\nbb +\n\npot) Fy,\n\npared \u2019\n\nM\n\nclosely\n\nat the\n\nor\nFi\n\n1\n\nee De: ee\n2\n\nfrom t\n\n1 \u2014 pape My\n\u00bb\neee\n1\nLu\n1\n\n(22\n\nratios\nmaint:\nThe\n\nIf the added transmission path is so designed that\n\nloop c\nstabili\nwhen\nbecau\nFORWARD-ACTING\n\nFEEDBACK\n\nTACHOMETER\n\nTACHOMETER\n\n\u2014\n\naR\n\u2014\n\nINPUT\n8;\n\n\u2014\u2014\n\n? kL\nee\n_SYNCHRO\nPAIR\nG Eger ES\nL\u2014/\n_\n\nbean\n\n|\n\nOUTPUT\n\n|\n\n85\n\n!\n\n|\n|\n\n4\n\nM\n\nCcctemaninaied\n\n4\n\n\\\u2014\u2014~<% b\n\nes\n\nF\n\n-~jJWR,O5\n\ny\n(FOLLOW-UP\nLOOP\n\n!\n|\n\n;\n\n_\n|\n|\n|\n\n|\n|\n|\n|\n|\n\nFOR ZERO DYNAMIC ERROR, Rt Ry\nFig. 20\u2014Forward-acting tachometer system,\n\nthen fF; = F;, and the dynamic error vanishes. Thus the desired form\nof the added transmission depends only upon the ye portion of the loop\ncharacteristic.\nIt will not be possible to satisfy the condition given by (23)\nexactly, especially at the higher frequencies where noise enhancement and\nparasitic effects will become increasingly important.\nHowever, it is often\npossible to obtain the proper form for wa over the range of frequencies responsible for the bulk of the dynamic error. If ue has the proper frequency\ncharacteristic but is too large by 10%, fcr instance, it may be seen from (22)\nthat there still remains a 10/1 increase .n dynamic accuracy.\nThe foregoing method is especially applicable when ye represents the transfer characteristic of a motor-drive system employing tachometer feedback,\nas shown in Fig. 20. Here the basic input-output comparison is obtained\nby means of the synchro pair, while a tachometer coupled to the input shaft\nprovides the error-reducing signal. Thus the transmission ya is equal to\njwRt, where R; is the tachometer transfer resistance. The expression for\n\nLINEAR\n\nSERVO\n\nTHEORY\n\n651\n\nus is given approximately by (21.2) as 1/jwR,. Thus, by (23), R, = R,\nfor substantial cancellation of the dynamic error (at frequencies small compared with Wm).\n\nThat is, the output voltages of the two tachometers must\n\nclosely annul each other when the input and output shafts are travelling\nat the same speed. Since the tachometers may be closely alike and excited\nfrom the same supply line, it is comparatively easy to keep their transfer\nratios closely matched.\nIn practice an error reduction ef 20/1 is readily\nmaintained by this method.\nThe error compensation scheme described above does not change the\nloop characteristic 48 of the basic servo loop, and thus does not create\n\nnew\n\nstability problems.\nIts use to obtain high servo accuracy is desirable\nwhen the input noise is small and when a high loop gain is difficult to obtain\nbecause of parasitic elements or equipment complexities.\n\nHCl-sa\nwith k\n\nwhich <\n\nAbstracts of Technical Articles by Bell System Authors\nQuai\nComputation of Interfacial Angles, Interzonal Angles, and Clinographic Pro\njection by Matrix Methods.' W.L. Bonp. A way of setting up the general\ncrystallographic axes a, b, c on unit orthogonal axes x, y, z is used to afford a\nmatrix method of computing interfacial angles and zonal angles. It also\naffords a method of making clinographic projections.\nA Current Distribution for Broadside Arrays which Optimizes the Relationship between Beam Width and Side-Lobe Level? C. L. Doupu. A oneparameter family of current distributions is derived for symmetric broadside\narrays of equally spaced point sources energized in phase. For each value\nof the parameter, the corresponding current distribution gives rise to a\npattern in which (1) all the side lobes are at the same level; and (2) the beam\nwidth to the first null is a minimum for all patterns arising from symmetric\ndistributions of in-phase currents none of whose side lobes exceeds that level.\nDesign curves relating the value of the parameter to side-lobe level as well\nas the relative current values expressed as a function of side-lobe level are\ngiven for the cases of 8-, 12-, 16-, 20-, and 24-element linear arrays.\n\ncompe\ncrysta\u2019\n\nence 0\nquartz\nof the\n\ndevelc\nimme\nour 1\nand r\ncles c\nTech\n\nTh\nquar!\nspeci\ncryst\n\nvaric\nothe\n\nthe |\n\nPaper Capacitors Containing Chlorinated Impregnants\u2014Mechanism of Stabilization L. EGerton and D. A. McLean.\nThe stabilization of paper\ncapacitors containing chlorinated aromatic impregnants with small quantities of organic additives is well established commercially.\nAlthough for\npractical reasons anthraquinone was chosen for initial commercial application, other quinones are also effective, as are the nitroaromatics, maleic\nanhydride, and sulfur. Evidence is given that the mechanism of stabiliza-\n\nmos:\nmen\ncrys\nmer\n\ntion consists of the formation of barrier films on the electrodes.\n\ntail\n\nThese bar-\n\nrier films, which in certain cases may cover only the active points on the\nelectrode surface, reduce the catalytic decomposition of the chlorinated impregnant by the electrode metal, prevent attack of the electrodes by liberated\nhydrogen chloride, and hinder electrolytic action. It appears likely that the\nfilm-forming properties of the stabilizers are dependent upon their oxidizing\npower. A secondary effect of stabilizers may be the formation of complexes\n\nwith aluminum chloride to diminish the activity of the latter or change the\nnature of the reactions which it induces.\n1\n\nConductivity\n\nAmerican Mineralogist, Vol. 31, pp. 31-42 (1946).\n\n2 Proc. I.R.E. and Waves and Electrons, June 1946.\n8 Indus. and Engg. Chemistry, May 1946.\n652\n\nmeasurements in\n\nthe\n\nan\nfor\nchi\nfou\narl\ntol\n\ncu\npr\n\nABSTRACTS\n\nOF\n\nTECHNICAL\n\nARTICLES\n\n653\n\nHCl-saturated chlorinated dipheny! containing soluble additives are in line\nwith known hydrogen-bonding tendencies of the additives. Compounds\nwhich are strong organic bases do not stabilize capacitors.\n\nral\nda\nlso\n\nQuartz Crystals for Electrical Circuits* R.A. Hetstnc.\nThis book is a\ncompendium of information, both theoretical and practical, on quartz\ncrystal plates, their design and manufacture.\nIt embodies the vast experience of the Bell Telephone Laboratories in research and in manufacture of\nquartz crystals. It originated from a series of lectures given by the members\nof the Laboratories technical staff who had carried out the early studies and\ndevelopments in this field. By this means, engineers were trained for the\nimmense expansion in crystal manufacture required to meet the demand of\nour military forces during the War. These lectures have been reorganized\nand rewritten, and are published together in this comprehensive book. Articles covering some of the various chapters have appeared in the Bell System\nTechnical Journal.\nThe treatment covers in full the theory and practice of the preparation of\nquartz crystals, the instruments used, including new types developed for\nspecial purposes, the problems encountered in the various uses of quartz\ncrystals, and the full details of the methods devised for their solution. The\nvarious processing chapters, dealing with cutting and grinding, plating and\nother topics of equal importance, include much information that appears for\nthe first time in any book. The account of the manufacturing process is\nmost complete. There are discussions of new practical methods of adjustment to frequency, of the new performance indicator, of a new type of\ncrystal cut that operates at very low frequencies, and many new developments that represent notable advances in crystal technology.\nGeometrical Characterizations of Some Families of Dynamical Trajectories.\nL. A. MacCott.\nThe chief problem considered in this paper is that of obtaining a set of geometrical properties which shall completely characterize\nthe five-parameter family of trajectories of an electrified particle moving in\nan arbitrary static magnetic field. A solution of the problem is found in the\nform of a set of four principal and four subsidiary properties. A geometrical\ncharacterization, in the form of a set of two properties, is also given of the\nfour-parameter family of trajectories of an electrified particle moving in an\narbitrary static magnetic field with an arbitrarily prescribed value of the\ntotal energy. Various other properties of the families of curves are discussed, and the paper closes with a brief consideration of some analogous\nproblems in which the particle moves in a fixed plane.\n\u2018Published by D. Van Nostrand Company, Inc., New York, N. Y., 1946.\n5 Amer. Math. Soc. Trans., July 1946.\n\n654\n\nBELI.\n\nSYSTEM\n\nTECHNICAL\n\nJOURNAL\n\nComparison of Natural and Synthetic Hard Rubbers\u00ae G. G. Wrisprax.\nGR-S, nitrile, and natur.|\nD. B. HERRMANN, F.S. MAtm, and A. R. Kemp.\nhard rubbers are compared as regards compounding, processing, vulcanizs\ntion, and physical and dielectric properties. Natural rubber and GR-S\ncompounds intermediate in sulfur content between hard and soft rubber also\nare compared.\nGR-S and nitrile rubber compositions suitable for commercial ebonite fabrication are described.\nExtensive breakdown of the basi:\ncopolymers has little effect on the physical properties of synthetic ebonites.\nThe time required for the beginning of exothermic reaction in vulcanization\nis longer for GR-S than for natural rubber ebonites. Rockwell hardness is\nSome GR-S ebonites are penetrated to the same depth\ngreater for GR-S.\nas natural ebonites, with a greater tendency toward instantaneous recovery,\nThe two are similar in impact strength, but the ability to withstand a sharp\nbend is characteristic of natural ebonites alone. The latter are superior to\nGR-S ebonites in heat deformation below 60\u00b0 C., but above this temperature\n\nthe reverse is true and nitrile ebonites are superior to both.\n\nWALL\n\n1936; \\\nbeen pr\nFREL\n\nversity.\n\nDSc. (\nElectri\nLabore\nLabor\n\nGR-S ebonites\n\nfundar\n\nare more stable and nitrile ebonites less stable chemically than natural\nebonites. GR-S ebonite dust as a filler increases brittleness. A diatomaceous earth improves the processing properties of GR-S hard rubbers. The\nadverse effect of ultraviolet light on surface resistivity is reduced when a\nGR-S hard rubber is filled with whiting. Natural and GR-S hard rubbers\nare alike in dielectric behavior.\n\nswitch\ndevice\nRol\n\njoinec\nhas s\nwork\n\nSignal and Noise Levels in Magnetic Tape Recording.\" D. E. WooLpRipce.\nThe primary object of the work described here was to determine what properties of the tape and associated magnetic elements are responsible for the\nnoise and signal output levels of magnetic recordings and, if possible, to display in specific equations the pertinent relationships connecting noise and\nsignal levels with the physical properties of the tape and polepieces. In the\ncourse of the study, methods appeared for decreasing the noise and increasing the useful signal reproduced from magnetic tape. These methods and\nsome of the use that Bell Telephone Laboratories and Western Electric have\nmade of them are mentioned in the discussion. While some of the work\ndescribed in this paper has implications for more than one type of magnetic\nrecording process, perpendicular recording on tape is the actual subject matter dealt with. In every case discussed, the record medium was 0.050 inch\nwide and 0.0022 inch thick. Except where otherwise noted, a chrome-steel\ntape was used at a speed of 16 inches per second.\nSIndus. and Engg. Chemistry, July 1946,\n7 Elec. Engg., Trans. Sec., June 1946.\n\nphon\nin co\nused\n\n1917\n\nEng\nCor\npar\n\nwo!\n\n~\nTe\nnet\n19:\ning\n\n656\n\nBELL\n\nSYSTEM\n\nTECHNICAL\n\nJOURNAL\n\nE. J. Ryper, E.E., Polytechnic InstituteofBrooklyn, 1935. Engineering\nDepartment, Western Electric Co., 1922-25; Bell Telephone Laboratori:\n1925-.\nIn the Physical Research Department engaged in contact studies\nand during the war the development of switching devices used in radar.", "title": "The Bell System Technical Journal 1946-10: Vol 25 Iss 4", "trim_reasons": ["leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise"], "year": 1946}
{"archive_ref": "pat-us4622672", "canonical_url": "https://patents.google.com/patent/US4622672A/en", "char_count": 17058, "collection": "archive-org-bell-labs", "doc_id": 29, "document_type": "patent", "id": "bella-qwen-pretrain-doc29", "record_count": 1, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": "pdftotext", "selected_extraction_score": 1.0, "source_family": "google_patents", "source_url": "https://patents.google.com/patent/US4622672A/en", "split": "test", "text": "[75] Inventors: Larry A. Coldren, Holmdel, N.J.;\nKarl J. Ebeling, Gleichen, Fed. Rep.\n\nof Germany\n\n[73] Assignee: AT&T Bell Laboratories, Murray\n\nHill, N.J.\n\n[21] Appl. No.: 572,682\n[22] Filed: Jan. 20, 1984\nS51] Unt. C14 ec ccccecsteeeeennenreeseerensesens HOIS 3/10\n[52] ULS. CL, woe eeecenceseensreeeennereneees 372/32; 372/92;\n\n| 372/97\n[58] Field of Search ............:::see 372/50, 29, 32, 92,\n\n372/97\n\n[56] References Cited\n\nPUBLICATIONS\n\nW. Bludau et al, \u201cCharacterization of Laser-to-Fiber\nCoupling Techniques by their Optical Feedback\u201d, Ap-\nplied Optics, vol. 21, No. 11, Jun. 1982, pp. 1933-1939.\nK. R. Preston, \u201cSimple Spectral Control Technique for\nExternal Cavity Laser Transmitters\u201d, Electronics Let-\nters, vol. 18, No. 25, (1982), pp. 1092-1094.\n\nW. Tsang, N. A. Olsson and R. A. Logan, \u201cHigh-Speed\n\nDirect Single-Frequency Modulation with Large Tun-\ning Rate and Frequency Excursion in Cleaved\u2014Cou-\npled-Cavity Semiconductor Lasers\u201d, Applied Physics\nLetters, vol. 42, No. 8, (1983), pp. 650-652.\n\nH. Kogelnik and C. V. Shank, \u201cStimulated Emission in\na Periodic Structure\u201d, Applied Physics Letters, vol. 18,\nNo. 4, (1971), pp. 152-154.\n\nK. J. Ebeling, L. A. Coldren, B. I. Miller and J. A.\nRentschler; \u2018\u201c\u201cGeneration of Single-Longitudinal-Mode\nSubnanosecond Light Pulses by High-Speed Current\nModulation of Monolithic Two-Section Semiconduc-\ntor Lasers\u201d, Electronics Letters, vol. 18, No. 21, (1982),\n\npp. 901-902.\nPrimary Examiner\u2014James W. Davie\n\n[57] ABSTRACT\n\nA control circuit for stabilizing the emission wave-\nlength of the coupled cavity semiconductor laser 1s\ndescribed which uses measurements of the voltage\nacross the laser cavity to maintain stable single longitu-\n\ndinal mode output.\n\n12 Claims, 3 Drawing Figures\n\n33 35\n| REFERENCE | | DC BLOCKING\nOSCILLATOR [oo *SY~CO*\u00ab\u00e9 RGU\nf | | kHz\nPHASE jf\nSENSITIVE dc\nDETECTOR a a\u2014(+)\u20140 If\n\u00bb REF\n\nU.S. Patent\n\nARB.\nUNITS\n\n-10\n\nFIG. 3\n\nNov. 11, 1986 4,622,672\n\u201cOVe, Ie.\n5, Pete\n\u201c3\nwr\n\u2014 . f\ng\nmA\n| 33) 399\nREFERENCE | OC BLOCKING\nOSCILLATOR [7 *\u00ab)~\u00ab3=s CIRCUIT\n3) } [kHz\nPHASE Sf\nSENSITIVE dc\nDETECTOR (+)e\u2014\u20140 Ip\n\u00bb REF\n\n4,622,672\n\nI\nSELF-STABILIZED SEMICONDUCTOR LASERS\n\nTECHNICAL FIELD\n\nThis invention relates generally to semiconductor\nlasers and particularly to such lasers that operate stably\nwith a single longitudinal mode output.\n\nBACKGROUND OF THE INVENTION\n\nOptical communications systems as presently con-\ntemplated and constructed use a light source, such as a\nsemiconductor laser or light emitting diode, which 1s\noptically coupled to a photodetector through a glass\ntransmission line. The transmission line is commonly\nreferred to as an \u201coptical fiber.\u201d Either amplitude or\nfrequency modulation may be used to convey informa-\ntion. If amplitude modulation is used, the information ts\ntransmitted as a bit stream of 1\u2019s and 0\u2019s with the bits\nbeing both transmitted and detected within predeter-\nmined time intervals.\n\nSuch systems presently operate at wavelengths be-\ntween approximately 0.8 wm and approximately 1.6 um\nwith the longer wavelengths, that is, the wavelengths\ngreater than appoximately 1.2 jum, presently being con-\nsidered most desirable for many systems applications\nbecause optical fibers presently used exhibit minimum\nloss and dispersion within this wavelength range. These\ncharacteristics facilitate design of optical communica-\ntions systems having desirable properties such as high\ntransmission rates, long distances between repeaters.\netc.\n\nHowever, to most efficiently utilize the minimum loss\nand dispersion properties of the optical fibers, the light\nsource should not only emit radiation within this wave-\nlength range but should also have its output concen-\ntrated in a narrow spectral range. In practical terms, this\nmeans that the light source should be a semiconductor\nlaser emitting output of a single longitudinal mode. A\nsingle longitudinal mode is narrow enough in spectral\nwidth that it may be thought of as being a single fre-\nquency. The output contains unwanted secondary\nmodes but these are greatly suppressed in intensity with\nrespect to the wanted primary mode. The requirement\nthat the output be a single longitudinal mode is easily\nunderstood by considering that a finite width pulse will\nspread, i.e., broaden, because of the dispersion proper-\nties of the fiber. If the spread becomes large enough,\nadjacent pulses will overlap and recovery of the trans-\nmitted information by the receiver will be impossible;\ni.e., the bits will not arrive at the photodetector within\nthe proper time interval.\n\nAccordingly, a variety of approaches has been taken\nin attempts to develop single longitudinal mode lasers.\nFor example, distributed feedback (DFB) lasers have\nbeen developed. See, for example, Applied Physics Let-\nters, 18, pp. 152-154, Feb. 15, 1971. Semiconductor\nlasers operating with an external cavity to produce\nsingle mode operation have been reported. See, for\nexample, Electronics Letters, 18, pp. 1092-1094, Dec. 9,\n1982. Additionally, coupled cavity lasers have been\ndeveloped and have been shown to produce single lon-\ngitudinal mode output even under high speed modula-\ntion. See for example, Electronics Letters, 18, pp.\n901-902, Oct. 14, 1982, and Applied Physics Letters, 42,\npp. 650-652, Apr. 15, 1983.\n\nThe operation of a coupled cavity laser is described in\ndetail in the preceding references and may be briefly\nsummarized as follows. A coupled cavity laser has two\n\n10\n\n15\n\n20\n\n25\n\n30\n\n35\n\n40\n\n45\n\n50\n\n35\n\n60\n\n65\n\n2\n\ncavities which are optically coupled to each other.\nEach cavity has separate electrical contacts, i.e., the\nlaser is a three-terminal device. One cavity operates as\nan oscillator while the other cavity acts as an etalon and\nprovides mode selection. That the laser produces a\nsingle mode output is understood, for the case of gap\nwidths approximately nA/2, where n is an integer and A\nis the wavelength of the radiation, by considering that\nthe mode of the coupled cavity laser must satisfy Fabry-\nPerot mode conditions in both cavities. More generally,\nmodes are enhanced at local loss minima because of the\nconstructive interference of optical energy reflected\nfrom the gap with that fed back from the other cavity,\ni.e., the etalon. This may occur near resonance or anti-\nresonance of the etalon as well as intermediate points.\nThis is possible only for a limited number of discrete\nfrequencies.\n\nVariation of the etalon current permits tuning of the\nwavelength across the gain profile of the laser. How-\never, a large number of mode hops may occur as the\netalon current is increased from zero to values above\nthreshold. Although stable output is observed between\nthe mode hops, the mode hop boundaries, as a function\nof etalon current, vary with temperature and also per-\nhaps with device aging. Thus, simply maintaining the\netalon current at a constant value will not guarantee\nstable single mode operation. Mode hops are undesir-\nable because they may cause information to be lost\nbecause, for example, bits are not transmitted as desired\nor they are lost because they are not received within the\nproper time interval. Thus, techniques to maintain sin-\ngle mode operation at a desired wavelength are desir-\nabie.\n\nWavelength stabilization techniques have been previ-\nously developed for both coupled cavity lasers as well\nas various other types of lasers such as semiconductor\nlasers operating with an external cavity. A common\nelement of these prior art schemes is the requirement of\nan external optical element, such as either a photodetec-\ntor or a spectrometer, to monitor the light output from,\ne.g., the laser. Feedback loops are then used to adjust,\nfor example, the laser current or a property of the exter-\nnal cavity such as its length if it is a passive cavity.\n\nSUMMARY OF THE INVENTION\n\nWe have found that wavelength stabilization of a\nsemiconductor laser having two cavities optically cou-\npled to each other, such as a coupled cavity laser, may\nbe obtained by using the lasing cavity as a photodetec-\ntor to provide a discrimination signal for a feedback\ncontrol circuit. The feedback control circuit comprises\nmeans: for measuring the voltage variation across the\nlaser cavity, means for comparing the measured voltage\nto a reference characteristic, and means for varying a\nproperty of a second cavity by an amount determined\nby said comparing step. In one embodiment, the laser is\na coupled cavity laser and the circuit comprises a dc\nblocking circuit, a phase-sensitive detector which is\nelectrically connected to the laser cavity, and a refer-\nence oscillator electrically connected to both the block-\ning circuit and the phase-sensitive detector. The output\nfrom the phase-sensitive detector goes to an amplifier\nwhere it is compared to a reference value. The outputs\nfrom the amplifier, which is a function of the difference\nbetween the measured and references voltages, and\nblocking circuit are superimposed on the etalon current.\n\n4,622,072\n\n3\nBRIEF DESCRIPTION OF THE DRAWING\n\nFIG. 1 1s a schematic representation of a coupled\ncavity semiconductor laser;\nFIG. 2 plots a typical differential voltage (laser)/cur-\n\nrent (etalon) characteristic, 1.e., dVz/dIz, vertically\nversus the etalon current horizontally in units of mA;\nand\n\nFIG. 3 is a schematic representation of a wavelength\nstabilized coupled cavity semiconductor laser accord-\ning to our invention.\n\nDETAILED DESCRIPTION\n\n10\n\nOur invention will be specifically described by refer- |\n\nence to the coupled cavity laser schematically depicted\nin FIG. 1. For reasons of clarity, the elements of the\ndevice are not drawn to scale. The laser comprises laser\ncavity 1, etalon 3, electrode 5, laser electrode 7, and\netalon electrode 9. Electrode 5 is a common electrode\nand contacts both the laser cavity and the etalon. Both\ncavities comprise a narrow active stripe 11 where elec-\ntron hole recombination takes place when the laser\ncavity and etalon are forward biased. The laser and\netalon cavities are spaced from each other but are opti-\ncally coupled to each other. The space between the\ncavities may be formed by, for example, etching or\ncleaving a unitary structure. The details of exemplary\nlaser and etalon cavities, as well as their fabrication, are\nwell known to those skilled in the art and need not be\nfurther described in detail.\n\nAs shown, currents I, and I\u00a2 flow through the laser\nand etalon cavities, respectively. The voltages across\nthe two cavities are Vr, and Vg for faser and etalon\ncavities, respectively.\n\nThe operation of the control circuit will be better\nunderstood if the operation and light output characteris-\ntics of the coupled cavity laser are first explained by\nassuming that a constant current, Iz, which is above\nthreshold, flows through the lasing section. If the mag-\nnitude of the etalon current is now varied, the changes,\ndiz, of the etalon current cause variations of the light\ndistribution within the coupled cavity system. These\nvariations also lead to variations, dV ,, of the lasing\njunction voltage as a consequence of the internal photo-\neffect. These variations are typically extremely small.\n\nFIG. 2 shows a typical differential signal, dV7/dlIz,\nplotted vertically in arbitrary units as a function of the\netalon current plotted horizontally in units of mA. The\nlaser was an InGaAsP/InP buried crescent coupled\ncavity laser emitting at 1.3 um. The lengths of the two\ncavities were 222.0 and 191.0 um and the space between\nthem was less than 1.0 um. The current through the\nlasing section was 20 mA above threshold. The laser\nproduces stable single mode output when the etalon\ncurrent is above 3 mA.\n\nAs the etalon current increases, mode hops occur to\nneighboring modes as well as across the entire gain\nprofile of the semiconductor laser. The arrows indicate\nmode hops across the entire gain profile. It was found\nthat a mode hop results in a sudden decrease in the\ndifferential signal with an abrupt mode hop producing a\nsharp negative peak while a more gradual mode transi-\ntion produces a comparatively smooth step. Between\nsuccessive mode transitions, the differential signal usu-\nally increases as the etalon current increases. Thus, the\nbehavior of the junction voltage, Vz, is a function of the\netalon current, Iz, and may be used to detect mode\nhopping. The points of maximum suppression, with\n\n13\n\n20\n\n25\n\n30\n\n35\n\n40\n\n45\n\n50\n\n35\n\n60\n\n65\n\n4\n\nrespect to the primary mode, of the unwanted second-\nary modes are indicated by the solid circles. These cir-\ncles generally lie approximately in the center of each\n\nsingle mode regime. The dV;/dlz slope may therefore\nbe used as a discriminating signal. The wavelength sta-\n\nbilization circuit of this invention tracks the etalon cur-\nrent to the point of optimum mode suppression and\nmaintains the current at that point by adding the ampli-\nfied discriminating signal to the preset etalon bias cur-\nrent. If the etalon current is preset before the feedback\n\nloop is turned on, the desired mode can be maintained.\n\nFIG. 3 1s a schematic representation of our wave-\nlength stabilization circuit. In addition to the laser ele-\nments described with respect to FIG. I, the stabilization\ncircuit further comprises a phase-sensitive detector 3I, a\nreference oscillator 33, a dc blocking circuit 35, and an\namplifier 37. The output from the reference oscillator\n\ngoes to both the de blocking circuit and the phase-sensi-\n\ntive detector. The phase-sensitive detector is connected\nto the laser cavity and to an amplifier which also has a\nreference input, A;yey The amplifier output is a function\nof the difference between the signal from the phase-sen-\nsitive detector and pret. The output of the amplifier, dc\nblocking circuit, and the etalon current, Iz, are com-\nbined and sent to the etalon cavity. Each of these ele-\nments is well known to those skilled in the art and need\nnot be described in further detail.\n\nThe laser cavity is driven above threshold and a small\ndither signal from the dc blocking circuit is superim-\nposed on the etalon current. A typical amplitude is 10\nuA and a typical frequency 1s 1 kHz. The correspond-\ning ac voltage change at the lasing cavity is detected\nwith the phase-sensitive detector. If the measured ac\nsignal deviates from a reference value, indicated as Ayey,\nrequired for optimum suppression of the unwanted sec-\nondary modes, a de correction current is fed from the\namplifier to the etalon thereby forcing the overall eta-\nlon current to the optimum value.\n\nThe laser section current, for the laser previously\ndescribed, was set at I, =20 mA\u2018above threshold and\nthe etalon current was set at 11.1 mA to select the de-\nsired mode. The temperature was then varied and the\noutput spectra recorded. With the wavelength control\ncircuit on, the emission stayed in the same mode over\nthe temperature range between 14 degrees C. and 26\ndegrees C. while several mode hops occurred with the\ncircuit off. )\n\nIt will be readily appreciated that our circuit may be\nused with other types of lasers. For example, the feed-\nback circuit may be used to adjust the length of the\nexternal cavity used with a semiconductor laser.\n\nWhat is claimed 1s:\n\n1. A wavelength-stabilized semiconductor laser com-\nprising a laser cavity, 4 second cavity, said laser cavity\nand said second cavity being optically coupled to each\nother, means for measuring a voltage variation across\nthe laser cavity, means for comparing the measured\nvoltage to a reference characteristic, and means for\nvarying a property of said second cavity by an amount\ndetermined by said comparing step.\n\n2. A laser as recited in claim 1 in which said laser\ncomprises a coupled cavity laser and said second cavity\ncomprises an etalon.\n\n3. A laser as recited in claim 2 in which said means for\nvarying changes the current to the etalon.\n\n4. A laser as recited in claim 1 in which said second\ncavity comprises an external cavity optically coupled to\nsaid laser cavity.\n\n4,622,672\n\n.\n\n5. A laser as recited in claim 4in which said means for\nvarying changes the optical length of said external cav-\nity.\n\n6. A laser as recited in claim 1 in which said means for\nmeasuring comprises a reference oscillator and a phase-\nsensitive detector electrically connected to said laser\n\nand to said oscillator.\n7 A laser as recited in claim 6 in which said means for\n\ncomparing comprises an amplifier having an input from\nsaid detector and a reference input, said amplifier hav-\n\ning an output.\n\n.)\n\n10\n\nI5\n\n20\n\n25\n\n30\n\n35\n\n40\n\n45\n\n50\n\n35\n\n60\n\n65\n\n6\n\n8. A laser as recited in claim 7 in which said means for\nvarying comprises the output of said amplifier.\n\n9. A laser as recited in claim 8 in which said laser\ncomprises a coupled cavity laser and said second cavity\n\ncomprises an etalon.\n10. A laser as recited in claim 9 in which said means\n\nfor varying changes the current to the etalon. |\n11. A laser as recited in claim 8 in which said second\n\ncavity comprises an external cavity optically coupled to\n\nsaid laser cavity.\n12. A jaser as recited in claim 1] in which said means\n\nfor varying changes the optical length of said external\n\ncavity.\n* * * ae *", "title": "Self-stabilized semiconductor lasers", "trim_reasons": ["leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise"], "year": 1986}
{"archive_ref": "bellsystemtechni01amerrich", "canonical_url": "https://archive.org/details/bellsystemtechni01amerrich", "char_count": 570867, "collection": "archive-org-bell-labs", "doc_id": 65, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc65", "record_count": 687, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bellsystemtechni01amerrich", "split": "test", "text": "Reprinted  with  the  permission  of  the  American  Telephone \nand  Telegraph  Company,  New  York\n\nDevoted  to  the  Scientific  and  Engineering  Aspects \nof  Electrical  Communication\n\nMODERN  industry  is  characterized  by  the  extent  to  which \nscientific  research  and  technique  based  on  precise  study  have \ncontributed  to  its  progress.  So  complete  has  been  the  adaptation \nof  and  reliance  on  scientific  research  in  many  industries  that  it  is \ndifficult  at  this  time  to  visualize  the  state  of  affairs  of  two  or  three \ndecades  ago,  when  substantially  all  industry  on  its  technical  side  was \ndependent  for  advancement  on  cut-and-try,  rule-of-thumb,  methods \nof  development.  Today  in  many  industries  the  management  would \nnot  think  of  embarking  on  a  new  project  without  consulting  their \nresearch   engineers.\n\nMany  industries  have  proved  the  benefits  to  be  derived  from  the \nutilization  of  that  organized  knowledge  provided  both  in  the  fields \nof  the  physical  sciences  and  in  those  newer  fields  which  have  to  do \nwith  psychology  and  economics.  There  are  still  greater  numbers  of \nindustrial  organizations  where  the  adoption  of  scientific  methods  has \nbeen  slow.  However,  the  time  will  undoubtedly  come  when  every  in- \ndustry will  recognize  the  aid  it  can  derive  from  scientific  research \nin  some  form  as  it  now  recognizes  its  dependence  for  motive  power \non  steam  or  electricity  rather  than  on  muscular  activity.\n\nUpwards  of  one  hundred  years  ago  there  was  adopted  in  earnest \nby  scientific  men,  principally  in  university  laboratories,  the  program \nof  searching  deeper  into  the  unknown,  to  discover  new  principles  and \nnew  relationships  of  a  kind  which  had  at  the  time  very  little  apparent \npractical  interest  to  mankind  as  a  whole.\n\nOut  of  this  work,  and  in  time,  have  grown  entirely  new  industries. \nFrom  the  fact  that  these  industries  sprang  directly  from  the  research \nlaboratory,  it  was  inevitable  that  they  should  be  conspicuous  because \nof  the  number  of  their  men  trained  in  the  methods  of  scientific  re- \nsearch. Equally  inevitable  was  it  that  these  new  fields  of  endeavor, \noriginating  as  they  did  and  being  staffed  as  they  were,  should  be  the \nground  where  industrial  research  would  find  its  first  and  largest \ndevelopment.  And  not  the  least  of  the  advantages  which  obtained \nin  these  newer  industries  was  the  absence  of  age-long  traditions  tend- \ning to  ultra-conservatism  as  to  new  undertakings,  and  more  par- \nticularly as  to  the  employment  of  the  new  types  of  mind.\n\nThe  results  up  to  the  present  indicate  clearly  that  the  electrical \nand  chem.ical  fields  in  industry  as  we  know  them  today,  are  the  places \nwhere  the  greatest  advances  have  been  made  in  the  utilization  of \nresearch  methods  and  research  men.  Other,  older  and  more  basic \nindustries  are  rapidly  following  the  general  path  marked  out  by  the \nsuccesses  already  obtained  in  these  fields.  Hence,  it  is  expected  that \nshortly  all  industrial  activities  will  be  based  on  the  results  obtained \nby  trained  investigators,  using  the  tools  of  modern  scientific  in- \nvestigation.\n\nJust  as  applied  electricity  is  a  leading  exemplar  of  the  benefits  to \nbe  obtained  by  an  intelligent  use  of  scientific  knowledge,  so  electrical \ncommunication  of  intelligence  is  a  leading  exemplar  in  the  field  of \napplied  electricity.  This  branch  of  applied  electricity  is  a  pioneer \namong  those  recognizing  the  practical  value  of  scientific  research. \nIt  is  interesting  to  note  that  electrical  communication  is  credited \nwith  having  organized  a  research  laboratory  prior  to  the  first  university \ncourse    in   electrical   engineering.\n\nMore  than  ever  before,  the  communication  engineer  must  seek \nexact  solutions  of  his  problems.  If  his  results  do  not  always  attain \nthe  certainty  he  desires,  the  reason  is  the  absence  of  complete  knowl- \nedge with  regard  to  one  or  more  essential  facts.  But  true  knowledge \nof  what  things  limit  the  solution  of  a  problem  is  frequently  more \nthan  half  the  battle  of  obtaining  the  missing  facts.  Sometimes  these \nunknown  facts  can  be  obtained  by  a  search  through  the  remoter \nparts  of  the  vast  scientific  storehouses  which  have  been  built  in  times \npast.     Frequently,  however,  the  search  discloses  the  entire  absence\n\nof  the  thing  sought  for,  and  new  researches  are  begun  with  definite \nends  in  \\'\\e\\\\.  Thus  it  has  come  about  that  the  communication \nengineer  has  become  an  original  investigator  and  is  extending  the \nboundaries  of  human  knowledge  and  supplementing  the  advances  of \npure  science  to  find  solutions  for  his  various  and  sundry  problems.\n\nHence,  while  well  equipped  physical  and  chemical  laboratories  are \nstill  a  necessary  part  of  the  communication  engineer's  equipment,  he \nis  equally  active  in  pushing  his  investigations  in  many  other  direc- \ntions. Questions  involved  in  the  making  of  proper  rate  schedules  and \nadequate  fundamental  plans  for  new  construction  are  originating \nprofound  researches  in  such  fields  as  political  science,  psychology  and \nmathematics.  A  casual  examination  of  recent  technical  literature \ndealing  with  electrical  communication  would  show  articles  which \ntouch  upon  almost  every  branch  of  human  activity,  which  we  designate \nas  science.\n\nWith  this  intense  and  growing  interest  in  the  proper  application  of \nscientific  methods  to  the  solution  of  the  problems  of  electrical  com- \nmunication, it  is  natural  that  a  widespread  desire  should  have  arisen \nfor  a  technical  journal  to  collect,  print  or  reprint,  and  make  readily \navailable  the  more  important  articles  relating  to  the  field  of  the  com- \nmunication engineer.  These  articles  are  now  appearing  in  some \nfifteen  or  twenty  periodicals  scattered  throughout  the  world  and  in \nthe  majority  of  instances  receive  their  first  and  last  printing  in  these \nwidely  separated  mediums.  The  need  already  felt  for  such  a  journal \nwill  grow  keener  as  new  developments  extend  the  scope  of  the  art  and \nthe  specialization  of  its  engineers  of  necessity  increases.  It  is  hoped \nthat  the  Bell  System  Technical  Journal  will  fill  this  need,  and \nas  implied  above,  it  is  intended  that  the  range  of  subjects  treated  in \nthe  Journal  will  be  as  broad  as  the  science  and  technique  of  electrical \ncommunication  itself.\n\nWhile  many  of  the  articles  which  will  appear  in  the  Journal  will \nbe  original  presentations  of  some  phase  of  the  research  or  develop- \nment or  other  technical  work  of  the  Bell  System,  it  is  not  intended \nthat  the  Journal  should  be  the  sole  means  by  which  this  work  is \npresented.  Just  as  in  the  past,  original  articles  and  papers  will  con- \ntinue to  be  presented  before  various  societies  and  in  different  technical \nand  non-technical  magazines.  Moreover,  the  Journal  will  reprint \narticles  on  important  research  and  development  work  in  the  communi- \ncation field  generally  so  that  the  results  of  such  work  may  be  given \ngreater  publicity  and  become  of  greater  value  to  communication \nengineers.\n\nSynopsis:  The  type  of  vacuum  tube  described  in  the  present  article  is \nlikely  to  become  one  of  the  most  remarkable  devices  of  modern  electrical \nscience.  Vacuum  tubes  capable  of  handling  small  amounts  of  power  have \nbeen  extensively  used  during  the  past  few  years  as  telephone  repeaters \nand  as  oscillators,  modulators,  detectors  and  amplifiers  in  radio  trans- \nmission and  other  fields.  Practically  all  such  tubes  have  depended  upon \nthermal  radiation  from  the  plates  to  dissipate  the  electrical  energy  which  in\n\nthe  device  necessarily  absorbs  during  its  operation.     With  present  methods  i^^\n\nof  construction,  and  using  glass  for  the  containing  V^ulb,  a  fairly  definite \nupper  limit  can  be  set  for  the  power  which  a  radiation  cooled  tube  can \nhandle;  as  the  author  points  out,  this  limit  gives  a  tube  capable  of  delivering \nabout  1  to  2  k.  w.  when  used  as  an  oscillator.\n\nbeen  made  possible  by  a  new  and  striking  development  in  the  art  of  seal- \nin?  metal  to  glass.  In  the  case  of  the  100  k.  w.  tube  the  seal  between  the \nc>lindrical  copper  anode  and  glass  portion  is  3.5  inches  in  diameter.\n\nThe  remarkable  character  of  these  copper-in-glass  seals  is  evidenced \nby  the  fact  that  they  do  not  depend  upon  a  substantial  equality  between \nthe  coefficient  of  expansion  of  the  metal  and  glass.  To  Mr.  W.  G.  Hous- \nkeeper  of  the  Bell  System  Research  Laboratory  at  the  Western  Electric \nCompany,  goes  the  credit  for  developing  the  copper-in-glass  seals.  As \nthe  article  brings  out,  Mr.  Houskeeper  has  also  invented  means  for  sealing \nheavy  copper  wire  and  strip  through  glass  in  such  a  way  that  the  best \nvacua  can  be  maintained  under  wide  changes  of  temperature. \u2014 pAlilor.\n\nTHE  development  of  wireless  telephony  and  the  use  of  continu- \nous wave  transmission  in  wireless  telegraphy  have  led  to  the \ngeneral  adoption  of  the  vacuum  tube  as  the  generator  of  high  fre- \nquency currents  in  low  power  installations.\n\nThe  ordinary  form  of  vacuum  tube  is,  however,  ill  suited  for  the \nhandling  of  large  amounts  of  power,  and  at  the  large  wireless  stations \nwhere  the  plant  is  rated  in  hundreds  of  kilowatts  either  the  arc  or \nthe  high  frequency  alternator  is  used.\n\nThe  undoubted  advantages  to  be  derived  from  the  use  of  vacuum \ntubes,  especially  in  the  field  of  wireless  telephony  where  the  output \npower  must  be  modulated  to  conform  to  the  intricate  vibration  pat- \ntern of  the  voice,  has  led  to  a  demand  for  tubes  capable  of  handling \namounts  of  power  comparable  with  those  in  use  at  the  largest  stations.\n\nThat  the  development  of  such  tubes  was  of  great  importance  was \nrecognized  by  the  engineers  of  the  Bell  Telephone  System  in  the \nearly  days  of  the  vacuum  tube  art.     The  experiments  at  Arlington,\n\nVirginia,  in  which  speech  was  first  transmitted  across  the  Atlantic \nto  Paris  and  across  the  Pacific  to  Honolulu,  required  the  use  of  nearly \n300  of  the  most  powerful  tubes  then  available,  each  capable  of  han- \ndling about  25  watts,  and  the  difficulties  encountered  in  operating  so \nmany  tubes  in  parallel  gave  added  impetus  to  the  development  of \nhigh  power  units.\n\nIt  is  the  object  of  the  present  paper  to  deal  with  the  various  steps \nin  the  development  of  high  power  tubes  as  carried  out  in  the  Bell \nSystem  research  laboratories  at  the  Western  Electric  Company.\n\nThe  usual  type  of  vacuum  tube  consists  of  an  evacuated  glass  vessel \nin  which  are  enclosed  three  elements,  the  filament,  the  plate,  and  the \ngrid.  When  the  tube  is  in  operation  an  electron  current  flows  between \nthe  filament  which  is  heated  by  an  auxiliary  source  of  power  and  the \nplate,  the  magnitude  of  this  current  being  controlled  by  the  grid.\n\nThe  passage  of  the  current  through  a  thermionic  tube  is  accompanied \nby  the  dissipation  in  the  plate  of  an  amount  of  power  which  is  compa- \nrable to  the  power  delivered  to  the  output  circuit  and  wrhich  manifests \nitself  in  the  form  of  heat.  This  causes  the  temperature  of  the  plate \nin  the  usual  type  of  tube  to  rise  until  the  rate  of  loss  of  heat  by  radia- \ntion is  equal  to  the  power  dissipated.  Some  of  the  heat  liberated  by \nthe  plate  is  absorbed  by  the  walls  of  the  containing  vessel  which  con- \nsequently rise  in  temperature.  These  factors,  together  with  a  con- \nsideration of  the  size  of  plate  that  can  be  conveniently  suspended  inside \na  glass  bulb  and  the  size  of  glass  bulb  that  can  be  conveniently  worked, \nset  a  limit  of  about  1  to  2  k.  w^  for  the  power  that  can  be  dissipated \nin  the  plate  of  a  commercial  vacuum  tube  of  this  type.  The  plates \nare  generally  constructed  of  molybdenum  or  some  other  refractory \nmetal  and  the  containing  vessel  made  of  hard  glass.\n\nThe  use  of  quartz  as  the  containing  vessel  offers  certain  advantages \nwhich  tend  to  raise  the  power  limit  somewhat  and  this  material  has \nbeen  used  for  power  tube  purposes  in  England.\n\nIt  is  apparent  then  that  in  the  development  of  vacuum  tubes  capa- \nble of  handling  large  amounts  of  power  means  other  than  radiation \nmust  be  used  for  removing  the  heat  dissipated  at  the  plate,  and \ndevelopment  of  tubes  along  these  lines  was  undertaken  by  Dr.  E.  R. \nStoekle  and  Dr.  O.  E.  Buckley.\n\nDr.  Stoekle  had  already  worked  for  some  years  on  the  problem  of \nremoving  the  heat  dissipated  at  the  anode  of  a  thermionic  tube  by \nmaking  the  anode  a  part  of  the  outside  wall  of  the  vessel  and  thus \nmaking  it  possible  to  convey  the  heat  directly  away  from  it  by  means \nof  circulating  water.  This  was  clearly  the  right  principle  but  as  is \nobvious  to  those  who  are  familiar  with  these  devices,  great  difficulties\n\npresented  themselves  in  the  mechanical  construction  of  large  tubes \nin  which  vacuum  tight  joints  must  be  made  and  maintained  between \nglass  and  large  masses  of  metal.  The  importance  of  the  problem, \nhowever,  was  such  that  Stoekle  and  Buckley  pushed  on  in  the  face  of \ndifficulties  to  the  construction  of  tubes  which  could  handle  kilowatts \nwhere  previous  tubes  could  only  handle  watts.\n\nA  step  in  the  direction  of  overcoming  these  difficulties  was  made  by \nMessrs.  Schwerin  and  Weinhart,  who  were  working  with  Dr.  Buckley \non  the  problem,  and  who  suggested  that  the  anode  might  be  made \nin  the  form  of  a  tube  or  thimble  of  platinum  sealed  into  a  glass  vessel \nand  kept  cool  by  passing  water  through  it.\n\nThis  suggestion  led  to  the  development  of  a  tube  which,  although \nnot  the  one  finally  adopted,  is  discussed  in  some  detail  since  it  was  the \nfirst  one  to  be  pushed  to  such  a  point  as  to  give  promise  of  economical \ncommefcial  manufacture.\n\nThe  tube  is  shown  in  Fig.  1.  The  anode  consists  of  a  platinum \ncylinder  A,  7\"  long  and  .625\"  wide,  which  is  scaled  into  the  center  of \nthe  glass  cylinder  B.  The  end  of  the  platinum  cylinder  remote  from \nthe  seal  is  closed.  The  anode  is  surrounded  by  the  grid  C  and  by \nthe  filament  D,  which  are  supported  by  the  glass  arbors  E.  The \ncurrent  for  the  filament  is  led  into  the  tube  through  the  platinum \nthimbles  F.\n\nThe  anode  is  kept  cool  by  means  of  a  supply  of  water  passing  into \nthe  anode  through  the  tube  G  and  leaving  by  the  tube  H.\n\nA  number  of  tubes  having  this  general  type  of  construction  were \nmade  up  and  it  was  found  possible  to  dissipate  as  much  as  15  k.  w. \nin  the  anode.\n\nAs  soon  as  the  pressure  of  work  more  directly  connected  with  the \nnecessities  of  the  war  would  permit,  Mr.  W.  G.  Houskeeper  and  Dr. \nM.  J.  Kelly  undertook  the  further  improvement  of  the  water-cooled \ntube,  the  former  assuming  the  task  of  developing  the  mechanical \nstructure,  and  the  latter  that  of  determining  the  electrical  design  and \nthe  process  of  tube  exhaust.\n\nMr.  Houskeeper  adopted  into  the  construction  of  the  tube  a  remark- \nable type  of  vacuum  seal  which  he  had  previously  developed.  These \nseals  are  made  between  glass  and  metal  and  can  be  made  in  any  desired \nsize.  They  are  capable  of  withstanding  repeated  heating  and  cooling \nover  wide  ranges  of  temperature,  from  that  of  liquid  air  to  350\u00b0  C, \nwithout  cracking  and  without  impairment  of  their  vacuum  holding \nproperties.\n\nIt  is  no  exaggeration  to  say  that  the  invention  of  these  seals  has \nmade  possible  the  construction  of  vacuum  tubes,  capable  of  handling \nin  single  units,  powers  of  any  magnitude  which  may  be  called  for  in \nwireless  telegraph  and  telephone  transmission.\n\nThe  underlying  principle  connected  with  the  making  of  this  seal \nconsists  in  obtaining  an  intimate  connection  between  the  glass  and \nmetal,  either  by  chemical  combination  or  by  mere  wetting,  and  in  so \nproportioning  the  glass  and  metal  portions  of  the  seal  that  the  stresses \nproduced  when  the  seal  is  heated  or  cooled  will  not  be  great  enough  to \nrupture  either  the  glass  or  the  junction  between  the  glass  and  metal.\n\nThe  three  principal  types  of  seals  developed  by  Mr.  Houskeeper \nare  known  as  the  ribbon  seal,  the  disc  seal  and  the  tube  seal.\n\nIf  a  copper  ribbon  is  directly  sealed  through  glass  it  is  found  that \nthe  glass  and  copper  adhere  along  the  flat  faces  of  the  seal  but  that \nruptures  occur  along  the  edges  as  shown  in  Fig.  2  (a).  This  is  due  to \nthe  fact  that  as  the  seal  cools  after  being  made,  the  glass  in  contact \nwith  metal  is  capable  of  resisting  the  shearing  and  tensile  stresses\n\nthat  occur  along  the  faces,  while  the  glass  wrapping  round  the  edges \nof  the  ribbon  is  called  upon  to  withstand  much  greater  tensile  stresses \nand  gives  way.  If  the  edges  of  the  ribbon  are  sharpened  as  shown  in \nFig.  2  (b),  a  tight  seal  results,  the  reason  being  that  the  forces  of\n\nadhesion  between  the  glass  and  copper  acting  along  the  flat  contact \nfaces  are  sufficient  to  stretch  the  thin  copper  at  the  edge  and  prevent \nits  drawing  away  when  cooled.  There  is  a  definite  relation  between \nthe  elastic  properties  of  the  metal  and  glass  and  the  angle  of  edge \nthat  can  be  used  for  a  successful  seal.\n\nBy  proper  shaping  of  the  metal  ribbon,  seals  have  been  successfully \nmade  up  to  very  large  sizes.  Some  of  these  are  shown  in  Fig.  3,  the \nthe  largest  in  the  photograph  being  about  1  in  width,  and  capable \nof  successfully  conducting  a  current  of  150  to  200  amperes.\n\nThe  principles  involved  in  the  making  of  the  disc  seal  are  the  same \nas  those  involved  in  making  the  ribbon  seal.  If  a  metal  disc  is  sealed \nwholly  into  glass  the  edges  must  be  sharpened  or  the  glass  and  copper \nbreak  away  from  each  other  as  in  the  case  of  the  ribbon  seal.\n\nIn  the  general  use  to  which  these  seals  are  put  there  is  no  necessity \nfor  having  the  glass  surround  the  circumference  of  the  copper  disc\n\nand  the  necessity  for  sharpening  the  edge  is  obviated  by  allowing  the \nglass  to  adhere  to  the  flat  portion  of  the  disc  only,  care  being  taken  to \nprevent  its  flowing  around  the  edge.  It  is  necessary  to  have  a  ring \nof  glass  on  both  sides  of  the  seal  in  order  to  equalize  the  bending  stresses\n\nwhich  would  otherwise  tend  to  break  the  glass  and  copper  away  from \neach  other.  Successful  disc  seals  have  been  made  with  copper  up  to \n1-10\"  thick.  There  is,  of  course,  a  certain  maximum  thickness  that \ncan  be  used  for  a  seal  of  a  given  diameter  and  it  is  preferable  to  keep \nwell  below  this  limit.\n\nThe  seals  shown  in  Fig.  4  close  the  ends  of  glass  tubes  to  the  other \nends  of  which  are  sealed  pilot  lamps  for  the  purpose  of  testing  the \nvacuum.  Tubes  sealed  in  this  way  have  been  kept  a  number  of \nyears  without  any  impairment  of  the  vacuum.\n\nThe  third  type  of  seal  and  the  most  important  in  connection  with \nthe  present  problem  is  the  tube  seal  shown  in  Fig.  5.  This  furnishes \nthe  means  of  joining  metal  and  glass  tubes  end  to  end  and  is  used  in \nthe  water-cooled  tube  to  attach  the  anode  to  the  glass  cylinder  which\n\nserves  to  insulate  the  other  tube  elements.  As  in  the  case  of  the \ndisc  seal,  it  can  be  made  either  with  the  edge  of  the  metal  not  in  con- \ntact with  the  glass,  as  shown  at  A,  or  with  the  metal  sharpened  to  a \nfine  edge  which  is  in  contact  with  the  glass.  The  glass  may  be \nsituated  either  inside  or  outside  of  the  metal,  see  B  and  C.\n\nThe  first  thermionic  tubes  in  which  these  seals  were  embodied  were \nmade  of. copper  and  were  designed  to  operate  at  10,000  volts  and  to \ngive  about  5  k.  w.  output.\n\nA  photograph  of  one  of  these  tubes  is  shown  in  Fig.  6;  and  the \nfilament  grid  assembly  is  shown  in  Fig.  7.\n\nThe  anode  consists  of  a  copper  tube  1.5''  in  diameter  and  7.5  long. \nA  copper  disc  is  wielded  to  one  end  forming  a  vacuum-tight  joint. \nThe  other  end  which  is  turned  down  to  a  knife  edge  is  fused  directly \nto  a  glass  tube.\n\nThe  filament  grid  assembly  consists  of  two  lavite  discs  D  and  E, \nspaced  b\"  apart  by  a  seamless  steel  tube.  The  grid  F  is  made  in  the \nform  of  a  helix,  and  is  held  in  position  by  allowing  the  ends  of  the  lon- \ngitudinal wires,  to  which  the  turns  of  the  helix  are  welded,  to  pass \nthrough  holes  in  the  lavite  blocks  D  and  E.  The  filament  G  is \nmounted  between  hooks  fastened  to  the  lavite  blocks  and  is  kept \ntaut  by  the  springs  H.  The  grid  lead  is  shown  at  J,  and  the  filament \nleads  at  K  K.  In  this  tube  platinum  seals  are  used  for  the  lead  wires. \nThe  use  of  the  springs  H  make  it  necessary  to  supply  the  filament \nwith  current  from  the  opposite  end  of  the  assembly  and  this  is  done \nby  passing  the  current  through  the  steel  support  tube  and  returning \nit  through  a  lead  passing  through  this  tube  and  insulated  from  it  by \na  quartz  tube.\n\nThe  whole  assembly  is  carried  by  two  supports  B  B.  These  sup- \nports are  welded  to  a  corrugated  nickel  collar  A  which  grips  the  glass \nstem  C.\n\nThe  pumping  of  these  tubes  at  first  presented  considerable  difficulty, \nchiefly  on  account  of  the  large  amount  of  occluded  gas  contained  by \nthe  metal  parts.  This  caused  the  time  of  pumping  of  the  tube  to  be \nvery  long  and  a  dangerous  warping  of  the  internal  structure  developed \nowing  to  the  fact  that  during  exhaust  the  tube  elements  are  maintained \nat  a  much  higher  temperature  than  they  are  subjected  to  during  nor- \nmal operation.  The  trouble  was  overcome  by  heating  the  various \nparts  of  the  tube  to  as  high  a  temperature  as  possible  in  a  vacuum \nfurnace,  prior  to  the  final  assembly,  and  thus  getting  rid  of  a  large \namount  of  the  occluded  gases.  The  anode  was  preheated  before \nthe  glass  seal  was  made  and  the  whole  filament  grid  assembly  was  pre- \nheated just  before  it  was  mounted  on  the  glass  stem.  The  preheating \nof  the  parts  brought  about  an  enormous  reduction  in  the  time  re- \nquired for  pumping  and  gave  a  much  more  uniform  product.\n\nAlthough  successful  from  the  standpoint  of  operation,  this  tube \nhad  several  undesirable  features  that  it  was  thought  well  to  eliminate. \nIn  the  first  place  the  welding  of  the  end  into  the  tube  was  not  particu- \nlarly desirable,  and  in  general  any  troubles  that  occurred  due  to  leaks \nin  the  metal  could  be  traced  to  this  point.  Further,  in  the  assembly \nof  the  tube  there  were  a  very  large  number  of  welds  to  be  made \nwhich  constituted  points  of  weakness  at  the  high  temperature  neces- \nsary for  the  evacuation  of  the  tubes.  It  was,  therefore,  decided  to \ngo  to  a  type  of  tube  in  which  the  anode  would  be  drawn  in  one  piece \nand  in  which  as  many  welds  as  possible  would  be  eliminated  in  the \nassembly  of  the  internal  elements.  At  the  same  time  it  was  considered \ndesirable  to  go  to  a  somewhat  larger  type  of  structure  in  which  high\n\ntension  insulation  could  be  more  easily  provided  and  a  larger  tube \nwas,  therefore,  designed  capable  of  delivering  10  k.  w.  to  an  antenna \nat  a  plate  voltage  of  10,000  volts.\n\nThe  anode  A  is  drawn  from  a  piece  of  sheet  copper  and  is  9\"  long \nand  2\"  in  diameter.  The  copper  flare  B  is  turned  down  to  a  sharp \nedge  and  a  glass  bulb  C  sealed  thereto.  The  grid  and  plate  assembly \nis  shown  at  D.  The  structure  is  supported  by  four  molybdenum \nrods,  which  are  threaded  and  secured  by  means  of  nuts  to  the  lavite \npieces  E  and  F.  The  filament  is  made  of  19.5\"  of  .025  pure  tungsten \nwire  purchased  from  the  General  Electric  Company  and  is  formed \nand  secured  to  two  of  the  molybdenum  rods  at  G  and   H.     The\n\npower  consumed  in  it  during  operation  is  .75  k.  w.  It  is  guided  by \nthe  hooks  J.  The  filament  leads  are  shown  at  K,  K  and  are  led \nthrough  the  glass  by  the  copper  disc  seals  L,  L.     The  grid  is  a  molyb-\n\ndenum  helix  and  is  supported  by  the  molybdenum  rods  M  which  are \nfixed  to  the  lavite  block  E  and  slide  on  the  outside  of  the  lavite  block \nF.     The  whole  structure  is  mounted  on  the  flare  R  by  means  of  the\n\nnickel  collar  N  and  the  support  rods  P.  The  grid  lead  is  brought \nout  through  the  tube  Q.  The  tube  is  completed  by  sealing  together \nthe  flare  R  and  the  bulb  C.\n\nIn  this  tube  all  welds  except  those  in  the  collar  N  are  eliminated, \nthe  assembly  being  bolted  together.     The  drawing  of  the  anode  does\n\naway  with  the  leaks  that  were  troublesome  in  the  older  tubes  and \nthe  manufacture  of  the  tube  can  be  carried  out  with  certainty.\n\nWith  this  tube  as  much  as  12  k.  w.  have  been  obtained  in  an  arti- \nficial antenna  working  at  12,000  volts.  This  power  was  obtained \nat  a  frequency  of  600,000  cycles  corresponding  to  500  meters  wave \nlength.  The  difficulties  of  obtaining  this  amount  of  power  at  this \nfrequency  using  a  number  of  smaller  tubes  in  parallel,  are  obvious \nto  anyone  who  is  acquainted  with  the  problem.  On  a  D.  C.  test  the \nanode  w^as  found  to  be  capable  of  dissipating  26  k.  w\\  when  cooled \nwith  water.\n\nThe  success  which  had  attended  the  development  of  a  tube  of  this \nhigh  power  capacity  indicated  the  possibility  of  constructing  still \nlarger  tubes  and  it  was  decided  to  proceed  with  the  development  of  a \ntube  capable  of  delivering  at  least  100  k.  w.  into  an  antenna.\n\nThe  development  proceeded  with  a  few  minor  alterations  along  the \nlines  of  the  smaller  tube,  nominally  rated  at  10  k.  w.  and  the  100  k.  w.\n\ntube  as  now  developed  is  shown  in  Figs.  10  and  11.  The  anode  which \nis  made  of  a  piece  of  seamless  copper  tubing  closed  by  a  copper  disc \nwelded  into  the  end,  is  14\"  long  and  3.5\"  in  diameter.  The  filament  is \nof  tungsten  and  is  .060\"  in  diameter  and  63.5\"  long.  The  current \nrequired  to  heat  it  is  91  amperes  and  the  power  consumed  in  it  6  k.  w.\n\nThe  filament  leads  are  of  copper  red  one  eighth  of  an  inch  in  diameter \nand  are  sealed  through  1\"  copper  disc  seals.  The  grid  is  of  molyb- \ndenum and  is  wound  around  three  molybdenum  supports.\n\nThe  handling  of  the  parts  of  this  tube  during  manufacture  presents \na  task  of  no  mean  magnitude  and  numerous  fixtures  have  been  devised \nto  assist  in  the  glass  working.  It  has  been  found  necessary  for  in- \nstance to  suspend  the  anode  in  gimbals  during  the  making  of  the  tube \nseal  owing  to  its  great  weight,  and  special  devices  have  been  made \nto  hold  the  filament  grid  assembly  in  place  while  it  is  being  sealed  in, \notherwise  the  strains  produced  by  its  weight  cause  cracking  of  the  seal.\n\nThe  significance  of  this  development  in  the  radio  art  cannot  be \noverestimated.  It  makes  available  tubes  in  units  so  large  that  only \na  very  few  would  be  necessary  to  operate  even  the  largest  radio  sta- \ntions now  extant,  with  all  the  attendant  flexibility  of  operation  which \naccompanies  the  use  of  the  vacuum  tube.\n\nFrom  the  standpoint  of  wireless  telephony  the  development  of  these \nhigh  power  tubes  gives  us  the  possibility  of  using  very  much  greater \namounts  of  power  than  have  ever  been  readily  available  before.  The \nfilaments  in  these  tubes  have  been  made  so  large  that  the  electron \nemission  from  them  will  easily  take  care  of  the  high  peak  currents  ac- \ncompanying the  transmission  of  modulated  power.\n\nThe  100  k.  w.  tube  by  no  means  represents  the  largest  tube  made \npossible  by  the  present  development.  There  is  no  doubt  that  if  the \ndemand  should  occur  for  tubes  capable  of  handling  much  larger \namounts  of  power  they  could  be  constructed  along  these  same  lines.\n\nSynopsis:  Direct  capacity,  direct  admittance  and  direct  impedance  are \ndefined  as  the  branch  constants  of  the  particular  direct  network  which  is \nequivalent  to  any  given  electrical  system.  Typical  methods  of  measuring \nthese  direct  constants  are  described  with  especial  reference  to  direct  ad- \nmittance; the  substitution  alternating  current  bridge  method,  due  to \nColpitts,  is  the  preferred  method,  and  for  this  suitable  variable  capacities \nand  conductances  are  described,  and  shielding  is  recommended.  Propo.sed \nmethods  are  also  described  involving  the  introduction  of  electron  tubes \ninto  the  measuring  set,  which  will  reduce  the  measurement  to  a  single  setting \nor  deflection.  This  gives  an  alternating  current  method  which  is  com- \nparable with  Maxwell's  single  null-setting  cyclical  charge  and  discharge \nmethod.  Special  attention  is  drawn  to  Maxwell's  remarkable  method \nwhich  is  entirely  ignored  by  at  least  most  of  the  modern  text-books  and \nhandbooks.\n\nTHE  object  of  this  paper  is  to  emphasize  the  importance  of  direct \ncapacity  networks;  to  explain  various  methods  of  measuring \ndirect  capacities;  and  to  advocate  the  use  of  the  Colpitts  substitution \nmethod  which  has  been  found  preeminently  satisfactory  under  the \nwide  range  of  conditions  arising  in  the  communication  field.\n\nAbout  thirty  years  ago  telephone  engineers  substituted  the  so- \ncalled  \"  mutual  capacity  \"  measurement  for  the  established  \"  grounded \ncapacity  \"  measurement;  this  was  a  distinct  advance,  since  the  trans- \nmission efficiency  is  more  closely  connected  with  mutual  capacity \nthan  with  grounded  capacity.  Mutual  capacity,  however,  can  give \nno  information  respecting  crosstalk,  and  accordingly,  about  twenty \nyears  ago,  I  introduced  the  measurement  of  \"  direct  capacity  \" \nwhich  enabled  us  to  control  crosstalk  and  to  determine  more  com- \npletely how  telephone  circuits  will  behave  under  all  possible  con- \nnections.\n\nFor  making  these  direct  capacity  measurements  alternating  cur- \nrents of  telephone  frequencies  were  introduced  so  as  to  determine \nmore  exactly  the  effective  value  of  the  capacity  in  telephonic  trans- \nmission, and  to  include  the  determination  of  the  associated  effective \ndirect  conductances  which  immediately  assumed  great  importance \nupon  the  introduction  of  loading.\n\nTelephone  cables  and  other  parts  of  the  telephone  plant  present \nthe  problem  of  measuring  capacities  which  are  quite  impossible  to \nisolate,  but  which  must  be  measured,  just  as  they  occur,  in  associa- \ntion with  other  capacities;  and  these  associated  capacities  may  be \nmuch   larger  than   the  particular  direct  capacity  which  it  is  neces-\n\n1  This  article  is  also  appearing  in  the  August  issue  of  the  Journal  of  the  Optical \nSociety  of  America  and  Review  of  Scientific  Instruments.  An  appendix  is  added  here \ngiving  proofs  of  the  mathematical  results.\n\nsary  to  accurately  measure,  and  have  admittances  overwhelmingly \nlarger  than  the  direct  conductance,  which  is  often  the  most  important \nquantity.  This  is  the  interesting  problem  of  direct  capacity  measure- \nment, and  distinguishes  it  from  ordinary  capacity  measurements \nwhere  isolation  of  the  capacity  is  secured,  or  at  least  assumed.\n\nThe  substitution  alternating  current  bridge  method,  suggested  to \nme  in  1902  by  Mr.  E.  H.  Colpitts  as  a  modification  of  the  potentiometer \nmethod,  has  been  in  general  use  by  us  ever  since  in  all  cases  where \naccuracy  and  ease  of  manipulation  are  essential.\n\nAfter  first  defining  direct  capacities  and  describing  various  methods \nfor  measuring  them,  this  paper  will  explain  how  this  may  all  be  general- \nized so  as  to  include  both  the  capacity  and  conductance  components \nof  direct  admittances,  and  the  inductance  and  resistance  components \nof  direct  impedances.\n\nIt  is  a  familiar  fact  that  two  condensers  of  capacities  Ci,  Ci,  when \nin  parallel  or  in  series,  are  equivalent  to  a  single  capacity  (Ci  +  d) \nor  Ci  C-i.l{C\\  -f  C2),  respectively,  directly  connecting  the  two  terminals. \nThese  equivalent  capacities  it  is  proposed  to  call  direct  capacities. \nThe  rules  for  determining  them  may  be  stated  in  a  form  having  general \napplicability,  as  follows:\n\nRule  I.  The  direct  capacity  which  is  equivalent  to  capacities  in \nparallel  is  equal  to  their  sum.\n\nRule  2.  The  direct  capacity  between  two  terminals,  which  is \nequivalent  to  two  capacities  connecting  these  terminals  to  a  con- \ncealed branch-point,  is  equal  to  the  product  of  the  two  capacities \ndivided  by  the  total  capacity  terminating  at  the  concealed  branch- \npoint, i.e.,  its  grounded  capacity.\n\nThese  rules  may  be  used  to  determine  the  direct  capacities  of  any \nnetwork  of  condensers,  with  any  number  of  accessible  terminals  and \nany  number  of  concealed  branch-points.  Thus,  all  concealed  branch- \npoints may  be  initially  considered  to  be  accessible,  and  they  are  then \neliminated  one  after  another  by  applying  these  two  rules;  the  final \nresult  is  independent  of  the  order  in  which  the  points  are  taken;  all \nmay,  in  fact,  be  eliminated  simultaneously  by  means  of  determinants^; \na  network  of  capacities,  directly  connecting  the  accessible  terminals, \nwithout  concealed  branch-points  or  capacities  in  parallel,  is  the  final \nresult.  Fig.  1  shows  the  two  elementary  cases  of  direct  capacities \nand  also,  as  an  illustration  of  a  more  complicated  system,  the  bridge\n\ncircuit,  with  three  corners  1,  2,  3  assumed  to  be  accessible,  and  the \nfourth  inaccessible,  or  concealed.  Generalizing,  we  have  the  following \ndefinition:\n\nThe  direct  capacities  of  an  electrical  system  with  n  given  accessible \nterminals  are  defined  as  the  n{n  \u2014  i)  /2  capacities  which,  connected \nbetween  each  pair  of  terminals,  will  be  the  exact  equivalent  of  the  system, \nin  its  external  reaction  upon  any  other  electrical  system  with  which  it  is \nassociated  only  by  conductive  connections  through  the  accessible  terminals.\n\nThe  total  direct  capacity  between  any  group  of  the  terminals  and  all \nof  the  remaining  accessible  terminals,  connected  together,  is  called \nthe  grounded  capacity  of  the  group.\n\nThis  definition  of  direct  capacity  presents  the  complete  set  of  direct \ncapacities  as  constituting  an  exact,  symmetrical,  realizable  physical \nsubstitute  for  the  given  electrical  system  for  all  purposes,  including \npractical  applications.  Direct  capacities  are  Maxwell's  \"  coefificients \nof  mutual  induction,\"  but  with  the  sign  reversed,  their  number  being \nincreased  so  as  to  include  a  direct  capacity  between  each  pair  of \nterminals.\n\nIn  considering  direct  capacities  we  exclude  any  direct  coupling, \neither  magnetic  or  electric,  from  without  with  the  interior  of  the \nelectrical  system,  since  we  have  no  concern  with  its  internal  structure; \nwe  are  restricted  to  its  accessible,  peripheral  points  or  terminals; \nsome  care  has  been  taken  to  emphasize  this  in  the  wording  of  the \ndefinition.\n\nConnecting  a  capacity  between  two  terminals  adds  that  capacity \nto  the  direct  capacity  between  these  terminals,  and  leaves  all  other \ndirect  capacities  unchanged.  Connecting  the  terminals  of  two  distinct \nelectrical  systems,  in  pairs,  gives  a  system  in  which  each  direct  capacity \nis  the  sum  of  the  corresponding  two  direct  capacities  in  the  individual \nsystems.  Joining  two  terminals  of  a  single  electrical  system  to  form \na  single  terminal  adds  together  the  two  direct  capacities  from  the  two \nmerged  terminals  to  any  third  terminal,  and  leaves  all  other  direct \ncapacities  unchanged,  with  the  exception  of  the  direct  capacity \nbetween  the  two  merged  terminals,  which  becomes  a  short  circuit. \nCombining  the  terminals  into  any  number  of  merged  groups  leaves \nthe  total  direct  capacity  between  any  pair  of  groups  unchanged,  and \nshort-circuits  all  direct  capacities  within  each  group.\n\nThese  several  statements  of  the  additive  property  of  direct  capaci- \nties show  the  simple  manner  in  which  direct  capacities  are  altered \nunder  some  of  the  most  important  external  operations  which  can  be \nmade  with  an  electrical  network,  and  explain,  in  part,  the  preeminent \nconvenience  of  direct  capacity  networks.\n\nSince  the  additive  property  of  direct  capacities  is  sufficient  for \nexplaining  the  different  methods  of  measuring  direct  capacities  we \nmay  now,  without  further  general  discussion  of  direct  capacities,  pro- \nceed to  the  description  of  the  more  important  methods  of  measurement.\n\nThe  unknown  direct  capacity  is  shifted  from  one  side  of  the  bridge \nto  the  other,  and  the  balance  is  restored  by  adjusting  the  capacity \nstandard  so  as  to  shift  back  an  equal  amount  of  direct  capacity. \nThe  method  is  therefore  a  substitution  method,  and  the  value  of  the \nbridge  ratio  is  not  involved.  Both  the  standard  and  the  unknown \nremain  in  the  bridge  for  both  settings,  so  that  the  method  involves \ntransposition  rather  than  simple,  ordinary  substitution.\n\nDetails  of  the  method  as  shown  by  Fig.  2  are  as  follows :  To  measure \nthe  direct  capacity  Cn  between  terminals  1  and  2  connect  one  terminal \n(1)  to  corner  5*of  the  bridge,  and  adjust  for  a  balance  with  the  other \nterminal  (2)  on  corner  (^  and  then  on  (?,  while  each  and  every  one  of \nthe  remaining  accessible  terminals  (3,  4,  .  .  .  )  of  the  electrical  system \nis  permanently  connected  during  the  two  adjustments  to  either  corner \n<^or  (?.  If  the  direct  capacities  in  the  standard  condenser  between \ncorners  CF  and  S)  are  C,  C\"  in  the  two  balances,\n\nwhere  C^  is  the  standard  condenser  reading  when  the  bridge  alone \nis  balanced.\n\nTwo  settings  are  required  by  this  method  for  an  individual  direct \ncapacity  measurement,  but  in  the  systematic  measurement  of  all  the \ndirect  capacities  in  a  system  the  total  number  of  settings  tends  to \nequal  the  total  number  of  capacities,  when  this  number  becomes  large. \nThe  number  of  settings  may  always  be  kept  equal  to  the  number  of \ncapacities  by  employing  an  equality  bridge  ratio,  and  using  the  ex- \npression for  the  direct  capacity  difference  given  above.  The  same \nremarks  also  hold  for  the  group  of  direct  capacities  connecting  any \none  terminal  with  all  the  other  terminals.\n\nIn  general,  ground  is  placed  upon  corner  C  of  the  bridge,  but  is \ntransferred  to  corner  2),  if  it  is  connected  to  one  terminal  of  the  re- \nquired direct  capacity.  The  arbitrary  distribution  of  the  other \nterminals  between  corners  <^  and  (?  may  be  used  to  somewhat  control \nthe  amount  of  standard  capacity  required;  or  it  may  be  helpful  in \nreducing  interference  from  outside  sources,  when  tests  are  made  upon \nextended  circuits.  The  grounded  capacity  of  a  terminal  or  group  of \nterminals  is  measured  by  connecting  the  group  to  (?,  and  all  of  the \nremaining  terminals  together  to  2).\n\nThe  excess  of  one  direct  capacity  C\\2  over  another  Cm  is  readily \ndetermined  by  connecting  terminals  1  and  5  to  corner  ^,  terminals \n3,  4,  7,  8,  .  .  .  to  corner  ^''or  d*,  and  then  balance  with  terminals  2 \nand  6  on  ^  and  C^,  respectively,  and  repeat,  with  their  connections \nreversed.\n\nThe  required  direct  capacity  C12  is  balanced  against  one  of  its \nassociated  direct  capacities,  augmented  by  a  standard  direct  capacity\n\nC ,  and  the  measurement  is  repeated  with  the  required  direct  capacity \nand  standard  interchanged.  Let  R\\  R\"  be  the  resistances  required \nin  arm  (?^  o[  the  bridge  for  the  first  and  second  balance,  then,  5 \nbeing  the  total  slide  wire  resistance  and  G\\  the  grounded  capacity \nof  terminal  1:*\n\nThis  ratio  method  requires  for  the  bridge  a  variable  or  slide  wire \nresistance  and  a  constant  condenser,  and  it  may  be  employed  as  an \nimprovised  bridge,  when  sufficient  variable  capacity  is  not  available \nfor  the  Colpitts  method.     Not  being  a  substitution  method,  however,\n\ngreater  precautions  are  necessary  for  accurate  results.  There  must \nbe  no  initial  direct  capacity  in  arm  CS),  or  a  correction  will  be  re- \nquired. Possibly  variable  capacity  ratio  arms  would  be  preferable \nto  resistances.\n\nAssuming  that  the  electron  tube  supplies  the  means  of  obtaining \nan  invariable  true  negative  resistance,  Fig.  4  shows  a  method  which \ndetermines  any  individual  direct  capacity  from  a  single  bridge  setting. \nThe  bridge  arms  are  replaced  by  a  Y  network  made  up  of  two  resist-\n\nances  R,  R  and  a  negative  resistance  \u2014  R/2;  the  Y  has  then  a  null- \nimpedance  between  corner  ^  and  corners  ^,  (?  connected  together^. \nThe  three  terminals  1,  2,  3  of  the  network  to  be  measured  are  con- \nnected to  corners  2),  6^,  6B  and  a  balance  obtained  by  adjusting  the \nvariable  standard  condenser  C .  Then  di  =  C  regardless  of  the \ndirect  capacities  associated  with  Cn  and  C\\  since  these  capacities \neither  are  short-circuited  between  corners  SB,  (2  ox  ^,(?  or  are  between \ncorners  58,  S)  and  thus  outside  of  the  bridge.\n\nCorrect  adjustment  of  the  negative  resistance  may  be  checked  by \nobserving  whether  there  is  silence  in  telephone  Ti  after  the  balance \nhas  been  obtained.  Assuming  invariable  negative  resistance,  this \ntest  need  be  made  only  when  the  bridge  is  set  up,  or  there  is  a  change \nin  frequency.  The  bridge  may  be  given  any  ratio  Z1/Z2  by  employing \na  Y  made  up  of  impedances  Zi,  Z2,  and  \u2014  Zi  Z2/(Zi  +  Z2).\n\nConnect  the  terminals  between  which  the  direct  capacity  C12  is \nrequired,  to  A,  B  and  the  remaining  accessible  terminals  of  this \nelectrical  system  to  D.  The  adjustable  standard  capacity  is  C  and \nany  associated  direct  capacities  in  this  standard  are  shown  as  C\",\n\nC\".  If  C\\2  is  a  direct  capacity  to  ground,  interchange  C  and  C12. \nBalancing  involves  the  following  repeated  cycle  of  operations:\n\n1.  Make  connections  1,  2,  3  and  7  for  an  instant  (thus  charging \nC12,  Ci3,  C\",  C  and  discharging  the  electrometer).\n\n2.  Make  connections  4,  5  and  then  6  (to  discharge  condensers  C13, \nC\" ,  mix  charges  of  C12,  C  with  polarities  opposed  and  connect \nelectrometer).\n\n3.  Adjust  C  to  reduce  the  electrometer  deflection  when  the  cycle \nis  again  repeated.\n\nWhen  a  null  deflection  is  obtained  Cn  =  C;  the  required  direct \ncapacity  is  equal  to  the  standard  direct  capacity  irrespective  of  the \nmagnitudes  of  the  four  associated  direct  capacities.  If  all  capacities \nare  free  of  leakage  and  absorption,  this  remarkable  method  accurately \ncompares  two  direct  capacities  by  means  of  a  single  null  setting,  and \nit  requires  the  irreducible  minimum  amount  of  apparatus.\n\nThis  is  defined  as  the  direct  capacity  between  two  given  terminals \nwith  all  other  terminals  left  floating  and  ignored,  after  a  hypothetical \nredistribution  of  the  total  direct  capacity  from  the  given  pair  of \nterminals  to  every  third  terminal  which  balances  the  two  sides  of  the \npair.  The  balanced-terminal  capacity,  as  thus  defined,  is  equal  to \nthe  direct  capacity  between  the  pair  augmented  by  one-quarter  of\n\nthe  grounded  capacity  of  the  pair,  neither  of  which  is  changed  by  the \nassumed  method  of  balancing.\n\nAs  illustrated  in  Fig.  6,  terminals  1,  2  are  the  given  pair  and  terminal \n3  includes  all  others,  assumed  to  be  connected  together.  A  bridge \nratio  of  unity  is  employed,  and  the  entire  bridge  is  shielded  from \nground  with  the  exception  of  corners  (?,  2)  which  are  initially  balanced\n\nFig.  6 \u2014 Bridge  for   Determining   Hypothetical   Capacity   Between   Two  Terminals \nwith  Other  Terminals  Balanced  and  Ignored\n\nto  ground  within  the  range  of  variable  condenser  /.  The  following \ntwo  successive  balances  are  made:\n\n1.  With  contacts  a,  a'  closed  and  h  open,  balance  is  secured  by \nvarying  condenser  /  (the  total  capacity  of  which  is  constant) \ngiving  the  reading  C  for  its  direct  capacity  in  parallel  with \nterminals  1,  3.\n\n2.  With  contacts  a,  a'  open  and  h  closed,  balance  is  obtained  by \nvarying  condenser  //,  obtaining  the  reading  C\"  for  its  direct \ncapacity  in  (S'2).\n\nIf  Co,  C'l  are  the  corresponding  readings  without  the  network,  the \nbalanced-terminal  capacity  C\\,  and  the  grounded  capacity  unbalance \nof  the  given  pair  of  terminals  are:^\n\nAny  failure  to  afljust  condenser  /  to  perfectly  balance  the  given \npair  of  terminals  will  decrease  the  measured  capacity  Ce,.  This  fact \nmay  be  utilized  to  measure  the  capacity  with  the  second  bridge  ar- \nrangement alone  (contacts  a,  a'  open  and  b  closed)  by  adjusting  con- \ndenser /  so  as  to  make  the  reading  C\"  of  condenser  II  a  maximum. \nThis  procedure  presents  no  difficulty,  since  the  correct  setting  for \ncondenser  /  lies  midway  between  its  two  possible  settings  for  a  balance \nwith  any  given  setting  of  condenser  //;  furthermore,  C\"  is  not  sensi- \ntive to  small  deviations  from  a  true  balance  in  C.\n\nBalanced-terminal  capacity  is  of  practical  importance  as  a  measure \nof  the  transmission  efficiency  to  be  expected  from  a  metallic  circuit, \nif  it  is  subsequently  transposed  so  as  to  balance  it  to  every  other \nconductor.  In  practice,  when  the  unbalance  of  the  section  of  open \nwire  or  cable  pair,  which  is  being  measured,  is  relatively  small,  it  is \nsufficient  to  set  condenser  /,  once  for  all,  to  balance  the  bridge  itself \nand  ignore  the  unbalance  of  the  pair.  This  favors  an  unbalanced \npair,  however,  by  the  amount  [Gi  \u2014  GiY/^  {On  -f  Geo)  where \nGi2  +  GcD  is  the  grounded  capacity  of  the  pair  augmented  by  that \nof  the  bridge.^  For  rapid  working,  condenser  II  is  graduated  to  read \n2C\"  and  by  auxiliary  adjustment  Cg  is  made  zero,  so  that  the  re- \nquired capacity  is  read  directly  from  the  balance.\n\nMeasurement  of  the  capacity  between  the  terminals,  taken  in \npairs  with  all  the  remaining  terminals  left  insulated  or  floating,  gives \nn(n  \u2014  l)/2  independent  results,  from  which  all  the  direct  capacities \nmay  be  derived  by  calculation  of  certain  determinants^.  Practically, \nhowever,  we  are  in  general  interested  in  determining  individual  direct \ncapacities  from  the  smallest  possible  number  of  measurements,  and \nthe  first  step  is  naturally  to  connect  all  of  the  remaining  conductors \ntogether,  so  as  to  reduce  the  system  to  two  direct  capacities  in  addi- \ntion to  the  one  the  value  of  which  is  required.  Three  measurements \nare  then  the  maximum  number  required,  and  we  know  that  two,  or \neven  one,  is  sufficient  if  particular  devices  are  employed.\n\nThe  three  measurement  method  of  determining  direct  capacities \nfrom  the  grounded  capacities  of  the  two  terminals  taken  separately \nGi,  Gi,  and  together  G12,  is  given  by  Maxwell.'\"  If  d  =  \u20ac',  Gn  = \na  +  C\\  and  G2  =  C\"  +  C\"\\\n\nwhich  indicates  a  method  by  which  large  grounded  capacities  can  be \nbalanced  against  three  variable  capacities,  only  one  of  which  need  be \ncalibrated,  and  that  one  need  be  no  larger  than  the  required  direct \ncapacity.\n\nTwo-setting  methods,  as  illustrated  by  the  Colpitts  and  potentio- \nmeter methods,  rest  upon  the  possibility  of  connecting  one  of  the \nassociated  direct  capacities  between  opposite  corners  of  the  bridge \nwhere  it  is  without  influence  on  the  balance,  and  not  altering  any \nassociated  direct  capacity  introduced  into  the  working  arms  of  the \nbridge.  Numerous  variations  of  these  methods  have  been  considered \nwhich  may  present  advantages  under  special  circumstances.  Thus, \nif  conductors  1,  2,  3  of  Fig.  7  are  in  commercial  operation,  and  it  is \nnot  permissible  to  directly  connect  two  of  them  together,  a  double \nbridge  might  be  employed  with  a  testing  frequency  differing  from\n\nthat  of  operation.  A  telephone  is  shown  for  each  ear,  and  a  constant \ntotal  direct  capacity  is  divided  between  the  three  branches  in  the \nproportion  required  to  silence  both  telephones.\n\nOne-setting    methods    attained    ideal    simplicity    in    the    Maxwell \ndischarge  method,  but  we  found  it  necessary  to  use  alternating  current\n\nmethods,  and  here  negative  resistances  make  a  one-setting  method \nat  least  theoretically  possible,  as  explained  above.  Of  possible  varia- \ntions it  will  be  sufficient  to  refer  to  the  ammeter  method  Fig.  8.    Termi-\n\nnals  1  and  2  of  the  required  direct  capacity  Cn  are  connected  to  the \nvoltmeter  and  ammeter  terminals,  respectively,  and  all  other  terminals \ngo  to  the  junction  point  at  3.     Then\n\nprovided  the  ammeter  actually  has  negligible  impedance.  The \nmethod  is  well  adapted  for  rapid  commercial  testing.  The  ammeter \nimpedance  may  be  reduced  to  zero  by  a  variable  negative  impedance \ndevice  (  \u2014  Z),  adjusted  to  reduce  the  shunted  telephone  to  silence.\n\nIn  the  discussion  of  the  bridge,  it  has  been  assumed  that  the  several \npieces  of  apparatus  forming  the  six  branches  of  the  bridge  have  no \nmutual  electrical  or  magnetic  reaction  upon  each  other,  except  as \nindicated.  In  general,  however,  a  balance  will  be  upset  by  changes \nin  position  of  the  pieces  of  apparatus,  or  even  by  movements  of  the \nobserver  himself,  whereas  these  motions  cannot  affect  any  of  the \nmutual  reactions  which  have  been  explicitly  considered.  The  skillful \nexperimenter,  understanding  how  these  variations  are  produced  by \nthe  extended  electric  and  magnetic  fields,  will  anticipate  this  trouble \nand  take  the  necessary  precautions,  possibly  without  slowing  down  his \nrate  of  progress.\n\nWhere  hundreds  of  thousands  of  measurements  are  to  be  made, \nhowever,  substantial  savings  are  effected  by  arranging  the  bridge  so \nthat  reliable  measurements  can  be  made  by  unskilled  observers,  and \nhere  it  is  necessary  to  shield  the  bridge  so  that  any  possible  movements \nof  the  observer  and  of  the  apparatus  will  not  affect  the  results.  Mag- \nnetic fields  of  transformers  are  minimized  by  using  toroidal  coils \nwith  iron  cases.     Electrostatic  fields  are  shielded  by  copper  cases;\n\nthe  principles  of  shielding  were  explained  in  an  earlier  paper /^  Fig.  13 \nof  that  paper  showing  the  complete  shielding  of  the  balance  as  con- \nstructed for  the  measurement  of  direct  capacity  by  the  Colpitts \nmethod.  Over  five  million  capacity  and  conductance  measurements \nhave  been  made  with  the  shielded  capacity  and  conductance  bridge \nand  in  a  forthcoming  paper  Mr.  G.  A.  Anderegg  will  give  details  of \nactual  construction  of  apparatus  and  of  methods  of  operation  as  well \nas  some  actual  representative  results.\n\nFor  simplicity,  the  preceding  definitions  and  methods  of  measure- \nment have  been  described  in  terms  of  capacity,  but  everything  may  be \ngeneralized,  with  minor  changes  only,  for  the  definition  and  measure- \nment of  direct  admittances  with  their  capacity  and  conductance \ncomponents.  The  essential  apparatus  change  is  the  addition,  in \nparallel  with  the  variable  capacity  standards  employed,  of  a  variable \nconductance  standard,  which  shifts  direct  conductance  from  one  side \nof  the  bridge  to  the  other,  without  changing  the  total  reactance  and \nconductance  in  the  two  sides  of  the  bridge.  This  may  be  practically \nrealized  in  a  great  variety  of  ways  as  regards  details,  which  it  will \nsuffice  to  illustrate  by  Fig.  9,  where  C ,  C\" ,  C\" ,  G' ,  G\",  indicate  the\n\npig_  9 \u2014 Variable  Direct  Conductance  and  Capacity  Standard  for  Direct  Admittance\n\ncontinuously  variable  capacity  and  conductance  standards  with \nenough  step-by-step  extensions  to  secure  any  desired  range.\n\nFor  the  continuously  variable  conductance  standard  a  slide  wire  is \nrepresented,  with  a  slider  made  up  of  two  hyperbolic  arcs  so  pro- \nportioned that,  as  the  slider  is  moved  uniformly  in  a  given  oblique \ndirection,  conductance  is  added  uniformly  on  the  left  and  just  enough \nof  the  wire  is  short-circuited  to  produce  an  equal  conductance  de- \ncrease on  the  other  side.  The  arcs  are  portions  of  the  hyperbola \nxy  =  (L^  \u2014  5^)74,  where  L,  S  are  the  total  length  of  the  wire  and  of \nthe  portion  to  be  traversed  by  the  slider,  and  the  coordinate  axes  are\n\nthe  slide  wire  and  the  direction  of  the  motion  of  the  slider  as  oblique \nasymptotic  axes.'^  L  =  GS/g  =  4  G/p(G^  \u2014  g^),  where  G  is  the \ntotal  conductance  and  {G  =*=  g)/2  the  limiting  direct  conductance  on \neither  side.\n\nIf  an  ordinary  slider  replaces  the  hyperbolic  arc  slider,  and  the \nscale  reading  is  made  non-uniform  so  as  to  give  one-half  of  the  differ- \nence between  the  direct  conductances  C?  to  3)  and  C  to  3),  the  con- \nductance standard  will  still  give  absolutely  correct  results  with  the \nColpitts  method,  provided  the  bridge  ratio  is  unity.  This  simplifica- \ntion in  connection  with  the  balancing  capacity  /  of  Fig.  6  would, \nhowever,  not  be  strictly  allowable.  For  improvised  testing  we  have \nfound  it  sufficient  to  use  two  equal  resistances  {R)  with  a  dial  re- \nsistance {r)  in  series  with  one  of  them,  and  take  the  defect  of  con- \nductance introduced  by  the  dial  resistance  as  equal  to  r/F?  or  to \n10~2r,  10~V,  r,  micromho  according  as  R  was  made  10000,  3162, \nor  1000  ohms. 13\n\nFor  a  step-by-step  conductance  standard.  Fig.  9  shows  a  set  of \n10  equal  resistances,  connected  in  series  between  corners  ^,  (?,  to \nthe  junction  points  of  which  there  is  connected  a  parabolic  fringe  of \nresistances,  the  largest  of  which  is  2.5  times  each  of  the  ten  resistances. \nWith  this  arrangement  the  direct  conductance  in  CFS)  may  be  ad- \njusted by  ten  equal  steps,  beginning  with  zero,  whi'e  the  conductance \nin  (?2)  is  decreased  by  equal  amounts  to  zero.  The  total  res'stance \nrequired  for  this  conductance  standard  's  only  21/25  of  the  res'stance \nrequired  to  make  a  single  isolated  conductance  equal  to  one  of  the \nten  conductance  steps;  the  ratio  may  be  reduced  to  1/2  by  doubl  ng \nthe  number  of  contacts, ^^  and  using  onsfringe  resistance  for  all  posi- \ntions. Resistance  may  be  still  further  economized  by  using  as  high  a \ntotal  conductance  as  is  permissible  in  the  bridge,  and  securing  the \nrequired  shift  in  conductance  from  a  small  central  portion  of  the \nparabolic  fringe.\n\nFig.  9  shows  the  variable  capacity  standards  as  well  as  the  variable \nconductance  standards  and  a  few  practical  points  connected  with  the \ncapacity  standards  may  be  mentioned  here.\n\nThe  revolving  air  condenser  standard  has  two  fixed  plates  connected \nto  <^  and  (?,  so  that  the  capacity  will  increase  as  rapidly  on  one  side \nas  it  decreases  on  the  other  side.  Since  perfect  constancy  of  the  total \ncapacity  is  not  to  be  expected,  on  account  of  lack  of  perfect  mechanical \nuniformity,    the   revolving   condenser   should    be   calibrated    to    read\n\none-half  of  the  difference  between  the  capacities  on  the  two  sides,  as \nexplained  above  in  connection  with  conductance.  The  capacity \nsections  employed  to  extend  the  range  of  the  revolving  condenser \ninclude  both  air  condensers  C  and  mica  condensers  C\" ,  the  latter \nbeing  calibrated  by  means  of  the  air  condensers  and  the  conductance \nstandard.\n\nA  novel  feature  of  our  standard  air  condensers  is  a  third  terminal \ncalled  the  leakage  terminal,  and  indicated  at  L  in  Figs.  4,  9.  Attached \nto  it  are  plates  so  arranged  that  all  leakages  either  over,  or  through, \nthe  dielectric  supports  from  either  of  the  two  main  terminals,  must \npass  to  the  leakage  terminal.  There  can  be  no  leakage  directly \nfrom  one  of  the  main  terminals  to  the  other.  There  is  thus  no  phase \nangle  defect  in  the  standard  direct  capacity  due  to  leakage,  and  that \ndue  to  dielectric  hysteresis  in  the  insulating  material  is  reduced  to  a \nnegligible  amount  by  extending  the  leakage  plates  beyond  the  dielec- \ntric, so  as  to  intercept  practically  all  lines  of  induction  passing  through \nany  support.  This  leakage  terminal  is  connected  to  corner  C  of  the \nbridge;  in  the  revolving  condensers,  it  is  one  of  the  fixed  plates.\n\nThe  reciprocal  of  a  direct  admittance  is  naturally  termed  a  direct \nimpedance;  substituting  impedance  for  capacity,  the  definition  of \ndirect  capacity,  given  above,  becomes  the  definition  of  direct  im- \npedance. The  complete  set  of  ciirect  impedances  constitutes  an  exact, \nsymmetrical,  physical  substitute  for  any  given  electrical  system. \nDirect  impedances  are  often,  in  whole  or  in  part,  the  most  convenient \nconstants  since  many  electrical  networks  are  made  up  of,  or  ap- \nproximate to,  directly  connected  resistances  and  inductances.  To \nmake  direct  impedance  measurements  which  will  not  involve  the \ncalculation  of  reciprocals,  we  naturally  employ  inductance  and  re- \nsistance standards  in  series,  the  associated  direct  impedances  being \neliminated  as  with  direct  capacities.\n\nIt  has  been  necessary  to  preface  the  description  of  methods  of \nmeasuring  direct  capacities  by  definitions  and  a  brief  discussion,  since \ndirect  capacities  receive  but  scant  attention  in  text-books  and  hand- \nbooks. By  presenting  direct  capacities,  direct  admittances,  and \ndirect  impedances  as  alternative  methods  of  stating  the  constants  of \nthe  same  direct  network,  employed  as  an  equivalent  substitute  for \nany  given  electrical  system,  it  is  believed  the  discussion  and  measure-\n\nmerit  of  networks  has  been  simplified.  In  another  paper  the  terminol- \nogy for  admittances  and  impedances  will  be  still  further  considered, \ntogether  with  their  analytical  correlation.\n\nIn  explaining  the  different  methods  of  measuring  direct  capacities \nit  is  necessary  to  start  with  a  clear  idea  of  what  direct  capacities  are, \nand  to  make  use  of  the  additive  property,  but  it  is  not  necessary  to \ngo  into  any  comprehensive  discussion  of  direct  capacities.  Accord- \ningly, the  mathematical  treatment  of  direct  capacities  has  been  reserved \nfor  another  paper,  but  it  seems  desirable  to  append  to  the  present \npaper  proofs  of  the  analytical  results  given  in  this  paper,  since  the \nmethod  of  approach  giving  the  simplest  proof  is  not  always  perfectly \nobvious.\n\n(1)  Reducing  the  number  of  terminals  which  are  considered \naccessible,  by  ignoring  terminals  p,  q,  r,  .  .  .  ,  changes  the  direct  and \ngrounded  capacities  from  (Qy,  G,)  to  (Qj ,  G,'),  the  latter  being  ex- \npressed in  terms  of  the  former  as  follows:\n\nTo  check  these  formulas  note  that  on  substituting  (G,-,  \u2014  Gy)  for \nMaxwell's  (9,-,-,  qij)  in  his  equations  (18)'*  the  coefficients  form  an \narray  in  which  the  grounded  capacity  G,-  is  the  ith  element  in  the  main \ndiagonal  and  \u2014  Cij  is  the  element  at  the  intersection  of  row  i,  column \nj.  The  array  may  be  supposed  to  include  every  terminal  symmetrically \nby  considering  the  earth's  potential  as  being  unknown  and  writing \ndown  the  redundant  equation  for  the  charge  on  the  earth.  Let  the \ncharge  be  zero  on  terminal  j  and  on  all  concealed  terminals;  let  there \nbe  a  charge  on  terminal  i  and  an  equal  and  opposite  charge  on  all  the \nremaining  accessible  terminals,  connected  together  to  form  a  single \nterminal  k.     Now  taking  the  potential  of  j  as  the  zero  of  reference\n\nand  calculating  the  potentials  of  i  and  k  and  then  allowing  the  direct \ncapacity  between  j  and  k  to  become  infinite,  the  direct  capacity \nbetween  terminals  i  and  j  is  dj  =  \u2014  Lim  {Qk  Vk  /Vi).  This  gives \nthe  above  formula  for  C,-^,  with  \u2014  C\u201e-  as  a  special  case.  This  method \nis  an  electrostatic  counterpart  of  the  ammeter  method  shown  in \nFig.  8  on  page  29.\n\nIf  there  is  but  one  ignored  terminal  the  determinant  solution  takes \non  a  simple  form  from  which  Rules  1  and  2  and  Fig.  1  may  be  checked.\n\nIf  all  but  two  terminals  are  ignored  the  equivalent  direct  network \nis  reduced  to  a  single  direct  capacity.  When,  for  each  pair  of  termi- \nnals, this  capacity  dj  is  known,  from  measurements  or  from  calcu- \nlations, the  direct  capacities  between  the  terminals  may  be  derived \nby  means  of  the  following  formulas\n\nwhere  Dij  is  the  cofactor  of  the  element  in  row  i  column  j  of  the \ndeterminant\n\nwhich  has  zeros  in  the  main  diagonal,  a  border  of  ones  in  the  last \nrow  and  column,  while  the  other  elements  are  5^  =  1/C,-,-,  that  is, \nthe  reciprocals  of  the  given  capacities.  The  5's  form  a  complete \nsymmetrical  system  of  network  constants;  Maxwell's  coefficients  of \npotential  pa  with  the  two  sufifixes  the  same  are  the  same  quantities, \nbut  he  employs  only  those  coefificients  of  this  type  which  are  associ- \nated with  the  earth,  his  system  being  completed  by  adding  the  co- \nefficients with  different  suffixes.  By  starting  with  Maxwell's  results \nthe  above  formula  may  be  deduced,  but  more  direct  proofs,  both \nphysical  and  mathematical,  will  be  given  in  the  theoretical  paper \nreferred  to  at  the  end  of  the  present  paper.\n\nThe  purpose  of  this  section  of  the  appendix  is  achieved  if  the  de- \nterminant solutions  are  made  so  clear  as  to  be  available  for  use  in \nany  particular  case.\n\n(2)  Starting  with  the  bridge  alone  balanced  at  reading  C**  the  other \ntwo  settings  involve,  in  the  capacity  standard,  increases  in  the  direct\n\ncapacity  on  the  left  of  (C  \u2014  C\u00b0)  and  (C\"  \u2014  C\u00b0),  with  equal  decreases \non  the  right.     Therefore\n\n(3)  The  condition  of  equal  impedance  ratios  on  the  two  sides,  as \nrequired  for  a  balance,  gives,  for  both  the  switches  up  and  down,\n\n(4)  The  Y  of  Fig.  4  has  unusual  properties  because  the  total \nconductance  connecting  the  concealed  branch-point  of  the  Y  to  the \nthree  bridge  corners  ^,  ^,  (?  is  zero.  Thus  the  conductance  between \nany  one  corner  and  the  remaining  two  corners  joined  together  is  infinite, \nor  in  other  words,  the  Y  acts  as  a  short  circuit  under  all  these  three \nconditions.  On  the  other  hand,  if  corner  ^,  ^,  or  (F  is  left  floating \nand  ignored  the  conductance  between  the  other  two  corners  is  2/R, \n1/2R  or  2/R,  respectively,  and  the  Y  is  not  a  short  circuit.  These \nstatements  are  verified  at  once  by  applying  the  familiar  expressions \nfor  resistances  in  parallel  and  in  series.\n\nOn  account  of  the  unusual  behavior  of  the  Y,  even  when  taken \nalone,  it  is  not  immediately  apparent  how  it  will  affect  the  operation \nof  the  bridge  of  Fig.  4  with  direct  capacities  between  corners  ^S^ \nand  S^C  For  this  reason  it  is  highly  desirable  to  find  an  equivalent \nnetwork  the  behavior  of  which  is  more  readily  comprehended.  It  is \nnot  feasible  to  employ  the  delta  network  which  is  equivalent  to  the\n\nY  for  this  has  indeterminate  characteristics,  being  made  up  of  three \ninfinite  conductances,  only  two  of  which  have  the  same  sign.  We \nmay,  however,  make  use  of  the  Y  which  is  equivalent  to  the  original\n\nY  and  direct  capacities  C^b  and  Cbc  taken  together.  This  is  found  as \nfollows:  Any  admittance  delta  may  be  replaced  by  a  star  having  ad- \nmittances equal  to  the  sum  of  the  products  of  the  delta  admittances \ntaken  in  pairs  divided  by  the  opposite  delta  admittance.  Applied  to  the \ndelta  of  Fig.  1,  we  find  that  the  star  that  is  equivalent  has  the  capacities\n\nwhich,  upon  substituting  the  value  of  d,  is  the  sum  of  16  terms,  each \nof  which  is  the  product  of  three  capacities,  every  combination  of  three \ncapacities  being  included  except  the  four  cases  in  which  the  three \ncapacities  would  form  a  closed  circuit.  By  allowing  the  capacities \nto  be  complex  quantities,  any  admittances  are  covered  by  the  formulas. \nIf  G4  =  0\n\nor  the  new  Y  arms  present  the  same  ratios  as  the  original  Y  arms \ntaken  alone;  that  is,  the  direct  capacities  C^b^  Cbc  of  Fig-  4  have \nno  effect  on  the  bridge  ratio.  Thus  the  constancy  of  the  bridge  ratio \nholds  for  all  null-impedance  bridges  regardless  of  the  ratio  Z1/Z2  and \nof  the  nature  of  the  direct  admittances  from  corners  A  and  C  to  B. \nIf  G4  =  0  and  also  C24  =  C14  and  C\\i  =  0,  then\n\nApplying  this  to  Fig.  4,  which  is  possible  since  the  bridge  ratio  is \nunity,  we  find  that  the  three  arms  of  the  equivalent  Y  may  be  con- \nsidered as  being  made  up  of  resistances  and  capacities  in  parallel. \nThe  resistances  are  R,  R,  \u2014R/2  and  the  associated  capacities  C,  C, \n\u2014  2C,  where  R  is  the  original  resistance  in  the  Y  and  C  is  one-half  the \nsum  of  the  two  actual  direct  capacities  from  ^  to  f^and  from  -S^  to  (?. \nThe  equivalent  bridge  thus  obtained  has  ratio  arms  made  up  of \nordinary  resistances  and  capacities  and  therefore  Fig.  4  used  as  a \nbridge  can  present  no  unexpected  characteristics;  the  negative \nresistances  and  capacities  of  the  equivalent  Y  merely  affect  the  current \nsupplied  to  the  bridge.\n\nAn  ideal  transformer,  if  such  a  device  existed,  might  replace  the  Y, \nfor  it  would  maintain  a  constant  ratio  between  the  currents  in  the \ntwo  windings  and  act  as  a  short  circuit  when  the  bridge  is  balanced. \nTo  determine  the  error  when  an  actual  transformer  with  impedances \nZp,  Zs,  Zps  is  employed,  take  the  general  expression  for  the  ratio \nof  the  capacities  derived  above  which  is\n\nChange  to  admittances  by  substituting  Y  for  C  and  G  throughout. \nAssume  the  transformer  replaced  by  its  equivalent  conductance  star \nso  that\n\nSubstituting  these  values  the  expression  for  the  actual  ratio  of  the \nbridge  arms  becomes\n\n(5)  When  the  bridge  alone  is  balanced  at  readings  Co  and  Co  , \nlet  CcD  and  Geo  be  the  direct  capacity  between  corners  C  and  2)  and \nthe  total  direct  capacity  between  these  corners  and  ground.  Since \nGcD  is  balanced,  the  effective  direct  capacity  between  corners  C,  2) \nwhen  earth  is  ignored,  is  by  Fig.  1,  {Ccd  +  G^cd/4)-  Now  connect  the \nthree  terminals  1,  2,  3,  as  shown  with  direct  capacities  C12,  G\\  \u2014  C\\i, \nGi  \u2014  C12;  Gi,  Gi  being  the  grounded  capacities  of  terminals  1  and  2. \nThe  first  balance  with  the  reading  C  requires  the  equality  of  the \ntotal  capacity  added  on  each  side,  i.e.,\n\nFor  the  second  balance  ground  may  again  be  considered  an  ignored \nterminal,  and  since  terminals  1  and  2  have  been  balanced  to  ground, \nand  their  total  direct  capacity  to  ground  is  G12  =  Gi  -\\-  d  \u2014  2Ci2, \nthe  effective  direct  capacity  added  to  the  bridge  between  corners \n(?  and  2)isCb=  Cn  +  Gn/4:.  Equating  the  added  capacities  on  the \ntwo  sides  of  the  bridge  when  balanced  at  the  reading  C\",  we  obtain \nG  =  2  (C\"  -  C).\n\nThe  direct  capacity  between  (?  and  2),  when  ground  is  considered \nan  accessible  terminal,  is  assumed  to  be  absolutely  independent  of \nthe  setting  of  the  condenser  /.  To  actually  meet  this  condition  will \nrequire  some  attention  in  the  design  of  the  variable  condenser.\n\n(6)  Here  the  bridge  itself  is  supposed  to  have  equal  direct  capaci- \nties from  corners  <?and  ^to  ground,  while  the  added  terminals  1  and  2 \nhave  different  direct  capacities  to  ground,  the  difference  being \n(Gi  \u2014  G2),  while  the  total  direct  capacity  to  ground  is  {Gn  +  Geo)- \nNow  two  capacities  in  series  may  be  replaced  by  their  product  divided\n\nby  their  sum,  which  is  equal  to  one-fourth  of  the  sum  minus  the  square \nof  the  difference  divided  by  four  times  the  sum.  The  correction \ndue  to  the  difference  is  thus  (G2  \u2014  Gi)V4(Gi2  +  G'cd),  as  stated.\n\n(7)  These  determinants  are  given  at  the  end  of  the  first  paragraph \nof  this  appendix.  These  expressions  for  the  direct  capacity  are  of \nmore  special  interest  in  the  analytical  discussion  of  networks.\n\n(8)  Assume  that  a  wire  resistance  is  to  be  employed  and  that  a \nsliding  contact  is  to  intercept  such  an  amount  of  resistance  that  the \nequivalent  conductance  will  vary  directly  with  the  motion  of  the \nslider  carrying  the  contact  point.  Then  if  the  wire  is  straight  and  the \nintercepted  portion  is  of  length  x  and  the  slider  motion  is  rectilinear \nand  its  extent  is  y  the  relation  which  holds  between  them  is  xy  =  con- \nstant, the  value  of  the  constant  depending  upon  the  units  employed.\n\nIn  the  paper  it  is  assumed  that  the  total  conductance  G,  the  total \nshifted  conductance  g,  and  the  resistance  of  unit  length  of  the  slide \nwire  p  are  given ;  the  total  length  of  wire  L  and  the  portion  traversed \nby  the  slider  S  are  then  calculated.  The  arc  employed,  for  each \nhalf  of  the  slider  of  Fig.  9,  extends  equally  both  ways  from  the  vertex \nto  the  points  where  the  values  of  x  and  y  are  (L  \u00b1  S)/2,  on  the  hyper- \nbola xy  =  {U  \u2014  5^)/4,  Substituting  for  L  and  5  the  values  given \nin  the  paper,  it  will  be  found  that  this  range  of  x  actually  gives  the \nrange  of  conductance  {G  \u00b1  g)/2,  as  required.\n\n(10)  At  mid-point  the  total  conductance  due  to  the  five  resistances \n(R)  on  each  side,  taken  in  parallel,  is  2/5R  and  to  give  this  same \nconductance  an  end  fringe  must  have  the  resistance  2.5R.  Assume \na  parabolic  fringe  having  the  resistance  (5  \u2014  nY  R/10  at  the  point \nconnected  to  C?  and  (?  by  resistances  nR  and  (10  \u2014  n)R.  This  gives \na  Y  network  and  by  Fig.  1  the  equivalent  direct  conductances  are \n(10  -  n)/25R,  n/25R,  (5  -  ny/250R  between  =0^,  ^(?,  d'C  re- \nspectively. The  sum  of  the  first  two  is  constant  and  the  first  de- \ncreases by  equal  steps,  of  l/25i?  each,  to  zero  as  the  second  increases. \nThe  parabolic  fringe,  therefore,  gives  the  required  conductances.\n\nThe  total  resistance  in  the  chain  of  ten  resistances  is  lOi?,  in  the \nfringe  Hi?,  and  in  the  largest  single  fringe  2.hR.  With  the  complete \nfringe  the  total  required  is  (10  +  11)  i?  =  2\\R\\  with  a  single  fringe, \nsubdivided  as  required,  only  (10  +  2.5)  R  =  12.5R  is  required. \nCompared  with  25R,  which  would  be  required  for  one  of  the  con- \nductance steps,  these  resistances  are  21/25  and  1/2.\n\nTHE  purpose  of  this  paper  is  to  present  a  simple  theoretical  treat- \nment of  those  features  of  the  Petersen  methcd  of  grounding  a \npower  network  which  are  of  principal  interest  from  the  standpoint \nof  inductive  effects  in  neighboring  communication  circuits.  In  this \nmethod,  the  neutral  of  the  system  is  grounded  through  an  inductance \nwhich  is  in  resonance,  at  the  fundamental  frequency,  with  the  total \ndirect  capacity  of  the  system  to  ground.  The  theory  of  the  behavior \nof  a  power  system  thus  grounded  at  times  of  accidental  faults  to \nearth  has  been  developed  by  Petersen  in  a  paper  published  in  1919,' \nin  which  the  results  of  field  tests  and  of  operating  experience  with  an \ninstallation  in  Germany  are  also  described.  The  methcd  has  also \nfound  application  in  other  places  in  Europe,  chiefly  in  Italy  and \nSwitzerland.  It  does  not  appear  in  any  of  these  cases  that  inductive \ninterference  was  a  factor  requiring,  or  at  any  rate  receiving,  consider- \nation. In  fact,  it  does  not  seem  that  either  Petersen  himself,  or  other \nengineers  in  Europe  who  have  made  use  of  his  scheme,  have  considered \nit  except  as  a  method  of  protecting  power  systems  from  the  effects  of \naccidental  grounds.\n\nThe  features  of  the  method  that  are  of  interest  from  the  viewpoint \nof  inductive  interference  relate  both  to  normal  operating  conditions \nof  the  power  system  and  to  the  phenomena  which  occur  when  a  phase \nof  the  system  is  grounded.  With  regard  to  the  former,  it  is  principally, \nthough  not  entirely,  the  effect  of  the  neutral  reactor  on  the  harmonics \nof  frequencies  within  the  voice  range  that  require  examination;  with \nrespect  to  the  latter,  the  things  of  chief,  though  not  exclusive,  import- \nance, are  the  ground  currents  and  unbalanced  voltages  to  ground  of \nfundamental  frequency,  which  are  possible  sources  of  disturbance \nin  exposed  communication  circuits.\n\nThese  features,  particularly  those  concerned  with  effects  at  funda- \nmental frequency,  are  more  or  less  closely  related  to  questions  of \nprimary  importance  from  the  standpoint  of  power  system  operation. \nIt  is  impossible  that  this  should  not  be  the  case.  Accidental  dis- \nturbances of  a  character  which  may  interrupt  service  or  endanger \napparatus  or  equipment  in  a  power  system  may  produce  inductive\n\ndisturbances  in  neighboring  communication  circuits,  and  the  question \nof  grounding  the  neutral,  in  whatever  manner,  or  of  leaving  it  isolated, \nexists  because  of  its  bearing  on  the  avoidance  or  limitation  of  such \npower  system  disturbances.  Thus  a  method  of  grounding  the  neutral, \nor  any  other  method  of  power  system  operation  designed  to  limit  the \nextent  or  the  severity  of  accidental  disturbances,  must  necessarily \npossess  importance  with  respect  to  inductive  effects  in  exposed  com- \nmunication circuits.\n\nIt  does  not  seem  necessary,  therefore,  to  apologize  for  the  discussion, \nin  the  first  section  of  the  paper,  of  the  behavior  of  the  power  system  at \ntimes  of  faults  to  earth.  In  this  section,  an  explanation  is  given  of \nthe  principal  characteristic  effect  of  the  reactor  which  differs  in  some \nrespects  from  that  set  forth  by  Petersen  in  the  paper  already  referred \nto.  The  matter  of  transient  over-voltage  on  a  non-grounded  phase \nis  also  examined  in  this  section,  and  the  bearing  of  these  and  the \nearlier  considerations  on  inductive  effects  is  discussed.\n\nIn  the  second  section  of  the  paper,  the  behavior  of  the  power  system \nwith  reactor  under  normal  operating  conditions  is  discussed  with \nreference  to  noise  and  other  inductive  effects  in  neighboring  com- \nmunication circuits.\n\nReferring  to  Fig.  1,  in  which,  and  in  the  following  discussion  it  is- \nassumed  that  the  three  admittances  to  ground  are  equal,  the  ad- \nmittance current  through  the  fault  from  the  two  sound  phases  is\n\nFig.  1 \u2014 Single-Phase  System  with  Neutral  Grounded  Through  Reactor.    Admittances \nto  Ground  Assumed  Balanced\n\nOn  comparison  of  this  expression  with  the  above  equation  (1)  for \nthe  charging  current,  which  constitutes  the  fault  current  if  the  system \nis  isolated,  it  is  seen  that,  if  the  losses  in  the  system  are  small,  the \neffect  of  the  coil  is  to  reduce  the  magnitude  of  the  current  in  the  fault\n\nunity.  Further,  as  equation  (2)  shows,  the  phase  of  the  fault  current \ncoincides  with  that  of  the  voltage  impressed  between  the  faulty  wire \nand  ground  to  the  degree  of  approximation  here  used,  i.e.,  to  the \nsecond  order  of  small  quantities.  (The  bracket  on  the  right-hand \nside  of  (2),  if  written  in  full,  would  include  quadrature  terms  in  the \nsquare  and  higher  even  powers  of  r\u201e/uL\u201e.)  With  the  system  isolated, \nthe  phase  displacement  is  nearly  90\".\n\nThe  coincidence  in  phase  of  the  fault  current  and  the  voltage  im- \npressed by  the  transformer  on  the  faulty  wire,  together  with  the  small \nmagnitude  of  the  former,  are  very  farorable  to  the  suppression  of  the\n\narc.  Following  its  extinction,  there  is  a  further  action  of  the  resonant \nsystem  consisting  of  the  coil  and  the  total  capacity  to  ground  which \nacts  in  such  a  way  as  to  prevent  any  over-voltage  and  to  restore  the \nnormal  potentials  to  ground  gradually,  thus  tending  to  prevent  the \narc  from  restriking.     This  action  is  as  follows:\n\nFig.  2 \u2014 Vector  Diagram  Showing  Relations  of  Voltages  and  Currents  at  and  Follow- \ning Extinction  of  Arc\n\nReferring  to  Fig.  2,  the  vectors  \u00a3io,  -E20,  -E30  represent  the  emfs. \nimpressed  between  lines  and  neutral  by  the  transformer  bank.  7^, \n/\u201e,  If  are  respectively  the  admittance  current  to  the  fault,  the  current \nsupplied  by  the  reactor  to  the  fault,  and  the  total  fault  current.  The \nfault  is  assumed  to  be  on  phase  3.  The  arc  will  go  out  when  If  passes \nthrough  zero.  At  this  instant,  \u00a330  is  also  zero  and  /\u201e  and  7^  have \nnearly  their  maximum  values.  Their  instantaneous  values  are,  how- \never, exactly  equal  and  opposite  in  the  sense  indicated  by  the  arrows \nin  Fig.  1,  i.e.,  regarded  as  currents  fed  to  the  fault  by  the  two  parallel \ncircuits  (1)  coil-fault-faulty  wire-Eso  and  (2)  admittance  of  sound \nphases-fault-faulty  wire-transformer  bank.  These  instantaneous \ncurrents  are  exactly  equal  in  magnitude  and  are  in  the  same  direction \nin  the  single  series  circuit  consisting  of  coil,  transformers,  admittance \nto  ground  of  the  three  phases  in  parallel,  and  ground.  Thus  the \ncondition  in  this  series  resonant  circuit  at  the  instant  of  extinction \nis  that  of  an  established  free  oscillation,  the  energy  of  the  oscillation \nbeing  at  this  instant  wholly  electromagnetic.\n\nThe  voltage  across  the  reactor  due  to  the  current  7\u201e,  in  the  direction \nof  \u00a330  (Fig.  1)  is  represented  in  Fig.  2  by  the  vector  \u00a3og-  This  is \n180\u00b0  out  of  phase  with  E30  and  initially  of  the  same  amplitude.  At \nthe  instant  the  arc  goes  out,  both  are  practically  zero.  As  the  oscilla- \ntion progresses,  Eq^  dies  away,  due  to  damping,  and  the  resultant\n\nvoltage  of  the  faulty  wire  to  ground,  viz.,  E?g  =  \u00a330  +  \u00a30,,  passes \ngradually  back  to  the  normal  value  \u00a330.  At  the  same  time  the  voltages \nto  ground  of  the  two  sound  phases  (\u00a3ig,  E,g,)  return  to  their  normal \nvalues  \u00a310,  \u00a320-  The  ends  of  the  three  vectors,  \u00a33^,  \u00a3,,,  \u00a3?,,  may  be \nthought  of  as  sliding  at  equal  rates  along  the  line  \u00a330  and  the  dotted \nlines  parallel  to  it.\n\nThe  efifectiveness  of  the  action  just  sketched  (it  has  been  assumed \nthat  the  frequency  of  the  series  resonant  circuit  is  accurately  that  of \nthe  system  fundamental),  of  course,  depends  on  the  accuracy  of  the \ntuning  and  the  amount  of  damping.  If  the  free  period  of  the  resonant \ncircuit  differs  considerably  from  the  fundamental  period  of  the  system, \nthe  impressing  on  the  faulty  wire  of  a  voltage  to  ground  in  excess  of \nnormal  may  result,  especially  if  the  damping  is  small.  The  effect  of \ninexact  tuning  is  discussed  by  Petersen  in  the  article  referred  to  above. \nHe  describes  some  experiments  in  which  the  capacity  of  the  power \nsystem  to  ground  was  varied  some  15  or  20%,  each  way  from  the \nvalue  corresponding  to  resonance,  without  apparent  effect  on  the \nquenching  action  of  the  reactor.\n\nTo  simplify  the  following  theoretical  discussion  a  single  phase \nsystem  is  treated.     This  is  represented  in  Fig.  3.     Referring  to  this\n\nFig.  3 \u2014 Three-phase  System  with  Neutral  Grounded  Through  Reactor,  with  Fault\n\nfigure,  when  one  phase  is  grounded  (represented  by  the  closing  of  the \nswitch  S),  the  following  equations,  in  which  D  denotes  differentiation \nwith  respect  to  time,  must  be  satisfied:\n\nTo  solve  the  equations,  the  cubic  equation  <!>  {D)  =  0  must  be  solved. \nAn  algebraic  solution  would  be  so  cumbersome  as  to  be  impracticable. \nThe  following  numerical  values  of  the  constants  have  therefore  been \ninserted,  as  what  is  desired  is  a  numerical  solution  representing  the \neffect  in  a  practical  case:\n\nwhich  may  be  denoted  by  \u2014a',  \u2014a  -\\-  jb,  \u2014  a  \u2014  jb  respectively. \nThe  resulting  equations  for  iig  and  i\u201e  are\n\nThe  relations  between  the  two  sets  of  arbitrary  constants  may  be \nobtained  by  inserting  these  solutions  in  the  following  differential \nequation  connecting  iig  and  i\u201e:\n\nand  the  three  independent  arbitrary  constants  so  found  are  determined \nby  the  following  conditions  when  ^  =  0  (it  is  assumed  that  breakdown\n\nThe  numerical  quantities  in  the  second  and  third  of  these  equations \nare  respectively  the  total  current  supplied  to  the  sound  wire  and  the \ntotal  charge  on  this  wire  at  the  instant  of  breakdown.  They  are \nobtained  by  solving  the  network  of  Fig.  3,  with  switch  S  open  and \ntaking  instantaneous  values  when  the  impressed  e.  m.  f.  is  a  maximum. \nThe  resulting  expressing  for  iig  is\n\nThe  non-oscillatory  term  0.3  X  10~\u00aee~^^\"*'  is  seen  to  be  negligible \ncompared  to  the  others.\n\nThe  voltage  between  the  sound  wire  and  ground  is  obtained  by \ndividing  the  above  result  by  g  =  0.37  X  10\"^,  and  is\n\nThis  equation  is  plotted  for  about  13^  cycles  of  fundamental  frequency \nin  Fig.  4.  The  non-oscillatory  term  is  negligible.  As  will  be  seen, \nthe  maximum  overvoltage  is  about  30%.\n\nFor  comparison  purposes,  the  voltage  to  ground  of  the  sound  phase \nwith  the  reactor  omitted  {i.e.,  with  the  system  isolated)  has  been  cal- \nculated, using  the  same  constants  as  before.     The  result  is\n\nThis  is  practically  identical  with  the  earlier  result;  that  is,  the  voltage \nto  ground  of  the  sound  phase  is  practically  the  same  with  the  reactor \nas  with  the  system  isolated.\n\n3.     Effects  with  Respect  to  Induction  in  Neighboring  Communication \nCircuits\n\nAn  estimate  of  the  value  of  the  Petersen  coil  must  involve  a  com- \nparison with  other  methods  of  grounding  the  neutral  (including \ngrounding  through  an  infinite  impedance,  i.e.,  the  isolated  system)  or \nof  otherwise  limiting  the  effects  of  abnormal  occurrences  in  a  power \nsystem.  As  regards  the  induction  of  fundamental  frequency  voltages \nin  exposed  communication  circuits,  the  methods  of  chief  importance \nin  such  a  comparison,  at  least  so  far  as  American  practice  is  con- \ncerned, are  that  in  which  the  neutral  is  grounded  either  directly  or \nthrough  a  low  resistance,  and  that  in  which  the  neutral  is  isolated.\n\nWhen  accidental  grounds  occur  on  a  power  system  with  neutral \ngrounded  through  zero  or  a  low  impedance,  the  resulting  heavy  short \ncircuit  currents  to  ground  may  produce  severe  disturbances  in  ex- \nposed communication  circuits.  Owing  to  the  fact  that  these  dis- \nturbances are  produced  by  electromagnetic  induction  in  a  circuit \nconsisting  of  the  communication  conductor  as  one  side  and  the  earth \nas  the  other,  they  cannot  be  avoided  by  enclosing  the  communication \nconductors  in  lead-sheathed  cable,  even  when  this  is  placed  under- \nground.\n\nWith  the  Petersen  coil,  according  to  the  explanation  in  the  first  part \nof  this  section,  the  neutral  current  of  fundamental  frequency  due  to  a \nfault  to  ground  is  made  equal  to  the  charging  current  of  the  system \nto  ground  with  one  phase  grounded,  and  this  is  generally  a  small \nfraction^  \u2014 a  few  per  cent,  or  less \u2014 of  the  neutral  current  in  an  identi- \ncal system  with  neutral  directly  grounded.\n\nThe  Petersen  coil  will  thus  in  many  cases  largely  prevent  the  electro- \nmagnetic inductive  effects  at  fundamental  frequency  which  appear \nwhen  a  fault  to  ground  occurs  on  a  system  grounded  solidly  or  through\n\n\u00bb  Exceptions  to  this  statement  exist  in  the  case  of  extensive  high  voltage  networks \nwhere,  with  a  ground  on  one  phase,  the  charging  current  to  ground  with  isolated \nneutral  may  be  of  the  same  order  of  magnitude  as  the  short  circuit  current  with  dead- \nerounded  neutral  if  the  fault  is  remote  from  a  point  of  main  power  supply.\n\na  low  resistance.  There  will  appear,  however,  electrically^  induced \nvoltages  of  fundamental  frequency  substantially  identical  with  those \nthat  would  occur  with  neutral  isolated.  Where  the  communication \ncircuits  are  in  underground  cable,  these  voltages  are  of  inappreciable \nmagnitude,  and  with  aerial  cables  (with  metallic  sheaths)  their  effects \ncan  in  general  be  controlled  without  great  difficulty.  With  open \nwire  communication  circuits,  electrically  induced  voltages  are  of \nmuch  more  consequence.  They  may  in  some  cases  equal  or  exceed \nthe  voltages  which  would  be  induced  electromagnetically  with  dead- \ngrounded  neutral.  However,  except  perhaps  in  cases  of  long  ex- \nposure to  high  voltage  power  circuits  at  close  separations,  their \neffects  are  generally  much  less  severe  than  the  electromagnetic  effects, \nbecause  of  the  smaller  amount  of  energy  transferred  to  the  disturbed \ncircuit.  This  is  in  general  accordance  with  experience  with  open \nwire  circuits  exposed  to  power  circuits  of  moderate  voltage.  As  is \nexplained  in  the  next  paragraph,  the  use  of  a  Petersen  reactor  to  ground \nthe  neutral  may  be  expected  to  lessen  the  severity  of  inductive  effects \nwhich  would  be  experienced  from  an  isolated  system,  by  preventing \nadditional  parts  of  the  power  system  from  becoming  involved.\n\nAs  compared  to  the  isolated  system,  the  use  of  the  Petersen  coil  in \nthe  connection  from  neutral  to  ground  may  be  expected  to  have  the \nadvantage,  according  to  the  theory  of  the  first  subdivision  of  this \nsection,  of  preventing  the  formation  of  an  \"arcing  ground.\"  As  ex- \nperience has  shown,  an  arcing  ground  in  an  isolated  system  is  fre- \nquently the  cause  of  serious  disturbances  which  may  involve  portions \nof  a  network  remote  from  the  location  of  the  original  trouble.  The \nadvantage  of  the  reactor  in  this  respect  is,  of  course,  a  fundamental \none  from  the  standpoint  of  power  operation.  It  is  in  general  of  pro- \nportionate importance  from  the  inductive  interference  point  of  view, \nat  least  where  a  power  network  is  involved  in  parallels  with  com- \nmunication circuits  at  several  places,  as  is  not  infrequenctly  the  case \nnear  large  cities.  A  breakdown  to  ground  in  the  power  network  on  a \ndifferent  phase  from  that  originally  involved,  and  in  a  different  locality, \nmay  lead  to  large  phase-to-phase  currents  in  the  earth,  from  the \nsecond  fault  to  the  first,  or  to  a  ground  intentionally  placed  on  the \nphase  first  involved,  in  order  to  short  circuit  the  arc.  The  inductive \neffects  thus  become  electromagnetic  in  character,  and  the  inter- \nference produced  in  this  manner  may  be  severe.\n\n'\"Electrically\"  is  used  here  and  elsewhere  in  this  paper  in  the  sense  in  which \n\"electrostatically\"  is  perhaps  more  commonly  used.  The  phenomena  involved \nare  not  static  and  the  latter  word  is  inappropriate  on  this  account.\n\novervoltage  on  a  sound  phase  at  the  instant  of  grounding  than  would \nbe  the  case  in  an  isolated  system  is,  of  course,  of  importance  in  this \nconnection.  This  has  been  examined  from  a  theoretical  standpoint \nin  the  second  subdivision  of  this  section,  with  the  conclusion  that \nthere  is  no  material  difference  in  this  respect.\n\nThere  remains  the  method  of  grounding  in  which  a  high  resistance \nis  employed  in  the  connection  from  neutral  to  ground.  By  \"high  \"  here \nmay  be  meant  the  \"critical\"^  resistance  or  one  of  smaller  magnitude, \nbut  still  so  large  that  in  the  event  of  a  solidly  grounded  phase,  the \nsound  phases  are  brought  to  subtsantially  full  delta  voltage  above \nground.  There  are  probably  few  cases  where  electromagnetic  indu- \ntive  effects  due  to  accidental  grounds  on  a  power  system  are  a  matter \nof  importance,  in  which  a  neutral  resistance  small  enough  to  avoid \nthis  rise  of  voltage  on  the  sound  phases  would  be  effective  as  a  meas- \nure of  relief.  This  method  of  grounding  would  thus  not  avoid  the \nelectrically  induced  voltages  which  arise  when  the  reactor  is  used, \nalthough  it  would  presumably  be  effective  in  preventing  the  spread \nof  trouble  to  other  parts  of  the  power  system  if  positive  operation  of \nselective  relays  is  secured.  Inasmuch  as  it  presents  fewer  difficulties \nfrom  this  last  point  of  view  than  the  Petersen  reactor,  grounding \nthrough  a  moderate  resistance  has  a  definite  advantage  over  the  latter \nmethod  from  the  standpoint  of  inductive  effects  at  fundamental \nfrequency,  provided  sufficient  resistance  can  be  used  to  limit  the \nelectromagnetically  induced  voltages  to  tolerable  values.\n\nWhere  this  is  impracticable  from  the  standpoint  of  power  system \noperation,  the  relative  merits  of  the  two  systems  would  have  to  be \ndecided  by  balancing  the  effective  suppression  of  the  transient  electro- \nmagnetic inductive  effects  by  means  of  the  reactor,  plus  the  expecta- \ntion of  occasional  disturbances  continuing  over  the  intervals  neces- \nsary for  the  location  and  disconnection  of  the  faulty  line,  against  the \nimperfect  suppression  of  the  former  effects  by  means  of  the  resistor, \nplus  the  limitation  to  very  brief  intervals  of  electrically  induced  dis- \nturbances otherwise  the  same  as  with  the  reactor.  It  is  obvious  that \nthe  factors  controlling  such  a  decision  would  vary  widely  in  different \ncases,  and  that  practical  experience  with  both  methods  would  be  of \ngreat  value  in  estimating  their  relative  importance.  It  is,  of  course, \npossible  that  future  development  may  remove  some  of  the  disad- \nvantage at  which  the  Petersen  coil  now  finds  itself  in  respect  to  the \nmatter  of  relay  protection  for  interconnected  networks.  Such  de- \nvelopment would  presumably  be  of  importance  also  to  the  critical\n\n*  Le.,  the  resistance  for  which  a  discharge  to  ground  passes  from  the  oscillatory \nto  the  non-oscillatory  type.\n\nresistance  which,  as  a  method  of  grounding  the  neutral,  would  gener- \nally suffice  to  prevent  interference  from  ground  currents  at  times  of \nfaults  to  ground,  but  which  apparently  presents  difficulties  from  the \nrelay  standpoint  similar  to  those  involved  in  the  use  of  the  Petersen \ncoil.\n\nReferring  to  Fig.  5  (we  now  take  account  of  inequalities  in  the  ad- \nmittances to  ground),\n\nFig.   5 \u2014 Three-phase  System  with   Neutral   Grounded  Through   Reactor.     Admit- \ntances to  Ground  Not  Balanced.\n\nwhere     F  =   Fi  +  F2  +  F3  =  total    direct    admittance    to    ground. \nIf  the  impressed  voltages  are  balanced\n\nThe  parenthesis  is  the  \"  residual  admittance  \"  ^  to  ground  and,  if \nthe    three    leakances    to    ground    are    equal,    it    can    vanish    only   if\n\n'  Inductive  Interference  between  Power  and  Communication  Circuits,  California \nRailroad  Commission,  p.  269.\n\nthe  three  direct  capacities  to  ground  are  the  same.  The  equation  is \nequivalent  to\n\nwhere  E^^  is  the  \"  characteristic  residual  voltage  \"  \u00ae  of  the  (isolated) \nsystem  and  is  equal  in  magnitude  to  3  \u00a3oi  times  the  ratio  of  the  residual \nadmittance  to  ground  to  the  total  admittance  to  ground. \nWe  have\n\nr\u201e  and  L\u201e  being  the  resistance  and  inductance  of  the  earth  coil,  and \ng  and  C  the  total  leakance  and  capacity  to  ground,  respectively. \nAlso,  for  resonance\n\nIf  in  this  equation  the  denominator  on  the  right  be  denoted  by  x \nand   the  fractional  unbalance   of   the   admittance    to   ground   by  y\n\nIf  the  losses  in  the  system  (including  the  earth  coil)  are  small,  x  is \nsmall  compared  to  1.     Thus  we  get,  for  the  absolute  value  of  V\n\nand  the  absolute  value  of  the  residual  voltage  is  three  times  this, \napproximately.\n\nIn  substance,  this  means  that,  at  fundamental  frequency  and  with \nsmall  losses,  the  fractional  admittance  unbalance  should  be  kept  small \ncompared  to  the  ratio  of  the  resistance  to  the  reactance  of  the  coil,  if \nunduly  high  voltages  to  ground  are  to  be  avoided.  (It  is  supposed \nthat  g'oiC  is  of  the  order  of  r\u201e/wL\u201e,  or  smaller \u2014 a  condition  probably \nsatisfied  even  with  the  best  practiable  design  of  coil,  except  under \nvery  wet  line  conditions.)  The  point  involved  here  is,  of  course,  an \nimportant  one  from  the  standpoint  of  power  system  operation.  It \nis  also  important  from  the  standpoint  of  electrically  induced  voltages \nin  exposed  communication  circuits.  The  admittance  unbalance  can \nbe  kept  within  the  necessary  limits  by  suitable  power  circuit  trans- \npositions.\n\nThe  absolute  magnitude  of  the  fundamental  frequency  neutral \ncurrent  is  obtained  from  the  expression  for  ]  F\u201e  ]  by  dividing  by  the \ncoil  impedance  (or  directly  from  equation  (3),  and  is,  approximately,\n\nFor  a  system  with  dead-grounded  neutral  the  fundamental  frequency \nresidual  current  is  yEoiY  and  is  thus  smaller  than  that  just  found  for \nthe  case  of  the  reactor  in  the  ratio  of  j  x  |  to  1,  approximately.  It  is \nevident,  how^ever,  that  the  magnitude  of  the  neutral  current  with \nreactor  is  controllable  by  means  of  transpositions,  as  in  the  case  of \nthe  neutral  and  residual  voltages.  The  inductive  effects  of  this \ncurrent  should  be  of  small  consequence  with  an  amount  of  transposing \nsufficient  to  keep  the  line  voltages  to  ground  within  limits  desirable \nfrom  the  standpoint  of  power  system  operation.\n\nIn  the  following  discussion,  we  retain  the  assumption  of  lumped \nconstants,  so  that  the  results  are  not  applicable  to  extensive  networks \nwithout  modification.\n\nWith  this  restriction,  at  harmonic  frequencies  other  than  the  third \nor  one  of  its  odd  multiples,  the  above  approximate  equation  for  V\u201e \nbecomes\n\nm  being  the  order  and  \u00a3oi  the  voltage  of  the  harmonic  and  x  and  y \naccented  to  denote  that  they  are  to  be  taken  for  the  frequency  in\n\nquestion.  For  small  losses  x'  may  be  neglected  in  comparison  with \nunity,  as  before,  giving\n\nThus,  even  for  the  fifth  harmonic,  the  right  hand  side  of  (5)  is  only \n4  per  cent,  in  excess  of  the  value  it  would  have  if  the  neutral  were \nisolated.\n\nFor  harmonics  whose  orders  are  not  divisible  by  three  the  residual \nvoltage  is  three  times  F\u201e.  Thus  from  the  standpoint  of  noise  inter- \nference from  voltages,  a  system  grounded  through  a  Petersen  coil \nbehaves  practically  as  though  the  neutral  were  isolated,  so  far  as  these \nharmonics  are  concerned.  As  with  the  fundamental,  power  circuit \ntranspositions  are  available  for  the  reduction  of  residual  voltages  of \nthese  frequencies.\n\nResidual  currents  of  frequencies  belonging  to  this  series  of  har- \nmonics, which  are  not  present  at  the  ends  of  the  line  with  isolated \nneutral,  are  introduced  by  grounding  through  the  reactor,  but  they \nare  of  minor  importance,  as  may  be  judged  by  comparing  the  neutral \ncurrents  with  the  reactor  and  with  dead  grounded  neutral.  With  the \nreactor,  the  neutral  current  of  a  harmonic  of  order  m  not  a  multiple \nof  3  is  found  from  (5)  to  be,  in  absolute  value,\n\nZ\u201e  being  the  coil  impedance  at  fundamental  frequency,  while  with \ndead-grounded  neutral,  it  would  be\n\nin  which  Y  is  the  total  admittance  to  ground  at  fundamental  fre- \nquency. Thus  the  magnitude  of  the  neutral  current  with  the  reactor \nis  approximately  l/(m^  \u2014  1)  of  its  magnitude  with  dead-grounded \nneutral.  The  noise  effects  of  residual  currents  of  these  magnitudes \nwill  generally  be  insignificant  compared  to  those  arising  from  other \nsources,  particularly  if  the  power  circuit  capacities  to  ground  are  well \nbalanced.\n\nin  which  the  symbols  for  coil  impedance  and  line  admittance  are \naccented  to  denote  that  they  refer  to  the  harmonic  freciuency  in \nquestion;\n\nThe  neutral  is  thus  subjected  to  a  third  harmonic  voltage  some  12 \nper  cent,  greater  than  if  it  were  isolated,  but  for  the  higher  harmonics \nbelonging  to  this  series  (of  the  third  and  its  odd  multiples),  the  differ- \nence is  inappreciable.\n\nFrom  the  standpoint  of  noise  interference  in  telephone  circuits, \nresiduals  of  the  series  consisting  of  the  third  harmonic  and  its  odd \nmultiples  are  frequently  troublesome  where  the  neutral  of  a  three- \nphase  system  is  grounded  directly  or  through  a  low  resistance.  These \nresiduals,  of  course,  are  not  affected  by  power  circuit  transpositions, \neither  as  to  their  magnitudes  or  as  to  their  inductive  effects  upon \nexposed  telephone  circuits.  It  is  therefore  of  interest  to  examine  xhe \nexpressions  just  obtained  for  the  case  in  which  the  neutral  is  grounded \nthrough  the  coil.     While  the  neutral  current  will  not  be  the  same  as\n\nthe  residual  current  except  when  only  one  line  is  supplied  from  the \ntransformer  bank,  the  effect  upon  the  former  should  in  general  be  at \nleast  approximately  proportional  to  the  effect  upon  the  residual  cur- \nrent in  any  line  supplied  from  the  bank.\n\nIn  writing  equation  (7),  any  difference  between  the  induced  voltage \njEoi  and  the  voltage  appearing  between  line  and  neutral  has  been \nignored.  To  the  extent  that  this  is  justifiable,  the  expressions  for  the \ncase  of  the  solidly  grounded  neutral  may  be  obtained  by  making  the \ndenominators  of  the  right  hand  sides  of  (8)  and  (9)  each  unity.  If \nthe  transformer  bank  is  provided  with  a  delta  winding  of  low  im- \npedance, in  particular  if  it  is  connected  delta  on  one  side,  this  pro- \ncedure gives  a  fair  approximation  to  the  correct  expressions,  since \nthe  impedance  through  which  the  voltage  \u00a3oi  regulates  is  in  this \ncase  merely  the  transformer  leakage  impedance.  The  resulting \nconclusions  with  respect  to  the  advantage  of  the  reactor \u2014 for  ex- \nample, that  the  third  harmonic  residual  voltage  or  neutral  current  is \n1/8  as  large,  the  ninth  1/80  as  large,  etc.,  with  the  reactor  as  with \nsolidly  grounded  neutral \u2014 should  not,  in  any  event,  be  unfavorable  to \nthe  latter  method  of  grounding  unless  the  electrical  length  of  the  line \napproaches  the  point  at  which  its  reactance  to  ground  becomes \npositive.\n\nIf  the  transformers  are  so  connected  as  to  provide  no  path  for  triple \nharmonic  magnetizing  currents  other  than  through  line  admittance \nto  ground  and  the  impedance  between  neutral  and  ground,  the  in- \nduced voltage  jEoi  is  not  the  same  for  the  two  methods  of  grounding \nunder  consideration,  because  the  impedance  to  the  triple  harmonic \nmagnetizing  currents  is  appreciably  different  in  two  cases.  A  con- \nvenient method  of  taking  this  effect  into  account  is  to  regard  the  in- \nduced voltage  as  due  to  a  fictitious  impedanceless  generator  of  de- \nterminate voltage  regulating  through  the  mutual  impedance  of  the \ntransformer  windings  for  the  frequency  in  question.''  If  Z'^  is  one- \nthird  of  this  mutual  impedance  and  Vo\\^  is  the  voltage  of  the  fictitious \ngenerator,  the  expression  for  the  neutral  current  with  ground  con- \nnection through  the  reactor  becomes\n\n'The  voltage  thus  assumed  is,  of  course,  not  identical  with  the  induced  voltage \nfor  which  the  symbol  Eoi  has  hitherto  been  used,  and  for  this  reason  the  new  symbol \nFoi  is  used  for  it.  The  corresponding  7?oi  would  be  Foi  diminished  by  the  drop  through \nthe  mutual  impedance.\n\nThe  corresponding  expressions  for  solidly  grounded  neutral  are  ob- \ntained by  omitting  m-  in  the  denominator  for  each  of  the  equations \njust  derived.  Thus  the  advantage  of  grounding  through  the  reactor \nrelative  to  grounding  directly  depends  on  the  magnitude  of  F'Z'^  as \ncompared  to  the  square  of  the  order  of  the  harmonic.  Z'\u201e  depends \nupon  the  voltage  and  the  kva.  capacity  of  the  transformers  and  is \nmostly  inductive  reactance.  For  high  voltage  transformer  banks \nof  small  capacity  feeding  very  extensive  networks,  the  gain  indicated \nby  equations  (10)  and  (11)  from  the  use  of\"  the  reactor  would  prob- \nably not  be  large.  It  would  be  important,  however,  where  the  aggre- \ngate capacity  of  the  supply  transformers  is  moderate  or  large  and  the \nconnected  network  is  of  moderate  extent  and  voltage.  For  instance, \nusing  the  data  of  the  example  considered  in  an  earlier  part  of  this\n\nfor  a  total  transformer  capacity  of  7,000  to  8,000  kva.,  with  line  voltage \nfrom  20,000  to  30,000,  we  should  have  L'^Coi'^  equal  to  about  4  at \n180  cycles/sec.  In  other  words,  in  this  case,  the  employment  of  the \nreactor  would  reduce  the  residual  voltage  and  the  neutral  current  of \nthe  third  harmonic  frequency  due  to  a  star-star  solidly  grounded \ntransformer  bank  by  about  75  per  cent.,  and  residuals  of  other  fre- \nquencies belonging  to  the  same  series  probably  by  larger  amounts.\n\nIn  the  earlier  discussion  relating  to  harmonics  not  belonging  to  the \ntriple  series,  comparison  was  made  between  a  system  grounded  through \na  Petersen  reactor  and  the  isolated  system.  In  a  similar  comparison \nwith  respect  to  the  triple  harmonic  series,  the  isolated  system  has  the \nadvantage,  since  residuals  of  this  series  theoretically  do  not  appear \nin  such  a  system,  as  the  voltages  are  not  impressed  between  wires. \nAs  a  practical  matter,  an  isolated  system  would  probably  not  be \nentirely  free  of  triple  harmonic  residuals,  owing  to  dissimilarities  in \ntransformers  or  elsewhere.  Such  accidental  effects  can  hardly  be \ntaken  into  account  in  a  theoretical  discussion.  However,  in  setting \nup  a  comparison  between  the  isolated  system  and  that  grounded \nthrough  the  reactor,  an  idea  of  the  relative  importance  of  the  triple \nharmonic  residual  voltages  existing  in  the  latter  case  can  perhaps  be \nobtained  by  comparing  their  theoretical  magnitudes  with  the  theoreti-\n\ncal  magnitudes  of  residual  voltages  in  the  isolated  system  of  non- \ntriple  frequencies.\n\nThe  residual  voltage  due  to  one  of  these  non-triple  frequencies, \nwhich  is  three  times  the  neutral  voltage,  is  ^y'  E'oi,  according  to \nequation  (6).  Here  y'  is  the  fractional  residual  admittance  and  E'oi \nmay  be  taken  as  the  induced  voltage  in  the  transformer  for  the  fre- \nquency in  question.  For  a  harmonic  belonging  to  the  triple  series, \nwith  neutral  grounded  through  the  reactor,  the  absolute  value  of  the\n\nthe  harmonic  and  Eqi  the  induced  voltage,  if  we  assume  the  trans- \nformer bank  provided  with  a  low  impedance  path  for  triple  harmonics, \nand  therefore  neglect  the  difference  between  the  induced  and  the \nterminal  voltages.     The  ratio  of  the  triple-series  residual  voltage  to\n\nIf  we  take  the  ninth  as  the  harmonic  of  the  triple  series  and  assume \nequal  values  of  the  induced  voltages  \u00a301  and  Fq'i  it  will  be  seen  that \nI  y'  I  must  be  of  the  order  of  0;01  if  the  residual  voltage  of  the  triple \nharmonic  series  is  to  be  as  large  as  the  other.  This  amount  of  un- \nbalance is  somewhat  larger  than  has  been  found  at  this  frequency \n(540  cycles/sec.)  in  an  actual  transposed  line.^  If  we  consider  the \nhigher  harmonics  of  the  triple  series,  |  y'  \\  would  have  to  be  made \nprogressively  smaller  in  order  that  the  ratio  might  remain  unity. \nThus,  for  the  21st  harmonic,  |  y'  \\  would  have  to  be  of  the  order  of \n0.002.  While,  of  course,  |  y'  \\  may  be  made  as  small  as  desired  by \nsufficiently  close  power  circuit  transpositions,  it  appears  that  in  prac- \ntical cases  where  transformer  banks  have  delta  windings,  one  may \nexpect  the  residual  voltages  of  the  triple  series,  introduced  by  changing \nfrom  an  isolated  system  to  one  grounded  through  the  reactor,  to  be \nrelatively  unimportant  except  in  the  case  of  the  third  harmonic  and \nperhaps  in  that  of  the  ninth.  This  statement  would  not  be  true  if, \nas  with  star-star  transformers  under  some  circumstances,  no  low \nimpedance  path  is  provided  for  magnetizing  currents  of  the \ntriple  harmonic  series.  Such  cases  are  not  common  in  operating \npractice.\n\nThe  method  of  estimating  comparative  effects  here  applied  to  the \ncase  of  triple  harmonic  residual  voltages  is  not  available  for  residual \ncurrents.     To  take  account  of  the  latter  in  comparing  the  isolated\n\n\u2022  Inductive  Interference  between  Power  and  Communication  Circuits,  California \nRailroad  Commission,  Technical  Report  No.  51.\n\nsystem  and  the  system  with  neutral  ^rouncled  through  the  reactor, \nrecourse  may  be  had  to  the  indirect  method  of  reference  to  the  solidly \ngrounded  neutral  system,  as  in  the  discussion  of  residual  currents  of \nthe  non-triple  series  on  page  52.  Such  a  procedure,  of  course,  in- \nvolves a  reference  to  general  experience  also.  It  has  been  shown  in \nthe  earlier  discussion  that  for  a  triple  series  harmonic  of  order  m \nthe  neutral  current  wMth  the  reactor  is  approximately  1/  {m^  \u2014  1) \nas  large  as  with  the  dead-grounded  neutral  if  a  low  impedance  path \nfor  triple  frequency  magnetizing  currents  is  provided,  as  by  a  delta \nwinding.  The  establishment  of  this  system  of  neutral  currents,  even \nthough  they  are  small,  when  a  previously  isolated  system  is  grounded \nthrough  a  Petersen  reactor,  constitutes  an  addition  to  the  residuals \nwhich  produce  induction  in  neighboring  circuits.  However,  it  is  not \nto  be  expected  that  the  added  inductive  effects  would  be  important. \nWhere  no  low  impedance  path  for  the  triple  series  magnetizing  cur- \nrents exists,  the  reactor  is  relatively  less  effective  in  suppressing \nresidual  currents  of  this  series.  The  triple  harmonic  neutral  currents \nof  a  power  system  connected  in  this  manner  and  grounded  through  a \nPetersen  coil  might  in  some  cases  lead  to  inductive  effects  of  some \nsignificance.\n\nIn  general,  for  harmonics  of  orders  not  divisible  by  three,  grounding \nthrough  a  moderate  resistance  (large,  however,  compared  to  other \nimpedances  involved  in  a  short  circuit  to  ground)  will  be  more  ad- \nvantageous as  regards  residual  voltages,  and  less  advantageous  as \nregards  residual  currents,  than  grounding  through  the  reactor. \nGrounding  through  zero  impedance  would,  of  course,  generally  lead \nto  the  smallest  residual  voltages  and  the  largest  residual  currents  of \nthese  frequencies.  For  frequencies  belonging  to  the  tr-iple  series, \ngrounding  through  the  reactor  will  be  considerably  more  advantageous \nthan  grounding  through  a  moderate  resistance  as  regards  both  residual \nvoltages  and  residual  currents.  It  may  be  expected  that  with  moder- \nate neutral  resistance,  residual  currents  and  voltages  of  the  triple \nseries  will  both  be  nearer  in  magnitude  to  those  obtaining  with  zero \nneutral  impedance  than  to  those  obtaining  with  the  Petersen  coil. \nThe  moderate  neutral  resistance  is  relatively  more  effective  at  the \nhigher  frequencies  in  reducing  residual  currents  of  all  harmonics \nand  residual  voltages  of  the  triple  series;  for  harmonic  residual \nvoltages  not  belonging  to  the  triple  series,  it  is  relatively  more  effective \nat  the  lower  frequencies.\n\nI  wish  to  express  my  gratitude  for  helpful  suggestions  and  criticism \nreceived  in  the  preparation  of  this  paper  from  Messrs.  L.  P.  Ferris \nand  R.  G.  McCurdy,  and  also  from  Mr.  R.  K.  Honaman.\n\n1.  At  times  of  a  fault  to  ground  on  a  power  system  with  neutral \ngrounded  through  a  Petersen  reactor,  the  action  of  the  latter  tends  to \nextinguish  the  arc  and  to  prevent  its  restriking.  Theoretical  con- \nsiderations, applied  to  a  practical  case,  indicate  that  the  transient \nover-voltage  on  a  sound  phase  at  the  instant  of  occurrence  of  the \nfault  is  substantially  the  same  as  in  a  system  with  isolated  neutral.\n\n2.  Grounding  the  neutral  through  a  Petersen  earth  coil  instead  of \ndirectly  or  through  a  low  resistance  would  largely  prevent  the  electro- \nmagnetic inductive  effects  to  which  exposed  communication  circuits \nare  liable  at  times  of  faults  to  ground  in  systems  grounded  in  the \nlatter  manner.  (Extensive  high  voltage  networks  are  perhaps  an \nexception  to  this  statement.  But  even  here,  the  electromagnetic \ninductive  effects  would  in  general  not  be  greater  with  the  reactor \nthan  with  isolated  neutral.)  However,  effects  due  to  electric  induc- \ntion similar  to  those  from  an  isolated  system  may  be  expected  to \nappear.  Except  for  long,  close  parallels  involving  open-wire  com- \nmunication circuits  these  effects  should  in  general  be  much  less  severe \nthan  the  electromagnetic  inductive  effects  from  a  system  with  dead- \ngrounded  neutral.  The  extent  and  severity  of  the  inductive  effects \nexperienced  from  the  system  grounded  through  the  reactor  would \nfurther  tend  to  be  smaller  than  with  the  isolated  system,  because  of \nthe  effect  of  the  reactor  in  preventing  arcing  grounds.\n\n3.  Grounding  the  neutral  through  a  resistance  large  compared  to \nother  impedances  involved  in  a  short  circuit  to  ground  should  have \nan  advantage  over  grounding  through  the  Petersen  reactor,  in  that \nthe  former  method  presents  fewer  difficulties  in  respect  to  power \nsystem  protective  relays,  so  that  it  would  reduce  the  possibility  of  the \ncontinuance  of  inductive  disturbances  over  considerable  periods  of \ntime,  which  might  be  involved  in  grounding  through  the  reactor, \nunder  present  relay  practice.  From  an  inductive  interference  stand- \npoint, a  choice  between  the  two  methods  would  depend  upon  the \ncircumstances  of  particular  cases.  Advances  in  the  art  of  relay \nprotection  would  improve  the  position  of  the  reactor  in  such  con- \nsiderations.\n\n4.  Under  normal  power  system  operating  conditions,  the  use  of  the \nreactor  may  lead  to  excessive  residual  voltages  of  fundamental  fre- \nquency if  the  admittamces  from  phases  to  ground  are  unbalanced. \nSuch  unbalance  may  be  reduced  to  the  extent  necessary  from  this \npoint  of  view  by  power  circuit  transpositions.\n\n5.  Under  normal  operating  conditions,  it  is  to  be  expected  that  the \nresidual  voltages  and  currents  of  the  triple  harmonic  series  occurring\n\nwith  neutral  grounded  through  zero  or  a  low  impedance  would  he \nlargely  reduced  by  grounding  through  the  Petersen  reactor  instead. \nResiduals  of  other  harmonic  frequencies  should  be  substantially  the \nsame  as  with  isolated  neutral,  and  are  controllable  by  means  of  power \ncircuit  transpositions.  The  method  of  grounding  the  neutral  through \na  resistance  of  moderate  value  is  favorable  to  the  reduction  of  residual \nvoltages  of  the  harmonics  whose  orders  are  not  multiples  of  three, \nbut  is  relatively  unfavorable  to  the  suppression  of  residual  currents \nof  these  frequencies.  It  is  also  considerably  less  effective  than  the \nreactor  in  preventing  residuals,  either  voltages  or  currents,  belong- \ning to  the  triple  harmonic  series,  which  are  not  amenable  to  treat- \nment by  transpositions.\n\nSynopsis:  Engineering  and  construction  features  involved  in  a  complete \ntelephone  cable  system  over  300  miles  in  length  and  connecting  Philadelphia \nand  Pittsburgh,  Pa.,  are  described  in  the  following  paper.  This  cable  is \ndesigned  to  operate  as  an  extension  of  the  Boston-Washington  under- \nground cable  system  with  which  it  connects  at  Philadelphia.  It  is  also \ndesigned  for  operation  in  connection  with  the  Pittsburgh-Chicago  cable \nnow  under  construction,  and  other  cable  projects  included  in  a  compre- \nhensive fundarnpntal  plan.\n\nBeginning  with  the  fundamental  factor  of  public  requirements  for  com- \nmunication service  between  cities  separated  by  various  distances,  there  are \nnext  considered  the  methods  available  to  provide  this  service.  Small- \ngage,  quadded,  aerial  cable,  which  was  decided  upon  for  use  in  this  section \nafter  careful  economic  studies,  is  described  in  a  general  way  and  the  im- \nportant advantages  of  the  application  of  loading  and  telephone  repeaters  are \noutlined.  The  use,  in  connection  with  this  cable,  of  the  recently  developed \nmetallic  telegraph  system  for  cables  is  referred  to  and  some  facts  are  given \nregarding  power  plants,  test  boards  and  buildings.  A  few  of  the  many \npossible  combinations  of  cable  and  equipment  facilities  into  complete \ntelephone  circuits,  which  will  furnish  the  service  required  as  economically \nas  now  possible,  are  illustrated.\n\nThe  necessity  of  complete  coordination  of  the  many  factors  involved \nin  a  project  of  this  kind  is  emphasized.\n\nTHE  placing  in  service  in  the  latter  part  of  1921  of  the  final  section \nof  a  continuous  telephone  cable  over  300  miles  in  length  between \nPhiladelphia  and  Pittsburgh  marked  a  new  point  of  achievement \nin  the  steady  development  and  construction  of  facilities  designed \nto  render  to  the  public  the  best  possible  long-distance  telephone \nservice.  Furthermore,  this  cable  forms  an  important  part  of  a  com- \nprehensive plan  of  long-distance  cable  construction  throughout  that \nsection  of  the  United  States  lying  in  general  east  of  the  Mississippi \nRiver  and  north  of  the  Ohio  and  Potomac  Rivers.\n\nIn  the  discussion  of  a  project  of  this  kind  which  involves  many  new \npractices  and  the  expenditure  of  several  millions  of  dollars  and  which, \nwith  related  work  already  completed,  forms  the  groundwork  for \nlarge  expenditures  in  the  future,  it  is  usual  to  inquire  first  into  the \nunderlying  reasons  for  carrying  out  the  project  and  then  into  the \nmethods  adopted.  In  the  following  discussion  an  endeavor  will \ntherefore  be  made  to  furnish  some  information  on  these  two  items \nin  their  relation  to  the  Philadelphia-Pittsburgh  cable,  although,  as \nwill  be  obvious,  the  many  different  points  can  be  covered  in  only\n\n'  Presented  at  a  meeting  of  the  Philadelphia  Section  of  the  A.  I.  E.  E.,  January  9, \n1922,  presented  at  the  Annual  Convention  of  the  A.  I.  E.  E.,  Niagara  Falls,  Ont., \nJune  26-30,  1922,  and  appearing  in  the  Journal  of  the  A.  I.  E.  E.  for  August,  1922.\n\nthe  most  general  way  in  the  space  available.  However,  before  going \nahead  with  the  discussion,  I  would  like  to  point  out  that  this  project \nis  not  unlike  many  others  in  that,  as  a  whole  and  in  the  component \nparts,  there  have  been  required,  first,  the  careful  consideration  and \ndecisions  of  the  executives,  then  the  underlying  work  of  many  scien- \ntists, inventors  and  engineers,  then  the  skilled  work  of  the  manu- \nfacturers and  construction  forces,  and  finally  the  maintenance  and \noperation  by  trained  people  who  are  responsible  for  the  continuous \nservice  so  vitally  necessary  to  the  industrial  and  social  structure  of \nthe  country.  The  point  to  be  emphasized  here  is  that  the  coordina- \ntion of  all  of  these  factors  and  the  close  cooperation  of  all  of  the  many \nhundreds  of  people  concerned  are  the  important  things.\n\nFig.  1  is  an  outline  map  of  a  section  of  the  United  States  and  shows \nthe  routes  of  existing  and  proposed  long  telephone  cables  of  the  Bell \nsystem.  It  will  be  noted  that  the  present  and  proposed  routes  follow \nin  a  general  way  the  routes  of  trunk-line  railroads.  This  general \nsection  contains  m.ore  than  50  per  cent  of  the  entire  population  of \nthe  United  States  but  less  than  15  per  cent  of  the  area,  and  the  in- \ndustrial and  telephone  development  is,  of  course,  very  great.  Fur- \nthermore, the  nearby  surrounding  states,  supplying  as  they  do  large \nquantities  of  food  products  and  raw  materials,  are  commercially \nrelated  to  this  section  in  a  very  peculiar  way  and  this  fact  greatly \ninfluences  the  long-distance  telephone  development  along  the  par- \nticular cable  routes  indicated.  The  routes  through  the  State  of \nPennsylvania  and  the  offices  at  Philadelphia  and  Pittsburgh,  which  are \nthe  terminals  of  the  cable  that  is  more  particularly  the  subject  of \nthis  discussion,  occupy  strategic  positions  in  this  system.\n\nCircuits  of  the  American  Telephone  and  Telegraph  Company  and \nthe  Bell  Telephone  Company  of  Pennsylvania  are  carried  over  these \nroutes  and  this  cable  was  jointly  planned  and  installed  by  these \ncompanies.\n\nFig.  2  is  an  outline  map  of  the  State  of  Pennsylvania  and  shows \nthe  situation  in  this  section  a  little  more  in  detail.  On  this  map  are \nshown  some  of  the  larger  cities  and  routes  of  the  longer  and  more \nimportant  toll  and  long-distance  telephone  lines.  As  indicated, \nthese  lines  are  mainly  of  the  familiar  aerial  wire  type  which  has  been \ngenerally  used  in  the  past  for  this  purpose  and  which  is  today  the \nmost  efficient  and  economical  type  of  construction  for  many  cases. \nIn    the    general    section    between    Philadelphia    and    Pittsburgh    the\n\nrequirements  for  circuits  are  very  heavy  and  in  addition,  as  is  well- \nknown,  the  topography  of  the  country  is  such  that  the  through  routes \nwhich  can  economically  be  used  for  pole  lines  are  limited.  At  present, \nthese  few  routes  are  fully  occupied  by  the  pole  lines  of  the  various \nutilities  and  included  in  these  lines  are  three  fully  loaded  telephone \ntrunk  lines.  Another  item  of  importance  in  the  consideration  of \naerial  wire  construction  is  the  very  severe  damage  frequently  experi- \nenced in  many  sections  of  the  country  on  heavy  aerial  wire  lines \nfrom  ice  and  wind  storms.  Even  lines  built  with  exceptional  strength \nfail  in  these  storms  and  the  interruptions  to  service  are  serious  mat- \nters to  the  users  as  well  as  to  the  telephone  companies.     The  restora-\n\nFig.  3 \u2014 Damage  to  Section  of  New  York-Boston  Main  Line  Near  Worcester,  Mass. \nStorm  of  November  28,  1921\n\ntion  costs  under  the  conditions  that  naturally  exist  at  such  times  are \nabnormally  high.\n\nFigs.  3  and  4  show  the  effects  at  one  point  of  the  ice  and  wind  storm \nin  New  England  on  November  28,  1921,  and  are  proof  that  this \nproblem  is  real.  This  particular  spot  is  near  Worcester,  Mass.,  and \nthe  line  is  a  section  of  one  of  the  principal  aerial  wire  routes  between \nNew  York  and  Boston.  In  this  storm,  many  thousands  of  poles \nwere  broken  and  even  where  a  few  poles  remained  standing  due  to \nspecially  strong  construction,  the  load  of  ice  combined  with  the  wind \nwas  too  great  for  the  wires  to  withstand.  There  is  therefore  a  prac- \ntical limit  to  the  number  of  wires  that  can  be  safely  and  economically \ncarried  on  a  pole  line.\n\nexisting  pole  lines  are  fully  loaded,  and  where  estimated  future  circuit \nrequirements  are  of  considerable  magnitude,  it  is  obvious  that  different \nmethods  of  providing  facilities,  if  available,  must  sooner  or  later \nbe  given  serious  consideration.  The  conditions  between  Phila- \ndelphia and  Pittsburgh  and  in  general  along  all  of  the  cable  routes \nshown  on  Fig.  1  are  now,  or  are  expected  within  a  few  years  to  be, \nsuch  as  to  make  the  use  of  some  type  of  construction  other  than  aerial \nwire  desirable  for  most  of  the  circuits.\n\nFig.  4 \u2014 Section  of  New  York-Boston  Main  Line  Showing  Wires  Heavily  Loaded  with\n\nand  of  the  methods  available  for  providing  long-distance  telephone \nfacilities,  which  in  general  are  aerial  wire  and  cable,  it  has  been  decided \nthat  for  relief  in  these  sections  the  cable  method  will  give  the  best \nand  most  economical  results.  Long  underground  cables,  as  is  well- \nknown,  have  been  in  operation  for  many  years  between  Boston,  New \nYork,  Philadelphia,  Baltimore  and  Washington,  Chicago  and  Mil- \nwaukee and  in  other  sections.  However,  the  type  of  cable  and  asso- \nciated apparatus  which  is  now  being  used  in  the  development  of  the \nmore  comprehensive  plan  is  quite  different  from  that  originally  used \nbetween  Boston  and  Washington  and  in  the  other  sections,  particu- \nlarly in  the  use  of  copper  conductors  of  a  smaller  gage  combined \nwith  improved  loading  coils,   the  vacuum  tube  telephone  repeater\n\nand  other  methods  and  apparatus  which  are  the  result  of  recent \ndevelopments.  Lead-covered  aerial  cable  supported  on  wooden  pole \nlines  is  to  be  used  in  general  on  all  of  the  routes  except  in  the  two \nsections  just  mentioned  and  through  cities  or  where  special  condi- \ntions exist  for  short  distances.  The  possibility  of  now  using  con- \nductors of  No.  16  and  No.  19  A.  W.  G.  instead  of  conductors  up  to\n\nNo.  10  A.  W.  G.  as  in  the  older  cables,  has  contributed  to  make  aerial \nconstruction  rather  than  underground  conduit  the  more  economical \nin  many  sections,  as  one  cable  will  provide  for  a  much  greater  number \nof  circuits  and  consequently  fewer  cables  will  be  required.\n\nThe  general  type  of  aerial  construction  which  was  used  for  over \n250  miles  of  the  total  distance  of  302  miles  from  Philadelphia  to  Pitts- \nburgh may  be  seen  from  Figs.  5  and  6  which  illustrate  the  poles,  steel \nsuspension  strand,  metal  supporting  rings  and  the  cable.  The  poles \nare  25-foot  untreated  chestnut  spaced  100  feet  apart  and  designed \nto  carry  additional  cables  in  the  future.  While  the  poles  are  new \nand  carry  only  one  cable  they  have  a  factor  of  safety  of  about  9  under \nthe  most  severe  storm  conditions  expected,  but  this  will,  of  course,  be \nreduced  as  other  cables  are  placed  and  will  gradually  be  decreased \non  account  of  decay  at  the  ground  line  until  it  becomes  necessary  to \nstart  replacing  the  poles.  Many  of  these  poles  were  grown  near  the \nlocations  where  they  now  stand.     In  other  sections,   it  is  planned\n\nto  use  butt-treated  chestnut  or  cedar  poles,  or  creosoted  pine  poles \nwhere  these  prove  to  be  the  more  economical.\n\nThe  galvanized  steel  suspension  strand  has  a  breaking  strength \nof  about  16,000  pounds  and  the  actual  tension  under  normal  condi- \ntions is  about  7,000  pounds.  In  placing  the  strand,  it  is  necessary \nto  pull  it  to  just  the  right  tension  in  order  that  when  the  cable  is \nhung  it  will  have  the  proper  sag.  The  correct  tension  is  readily \ndetermined  by  what  is  known  as  the  \"oscillation\"  method.  The \nmetal  rings  are  spaced  16  inches  apart  and  the  cable  weighs  about \n7>^  pounds  per  foot.\n\nThe  size  and  make-up.of  the  cable  vary  somewhat  with  the  number  of \ncircuits  of  the  various  types  that  are  to  be  provided  in  the  different \nsections,  but  in  general  it  is  full  size,  that  is,  its  over-all  diameter  is\n\n2^  in.  which  is  about  the  maximum  size  of  telephone  cable.  The \nsheath  is  of  lead-antimony  alloy,  one-eighth  of  an  inch  thick,  and \nunder  normal  conditions  it  is,  of  course,  air-tight  to  keep  moisture \nfrom  entering.  The  cable  for  the  aerial  section  was  received  from  the \nfactory  in  500-foot  lengths,  this  being  largely  determined  by  the \narrangement  necessary  to  permit  the  proper  installation  tests.\n\nWe  might  next  consider  the  route  selected  and  for  this  purpose \nFig.  2  will  again  be  helpful.     It  will  be  noted  that  starting  at  Phila-\n\ndelphia,  the  ca]:)le  is  routed  to  Reading  touching  Pottstown,  Phoenix- \nville  and  other  points.  From  Reading  to  Harrisburg  the  cable  follows \nclosely  the  William  Penn  Highway,  although  in  sections  it  was  neces- \nsary to  obtain  private  right-of-way  or  to  use  longer  routes  removed \nfrom  this  highway  on  account  of  the  lines  of  various  kinds  already \nin  operation  there.  It  is  very  desirable  for  economic  reasons  to \nkeep  the  length  of  these  cables  as  short  as  possible  and  in  some  cases \nthis  is  absolutely  necessary  to  obtain   proper  operating  conditions.\n\nHow^ever,  the  most  direct  routes  cannot  always  be  used,  for  many \nobvious  reasons,  and  this  problem  required  careful  consideration \nin  all  sections  of  the  cable.\n\nBetween  Harrisburg  and  Pittsburgh  the  Allegheny  Mountains \nhad  to  be  crossed  and  for  this  crossing  only  two  general  routes  were \nfound  practicable,  the  first  following  an  existing  pole  line  which  is \nthe  New  York-Chicago  telephone  line  through  Lewuston,  Altoona, \netc..  and  which  we  may  call  the  northern  route,  and  second  a  southern \nroute  through  Shippensburg,  Bedford  and  Ligonier  for  the  most  part \nalong  the  Philadelphia-Chicago  line  and  also  the  Lincoln  Highway. \nA  middle  route  which  is  now  used  for  the  Harrisburg-Pittsburgh  line \nwas  not  seriously  considered  as  the  country  was  too  rough  for  econom- \nical construction  and  maintenance  and  no  important  advantages \nwere  to  be  obtained.  After  careful  surveys  and  cost  studies,  taking \ninto  account  all  existing  and  anticipated  conditions,  such  as  circuit\n\nrequirements  and  towns  to  be  reached,  length  of  practicable  routes, \nmaintenance  conditions,  freedom  from  proiiable  physical  and  electrical \ninterference,  etc.,  it  was  decided  to  build  on  the  southern  route.\n\nThis  route,  while  of  nearly  the  same  length  as  the  northern  one  and \noffering  some  important  advantages,  was  not  free  from  difficulties \nas  it  crosses  the  Allegheny  Mountains  within  a  few  miles  of  the  highest\n\npoint.  Fig.  7  shows  the  cable  line  on  what  is  know  n  as  the  seven-mile \nstretch  of  the  Lincoln  Highway  east  of  Ligonier,  and  here  the  going \nwas  fairly  good.  The  Philadelphia-Chicago  aerial  wire  line  is  also \nshown  and  two  of  the  crossarms  carrying  10  wires  each  are  to  be \nremoved  in  the  near  future  and  the  circuits  operated  in  the  cable. \nIt  is  planned  to  remove  the  remaining  two  crossarms  later  on.  Fig.  8 \nshows  the  cable  across  a  valley  and  is  taken  from  the  point  on  the \nLincoln  Highway  called  Grand  View\\  Fig.  9  shows  the  crossing \nof  the  Juniata  River  east  of  Bedford  where  special  construction  was \nused.  Fig.  10  shows  just  one  example  of  the  conditions  encountered \nin  crossing  the  many  mountains  and  a  photograph  does  not  do  the \nscenery  or  the  construction  difificulties  justice.  On  account  of  the \nsteep  slopes,  clamps  are  used  at  many  points  to  fasten  the  cable  to \nthe  strand.\n\nNarrow-gage  timber  railroads  were  used  in  the  mountains  where \npossible  to  get  material  to  the  job  and  Fig.  11  shows  one  of  the  regular\n\nflat  cars  adapted  for  our  purpose.  Fig.  12  shows  two  5-ton  tractors \nin  action  on  top  of  one  of  the  mountains.  As  many  sections  of  the \ncountry  are  very  rough  and  highways  several  miles  distant  it  seemed \nthat  no  other  method  of  transporting  the  cable  reels,  wiiich  weigh\n\nnearly  5,000  pounds,  could  possibly  be  used,  and  certainly  no  other \nmeans  would  have  been  as  satisfactory.  Even  with  these  methods \nthe  cable  reels  could  not  always  be  delivered  where  desired  and  in \nsome  cases  it  was  necessary  to  pull  the  sections  of  cable  through  the \nrings  for  a  distance  of  nearly  a  mile  to  get  them  in  place.\n\nAs  stated  before,  the  make-up  of  the  cable  varies  somewhat  with \nthe  circuit  requirements  in  the  different  sections  but  the  wires  and \narrangement  in  a  typical  section  of  cable  are  roughly  illustrated  in \nFig.  13.\n\nThe  cable  is  of  quadded  construction,  that  is,  the  wires  are  first \nwrapped  with  dry  paper  for  insulation  and  twisted  into  pairs  and  then \ntwo  pairs  are  twisted  into  what  is  called  a  quad.     These  quads  are\n\narranged  in  concentric  layers  as  shown  and  great  care  and  skill  are \nrequired  in  the  design  and  manufacture  or  there  is  certain  to  be  serious \ncross-talk  between  the  several  hundred  circiiits  when  used  for  long- \ndistance service.  Even  after  the  application  of  the  best  present  manu- \nfacturing methods,  tests  are  made  on  all  circuits  at  three  points  in \neach  loading  section  of  6000  feet  while  the  cable  is  being  spliced. \nThese  tests  are  made  in  order  to  determine  the  best  possible  arrange-\n\nFig.  14 \u2014 General  Phantom  Circuit  Arrangement.   Four  wires  providing  three  circuits\n\nment  of  conductors  for  still  further  reducing  cross-talk  between  circuits, \nand  the  splicing  is  done  accordingly.\n\nThere  are  19  quads  of  No.  16  A.  W.  G.  and  120  quads  of  No.  19 \nA.  W.  G.  pure  copper  conductors  in  one  of  the  principal  sections,  and \nthe  arrangement  of  the  four  wires  in  each  quad  is  such  that  two  physi- \ncal circuits  and  one  phantom  circuit  are  made  available.  The  method \nof  obtaining  three  telephone  circuits  from  two  pairs  of  wires  is  old  and \nextensively  used.  It  is  illustrated  in  Fig.  14.  The  method  results \nin  a  50  per  cent  increase  in  the  number  of  available  circuits  and  its \napplication  to  this  project  is  therefore  of  very  great  economic  impor- \ntance. Now  the  total  of  139  quads  multiplied  by  3  gives  417  circuits \nor  as  many  as  could  be  carried  on  about  14  heavily  loaded  pole  lines \nif  aerial  wire  were  used,  but  as  will  be  described  later,  we  will  have  to \nuse  two  of  these  circuits  to  make  one  telephone  circuit  in  some  cases \nwhere  the  distances  are  comparatively  great,  so  it  is  expected  that \nonly  about  300  telephone  circuits  will  be  obtained  for  regular  service. \nThis  is  as  many  as  could  be  carried  on  10  heavily  loaded  pole  lines  if \naerial  wire  were  used.  It  is  now  thought  that  in  some  sections  this \nnumber  of  circuits  will  take  care  of  future  demands  for  about  10  years \nafter  allowing  for  the  dismantling  of  some  existing  aerial  wire.\n\nAs  these  cables  can  be  obtained  in  any  size  desired  up  to  the  maxi- \nmum, the  period  for  which  they  should  be  engineered  can  be  determined \nfrom  studies  of  circuit  requirements  and  costs.  These  studies  are  of \nvery  great  importance  and  the  cost  considerations  include,  of  course,\n\nannual  costs  of  the  various  plans  over  proper  periods  as  well  as  first \ncosts.\n\nLoading  coils  are  now  connected  to  many  of  the  circuits  and  all  of \nthe  circuits  in  this  cable  are  intended  to  be  equipped  with  coils  located \nat  6000-foot  intervals.  The  theory  and  practice  of  loading  are  de- \nscribed in  papers  previously  presented  before  the  Institute^  and  for \nour  purpose  it  will  be  sufficient  to  state  that  these  loading  coils  very \nmaterially  reduce  the  attenuation  losses  and  improve  the  quality  of \ntransmission  as  compared  to  cable  circuits  not  so  equipped.  The  im- \nprovement in  so  far  as  the  attenuation  losses  are  concerned,  varies \nwith  the  type  of  circuit  and  loading  coils,  but  with  one  of  the  No.  19\n\nA.  W.  G.  circuits  in  this  cable  loaded  with  coils  having  an  inductance  of \n0.175  henry  located  at  6000-ft.  intervals,  the  losses  are  only  about  one- \nthird  as  great  as  in  a  similar  circuit  without  the  coils.  The  connections \nand  arrangements  of  the  coils  are  shown  in  Fig.  15  and  it  will  be  noted \nthat  coils  have  been  connected  to  both  the  physical  and  phantom  cir- \ncuits. The  arrangement  is  such  that  there  is  no  appreciable  inter- \nference between  circuits  due  to  magnetic  action  in  the  iron  cores  of  the \ndifferent  coils  or  to  the  necessarily  close  electrical  relation  in  the \nwindings.\n\nThe  loading  coils  are  potted  and  sealed  in  iron  pots,  two  of  which \nare  shown  in  Fig.  16,  and  in  the  country  these  are  mounted  on  pole \nfixtures.  Each  pot  contains  36  groups  of  3  coils  each.  The  pots  are \nnearly  30  inches  in  diameter  at  the  flange,  52  inches  high  and  weigh \nabout  2700  pounds.  The  pots  can  be  obtained  in  different  sizes \ndepending  upon  the  number  of  coils  which  it  is  desired  to  install  at \none   time.     When   the  cable  was  installed,  extra  lead  sleeves  were\n\nplaced  at  the  loading  points  and  a  little  slack  left  in  the  wire  to  facili- \ntate the  connection  of  four  additional  loading  pots  to  the  cable  at\n\nsome  later  date  when  the  circuits  are  needed.  The  loading  points \nmust  be  uniformly  spaced  in  order  to  obtain  the  proper  impedance \ncharacteristics  in  the  circuits  as  will  be  referred  to  later.  Fig.  17 \nshows  the  iron  core  of  a  loading  coil  and   Fig.   18  shows  this   core\n\nwound  with  insulated  wire  and   then  wrapped  with   cloth  and   the \nterminals  brought  out  nearly  ready  for  potting.     Fig.  19  shows  several\n\ncoils  arranged  on  one  of  the  spindles  which  will  be  placed  in  the  iron \npot  also  shown.  This  particular  pot  will  hold  7  spindles  and  when \nthey  are  in  place,  the  pot  will  be  filled  with  compound  and  thoroughly- \nsealed.\n\nTelephone  Repeaters \nEven  with   the  improvement  in   the  quality  of  transmission  and \nreduced  attenuation  losses  effected  by  the  use  of  loading  coils,  loaded\n\ncable  circuits  alone  of  No.  16  and  No.  19  A.  W.  G.  could  be  satisfac- \ntorily operated  for  distances  less  than  100  and  60  miles,  respectively, \nand  this  is  far  short  of  our  requirements  in  this  case.  In  fact,  we  wish \nto  operate  some  telephone  circuits  on  these  conductors  and  through  this \ncable  and  future  cables  up  to  at  least  1000  miles  in  length.     This\n\ncan  be  accomplished  1)\\'  tlie  use  of  telephone  repeaters  connected  to \nthe  loaded  conductors.\n\nall  the  while  the  original  wave  shape.  Therefore,  if  one  or  more  tele- \nphone repeaters  are  properly  inserted  in  circuits  adapted  to  their  use, \nthe  range  of  satisfactory  transmission  can  be  greatly  extended.  As \nmany  hundreds  of  vacuum-tube  repeaters  are  in  operation  on  the \nPhiladelphia-Pittsburgh  cable  and  connected  cables,  and  as  a  great \nmany  more  are  planned  for  future  installation,  we  will  briefly  consider \nthe  elementary  features  of  some  of  the  types  of  repeaters  used.\n\nelement  of  this  type  of  repeater.  It  is  a  small  glass  bulb  with  a \nvacuum  that  is  as  good  as  is  practicable  to  obtain.  In  the  tube \nis   a   filament   which   is   heated   to   incandescence   during   operation,\n\nand  a  grid  and  plate.  The  circuits  directly  associated  with  the  tube \nare  shown  in  more  detail  in  Fig.  21,  and  this  would  constitute  a  device \nfor  amplifying  currents  from  one  direction.  As  is  well  understood, \nany  change  in  the  potential  impressed  on  the  grid  causes  a  change  in \nthe  current  flowing  in  the  plate-filament  circuit.  To  obtain  complete \ntwo-way  repeater  action  two  of  these  amplifier  arrangements  are \ncombined  with  the  circuits  shown  in  Fig.  22.\n\nIt  will  be  noted  that  the  line  circuit  from  one  direction,  for  instance, \nthe  one  designated  \"line  west,\"  is  connected  through  a  three-winding \ntransformer  to  a  balancing  network  which  is  so  made  up  as  to  balance\n\nthe  line  as  nearly  as  possible  at  telephone  frecjuencies.  This  balance \nis  essential  to  proper  repeater  operation.  The  circuit  arrangement  is \nsuch  that  part  of  the  incoming  energy  is  diverted  to  that  part  of  the \ncircuit  containing  the  input  coil  directly  associated  with  this  three- \nwinding  transformer.  By  the  action  of  the  vacuum-tube  arrangement \namplified  energy  is  transmitted  to  the  line  east.  That  part  of  the \noriginal  incoming  energy  from  the  line  west  that  goes  through  the \nbalancing  network  or  the  output  coil  is  not,  of  course,  transmitted \nalong  into  the  line  east.  The  operation  in  the  case  of  currents  incom- \ning from  the  line  east  is  similar  and  it  w^ill  be  noted  that  the  complete \nrepeater  circuit  is  made  up  of  two  symmetrical  parts.  This  circuit \narrangement  constitutes  what  is  known  as  a  two-wire  repeater  and  the \napparatus  is,  of  course,  all  closely  associated  in  the  same  office.\n\nEquipped  with  TelepNsne  Repeaters  end \nAi-ranged  for  Connection  to  Wn  Wire  Circyiti  ft  the  Terminil*\n\nFig.   23 \u2014 Four-Wire  Circuit   equipped  with  telephone  repeaters  and  arranged  for \nconnection  to  two  wire-circuits  at  the  terminals\n\nSeveral  of  these  repeaters  may  be  inserted  in  tandem  at  appropriate \npoints  in  a  circuit,  but  there  is  a  limit  to  the  length  of  circuit  that  can \nbe  satisfactorily  operated  with  this  arrangement,  this  length  depend- \ning upon  the  type  of  the  facilities  used.  When  longer  circuits  are \nrequired,  a  four-wire  arrangement  is  used,  as  shown  in  Fig.  23.  It \nwill  be  noted  that  in  this  arrangement  the  three-winding  transformers \nare  not  located  in  the  same  office  but  may  be  in  offices  several  hundred \nmiles  apart.  At  each  of  the  intermediate  stations  a  vacuum-tube \namplifier  arranged  for  amplification  in  one  direction  only  is  connected \nto  each  of  the  two  branches  of  the  circuit.  Two  circuits  are,  of \ncourse,  required  between  the  terminals  and  these  may  be  either \nphysical  or  phantom  circuits.\n\nnot  required  at  each  repeater  station  and  the  general  matter  of  balance \nand  consequently  good  repeater  operation  in  the  circuit  as  a  whole  is \ngreatly  simplified.  This  arrangement  can,  therefore,  be  satisfactorily \nused  for  long  circuits  where  two-wire  operation  might  be  impracticable,\n\nand  examples  would  be  such  circuits  as  New  York-Pittsburgh  or  New \nYork-Chicago.\n\nBoth  of  these  types  of  circuits  may  be  operated  on  No.  19  A.  W.  G. \nfour-wire  facilities  which  may  be  either  physical  or  phantom  circuits.\n\nFig.  24  shows  a  group  of  repeaters  installed  in  the  office  at  Reading, \nPa.,  and  Fig.  25  shows  one  of  the  four-wire  repeater  units  in  somewhat \ngreater  detail.\n\nIn  order  that  networks  may  be  used  to  balance  the  lines  for  repeater \noperation,  it  is  necessary  as  a  practical  proposition  that  the  impe- \ndance characteristics  of  the  lines  be  fairly  uniform  over  the  range  of \ntelephone  frequencies.  The  solid  line  in  Fig.  26  shows  the  resist- \nance component  of  the  impedance  of  a  No.  19  loaded  cable  circuit \nwith  all  loading  coils  in  place.     The  solid  line  in  Fig.  27  shcnss  the\n\nresistance  component  found  in  impedance  measurements  on  the  same \ncircuit  with  one  coil  omitted  at  the  thirteenth  loading  point  from \nthe  end  at  which  the  tests  were  made.  It  will  be  noted  that  in  the \nlatter  case  the  characteristics  of  the  circuitsvary  greatly  with  frequency. \nIt  would  therefore  be  very  difficult  as  a  practical  proposition  to  build \nup  a  network  that  would  balance  lines  in  this  condition,  and  such \nvariations  in  the  electrical  characteristics  of  a  circuit  impair  the \nquality  of  telephone  transmission,  as  the  currents  of  different  fre- \nquencies are  differently  affected.  The  necessity  for  careful  main- \ntenance work  in  promptly  replacing  loading  coils  which  may  become \ndefective  or  preventing  other  irregularities  from  creeping  into  the  plant \nwill  therefore  be  clear.\n\nThe  resistance  of  small-gage  cable  conductors  is  one  of  the  important \nfactors  that  determine  the  transmission  losses  of  a  circuit.  The \nresistance  of  a  No.  19  A.  W.  G.  pair  is  about  88  ohms  per  mile  so  that\n\nFig.  27 \u2014 Cable  Circuit  with  Loading  Coil  Missing  at  Thirteenth  Loading  Point  from\n\nin  a  long  circuit  this  factor  of  line  resistance  reaches  considerable \nproportions.  Now  as  most  of  the  cable  is  aerial,  the  resistance  of \nthe  conductors  is  of  course  affected  by  changes  in  temperature  both \ndaily  and  seasonal  and  the  transmission  losses  vary  accordingly. \nThese  changes  in  transmission  values  are  of  such  magnitude  that \nautomatic  transmission  regulators  are  being  provided  for  certain \ngroups  of  longer  circuits.  All  changes  in  the  transmission  equivalents \nof  the  circuits  from  whatever  causes  must  be  carefully  watched  and \nnecessary  adjustments  made  or  the  service  will  be  seriously  affected.\n\nIn  the  section  between  Philadelphia  and  Pittsburgh  practically  all \nof  the  existing  long  aerial  wire  circuits  are  composited,  that  is,  they \nare  arranged  for  simultaneous  telephone  and  telegraph  operation. \nThe  telegraph  circuits  thus  obtained  are  generally  used  in  furnishing \nwhat  is  sometimes  called  \"leased  wire\"  service.  The  ground  return \nsystem  providing  either  full  duplex  or  single-line  operation  is  used \nand  the  line  currents  average  about  75  milliamperes.  This  grounded \ntelegraph  system  cannot  be  used  where  simultaneous  telephone  and \ntelegraph  service  is  desired  on  loaded  cible  circuits  of  the  length \ninvolved  in  this  cable,  and  as  a  part  of  the  work  of  carrying  out  the \ncomprehensive  toll  cable  plans  of  the  Bell  system,  a  new  telegraph \nsystem  had  to  be  developed.  It  was  found  preferable  to  use  a  metallic \nreturn  circuit  and  to  limit  the  line  current  to  a  value  between  3  and  5 \nmilliamperes  in  order  to  prevent  serious  interference  to  the  telephone \ncircuits  due  to  the  \"flutter  effect, \"^  Morse  thump,  and  mutual  inter- \nference between  telegraph  circuits.  Morse  thump  results  when  the \ncomposite  sets,  that  is,  the  apparatus  used  for  separating  the  tele- \nphone and  telegraph  currents,  do  not  completely  prevent  the  latter \nfrom  entering  the  telephone  circuit,  thus  causing  interference.  The \ntelegraph  repeaters  are  located  at  about  100-mile  intervals  on  the \nNo.  19  circuits  and  at  somewhat  less  frequent  intervals  on  No.  16 \ncircuits.  The  telegraph  apparatus  is  of  course  located  in  the  same \nbuildings  that  are  used  to  house  the  telephone  repeaters,  and  on  the \nPhiladelphia-Pittsburgh  cable  telegraph  repeaters  will  be  located \ninitially  at  Philadelphia,  Harrisburg,  Bedford  and  Pittsburgh.\n\nAll  of  the  conductors  in  the  cables  are  carried  into  stations  located \nat  about  50-mile  intervals  and  apparatus  is  provided  in  these  stations \nfor  making  regular  tests  to  ascertain  the  condition  of  the  cable  and  to \nlocate  trouble  quickly.  At  these  offices  the  different  kinds  of  operat- \ning apparatus  are  also  connected  to  the  cable  conductors;  examples \nof  this  apparatus  are  phantom  repeating  coils,  composite  sets  to  per- \nmit simultaneous  telephone  and  telegraph  operation,  telegraph \nrepeaters,  telephone  repeaters  and  associated  balancing  equipment, \nsignaling  apparatus,  and  where  required,  the  switchboards  necessary \nfor  making  the  telephone  connections  involved  in  furnishing  service. \nIt  is  necessary  that  this  apparatus  which  is  installed  in  large  quantities\n\nbe  systematically  arranged  and  facilities  provided  for  making  quick \nchanges  in  the  circuit  arrangement.  The  circuits  are  wired  through \njacks  installed  in  groups  in  test  boards  for  this  purpose  and  to  facilitate \ntesting.  One  of  these  boards  is  illustrated  in  Fig.  28.  This  particular \nboard  is  located  in  one  of  the  larger  offices.  The  test  boards  in  one \nof  the  repeater  stations,  such  as  Bedford,  would  consist  of  a  smaller\n\nnumber  of  positions.     A  position  is  three  feet  in  length, \neach  position  bears  a  number.\n\nTelephone  repeaters  of  either  the  two-wire  or  four-wire  type  are \nconnected  to  the  circuits  at  approximate  intervals  of  either  50  or  100 \nmiles,  depending  upon  the  type  of  facilities  which  it  is  economical \nto  use  in  the  different  circuits  and  the  kind  of  service  for  which  a  given \ncircuit  is  intended.  As  mentioned  above,  telegraph  repeaters  are \ninstalled  at  about  100-mile  intervals.  At  some  of  these  points  existing \noffices  are  used  while  in  a  number  of  cases  it  was  necessary  to  erect \nbuildings  for  the  sole  purpose  of  housing  the  repeaters,  testing  appa- \nratus and  other  equipment  associated  with  the  cable.  For  example, \nnew  buildings  of  fire-proof  construction  were  erected  at  Shippensburg, \nBedford  and  Ligonier.  Fig.  29  is  a  view  of  the  building  at  the  latter \npoint  and  the  other  two  buildings  are  similar  to  this  one,  the  dimensions \nbeing  about  50  by  80  feet.     Power  plants  are  installed  in  these  build-\n\nings  to  furnish  current  of  the  proper  characteristics  for  operating  the \napparatus,  and  storage  batteries  are  provided  to  insure  uninter- \nrupted service.  As  an  indication  of  the  size  of  these  plants  the  24- \nNolt  storage  batteries  installed  for  the  initial  load  at  Bedford  have \na  capacity  of  2240  ampere-hours  and  this  provides  about  one  day's \nreserve.  The  capacity  can,  of  course,  be  increased  as  repeaters  are \nadded  from  time  to  time  when  additional  circuits  are  needed.  Storage \nbatteries  of  smaller  sizes  supplying  current  at  potentials  of  30,  120 \nand  130  volts  are  also  provided.\n\nFig.  30  shows  two  possible  methods  of  building  up  a  Philadelphia- \nPittsburgh  terminal  circuit  and  Fig.  31,  a  method  of  building  up  a \nNew  York-Pittsburgh  terminal  circuit.  In  all  three  cases  these \ntelephone  circuits  are  intended  to  have  a  transmission  equivalent  of \nabout    12   miles   of   standard   cable.     Some   Philadelphia-Pittsburgh\n\nterminal  circuits  of  the  first  type  have  been  in  everyday  operation \nfor  several  months,  but  it  is  not  the  most  economical  arrangement \nthat  it  is  possible  to  obtain  for  general  use  in  providing  this  or  similar \nservice.  It  will  be  noted  that  No.  19  four-wire  facilities  are  used \nbetween  Philadelphia  and  Harrisburg  and  four-wire  repeaters  are \nlocated  at  these  two  points.     At  Harrisburg  the  four-wire  circuit  is\n\nconnected  to  a  No.  16  two-wire  circuit  with  a  two-wire  repeater  at \nBedford.  This  arrangement  was  used  in  order  to  start  service  through \nthe  cable  with  the  facilities  available,  but  it  is  intended  later  on  to  use \nthe  arrangement  shown  in  example  No.  2.\n\nIn  example  No.  2,  No.  16  heavily  loaded  conductors  are  used  and \ntwo-wire  repeaters  are  located  at  Reading,  Shippensburg  and  Ligonier. \nThe  total  transmission  equivalent  of  this  circuit  without  repeaters  is\n\nPITTSBUnCH  LICONIER  BCOrORD  SHIPPCNSBURC  HtRRISBURC  READING  PHILIDCLPHI*\n\nabout  50  miles  of  standard  cable  so  that  in  order  to  obtain  a  net \nequivalent  of  12  miles  for  the  circuit  each  of  the  three  repeaters  must \ngive  a  transmission  gain  of  nearly  13  miles  of  standard  cable.  This \ncircuit  could  not  of  course  be  used  for  telephone  purposes  without \nrepeaters.\n\nThe  third  example  shows  how  it  is  expected  to  operate  New  York- \nPittsburgh  circuits  intended  for  business  between  these  two  terminals.\n\nFour-wire  No.  19  loaded  cable  facilities  are  used  with  four-wire  tele- \nphone repeaters  located  at  New  York,  Philadelphia,  Harrisburg, \nBedford  and  Pittsburgh.\n\nEven  with  conductors  of  only  two  gages  in  the  cable,  it  is  clear  that \nmany  different  combinations  of  facilities  can  be  built  up  into  tele- \nphone circuits  and  an  endeavor  is  always  made  to  use  the  most  eco-\n\nnomical  arrangement  that  will  furnish  the  service  required  over  each \ncircuit  group.  The  examples  described  above  are  of  circuits  used  for \nbusiness  between  the  terminals  indicated  and  if  these  circuits  were \nto  be  connected  to  others  extending  to  points  considerable  distances \nbeyond  these  terminals  different  arrangements  would  be  required. \nThe  cable  conductors  used  in  building  up  these  telephone  circuits \ncan  be  ccmposited  and  telegraph  circuits  are  thus  provided  for  simul- \ntaneou!^  operation  with  the  telephone  circuits.\n\nIn  the  above  discussion,  an  effort  has  been  made  to  furnish  some \ndescriptive  information  regarding  a  complete  cable  system  recently \ncompleted  and  now  in  successful  operation  between  Philadelphia  and \nPittsburgh  and  designed  for  long-distance  telephone  and  telegraph \nservice.  In  one  sense  this  discussion  may  be  considered  a  report  of \nthe  present  status  of  the  toll  cable  plant  intended  to  connect  Atlantic \nSeaboard  cities  wdth  Chicago  and  other  cities,  and  extensions  are  now \nunder  construction.  However,  most  of  the  general  methods  which \nit  is  planned  to  use  in  these  extensions  are  not  expected  to  differ \ngreatly  from  those  described.\n\nAn  important  feature  of  this  cable  project  is  the  fact  that  while \nmany  new  developments  and  practices  are  utilized,  the  design  of  the \nsystem  as  a  whole  is  such  as  to  fit  in  economically  with  existing  wire \nand  cable  systems  and  proposed  extensions.\n\nSynopsis:  The  present  paper  presents  an  extensive  theoretical  investiga- \ntion of  the  impedance  of  the  \"sea  return\"  of  various  types  of  submarine \ncables.  In  the  case  of  the  cables  used  for  submarine  telegraphy  the  im- \npedance of  the  sea  return  has  been  practically  negligible  because  of  the  low \nfrequencies  involved.  For  these  low  frequencies  the  cross-section  of  the  re- \nturn path  is  very  large  and  its  resistance  low,  even  though  the  specific \nresistance  of  sea  water  is  of  the  order  of  ten  million  times  that  of  copper. \nAs  the  frequency  of  the  cable  current  is  raised,  however,  the  return  cur- \nrents crowd  in  nearer  the  cable  and  the  resistance  of  the  return  path  is \nincreased.  For  frequencies  in  and  above  the  telephone  range,  the  return \ncurrents  are  forced  into  the  steel  armor  wires  around  the  cable  and  into  the \nwater  just  outside  of  the  insulation.  The  small  cross-section  of  the  water \ninvolved  and  the  loss  in  the  armor  wires  cause  the  resistance  of  the  return \npath  to  become  a  very  large  part  of  the  total  resistance  of  the  circuit.\n\nThe  present  investigation  led  to  the  conclusion  that  the  resistance  of \nthe  return  path  could  be  greatly  diminished  by  winding  a  low  resistance \nconductor  in  the  form  of  a  copper  tape  immediately  around  the  gutta \npercha  insulation  applied  to  the  core  of  the  cable.  The  concentric,  cylin- \ndrical conductor  thus  formed  lies  within  the  armor  wires  but  is  not  insu- \nlated from  them  and  the  sea  water.  Estimates  of  the  sea  return  which \nwould  have  been  obtained  in  the  Key  West-Havana  cable  if  no  copper \ntape  had  been  provided  give  values  of  4,  6.5,  and  8  ohms  per  nautical  mile \nat  1,000,  3,000  and  5,000  cycles.  The  resistance  actually  obtained  with \nthe  copper  tape  does  not  exceed  1.7  ohms  at  5,000  cycles.  The  greater \nvalues  would  have  increased  the  attenuation  by  approximately  30%  at \n1,000  cycles  and  by  50%  at  the  two  higher  frequencies.  The  present \ncable  permits  of  the  operation  of  a  carrier  telegraph  channel  at  3,800  cycles, \nthis  lying  above  the  range  of  telephone  frequencies.\n\nThe  paper  gives  a  comparison  of  the  theoretical  conclusions  with  ex- \nperimental data  and  the  agreement  is  so  satisfactory  as  to  indicate  that \nthe  theory  is  a  reliable  guide  in  the  design  of  such  a  cable. \u2014 Editor.\n\nTHE  transmission  characteristics  of  a  conducting  system,  such  as \na  submarine  cable  circuit,  are  determined  by  its  propagation \nconstant,  F,  and  characteristic  impedance,  K,  which  may  be  calcu- \nlated for  the  frequency  p  12-k  from  the  formulas:\n\nwhere  R,  L,  G  and  C  are  the  four  fundamental  line  parameters,  re- \nsistance, inductance,  leakance,  and  capacity,  all  per  unit  length.  These \nformulas  are  rigorous  for  all  types  of  transmission  systems;  but  the \ndetermination  of  the  line  parameters  is  not  always  possible  by  ele- \nmentary methods,  and  may  indeed  be  a  matter  of  considerable  com-\n\nplexity  and  involve  rather  difficult  analysis.  In  the  case  of  the  sub- \nmarine cable,  exact  formulas  are  available  for  calculating  the  capacity \nand  leakage  and  the  core  impedance.  Considerable  uncertainty  is \nintroduced  into  the  theory,  however,  on  account  of  the  lack  of  a \nmethod  of  determining  the  \"return  impedance,\"  that  is,  the  con- \ntribution of  the  \"sea  return\"  (sea  water,  armor  wires,  etc.)  to  the \neffective  resistance  and  inductance  of  the  circuit.  An  investigation \nof  this  problem  was  undertaken  by  the  writers  in  connection  with  the \nresearch  program  of  the  American  Telephone  and  Telegraph  Company \nand  the  Western  Electric  Company.\n\nThe  purpose  of  the  present  paper  is  to  discuss  transmission  over \nthe  submarine  cable,  and,  more  particularly,  to  develop  rigorous \nformulas  for  the  calculation  of  the  impedance  of  the  return  con- \nductor of  the  cable.  The  results  of  theoretical  calculations  are  then \ncompared  with  actual  experimental  data;  and  the  agreement  between \ntheory  and  experiment  is  so  satisfactory  as  to  indicate  that  the  former \nis  a  reliable  guide  in  the  design  and  predetermination  of  the  cable.\n\nBesides  providing  a  method  for  accurately  calculating  the  trans- \nmission characteristics  of  a  submarine  cable,  the  present  analysis \nleads  to  the  following  general  conclusions:\n\n(1)  Contrary  to  usual  assumption,  the  \"sea  return\"  impedance \nis  by  no  means  negligible.  Even  at  quite  moderate  frequencies  there \nis  a  considerable  crowding  of  the  return  current  into  the  immediate \nneighborhood  of  the  cable,  with  a  consequent  rapid  increase  of  the \nresistance  and  a  corresponding  decrease  of  the  inductance  of  the \ncircuit.  Except  at  the  lowest  frequencies,  therefore,  the  impedance \nof  the  \"sea  return\"  is  a  very  important  factor.\n\n(2)  The  armor  wires  which  surround  the  cable,  and  which  are \nnecessary  for  mechanical  protection,  have  a  very  pronounced  effect \non  the  impedance  of  the  sea  return,  and  even  at  moderate  frequencies \nmay  become  the  controlling  factor.  Their  action  is  to  screen  the \ncurrent  from  the  sea  water  itself,  and,  as  the  frequency  increases, \nto  carry  more  and  more  of  the  return  current,  until  it  is  almost  en- \ntirely confined  to  the  armor  wires  and  excluded  from  the  sea  water.\n\n(3)  The  rapid  increase  in  the  impedance  of  the  armor  wires  with \nfrequency,  and  their  pronounced  and  even  controlling  effect  on  trans- \nmission makes  a  thorough-going  study  of  their  role  in  the  electrical \nsystem  a  matter  of  first-class  importance.  Heretofore  they  appear \nto  have  been  regarded  only  as  a  mechanical  protection,  and  their \neffect  on  transmission  has  been  ignored.  The  accurate  method  of \ncalculating  their  impedance  which  is  developed  in  the  following  pages \nis  believed  to  have  considerable  value  in  this  connection.\n\n(4)  At  relatively  high  frequencies,  the  return  impedance,  and \nhence  the  attenuation  and  the  distortion,  may  be  very  greatly  de- \ncreased by  a  correctly  designed  thin  metallic  sheath  concentric  with \nthe  core,  and  in  electrical  contact  with  the  armor  wires.  The  very \nimportant  action  of  such  a  sheath,  even  when  extremely  thin,  does  not \nappear  to  have  been  adequately  recognized  or  studied.  It  is  suggested \nthat  the  introduction  of  such  a  sheath  alifords  a  means  of  greatly \nincreasing   the   range   of   frequencies  which  the  cable   can   transmit.\n\nThe  general  problem  of  determining  the  transmission  character- \nistics of  a  system  consisting  of  an  insulated  conductor  surrounded  by  a \nconcentric  ring  of  armor  wires  immersed  in  sea  water  is  of  consider- \nable difficulty,  since  in  this  case  the  propagated  wave  must  be  repre- \nsented as  a  set  of  component  waves  centered  upon  or  diverging  from \nthe  axes  of  the  core  and  of  the  individual  armor  wires.  The  problem \nwas  first  simplified  by  replacing  the  ring  of  armor  wires  by  a  cylindrical \nsheath,  thus  giving  circular  symmetry  to  the  structure.  The  analysis \nof  this  case,  however,  showed  that  the  effect  of  the  iron  sheath  re- \nplacing the  armor  wires  was  so  pronounced  as  to  make  this  simplifying \nassumption  of  doubtful  validity.  The  general  problem  was  there- \nfore attacked,  and  rigorous  methods  developed  for  calculating  the \neffect  of  the  armor  wires  upon  transmission.  The  results  in  this \ncase  differ  markedly  from  those  obtained  for  the  case  of  a  continuous \niron  sheath,  which  indicates  that  great  caution  must  be  used  in  making \nassumptions  regarding  the  physical  structure  of  the  armoring.\n\nThe  present  paper  follows  rather  closely  the  course  of  the  writers' \ninvestigation.  In  Section  II  is  analyzed  the  problem  of  transmission \nover  a  system  consisting  of  n  coaxial  cylindrical  conductors,  which \nmay  be  either  in  electrical  contact  at  their  adjacent  surfaces  or  sepa- \nrated from  each  other  by  dielectric  spaces.  The  outermost  con- \nductor, consisting  of  the  sea  water,  is  assumed  to  extend  to  infinity. \nThis  analysis  is  then  applied,  in  Section  III,  to  the  case  of  a  sub- \nmarine cable  which  is  armored  with  a  continuous  iron  sheath.  This \nproblem  is  not  only  of  interest  in  itself,  but  serves  as  a  first  approxi- \nmation to  the  case  of  an  actual  cable,  and  gives  a  clear  qualitative \nidea  of  the  effect  of  the  various  factors  on  transmission.  In  Section \nIV  the  problem  of  the  submarine  cable  armored  with  a  ring  of  iron \nwires  is  attacked  and  solved  by  rigorous  methods,  and  the  theoretical \nresults  are  then  compared  with  experimental  data.\n\nThe  solution  of  the  problem  of  transmission  of  periodic  currents \nover  a  system  comprising  n  coaxial  cylindrical  conductors  consists\n\nin  finding  the  particular  solution  of  Maxwell's  equations  which  satis- \nfies the  boundary  conditions \u2014 continuity  of  tangential  electric  and \nmagnetic  forces  at  the  surfaces  of  the  conductors.  Let  the  common \naxis  of  the  conductors  coincide  with  the  Z  axis  of  a  system  of  polar \ncoordinates,  R,  <I>,  Z,  and  let  the  electric  and  inagnetic  variables  in- \nvolve the  common  factor  exp  (  \u2014 F  2  +  ipt),  T  is  therefore  the  propa- \ngation factor  characterizing  transmission,  and  p  is  2ir  times  the  fre- \nquency. This  factor  will  not  be  explicitly  written  in  any  of  the  work \nthat  follows,  but  it  will  be  assumed  to  be  incorporated  in  each  of  the \nelectric  variables  so  that\n\nFrom  symmetry,  it  is  evident  that  the  component  of  electric  field \nintensity  in  the  direction  of  <^  vanishes,  and  that  the  magnetic  lines \nof  force  are  circles  lying  in  planes  perpendicular  to  the  axis  of  the \nsystem,  and  centered  on  that  axis.  Also,  the  axial  and  radial  electric \nforces  are  independent  of  0.  It  can  be  shown  that  the  radial  compo- \nnent of  electric  field  intensity  in  the  conductors  is  negligibly  small  com- \npared with  the  axial  component.  The  latter,  for  a  given  conductor, \nis  of  the  form  E  exp{  \u2014  Tz  -\\-  ift),  where  \u00a3  is  a  solution  of  the  differen- \ntial equation\n\nHere  X  and  m  are  the  electrical  conductivity  and  the  magnetic  perme- \nability of  the  particular  conductor,  measured  in  absolute  electro- \nmagnetic units,  and  \u00a3  is  a  function  of  r  alone.\n\nFor  the  frequencies  in  which  we  are  interested  it  may  be  shown \nthat  T-/4Tr\\ijLp  is  exceedingly  small,  so  that  (2)  may  be  written\n\nWe  will  designate  by  the  subscript  j  all  quantities  pertaining  to  the \nj'^  conductor,  counting  from  the  axis.  The  solution  of  (3)  for  this \nconductor  may  then  be  written\n\nwhere  /<,  and  K^  are  Bessel  functions  of  zero  order,  Aj  and  Bj  are \narbitrary  constants  and\n\nwhere  the  prime  indicates  differentiation  with  respect  to  pj.  Taking \nthe  line  integral  of  both  sides  of  (5)  around  circular  paths  in  con- \nductor j  lying  close  to  the  inner  and  outer  surfaces  of  the  cylinder  we \nobtain\n\nThe  values  of  the  electric  field  intensity  at  the  inner  and  outer  sur- \nfaces of  the/''  conductor  can  be  written,  from  (4)\n\nWe  liave  now  succcecled  in  cxpressin;j;  tlu'  electric  forces  in  the \nconductors  as  linear  functions  of  the  currents  /i  .  .  .  /\u201e,  the  coeffi- \ncients being  of  the  nature  of  inipxxlances,  by  a  method  which  is  simply \nan  application  of  the  principle  of  continuity  of  magnetic  field  intensity. \nThe  remaining  boundar\\-  condition,  continuity  of  the  tangential \ncomponent  of  electrical  field  intensit\\-  gives,  where  two  consecutive \ncylinders  are  in  electrical   contact,\n\nThis  gives  m  relations  between  the  n  currents  of  the  system,  m \nbeing  the  number  of  contacts  between  successive  cylinders.  In  the \ncase  where  the  j  and  (_;  +  l)st  conductors  are  separated  by  a  layer \nof  dielectric  material,  a  relation  between  the  boundary  values  of \nelectric  field  intensity  may  be  obtained  as  follows:\n\nis  the  potential  difference  between  the  j  and  {j  +  l)st  conductors, \nin  the  sense  employed  in  ordinary  circuit  theory.  If  we  now  apply  the \nlaw\n\nwhere  Qj  is  the  charge  on  the  f^  conductor  and  kj  is  the  dielectric \nconstant  of  the  medium,  whence,\n\nwhere   the   last   term   represents   the   leakage   current,   gj   being   the \nspecific  conductivity  of  the  dielectric. \nFrom  (13)  we  have\n\nAn  equation  of  this  sort  may  be  obtained  for  each  layer  of  dielectric \nand  these  combined  with  equations  (9)  and  the  condition  that  the \nelectric  field  intensity  in  the  sea  water  must  vanish  at  infinity,\n\ngive  n  relations  between  /i  .  .  .  /\u201e.     In  order  that  these  shall   be  con- \nsistent, the  determinant  of  the  coef^cients  must  vanish.\n\nThis  is  an  equation  in  F^  of  degree  equal  to  the  number  of  dielectric \nlayers;  consequently,  there  are  as  many  independent  modes  of  propaga- \ntion in  the  system  as  there  are  branches  in  the  network  of  conductors. \nFrom  this  point  the  method  of  determining  the  behavior  of  the \nsystem  depends  upon  conditions  in  the  particular  problem.  For  the \ncase  where  there  are  k  dielectric  layers  separating  the  conductors \ninto  ^  +  1  groups  the  current  on  the  j'^  group  may  be  written  in  the \nform\n\nwhere   T^  .  .  .  T^k  are  the  k  roots  of  the  determinant   (17)   and    Aji \n,  .  .  A^',,  Bji  .  .  .  Bp  are   constants.     These  constants  are  not  all  in-\n\ndependent,  however,  since,  for  each  value  of  F,  Fi  for  instance,  there \nexist  k  relations  of  the  form  (16)  which  the  corresponding  set  of \nconstants  Aw,  A21,  \u25a0  \u25a0  \u25a0  Aki  must  satisfy.  The  remaining  2k  inde- \npendent constants  can  then  be  determined  from  a  knowledge  of  the \nconditions  at  the  terminals  of  the  conductors.\n\nIt  is  important  to  observe  that  the  transmission  characteristics  of \na  system  of  coaxial  conductors  are  influenced  to  a  great  extent  by \nthe  manner  of  connecting  the  various  members  of  the  system. \nAnomalies  in  the  impedance  of  a  complicated  network  such  as  a \nsubmarine  cable  with  several  conducting  sheaths  in  the  return  path, \nmay  often  be  traced  to  lack  of  proper  connections  between  the  sheaths, \nor  to  faulty  joints.\n\nThe  submarine  cable  armored  with  a  continuous  coaxial  sheath, \nas  shown  in  Fig.  2,  is  a  particular  case  of  the  foregoing,  and  one  which \npresents  a  clearer  idea  of  the  physical  significance  of  the  various  steps \nin  the  general  theory.     There  are  only  two  groups  of  conductors,  the\n\nfirst  consisting  of  the  core  conductor,  and  the  second  comprising  the \niron  sheath  and  the  sea  water,  the  two  groups  being  separated  by  the \ninsulating  material  and  the  layer  of  jute.  Consequently,  there  is \nonly  one  mode  of  propagation,  and  the  analysis  is  considerably  sim- \nplified.\n\nThe  jute  is  assumed  to  contain  sufficient  sea  water  so  that  although \nit  conducts  practically  no  current  axially,  it  maintains  equality  of \npotential  between  the  outer  surface  of  the  gutta  percha  and  the  inner\n\nwhere  \u00a3\"i  and  E'-i  are  the  values  of  electric  field  intensity  at  the \nouter  surface  of  the  core  conductor  and  the  inner  surface  of  the  iron, \nrespectively,  V  is  the  potential  difTerence  between  these  two  surfaces, \nand  $  is  the  magnetic  flux  threading  unit  length  of  the  gutta  percha \nand  jute.     Also,  from  (14)\n\nwhere  gu  and  ki2  are  the  electrical  constants  of  the  gutta  percha, \nand  b  is  the  external  radius  of  the  core.  It  is  evident,  that  G  and  C \nare  respectively  the  leakage  and  capacity  of  unit  length  of  the  cable. \nTherefore,  from  (1),\n\nwhere  R  and  L  are  the  resistance  and  inductance  of  unit  length  of  the \ncable,  including  the  sea  return.     Equation  (18)  may  then  be  written\n\nwhere  Zi  may  be  termed  the  \"  internal  impedance  \"  per  unit  length \nof  this  conductor.     In  fact,  when  we  place  Vi  =  0  in  (8)  we  obtain\n\nwhich  is  the  usual  formula  for  the  internal  impedance  of  a  c\\lindrical \nconductor. \nSimilarly\n\nminus  sign  being  due  to  the  fact  that  the  current  in  the  return  is \nin  the  negative  direction  of  z.\n\nwhere  I2  is  the  current  in  the  iron  sheath.  The  value  of  this  current \ncan  be  found  by  applying  the  condition  of  continuity  of  electric \nfield  intensity  at  the  common  surface  of  the  iron  and  the  sea  water, \nas  in  equation  (9).     This  gives\n\nin  which  I3  is  the  current  in  the  sea  water.  From  (8)  it  can  be  seen \nthat  Z33  =  0,  since  X3  =  00 ,  therefore\n\nas  the  internal  impedance  of  the  return  conductor.     The  resistance \nand  reactance  per  unit  length  of  this  portion  of  the  circuit  are  then \nrepresented  by  the  real  and  imaginary  parts  of  (28)  respectively. \nWe  may  then  determine  R  and  L  from  the  formula\n\n62  and  ai  being  the  inner  radius  of  the  iron  and  the  outer  radius  of \nthe  core  conductor,  respectively.\n\nFor  purposes  of  comparison,   the  return   impedance  is  calculated \nfor  the  case  where  the  iron  armoring  is  absent,  the  return  current\n\nTVie   expression    for   Z2   simplifies   considerably.     The   electric   field \nintensity  in  the  sea  water  may  be  written,  from  (4),\n\nthe  term  in  J^  being  absent  in  order  to  permit  \u00a32  to  vanish  at  infinity. \nAlso,  from  (6),\n\nThe  resistance  and  inductance  of  the  sea  return  of  a  submarine \ncable  were  calculated  from  formula  (28).  employing  the  following \nvalues  for  the  constants:\n\nThe  armoring  was  then  assumed  to  be  replaced  by  sea  water,  and \nthe  resistance  and  inductance  of  the  cable  were  calculated  from  (33).\n\nIt  is  evident  from  these  curves  that  the  effect  of  the  iron  armoring \nis  to  increase  considerably  the  impedance  of  the  return  path.     The\n\n/4  -  Resistance  ~  Iron  Sheet \n8  -  Inductance  -  Iron  Sheat \nC  -  Resistance  -  No  Armor \nD  ~  Inductance   ~  No  Armor\n\nphysical  explanation  of  this  fact  is  that  the  iron  acts  as  a  shield  to \nscreen  from  the  sea  water  the  electromagnetic  effects  of  the  current \nflowing  in  the  cable  conductor.  Energy  is  dissipated  in  the  armoring \nand  is  prevented  from  spreading  out  through  the  surrounding  medium. \nThe  assumption  that  the  armor  wires  could  be  replaced  by  a  solid \ncylinder  of  iron  is,  therefore,  subject  to  question,  since  it  is  possible \nthat  the  larger  surface  area  of  the  assemblage  of  armor  wires,  and \nthe  gaps  between  these  wires  may  be  effective  in  diminishing  the \nenergy  dissipated  in  the  armoring  and  consequently  diminishing  the \nscreening  effect.     This  problem  is  investigated  in  the  following  section.\n\nThe  physical  system  under  consideration  is  shown  schematically \nin  cross-section  in  Fig.  4,  and  consists  of  an  insulated  conductor  and\n\nprotective  covering  of  jute,  surrouiuied  !)>\u2022  a  riiij^  of  N  armor  wires \nimmersed  in  sea  water.  The  method  of  solution  is  essentially  similar \nto  that  given  in  the  preceding  pages,  and  consists  in  determining  the \nvalues  of  electric  field  intensity  at  the  outer  surface  of  the  core  con- \nductor and  the  inner  surface  of  the  return  conductor,  from  which  the \ninternal  impedances  of  the  two  conductors  can  be  found.\n\nThe  main  dif^culty  in  the  analysis  is  caused  by  the  lack  of  uniaxial \nsymmetry  in  the  return  conductor.  This  was  overcome  by  employ- \ning a  method  developed  by  one  of  the  authors^  in  a  study  of  trans- \nmission in  parallel  wires.\n\nThe  electric  field  intensity  in  the  sea  w^ater  satisfies  the  differential \nequation\n\nAssuming  that  the  current  distribution  in  the  core  conductor  is \nindependent  of  the  angle  0,  that  is,  neglecting  the  individual  char- \nacter of  the  armor  wires  only  in  their  efifect  on  the  current  distribu- \ntion in  the  core,  the  effect  due  to  the  current  in  the  core  is  represented\n\n'\"Wave  Propagation  over  Parallel  Wires;  The  Proximity  Effect.\"  John  R. \nCarson,  PJiil.  Mag.,  vol.  xli,  p.  607  (1921).\n\nby  the  first  term  of  such  a  series,  and  the  total  field  intensity  may  be \nwritten\n\nPj  and  4:j  being  referred  to  the  axis  of  wire  j,  as  shown  in  Fig.  5.  That \nis,  the  resultant  field  is  expressible  as  a  set  of  waves  centered  on  the \naxis  of  the  cable  and  the  axes  of  the  N  armor  wires.\n\nIn  the  neighborhood  of  the  armor  wires  the  arguments  of  the  Bessel \nfunctions  are  sufficiently  small  ^  to  permit  of  the  approximations\n\nmagnetic  intensity  in  the  sea  water  can  be  obtained  by  differentiation.\n\nX  and  y.  being  the  electrical  conductivity  and  the  magnetic  perme- \nability, respectively,  of  the  material  of  the  armor  wire.  The  quan- \ntities a  and  4>  are  centered  on  the  axis  of  the  wire.\n\nIn  order  to  determine  the  coefficients  A,  B^,  Bi,  \u2014  ,  Co,  Ci,  \u2014  we \nmake  use  of  the  fact  that  the  electric  and  the  magnetic  field  intensi- \nties are  continuous  at  the  surface  of  the  wire.  It  is  obvious,  however, \nthat  nothing  can  be  learned  by  equating  (35)  and  (36)  since  they \nare  formally  dissimilar.  We  therefore  transform  \u25a0*  the  various  terms \nof  (35)  to  a  common  axis  which  coincides  with  the  axis  of  one  of  the \narmor  wires,  hereafter  called  wire  \"  zero,\"  and  the  electric  field \nintensity  in  the  sea  water,  close  to  the  surface  of  the  armor  wire,  is\n\nThe  tangential   magnetic  field  intensity  in   the  sea  water  at  the \nsurface  of  wire  \"  zero  \"  is,  therefore,\n\nTo  satisfy  the  condition  of  continuity  of  electric  and  magnetic \nfield  intensities  at  the  surface  of  the  armor  wire  it  is  necessary  that \nthe  coefftcients  of  the  corresponding  terms  of  (36)  and  (38)  and  of \n(37)  and  (41)  be  equal.     This  gives\n\nFrom  these  expressions  the  quantities  Bi  .  .  .  Ci  .  .  .  can  be  de- \ntermined.    Multiplying  (43)  by  tui  and  subtracting  (45)  gives\n\nwhich  expresses  C\u201e  in  terms  of  </\u201e.     Multiplying  (43)  by  f/\u201e'  (f)  and \n(45)  by  /\u201e  (f),  and  subtracting  gives\n\nFrom  the  infinite  set  of  simultaneous  equations   (47)   the  infinitely \nmany  variables  q\u201e  may  be  determined  in  terms  of  A  and  B^.^\n\nWe  have  thus  determined  the  arbitrary  constants  C^  \u25a0  .  .  C\u201e  and \nqi  .  .  .  g\u201e  (or  Bi  .  .  .  B\u201e)  as  functions  of  A  and  Bg.  It  remains  to \nexpress  the  latter  quantities  in  terms  of  physical  quantities.     If  /i\n\nWe  can,  therefore,  express  all  the  arbitrary  constants  as  linear,  homo- \ngeneous functions  of  !\u201e  and  Zi.\n\nHaving  shown  that  the  constants  A,  B^  .  .  .  of  the  series  (35)  are \nproportional  to  I^,  we  can  express  the  electric  field  intensity  at  the \ninner  surface  of  the  return  conductor  in  the  form\n\nThe  computation  of  Z2  is  facilitated  by  transforming  the  terms  of \n(35)  to  the  axis  of  the  core  conductor  ^  and  placing  r  =  c  \u2014  a.  We \nthus  obtain\n\nE2=-Z2lo={A+NBo)K-A\\og{c-a)-NBo\\ogc-N{qi-q2  +  qz...) \n+  (terms  containing  cos  d,  cos  26,  etc.,  as  factors).     (53)\n\nWe  have,  by  applying  the  curl  law  to  an  elementary  contour  which \nlinks  the  core  conductor  and  the  return.\n\n>o  and  Ho  being  the  electrical  constants  of  the  core  conductor  and \nOo  its  radius.  The  value  given  above  for  $12  holds  only  for  the  con- \ntour on  which  \u00a32  is  independent  of  the  angle  d,  that  is,  when  the  terms \nof  (53)  that  contain  cos  6,  cos  26,  etc.,  vanish.  The  value  of  Z2  to \nbe  used  in  (54)  is  therefore  determined  from\n\nwhere  R  and  L  are  the  resistance  and  inductance  per  unit  length  of \nthe  cable,  including  the  sea  return. \nWe  have  then  from  (54),\n\n(1)  Determine  from  (47)  the  quantities  qi  .  .  .  q\u201e  in  terms  of  A  and \nBo,  and  then  in  terms  of  /i  and  /\u201e  by  (49)  and  (50).\n\n(4)  Eliminate  /i  from  these  two  relations,  thus  obtaining  \u00a32  in \nterms  of  I^-     Then  Zo  =  \u2014  E2/ 1^.\n\n(5)  Substitute  this  value  of  Z2  and  the  value  of  Zi  calculated \nfrom  (55)  in  equation  (58).\n\n(6)  The  resistance  and  the  inductance  per  unit  length  of  the  cable \nmay  then  be  determined  from  the  real  and  imaginary  parts  of  the \nlatter  equation.\n\neffect  of  the  presence  of  the  iron  upon  the  resistance  of  the  return \nconductor  is  still  noticeable,  although  it  is  much  less  than  in  the  case \nof  the  continuous  iron  sheath.  The  reason  for  this  is  evident  after \ninspection  of  the  curves  of  Fig.  6,  which  show  the  percentage  of  return \ncurrent  carried  by  the  armor  in  the  two  cases.  Especially  at  the \nlower  frequencies,  the  return  current  is  much  more  confined  by  the \ncontinuous  sheath  than  it  is  by  the  wires.\n\nAs  a  check  of  the  method,  the  resistance  and  inductance  of  the \nSeattle-Sitka  cable  of  the  United  States  Signal  Corps  were  calcu- \nlated for  frequencies  in  the  range  50  to  600  cycles  per  second,  and\n\nA- Resistance' Small   Diameter  Armor  Ring. \nB- Resistance 'Large  Diameter  Armor   Ring \nC  -  Inductance  ~  Small  Diameter  Armor  Ring. \nD- Inductance -Large  Diameter  Armor   Ring. \nO'Experimental  Values\n\nthe  values  so  obtained  were  then  compared  with  the  results  of  meas- \nurements recently  made  upon  this  cable. ^  The  constants  used  in \nthe  calculations  were  as  follows:\n\n'  \"The  Use  of  Alternating  Currents  for  Submarine  Cable  Transmission,\"  Frederick \nE.  Pernot,  Jour,  of  the  Franklin  Instilute,  vol.  190,  p.  323,  1920.\n\nOwing  to  lack  of  information  concerning  the  mean  radius  of  the \nring  of  armor  wires,  two  sets  of  data  were  computed  employing  the \nvalues  c  =  0.6148  and  c  =  0.920,  which  correspond,  respectively,  to \nzero  and  maximum  separation  of  the  armor  wires.\n\nThe  results  of  the  calculations  are  shown  in  Fig.  7.  The  experi- \nmental values  are  indicated  by  small  circles,  and  agree  well  with  the \ntheoretical    values    throughout    the    range    of    frequencies.     The    re-\n\nsistance  of  the  sea  return  increases  most  rapidly  in  the  region  of \nfrequencies  used  in  ordinary  telegraphy,  0  to  100  cycles  per  second. \nIn  this  range  the  inductance  of  the  cable  also  has  its  greatest  values, \nand  these  two  effects  have  considerable  influence  in  determining  the \ntransmission  characteristics  of  the  cable.\n\nThe  percentage  of  the  return  current  that  is  carried  by  the  armor \nwires  is  shown  in   Fig.  8.\n\nAs  was  previously  pointed  out,  the  effect  of  the  shielding  action \nof  the  iron  armor  of  a  submarine  cable  is  to  diminish  the  electro- \nmagnetic field  which  is  propagated  through  the  sea  water,  and  which \ngives  rise  to  the  return  current.  Combined  with  this  effect  is  the \nshielding  action  of  the  sea  water  adjacent  to  the  cable,  upon  the  dis- \ntant portions.  The  total  shielding  effect  increases  with  the  fre- \nquency until  a  point  is  reached  where  practically  the  whole  of  the \nreturn  current  is  carried  by  the  armor  wires.\n\nSeveral  remedies  have  been  suggested  for  diminishing  the  damping \neffect  of  the  armor  wires.  It  can  be  proved,  for  example,  that  for  a \ngiven  size  of  core  and  .weight  of  armor,  the  number  and  size  of  armor \nwires  can  be  chosen  so  as  to  give  a  minimum  value  of  return  im- \npedance. A  proper  choice  of  the  electrical  constants  of  the  ma- \nterial of  which  the  armor  is  constructed  would  also  be  of  advantage, \nsince  the  return  impedance  is  somewhat  larger  for  iron  than  it  is  for \nmaterial  of  higher  or  lower  conductivity.\n\nAnother  method  of  diminishing  the  return  impedance,  which  has \nbeen  used  in  practice,  is  to  wrap  the  cable  core  with  a  number  of  con- \ncentric layers  of  conducting  tape  before  it  is  covered  with  jute.  The \nreturn  current,  as  it  crowds  in  toward  the  core  with  increasing  fre- \nquency, will  then  have  a  path  of  comparatively  low  impedance,  and \nat  the  higher  frequencies  only  a  small  portion  of  the  current  will  be \ncarried  by  the  armor  wires  and  the  sea  water.  The  impedance  of \nthe  return  path  can  be  calculated  for  this  case  by  the  methods  given \nin  the  preceding  pages.  The  following  table  compares  the  values \nof  the  resistance  of  the  return  conductor  calculated  by  three  different \nmethods,  and  determined  experimentally,  for  a  cable  provided  with \na  brass  tape  5  mils  in  thickness.\n\nThe  experimental  values  are  the  results  of  a  series  of  measure- \nments made  by  the  Department  of  Development  and   Research  of\n\n*  This  is  an  empirical  formula  which  has  been  found  to  be  fairly  close  in  most \ncases.  The  correction  factor  suggested  itself  in  that  it  takes  care  of  the  increased \nsurface  of  the  armor  wires,  as  compared  with  the  corresponding  continuous  sheath.\n\nthe  American  Telephone  and  Telegraph  Company  upon  the  Victoria- \nVancouver  submarine  cable.  The  calculated  values  were  obtained \nby  both  the  approximate  and  the  exact  methods,  discussed  in  the \npreceding  pages,  in  which  the  armor  of  the  cable  is  treated,  respec- \ntively, as  a  continuous  sheath  and  as  a  ring  of  wires.  The  modifica- \ntions which  must  be  introduced  to  include  the  effect  of  the  conduct- \ning tape  are  outlined  in  the  discussion  of  the  general  theory.  The \nagreement  between  the  calculated  and  the  measured  values  of  return \nresistance  proves  that  the  method  developed  in  the  present  paper \nis  accurate  even  at  the  highest  frequencies  employed  in  telephony.\n\nThe  Bessel  Functions  of  zero  order  of  the  first  and  second  kinds, \nJo  (p)  and  Ko  (p),  used  in  the  preceding  work  are  all  to  a  complex \nargument  p  =  ^'gvTwhere  5  is  a  real  number  and  i  =  V-  1.  The \nfollowing  formulas^  may  be  used  for  determining  the  values  of  these \nfunctions:\n\nThe  reports  of  the  British  Association  for  1912  and  1915  give  the \nvalues  in  this  range  of  the  functions  ber  q,  ber'  q,  bei  q,  bei'  q,  ker  g, \nker'q,  kei  q,  kei'q  which  are  defined  by  the  relations\n\n\u00bb  It  is  to  be  noted  that  this  approximation  for  Ko  (p)  differs  from  the  expression \nused  by  J.  J.  Thomson,  \"Recent  Researches  in  Electricity  and  Magnetism,\"  p.  263. \nThomson's  formula  (2)  from  which  his  approximation  was  derived,  contains  a \nnumber  of  errors  and  should  read\n\nIn   problems  involving  Fourier-Bessel  expansions  it  is  sometimes \nnecessary  to  transform  quantities  of  the  form\n\nfrom  the  system  of  coordinates  pj,  (pj  to  the  systems  p,  4>  or  r,  d  which \nare  related  as  shown  in  Fig.  5.\n\nThe  vectors  Zj,  Z  and  C,  as  may  be  seen  from  Fig.  5,  have  the \nlengths  Pj,  p  and  c,  respectively,  and  the  directions  indicated  by  the \narrows.\n\nThe  values  qi,  52  \u2022  \u2022  \u2022  may  be  determined  by  a  method  of  approxi- \nmations, g\u201e  being  the  limit  of  the  sequence\n\nThis  method,  however,  while  formally  simple  and  direct  is  not \nusually  well  adapted  for  numerical  solution.  For  all  sizes  of  armor \nwire  and  for  frequencies  of  practical  importance  the  argument  f  in \nthe  expression  (48)  is  small  compared  with  /i  and  the  quantities,\n\nare  all  nearly  unity.     This  suggests  the  use  of  the  following  method \nof  solution  of  equations  (47).\n\nin  which  Cu,  etc.,  are  numerics.  This  solution  is  effected  by  retain- \ning a  finite  number  of  equations  and  an  equal  number  of  variables \nand  solving  by  the  usual  methods.  It  will  be  found  that  except  in \nextreme  cases,  a  very  good  approximation  can  be  gotten  by  ignoring \nall  the  p's  except  the  first  four.  The  qs  may  then  be  obtained  by \nthe  relation\n\nThis    system    is    easily    adapted    to    solution    by    successive    ap- \nproximations,\n\nC\u201ei,  etc.,  being  the  numerical  coefficients  which  appear  in  the  ex- \npressions for  pi,  p2  .  .  .  .\n\nSynopsis:  The  frequency  distribution  of  energy  in  speech  has  been  de- \ntermined for  six  speakers,  four  men  and  two  women,  for  a  50-syllable \nsentence  of  connected  speech,  and  also  for  a  list  of  50  disconnected  syllables. \nThe  speech  was  received  by  a  condenser  transmitter  whose  voltage  output, \namplified  3,000  fold,  was  impressed  on  the  grids  of  twin  single  stage  amplifi- \ners. The  unmodified  output  of  one  of  these  amplifiers  was  measured  by  a \nthermocouple  and  was  a  known  function  of  the  total  energy  received  by  the \ntransmitter,  corrections  being  made  for  the  slight  variation  with  frequency \nof  the  response  of  the  circuit.  The  output  of  the  other  amplifier  was  limited \nby  a  series  resonant  circuit  to  a  narrow  band  of  frequencies,  the  energy  in \nthis  band  being  measured  by  a  second  thermocouple.  The  damping  of  the \nresonant  circuit  was  so  chosen  that  sufficient  resolving  power  and  suiificient \nenergy,  sensitiveness  were  obtained  over  the  range  from  75  to  5,000  cycles \nper  second;  and  23  frequency  settings  were  made  to  cover  this  range.  For \neach  syllable  simultaneous  readings  were  recorded  on  the  two  thermocouples \nat  each  frequency  setting.  The  consecutive  syllables  were  pronounced  de- \nliberately by  each  speaker,  maintaining  as  nearly  as  possible  the  normal \nmodulation  of  the  voice.  Corrections  were  applied  to  offset  the  unavoidable \nvariations  in  total  energy  incidental  to  repetition  of  a  given  syllable. \n13,800  observations  were  made  for  all  speakers.  The  energy  distribution \ncurves  obtained  are  essentially  the  same  for  connected  as  for  disconnected \nspeech,  and  indicate  that  differences  between  individuals  are  more  important \nthan  variations  due  to  the  particular  test  material  chosen.  A  composite \ncurve  drawn  from  the  individual  curves  shows  a  great  concentration  of \nspeech  energy  in  the  low  frequencies,  a  result  which  would  not  be  expected \nfrom  data  previously  published  by  others.  The  actual  results  contain \na  factor  due  to  standing  waves  between  the  speaker's  mouth  and  the \ntransmitter,  a  complication  always  present  in  telephoning;  this  could  not \nbe  eliminated.\n\nThe  rate  of  energy  output  in  speech  for  the  normally  modulated  voice,  was \ndetermined  from  the  readings  for  total  energy  and  was  found  to  be  about \n125  ergs  per  second.\n\nIN  the  study  of  speech  and  its  reproduction  by  mechanical  apparatus \nit  is  necessary  to  consider  its  composition  from  several  different \npoints  of  view.  We  desire  first  of  all  to  know  the  actual  frequency \ndistribution  of  the  total  energy  in  speech,  as  well  as  the  separate  dis- \ntributions for  each  individual  sound.  We  also  desire  to  know  the \napparent  distribution  of  energy,  that  is,  the  distribution  as  perceived \nby  the  ear.  Finally,  we  wish  to  know  the  importance  of  each  fre- \nquency, that  is,  the  contribution  to  \"articulation\"  or  \"quality\" \nin  the  exact  reproduction  of  speech  which  can  be  traced  to  the  energy \nof  each  elementary  band  of  frequencies  in  the  speech  range.  In  all \nthree  cases  certain  frequency  functions  are  used  to  represent  these \ndistributions.  The  advantage  of  considering  these  difTerent  frequency \ndistribution  functions  separately  has  already  been  indicated  by  one \nof  the  present  writers.^\n\n1  Reprinted  from  The  Physical  Review,  N.S.,  Vol.  XIX.  No.  3,  March,  1922. \n\u00bb\"The  Composition  of  Speech,\"  Phys.  Rev.,  X,  p.  74,  1917.\n\nIn  our  judgment  the  most  important  of  these  data  of  speech  study  is \nthe  actual  energy  distribution,  considering  speech  as  \"a  continuous  Mow \nof  distributed  energy, \"  in  accorchmce  with  the  ideas  expressed  in  the \nearlier  paper.  The  present  paper  offers  a  determination  of  this \nfundamental  factor.\n\nTo  determine  the  energy  distribution  in  speech  to  a  high  degree  of \naccuracy  it  would  be  desirable  to  analyze  a  certain  amount  of  con- \nnected speech  and  take  a  time  average  of  the  energy  distribution  of \nthe  whole.  This  is  not  feasible  at  the  present  time,  but  a  very  close \napproach  to  this  result  has  been  made.  The  method  consists  in \nanalyzing  the  speech  waves  as  impressed  on  a  condense  transmitter, \nusing  a  tuned  circuit  to  transmit  narrow  frequency  bands  of  energy  and \npronouncing  the  separate  syllables  of  the  connected  speech  so  slowly \nthat  the  kick  of  a  direct  current  galvanometer  connected  to  an  A.  C. \nthermocouple  can  be  separately  read  for  each  syllable.  Using  a  suit- \nable calibration  for  the  whole  apparatus,  the  magnitude  of  this  kick \ncan  be  interpreted  in  terms  of  the  time  integral  of  the  energy  at  a \nparticular  frequency  setting  for  each  syllable.  A  mean  of  the  read- \nings for  all  the  syllables  in  the  \"speech\"  at  any  frequency  setting \ngives  the  relative  energy  at  that  frequency.\n\nThe  present  method  is  a  modification  of  an  earlier  method  in  which \napproximate  analyses  of  speech  sounds  were  made,  using  a  condenser \ntransmitter,  tuned  circuit,  an  amplifying-rectifying  circuit,  and  ballis- \ntic -galvanometer.  The  method  is,  however,  much  improved  as  we \nnow  have  very  accurately  calibrated  condenser  transmitters  of  better \ndesign,^  and  a  great  deal  of  care  has  been  taken  to  calibrate  the  suc- \ncessive elements  of  the  train  of  apparatus,  and  increase  the  resolving \npower.\n\nSound  waves  emitted  from  the  mouth  of  the  speaker  are  allowed  to \nfall  upon  the  diaphragm  of  a  condenser  transmitter,  connected  in  the \nconventional  manner  to  the  input  of  a  three-stage  amplifier.  The \noutput  of  this  is  impressed  upon  the  input  circuits  of  twin  single \nstage  amplifiers,  potentiometers  being  interposed  to  permit  regula- \ntion of  the  grid  voltages  of  the  twin  amplifier  tubes.\n\nThe  output  circuits  of  the  fourth  stage  consist  of  the  high  windings \nof  two  step  down  ironclad  transformers.  These  step  down  trans- \nformers have  a  voltage  ratio  of  1 1 :  1  and  are  designed  to  work  between \nimpedances  of  6,000  and  50  ohms.  The  low^  impedance  winding  of \none  of  these  transformers  operates  into  a  thermocouple  heater  of,\n\n'  The  present  design  of  the  condenser  transmitter  and  its  calibration  are  fully \ntreated  in  a  paper  by  Dr.  E.  C.  Wente  which  will  appear  shortly  in  this  Journal.\n\nroughly,  40  ohms  resistance.     The  low  side  of  the  other  transformer \noperates  through  a  tuned  circuit  into  a  similar  thermocouple  heater.\n\n-Circuit  Used  for  the  Analysis  of  Speech.     (The  Usual  Details  of  the  Three- \nStage  Amplifier  Are  Not  Shown)\n\nWhen  the  diaphragm  of  the  condenser  transmitter  is  set  in  vibration \nby  speech  a  current  made  up  of  a  range  of  frequencies  flows  in  the \nheater  of  thermocouple  I.,  while  the  heater  of  thermocouple  II  is \ntraversed  only  by  such  a  band  of  frequencies  as  the  resonant  circuit \nallows.     Fig.  2  shows  a  number  of  typical  resonance  curves  obtained\n\nin  the  course  of  calibrating  this  apparatus.  These  curves  are  such \nthat  the  tuned  circuit  functions  as  a  filter  transmitter  only  a  narrow \nregion  of  frequencies.     One  side  of  the  twin  amplifier  transmits  the\n\nentire  electrical  response  of  tl.c  system;  the  other  side  suppresses \nall  save  a  band  of  frequencies,  the  center  of  this  hand  being  shifted \nby  resetting  the  condenser  and  inductometer.\n\nHaving  chosen  for  analysis  a  piece  of  connected  discourse,  the \nspeaker  utters  the  successive  syllables  separately  but  as  nearly  as \nmay  be  with  the  same  inflection  and  volume  as  if  the  syllables  were \ncontinuously  spoken.  Two  observers  record  the  readings  of  microam- \nmeters  in  the  couple  circuits  of  the  thermocouples.  One  of  these \ninstruments  gives  a  deflection  corresponding  to  the  total  energy  of \nthe  syllable  uttered;  the  deflection  of  the  other  instrument  corre- \nsponds to  the  energy  of  the  syllable  lying  within  the  limits  of  trans- \nmission of  the  tuned  circuit.\n\nPreliminary  experiments  were  carried  out  to  determine  the  relation \nbetween  momentary  deflection  read  on  the  microammeter,  and  the \ncurrent  momentarily  flowing  in  the  thermocouple  heater.  Currents  of \ndifferent  values  were  caused  to  flow  for  intervals  of  time  varying \nfrom  0.2  second  to  1.2  seconds,  and  the  deflections  were  found  nearly \nproportional  to  the  product  of  current  squared  and  time  interval; \nthis  proportionality  was  most  nearly  exact  when  the  current  was \nweak  and  the  time  intervals  short.  For  all  cases  likely  to  be  dupli- \ncated in  the  speech  analysis  work  the  error  might  be  taken  as  about \n5  per  cent,  a  quantity  small  in  comparison  with  the  inevitable  un- \ncertainties due  to  other  causes.\n\nQuite  low  damping  is  attained  in  the  resonant  circuit.  The  values \nof  inductance  used  ranged  from  0.20  to  0.66  henry  and  the  total  re- \nsistance of  the  circuit\u2014 transformer  winding,  inductometer  coil, \nthermocouple  heater\u2014 is  of  the  order  of  100  ohms.  The  damping \nthus  ranges  from  75  to  250.\n\nA  switch  is  so  introduced  that  it  is  possible  to  include  in  series  with \nthe  thermocouple  the  resonant  circuit,  or  replace  it  by  a  non-in- \nductive resistance  whose  value  is  approximately  that  of  the  A.  C. \nresistance  of  the  inductometer  winding.  With  the  tuned  circuit \nexcluded,  an  alternating  current  of  suitable  magnitude  is  caused  to \nflow  in  the  thermocouple  heater;  the  tuned  circuit  is  then  substituted \nand  the  new  value  of  the  current  observed,  the  input  voltage  remaining \nconstant.  The  ratio  of  current  squared  \"tuned  circuit  in\"  to  current \nsquared  \"tuned  circuit  out\"  is  plotted  against  frequency,  yielding  a \ncurve  for  energy  transmission.\n\nTwenty-three  bands  in  all  were  considered  adequate  for  the  analysis \nof  energy  distribution  in  speech;  the  centers  of  these  were  at  75,  100, \n200,  300  cycles,  400  to  3,200  cycles  by  steps  of  200;  3,500,  4,000,  4,500,\n\n5,000  cycles  per  second.  Beyond  5,000  cycles  per  second,  the  energy \nis  so  low  as  to  be  impossible  of  measurement  with  the  apparatus  used. \nA  Weston  Type  322  microammeter  recorded  the  couple  current  for  the \ntuned  circuit  side  of  the  twin  single  stage  amplifier.  With  this  instru- \nment and  the  thermocouple  used,  0.2  microampere  in  the  couple  circuit \ncorresponds  to  one-quarter  of  a  milliampere  in  the  heater,  and  this  is \nthe  lowest  readable  deflection  of  the  Weston  instrument.\n\nThree  corrections  have  to  be  made,  the  first  being  the  correction \nfor  varying  volume.\n\nSimultaneous  observations  are  made,  at  each  setting  of  the  tuned \ncircuit,  of  the  filtered  and  the  unfiltered  energy  of  each  syllable.  It \nis  not  possible  to  utter  a  given  syllable  with  the  same  intensity  and \nat  the  same  distance  from  the  transmitter  for  every  one  of  twenty-\n\nFig.  3 \u2014 Illustrating  Correction  of  ObservaLions,  Necessary  Because  of  Variation  in \nResolving  Power  with  Frequency  Setting\n\nthree  times.  Accordingly,  the  \"unfiltered\"  readings  are  averaged  and \neach  of  the  filtered  readings  for  each  syllable  reduced  from  the  value \nactually  observed  to  the  value  that  would  have  been  read  had  the \nvolume  and  distance  been  such  as  to  give  the  average  \"unfiltered\" \nreading.  This  procedure  is  quite  legitimate  if  it  be  granted  possible \nto  maintain  a  definite  composition  of  the  syllable  in  question  through- \nout the  changes  of  the  tuned  circuit  setting.\n\nA  second  correction  was  made  for  the  varying  area  of  tuned  circuit \ncurves.\n\nIn  Fig.  3  let  S{j)  be  the  speech  spectrum  determined  by  'deal \nmethods;  \"i?\"  the  transmission  curve  of  the  tuned  circuit,  set  for  a \nresonant  frequency/.  An  ideal  transmission  curve  would  be  a  rectangle \nwhen  plotted  in  this  figure,  of  height  \"/^\"  and  transmission  range  A/.\n\nThe  true  amount  of  energy  5(0  associated  with  frequency/,  and  the \nexperimentally  determined  value  which  we  may  call  (5/)  are  con- \nnected by  the  relation\n\nwe  may  take  for  all  practical  purposes  S{J)  =  Sif),  considering  the \nnarrowness  of  the  transmission  range.  We  must  therefore  find  the \nfactor  h^f,  proportional  to  the  area  of  each  tuned  circuit  curve  and \ndivide  the  energy  received  through  the  filtered  side  by  /?A/,  in  order  to \nobtain  S{f).  This  treatment  may  be  gone  through  for  each  syllable \nindividually,  but  it  is  more  convenient  to  sum  the  tuned  circuit  read- \nings for  all  the  syllables  used,  corrected  one  at  a  time  for  varying \nvolume,  and  then  apply  the  curve  area  correction  to  this  sum.\n\nA  third  correction  was  made  for  the  varying  frequency-sensitivity  of \nthe  whole  apparatus.  Thus  far  we  have  discussed  only  the  electrical \nenergy  in  the  output  circuit  of  the  fourth  stage.  It  remains  to  show  in \nwhat  way  this  is  related  to  the  mechanical  energy  of  the  diaphragm, \nand  this  in  turn  to  the  incident  sound  energy.\n\nThe  calibration  of  the  circuit  as  a  whole  was  made  by  introducing  a \nsmall  resistance  carrying  alternating  current  in  series  with  the  con- \ndenser transmitter,  thus  introducing  a  known  potential  drop  in  the \nundisturbed  input  mesh  of  the  circuit.\n\nAn  amplification  curve  is  appended  {A,  Fig.  4)  which  gives  to  an \narbitrary  scale  the  ratio  of  volts  output  to  volts  input  as  a  function  of \nfrequency,  for  the  system  as  actually  operated.  The  calibration  of  the \ncondenser  transmitter,  shown  in  Fig.  4,  Curve  C,  gives  the  open \ncircuit  voltage  of  the  transmitter  per  unit  pressure  on  the  diaphragm \nas  a  function  of  frequency.  The  product  of  these  curves  is  the  volts \noutput  per  unit  alternating  pressure  on  the  diaphragm,  and  the  square \nof  this  product,  curve  E  is  proportional  to  the  electrical  energy  output \nper  unit  sound  energy  incident  on  the  diaphragm,  if  we  assume  that \nthe  sound  energy  is  proportional  to  the  square  of  the  alternating \npressure.  This  point,  however,  requires  some  further  discussion, \nwhich  will  be  given  later  on.\n\nIt  is  plain  from  curve  E  that  the  response  of  the  system  is  a  maxi- \nmum of  frequencies  in  the  neighborhood  of  2,250  cycles.  If,  now,  the \nobservations  already  corrected  for  varying  volume  and  for  area  of \nresonance  curves,  are  subjected  to  further  correction  for  the  exaggera- \ntion of  these  frequencies,  it  is  possible  to  draw  a  curve  which  shall\n\nexhibit  the  mean  square  of  the  excess  pressure  on  the  diaphragm,  as  a \nfunction  of  frequency  in  the  voice  exciting  the  vibration.  We  obtain \nthis  corrected  curve  by  dividing  the  results,  after  the  first  and  second \ncorrections  above  have  been  made,  by  the  ordiiiates  of  curve  R.\n\nIn  order  to  investigate  the  possibilities  of  this  method  it  was  decided \nto  work  with  a  rather  short  piece  of  connected  speech,  and  to  use  a \nlimited  number  of  observers,  on  account  of  the  large  number  of  ob- \nservations which  are  required  for  each  separate  syllable.  With  six \nspeakers  (four  men  and  two  women)  each  pronouncing  the  test  sentence \nof  fifty  syllables  for  each  of  the  twenty-three  frequency  settings,  6,900 \nseparate  observations  were  required.  It  is  believed  that  representa- \ntive results  have  been  obtained  from  these  observations,  but  if  this \nis  not  the  case  then  some  method  of  graphical  registration  of  the  energy- \ntime  curve  of  speech  for  the  different  frequency  settings  must  be \napplied  in  order  to  handle  the  vast  amount  of  data  involved  in  work \non  an  appreciably  larger  scale.\n\n\"Quite  four  score  and  seven  years  ago  our  father  brought  forth  on  this  continent, \na  nice  new  nation,  conceived  in  liberty,  and  dedicated  to  the  proposition  that  all \nmen  are  created  equal.\"\n\nThe  two  italicized  words  were  added  to  the  first  sentence  of  the \n\"Gettysburg  Address\"  in  order  to  bring  the  total  up  to  fifty  syllables, \nand  improve  the  balance  between  the  vowel  sounds.\n\nplotted  so  that  J^  5  (/)  df\u2014  1  in  each  case.  In  Fig.  5  the  individual \ncurves  for  each  of  the  six  speakers  are  shown  on  a  small  scale ;  in  Fig.  6 \nthe  composite  curve  for  the  men  and  the  composite  curve  for  the \nwomen,  drawn  separately,  and  in  Fig.  7  the  composite  curve  for  all \nsix  speakers,  giving  the  data  of  curves  QA  and  65  equal  weight.\n\nThese  curves  are  very  similar  to  a  curve  obtained  by  Dr.  Fletcher  of \nthis  laboratory,  using  block  filters  and  based  on  the  simple  calling \nsentence  \"Now  we're  off  on  one.\"  A  general  consideration  of  this  fact \nand  of  the  data  shown  leads  us  to  believe  that  the  differences  between \ncurves  of  this  sort,  made  by  the  method  described  are  due  rather \nmore  to  differences  between  the  voices  of  the  individual  speakers \nthan  to  the  particular  piece  of  connected  speech  which  is  chosen,  pro- \nvided the  speech  is  of  reasonable  length.  The  differences  between \nthe  different  voices  are  so  marked  that  we  should  expect  them  to  remain \neven  though  we  used  as  test  material  a  connected  speech  ten  or  fifty \ntimes  as  long  as  the  sentence  used.\n\nAn  interesting  comparison  may  be  made  between  the  curves  shown \nfor  the  energy  distribution  of  \"continuous  speech\"  and  certain  specu- \nlative curves  previously  constructed  to  indicate  the  energy  distribu-\n\ntion.  One  of  these  curves  is  shown  in  Fig.  8.  Curve  A  was  constructed \nby  one  of  the  writers  in  1916  in  an  attempt  to  synthesize  the  energy \ncurve  from  the  energy  distributions  of  the  vowel  sounds,  using  the \nvowel  analyses  of  Dr.  Dayton  C.  Miller.  Curve  C  is  the  composite \n\"continuous  speech\"  curve  of  Fig.  7.    The  vowel  sounds  analyzed  by\n\nMiller  were  intoned  and  the  vowel  sounds  analyzed  by  us  were  spoken, \nbut  Miller's  work  seemed  to  show  that  there  was  no  essential  differ- \nence between  intoned  and  spoken  vowel  sounds.  There  is,  however,  a \nvery  noticeable  difference  between  Curve  A  and  Curve  C,  the  energy \nin  the  fundamental  tone  of  the  speaker's  voice  coming  out  much \nmore  strongly  in  Curve  C.  We  should  expect  that  our  improved  ap- \nparatus would  record  the  energy  in  the  lower  frequencies  more  cor- \nrectly than  the  apparatus  heretofore  used  but  as  we  used  different \ntest  material  (connected  speech  instead  of  disconnected  syllables  or \nvowel  sounds)  it  is  not  immediately  evident  which  of  these  two  factors \nis  responsible  for  the  differences  between  the  A  and  the  C  curves.\n\nIn  order  to  investigate  this  point  more  fully  the  testing  routine  for \nall  six  speakers  was  repeated,  using  instead  of  the  fifty-syllable  sen- \ntence, the  fifty  disconnected  syllables  of  one  of  the  standard  articula- \ntion testing  lists,  as  used  by  Dr.  Fletcher  ii  this  laboratory.  The \nresults  for  energy  distribution  are  shown  in  Fig.  9,  Curve  A  being\n\nthe  mean  energy  distribution  for  the  four  male  speakers,  using  the \nsyllables,  while  Curve  B  is  the  mean  energy  distribution  of  the  two \nfemale  speakers.  Curves  ^A  and  95  may  be  compared  with  Curves \n6-4  and  65  which  represent  the  sentence  of  continuous  speech.  The \ntwo  sets  of  curves  are  essentially  the  same  as  shown  in  Fig.  8,  C  and  B\n\nbeing  respectively  the  composite  curves  for  all  speakers,  using  con- \nnected and  disconnected  speech.\n\nSuch  small  differences  as  exist  between  Curves  C  and  B  of  Fig.  8 \nmay  probably  be  due  to  differences  in  the  distribution  of  the  vowel \nsounds  in  the  connected  and  disconnected  test  material.  This  dis- \ntribution is  given  in  the  following  table:\n\nThe  similarity  between  Curves  C  and  B  of  Fig.  8  is  evidence  of  the \ngeneral  reliability  of  the  method,  and  leads  to  two  rather  important \nconclusions.\n\nIn  the  first  place,  characteristic  results  have  been  obtained  for  a \ngiven  set  of  speakers,  using  two  different  types  of  test  materials.    This\n\nseems  to  show  that  the  choice  of  test  material  does  not  require  especial \nconsideration,  provided  it  is  of  sufficient  length.  It  seems  to  be  a \nmatter  of  rather  greater  importance  to  increase  the  number  of \nobservers.\n\nIn  the  second  place,  it  seems  that  for  the  actual  energy  distribution, \nthe  results  previously  obtained  from  the  vowel  analyses  are  definitely \nin  error,  in  that  they  show  relatively  little  energy  associated  with  the \nlower  voice  frequencies.\n\nThe  foregoing  treatment  provides  a  curve  showing  the  frequency \ndistribution  of  the  square  of  the  excess  pressure  on  the  diaphragm. \nIn  an  undisturbed  field  of  sound  energy  we  have  for  the  intensity\n\nin  which  p  is  the  mean  density  of  the  medium,  a  the  velocity  of  sound \nin  the  medium  and  P  the  maximum  excess  pressure.\n\nIt  remains  for  us  to  consider  in  how  far  the  results  obtained  represent \nthe  frequency  distribution  of  sound  energy  in  speech.\n\nDue  to  the  fact  that  at  frequencies  where  the  sound  wave-length  is \nshort  and  comparable  with  the  diameter  of  the  transmitter,  consid- \nerable reflection  takes  place,  and  the  pressure  on  the  diaphragm  is \nproportionately  greater  for  these  frequencies  than  for  those  which  are \nnot  accompanied  by  strong  reflection.  In  this  respect  again  the \nhigher  frequencies  provoke  the  greater  response  in  the  system.\n\nThe  following  experiment  was  tried  to  investigate  this  variation.  A \nwall  six  feet  square,  with  a  central  hole  to  fit  over  the  condenser  trans- \nmitter, was  brought  up  to  make  the  transmitter  a  part  of  a  plane  wall. \nThe  clearance  around  the  periphery  of  the  transmitter  was  tightly \nclosed,  and  reflection  was  to  be  expected  at  all  frequencies.  Where \ntotal  reflection  takes  place,  a  given  quantity  of  sound  energy  result? \nin  twice  the  alternating  pressure  on  the  diaphragm  as  when  no  reflection \noccurs.  That  is,  the  resulting  electrical  energy  observed  should  be  four \ntimes  as  great  for  total  reflection  as  for  no  reflection.  The  wall  was \nexpected  to  cause  reflection  at  all  frequencies,  and  the  experiment \nconsisted  in  reading  the  electrical  response,  with  and  without  the  wall, \nthe  condenser  transmitter  being -exposed  to  tones  of  frequencies  from \n200  to  10,000  cycles  per  second  under  definite  adjustments  of  the \nsupply  circuit  of  a  receiver  producing  this  tone.  When  the  frequency \nis  low,  little  reflection  takes  place  from  the  transmitter  standing  alone, \nand  bringing  up  the  wall  should  cause  a  great  increase  in  the  response\n\nof  the  system.  At  high  frequencies  the  transmitter  should  reflect \nnearly  as  much  alone  as  when  part  of  a  large  wall,  and  the  readings \nwith  and  without  the  wall  should  be  nearly  equal.  Plotting  ratio  of \nresponse  without,  to  response  wuth  the  wall  was  expected  to  yield  a \ncurve  which  could  be  used  to  make  the  final  reduction  of  electrical \noutput  to  incident  sound  energy,  and  so  permit  a  more  accurate  de- \ntermination of  the  spectrum  of  sound  energy  of  the  voice.\n\nThus,  the  curves  finally  obtained  show  no  more  than  the  frequency \ndistribution  of  energy  in  speech  in  terms  of  the  mechanical  energy  of  a \nmore  or  less  ideal  transmitter  diaphragm.  However,  this  information \nhas  its  value  because  in  any  given  configuration  of  transmitter, \nspeaker,  and  room,  there  is  a  definite  correspondence  between  the  sound \nenergy  of  the  voice  and  the  force  acting  on  the  diaphragm  on  which  it \nfalls,  and  in  telephony  at  any  rate  it  is  this  action  on  the  diaphragm \nwith  which  we  are  immediately  concerned.\n\nIn  conclusion  we  may  give  a  determination  of  the  total  energy  rate \nof  speech,  obtained  as  a  by-product  of  the  preceding  investigation. \nKnowing  the  calibration  of  the  system  in  absolute  units,  it  is  possible \nto  determine  the  alternating  pressure  on  the  condenser  transmitter \ndiaphragm  exposed  to  continuous  speech  from  the  normally  modulated \nvoice  under  the  conditions  of  the  experiment.  Using  the  mean  of  the \nvalues  obtained  with  9  observers  we  find  for  the  alternating  pressure \n11.3  dynes  per  sq.  cm.  (r.m.s.)  for  a  distance  of  2.5  cm.  from  mouth  to \ndiaphragm.  This  corresponds  to  an  energy  flow  of  3.2  ergs  per  sq. \ncm.  per  second.  Assuming  that  this  energy  flow  is  distributed  uni- \nformly over  a  hemisphere  of  2.5  cm.  radius,  we  may  take  125  ergs \nper  second  as  the  total  sound  energy  flow  from  the  lips  with  the \nnormally  modulated  voice.\n\nVARIOUS  phases  of  this  subject  have  received  serious  study  by \nphoneticians,  otologists,  and  physicists.  On  account  of  its  universal \ninterest,  it  has  received  attention  from  men  in  many  branches  of \nscience.  In  spite  of  the  large  amount  of  time  devoted  to  the  sub- \nject, the  progress  in  understanding  its  fundamental  aspects  has  been \nrather  slow.  At  the  present  time  the  physical  properties  which \ndifferentiate  the  various  fundamental  speech  sounds  are  understood \nin  only  a  very  fragmentary  way.  Some  very  interesting  and  pains- \ntaking work  has  been  done  on  the  physical  analysis  of  vowel  sounds, \nbut  the  results  to  date  are  far  from  conclusive.  Although  several \ntheories  have  been  advanced  to  explain  the  way  in  which  the  ear \ninterprets  sound  waves,  they  are  still  in  the  controversial  stage.\n\nThe  material  which  is  presented  here  is  the  result  of  an  investi- \ngation which  has  been  carried  on  in  the  Research  Laboratories  of \nthe  American  Telephone  and  Telegraph  Company  and  Western \nElectric  Company  during  the  past  few  years.\n\nTo  make  a  quantitative  study  of  speech  and  hearing  it  is  necessary \nto  obtain  the  speech  sounds  at  varying  degrees  of  loudness  and  with \ndefinitely  known  amounts  of  distortion.  The  main  reason  why  so \nfew  real  results  have  been  obtained  in  the  investigation  of  speech \nsounds  is  due  to  the  fact  that  it  is  extremely  difficult  to  change  the \nvolume  and  distortion  of  these  sounds  by  acoustic  means.  Due  to \nrecent  developments  in  the  electrical  transmission  of  speech  it  is \npossible  to  produce  the  equivalent  of  these  changes  by  electrical  means. \nFor  this  purpose  a  telephone  system  was  constructed  which  repro- \nduced speech  with  practically  no  distortion.  It  was  arranged  so  that \nby  means  of  distortionless  attenuators  the  volume  of  reproduced \nspeech  could  be  varied  through  a  very  wide  range,  and  so  that  by \nintroducing  various  kinds  of  electrical  apparatus  the  transmitted \nspeech  wave  could  be  distorted  in  definitely  known  ways.\n\nA  method  was  developed  for  measuring  quantitatively  the  ability \nof  the  ear  to  interpret  the  transmitted  speech  sounds  under  different \nconditions  of  distortion  and  loudness.  By  choosing  these  conditions \nproperly,  considerable  information  was  gained  concerning  both  speech \nand  hearing.     This  indirect  method  of  attack  has  a  distinct  advantage\n\n*  Presented  at  a  meeting  of  the  Electrical  Section  of  the  Franklin  Institute  held \nThursday,  March  30,  1922.  Reprinted  from  the  Journal  of  the  Franklin  Institute \nfor  June  1922.\n\nfor  engineering  purposes,  in  that  it  measures  directly  the  thing  of \nmost  interest,  namely,  the  degrading  effect  upon  telephone  conversa- \ntion of  introducing  electrical  distortion  into  the  transmission  circuit. \nHowever,  the  application  of  the  results  is  not  limited  to  this  particular \nfield.\n\nBriefly  stated  the  method  consists  in  pronouncing  detached  speech \nsounds  into  the  transmitting  end  of  the  system  and  having  observers \nwrite  the  sounds  which  they  hear  at  the  receiving  end.  The  com- \nparison of  the  called  sounds  with  those  observed  shows  the  number \nand  kinds  of  errors  which  are  made.  The  per  cent  of  the  total  sounds \nspoken  which  are  correctly  received  is  called  the  articulation  of  the \nsystem.\n\nlip  against  lip \ntongue  against  teeth \ntongue  against  hard  palate \ntongue  against  soft  palate\n\nIn  order  to  understand  the  construction  of  the  articulation  lists \nand  also  to  interj^ret  the  results  of  this  investij^ation,  I  desire  to \ngive  here  a  brief  classification  of  the  speech  sounds,  which  is  based \nupon  tiie  position  of  the  various  speech  organs  when  the  sounds  are \nbeing  produced.     It  is  shown  in  the  accompanying  table  (Table  I).\n\nThe  pure  vowels  are  arranged  in  the  vowel  triangle,  which  is  familiar \nto  phoneticians.  Starting  with  the  sound  u  the  lips  are  rounded \nand  there  is  formed  a  single  resonant  cavity  in  the  front  part  of  the \nmouth.  Passing  along  the  left  side  of  the  triangle  from  u  to  a  the \nmouth  is  gradually  opened  with  the  tongue  lowered  to  form  the  suc- \ncessive vowels.  Going  along  the  right  side  of  the  triangle  from  a  to \ne,  the  tongue  is  gradually  raised  to  the  front  part  of  the  mouth  form- \ning two  resonant  chambers  in  the  mouth  cavity.  An  infinite  number \nof  different  shadings  of  these  vowels  may  be  produced  by  placing \nthe  mouth  in  the  various  intermediate  positions,  but  the  ones  which \nare  shown  were  chosen  as  being  the  most  distinct.\n\nThe  sounds  w,  y,  ou,  I  and  h  are  classed  as  combinational  and \ntransitional  vowels.  As  the  mouth  is  placed  in  the  position  to  say  u \nand  then  suddenly  changed  so  as  to  form  any  other  vowel  in  the \ntriangle,  the  result  obtained  is  signified  in  writing  by  placing  the \nktter  w  before  the  vowel.  In  a  similar  way  we  get  the  effect  usually \ndesignated  by  y  if  the  position  of  the  vowel  suddenly  changes  from \ne  to  any  other  vowel.  An  infinite  variety  of  dipthongs  can  be  formed \nby  changing  the  position  of  the  mouth  necessary  to  form  one  vowel \nto  that  to  form  another  without  interrupting  the  voice.  The  most \ndistinct  and  principal  ones  used  in  our  language  are  formed  by  passing \nfrom  the  sound  a  to  either  extreme  corner  of  the  triangle  and  are \nknown  as  ou  and  I.  When  a  vowel  commences  a  syllable  it  is  formed \nby  suddenly  opening  the  glottis,  permitting  the  air,  which  has  been \nheld  in  the  lungs,  to  escape  into  the  mouth,  which  is  formed  for  the \nproper  vowel.  If  the  glottis  remains  open  and  the  vowel  is  started \nby  the  sudden  contraction  of  the  lungs,  we  have  the  effect  which  is \nrepresented  in  writing  by  placing  an  h  before  the  vowel.  The  sounds \n1  and  r  are  called  semi-vowels  because  the  voice  train  is  partially \ninterrupted,  although  the  sound  can  be  continued.  The  stop  and \nfricative  consonants  are  classified  in  a  manner  which  is  familiar  to \nphoneticians.\n\nIt  will  be  noticed  that  the  markings  are  not  those  used  in  the  inter- \nnational phonetic  alphabet  which  were  entirely  too  complicated  for \npractical  use.  Only  the  bar  and  accent  stroke  are  used.  These \ncan  be  written  quickly  and  with  little  chance  of  error.\n\nbe  combined  into  syllables.  For  the  purpose  of  this  investigation \nthey  were  combined  into  mono-syllables  of  the  simple  types  con- \nsonant-vowel,  vowel-consonant,   and   consonant-vowel-consonant.\n\nTo  eliminate  memory  efTects  every  possible  combination  of  the \nsounds  into  these  types  of  syllables  was  used  unless  there  was  a  good \nreason  for  excluding  it.  The  complete  list  contained  8700  syllables. \nFor  convenience  of  testing  these  syllables  were  divided  into  groups \nof  fifty.  Each  group  contained  the  same  kind  and  number  of  syllable \nforms  and  an  equal  number  of  each  of  the  fundamental  vowel  and \nconsonant  sounds.\n\nTo  illustrate  the  technique  of  articulation  testing  a  sample  list \nis  given  in  Table  II.  fn  the  first  column  the  syllable  is  given  in  its \nphonetic  form.  A  key-word  showing  how  each  syllable  is  pro- \nnounced is  given  in  the  second  column.  These  syllables  were  written \non  cards  which  were  shuffled  each  time  before  they  were  used,  so \nthat  the  order  in  which  they  were  pronounced  was  entirely  hap- \nhazard. One  hundred  and  seventy-four  similar  lists  were  used  in \nthis    work.     In    order    to    eliminate    personal    peculiarities,    several\n\ncallers  and  observers  were  used.  In  Table  III  are  shown  the  results \nobtained  by  an  observer  when  this  list  was  transmitted  over  a  system \nwhich  eliminated  all  frequencies  above  1250  cycles  per  second.\n\nThe  correct  word  is  written  opposite  all  of  the  syllables  which \nwere  recorded  incorrectly.  The  errors  for  each  of  the  fundamental \nsounds  were  taken  from  this  original  sheet  and  recorded  on  an  analysis\n\nsheet  as  shown  in  Table  IV.  for  example  it  will  he  noticed  that  p \nwas  recorded  as  k  24.4  per  cent,  as  p  45  per  cent,  and  as  t  22.2  per  cent \nof  the  times  called.  On  the  other  hand  the  sound  w  was  only  recorded \nincorrectly  1  per  cent  of  the  times  called.\n\nFor  this  system  the  consonant  articulation  was  65.8  and  the  vowel \narticulation  83.4.\n\nThe  telephone  system  used  in  this  investigation  is  probably  more \nnearly  perfect  than  any  other  which  has  yet  been  built.  Its  essential \nelements  are  a  condenser  transmitter  to  receive  the  speech  waves \nand  transform  them  into  the  electrical  form,  an  amplifier  for  magni- \nfying the  intensity  of  the  electrical  speech  currents,  an  attenuator \nfor  controlling  the  intensity,  an  equalizing  network,  and  a  receiver \nfor  delivering  the  speech  to  the  ear.  A  schematic  arrangement  of \nthe  circuit  is  shown  in  Fig.  1.\n\nA  detailed  description  of  the  construction  and  operation  of  the \ncondenser  transmitter  has  been  given  by  Crandall  and  Wente  and \npublished  in  the  Physical  Review}  It  is  simply  an  air  condenser, \none  of  its  plates  being  a  flexible  metal  diaphragm.\n\nA  five-stage  vacuum  tube  amplifier  was  used.  Particular  care \nwas  taken  in  coupling  the  stages  together,  so  that  the  amplifier  was \npractically  free  from  frequency  distortion.\n\nThe  attenuator  consisted  of  a  potentiometer  arrangement  which \ncould  reduce  the  amplitude  of  the  speech  waves  to  approximately \none-millionth  of  their  maximum  values.\n\ndensers  and  inductance  coils  having  a  frequency  selectivity  which \nwas  the  complement  of  that  of  the  rest  of  the  system.\n\nThe  telephone  receiver  was  a  bipolar  type  having  a  special  con- \nstruction which  was  designed  to  broaden  the  range  of  frequency \nresponse.\n\nThe  reproducing  efficiency  of  the  system  from  the  mouth  of  the \nspeaker  to  the  ear  of  the  listener  for  each  frequency  is  shown  in  Fig.  2.\n\nThe  pitch  or  frequency  of  the  tone  is  given  on  the  X  axis.  The \nordinates  represent  amplitude  ratios  or  the  number  of  times  the \namplitude  of  the  tone  reaching  the  ear  was  greater  than  that  which \nentered  the  transmitter.  It  will  be  seen  that  this  high  quality  system \nhas  practically  a  uniform  response  for  all  frequencies  throughout  the \nspeech  range.\n\nIn  order  that  its  uniformity  may  be  appreciated,  a  comparison \ncurve  is  given.  This  curve  shows  the  deviation  in  the  sensitivity \nof  a  typical  individual  ear  from  the  average  sensitivity  of  a  large \nnumber  of  ears.  The  ordinates  represent  the  ratio  of  amplitudes \nat  the  various  pitches  which  was  necessary  to  bring  the  tone  to  the \nthreshold  of  audibility.  It  is  evident  that  this  deviation  is  much \nlarger  than  the  departure  of  the  high  quality  circuit  from  uniformity.\n\nTo  show  that  this  particular  individual's  curve  is  typical,  the \ncurves  for  both  ears  of  20  women  are  given  in  Fig  3.  For  convenience \nthese  curves,  are   plotted   on   logarithmic   paper.     If   an   arithmetic\n\nscale  is  used,  all  of  the  eur\\es  below  the  mean  are  crowded  together \nin  the  small  space  between  zero  and  one,  and  all  those  above  the \nmean  are  stretched  out  from  one  to  infinity.  By  using  a  logarithmic \nplot  a  symmetrical  distribution  is  obtained.     The  method  of  obtain-\n\nIt  is  interesting  to  note  that  they  indicate  that  each  individual \nhas  a  hearing  characteristic  which  is  quite  difTerent  from  other  in- \ndividuals. Consequently  speech  sounds  differently  to  difTerent \npersons.  Any  distortions  of  the  speech  sounds  will  necessarily \naffect  some  persons  differently  from  others.  It  is  evident  then  that \nin  discussing  speech  and  hearing  we  must  deal  w^ith  statistical  averages.\n\nExperimental  articulation  tests  showed  that  the  ear  interpreted \nthe  speech  which  was  transmitted  over  this  high  quality  system \npractically  as  well  as  that  transmitted  through  the  air.  Some  may \nwonder  why  such  good  quality  is  not  furnished  telephone  users  in \ncommercial  practice:  Scientifically  speaking,  it  is  possible  to  furnish \nsuch  quality,  but  it  is  evident  that  the  equipment  involved  is  so  com-\n\nplicated  that  such  service  would  be  altogether  too  costly  for  com- \nmercial use;   people  could  not  afford  to  pay  for  it.\n\nThe  Relation  Between  the  Volume  and  Articulation  of \nUndistorted  Speech\n\nin  Fig.  4.  The  abscissas  in  this  curve  represent  loudness  and  are \nexpressed  as  the  natural  logarithm  of  the  number  of  times  the  speech \nwave  amplitude  has  been  decreased  from  the  initial  intensity  at \n]/^  inch  in  front  of  the  mouth  of  the  callers.  This  unit  of  loudness \nhas  never  been  given  a  name,  and  as  a  matter  of  convenience  in \nthis  work  it  is  called  a  napier.  It  will  be  noticed  that  when  the \nvolume  is  reduced  11}/^  napiers  below  the  initial  speech  intensity \nthe  articulation  becomes  zero.  This  point  also  represents  the  value \nat  which  the  speech  becomes  inaudible  and  corresponds  to  approxi- \nmately 1/1000  dynes  per  square  centimetre  pressure  variation  against \nthe  ear  drum.  In  energy  units  it  is  a  reduction  of  ten  billion  times \nbelow  the  initial  speech  intensity.  For  very  loud  initial  speech \nthis  point  is  shifted  about  1  napier.  For  purposes  of  comparison \nthe  intensity  reductions  are  also  indicated  on  the  loudness  axis.\n\nAt  3  napiers  below  or  at  about  1/1000  of  the  initial  speech  in- \ntensity the  articulation  becomes  a  maximum.  Louder  speech  than \nthis  seems  to  deaden  the  nerves  so  that  a  person  makes  a  less  accurate\n\ninterpretation  of  the  received  speech.  These  results  were  obtained \nin  a  room  which  was  especially  constructed  to  exclude  outside  noise. \nWhen  noise  is  present  at  the  receiving  station  the  optimum  loudness \nincreases  as  the   noise  increases.\n\nThe  articulation  data  were  analyzed  so  as  to  show  the  errors  of \neach  of  the  fundamental  sounds.  The  curves  given  in  big.  5  show \nthe  results  of  this  analysis.  It  will  be  noticed  that  the  \\olume  at \nwhich  errors  begin  to  be  appreciable  is  different  for  the  different \nsounds  and  is  usualK'  higher  for  the  consonants  than  for  the  vowels.\n\nWithin  the  precision  of  the  test  the  intersection  point  on  the  X  axis \nwas  the  same  for  all  the  sounds,  namely  at  11.5  napiers.\n\nIt  will  be  noticed  that  the  consonants  are  usually  harder  to  hear \nthan  the  vowels.  However,  the  speech  sounds  e  and  I,  r,  ng  form \nnotable  exceptions  to  this  general  rule,  since  the  former  is  among \nthe  most  difficult,  while  the  latter  are  among  the  very  easiest  speech \nsounds.  The  order  in  which  the  speech  sounds  are  given  here  rep- \nresents their  relative  difficulty  of  interpretation  when  received  at \naverage  intensities.  At  all  intensities,  the  sounds  th,  f  and  v  are \nthe  most  difficult.  Z,  h  and  s  become  very  difficult  at  weak  volumes. \nThe  sounds  i,  ou,  er  and  6  are  missed  less  than  10  per  cent  of  the \ntime,  even  with  \"very  weak\"  intensity.  At  \"average\"  volumes \nthere  are  only  three  sounds  more  difficult  than  e  while  at  \"very \nweak\"  volumes  there  are  23  sounds  more  difficult.  At  very  weak \nvolumes  1,  which  is  the  easiest  sound  at  \"average\"  volumes  is  missed \nthree  times  as  often  as  e.\n\nWe  will   now  pass  to  a  consideration  of  the  effect  of  distortion \nupon  the  articulation  of  the  sounds.\n\nIn  order  to  investigate  distortion  we  would  like  to  be  able  to  take \nthe  train  of  speech  waves  going  from  the  mouth  to  the  ear  and  oper- \nate upon  it  in  various  ways  such  as  eliminating  frequencies  in  certain \nregions  without   marring  or  disturbing  other  frequencies.      For  ex-\n\nample,  if  all  frequencies  above  1000  were  eliminated,  it  would  be \npossible  to  determine  what  intelligibility  is  carried  by  this  range  of \nfrequencies.\n\nFortunately  one  of  the  recent  electrical  inventions  is  admirably \nadapted  for  this  purpose,  namely,  the  electrical  wave  filter  invented \nby  Dr.  G.  A.  Campbell.  This  device  was  used  extensively  in  this \ninvestigation.\n\nThe  schematic  circuit  diagrams  of  the  two  types  of  filters  which \nwere  used  are  given  in  Fig.  6.\n\nThis  arrangement  of  coils  and  condensers  produces  an  electrical \nconductor  with  the  unusual  properties  that  it  transmits  without \nappreciable  diminution  in  amplitude  any  frequency  between  certain \nlimits  and  reduces  the  amplitude  of  all  frequencies  outside  these \nlimits  to  less  than  1/1000  of  their  original  value.  By  varying  the \nnumerical  values  of  the  inductances  and  capacities  this  transmitted \nrange  can  be  placed  at  any  desired  position.  In  the  arrangement \nwhich  was  used  in  the  investigation  these  coils  and  condensers  were\n\nhoused  in  two  boxes.  The  switching  mechanism  was  arranged  so \nthat  by  turning  a  dial  the  condensers  and  coils  were  connected  in \nsuch  a  way  that  the  filter  transmitted  different  frequency  bands.\n\nIn  Fig.  7  are  shown  the  transmission  properties  of  the  low  pass \nfilter  when  the  dial  is  set  to  transmit  frcciuencics  from  0  to  1500. \nIt  is  seen  that  for  frequencies  below  1400  the  amplitudes  of  the  trans- \nmitted tones  are  always  greater  than  .8  of  their  initial  values,  while \nfor  frequencies  above  1500  the  amplitudes  are  decreased  to  less  than \n.001  of  their  initial  values.  These  electrical  filters  were  connected \ninto   the  high   quality  circuit  between   the   third  and   fourth   stages\n\nof  the  amplifier  as  indicated  in  Fig.  1.  This  combination  formed  a \nsystem  which  would  pick  up  a  complex  sound  wave  and  transmit \nfaithfully  to  the  ear  those  component  frequencies  in  any  desired \nregion  and  eliminate  all  other  frequencies.\n\nArticulation  tests  were  made  with  these  filter  systems  and  the \nresults  analyzed  as  described  above.  In  Fig.  8  the  syllable  articu- \nlation results  are  shown  in  graphical  form.  The  ordinates  for  the \nsolid  curves  represent  the  per  cent  of  the  articulation  syllables  called \ninto  the  system  which  were  correctly  recorded  at  the  observing  end. \nThe  abscissas  represent  the  so-called  \"cut  off\"  frequency  of  the \nfilter.  For  example  on  the  curve  labelled  \"Articulation  L\"  the \npoint  (1000,  40)  means  that  a  system  w^hich  transmits  only  frequencies\n\nbelow  1000  cycles  per  second  has  a  syllable  articulation  of  40  per \ncent.  Similarly  on  the  curve  labelled  \"Articulation  H\"  the  point \n(1000,  86)  means  that  a  system  which  transmits  only  frequencies \nabove  1000  cycles  per  second  has  a  syllable  articulation  of  86  per \ncent.  The  dotted  curves  show  the  per  cent  of  the  total  speech  energy \nwhich  is  transmitted  through  the  filter  systems  used  in  the  articula- \ntion tests.  These  curves  are  derived  from  the  results  of  Crandall \nand  MacKenzie  which  were  recently  published.^\n\nIt  will   be  seen   that  although   the   fundamental   cord   tones  with \ntheir  f^.rst  few  harmonies  carry  a  large  portion  of  the  speech  energy,\n\nEffect  upon  the  Articulation   and  the  Energy  of  Speech \nof  Eliminating  Certain  Frequency  Regions\n\nthey  carry  practically  none  of  the  speech  articulation.  A  filter \nsystem  which  eliminates  all  frequencies  below  500  cycles  per  second \neliminates  60  per  cent  of  the  energy  in  speech,  but  only  reduces  the \narticulation  2  per  cent.  A  system  which  eliminates  frequencies \nabove  1500  cycles  per  second  eliminates  only  10  per  cent  of  the  speech \nenergy,  but  reduces  the  articulation  35  per  cent.  A  system  which \neliminates  all  frequencies  above  3000  cycles  per  second  has  as  low  a \nvalue  for  the  articulation  as  one  which  eliminates  all  frequencies \nbelow  1000  cycles  per  second.  This  last  statement  may  appear \nrather  astonishing  since  it  is  contrary  to  the  popular  notion  of  the \nrelative  importance  of  various  voice  frequencies  from  an  interpre- \ntation standpoint.\n\nThe  two  solid  curves  intersect  on  the  1550  cycle  abscissa  and  at \n65   per  cent  articulation,  which  shows  that  using  only  frequencies\n\nabove  or  frequencies  below  1550  cycles  an  articulation  of  65  per \ncent  will  be  obtained.  The  two  dotted  curves  necessarily  intersect \nat  50  per  cent.\n\nThe  cur\\es  in  Fig.  9  show  how  the  articulation  of  some  of  the \nfundamental  speech  sounds  was  affected  by  eliminating  certain \nfrequency  regions.  The  ordinate  gives  the  number  of  times  the \nsound  was  written  correctly  per  100  times  called.     As  in  Fig.  8  the\n\nleft  hand  curve  shows  the  effect  of  eliminating  all  frequencies  below \nand  the  right  hand  curve  the  effect  of  eliminating  all  frequencies \nabove  the  frequency  specified  by  the  abscissa.\n\nThese  nine  speech  sounds  were  chosen  as  representing  three  im- \nportant classes.  It  is  seen  that  the  long  vowels  e,  1  and  i  can  be  trans- \nmitted with  an  error  of  less  than  3  per  cent  when  using  either  half \nof  the  range  of  frequencies.  When  using  either  frequencies  from \n0  to  1700  or  from  1700  to  infinity  e  was  interpreted  correctly  98  per \ncent  of  the  time.  Similarly  1  was  interpreted  correctly  97  per  cent \nof  the  time  when  using  either  the  range  from  0  to  1000  or  1000  to \ninfinity,  and  i  96  per  cent  of  the  time  when  using  either  the  range \nfrom  0  to  1350  or  from  1350  to  infinity.  The  short  vowels,  u,  o  and \ne  are  seen  to  have  important  characteristics  carried  by  frequencies \nbelow  1000.     More  than  a  20  per  cent  error  is  made  on  any  of  these\n\nthree  sounds  when  frequencies  below  1000  are  eliminated.  The \nelimination  of  frequencies  above  2000  produces  almost  no  effect.\n\nThe  fricative  consonants  s,  z  and  th  are  seen  to  be  affected  very \ndifferently  from  those  in  the  other  two  classes.  These  sounds  are \nvery  definitely  affected  when  frequencies  above  5000  are  eliminated. \nThe  sounds  s  and  z  are  not  affected  by  the  elimination  frequencies \nbelow  1500.  It  is  principally  due  to  these  three  sounds  that  the \nsyllable  articulation  is  reduced  from  98  per  cent  to  82  per  cent  when \nfrequencies  above  2500  cycles  are  eliminated.\n\nA  more  detailed  analysis  of  the  articulation  results  on  all  the  speech \nsounds  showing  the  kind  as  well  as  the  number  of  errors  will  be  given \nin   a   future   paper.\n\nIn  conclusion  then  we  see  that  the  intensity  of  undistorted  speech \nwhich  is  received  by  the  ear  can  be  varied  from  100  times  greater \nto  one-millionth  less  than  the  initial  speech  intensity  without  notice- \nably affecting  its  interpretation.  The  intensity  must  be  reduced \nto  one-ten-billionth  of  that  initial  speech  intensity  to  reach  the  thres- \nhold of  audibility  for  the  average  ear.  Also  it  is  seen  that  any  ap- \nparatus designed  to  reproduce  speech  and  preserve  all  of  its  char- \nacteristic qualities  must  transmit  frequencies  from  100  to  above \n5000  cycles  with  approximately  the  same  efficiency.  Although  most \nof  the  energy  in  speech  is  carried  by  frequencies  below  1000,  the \nessential  characteristics  which  determine  its  interpretation  are  carried \nmostly  by  frequencies  above  1000  cycles.  In  ordinary  conversation \nthe  sounds  th,  f  and  v  are  the  most  difficult  to  hear  and  are  responsi- \nble for  50  per  cent  of  the  mistakes  of  interpretation.  The  character- \nistics of  these  sounds  are  carried  principally  by  the  very  high \nfrequencies.\n\nIt  is  evident  that  progress  in  the  knowledge  of  speech  and  hearing \nhas  a  great  human  interest.  It  will  greatly  aid  the  linguists,  the \nactors,  and  the  medical  specialists.  It  may  lead  to  improved  devices \nwhich  will  alleviate  the  handicaps  of  deaf  and  dumb  persons.  Fur- \nthermore this  knowledge  will  be  of  great  importance  to  the  telephone \nengineer,  and  since  the  telephone  is  so  universally  used,  any  improve- \nment in  its  quality  will  be  for  the  public  good.\n\nThese  humanitarian  and  utilitarian  motives  as  well  as  the  pure \nscientific  interest  have  already  attracted  a  number  of  scientists  to \nthis  field.  Now  that  new  and  powerful  tools  are  available,  it  is \nexpected  that  in  the  near  future  more  will  be  led  to  pursue  research \nalong  those  lines.\n\nWilliam  Wilson,  Victoria  University  of  Manchester,  1904-10; \nM.Sc,  1908;  Cavendish  Laboratory,  Cambridge  University,  1910-12, \nB.A.,1912;  Lecturer  in  Physics,  Toronto  University,  1912-14;  D.Sc \nManchester,  1913.  Engineering  Department  W'estern  Electric  Com- \npany, 1914-  .  Dr.  Wilson  has  pubHshed  numerous  papers  on  radio \nactivity  and  thermionics  and  since  1917  has  been  in  direct  charge  of \nvacuum  tube  design.\n\nGeorge  A.  Campbell,  B.S.,  Massachusetts  Institute  of  Technology, \n1891;  A.B.,  Harvard,  1892;  Ph.D.,  1901;  Gottingen,  Vienna  and \nParis,  1893-96.  Mechanical  Department,  American  Bell  Telephone \nCompany,  1897;  Engineering  Department,  American  Telephone  and \nTelegraph  Company,  1903-1919;  Department  of  Development  and \nResearch,  1919 \u2014 ;  Research  Engineer,  1908 \u2014 .  Dr.  Campbell  has \npublished  papers  on  loading  and  the  theory  of  electric  circuits  and  is \nalso  well-known  to  telephone  engineers  for  his  contributions  to  re- \npeater and  substation  circuits.  The  electric  filter  which  is  one  of  his \ninventions  plays  a  fundamental  role  in  telephone  repeater,  carrier \ncurrent  and  radio  systems.\n\nH.  M.  Trueblood,  B.S.,  Earlham,  1902;  Haverford,  1903;  Massa- \nchusetts Institute  of  Technology,  1908-09;  Ph.D.,  Harvard,  1913; \naid  and  assistant  United  States  Coast  and  Geodetic  Survey,  1903-08; \nassistant  in  physics,  Harvard^  1912-14;  Joule-Thomson  effect  in \nsuper-heated  steam;  instructor  and  assistant  professor  electrical \nengineering,  University  of  Pennsylvania,  1914-17;  Department  of \nDevelopment  and  Research,  American  Telephone  and  Telegraph  Corn- \ncompany,  1917 \u2014 ;  work  on  inductive  interference.\n\nJ.  J.  PiLLiOD,  E.E.,  Ohio  Northern  University,  1908;  American \nTelephone  and  Telegraph  Company,  Toledo  Home  Telephone  Com- \npany, and  Union  Switch  and  Signal  Company,  short  periods,  1904-08; \nAmerican  Telephone  and  Telegraph  Company,  Long  Lines  Depart- \nment, 1908-11;  Engineering  Department,  1912-13;  Division  Plant \nEngineer,  Long  Lines  Department,  1914-17 ;  Engineer  of  Transmission, \n1918-19;  Engineer,  1920-.  As  Engineer  of  the  Long  Lines  Depart- \nment, Mr.  Pilliod  has  been  in  general  charge  of  engineering  work  in- \nvolved in  the  planning  and  installation  of  the  newer  sections  of  the \ncable  project  described.\n\nJoHX  R.  Carson,  B.S.,  Princeton,  1907;  E.E.,  1909;  M.S.,  1912; \nResearch    Department,  Westinghouse    Electric    and    Manufacturing\n\nCompany,  1910-12;  instructor  of  physics  and  electrical  engineering, \nPrinceton,  1912-14;  American  Telephone  and  Telegraph  Company, \nEngeneering  Department,  1914-15;  Patent  Department,  1916-17; \nEngineering  Department,  1918;  Department  of  Development  and \nResearch,  1919-.  Mr.  Carson's  work  has  been  along  theoretical  lines \nand  he  has  published  several  papers  on  theory  of  electric  circuits  and \nelectric  wave  propagation.\n\nJ.  J.  Gilbert,  A.B.,  University  of  Pennsylvania,  1909;  Harvard, \n1910-11;  Chicago,  1911-12;  E.E.,  Armour  Institute,  1915;  in- \nstructor of  electrical  engineering,  Armour,  1912-17;  Captain  Signal \nCorps,  1917-19;  Engineering  Department,  Western  Electric  Com- \npany, 1919,  since  when  he  has  worked  on  submarine  cable  problems.\n\nI.  B.  Crandall,  A.B.,  Wisconsin,  1909;  A.M.,  Princeton,  1910; \nPh.D.,  1916;  Professor  of  Physics  and  Chemistry,  Chekiang  Pro- \nvincial College,  1911-12;  Engineering  Department,  Western  Electric \nCompany,  1913-.  Dr.  Crandall  has  published  papers  on  infra-red \noptical  properties,  condenser  transmitter,  thermophone,  etc.  More \nrecently  he  has  been  associated  with  studies  on  the  nature  and  analysis \nof  speech  which  have  been  in  progress  in  the  Laboratory.\n\nDonald  MacKenzie,  A.B.,  Johns  Hopkins,  1908,  A.M.,  1911; \nPh.D.,  1914;  assistant  astronomy,  1914-17;  associate  physicist. \nBureau  of  Standards,  1918-20;  Engineering  Department,  Western \nElectric  Company,  1920-.\n\nHarvey  Fletcher,  B.S.,  Brigham  Young,  1907;  Ph.D.,  Chicago, \n1911;  instructor  of  physics,  Brigham  Young,  1907-08;  Chicago, \n1909-10;  Professor,  Brigham  Young,  1911-16;  Engineering  Depart- \nment, Western  Electric  Company,  1916-.  The  present  paper  by  Dr. \nFletcher  gives  some  of  the  results  of  an  investigation  which  is  being \nmade  of  the  relation  between  the  frequency  characteristics  of  tele- \nphone circuits  and  the  intelligibility  of  transmitted  speech.  Dr. \nFletcher  has  also  published  on  Brownian  movements,  ionization  and \nelectronics.\n\nNote  :  The  electric  wave-filter,  an  invention  of  Dr.  Campbell,  is  one  of  the \nmost  important  of  present  day  circuit  developments,  being  indispensable \nin  many  branches  of  electrical  communication.  It  makes  possible  the \nseparation  of  a  broad  band  of  frequencies  into  narrow  bands  in  any  desired \nmanner,  and  as  will  be  gathered  from  the  present  article,  it  effects  the \nseparation  much  more  sharply  than  do  tuned  circuits.  As  the  communica- \ntion art  develops,  the  need  will  arise  to  transmit  a  growing  number  of  tele- \nphone and  telegraph  messages  on  a  given  pair  of  line  wires  and  a  grow- \ning number  of  radio  messages  through  the  ether,  and  the  filter  will  prove \nincreasingly  useful  in  coping  with  this  situation.  The  filter  stands  beside \nthe  vacuum  tube  as  one  of  the  two  devices  making  carrier  telegraphy  and \ntelephony  practicable,  being  used  in  standard  carrier  equipment  to  separate \nthe  various  carrier  frequencies.  It  is  a  part  of  every  telephone  repeater \nset,  cutting  out  and  preventing  the  amplification  of  extreme  line  frequencies \nfor  which  the  line  is  not  accurately  balanced  by  its  balancing  network. \nIt  is  being  applied  to  certain  types  of  composited  lines  for  the  separation \nof  the  d.c.  Morse  channels  from  the  telephone  channel.  It  is  finding  many \napplications  to  radio  of  which  multiplex  radio  is  an  illustration.  The  filter \nis  also  being  put  to  numerous  uses  in  the  research  laboratory.\n\nThe  present  paper  is  the  first  of  a  series  on  the  electric  wave-filter  to \nbe  contributed  to  the  Technical  Journal  by  various  authors.  Being  an \nintroductory  paper  the  author  has  chosen  to  discuss  his  subject  from  a \nphysical  rather  than  mathematical  point  of  view,  the  fundamental  char- \nacteristics of  filters  being  deduced  by  purely  physical  reasoning  and  the \nderivation  of  formulas  being  left  to  a  mathematical  appendix. \u2014 Editor.\n\nTHE  purpose  of  this  paper  is  to  present  an  elementary,  physical \nexplanation   of  the  wave-filter  as  a  device   for  separating  sin- \nusoidal  electrical   currents  of  different   frequencies.     The  discussion\n\nwill  be  general,  and  will  not  Involve  assumptions  as  to  the  detailed \nconstruction  of  the  wave-filter;  but  in  order  to  secure  a  certain  nu- \nmerical concreteness,  curves  for  some  simple  wave-filters  will  be  in- \ncluded. The  formulas  employed  in  calculating  these  curves  are \nspecial  cases  of  the  general  formulas  for  the  wave-filters  which  are, \nin  conclusion,  deduced  by  the  method  employed  in  the  physical \ntheory.\n\nAll  the  physical  facts  which  are  to  be  presented  in  this  paper, \ntogether  with  many  others,  are  implicitly  contained  in  the  compact \nformulas  of  the  appendix.  Although  only  comparatively  few  words \nof  explanation  are  required  to  derive  these  formulas,  they  will  not  be \npresented  at  the  start,  since  the  path  of  least  resistance  is  to  rely \nimplicitly  upon  formulas  for  results,  and  ignore  the  troublesome  ques- \ntion as  to  the  physical  explanation  of  the  wave-filter.  In  order  to \nexamine  directly  the  nature  of  the  wave-filter  in  itself,  as  a  physical \nstructure,  we  proceed  as  though  these  formulas  did  not  exist.\n\nIt  is  intended  that  the  present  paper  shall  serve  as  an  introduction \nto  important  papers  by  others  in  which  such  subjects  as  transients  on \nwave-filters,  specialized  types  of  wave-filters,  and  the  practical  design \nof  the  most  efificient  types  of  wave-filters  will  be  discussed.^\n\nA  wave-filter  is  a  device  for  separating  waves  characterized  by  a  dif- \nference in  frequency.  Thus,  the  wave-filter  differentiates  between \ncertain  states  of  motion  and  not  between  certain  kinds  of  matter, \nas  does  the  ordinary  filter.  One  form  of  wave-filter  which  is  well \nknown  is  the  color  screen  which  passes  only  certain  bands  of  light \nfrequencies;  diffraction  gratings  and  Lippmann  color  photographs \nalso  filter  light.  Wave-filters  might  be  constructed  and  employed \nfor  separating  air  waves,  water  waves,  or  waves  in  solids.  This \npaper  will  consider  only  the  filtering  of  electric  waves;  the  same \nprinciples  apply  in  every  case,  however.\n\n^  I  take  pleasure  in  acknowledging  my  indebtedness  to  Mr.  O.  J.  Zobel  for  specific \nsuggestions,  and  for  the  light  thrown  on  the  whole  subject  of  wave-filters  by  his \nintroduction  of  substitutions  which  change  the  propagation  constant  without  chang- \ning the  iterative  impedance.\n\nplied  with  its  assigned  narrower  range  of  frequencies.  Thus,  for \ninstance,  with  three  wa\\e-tilters  the  band  of  frequencies  necessary \nfor  ordinary  telephony  might  ^be  transmitted  to  one  receiving  device, \nall  lower  frequencies  transmit^jted  to  a  second  device,  and  all  higher \nfrequencies  transmitted  to  a\"  third  de\\'ice \u2014 separation  being  made \nwithout  serious  loss  of  energy  in  any  one  of  the  three  bands.\n\nBy  means  of  wave-filters  ii^terference  between  difYerent  circuits  or \nchannels  of  communication  irv  telephony  and  telegraphy,  both  wire  and \nradio,  can  be  reduced  provided  they  operate  at  difTerent  frequencies. \nThe  method  is  furthermore  applicable,  at  least  theoretically,  to  the \nreduction  of  interference  between  powder  and  communication  circuits. \nThe  same  is  true  of  the  simultaneous  use  of  the  ether,  the  earth  return, \nand  of  expensive  pieces  of  apparatus  employed  for  several  power  or \ncommunication  purposes.  In  all  cases  the  principle  involved  is  the \nsame  as  that  of  confining  the  transmission  in  each  circuit  or  channel \nto  those  frequencies  which  serve  a  useful  purpose  therein  and  exclud- \ning or  suppressing  the  transmission  of  all  other  frequencies.  In  the \nfuture,  as  the  utility  of  electrical  applications  becomes  more  widely \nand  completely  appreciated,  there  will  be  an  imperative  necessity \nfor  more  and  more  completely  superposing  the  varied  applications  of \nelectricity;  it  will  then  be  necessary,  to  avoid  interference,  to  make \nthe  utmost  use  of  every  method  of  separating  frequencies  including \nbalancing,  tuning,  and  the  use  of  wave-filters.\n\nThe  w^ave-filter  problem  in  this  paper  is  discussed  as  a  phase  of \nthe  artificial  line  problem,  and  it  is  desirable  to  start  with  a  some- \nwhat generalized  definition  of  the  artificial  line.  The  definition \nwill,  however,  not  include  all  wave-filters  or  all  artificial  lines,  since \na  perfectly  general  definition  is  not  called  for  here.  Even  if  an  ar- \ntificial line  is  to  be,  under  certain  wave  conditions,  an  imitation  of, \nor  a  substitute  for,  an  actual  line  connecting  distant  points,  hardly \nany  limitation  is  thereby  imposed  upon  the  structure  of  the  device; \nan  actual  line  need  not  be  uniform  but  may  vary  abruptly  or  gradually \nalong  its  length  and  may  include  two,  three,  four  or  more  transmis- \nsion conductors  of  which  one  may  be  the  earth.  Having  indicated \nthat  w^ave-filters  partake  of  somewhat  this  same  generality  of  struc- \nture, the  present  paper  is  restricted  to  wave-filters  coming  under \nthe  somewhat  generalized  artificial  line  specified  by  the  following \ndefinition  :\n\nAn  artificial  line  is  a  chain  of  networks  connected  together  in  sequence \nthrough  two  pairs  of  terminals,  the  networks  being  identical  hut  other-\n\nwise  unrestricted.  This  generalized  artificial  line  possesses  the  well- \nknown  sectional  artificial  line  structure  but  it  need  not  be  an  imita- \ntion of,  or  a  substitute  for,  any  known,  real,  transmission  line  con- \nnecting together  distant  points.  The  general  artificial  line  is  shown \nby  Fig.  1  where  N,  N,.  .  .  . are  the  identical  unrestricted  networks \nwhich  may  contain  resistance,  self-inductance,  mutual  inductance, \nand  capacity.\n\nIn  discussing  this  type  of  structure  as  a  wave-filter,  the  point  of \nview  of  an  artificial  line  is  adopted  for  the  reason  that  it  is  advan- \ntageous to  regard  the  distribution  of  alternating  currents  as  being \ndependent  upon  both  propagation  and  terminal  conditions,  which \nare  to  be  separately  considered.     In   this  way   the  attenuation,   or\n\nFig.    1 \u2014 Generalized  Artificial   Line    as    Considered   in   the   Present   Paper,   where \nN,  N,  .  .  .  are  Identical  Arbitrary  Electrical  Networks\n\nfalling  off,  of  the  current  from  section  to  section  may  be  most  directly \nstudied.  Terminal  effects  are  not  to  be  ignored,  but  are  allowed \nfor,  after  the  desired  attenuation  effects  have  been  secured,  possibly \nby  an  increase  in  the  number  of  sections  to  be  employed.\n\nThe  fundamental  property  of  this  generalized  artificial  line,  which \nincludes  uniform  lines  as  a  special  case,  is  the  mode  in  which  the \nwave  motion  changes  from  one  section  to  the  next,  and  may  be  stated \nas  follows:\n\nUpon  an  infinite  artificial  line  a  steady  forced  sinusoidal  disturbance \nfalls  off  exponentially  j ram  one  section  to  the  next,  while  the  phase  changes \nby  a  constant  amount.  Reversing  the  direction  of  propagation  does \nnot  alter  either  the  attenuation  or  phase  change.  When  complex  quan- \ntities are  employed  the  exponential  includes  the  phase  change.\"^  This \ntheorem  is  proved,  without  mathematical  equations,   by  observing\n\n^  This  theorem  is  not  new,  but  it  is  ordinarily  derived  by  means  of  differential  or \ndifference  equations  whereas  it  may  be  derived  from  the  most  elementary  general \nconsiderations,  thus  avoiding  all  necessity  of  using  differential  or  difference  equa- \ntions, as  illustrated  in  my  paper  \"On  Loaded  Lines  in  Telephonic  Transmission\" \n{Phil.  Mag.,  vol.  5,  pp.  313-331,  1903).  In  that  discission,  as  well  as  in  this  present \none,  it  is  tacitly  assumed  that  the  line  is  either  an  actual  line  with  resistance,  or  the \nlimit  of  such  a  line  as  the  resistance  vanishes,  so  that  the  amplitude  of  the  wave \nnever  increases  towards  the  far  end  of  an  infinite  line.\n\nthat  the  percentage  reduction  in  amplitude  and  the  change  in  phase, \nin  passing  from  the  end  of  one  section  to  the  corresponding  point  of \nthe  next  section,  do  not  depend  upon  either  the  absolute  amplitude \nor  phase;  they  depend,  instead,  only  upon  the  magnitudes,  angles \nand  interconnections  of  the  impedances  between  the  two  points  and \nof  the  impedances  beyond  the  second  point.  These  impedances \nare,  since  the  line  is  assumed  to  be  periodic  and  infinite,  identically \nthe  same  for  corresponding  points  between  all  sections  of  the  line, \nand,  therefore,  the  relative  changes  in  the  wave  will  be  identical  at \ncorresponding  points  in  all  sections.  This  proves  the  exponential \nfalling  off  of  the  disturbance  and  the  constancy  of  phase  change;  the \nordinary  reciprocal  property  shows  that  the  wave  will  fall  off  identic- \nally whichever  be  the  direction  of  propagation.  By  the  superposition \nproperty  it  follows  that  the  steady  state  on  any  finite  portion  of  a \nperiodic  recurrent  structure  must  be  the  sum  of  two  equally  attenuated \ndisturbances,  one  propagated  in  each  direction.\n\nThe  fundamental  w^ave  propagation  theorem  may  be  generalized \nfor  any  periodic  recurrent  structure  irrespective  of  the  number  and \nkind  of  connections  between  periodic  sections,  provided  the  dis- \nturbance is  such  as  to  remain  similar  to  itself  at  corresponding  points \nof  each  of  these  connections.\n\nSince,  at  a  given  frequency,  any  network  employed  solely  to  con- \nnect a  pair  of  input  terminals  with  a  pair  of  output  terminals  may  be \nreplaced  by  either  three  star-connected  impedances  or  three  delta- \nconnected  impedances,  the  general  artificial  line  of  Fig.    1   may  be\n\nFig.   2 \u2014 Equivalent   Artificial   Line  Obtained   by  Substituting  Star   Impedances\n\nFig.    3 \u2014 Equivalent   Artificial    Line   Obtained    by   Substituting   Delta    Impedances\n\nreplaced  by  the  equivalent  artificial  line  of  either  Fig.  2  or  Fig.  3. \nBy  combining  the  series  impedances  in  Fig.  2  and  the  parallel  im- \npedances in  Fig.  3,  the  equivalent  line  in  Fig.  4  is  obtainable.  The \ntwo  ways  of  arriving  at  Fig.  4  give  different  values  for  the  series  and \nshunt  impedances  Zi,  Z2,  and  different  terminations  for  the  line,  but \nthe  propagation  of  the  wave  is  the  same  in  both  cases,  since  the  assumed \nsubstitutions  are  rigorously  exact.  While  Fig.  4  may  be  considered \nas  the  generalized  artificial  line  equivalent  to  Fig.  1,  this  requires \nincluding  in  Zi  and  Z2  impedances  which  cannot  always  be  physically \nrealized  by  means  of  two  entirely  independent  networks,  one  of \nwhich  gives  Zi  and  the  other  Z2.  This  restriction  is  of  no  importance \nwhen  we  are  discussing  the  behavior  of  the  generalized  artificial  line \nat  a  single  frequency;  accordingly,  the  ladder  artificial  line  is  suitable \nfor  this  part  of  the  discussion.  When  we  come  to  the  more  specific \ncorrelation  of  the  behavior  of  the  generalized  artificial  line  at  different \nfrequencies,  it  will  be  found  more  convenient  to  replace  the  ladder \nartificial  line  by  the  lattice  artificial  line,  which  avoids  the  necessity \nof  considering  any  impedances  which  are  not  individually  physically \nrealizable.\n\nThe  equivalence  between  Figs.  1  and  4  is  implicitly  based  upon \nthe  assumption  that  it  is  immaterial,  for  artificial  line  iises,  what \nabsolute  potentials  the  terminals  1,  2;  3,  4;  5,  6;  etc.  have \u2014 this  leaves \nus  at  liberty  to  connect  2,  4,  6,  etc.,  together,  so  long  as  we  main- \ntain unchanged  the  differences  in  potential  between  1  and  2,  3  and  4, \netc.  Instead  of  connecting  2  and  4  we  might  equally  well  connect \n2  and  3,  and  then  Zi  would  connect  1  and  4    as  in  Fig.  5;  with  these\n\ncross-connections  the  propagation  still  remains  unchanged.  We  have \nagain  obtained  Fig.  4  with  no  circuit  difference  except  the  inter- \nchange of  terminals  8  and  7  with  terminals  4  and  8;  or,  if  this  is  ignored, \na  reversal  in  the  sign  of  the  current  at  alternate  pairs  of  terminals. \nThis  shows  that  the  reveisal  of  the  current  in  alternate  sections  of \nFig.  4  may  not  be  of  primary  significance,  since  networks  which  are \nessentially  equivalent  have  reversed  currents.\n\nIn  order  to  deal,  at  the  start,  with  only  the  simpler  terminal  con- \nditions, we  may  consider  the  line  to  begin  with  only  one-half  of  the \nseries  impedance  Zi,  or  only  one-half  of  the  bridged  admittance \n1  'Zo.  These  mid-points  are  called  the  mid-series  and  mid-shunt \npoints;  knowing  the  results  of  termination  at  either  of  these  points, \nthe  effect  of  termination  at  any  other  point  may  be  readily  deter- \nmined. For  Fig.  4  termination  at  mid-shunt  has  been  chosen  so \nthat  each  section  of  the  line  adds  a  complete  symmetrical  mesh  to \nthe  network.\n\nAn  alternator,  introducing  an  impedance  Z^,  is  shown  as  the  source \nof  the  steady-state  sinusoidal  current  in  Fig.  4.  Assume  that  the \nimpedance  Z^  is  variable  at  pleasure,  and  that  it  is  gradually  adjusted \nto  make  the  total  impedance  in  the  generator  circuit  vanish, \u2014 in  this \ncase  no  e.m.f.  will  be  required  to  maintain  the  forced  steady-state \nwhich  becomes  a  free  oscillation.  If,  in  addition,  it  is  assumed  that \nthe  line  has  an  infinite  number  of  sections,  this  required  value  of  Z^ \nwill  be  the  negative  of  the  mid-shunt  iterative  impedance^  of  the  ar- \ntificial line,  which  will  be  designated  as  K^.  The  first  shunt  on  the \nline  now  includes  \u2014 X2  in  parallel  with  2Z2  so  that  its  total  impedance \nis,  say,  Z'=  \u20142ZiK-i.l(^Zi  \u2014  K'^.  The  infinite  line  with  its  first \nshunt  given  the  special  value  Z'  is  thus  capable  of  free  oscillation.\n\nIt  is  possible  to  simplify  this  infinite  oscillating  circuit  by  cutting\n\n^  The  \"iterative  impedance\"  of  an  artificial  line  is  the  impedance  which  repeats \nitself  when  one  or  more  sections  of  the  artificial  line  are  inserted  between  this  im- \npedance and  the  point  ot  measurement.  It  is  thus  the  impedance  of  an  infinite \nlength  of  any  actual  artificial\" line,  regardless  of  the  termination  of  the  remote  end \nof  the  line.  In  general,  its  value  is  different  for  the  two  directions  of  propagation, \nbut  not  when  the  line  is  symmetrical,  as  at  mid-series  and  mid-shunt.  The  values \nat  these  points  are  denoted  by  Kx  and  Ki.  \"Iterative  impedance\"  is  employed \nbecause  it  is  a  convenient  term  which  is  distinctive  and  describes  the  most  essential \nproperty  of  this  impedance;  it  seems  to  be  more  appropriate  than  \"characteristic \nimpedance,\"  \"surge  impedance\"  and  the  other  synonyms  in  use.\n\n-\\-Ki  in  parallel,  which  have  the  impedance  Z\"  =  +2Z2i^2/(2Z2+A'2)- \nRemoving  Z'  together  with  the  infinite  line  on  the  right  there  remains \non  the  left  a  closed  circuit  made  up  of  the  three  impedances  Zi,  Z' \nand  Z\"  in  series.\n\nAfter  the  division,  the  infinite  line  on  the  right  will  continue,  with- \nout modification,  to  oscillate  freely,  since  it  is  an  exact  duplicate  of \nthe  original  oscillating  line,  and  so  must  maintain  the  free  oscillation \nalready  started.  Since  it  oscillates  freely  by  itself,  it  had  originally \nno  reaction  upon  the  simple  circuit  from  which  it  was  separated; \nthis  simple  circuit  on  the  left  must  thus  also  continue  its  own  free \noscillations  without  change  in  period  or  phase.\n\nWe  might  continue  and  subdivide  the  entire  infinite  line  into \nidentical  simple  circuits  but  it  is  sufficient  to  consider  this  one  detached \ncircuit,  which  is  shown  separately  in  two  ways  by  Fig.  6,  since  from\n\nits  free  oscillations  the  mathematical  formulas  for  the  steady-state \npropagation  in  the  artificial  line  may  be  derived.  This  is  deferred, \nhowever,  until  after  the  physical  discussion  is  completed,  so  as  to \nleave  no  room  for  doubt  that  the  essentials  of  the  physical  theory \nare  really  deduced  without  the  aid  of  mathematical  formulas.\n\nThe  generalized  artificial  line,  if  made  up  entirely  of  pure  resist- \nances, will  attenuate  all  frequencies  alike,  and  the  entire  wave  will \nbe  in  the  same  phase;  this  remains  true,  whatever  be  the  impedance \nof  the  individual  branch  of  the  network,  provided  the  ratio  of  the \nimpedances  of  all  branches  is  a  constant  independent  of  the  frequency. \nThis  is  precisely  the  condition  to  be  avoided  in  a  wave-filter;  branches \nmust  not  be  similar  but  dissimilar  as  regards  the  variation  of  impedance \nwith  frequency.  This  calls  for  inductance  and  capacity  with  neg- \nligible resistance,  so  that  there  is  an  opportunity  for  the  positive \nreactance  of  one  branch  to  react  upon  the  negative  reactance  of \nanother  branch,  in  different  proportions  at  diff^erent  frequencies. \nAssuming  the  unit  network  A^^  of  Fig.  1  to  be  made  up  of  a  finite \nnumber  of  pure  reactances,  the  equivalent  impedances  Zi  and  Z2  of \nFigs.  4  and  5  must  also  be  pure  reactances.     Under  this  assumption\n\nlet  us  consiticr  tlic  free  oscillations  of  I^'ig.  0;  first,  with  A'o  assumed \nto  be  a  pure  reactance;  second,  with  A'2  assumed  to  be  a  pure  resist- \nance; and  third,  in  order  to  show  that  this  third  assumption  is  con- \ntrary to  fact,  with  A'o  assumed  to  be  an  impedance  with  both  resist- \nance and  reactance.\n\nWith  Ao  a  reactance,  the  circuit  contains  nothing  but  reactances, \nand  free  oscillations  are  possible  if,  and  only  if,  the  total  impedance \nof  the  circuit  is  zero.  The  end  impedances  Z'  and  Z\"  being  different, \nthe  potentials  at  the  ends  of  the  mesh  will  be  dilTerent,  and  this  means \nthat  the  corresponding  wave  on  the  infinite  line  will  be  attenuated, \nsince  the  ratio  between  these  potentials  is  the  rate  at  which  the  am- \nplitudes fall  ofY  per  section.\n\nWith  Ao  a  pure  resistance,  a  free  oscillation  is  possible  only  if  the \ndissipation  in  the  positive  resistance  at  the  right  end  of  the  circuit \nis  exactly  made  up  by  the  hypothetical  source  of  energ\\'  existing  in \nthe  negative  resistance  \u2014  A2  at  the  left  end  of  the  circuit.  An  exact \nbalance  between  the  energy  supplied  at  one  end  and  that  lost  at  the \nother  end  is  possible,  since  the  equal  positive  and  negative  resistances \nA2,  \u2014  Ao  carry  equal  currents.  This  continuous  transfer  of  energy \nfrom  the  left  of  the  oscillating  circuit  of  Fig.  6  to  the  right  end  is  the \naction  which  goes  on  in  every  section  of  the  infinite  artificial  line,  and \nserves  to  pass  forward  the  energy  along  the  infinite  line.\n\nIf  Ao  were  complex,  \u2014 Ao  on  the  left  of  Fig.  6  and  +A2  on  the \nright  would  not  carry  the  same  fraction  of  the  circulating  current  /, \nsince  they  are  each  shunted  by  a  reactance  2Z2  which  would  allow \nless  of  the  current  to  flow  through  +A2  than  through  \u2014  Ao,  if  2Z2 \nmakes  the  smaller  angle  with  +A2,  and  vice  versa.  No  balance \nbetween  absorbed  and  dissipated  energy  is  possible  under  these  con- \nditions when  the  equal  and  opposite  resistance  components  carry \nunequal  currents.  A  complex  A2,  therefore,  gives  no  free  oscilla- \ntion, and  cannot  occur  with  a  resistanceless  artificial  line.\n\nIt  is  perhaps  more  instructive  to  consider  the  transmission  on  the \nline  as  a  whole,  rather  than  to  confine  attention  exclusively  to  the \noscillations  of  the  simple  circuit  of  Fig.  6  and  so,  at  this  point,  w^ith- \nout  following  further  the  conclusions  to  be  drawn  directly  from  this \noscillating  circuit,  the  fundamental  energy  theorem  of  resistanceless \nartificial  lines  will  be  stated,  and  then  proved  as  a  property  of  an \ninfinite  artificial  line.\n\nUpon  an  infinite  line  of  periodic  recurrent  structure,  and  devoid  of \nresistance,  a  sinusoidal  e.m.f.  produces  one  of  two  steady  states,  viz.:\n\n1.  A  to-and-jro  surging  of  energy  without  any  resultant  transfer \nof  energy;  currents  and  potential  differences  each  attenuated  from \nsection  to  section,  but  everywhere  in  the  same  or  opposite  phase  and \nmutually  in  quadrature,  or,\n\n2.  A  continuous,  non- attenuated  flow  of  energy  along  the  line \nto  infinity  with  no  energy  surging  between  symmetrical  sections; \ncurrent  and  potential  non-attenuated,  but  retarded  or  advanced  in \nphase  from  section  to  section,  and  mutually  in  phase  at  mid-shunt \nand  mid-series  points.\n\nThe  critical  frequencies  separating  the  two  states  of  motion  are  the \ntotality  of  the  resonant  frequencies  of  the  series  impedance,  the  anti- \nresonant  frequencies  of  the  shunt  impedance,  and  the  resonant  frequencies \nof  a  single  mid-shunt  section  of  the  line.\n\nTo  prove  the  several  statements  of  this  theorem  let  us  consider \nfirst  the  consequences  of  assuming  that  the  wave  motion,  in  progress- \ning along  the  line,  is  attenuated,  and  next  the  consequences  of  assum- \ning that  the  wave  motion  changes  its  phase.  If  the  wave  is  atten- \nuated, however  little,  at  a  sufficient  distance  it  becomes  negligible, \nand  the  more  remote  portions  of  the  line  may  be  completely  removed \nwithout  appreciable  effect  upon  the  disturbance  in  the  nearer  portion \nof  the  line.  That  part  of  the  line  which  then  remains  is  a  finite  net- \nwork of  pure  reactances,  and  in  any  such  network  all  currents  are \nalways  in  the  same,  or  opposite,  phase;  so,  also,  are  the  potential \ndifferences;  moreover,  the  two  are  mutually  in  quadrature;  there  is \nno  continuous  accumulation  of  energy  anywhere,  but  only  an  ex- \nchange of  energy  back  and  forth  between  the  inductances,  the  ca- \npacities and  the  generator.  Continuously  varying  the  amount  of \nthe  assumed  attenuation  will  cause  a  continuous  variation  in  the \ncorresponding  frequency.  The  motion  of  the  assumed  character \nmay,  therefore,  be  expected  to  occur  throughout  continuous  ranges \nor  bands  of  frequencies  and  not  merely  at  isolated  frequencies.\n\nThe  question  may  be  asked \u2014 How  far  does  the  energy  surge?  Is \nthe  surge  localized  in  the  individual  section,  or  does  the  surge  carry \nthe  energy  back  and  forth  over  more  than  one  section,  or  even  in  and \nout  of  the  line  as  a  whole?  To  answer  this  question,  it  would  be \nnecessary,  as  we  will  now  proceed  to  prove,  to  know  something  about \nthe  actual  construction  of  the  individual  section.  If  each  section  is \nactually  made  up  as  shown  in  Fig.  6,  and  this  is  entirely  possible  in \nthe  present  case  (since  only  positive  and  negative  reactances  would \nbe  called  for) ,  then  the  section  is  capable  of  free  oscillation,  as  explained\n\nabove,  and  the  surging  is  localized  within  the  section;  twice  during \neach  cycle  the  amount  of  energy  increases  on  the  right  and  decreases \non  the  left.  But  we  do  not  know  that  the  section  is  made  up  like \nFig.  G;  we  only  know  that  it  is  ecjuivalent  to  Fig.  6  as  regards  input \nand  output  relations.  As  far  as  these  external  relations  go,  the  actual \nnetwork  may  be  made  exclusively  of  either  inductances  or  capacities \nwith  the  connections  shown  in  Fig.  4  or  with  the  cross-connections \nof  Fig.  5,  according  as  the  current  is  to  have  the  same  or  opposite \nsigns  in  consecutive  sections.  In  any  network  made  up  exclusively \nof  inductances  or  of  capacities,  the  total  energy  falls  to  zero  when \nthe  current  or  the  potential  falls  to  zero,  respectively.  Twice,  there- \nfore, in  every  cycle  the  total  energy  surges  into  this  line  and  then  it \nall  returns  to  the  generator.  With  other  networks,  surgings  inter- \nmediate betw'een  these  two  extremes  will  occur.  The  theorem, \ntherefore,  does  not  limit  the  extent  of  the  surging.\n\nUnder  the  second  assumption,  the  phase  difference  between  the \ncurrents  at  two  given  points,  separated  by  a  periodic  interval,  is  to \nbe  an  angle  which  is  neither  zero  nor  a  multiple  of  \u00b1x.  The  assumed \ndifference  in  phase  can  only  be  due  to  the  infinite  extension  of  the \nartificial  line  since,  as  previously  noted,  no  finite  sequence  of  induct- \nances and  capacities  can  produce  any  difference  in  phase.  That \ninfinite  lines  do  produce  phase  differences  is  well-known;  in  particular, \nan  infinite,  uniform,  perfectly  conducting,  metallic  pair  shows  a \ncontinuous  retardation  in  phase.  If  the  infinitely  remote  sections \nof  the  artificial  line  are  to  have  this  controlling  effect  oti  the  wave \nmotion,  the  wave  motion  must  actually  extend  to  infinity,  that  is, \nthere  can  be  no  attenuation.  The  wave  progressing  indefinitely  to \ninfinity  without  attenuation  must  be  supplied  continuously  with \nenerg\\';  this  energy  must  flow  along  the  entire  line  with  neither  loss \nnor  gain  in  the  reactances  it  encounters  on  the  way.  This  continuous \nflow  of  energ>'  can  take  place  only  provided  the  currents  and  poten- \ntials are  not  in  quadrature;  they  may  be  in  phase.  In  considering \nthe  free  oscillations  of  Fig.  6  it  was  shown  that  K2  is  real  if  it  is  not \npure  reactance.  That  is,  for  the  mid-shunt  section  the  current  and \npotential  are  in  phase.  It  is  easy  to  show  that  they  are  also  in  phase \nat  the  mid-series  point  which  is  also  a  point  of  symmetry.\n\nThis  flow-of-energy  state  of  motion  thus  necessarily  characterizes \na  phase-retarded  wave  on  a  resistanceless  artificial  line,  regardless \nof  the  amount  of  the  assumed  positive  or  negative  retardation,  which \nmay  be  taken  to  have  any  value  betw'een  zero  and  exact  opposition \nof  phase.  Continuously  varying  the  retardation  throughout  the  180 \ndegrees  will,  in  general,  call  for  a  continuous  change  in  the  frequency\n\nof  the  wave  motion.  The  second  state  of  motion  occurs,  therefore, \nthrorghout  continuous  ranges  or  bands  of  frequencies.\n\nNo  other  state  of  motion  is  possible.  With  given  initial  amplitude \nand  pha^e  any  possible  wave  motion  is  completely  defined  by  its \nattenuation  and  phase  change.  All  possible  combinations  of  these \ntwo  elements  have  been  included  in  the  two  states,  since  the  excluded \nconditions  on  each  assumption  have  been  included  as  a  consequence \nof  the  other  assumption.  Thus,  the  exclusion  of  no  attenuation  in \nthe  first  assumption  was  found  necessarily  to  accompany  the  phase \nchange  of  the  second  assumption;  currents  in  phase  or  opposed, \nwhich  were  excluded  from  the  second  assumption,  were  found  to  be \nnecessary  features  accompanying  the  first  assumption.  There  remains \nonly  to  consider  the  critical  frequencies  separating  the  two  states  of \nmotion.  At  these  frequencies  there  can  be  no  attenuation  and  lag \nangles  of  multiples  of  ^tt,  including  zero,  only.  At  symmetrical \npoints  the  iterative  impedance  of  the  line  must  be  a  pure  reactance \nto  satisfy  the  first  state  of  motion,  and  a  pure  resistance  to  satisfy \nthe  second  state  of  motion.  The  only  iterative  impedances  which \nsatisfy  these  conditions  are  zero  and  infinity.\n\nSome  details  relating  to  the  pass  and  stop  bands  and  the  criti- \ncal frequencies  are  brought  together  in  the  following  table,  where \n\u25a0\"  stop  (  \u00b1  )\"  refers  to  stop  bands,  the  current  being  in  phase  or  op- \nposed in  successive  sections,  and  where  7  and  k  refer  to  the  line  obtained \nby  uniformly  distributing  I/Z2  with  respect  to  Zi.\n\nIt  is  not  necessary  to  check  the  table  item  by  item,  many  of  which \nhave  already  been  proven,  but  it  will  be  instructive  to  check  some  of \nthe  items  by  assuming  that  Z1/4Z2,  called  the  ratio  for  brevity,  is \npositive  to  begin  with,  and  that  a  continuous  increase  in  frequency \nreduces  the  ratio  to  zero  and  back  through  =f  00  to  its  original  posi- \nti\\e  value.  This  cycle  starts  with  a  stop  (  +  )  band  since  the  artificial \nline  is  in  effect  a  network  of  reactances,  all  of  which  have  the  same \nsign ;  there  is  attenuation  and  the  iterative  impedances  are  imaginary. \nWhen  the  ratio  decreases  to  zero,  there  must  be  either  resonance \nwhich  makes  Z\\  =  0,  or  anti-resonance  which  makes  Z2  =  00  ;  in \neither  case  the  artificial  line  has  degenerated  into  a  much  simpler  cir- \ncuit; it  is  a  shunt  made  up  of  all  Z2's  combined  in  parallel,  or  a  simple \nseries  circuit  made  up  of  all  Zi's,  respectively;  the  iterative  imped- \nances are  0  and  ^  ,  respectively ;  there  is  no  attenuation  in  either  case.\n\nWith  a  somewhat  further  increase  of  the  frequency  the  ratio  will \nassume  a  small  negative  value  with  the  result  that  the  artificial  line \nwill  have  both  kinetic  and  potential  energy.  An  analogy  now  exists \nbetween  the  artificial  line  and  an  ordinary  uniform  transmission  line, \nwhich  possesses  both  kinetic  and  potential  energy,  and  is  ordinarily \nvisualized  as  being  equivalent  to  many  small  positive  reactances,  in \nseries,  bridged,  to  the  return  conductor,  by  large  negative  reactances. \nThe  fact  that  uniform  lines  do  freely  transmit  waves  is  a  well-known \nphysical  principle,  and  it  is  not  necessary  to  repeat  here  the  physical \ntheory  of  such  transmission  merely  to  show  that  the  same  phenomenon \noccurs  with  the  identical  structure  when  it  is  called  an  artificial  line \nor  wave-filter.\n\nIn  order  to  determine  just  how  far  the  ratio  may  depart  from  zero, \non  the  negative  side,  without  losing  the  property  of  free  transmission, \nwe  look  for  any  change  in  the  action  of  the  individual  section  of  the \nartificial  line  which  is  fundamental ;  nothing  less  than  a  fundamental \nchange  in  the  behavior  of  the  individual  section  can  produce  such  a \nradical  change  in  the  line  as  an  abrupt  transition  from  the  free  trans- \nmission of  a  pass  band  to  the  to-and-fro  surging  of  energy  in  a  stop \nband.  Now  as  the  ratio  is  made  more  and  more  negative  by  the \nassumed  increase  of  frequency,  the  value  \u20141  is  reached,  at  which \nfrequency  the  symmetrical  section  (Fig.  6)  of  the  artificial  line  is \ncapable  of  free  oscillation  by  itself.  This  is  well  recognized  as  a \nmost  fundamental  change  in  the  properties  of  any  network,  and  it \nafifords  grounds  for  expecting  a  complete  change  in  the  character  of \nthe  propagation  over  the  artificial  line.  The  change  must  be  to  a \nstop  band  with  currents  in  opposite  phase,  since  at  resonance  the \npotentials  at  the  two  ends  of  a  section  are  in  opposite  phase.\n\nFurther  increase  in  the  frequency  cannot  make  any  change  in  the \nabsolute  difference  in  phase  between  the  two  ends  of  the  other  section, \nsince  opposition  is  the  greatest  possible  difiference  in  phase;  the  wave \nnow  adapts  itself  to  increasing  frequency  by  altering  its  attenuation.\n\nUpon  continuing  the  increase  of  frequency,  so  as  to  reduce  the \nratio  to  \u2014  00  ,  we  arrive  at  either  anti-resonance  corresponding  to \nZi  =  00  or  resonance  corresponding  to  Z2  =  0;  the  artificial  line  has \nnow  degenerated  into  a  row  of  isolated  impedances  Z2,  or  into  a  series \nof  impedances  Zi  short-circuited  to  the  return  wire;  in  either  case \nthe  attenuation  is  infinite  since  no  wave  is  transmitted.  Passing \nbeyond  this  critical  frequency  the  ratio  becomes  positive,  according \nto  our  assumption,  and  we  are  again  in  a  stop  (-f )  band.\n\nWhile  in  this  rapid  survey  of  what  happens  during  this  frequency \ncycle  little  has  been  actually  proven,  it  should  have  been  made \nphysically  clear  why  abrupt  changes  in  the  character  of  the  trans- \nmission occur  at  the  frequencies  making  the  ratio  equal  to  0,  \u2014  1  or \n00 ,  since  the  line  degenerates  into  a  simpler  structure,  or  the  phase \nchange  reaches  its  absolute  maximum,  on  account  of  resonance,  at \nthese  particular  frequencies.\n\nInformation  as  to  the  location  of  the  bands  is  often  obtained  most \nreadily  by  plotting  both  Zi  and  \u2014AZ^,  as  illustrated  in  Fig.  7,  and \ndetermining  the  critical  frequencies  by  noting  where  the  curves  cross \neach  other  and  the  abscissa  axis,  as  well  as  where  they  become  in- \nfinite. Any  particular  band  is  then  a  pass  band,  a  stop  (  +  )  band \nor  a  stop  (  \u2014 )  band,  according  as  Zi,  the  abscissa  axis,  or  \u2014  4Z2  lies \nbetween  the  other  two  of  the  three  lines.  In  Fig.  7  the  pass  bands \nare  Pi,  P2,  Pz,  Pi\\  the  stop  (  +  )  bands  are  52,  54,  56;  and  the  stop \n(  \u2014 )  bands  are  5i,  S^,  St,  5?,  and  they  illustrate  quite  a  variety  of \nsequences.  By  altering  the  curves  the  bands  may  be  shifted,  may \nbe  made  to  coalesce,  or  may  be  made  to  vanish.\n\nThe  pass  band  and  stop  band  characteristics  of  wave-filters  are \nconcretely  illustrated  for  a  few  typical  cases  by  the  curves  of  Figs. \n8-13,  which  show  the  attenuation  constant  A,  the  phase  constant  B, \nand  both  the  resistance  R  and  reactance  X  components  of  the  itera- \ntive impedance  for  a  range  of  frequencies  which  include  all  of  the \ncritical  frequencies,  except  infinity.  The  heavy  curves  apply  to  the \nideal  resistanceless  case,  while  the  dotted  curves  assume  a  power \nfactor  equal  to  l/(207r)  for  each  inductance  which  is  a  value  readily \nobtained  in  practice.  This  value  is,  however,  not  sufficiently  large  to \nmake  these  small  scale  curves  entirely  clear,  since  considerable  por- \ntions of  the  dotted  curves  appear  to  be  coincident  with  the  heavy  line \ncurves;  but  this,  as  far  as  it  goes,  proves  the  value  of  the  present  dis- \ncussion which  rests  upon  a  close  approximation  of  actual  wave-filters \nto  the  ideal  resistanceless  case.\n\nThe  low  pass  resistanceless  wave-filter,  as  shown  by  Fig.  8,  pre- \nsents no  attenuation  below  1,000  cycles;  above  this  frequency  the \nattenuation  constant  increases  rapidly,  in  fact,  the  full  line  attenuation \ncurve  increases  at  the  start  with  maximum  rapidity,  since  it  is  there \nat  right  angles  to  the  axis.  The  dotted  attenuation  curve,  which  in- \ncludes the  effective  resistance  in  the  inductance  coils,  follows  the \nideal  attenuation  curve  closely,  except  in  the  neighborhood  of  1,000 \ncycles,  where  resistance  rounds  ofi  the  abrupt  corner  which  is  present \nin  the  ideal  A  curve.  The  phase  constant  B  is,  at  the  start,  propor- \ntional to  the  frequency,  as  for  an  ordinary  uniform  transmission  line; \nits  slope  becomes  steeper  as  the  critical  frequency  1,000  is  approached \nwhere  the  curve  reaches  the  ordinate  tt,  at  which  value  it  remains \nconstant  for  all  higher  frequencies.  As  shown  by  the  dotted  B \ncurve,  resistance  rounds  off  the  corner  at  the  critical  frequency,  but\n\nFig.  8\u2014 Low  Pass  Wave-Filter:  L  =  I/tt  Henry,  C'  =  1  V  Microfarad \nFig.  9\u2014 Complementary  High  Pass  Wave-Filter:  Z,  =  l/47r,  C  =  IAtt\n\nOtherwise  leaves  the  curve  approximately  uinhaiigcd.  The  full  line \ncurves  for  R^  and  A'l  show  that  in  the  ideal  case  the  iterative  im- \npedance is  pure  resistance  and  pure  reactance  in  the  pass  and  stop \nbands  respectively,  and  that  resistance  smooths  the  abrupt  transition \nat  the  critical  frequency.\n\nThe  high  pass  wave-filter  shown  by  Fig.  9  passes  the  band  which \nis  stopped  by  the  low  pass  wave-filter  of  Fig.  8,  and  vice  versa.  For \nthis  reason  the  two  wave-filters  are  said  to  be  complementary.\n\nAnother  set  of  two  complementary  wave-filters  is  shown  by  Figs. \n10  and  11,  one  of  which  passes  only  a  single  band  of  frequencies, \nnot  extending  to  either  zero  or  infinity,  while  the  other  passes  the \nremaining  frequencies  only.  The  single  pass  band  of  Fig.  10,  em- \nbracing a  total  phase  change  27r  on  the  B  curve,  is  actually  a  case  of \nconfluent  pass  bands,  each  of  which  embraces  the  normal  angle  tt. \nThe  tendency  of  the  two  simple  pass  bands  to  separate,  and  leave  a \nstop  band  between  them,  is  shown  by  the  hump  in  the  dotted  at- \ntenuation constant  curve  at  1,000  cycles.  If,  instead  of  the  two \nsimple  bands  having  been  brought  together,  one  of  them  had  been \nrelegated  to  zero  or  infinity,  the  single  remaining  pass  band  would \nhave  exhibited  the  normal  angular  range  r  in  the  B  curve,  and  there \nwould  have  been  no  hump  in  the  dotted  A  curve.  The  stop  band  of \nFig.  11  also  illustrates  peculiarities  which  are  not  necessary  features \nof  a  wave-filter  with  a  single  stop  band  in  this  position.  This  wave- \nfilter  is  obtained  from  Fig.  7  by  making  all  bands  vanish  except \nPi,  Sz,  So  and  P3, \u2014 by  extending  P2  to  zero,  P3  to  infinity,  and  making \nS3  and  Si  coalesce,  so  that  the  attenuation  becomes  infinite  in  the \nstop  band  without  passing  from  a  stop  (  \u2014 )  to  a  stop  (-f-)  band. \nThe  coalescing  stop  bands  are  responsible  for  the  rapid  changes  in \nthe  B,  Ri,  and  A'l  curves  of  Fig.  11  which  would  not  have  appeared \nif,  in  Fig.  7,  the  same  pass  band  had  been  obtained  by  retaining  Pi, \n52  and  P2  and  making  all  other  bands  vanish.\n\nAn  extreme  case  of  complementary  wave-filters  is  shown  by  Figs. \n12  and  13,  where  no  frequencies  and  all  frequencies  are  passed  re- \nspectively. The  first  result  is  obtained  by  combining  inductances \nalone,  which,  as  has  been  pointed  out  above,  can  give  only  an  at- \ntenuated disturbance  devoid  of  wave  characteristics.  The  wave- \nfilter  shown  for  passing  all  frequencies  has  inductance  coils  in  the \nline,  and  capacities  diagonally  bridged  across  the  line.  This  wave- \nfilter  combines  a  constant  iterative  impedance  with  a  progressive \nchange  in  phase  which  is  sometimes  useful.\"*     An  outstanding  char-\n\n*  A  theoretical  use  of  the  phase  shifting  afforded  by  the  lattice  artificial  line  was \nmade  at  page  253  of  \"Maximum  Output  Networks  for  Telephone  Substation  and \nRepeater  Circuits,\"  Trans.  A.  I.  E.  E.,  vol.  39,  pp.  231-280,  1920.\n\nFig.  12\u2014 No  Pass  Wave-Filter:  L,  =  1/t,  Lj  =  l/4x \nFig.  13 \u2014 Complementary  All  Pass  Wave-Filter:  L  =  l/27r,  C  =  l/2j\n\nacteristic  of  this  type  of  artificial  line  is  that  it  has,  for  all  frequencies, \nthe  same  iterative  impedance  as  a  uniform  line  with  the  same  total \nseries  and  shunt  impedances.  This  artificial  line  will  be  considered \nin  more  detail  in  the  next  section  of  this  paper.\n\nUp  to  this  point  we  have  considered  the  properties  of  artificial  line \nnetworks  which  were  supposed  to  be  given.  In  practice  the  problem \nis  ordinarily  reversed,  and  we  ask  the  questions:  May  the  locations \nof  the  bands  be  arbitrarily  assigned?  May  additional  conditions  be \nimposed?  How  may  the  corresponding  network  be  determined,  and \nwhat  is  its  attenuation  in  terms  of  the  assigned  critical  frequen- \ncies?    These  questions  might  be  answered  by  a  study  of  Fig.  7,  in\n\nall  its  generality,  but  it  seems  simpler  to  base  the  discussion  upon \nthe  artificial  line  shown  in  Fig.  14,  which  is  to  be  a  generalization \nof  Fig.  13  to  the  extent  of  making  the  two  impedances  Zi  and  Z2 \nany  possible  actual  driving-point  impedances.  It  is  sometimes \nilluminating  to  regard  this  artificial  line  as  a  nest  of  bridges,  one \nwithin  another,  as  shown  by  Fig.  15.\n\nOn  interchanging  terminals  3  with  4  and  7  with  8  in  Fig.  14  the \nnetwork  of  lines  remains  unchanged;  thus,  Zi  and  4Z2  may  be  inter- \nchanged in  the  formulas  for  the  artificial  line  with  no  change  in  the \nresult,  except,  possibly,  one  corresponding  to  a  reversal  of  the  current \nat  alternate  junction  points.  Another  elementary  feature  of  this  arti- \nficial line  is  that  it  degenerates  into  a  simple  shunt  or  a  simple  series \ncircuit  at  the  resonant  or  anti-resonant  frequencies,  respectively,  of \neither  Zx  or  Z2,  and  these  are  the  critical  frequencies,  terminating \nthe  pass  bands.  At  other  frequencies,  a  positive  ratio  Z1/4Z2  must \ngive  a  stop  band,  since  the  reactances  are  all  of  one  sign.  If  a  small \nnegative  value  of  this  ratio  gives  free  transmission,  as  we  naturally \nexpect,  there  will  be  identical  transmission,  except  for  a  reversal  of\n\nsign,  when  the  ratio  has  the  reciprocal  value,  which  will  be  a  large \nnegative  quantity,  since  we  may  always  interchange  Zi  and  4Z2. \nThe  consequences  of  this  and  of  other  elementary  properties  of  this \nartificial  line  are  brought  together  in  the  following  table:\n\nThe  cycle  of  bands:  stop   (  +  ),  pass,  stop  (  \u2014 ),  adopted  for  the \ntable,  carries   the    attenuation  factor    e~^    around  the  periphery  of\n\na  unit  semi-circle;  in  the  stop  (  +  )  band  it  traverses  the  radius  from \n0  to  1,  in  the  pass  band  it  travels  along  the  unit  circle  through  180 \ndegrees  to  the  value  \u20141,  completing  the  cycle  from  \u20141  to  0  in  the \nstop  (  \u2014 )  band.  In  this  cycle  there  are  four  points  of  special  interest, \ncorresponding  to  ratio  values  1,  0,  \u20141  and  oo,  for  which  the  wave  is \ninfinitely  attenuated,  unattenuated  with  an  angular  change  of  0, \nof  90,  and  of  180  degrees,  respectively.  It  is  at  the  90  degree  angle \nthat  resonance  of  the  individual  section  occurs;  the  iterative  im- \npedance is  then  equal  to  2\\Z2\\.\n\nIf  we  plot  Z\\  and  4Z2  the  pass  bands  are  shown  by  the  points  where \nthe  curves  become  zero  or  infinite,  and  the  intersections  of  the  two\n\nFig.  16 \u2014 Graph  for  Locating  the  Pass  and  Stop  Bands  of  the  Lattice  Artificial  Line,\n\ncurves  show  the  frequencies  at  which  the  attenuation  becomes  in- \nfinite. These  intersections  must  be  at  an  acute  angle  since  each \nbranch  of  the  two  curves  has  a  positive  slope  throughout  its  entire \nlength ;  for  this  reason  it  may  be  desirable  to  plot  the  ratio  rather  than \nthe  individual  curves;  this  is  especially  desirable  in  cases  where  the \ntwo  curves  do  not  intersect,  but  are  tangent.  Fig.  16  is  for  a  lattice \nnetwork  equivalent  to  two  sections  of  the  ladder  type  illustrated  by \nFig.  7,  and  so  cannot  include  a  stop  (  \u2014 )  band.  Accordingly,  the \nratio  does  not  go  above  unity,  although  it  reaches  unity  at  the  two \nfrequencies  300  and  400,  corresponding  to  the  infinite  attenuation \nwhere  stop   (  \u2014 )    and   stop    (  +  )    bands  meet  in   Fig.  7.     It   is  also\n\nunity  at  the  extreme  frequencies  zero  and  infinity.     The  four  pass \nbands  have,  of  course,  the  same  locations  as  in  Fig.  7.\n\nMultiplying  the  ratio  by  a  constant  greater  than  unity  introduces \nstop  (  \u2014 )  bands  along  with  the  stop  (  +  )  bands;  multiplying  it  by \na  constant  less  than  unity  removes  all  infinite  attenuations;  these \nchanges  within  the  stop  bands  are  made  without  altering  the  loca- \ntions of  the  four  pass  bands.\n\nIn  connection  with  practical  applications  we  especially  desire  to \nknow  what  latitude  is  permitted  in  the  preassignment  of  properties \nfor  a  wave-filter.  If  we  consider  first  the  ideal  lattice  wave-filter, \nits  limitations  are  those  inherent  in  the  form  which  its  two  inde- \npendent resistanceless  one-point  impedances^  Zi  and  Zo  may  assume. \nThe  mathematical  form  of  this  impedance  is  shown  by  formula  (7) \nof  the  appendix,  which  may  be  expressed  in  words  as  follows:\n\nWithin  a  constant  factor  the  most  general  one-point  reactance  obtain- \nable by  means  of  a  finite,  pure  reactance  network  is  an  odd  rational \nfunction  of  the  frequency  which  is  completely  determined  by  assigning \nthe  resonant  and  anti-resonant  frequencies,  subject  to  the  condition  that \nthey  alternate  and  include  both  zero  and  infinity.\n\nThe  corresponding  general  expressions  for  the  quotient  and  product \nof  the  impedances  Zi  and  Zj  are  shown  by  formulas  (8)  and  (9). \nDefinite,  realizable  values  for  all  of  the  2n-f2  parameters  and \n2n  +  l  optional  signs  occurring  in  these  formulas  may  be  deter- \nmined in  the  following  manner:\n\n(a)  Assign  the  location  of  all  n  pass  bands,  which  must  be  treated \nas  distinct  bands  even  though  two  or  more  are  confluent;  this \nfixes  the  values  of  the  2n  roots  pi  .  .  .  p2n  which  correspond \nto  the  successive  frequencies  at  the  two  ends  of  the  bands.\n\n(b)  Assign  to  the  lower  or  upper  end  of  each  pass  band  propagation \nwithout  phase  change  from  section  to  section;  this  fixes  the \ncorresponding  optional  sign  in  formula  (8)  as  +  or  \u2014 ,  respec- \ntively.\n\n(c)  Assign  a  value  to  the  propagation  constant  at  any  one  non- \ncritical  frequency  (that  is,  assign  the  attenuation  constant  in  a\n\n*  A  one-point  impedance  of  a  network  is  the  ratio  of  an  impressed  electromotive \nforce  at  a  point  to  the  resulting  current  at  the  same  point \u2014 in  contradistinction  to \ntwo-point  impedances,  where  the  ratio  applies  to  an  electromotive  force  and  the \nresulting  current  at  two  different  points.\n\nstop  band  or  the  phase  constant  in  a  pass  band) ;  this  fixes \nthe  value  of  the  constant  G  and  thus  completely  determines \nformula  (8)  on  which  the  propagation  constant  depends.\n\n(d)  Assign  to  the  lower  or  upper  end  of  each  stop  band  the  iterative \nimpedance  zero;  this  fixes  the  corresponding  optional  sign  in \nformula  (9)  as  +  or  \u2014 ,  respectively.\n\n(e)  Assign  the  iterative  impedance  at  any  one  non-critical  fre- \nquency (subject  to  the  condition  that  it  must  be  a  positive \nresistance  in  a  pass  band  and  a  reactance  in  a  stop  band) ; \nthis  fixes  the  constant  H  and  thereby  the  entire  expression  (9) \nupon  which  the  iterative  impedance  depends.\n\nThe  quotient  and  product  of  the  impedances  Zi  and  Z2  are  now \nfully  determined;  the  values  of  Zi  and  Zi  are  easily  deduced  and  also \nthe  propagation  constant  and  iterative  impedance  by  formulas  (11) \nand  (12);  Zi  and  Zi  are  physically  realizable  except  for  the  necessary \nresistance  in  all  networks.\n\nA  lattice  wave-filter  having  any  assigned  pass  bands  is  physically \nrealizable;  the  location  of  the  pass  bands  fully  determines  the  propagation \nconstant  and  iterative  impedance  at  all  frequencies  when  their  values \nare  assigned  at  one  non-critical  frequency,  and  zero  phase  constant  and \nzero  iterative  impedance  are  assigned  to  the  lower  or  upper  end  gf  each \npass  band  and  stop  band,  respectively.\n\nLattice  Artificial  Line   Equivalent  to  the   Generalized \nArtificial  Line  of  Fig.  1\n\nSince  any  number  of  arbitrarily  preassigned  pass  bands  may  be \nrealized  by  means  of  the  lattice  network,  it  is  natural  to  inquire \nwhether  this  network  does  not  present  a  generality  which  is  essen- \ntially as  comprehensive  as  that  obtainable  by  means  of  any  network \nN  in  Fig.  1,  provided  the  generalized  line  is  so  terminated  as  to  equalize \nits  iterative  impedances  in  the  two  directions.  This  proves  to  be \nthe  case.\n\nIf  network  A^  has  identical  iterative  impedances  in  both  directions, \nthe  lattice  network  equivalent  to  two  sections  of  N  is  shown  by  Fig. \n17;  each  lattice  impedance  is  secured  by  using  an  N  network;  the  N's \nplaced  in  the  two  series  branches  of  the  lattice  have  their  far  terminals \nshort-circuited  so  that  they  each  give  the  impedance  denoted  by \nZo;  the  N's  in  the  two  diagonal  branches  have  their  far  ends  open \nand  they  each  give  the  impedance  denoted  by  Zoo.\n\nThe  lattice  network  of  l*'ig.  IS  lias  in  each  branch  a  one-p(jinl  im- \npedance obtained  by  means  of  a  duplicate  of  the  given  network  A^ \nand  an  ideal  transformer.  The  two  lattice  branch  impedances  are \nZq-\\-Zr^2Zqr  whcrc  the  three  impedances  Z,,  Zr,  Z^r  are  the \nefTective  self  and  mutual  impedances  of  the  network  N  regarded  as  a \ntransformer.     This  lattice  network  has  identicalK'  the  same  propaga-\n\ntion  constant  as  the  single  network  N  shown  on  the  left.  Since  the \nlattice  cannot  have  difTerent  iterative  impedances  in  the  two  direc- \ntions, it  actually  compromises  by  assuming  the  sum  of  the  two  itera- \ntive impedances  presented  by  N.     A  physical  theory  of  the  equival-\n\nFig.   18 \u2014 Lattice  Network  Having  the  Same  Propagation  Constant  as  N  and  an \nIterative   Impedance  Equal  to  the  Sum  of   the  Two   Iterative    Impedances   of    A'^\n\nences  shown  in  Figs.  17  and  18  has  not  been  worked  up;  the  analytical \nproofs  were  made  by  applying  the  formulas  given  in  the  appendix \nunder  lattice  networks.\n\nWithout  going  to  more  complex  networks  it  is,  of  course,  not  pos- \nsible to  get  a  symmetrical  iterative  impedance,  but  that  is  not  necessary \nfor  our  present  purposes  where  we  are  concerned  primarily  with  the\n\npropagation  constant.     It  has  now  been  shown  with  complete  gener- \nality that:\n\nThe  lattice  artificial  line,  with  physically  realizable  branch  impedances , \nis  identically  equivalent  in  propagation  constant  and  mean  iterative \nimpedance  to  the  chain  of  identical  physically  realizable  networks  con- \nnected, together  in  sequence  through  tivo  pairs  of  terminals.\n\nTo  complete  this  simplification  of  the  generalized  artificial  line  it  is \nnecessary  to  know  the  simplest  possible  form  of  the  one-point  im- \npedances employed  in  the  branches  of  the  lattice  network.  The \ndiscussion  of  the  most  general  one-point  impedance  obtainable  by \nmeans  of  any  network  of  resistances,  self  and  mutual  inductances, \nleakages  and  capacities  will  find  its  natural  place,  together  with \nallied  theorems,  in  a  paper  on  the  subject  of  impedances.  For  the \npresent  purpose  it  is  sufficient  to  state:\n\nThe  most  general  branch  impedance  of  the  lattice  network  may  he \nconstructed  by  combining,  in  parallel,  resonant  circuits  having  im- \npedances of  the  form  R-\\-iLp-\\-{G-{-iCp)~^;  or  they  may  equally \nwell  be  constructed  by  combining,  in  series,  anti-resonant  circuits  having \nimpedances  of  the  form  \\G-\\-iCp-\\-{R-^iLp)~^Y~^\n\nThe  wave-filter  under  discussion  approximates  to  a  resistanceless \nartificial  line,  and  such  an  ideal  artificial  line  is  capable  of  two,  and \nonly  two,  fundamentally  distinct  states  of  motion.  In  one  state  the \ndisturbance  is  attenuated  along  the  line,  and  there  is  no  flow  of  energy \nother  than  a  back  and  forth  surging  of  energy,  the  intensity  of  which \nrapidly  dies  out  along  the  line.  In  the  other  state  there  is  a  free \nflow  of  energy,  without  loss,  from  section  to  section  along  the  line, \nwith  no  surge  of  energy  between  symmetrical  sections.  Each  state \nholds  for  one  or  more  continuous  bands  of  frequencies;  these  bands \nhave  been  distinguished  as  stop  bands  and  pass  bands.\n\nA  high  degree  of  discrimination,  between  different  frequencies, \nmay  be  obtained,  even  if  each  section,  taken  alone,  gives  only  a \nmoderate  difference  in  attenuation,  by  the  use  of  a  sufificient  number \nof  sections  in  the  wave-filter,  since  the  attenuation  factors  vary  in \ngeometrical  progression  with  the  number  of  sections.\n\nAny  number  of  arbitrarily  located  pass  bands  may  be  realized  by \nmeans  of  the  lattice  artificial  line;  furthermore,  the  propagation \nconstant  at  one  frequency,  and  the  iterative  impedance  at  one  fre- \nquency may  both  be  assigned,  while  the  location  of  zero  phase  con-\n\nstant  and  zero  iterati\\e  impedance  at  the  lower  or  upper  end  of  each \npass  band  and  stop  band,  respectively,  is  also  optional.  This  com- \npletely determines  the  lattice  artificial  line.  No  additional  condition, \nother  than  iterati\\e  impedance  asymmetry,  can  be  realized  by  re- \nplacing the  lattice  network  b>-  an\\-  four  terminal  network.\n\nFormulas  for  the  propagation  constant  and  iterative  impedance  of \nthe  generalized  artificial  line,  expressed  in  a  number  of  equivalent \nforms,  have  already  been  given  in  my  paper  on  Cisoidal  Oscillations,^ \nbut  it  seems  worth  while  to  deduce  the  formulas  anew  here  from  the \nfree  oscillations  of  the  detached  unit  circuit  of  Fig.  6,  so  as  to  complete \nthe  physical  theory  by  deducing  the  comprehensive  mathematical \nformulas  by  the  same  method  of  procedure.\n\nLadder  Network  Formulas \nNotation: \nZi,  Z2  =  series  impedance   and   shunt   impedance   of   the   section   of\n\nFig.  L \nV  =  A  -{-  iB  =  propagation  constant  per  section. \nK\\,  A'o  =  iterative  impedances  at  mid-series  and  mid-shunt. \nT  =  \u00ab  +  '^  =  V  Zi/  Z2  =  propagation  constant    for    uniform   distri- \nbution of  Zi  and  1  Z2,  per  unit  length. \n^  ~   V  ZiZo  =  iterative  impedance  of  this  same  uniform  line.\n\nthis  condition  it  is  sufficient  to  make  use  of  two  other  simple  relations:\n\nthe  proportionality  of  the  potential  drops  in  the  direction  of  the  current\n\n\u00ab\"Cisodial  Oscillations,\"  Trans.  A.  I.  E.  E.,  vol.  30,  pp.  873-90),  1911.  In  the \nlowest  row  of  squares  of  Table  I,  the  iterative  impedances  and  propagation  constant \nof  any  network  are  given  in  fi\\-e  different  wa\\s,  involving  one-point  and  two-point \nimpedances,  equivalent  star  impedances,  equivalent  delta  impedances,  equivalent \ntransformer  impedances,  or  the  determinant  of  the  network.  The  only  typo- \ngraphical errors  in  Table  I  appear  to  be  the  four  which  occur  in  the  first,  third  and \nfifth  squares  of  this  row:  in  tfie  values  for  K,  replace  {S,  \u2014  S^)  by  (5,  \u2014  Sr)  and \nplace  a  parenthesis  before  U,  \u2014  Ur)\\  in  the  first  value  of  AV  replace  5,^  by  S^-;  in \nthe  last  value  for  T^  add  a  minus  sign  so  that  it  reads  cosh\"'.\n\nKi,  the  mid-series  iterative  impedance  of  the  artificial  Hne,  to  the \ntotal  impedance  on  the  right  of  the  mid-point  of  the  series  impedance \nZi.     These  three  relations,  which  can  be  written  down  at  once,  are:\n\nand  the  formulas  for  Zi  and  Zo  in  terms  of  F  and  Kx  or  7^2  are  likewise \nfound  to  be:\n\nFormulas  (3)  and  (4)  are  in  the  nature  of  design  formulas  in  that \nthey  determine  the  impedance  Zi  and  Z2,  at  assigned  frequencies, \nwhich  will  ensure  the  assigned  values  of  V  and  K  at  these  frequencies. \nIn  general,  however,  it  Avould  not  be  evident  how  best  to  secure  these \nrequired  values  of  Zi  and  Zo;  complicated  or  even  impossible  net- \nworks might  be  called  for,  even  to  approximate  values  of  Zi  and  Z2 \nassigned  in  an  arbitrary  manner.  Fortunately,  practical  require- \nments are  ordinarily  satisfied  by  meeting  maximum  and  minimum \nvalues  for  the  attenuation  constant  throughout  assigned  frequency \nbands.  Formulas  (8)  and  (9)  may  be  employed  for  this  purpose  as \nexplained  below.\n\nIt  is  convenient  to  have  formulas  (1)  and  (2)  expressed  in  a  variety \nof  ways,  since  no  one  form  is  well  suited  for  calculation  throughout \nthe  entire  range  of  the  variables.  Accordingly,  the  following  analyti- \ncally equivalent  expressions  are  here  collected  together  for  reference:\n\nThe  formulas  leave  indeterminate  the  signs  of  7,  k,  F,  and  K,  and \nalso  a  term  ^i^ttw  in  7  and  T.  The  signs  are  to  be  so  chosen  that \nthe  real  parts  are  positive,  or  become  positive  when  positive  re- \nsistance is  added  to  the  system.  The  indeterminate  =*=  i27rn  can \nbe  made  determinate  only  after  knowing  something  of  the  internal \nstructure  of  the  unit  network  of  which  the  artificial  line  is  composed; \nthe  conditions  to  be  met  are \u2014 absence  of  phase  differences  when  all \nbranches  of  the  unit  network  A'^  of  Fig.   1   are  assumed  to  be  pure\n\nresistances  and  continuity  of  phase  as  reactances  are  gradually  intro- \nduced to  give  the  actual  network.\n\nFormula  (5)  is  adapted  for  use  in  the  pass  bands,  since  the  ex- \npressions are  real  when  y^  is  real,  negative  and  not  less  than  \u2014  4; \nsimilarly,  formulas  (5a)  and  (5b)  are  adapted  for  use  in  the  stop  (\u00b1) \nbands,  that  is,  when  7^  is  positive  and  less  than  \u2014  4  respectively.\n\nFrom  the  theory  of  impedances  we  know  that  any  resistanceless \none-point  impedance  is  expressible  in  the  form\n\nwhere  the  factor  D  and  the  roots  px,  pi,  .  .  .  pin  are  arbitrary  positive, \nreals  subject  only  to  the  condition  that  each  root  is  at  least  as  large \nas  the  preceding  one.  This  enables  us  to  write  down  the  forms  which \nthe  quotient  and  product  of  two  resistanceless  one-point  impedances \nmay  assume,  which  are  as  follows:\n\nIf  in  formulas  1,  2,  5  and  6  we  substitute  for  Z1/Z2  =  y\"^  and \nZi  Zi  =  k\"^  the  right-hand  side  of  formulas  (8)  and  (9),  respectively, \nwe  obtain  formulas  for  the  propagation  constant  and  iterative  im- \npedance of  an  artificial  resistanceless  line  in  terms  of  frequencies  at \nwhich  the  propagation  constant  becomes  zero  or  infinite.  Ordi- \nnarily, however,  we  are  more  interested  in  having  expressions  in \nterms  of  the  frequencies  which  terminate  the  pass  bands.  To  secure \nthese  the  substitutions  should  be  4[8]/(4  -  [8]  )  and  [9]  (1  -  [8]/4)^S \nwhere  [8]  and  [9]  stand  for  the  entire  right-hand  sides  of  formulas \n(8)  and  (9).  This  substitution  amounts  to  obtaining  the  lattice  net- \nwork giving  the  required  pass  bands,  and  then  transforming  to  the\n\nladder  network  having  the  same  propagation  constant  and  the  same \niterative  impedance  at  mid-series  or  mid-shunt.\n\nThe  impedances  of  a  single  section  between  terminals  1  and  2,  with \nthe  far  end  of  the  section  3  and  4  either  short-circuited  or  open,  are \nreadily  seen  to  be\n\nSince  vZoZ 00  and  'V  Zq/Zco  are  the  iterative  impedance  and  the \nhyperbolic  tangent  of  the  propagation  constant  for  any  symmetrical \nartificial  line,  we  have  the  following  analytically  equivalent  formulas \nfor  the  lattice  network  where  7  =  v  Z1/Z2,  and  k  =  V  Z1Z2  as  for \nthe  ladder  type.\n\nIn  these  formulas  Z1/Z2  =  7^  and  Zi  Z2  =  k\"^  might  be  expressed \nin  terms  of  the  resonant  and  anti-resonant  complex  frequencies  of \nZi  and  Z2,  the  frequencies  being  made  complex  quantities  so  as  to \ninclude  the  damping.  Where  there  is  no  damping,  that  is,  where  all \nnetwork  impedances  are  devoid  of  resistance,  the  simplified  forms \nof  these  expressions  are  given  by  formulas  (8)  and  (9).  The  use  of \nthese  formulas  for  designing  wave  filters  having  assigned  pass  bands \nis  explained  at  page  23.\n\nNote:  Much  has  been  written  on  the  subject  of  the  binaural  location \nof  pure  tones  but  the  case  of  complex  sounds  has  received  little  attention \nin  recent  literature.  The  purpose  of  the  present  paper  is  to  bring  the  dis- \ncussion of  complex  sounds  abreast  of  that  relating  to  pure  tones.  Those \nwho  wish  to  acquaint  themseUcs  with  the  work  on  pure  tones  will  be  inter- \nested in  reading  the  theoretical  work  of  the  authors  and  the  experimental \nstudies  carried  out  by  G.  VV.  Stewart  and  students  working  under  his \ndirection.  This  work  has  been  reported  in  various  papers,  most  of  which \nhave  appeared  during  recent  years  in  the  Physical  Review  and  the  Physi- \nkalische  Zeitschrift.\n\nA  resume  of  the  present  paper  is  given  by  the  authors  in  their  concluding \nparagraph. \u2014 Editor.\n\nTHE  need  of  determining  the  location  of  enemy  submarines  and \naeroplanes  during  the  war  brought  into  use  practical  methods \nfor  locating  a  sound  source  which  depend  upon  differences  between \nthe  sound  waves  reaching  the  two  ears.  This  stimulated  a  general \nstudy  of  the  phenomena  involved  in  binaural  sound  location.  The \nfoundation  for  this  study  had  already  been  laid  in  the  work  of  Lord \nRayleigh  and  others,  who,  following  more  or  less  in  his  footsteps, \nhad  accumulated  a  considerable  amount  of  information  of  both \ntheoretical  and  experimental  sorts.  Of  this  information  almost  all \nthat  was  of  a  theoretical  nature  and  a  considerable  portion  of  the \nexperimental  kind  dealt  only  with  the  location  of  pure  tones,  the  more \ncomplicated  and  in  some  respects  more  important  problem  of  complex \nsounds  being  almost  entirely  neglected.  Such  advances  as  were \nmade  in  the  theoretical  aspects  of  the  problem  during  the  war  were \nsubject  to  the  same  restriction  so  that  even  to-day  no  comprehensive \ntheory  has  been  advanced  which  adequately  covers  the  problem  of \nthe  location  of  such  sounds  as  occur  in  every-day  life,  and  in  the \npractical  applications  of  binaural  methods.  However,  the  results \nobtained  with  pure  tones  can  be  made  to  throw  considerable  light \nupon  the  problem,  and  it  is  primarily  from  this  standpoint  that  the \nfollowing  discussion  is  written.\n\nIt  may  be  well  at  the  outset  to  review  some  of  the  outstanding \ndifferences  between  the  observed  phenomena  in  the  two  cases.  The \naccuracy  of  location  is  much  less  for  pure  tones,  as  is  also  the  sense \nof  definiteness  of  the  sound  image.  The  location  of  pure  tones  is  almost \nwholly  binaural  as  is  evidenced  by  the  inability  of  persons  deaf  in  one \near  to  locate  such  a  tone.  With  complex  sounds  not  only  is  the \nlocation  by  binaural  effects  more  accurate  and  definite,  but  also  the \nobserver  is  not  dependent  on   these  alone.     Persons  who  are  deaf\n\nin  one  ear  can  locate  familiar  complex  sounds  almost  as  well  as  those \nwith  normal  hearing.\n\nPractically  all  theories  of  sound  location  start  from  the  assumption \nthat  the  listener  subconsciously  observes  certain  sound  characteris- \ntics which  depend  upon  the  position  of  the  source  and  forms  a  judg- \nment of  where  the  source  must  be  by  comparing  these  characteristics \nwith  information  which  he  has  stored  up  as  a  result  of  his  past  ex- \nperience with  cases  in  which  the  position  of  the  source  was  known. \nIn  order  to  fix  the  position  of  the  source  he  must  assign  to  it  three \ncoordinates  such  as  its  distance  and  some  two  angles  which  define \nits  direction.  To  do  this  he  must  be  able  to  observe  at  least  three \nindependent  properties  of  the  sound  which  are  functions  of  the  posi- \ntion of  the  source.  If  fewer  than  three  are  available  some  difficulty \nin  location  is  certain  to  arise.  If  more  than  three  are  available  there \nis  the  possibility  of  a  number  of  simultaneous  independent  determina- \ntions of  the  three  coordinates.\n\nIf  the  sounds  of  every-day  life  were  never  distorted  in  transmission \nall  of  these  determinations  would  yield  the  same  set  of  coordinates \nand  the  only  advantage  which  the  listener  would  gain  from  the  addi- \ntional information  available  would  lie  in  the  fact  that  some  one  set \nmight  be  peculiarly  sensitive  to  slight  differences  in  the  position \nof  the  source,  and  therefore  might  lead  to  increased  certainty  on  the \npart  of  the  observer.  Owing  to  reflection  from  the  walls  of  buildings \nand  the  like,  the  sounds  of  every-day  life  seldom  arrive  undistorted, \nso  that  the  observer  must  always  be  somewhat  uncertain  as  to  whether \nor  not  the  coordinates  of  the  sound  source  are  actually  those  which \nhe  deduces  from  the  properties  of  the  sound  wave  as  it  reaches  his \nears.  If  enough  properties  are  available  to  permit  him  to  make \ntwo  independent  determinations  he  may  use  one  of  them  to  check  the \nother,  and  if  they  agree  he  is  justified  in  a  feeling  of  increased  cer- \ntainty as  to  the  accuracy  of  his  judgment.  The  more  independent \ndeterminations  he  can  make  the  more  checks  he  will  be  able  to  apply \nand  consequently  the  more  confident  he  will  be.^\n\nIt  should  not  be  inferred,  however,  that  it  is  only  the  sounds  of \nthe  street  which  reach  the  observer  in  a  distorted  form.  In  a  great \nmany  laboratory  experiments  the  characteristics  of  the  sounds  have\n\n^  It  is  interesting  to  note  in  this  connection  that  it  is  not  surprising  that  an  observer \nlocates  a  complex  tone  with  much  greater  certainty  than  a  pure  tone  when  we  con- \nsider how  rapidly  the  number  of  independent  sets  of  data  increases  with  increase \nin  complexity  of  sound.  We  have  already  said  that  three  indepsndent  pro  jerties \nare  needed  for  the  determination  of  the  three  coordinates  of  the  source.  Hence \nif  only  three  are  available,  only  one  determination  can  be  made  and  no  checks  are \npossible.  On  the  other  hand,  if  four  are  available,  four  groups  of  three  each  can \nbe  formed  and  therefore  four  separate  determinations  can  be  made.  Similarly, \n10  determinations  can  be  made  from  5  properties,  20  from  6,  and  120  from  10.\n\nbeen  inconsistent,  and  in  some  cases  thc>'  have  not  e\\en  corresponded \nto  any  actual  source  whatever.  Under  these  circumstances,  if  an \nimage  is  formeci  at  all,  some  purely  ps\\'chological  factors  must  enter \nin.  For  pure  tones  it  has  been  found  possible  to  explain  much  of  the \nexperimental  data  obtained  under  circumstances  such  as  this  by \nassuming  that  the  observer  subconsciously  judges  one  or  more  of  the \ncharacteristics  to  be  in  error  and  applies  such  corrections  as  will \nmake  all  of  the  data  correspond  to  an  actual  source.  As  a  criterion \nfor  determining  which  characteristics  will  be  altered,  it  is  assumed \nthat,  in  general,  those  are  chosen  which  require  the  smallest  changes.\n\nLet  us  now  consider  what  characteristics  are  available  for  locating \nsounds  of  different  kinds.  A  pure  tone  from  a  source  at  rest  with \nrespect  to  the  observer  has  at  any  point  only  two  physical  character- \nistics which  are  subject  to  change  with  the  position  of  the  source. \nThey  are  its  amplitude  and  phase.  Corresponding  to  each  position \nof  the  source  there  is  a  particular  amplitude  and  phase  at  each  of  the \ntwo  ears  so  that  a  total  of  four  properties \u2014 the  loudness  of  the  sound, \nthe  average  phase,  the  difference  in  amplitude  (which  mayconven- \niently  be  expressed  as  a  ratio)  and  the  difference  in  phase  at  the  two \nears\u2014 are  available  for  determining  the  position  of  the  source.  It  is \ninconceivable  that  the  average  phase  can  have  anything  to  do  with \nthe  location  of  the  sound  since  it  may  be  changed  at  will  without \naltering  the  position  of  the  source.  The  same  remark  applies  to  the \nloudness  of  the  sound  except  in  those  instances  where  the  observer \nis  familiar  with  the  source  to  such  an  extent  as  to  know  how  loud \nit  may  be  expected  to  be.  Hence,  if  we  restrict  ourselves  to  the \ncases  in  which  prejudicial  information  of  this  sort  does  not  exist, \nwe  find  that  the  observer  has  only  two  quantities  from  which  he  may \ndeduce  the  position  of  the  source.  We  should  therefore  expect  that \nthese  two  quantities  would  make  it  possible  to  locate  the  tone  with \nrespect  to  two  coordinates  only.  This  is  found  to  be  in  general \nagreement  with  experiment,  for  most  observers  locate  all  sources  of \npure  tones  in  the  same  horizontal  plane  with  their  heads  and  determine \nonly  the  distance  and  angular  departure  from  the  median  plane.  If \nthe  source  is  more  than  a  few  yards  away  the  intensity  ratio  and  phase \ndifference  change  very  slowly  with  distance  so  that  in  this  case  even \nthe  sense  of  distance  is  not  keen  and  a  feeling  of  certainty  exists  with \nrespect  to  the  direction  only.\n\nIn  many  experiments  the  tones  at  the  two  ears  have  been  varied \narbitrarily  so  as  to  give  combinations  having  equal  phases  and  un- \nequal intensities  or  vice  versa \u2014 combinations  which  cannot  arise \nfrom  actual   physical   sources  in   the  absence  of  distortion.     Under\n\nthese  conditions  the  observer  generally  corrects  one  to  a  value  con- \nsistent with  the  other  except  in  extreme  cases  where  the  correction \nrequired  for  this  purpose  would  be  inordinately  large.  When  this \noccurs  he  may  either  assume  both  to  be  correct  and  form  two  images \u2014 \none  based  on  the  phase  difference  together  with  a  mentally  supplied \nintensity  ratio  consistent  with  it,  and  the  other  similarly  derived \nfrom  the  observed  intensity  ratio \u2014 or  he  may  fail  to  have  a  sense \nof  location  at  all.\n\nBefore  considering  the  available  characteristics  of  complex  sounds \nin  general  let  us  confine  our  attention  for  a  time  to  those  which  are \nmade  up  of  a  limited  number  of  sustained  pure  tones  such  as  an  organ \nnote  with  its  series  of  overtones,  or  a  group  of  tuning  forks.  Here  the \nnumber  of  characteristics  increases  rapidly  with  the  number  of  com- \nponent tones.  For  each  component  tone  there  are  two  quantities: \nintensity  ratio  and  phase  difference.  In  addition,  at  either  ear  alone \nthe  relative  intensities  of  any  two  of  the  tones  changes  with  the \nposition  of  the  source,  owing  to  the  diffraction  of  the  sound  waves \naround  the  head  being  different  for  different  frequencies.  There  are \ntherefore  as  many  of  these  observable  intensity  ratios  as  there  are \npairs  of  components.  Similarly,  for  any  two  tones  whose  frequencies \nare  commensurable,  the  relative  phases  of  the  two  at  the  same  ear \ndepend  upon  the  position  of  the  source.\n\nNot  all  of  these  characteristics  are  capable  of  contributing  to \nbinaural  as  distinct  from  monaural  location.  In  fact,  only  the  phase \ndifferences  and  intensity  ratios  of  the  separate  components  are  bi- \nnaural. A  man  who  is  deaf  in  one  ear  has  available  all  of  the  rela- \ntions between  the  intensities  and  phases  of  the  various  components \nat  his  normal  ear.  That  these  relations  do  actually  contribute  to \nsound  location  is  supported  by  experimental  evidence.  Myers  ^ \nfound  that,  after  familiarizing  himself  with  a  complex  sound,  a  blind- \nfolded observer  could  locate  its  position  with  considerable  accuracy, \neven  when  it  was  moved  about  in  the  median  plane,  but  that  his \naccuracy  could  be  destroyed  by  varying  the  relative  intensities  of  the \ncomponents.'  It  is  not  surprising  then,  that  for  complex  sounds  the \naccuracy  is  about  the  same  whether  the  location  is  binaural  or  mo- \nnaural.*    The  observed  failure  of  monaural  location  in  the  case  of  a\n\n'  It  should  be  noticed  that  this  effect  must  have  been  purely  psychological  since  it \ncould  be  produced  without  moving  the  source  at  all.  It  therefore  lends  plausibility \nto  the  assumption  upon  which  our  theory  is  based:  that  when  discordant  or  unusual \nstimuli  are  experienced,  a  mental  readjustment  of  the  stimuli  is  made  in  order  to \nrender  them  more  nearly  consistent  with  every-day  experience.\n\n*  As  shown  by  the  experiments  of  Angell  and  Fite  upon  persons  deaf  in  one  ear. \nPsychol.  Rev.,  vol.  8,  pp.  225-246,  1911.\n\npure  tone  follows  directly  from  the  absence  of  other  frequencies  with \nwhich  the  pure  tone  may  be  compared.\n\nAs  we  are  here  concerned  with  binaural  phenomena  we  shall  con- \nfine our  attention  to  the  relative  phases  and  intensities  at  the  two  ears. \nThe  question  at  once  arises:  does  the  observer  actually  hear  the  differ- \nent tones  separately,  and  if  so,  does  he  assign  a  location  to  each \nseparately?\n\nTo  what  extent  the  listener  locates  each  component  separately \ndepends  upon  the  ease  with  which  the  tones  can  be  distinguished. \nThe  experiments  which  bear  most  directly  upon  this  point  are  those \nin  which  the  component  tones  at  the  two  ears  are  arbitrarily  adjusted \nto  give  values  of  phase  difference  corresponding  to  different  locations. \nThis  is  done  under  conditions  where  the  location  of  each  component \nseparately  is  largely  determined  by  the  phase  difference.  More  ^ \nexperimented  with  two  tones,  transmitting  them  to  the  ears  through \ntubes  of  adjustable  lengths.  This  permitted  him  to  change  the  phase \ndifference  at  the  two  ears  while  keeping  the  intensities  substantially \nequal.  He  observed  the  apparent  location  for  various  settings  when \neach  tone  was  applied  by  itself  and  when  both  were  applied  together, \nusing  forks  of  256  and  320  cycles.  With  the  paths  equal  the  tones \ncombined  into  a  chord  located  in  the  median  plane  and  the  separate \ncomponents  could  not  be  heard.  With  a  setting  for  which  the  two \ncomponents  separately  appeared  on  opposite  sides  of  the  head,  one \ncomponent  was  heard  distinctly  by  the  right  ear  only  on  the  right \nside,  and  the  other  by  the  left  ear  only  on  the  left  side.  At  the  same \ntime  the  chord  was  heard  rather  indistinctly  near  the  median  plane \nbut  tending  slightly  toward  the  side  of  the  lower  tone.\n\nApparently  the  observer  does  not  consciously  separate  the  chord \ninto  its  components  unless  he  is  forced  to  do  so  by  some  inordinate \ndiscrepancy  between  the  positions  of  the  images  formed  from  them. \nThere  is  no  evidence  in  the  case  of  equal  paths  to  show  that  he  did \nor  did  not  subsconsciously  locate  the  separate  components  and  find \nthem  to  be  in  agreement.  In  view  of  the  second  experiment  it  seems \nprobable  that  he  did.  In  this  latter  experiment  he  obviously  found \nthat  the  two  components  corresponded  to  different  locations  and \nassigned  different  sources  to  each.  At  the  same  time  his  experience \ntold  him  that  tones  which  would  combine  to  form  a  musical  sound \ngenerally  have  a  common  source.  Hence  he  may  have  concluded \nsubconsciously  that  the  sound  waves  had  probably  been  distorted \nin  coming  from  a  common  source  and  so  he  corrected  his  observations \n*  Louis  T.  More:  Phil.  Mag.  XVIII,  1909,  p.  308.\n\non  both  tones  to  make  them  consistent  and  arrived  at  an  image  of \nthe  chord  between  the  other  two.\n\nSimilar  results  were  obtained  with  forks  of  256  and  384  cycles \nper  second,  except  that  in  general  the  lower  tone  was  completely \nblotted  out.  The  higher  tone  was  usually  quite  distinct  and  defin- \nitely located.  The  image  of  the  chord  was  nearer  to  the  image  formed \nwhen  the  higher  component  was  sounded  by  itself  than  to  the  image \nformed  from  the  lower  one  alone.  With  settings  for  which  the  direc- \ntions of  the  tones  separately  were  the  same,  whether  right,  left,  or \nmiddle,  the  upper  tone  disappeared  leaving  only  the  chord.  In \nexperiments  with  forks  of  256  and  512  cycles  it  was  difificult  to  dis-  f \ntinguish  the  separate  notes.  With  settings  for  which  the  two  separ- \nately were  on  opposite  sides  the  combination  was  on  the  side  of  the \nlower  fork.  This  can  be  interpreted  as  meaning  that  the  octave \nrelationship  is  inherently  difficult  to  resolve,  or  else  that  tones  an \noctave  apart  so  generally  come  from  a  common  source  that  the  ob- \nserver was  unwilling  to  make  any  other  assumption.\n\nAlthough  the  explanation  of  these  results  is  not  yet  thoroughly \nunderstood,  they  show  very  definitely  that  in  locating  complex  sounds \nmade  up  of  pure  tones  the  observer  does  within  limits  locate  the \ncomponents  separately.  If  they  agree,  a  single  image  is  formed;  if \nthey  do  not,  he  may  either  locate  the  tones  separately  or  form  a  single \ncompromise  image  or  do  both.\n\nIt  is  in  this  way  that  the  theory  developed  for  pure  tones  is  ap-  | \nplied  to  complex  sounds  made  up  of  pure  tones.  The  next  step  is  to \nextend  it  so  as  to  include  complex  sounds  in  general.  To  do  this  we \nmust  picture  the  observer  as  resolving  each  sound  into  sinusoidal \ncomponents  locating  the  components  separately  and  forming  one  or \nmore  images  based  on  a  combination  of  the  apparent  sources  as  in- \ndicated by  the  separate  components.  While  it  is  fairly  easy  to  effect \nsuch  a  resolution  mathematically  it  is  somewhat  less  easy  to  interpret \nthe  result  in  a  manner  satisfactory  to  our  intuitive  conceptions  of  the \nphenomena  involved;  also,  granted  the  theoretical  possibility  of  the \nresolution,  there  remains  the  question  of  what  physical  or  psycho- \nlogical limitations  there  may  be  to  its  application.\n\nIn  view  of  the  fact  that  a  really  pure  component  tone  has  no  begin- \nning or  end,  and  no  fluctuations  in  its  amplitude,  it  is  not  at  once \napparent  how  a  single  discrete  sound  such  as  the  bark  of  a  dog  can  be \nresolved  into  components  of  that  nature.  However,  if  enough  com- \nponents are  available  it  has  been  established  beyond  question  that \nby  properly  choosing   their   frequencies,   amplitudes,   and   phases,   a\n\ncombination  may  be  anixcd  at  in  which  the  algebraic  sum  of  all  the \ncomponents  is  zero  for  all  instants  before  and  after  the  period  occupied \nby  the  sound  and  equal  to  the  instantaneous  value  of  the  sound \nwave  for  instants  within  that  period.  This  combination  is  known  to \nmathematicians  as  the  Fourier  Integral  corresponding  to  the  wave, \nand  the  formula  for  the  phase  and  amplitude  of  each  component \nsinusoid  is  known.  It  is  an  extension  of  the  well  known  Fourier \nseries  expansion  used  for  resolving  sustained  periodic  disturbances.\n\nThe  physical  interpretation  of  this  integral  may  be  facilitated  by \nreviewing  the  steps  in  its  evolution  from  the  Fourier  series.  It  is \nwell  known  that  if  the  sound  in  question  were  repeated  at  regular \nintervals  the  resulting  periodic  wave  could  be  resolved  by  Fourier \nanalysis  into  a  series  of  sinusoidal  components,  the  frequencies  of  all \nof  which  are  integral  multiples  of  the  frequency  of  repetition  of  the \nsound.  Successive  components  therefore  differ  in  frequency  by  an \namount  equal  to  this  frequency  of  repetition.  Now  it  is  not  essential \nthat  the  repetitions  of  the  sound  follow  each  other  immediately. \nInstead,  they  may  be  separated  by  intervals  of  silence.  The  effect \nof  such  silent  intervals  is  to  reduce  the  frequency  of  repetition  and \ntherefore  also  the  fundamental  frequency.  As  a  result  the  com- \nponent frequencies  are  brought  closer  together  and  the  number  within \nany  particular  frequency  range  is  increased.\n\nSuppose  now  that  the  interval  between  repetitions  is  indefinitely \nincreased.  As  this  is  done  the  efTect  of  any  one  occurrence  of  the \nsound  becomes  more  and  more  independent  of  the  others,  and  in  the \nlimit  when  the  sounds  next  preceding  and  next  following  the  one \nunder  consideration  are  infinitely  far  removed,  we  have  the  case  of  a \ndiscrete  sound.  As  this  limiting  case  is  approached  the  fundamental \nfrequency  becomes  smaller  and  smaller  and  the  component  frequen- \ncies, which  are  multiples  of  it,  are  separated  by  infinitesimal  frequency \ndifferences.  While  the  amplitude  of  each  component  also  decreases, \nthe  number  of  components  increases  at  such  a  rate  that  the  aggregate \nenergy  of  all  the  components  within  a  given  frequency  range  remains \nfinite.  In  this  way,  the  distribution  of  the  sound  energy  over  various \nfrequencies \u2014 that  is,  the  \"energy  spectrum\" \u2014 can  be  obtained.\n\nIt  is  evident,  then,  that  when  an  aperiodic  complex  sound  is  resolved \nmathematically  there  results  an  infinity  of  component  tones,  each \nhaving  a  characteristic  intensity  and  phase.  If  an  observer  were \ncapable  of  an  equally  complete  resolution  he  would  have  at  his  dis- \nposal an  infinity  of  sets  of  data  from  which  an  infinity  of  images \ncould  be  formed.  In  the  absence  of  distortion  these  should  all  co- \nincide.\n\nPractically,  of  course,  no  such  refinement  of  resolution  is  possible. \nThe  ability  to  distinguish  differences  in  pitch  varies  from  person  to \nperson,  but  the  minimum  intervals  employed  in  musical  composition \nprobably  give  a  rough  measure  of  the  normal  resolving  power  of  the \near.  Even  with  this  limitation  the  broad  sound  spectrum,  such  as  an \nirregular  sound  produces,  is  capable  of  yielding  a  very  large  number  of \nseparable  components;  and  hence  a  large  number  of  individual  im- \nages. It  is  this  fact \u2014 that  with  a  very  complex  sound  the  number  of \nindependent  determinations  of  the  image  is  limited  only  by  the  re- \nsolving power  of  the  observer \u2014 which  makes  his  accuracy  of  binaural \nlocation  as  well  as  his  sense  of  certainty  much  greater  for  such  sounds \nthan  for  pure  tones.\n\nSo  long  as  the  images  of  all  the  components  coincide,  it  is  of  little \nimportance  how  fine  the  resolution  is,  for  further  refinement  only \nserves  to  increase  the  sense  of  certainty  by  adding  to  the  volume  of \naccordant  evidence.  However,  when  the  images  are  not  in  agree- \nment the  problem  is  more  complicated  and  the  degree  of  resolution \nbecomes  important.  Here  also  purely  physical  considerations  cease \nto  be  adequate  and  psychological  factors  must  be  considered  similar \nto  those  involved  in  the  location  of  a  pure  tone  for  which  the  intensity \nratio  and  phase  difference  do  not  correspond  to  any  actual  source. \nWhen  an  observer  is  faced  with  discordant  results  he  must  make \nsome  subconscious  judgment.  For  small  discrepancies  such  as  occur \nin  every-day  experience,  he  probably  assumes  those  images  which \ndepart  most  from  the  rest  to  be  misplaced  because  of  distortion  dur- \ning transmission  and  so  either  corrects  or  ignores  them.  If  the \ndiscrepancies  are  large  he  may  find  it  difficult  on  the  ground  of  ex- \nperience to  believe  that  so  much  distortion  could  occur.  In  such  an \nevent  he  will  most  likely  form  several  images  from  different  com- \nponents or  in  extreme  cases  lose  the  sense  of  location  altogether.\n\nBowlker  found  separate  images  to  occur  experimentally  both  for \nband  music,  which  approaches  a  collection  of  tones  and  for  the \nirregular  barking  of  dogs.  He  placed  tubes  of  unequal  length  to  his \ntwo  ears  thereby  upsetting  the  normal  diffraction  around  the  head \nand  interposing  a  longer  path  on  one  side  than  on  the  other.  Obvi- \nously, the  distortion  produced  in  this  manner  is  of  a  type  not  likely \nto  be  met  in  every-day  life  and  affects  different  frequencies  in  widely \ndifferent  fashions.  He  reports  that  when  listening  to  \"a  band  of \nthree  or  four  instruments  played  in  the  open \u2014 the  notes  will  be  found \nto  be  scattered  over  a  wide  range,  most  being  to  the  side  of  the  short \ntube,  some  being  in  front  and  some  being  to  the  side  of  the  long  tube. \nIn  listening  with  such  a  pair  of  tubes  to  two  dogs  furiously  barking\n\nthe  effect  is  at  first  quite  alarming \u2014 one  seems  to  be  in  the  middle \nof  a  pack  of  dogs  some  of  which  are  rushing  viciously  at  one's  throat.\" \nAn  illustration  of  failure  to  form  any  image  is  found  in  a  phenomenon \nobserved  in  the  use  of  binaural  compensators  for  determining  the \ndirection  of  submarine  sounds.  The  sound  is  picked  up  by  two  sub- \nmarine telephone  transmitters  and  led  to  the  ears  through  inde- \npendent paths.  By  adjusting  the  lengths  of  the  paths  the  image \ncan  be  shifted  from  side  to  side  and  for  practical  purposes  the  setting \nof  the  instrument  is  made  by  bringing  the  image  exactly  to  the  middle. \nA  fairly  definite  sound  image  is  formed,  but  observers  report  that  part \nof  the  sound  does  not  merge  into  this  sound  image  and  move  in  re- \nsponse to  the  adjustment,  but  instead  appears  as  a  diffuse  back- \nground of  noise.\u00ae  This  may  be  explained  on  the  assumption  that, \nwhile  the  images  formed  from  most  of  the  sound  components  agree \nsufificiently  well  that  the  observer  corrects  them  to  a  single  position, \ncertain  components  are  so  distorted  by  resonance  effects  inherent \nin  the  apparatus  that  their  images  are  scattered  more  or  less  at  ran- \ndom. The  lack  of  agreement  among  any  considerable  number  of \nthese  prevents  the  formation  of  a  second  image  and  causes  the  sense  of \ndiffusedness.\n\nAs  the  distortion  becomes  still  more  extreme  we  should  expect \nthe  experimental  results  to  depend  more  and  more  upon  the  observ- \ner's power  of  resolution,  for  as  the  distortion  is  progressively  in- \ncreased a  condition  must  finally  be  reached  where  the  positions  of  the \nimages  are  appreciably  different  for  two  components  whose  frequencies \nare  so  nearly  alike  as  to  make  their  recognition  as  separate  tones \ndillficult  if  not  impossible.  This  condition  actually  occurred  in  an \nexperiment  of  Baley's  with  a  sound  consisting  of  a  mixture  of  sus- \ntained tones.  Its  effect  on  the  listener  is  interesting  from  the  stand- \npoint of  subconscious  readjustment  of  discordant  data.\n\nBaley's  ^  experiment  consisted  in  applying  a  number  of  sustained \ntones  to  one  ear  of  a  musically  trained  observer  and  a  number  of  differ- \nent tones  to  the  other  ear,  and  testing  his  ability  to  assign  them  to \ntheir  proper  sides.  So  long  as  the  intervals  between  the  tones  were \nfairly  large,  the  observer  never  failed  to  locate  them  correctly.  Con- \nsidering the  entire  stimulus  as  a  complex  sound  we  may  think  of  the \nobserver  as  locating  the  tones  individually  and  finding  them  to  fall \ndefinitely  into  two  groups  whose  images  are  located  one  at  each  ear.\n\n*  This  interesting  phenomenon  was  called  to  our  attention  by  Mr.  Richard  D. \nFay  of  the  Submarine  Signalling  Corporation  who  tells  us  that  it  has  been  noted \nby  a  large  number  of  observers.\n\nHowever,  when  he  used  six  tones  which  were  separated  from  each \nother  by  a  single  tone  interval,  the  separate  components  could  not \nbe  distinguished  and  a  painful  sensation  was  produced.  The  ob- \nserver was  apparently  faced  with  the  situation  that  to  make  the  ob- \nserved intensity  ratios  and  phase  differences  correspond  to  a  single \nsource  would  involve  extremely  large  corrections  in  the  observed \ndata.  On  the  other  hand,  his  power  of  tone  resolution  was  insufficient \nto  separate  the  components  and  assign  them  to  different  sources.  It \nis  not  surprising,  then,  that  the  difficulty  manifested  itself  by  painful \nsensations.  While  this  illustration  is  taken  from  an  extreme  condi- \ntion of  laboratory  experiment  and  may  appear  to  have  little  bearing \non  the  every-day  location  of  sounds,  it  is  really  significant  because  of \nthe  manner  in  which  it  illustrates  the  importance  of  psychological \nfactors  in  all  cases  in  which  the  sound  waves  are  distorted.\n\nIn  the  foregoing  discussion  an  attempt  has  been  made  to  bring  out \nthe  main  features  involved  in  extending  the  theory  of  the  binaural \nlocation  of  pure  tones  to  cover,  qualitatively  at  least,  the  location \nof  complex  sounds.  It  has  virtually  been  assumed  that  the  latter \ninvolves  three  processes:  first,  the  resolution  of  the  sound  into  its \ncomponent  tones;  second,  the  independent  (generally  subsconscious) \nlocation  of  each  separate  component;  and  third,  the  formation  of  a \nconscious  judgment  of  the  position  of  the  source  based  on  the  locations \nof  the  individual  images.  The  greatly  increased  amount  of  data \navailable  when  the  sound  is  complex  has  quite  different  effects  on  the \nfinal  result  according  as  the  different  images  do  or  do  not  coincide. \nIf  they  do,  the  accuracy  of  location  and  the  sense  of  certainty  are \nincreased.  If  they  do  not,  confusion  arises,  subconscious  corrections \nare  called  for,  and  the  final  result  is  likely  to  depend  very  consider- \nably on  the  psychological  processes  and  individual  prejudices  of  the \nparticular  observer.\n\nSynopsis:  The  art  of  electrical  communication  owes  a  great  and  incrcas- \ninj;!y  recognized  debt  to  Oliver  Heaviside  for  his  work  in  developing  and \nemphasizing  a  correct  theory  of  electrical  transmission  along  wires  and  in \nparticular  for  his  insistance  on  the  importance  of  inrluctance.  His  oper- \national methods  of  solving  the  differential  etjuations  which  are  fundamental \nof  the  theory  of  electric  circuits,  although  not  widely  known,  are  important. \nThese  methods  are  peculiarly  applicable  to  many  important  problems  of \nelectrical  transmission.  The  present  paper,  while  theoretical  in  char- \nacter, therefore  deals  with  a  subject  of  practical  importance  to  the  com- \nmunication engineer.\n\ndt \nbraic  equations  by  replacing  the  differential  operator  by  the  symbol  p  and \nby  this  expedient  a  purely  symbolic  solution  is  obtained.     This  syrnbolic \nsolution  is  called  the  operational  formula  of  the  problem.\n\nIn  order  to  interpret  the  purely  symbolic  operational  formula,  Heaviside \nproceeded  as  fellows:  By  direct  comparison  of  the  operational  fornmla  of \nspecific  problems  with  their  known  explicit  solutions  he  was  led  to  assign \na  definite  significance  to  the  operator  p.  Thereupon,  he  obtained  by  in- \nduction generalized  specific  criteria  or  rules  for  solving  the  operational \nformula.\n\nThe  present  paper,  by  attacking  the  problem  from  a  different  standpoint, \nshows  that  the  Heaviside  operational  formula  is  a  shorthand  equivalent  of \nan  integral  equation  from  which  the  methods  and  rules  of  his  operational \ncalculus  are  deducible. \u2014 Editor.\n\nAVERY  interesting  and  by  no  means  the  least  valuable  part  of \nHeaviside's  researches  relates  to  operational  methods  of  solving \nthe  differential  equations  of  a  class  of  physical  problems  of  which \nelectric  circuit  theory  problems  are  typical;  in  fact  Volume  II  of  his \nElectromagnetic  Theory  is  almost  entirely  devoted  to  this  subject. \nThe  methods  of  solution  which  he  originated  and  employed  are  of \nextraordinary  directness  and  simplicity  in  a  very  large  class  of  prob- \nlems in  applied  mathematics.  In  fact  it  would  be  difhcult  to  exagger- \nate the  value  of  his  work  along  this  line,  and  nowhere  is  it  more  im- \nmediately and  usefully  applicable  than  in  the  theoretical  problems \nof  electro-technics.\n\nHeaviside  is,  however,  by  no  means  easy  reading  and,  in  spite  of \nthe  considerable  number  of  published  studies  relating  to  his  opera- \ntional calculus,  it  is  less  generally  understood  and  applied  than  its \nvalue  warrants.  The  writer  has  had  occasion  to  apply  Heaviside's \nmethods  quite  extensively  in  electrical  problems  and  in  the  course \nof  his  study  was  led  to  a  general  formula  which  to  him  at  least,  has \nproved  useful  in  interpreting  and  rationalizing  the  operational  cal-\n\ncuius.  This  formula  is  presented  and  discussed  in  the  present  paper \nwith  the  hope  that  it  may  be  of  service  to  students  of  Heaviside  in \nunderstanding  and  applying  his  methods.  The  paper  may  also  serve \nas  a  brief  recapitulation  of  some  of  the  outstanding  methods  ap- \nplicable to  the  solution  of  problems  in  electric  circuit  theory.\n\nThe  type  of  problem  to  which  the  operational  calculus  is  applicable \nand  Heaviside's  method  of  solution  may  be  illustrated  in  a  sufficiently \ngeneral  manner  for  didactic  purposes  in  connection  with  the  solution \nof  the  system  of  equations\n\nWe  are  concerned  with  the  determination  of  the  variables  Xi  .  .  x\u201e \nas  functions  of  the  independent  variable  t  for  the  following  boundary \nconditions;  the  known  functions  F]  .  .  F\u201e  and  the  variables  Xi  .  .  :v:\u201e \nare  identically  zero  for  values  of  the  independent  variable  t^O.  In  other \nwords,  the  system  is  initially  in  a  state  of  equilibrium  when  the \n\"  forces  \"  Fi  .  .  .  F\u201e  are  applied.  These  boundary  conditions  are \nextremely  important  in  physical  problems.\n\nOwing  to  the  linear  character  of  the  equations  we  may  without  loss \nof  generality  set  all  the  F  functions  equal  to  zero  except  one,  say \nFi{t),  and  write\n\nThe  function  on  the  right  hand  side,  written,  in  accordance  with \nthe  Heaviside  notation,  as  unity  is  identically  zero  for  /<0  and \nunity  for  />0  and  hi  .  .  .  h\u201e  are  identically  zero  for  /<0.\n\nH^quations  (4)  formulate  the  problem  actually  dealt  with  by  Heavi- \nside who  did  not  explicitly  consider  the  more  general  equations  (3). \nHis  method  of  attack  was  as  follows;  Writing  p\"  for  the  differential \noperator  d\"/dt\"  equations  (4)  become  formally  algebraic  and  yield  a \npurely  symbloic  solution\n\nEquation  (6)  is  the  Heaviside  operational  formula;  as  it  stands, \nhowever,  it  is  purely  symbolic  and  the  problem  remains  to  find  the \nsignificance  of  the  equation  and  to  deduce  therefrom  the  value  of \nh  =  h(t)  as  a.  function  of  /.\n\nHeaviside's  method  from  this  point  on  was  one  of  pure  induction. \nFrom  the  known  solution  of  specific  problems  he  inferred  general \nrules  for  expanding  and  interpreting  the  operational  formula:  the \nbody  of  rules  thus  developed  for  solving  the  operational  equation \nmay  be  appropriately  termed  the  Heaviside  Operational  Calculus.\n\nThe  contribution  of  the  present  paper  to  the  theory  of  the  Heavi- \nside operational  calculus  depends  on  the  following  proposition  and \nits  immediate  corollary.^\n\nThe  integral  equation  is  an  identity  for  all  positive  real  values  of  p  and \nconsequently  determines  hj(t)  uniquely.\n\n'  This  formula  has  been  established  in  previous  papers.     It  is  briefly  discussed \nin  .Appendix  I.\n\nIt  follows  as  an  immediate  corollary  that  the  Heaviside  operational \nequation\n\nThe  significance  of  the  operational  equation  and  the  rules  of  the  Heaviside \noperational  calculus  are  therejore  deducible  from  the  latter  equation. \nThe  whole  problem  is  thus  reduced  to  the  purely  mathematical  problem \nof  solving  the  integral  equation.\n\nIt  should  be  remarked  in  passing  that,  while  the  Heaviside  opera- \ntional calculus  has  been  elucidated  in  connection  with  the  solution \nof  a  set  of  differential  equations  involving  a  finite  number  of  variables, \nit  is  not  so  limited  in  its  applications.  It  is  applicable  also  when  the \nnumber  of  variables  is  infinite  and  to  such  partial  differential  equ- \ntions  as  the  telegraph  equation.  The  foregoing  theorem  applies  also \nto  all  such  physical  problems  where  an  operational  formula  A  =  \\/H{p) \nis  derivable.\n\nBefore  discussing  the  solution  of  the  integral  equation  (9)  and  de- \nducing therefrom  some  of  the  rules  of  the  operational  calculus,  a \nsimple  but  interesting  and  instructive  example  of  the  way  the  oper- \ntional  formula  is  set  up  will  be  given.\n\nConsider  a  transmission  line  of  infinite  length  along  the  positive \nX  axis  and  let  it  have  a  distributed  inductance  L  and  capacity  C  per \nunit  length.  Let  a  unit  voltage  be  applied  to  the  line  at  the  origin \nre  =  0  at  time  /  =  0;  required  the  line  current  /  and  voltage  V  at \nany  point  x  at  any  subsequent  time  /.\n\nNow  by  the  conditions  of  the  problem  Vq  is  zero  before,  unity  after \ntime  /\u25a0  =  0;  hence  the  foregoing  equations  are  operational  formulas \nand  by  (9)\n\nwhich  are,  of  course,  the  well  known  solutions  of  the  problem.  The \ndirectness  and  simplicity  of  the  solution  from  the  definite  integrals  is, \nhowever,  noteworthy.\n\nBy  virtue  of  the  foregoing  analysis  the  Heaviside  operational  cal- \nculus becomes  identical  with  the  methods  and  rules  for  the  solution \nof  integral  equations  of  the  type\n\nAn  integral  equation  is,  of  course,  one  in  which  the  unknown  func- \ntion appears  under  the  sign  of  integration;  the  process  of  determining \nthe  unknown  function  is  the  solution  of  the  equation.  Integral \nequations  of  the  form  of  (9)  were  first  employed  by  Laplace  and  may \nbe  referred  to  as  equations  of  the  Laplace  type.  More  recently  they \nhave  become  of  importance  in  the  modern  theories  of  divergent  series \nand  summability.  The  solution  of  a  large  number  of  integral  equa- \ntions of  the  Laplace  type  has  been  worked  out;  however  the  procedure \nis  usually  peculiar  to  the  particular  problem  in  hand.  In  this  con- \nnection it  is  noteworthy  that,  from  a  purely  mathematical  stand- \npoint, Heaviside's  operational  calculus  is  a  valuable  contribution  to \nthe  systematic  solution  of  this  type  of  integral  equations.  That  is  to \nsay,  methods  which  he  developed  for  the  solution  of  his  operational \nequation  suggest  systematic  procedure  in  the  solution  of  the  integral \nequation  (9),  as  might  be  expected  from  the  relationship  pointed  out \nin  the  present  paper.\n\nAs  stated  above  a  large  number  of  infinite  integrals  of  the  type \nappearing  in  equation  (9)  have  been  worked  out.  Consequently  the \nsolution  of  (9)  can  frequently  be  written  down  by  inspection.  When \nthis  is  not  the  case,  however,  the  appropriate  procedure  is  usually \nto  expand  the  function  \\/pH{p)  in  such  a  form  that  the  individual \nterms  are  recognizable  as  identical  with  infinite  integrals  of  the  re- \nquired type.\n\nAn  interesting  expansion  of  this  kind  and  one  which  is  applicable \nto  a  large  number  of  physical  problems  is  as  follows:\n\nThis  expansion  is  purely  formal  and  the  series  is  divergent.  It  is \nsummable,  however,  in  the  sense  that  it  may  be  identified  with  its \ngenerating  function  l/pH(p).  It  is  also  summable  in  accordance \nwith  Borel's  definition  of  the  sum  of  a  divergent  series  by  the  Borel \nintegral  *\n\nJ     dt  e-^^^y\\  ant^/n\\ \nThis  suggests  that  these  two  series  are  equal  and  consequently  that\n\nprovided  this  series,  which  is  called  by  Borel  the  associated  function  of \nthe  divergent  expansion,  is  itself  convergent.  This  is  the  case  in  all \nphysical  problems  to  which  this  form  of  expansion  has  been  applied.^\n\nThe  foregoing  will  be  recognized  as  identical  with  Heaviside's \npower  series  solution,  obtained  by  the  empirical  rule  of  identifying \n\\/p^  with  t\"/n\\  in  the  asymptotic  expansion  of  l/H{p).\n\nAnother  form  of  solution  of  very  considerable  practical  value \ndepends  on  a  partial  fraction  expression  which  can  be  carried  out  in  a \nlarge  number  of  physical  problems.     It  is\n\nwhich    will    be   recognized    as    the   celebrated    Heaviside    Expansion \nSolution.\n\nAs  illustrating  the  flexibility  of  the  integral  identity  (9),  another \nform  of  solution  will  be  given  which  is  often  of  value  in  practical \nproblems  where  an  explicit  solution  cannot  be  obtained.  Suppose \nthat   l/pH(p)   can   be  written  as\n\n'The  terms  a  +  c/p^  in  this  expansion  were  suggested   by  Dr.  O.  J.  Zobel  and \nmust  be  included  in  a  number  of  important  problems  in  electric  circuit  theory.\n\nAs  a  final  example  of  the  foregoing  discussion  we  shall  consider  a \nspecific  problem  of  some  practical  interest  in  itself  and  which  involves \nHeaviside's  so-called  \"  fractional  differentiation  \"  and  his  resulting \nasymptotic  solutions.  The  physical  problem  is  as  follows:  a  \"  unit- \nvoltage  \"  (zero  before,  unity  after  time  /  =  0)  is  applied  through  a \nterminal  condenser  Co  to  an  infinitely  long  cable  of  resistance  i?  and \ncapacity  C  per  unit  length.  Required  the  Voltage  V  at  the  cable \nterminals.\n\nTaking  the   last   form   of   l/pH{p),   expanding    asymptotically    and \nrecognizing  that\n\nup.  VP  -fe-'-  (2\u201e  -  1)  [zf-  3) .  .  .  1  7=i \nthe  resulting  series  solution  can  be  recognized  and  summed  as\n\n\u2022This  formula  is  quite  useful;  it  is  applied  in  the  solution  of  the  last  example  of \nthie  present  paper.\n\nthe  series  is  truly  asymptotic  in  the  sense  that  the  error  is  less  than \nthe  last  term  included.\n\nAnother  mode  of  procedure,  however,  suggests  itself,  which,  by \nthe  aid  of  equation  (10)  gives  the  solution  directly  without  series \nexpansion.     We  have\n\nThe  solution  h{t)  =  hi{i)  +  hiit)  agrees  with  the  preceding  derived \nfrom  the  asymptotic  expansion,  and  is  considerably  more  direct  and \nsimple.\n\nIt  is  interesting  to  compare  this  solution  with  Heaviside's  own \noperational  solution  (Electromagnetic  Theory  Vol.  II,  p.  40)  w^hich \namounts  to  the  following.     The  operational  formula  is  written\n\nwhich  agrees  with   the  foregoing  and   is  the  actual   asymptotic  ex- \npansion.^\n\nThe  foregoing  discussion  is  sufficient,  it  is  hoped,  to  show  the  place \nof  the  integral  formula  (9)  in  relation  to  the  Heaviside  operational \ncalculus.  It  is  believed  to  be  particularly  applicable  in  connection \nwith  a  number  of  questions  relating  to  divergent  series  and  solutions \nwhich  Heaviside's  work  has  raised  and  which  have  received  too  little \nattention  from  mathematicians.\n\nThis  equation  may  be  regarded  as  well  established  and  can  in  fact \nbe  deduced  in  a  quite  general  manner  by  synthetic  arguments.  It  is \nderived  and  employed  in  papers  by  the  writer  (Trans.  A.  I.  E.  E., \n1911,  pp.  345-427,  and  Phys.  Rev.  Feb.  1921,  pp.  116-134)  and  is \ndeducible  at  once  from  the  work  of  Fry  (Phys.  Rev.  Aug.  1919,  pp. \n115-136).\n\nOn  the  basis  of  equation  (5)  the  deduction  of  formula  (9),  in  which, \nhowever,  no  pretense  to  rigor  is  made,  proceeds  as  follows;\n\nIf  the  function  F{t)  in  equations  (3)  is  set  equal  to  e^',  the  complete \nsolution  (5)  includes  the  particular  solution  ^\n\nwhich  irfvolves  /  only  through  the  exponential  term.     The  complete \nsolution  must,  therefore,  admit  of  reduction  to  the  form\n\n'  The  procedure  by  which  Heaviside  arrived  at  the  foregoing  asymptotic  solution \nis  not,  however,  always  so  fortunate  For  example  if  a  terminal  inductance  is  sub- \nstituted for  the  terminal  condenser  of  the  preceding  problem,  precisely  the  same \nprocedure  gives  an  incomplete  result.  Heaviside  recognized  this  and  added  an \nextra  term  without  explanation  (Elm.  Th.  Vol.  H,  p.  42)  but  his  solution  appears \nto  be  doubtful  in  the  light  of  some  recent  work  by  the  writer  in  applying  the  formula \nof  the  present  paper  to  the  same  problem.\n\n*  Provided  H  {p)  '^  o.  This  restriction  is  of  no  consequence  in  physical  prob- \nlems, where  the  roots  of  H  (p)  are  in  general  complex  with  real  part  negative.\n\nNow  the  first  term  of  the  expression  involves  /  only  through  the  ex- \nponential term  while  the  second  term  involves  t  through  the  lower \nlimit  of  the  integral  which  ultimately  vanishes  and  therefore  includes \nno  term  involving  /  only  through  the  exponential.  Consequently \nthe  first  term  of  {b)  is  identifiable  as  the  particular  solution  of  (a)  and \nby  direct  equation  it  follows  that\n\nThe  most  important  restriction  which  is  implicit  in  the  foregoing \nis  that  in  splitting  up  the  definite  integral  of  (5)  we  have  tacitly  as- \nsumed that  hit)  is  finite  for  all  values  of  /;  a  restriction  which  is \nnecessary  in  order  that  the  infinite  integral  shall  be  convergent  for \nall  positive  real  values  of  p.  This  condition  is  satisfied  in  all  physical \nproblems  and  therefore  introduces  no  practical  limitation  of  im- \nportance.\n\nHowever,  even  w^hen  this  restriction  does  not  hold  formula  (9)  may \nbe  valid  and  uniquely  determine  h{t)  if  p  is  restricted  to  values  which \nmake  the  infinite  integral  convergent,  or  when  the  problem  is  such \nthat  e~^'h{t)   is  an  exact  derivative.     As  an  example,  suppose   that\n\nwhere  a  is  a  real  positive  quantity.     It  may  be  otherwise  shown  that \nk{t)  =  e\u00b0'  and  formula  (9)  becomes\n\nThe  discussion  in  the  text  does  not  pretend  to  be  a  proof  of  the \npower  series  expansion  in  any  strict  sense.  A  more  satisfactory \n<iiscussion  proceeds  as  follows:\n\nWe  shall  here  introduce  a  necessary  restriction  on  the  function  1/H{p). \nIt  must  include  no  function  which  is  represented  asymptotically  by \na  series  all  of  whose  terms  are  zero;  that  is  a  function  <p{p)  such  that \nthe  limit,  as  p  approaches  oo,  of  p\"(l>ip)  is  zero  for  every  value  of  n. \nThe  function  e~^  is  such  a  function.  (See  Whittaker  &  Watson, \np.  154.)\n\nWith  this  restriction  understood,  start  with  the   integral   (9)  and \nintegrate  by  parts;  we  get\n\nNow  let  p  again  approach  infinity;  in  the  limit  the  integral  vanishes, \nand  the  right  hand  side,  by  virtue  of  the  asymptotic  expansion, \napproaches  the  limit  Ui,  whence  h''^^  (o)  =  ai.  Proceeding  in  this \nmanner,  repeated  integrations  by  parts  establish  the  relation \nh^\"^  (o)  \u2014  a\u201e.     But  provided  the  series  is  absolutely  convergent,  then\n\nThe  power  series  solution  is  applicable  to  a  large  class  of  physical \nproblems  and  has  been  rigorously  established  under  certain  restric- \ntions by  other  methods  than  that  employed  above  (see  papers  by \nBromwich,  Phil.  Mag,  May  1920,  p.  407;  Fry,  Phys.  Rev.  Aug. \n1919,  p.  115;  and  the  writer,  Trans.  A.  I.  E.  E.  1919,  p.  345).\n\nOn   the  basis  of   tlie   preceding  and   with   the  aid   of  fornuihi   (10), \nexpansions  of  the  type\n\n!//)//(/))  ~ -i-    ^bn/P\"  +  '  =  f^     e~''h{l)dt \nwhich  occur  in  physical  problems,  can  be  dealt  with.     For  since\n\nThe  Physical  Characteristics  of  Audition  and \nDynamical  Analysis  of  the  External  Ear\n\nSynopsis:  This  paper  discusses  some  of  the  characteristics  of  the \near  which  have  become  important  in  the  design  and  development  of  tele- \nphone apparatus  and  circuits.  The  field  of  audition,  bounded  by  the \ncurves  of  minimum  and  maximum  loudness  as  functions  of  frequency, \nhas  been  determined  for  a  large  number  of  ears,  and  the  smaller  included \narea  most  used  in  speech  has  been  mapped.  The  nature  of  these  fields  in \ncertain  cases  of  abnormal  hearing  has  also  been  determined  and  the  condi- \ntions which  must  be  observed  in  designing  apparatus  to  satisfactorily \nrelieve  deafness  are  discussed.\n\nThe  sensitivity  of  the  ear  is  given  in  terms  of  the  r.  m.  s.  pressure \nmeasured  by  a  calibrated  condenser  transmitter.  It  is  printed  out  in  the \nappendix  that  this  pressure  is  not  necessarily  equal  to  that  which,  when \napplied  to  the  ear  drum,  would  just  give  rise  to  the  sensation  of  sound. \nHowever,  it  is  the  nearest  approach  to  the  value  of  this  pressure  which  can \nbe  determined  at  present,  and  as  the  dynamical  properties  of  the  ear \nbecome  more  fully  known  it  is  pointed  out  how  the  relation  between  the \ntwo  pressures  can  be  more  accurately  stated. \u2014 Editor.\n\n1.  Introduction.  It  has  become  important  in  the  design  and \ndevelopment  of  telephone  apparatus  and  circuits  to  know  quantita- \ntively the  various  functional  characteristics  of  the  ear  since  the  ear  is \nan  important  dynamical  unit  in  the  long  series  of  vibration  transmitting \napparatus  constituting  a  telephone  system.  A  complete  analysis  of \nthis  problem  involves  not  only  the  properties  of  the  physical  circuit, \nbut  also  the  characteristics  of  the  ear  and  voice  and  of  the  air  passages \nbetween  the  mouth  and  transmitter  and  between  the  ear  and  receiver. \nIt  is  the  purpose  of  the  present  paper  to  discuss  some  of  the  character- \nistics of  the  ear  and  its  outer  air  passages.\n\nMuch  has  been  learned  about  the  normal  ear  by  the  investigation \nof  the  characteristics  of  abnormal  ears.  This  has  incidentally  had \nan  application  to  otological  diagnosis  and  the  design  and  building  of \namplifying  apparatus  for  the  deaf.\n\nThis  paper  is  a  summary  of  the  conclusions  reached  to  date  regard- \ning the  absolute  sensitivity  of  normal  and  abnormal  ears,  the  maximum \nsound  to  which  the  ear  can  accommodate  itself,  the  much  discussed \npoints  of  \"upper  and  lower  frequency  limits  of  audition,\"  the  \"quality\" \nof  audition,  a  brief  mention  of  the  binaural  sense  and  the  principles  of \nrigorous  dynamical  analysis  of  the  ear  as  a  mechanism.  A  brief \ndescription  of  the  apparatus  used  is  also  given.\n\nThe  function  of  the  auditory  sense  is  to  detect  sounds  of  various \nkinds  and  wave  shapes  varying  over  a  range  of  pressure  on  the  ear \ndrum  of  from  about  .001  to  1,000  dynes  per  cm^  and  over  a  consider- \nable part  of  this  range  to  differentiate  with  certainty  between  complex\n\nsounds  so  ncarl>-  alike  that  no  existing  physical  apparatus  can  separate \nthem.  The  binaural  feature  adds  a  sense  of  orientation  with  respect \nto  a  source  and  uniform  sensitivity  for  sounds  approaching  from  differ- \nent directions.  The  abnormal  auditory  sense  may  be  regarded  as \nlacking  niore  or  less  in  (a)  range  of  sensation  (frequency  and  intensity) ; \n(b)  quality  of  sensation  in  various  regions  of  the  range;  (c)  the  bi- \nnaural sense.  Apparatus  and  methods  have  been  developed  by  means \nof  which  the  outstanding  features  of  these  functions  can  be  measured \nand  to  a  limited  extent  compensated  for.\n\n2.  Minimum  Audibility.     Fig.  1  shows  a  plot  of  the  logarithmic \naverage  of  minimum  audible  pressure  on  72  normal  ears  taken  through-\n\nout  a  range  of  frequency  from  60  to  4,0C0  cycles.^  Both  the  intensity \nand  frequency  scales  are  logarithmic.  Although  all  skew  errors  in  the \ndetermination  of  the  average  curve  have  not  been  eliminated,  an \ninvestigation  has  shown  that  they  are  so  small  as  not  to  affect  the \nutility  of  the  curve  for  the  purpose  of  measuring  deafness.  Among \nthe  errors  which  obviously  tend  to  raise  this  curve  might  be  men- \ntioned, noise  in  the  observing  room,  abnormality  of  hearing,  lack  of \nattention,  and  low  mentality  of  the  observer.  Care  was  taken  to \nreduce  these  errors  to  a  minimum  without  actually  making  separate\n\n'This  curve  has  already  been  pubhshed;  The  Frequency  Sensitivity  of  Normal \nEars,  by  H.  Fletcher  and  R.  L.  Wegel,  Proceedings  pf  the  National  Academy  of  Sciences, \nJanuary,  1922,  and  Physical  Review,  June,  1922.\n\nquantitative  measurements  of  each  of  them  on  a  rigorous  statistical \nbasis.\n\nThe  statistical  deviation  from  the  mean  varies  irregularly  with  fre- \nquency; very  likely  this  is  due  mostly  to  the  external  anatomical \nvariations  in  ears  which  cause  deviations  in  the  dynamical  constants \nof  the  transmission  system  from  sound  source  to  the  ear  drum.  The \ndotted  lines  following  the  curve  of  minimum  audibility  represent \napproximately  the  \"standard  deviation.\"\n\n3.  Maximum  Andibility.  The  curve  marked  \"Maximum  audi- \nbility\" represents  the  logarithmic  average  pressure  on  48  normal  ears \nrequired  to  produce  the  sensation  of  feeling.  This  represents  the \nthreshold  of  feeling  in  the  same  way  that  the  minimum  audibility \ncurve  represents  the  threshold  of  audition.  A  sound  much  louder \nthan  this  is  painful.  The  measurements  were  taken  through  a  range \nof  from  60  to  3,000  cycles.  The  standard  deviation  lines  are  also \ngiven  from  which  it  will  be  seen  that  this  curve  is  quite  as  definite  as \nthat  of  minimum  audibility.  While  this  point  of  feeling  probably  has \nno  relation  to  the  auditory  sense  it  does  serve  as  a  practical  limit  to \nthe  range  of  auditory  sensation.  A  few  observations  indicate  that \npeople  with  abnormal  ears  have  a  point  of  feeling  sound  which  is  not \ngreatly  different  from  that  of  normal  ears,  but  this,  of  course,  depends \non  the  type  of  abnormality.  The  intensity  for  feeling  is  about  equal \nto  that  required  to  excite  the  tactile  nerves  in  the  finger  tips.\n\n4.  Loiver  and  Upper  Frequency  Limits  of  Hearing.  The  curves  of \nminimum  and  maximum  audibility  in  Fig.  1  will  be  seen  to  have  been \nextrapolated  to  the  points  of  intersection  at  high  and  low  frequencies. \nThe  feeling  sensation  in  the  middle  range  of  frequency  is  first  a  tickling \nsensation  and  then  becomes  acutely  painful  as  the  loudness  is  in- \ncreased. As  the  frequency  is  decreased  the  sensation  of  feeling  becomes \nmilder  until  frequencies  around  60  cycles  it  is  sensible  as  a  flutter, \nbut  still  quite  different  from  the  sense  of  audition.  As  the  frequency \nis  still  further  decreased  to  a  point  where  the  hearing  and  feeling  lines \nappear  to  intersect,  it  is  difficult  to  distinguish  between  the  sense  of \nhearing  and  that  of  feeling.  The  low  point  of  intersection  of  the  two \nnormal  curves  of  minimum  audibility  and  feeling  sense  may,  there- \nfore, be  taken  arbitrarily  as  the  lower  tone  limit  of  audibility.  For \nfrequencies  lower  than  this  it  is  easier  to  feel  than  to  hear  the  air \nvibration.  The  point  of  intersection  cannot  be  determined  by  direct \nobservation  due  to  the  diflficulty  in  distinguishing  between  the  two \nsensations.  A  .similar  intersection  of  the  two  curves  occurs  at  some \nvery  high  frequency.  Sound  waves  of  frequencies  below  the  lower \nintersection  and  above  the  upper  intersection  are  more  easily  sensed\n\nby  feeling.    Soiiiul  waxes  between  these  limits  are  more  easily  sensed \nby  audition. 2\n\nThis  suggests  a  rational  way  of  defining  and  determining  the  two \nfrequency  limits  of  audibility.  Measurements  of  these  limits  which \nhave  been  made  in  the  past  are  questionable  because  the  intensity \nfactor  has  been  neglected.  At  the  lower  limit  of  audibility  the  ex- \ncursions of  the  diaphragm  and  ossicles  of  the  middle  ear  are  probably \nso  large  that  the  nerves  feeding  these  movable  parts  are  stimulated. \nThis  observation  at  low  frequencies  as  indicated  in  this  work  lends \ncolor  to  the  hypothesis  of  otologists  that  abnormalities  in  the  hearing\n\nfrequencies  is  considered  an  indication  of  obstructive  deafness  if  there\n\n^  The  extrapolation  upward  of  the  curve  of  minimum  audibility  is  consistent \nwith  some  recent  observations  of  Mr.  C.  E.  Lane  at  the  University  of  Iowa,  Physi- \ncal Review,  May,  1922.\n\ncharacteristic  of  the  normal  ear.  Any  point  within  this  area  repre- \nsents a  definite  auditory  sensation  in  frequency  and  intensity.  The \narea  of  sensation  is  analogous  to  the  field  of  vision  of  the  eye.  The \npart  of  this  area  which  is  most  utilized  in  the  interpretation  of  speech \nis  represented  approximately  by  the  shaded  area  in  Figs.  1  and  2 \nand  corresponds  in  a  way  to  the  center  of  the  field  of  vision.  A \nnormal  listener  tries,  by  keeping  at  a  certain  distance  from  a  speaker, \nto  bring  this  part  of  his  sensation  area  into  play  in  the  same  way  that \nwhen  examining  an  object  he  directs  his  eyes  so  that  it  falls  in  the \ncenter  of  the  field  of  vision.\n\nAn  abnormal  ear  may  be  regarded  as  having  an  area  of  sensation \nwhich  is  smaller  than  the  normal  area  but  included  within  it.  Fig.  2 \nis  a  plot  of  the  minimum  audibility  of  the  right  and  left  ears  for  a \nman  (CHK)  having  a  \"catarrhal\"  deafness.  The  areas  between \nthese  curves  of  minimum  audibility  and  the  curve  of  feeling  are  his \nareas  of  sensation.  It  will  be  seen  that  CHK  retains  about  50  or  60 \nper  cent  of  the  normal  amount  of  sensation.  He  hears  and  interprets \nconversation  with  some  difficulty.\n\nSince  the  CHK  curves  pass  through  the  speech  region,  part  of  it ' \nis  entirely  inaudible  and  the  remainder  is  near  minimum  audibility' \nfor  him.  In  order  to  make  him  hear  well,  the  speech  area  must  be' \nraised  to  a  higher  level  of  intensity  or  loudness  as  indicated  by  the \ndotted  curve.\n\nIn  general  it  takes  a  loss  of  about  20  per  cent  of  the  sensation  area \nto  become  noticeable  and  much  more  is  disagreeable.  A  loss  of  50 \nper  cent  requires  the  use  of  deaf  apparatus.  A  loss  of  75  per  cent \ncan  be  aided  considerably  by  the  use  of  high  powered  amplifying \napparatus.\n\n6.  Importance  of  Various  Intensities  in  Speech.  It  is  interesting \nto  speculate  on  how  CHK  interprets  speech.  It  has  been  shown  ' \nthat  the  intensity  of  speech  may  be  varied  over  perhaps  70-80  per \ncent  of  the  range  of  sensation  without  serious  loss  of  intelligibility \nto  the  normal  ear.  As  the  sound  intensity  is  decreased,  the  in- \ntelligibility drops  very  suddenly  to  zero  at  minimum  audibility.  A \nsimilar  drop  is  to  be  expected  at  an  intensity  so  loud  as  to  be  painful. \nIt  is  evident,  therefore,  that  the  range  in  which  speech  is  intelligible  for \nCHK  is  very  considerably  limited  as  compared  to  normal.  It  is \npossible  to  design  a  deaf  set  which  raises  the  intensity  of  the  principal \nspeech  region  to  any  desired  place  within  the  abnormal  sensation  area \nand  so  in  a  measure,  compensate  for  this  narrowed  range.  The \nregion  in  Fig.  1  \"Region  of  Lesser  Importance  in  Speech,\"  corre- \n'  H.  Fletcher,  Journal  of  the  Franklin  Institute,  June,  1922.\n\nspends  to  stimuli  in  conversation  of  lesser  energy  content,  such  as \nthe  minor  shadings  and  fainter  consonant  sounds.  While  it  is  physi- \ncally possible  to  produce  an  amplification  of  speech  so  that  this  region \nis  raised  into  the  diminished  area  it  is  impracticable  to  do  so  because \nof  the  pain  which  would  be  caused  by  the  louder  components.  A \ndiminution  in  sensation  area  can,  therefore,  be  only  partially  com- \npensated for.  In  case  the  area  is  extremely  narrow  a  deaf  set  fur- \nnishing optimum  volume  can  only  serve  as  an  aid  to  lip  reading.\n\n\"sustained,\"  \"smooth,\"  \"pure,\"  etc.  Such  a  description  may,  in \nfact,  be  taken  as  a  reasonable  indication  that  the  quality  of  sensation \nat  the  point  in  question  is  normal.  Abnormal  ears  sometimes  ex- \nperience a  subjective  degeneration  of  quality  of  pure  stimuli  which \nthey  describe  as  \"rough,\"  \"harsh,\"  \"sharp,\"  \"buzzing,\"  \"vibrating,\" \n\"hissing,\"  etc.  This  subjective  degeneration  is  independent  of  any \ntinnitus  or  head  noises  which  the  patient  may  have.  Fig.  3  shows \nvarious  regions  of  the  sensation  area  which  are  degenerated  in  the \ncase  of  CHK,  left  ear.  The  shaded  area  was  not  explored.  The \nboundaries  of  the  degenerated  regions  are  usually  more  sharply \nmarked  than  the  outer  boundaries  of  the  sensation  area.  The  sensa- \ntions in  these  areas  are  so  radically  different  from  the  sensation  of  a\n\npure  tone  that  it  is  with  difficulty  that  the  patient  is  convinced  that \nthe  stimulation  is  the  same  pure  tone  to  which  he  has  been  listening \nat  the  other  intensities.  The  subject  of  these  tests  is  a  violinist  and \ncapable  of  better  descriptions  and  finer  distinctions  than  average.\n\nSince  all  speech  sounds  may  be  considered  as  stimuli  composed  of \nvarious  frequency  components  of  certain  intensities,  the  sensation \ncaused  by  such  a  sound  may  be  represented  on  this  plot  by  points, \nor  by  a  line  provided  the  sound  has  a  band  spectrum.  If  the  points \nor  line,  falls  within  the  sensation  area  the  sound  is  audible.  It  is \neasy  to  see  that  if  the  points  or  the  part  of  the  line  which  represent \nthose  frequency  components  most  essential  to  interpretation  of  the \nsound,  fall  within  any  of  these  abnormal  areas,  the  sound  is  very \nlikely  to  be  misinterpreted.  This  adds  a  further  source  of  loss  in \nintelligibility  to  that  already  observed  due  to  a  narrowing  of  the \nsensation  range.  When  an  amplifying  deaf  set  is  designed,  due  care \nshould  be  taken  to  raise  the  principle  speech  region  in  such  a  way \nas  to  cause  a  minimum  overlapping  with  the  abnormal  areas.\n\nMany  practically  normal  ears  have  verv  small  abnormal  areas. \nThey  have  always  been  found  near  minimum  audibility  and  \"f  this  is \nalways  true  would,  therefore,  have  little  influence  on  the  hearing  of \nthe  individual.  They  seem  to  be  associated  with  \"catarrhal\"  con- \nditions although  this  cannot  be  stated  positively.\n\n8.  Binaural  Sense.  The  normal  individual  has  learned  to  interpret \nthe  differential  sensations  of  the  two  ears  to  advantage.  It  helps \nhim  to  locate  the  direction  from  which  sounds  come,  to  have  a  sort  of \nsense  of  orientation  with  respect  to  sounds  approaching  from  different \ndirections,  and  whether  for  physical  or  for  purely  psychological \nreasons  to  assist  in  focusing  of  the  attention  on  one  sound  of  a  large \nnumber.  Two  ears  also  assist  the  individual  in  perceiving  equally \nwell  sounds  coming  from  different  directions.\n\nWhen  one  ear  becomes  less  sensitive,  even  though  the  loss  is \nsmall,  the  use  of  the  binaural  sense  disappears  an]  after  a  time  is  not \nmissed,  the  subject  depending  upon  other  means  of  locating  sounds. \nFor  the  binaural  sense  to  be  most  effectively  utilized  it  is  necessary \nthat  the  ears  be  very  nearly  alike.  When  a  binaural  deaf  set  is  made \nand  fitted  to  a  person  with  compensating  sensitivity  for  the  two  ears \nso  that  both  hear  the  sounds  equally  loud,  the  sensation  is  usually \nso  novel,  that  if  the  patient  is  actually  able  to  experience  a  binaural \nsensation  he  is  very  much  pleased.  Usually,  however,  he  has  not \nused  his  binaural  sense  for  so  long  a  time  that  it  takes  a  considerable \namount  of  practice  before  he  is  able  to  have  binaural  experiences.  It \nmay  be  noted   in   this  connection   that  the  same  experience  is  en-\n\ncountered  in  fittinj^  the  eyes  with  jjlasses.  It  is  found  that  people \nwith  two  eyes  which  are  slightly  different  do  not  see  stereoscopically \nbut  if  glasses  are  made  so  as  to  compensate  and  make  the  eyes  nearly \nalike,  it  usually  takes  a  certain  amount  of  practice  before  the  sense \nof  perspective  can  be  brought  back.\n\nAs  a  source  of  sound,  a  small  thermal  receiver  unit  was  used.  This \nconsisted  ot  about  twenty  very  small  loops  of  Wollaston  wire  con- \ntained in  a  brass  case  small  enough  to  be  inserted  in  the  external  ear \ncanal  and  entirely  stop  it  up.  In  the  average  ear  a  volume  of  about \n1  cm.^  of  air  is  included  between  it  and  the  drum  membrane.  A  direct \nheating  current  is  passed  through  the  receiver  and  an  alternating \ncurrent  of  the  desired  frequency  and  intensity  is  superimposed  to \nmodulate  its  temperature.  This  modulation  in  temperature  causes \nalternate  expansion  and  contraction  of  a  very  thin  film  of  air  covering \nits  surface  and  so  produces  alternations  in  pressure  in  the  ear  canal \nof  the  frequency  of  the  impressed  alternating  current.  The  intensity \nis  proportional  to  this  alternating  current  if  it  is  maintained  small \ncompared  with  the  direct  current.  This  arrangement  permits  of \nproducing  alternating  or  sound  pressure  on  the  ear  drum  with  a \ncomparatively  simple  dynamical  relation  between  the  source  of \nsound  and  the  ear  drum.  The  thermal  receiver  is  also  dynamically \none  of  the  simplest  sources  of  sound  known.\n\nThe  sound  or  alternating  pressure  was  determined  by  calibration. \nThis  was  done  by  inserting  the  thermal  receiver  in  an  air  cavity  of \n1  cm.^  volume  in  front  of  a  condenser  transmitter  diaphragm  by  which \nthe  alternating  pressure  developed  by  a  given  current  in  the  receiver \ncould  be  measured.^  By  measurement  of  the  current  for  minimum \naudibility  or  \"maximum  audibility\"  or  for  any  other  intensity  the \npressure  in  the  ear  canal  is  determined.\n\n10.  Dynamical  Principles  of  Ear  Measurements.  From  a  dynamical \nstandpoint  the  phrase  \"sensitivity  of  the  ear\"  as  it  is  usually  used  is\n\n*  For  further  details,  see  \"The  Frequency  Sensitivity  of  Normal  Ears,\"  H.  Fletcher \nand  R.  L.  Wegel,  Physical  Review,  June,  1922.\n\n*  For  the  method  of  calibration  of  the  condenser  transmitter,  see  article  by  H.  D. \nArnold  and  I.  B.  Crandail,  Physical  Review,  July,  1917.\n\nrather  indefinite.  When  a  figure  is  given  in  ergs  per  second,  the  rate \nof  flow  of  energy  through  an  area  equal  to  that  of  the  ear  opening  in \nan  unobstructed  wave,  is  commonly  meant.  This  has  no  simple \nrelation,  theoretically  at  any  rate,  to  the  net  rate  of  flow  of  energy \ninto  the  ear  when  the  head  is  placed  as  an  obstruction  to  the  wave. \nThe  distortion  of  the  sound  field  by  the  head  varies  greatly  with \nfrequency.  Similarly,  there  is  no  simple  relation  between  the  energy \nflowing  into  the  ear  and  that  transmitted  to  and  absorbed  by  the  ear \ndrum  or  by  the  cochlea.  In  the  experiments  recorded  above,  atten- \ntion was  paid  to  the  experimental  set-up  so  as  to  make  the  figures \ngiven  have  a  more  definite  dynamical  significance.  Sensitivity  is \ngiven  in  terms  of  the  alternating  (root  mean  square)  pressure  to \nproduce  a  minimum  audible  sensation.  The  term  \"pressure\"  has  so \nfar  been  used  in  a  rather  loose  sense.  Just  why  this  is  so  will  be  seen \nfrom  the  following  argument.\n\nr  =  frictional  resistance  to  motion,  with  respect  to  a  station- \nary body  and  involves  dissipation  of  energy  at  a  rate \nof  Pr  where  x  is  the  root  mean  square  value  of  the \nrelative  velocity.  The  velocity  x  will  be  assumed \nsimply  sinusoidal  in  what  follows,\n\nni  =  mass  or  inertia  constant  involving  an  average  storage \nkinetic  energy  of  x^m  through  one  cycle,\n\ns  =  stiffness  constant  involving  an  average  storage  of \npotential  energy  through  one  cycle  of  x^s/w^.\n\nIf  the  r.  m.  s.  alternating  force  acting  is  F,  the  motion  at  any \nfrequency  is  given  by\n\nas  \"lunipc'd,\"  that  is  in  which,  for  the  pur[K)sc  of  {iractical  sohition, \na  finite  number  of  degrees  of  freedom  may  be  assumed,  the  method \nis  to  find  the  most  useful  way  of  \"lumping\"  these  constants.  The \nmotions  are  then  represented  by  a  series  of  equations,  one  for  each \ndegree  of  freedom,  between  the  forces  acting  and  the  impedances \nand  velocities.  The  determinant  of  the  coefficients  of  these  equations \nis  the  Lagrange  determinant  of  the  system.  The  only  caution  to  be \nobserved  in  lumping  the  constants  is  that  the  reciprocal  relation, \nwhich  is  a  property  of  any  linear  system  holds  also  for  the  physical \nsystem  which  the  assumed  Lagrange  determinant  is  supposed  to \nrepresent.\n\nThe  dynamical  system  used  in  calibration  with  the  condenser \ntransmitter  consists  of  three  parts:\n\n(a)  The  very  thin  pulsating  air  film  over  the  thermal  receiver \nfilaments.  The  expansion  of  air  around  the  wires  is  represented  by \nthe  \"diffusion\"  equation,  the  solution  of  which  in  such  a  case  of \ncylindrical  symmetry  is  given  as  a  Bessel's  function  of  the  distance \nfrom  the  wire.^  This  wave  is  so  quickly  damped  in  travelling  away \nfrom  the  wire  as  to  be  negligible  beyond  the  first  zero  point  of  the \nBessel's  function.  The  vibrating  system  of  this  receiver  may  then \nbe  considered  as  a  cushion  of  air  next  to  the  wire  of  a  thickness  a  little \nless  than  the  first  half  wave  length  of  the  heat  w'ave.  The  thickness \nof  this  cushion  is  an  inverse  function  of  the  frequency.\n\n(b)  The  air  chamber  between  the  thermal  receiver  and  con- \ndenser transmitter  diaphragm  having  a  volume  of  1  cm.'  and  en- \nclosed by  practically  unyielding  w^alls  w^ith  no  openings.\n\n(c)  The  condenser  transmitter  diaphragm,  being  stretched  very \ntightly  and  air  damped.  It  may  also  be  regarded  as  unyielding,  or  as \nhaving  an  impedance  very  high  compared  to  that  of  the  connecting \nair  chamber.\n\nIf  for  simplicity  the  mass  reaction  and  internal  losses  in  the  air \nchamber  may  be  neglected,  it  may  be  seen  that  the  moving  system \nof  the  receiver  may  be  regarded  as  a  weightless  and  frictionless \n\"diaphragm\"  surrounding  the  wires  at  a  distance  equal  to  the  effective \nthickness  of  the  active  air  film  and  may  be  shown  to  have  an  intrinsic \nstiffness  reactance  of:\n\nIn  this  expression,  7  is  the  adiabatic  constant  of  air,  po  the  atmos- \npheric pressure,  ai  the  area  of  the  fictitious  diaphragm,  and  Vi  the \nvolume  of  air  in  the  film.  This  diaphragm  is  loaded  externally  by \nthe  air  chamber,  when  the  transmitter  diaphragm  is  prevented  from \nmoving,  by  a  stiffness  reactance  of\n\nin  which  v  is  the  volume  of  the  air  chamber.  Similarly,  the  load  of \nthe  air  chamber  on  the  transmitter  diaphragm,  whose  area  is  02,  is\n\nThe  air  chamber  also  acts  as  a  mutual  impedance  between  the \nthermal  unit  and  the  transmitter  diaphragm  equal  to\n\nIf,  further,  the  intrinsic  impedance  of  the  transmitter  diaphragm, \nwhich  may  be  any  function  of  frequency,  be  denoted  by  Z2,  the \nequations  of  motion  of  the  system  may  be  written\n\nIn  these  equations,  F  is  the  force  acting  on  the  thermal  receiver \n\"diaphragm\"  due  to  alternating  current,  Xi  the  velocity  of  its  motion \nand  Xi,  the  velocity  of  motion  of  the  condenser  transmitter  diaphragm. \nA  rough  calculation  shows  that  Vi  is  very  small  compared  with  v, \nso  that  Si  may  be  neglected  compared  to  Si  and  that  the  reaction \nyiyiXi  may  be  neglected.  The  analysis  of  the  condenser  trans- \nmitter shows  Z2  to  be  very  large  compared  to  M'^.  These  equations \nmay  then  be  rewritten\n\nThe  equations  of  motion,  when  the  receiver  is  inserted  in  the  ear, \nmay  be  derived  in  a  similar  way.  In  this  case,  although  the  volume \nof  air  between  the  receiver  and  ear  drum  is  the  same  as  before,  the\n\nwalls  may  yield  appreciably,  particularly  in  some  frequency  ranges. \nThe  mutual  impedance  between  the  receiver  and  ear  drum,  is,  there- \nfore, not  necessarily  a  simple  stiffness  reactance.  Also  the  loads  due \nto  it  on  the  thermal  receiver  and  ear  drum,  which  in  this  case  takes  the \nplace  of  the  transmitter  diaphragm,  are  not  simple  stiffness  reactances. \nThe  constants  in  the  case  of  the  ear  system  will  be  denoted  by  the \nsame  letters  as  those  used  in  the  calibration  but  with  the  primes \ndropped,  with  the  exception  that  the  intrinsic  impedance  of  the  ear \ndrum  is  denoted  by  D.  D  includes  the  reactions  of  the  ossicles  of \nthe  middle  ear  and  the  cochlea  and  is  probably  a  complicated  function \nof  frequency.  If,  as  may  be  expected,  nature's  design  is  efficient, \nthen  D  must  be  of  the  same  general  order  of  magnitude  as  the  load  on \nthe  ear  drum,  M12,  of  the  ear  canal.  This  probably  constitutes  the \nlargest  difference  between  the  calibration  and  the  observational \nsystems.  Strictly,  of  course,  the  condition  for  maximum  power \nabsorption  by  the  ear  drum  from  the  air  is  that  D  be  the  conjugate \nof  the  impedance  of  the  load  on  it  due  to  the  unobstructed  ear  canal. \nThis  condition  is  not  obtained  in  nature  because  of  such  requirements \nplaced  on  the  design  as  protection  from  injury,  etc.\n\nIn  the  case  of  the  ear,  Mi  may  again  be  neglected,  compared  to \nZi,  and  the  reactance,  M12  ^2  may  be  neglected.     Then\n\nwhere  .V2  represents  the  velocity  of  motion  of  the  ear  drum.  Suitable \nvariations  with  frequency  are  implied  in  each  of  the  \"  constants  \" \nof  this  system.\n\nWe  are  now  in  a  position  to  see  just  what  has  been  measured  and \ncalled,  for  the  sake  of  brevity  or  want  of  a  better  name,  \"  minimum \naudible  pressure  \"  in  the  first  part  of  this  paper.\n\nLet  X\\  now  represent  the  velocity  of  the  receiver  diaphragm  in \nboth  systems  corresponding  to  that  necessary  to  obtain  a  minimum \naudible  sensation  in  the  ear,  and  F  the  corresponding  force.  Then \nX2  will  be  the  velocity  of  the  ear  drum  corresponding  to  minimum \naudibility  in  equation  (2).  In  the  calibration,  the  pressure  P'  on  the \ncondenser  transmitter  diaphragm  corresponds  to  Xi.  The  total  force \nacting  on  this  diaphragm  is  p'a'  where  now  a'  designates  its  area. \nSince  this  force  is  relieved  by  the  motion  of  the  diaphragm,  it  is  seen \nfrom  equation  (1)  to  be  equal  to  \\\n\nSimilarly  if  the  actual  pressure  on  the  ear  drum  is  p,  and  its  effec- \ntive area,  a,  the  total  force  on  the  ear  drum  pa  =  D  x^-  Combining \nequations  (2)  and  (3)  gives\n\nThe  pressure  p  is  the  actual  pressure  on  the  ear  drum.  The \npressure  p'  is  that  measured  and  plotted  in  the  diagram.  If  the  walls \nof  the  ear  canal  and  the  ear  drum  were  unyielding,  p  and  p'  would  be \nidentical  for  then  M  =  M'  and  M^  Xija  would  vanish.  If  the  yield \nof  the  ear  canal  walls  were  such  as  to  relieve  half  the  pressure  in  the \ncanal  and  that  of  the  ear  drum  about  the  same,  the  difference  would  be \nconsiderably  less  than  one  of  the  divisions,  in  the  diagrams,  on  the \nintensity  scale.  If  the  drum  impedance  D  should  be  found  to  be \nnegligible  compared  to  its  load  Mi  the  difference  would  be  consider- \nable. This,  however,  is  hardly  to  be  expected  even  through  narrow \nranges  of  frequency.  If  the  impedances  in  the  formulas  were  measured \nthe  energy  flow  into  the  ear  drum  could  be  computed.\n\nIn  conclusion,  the  present  status  of  the  ear  problem  may  be  sum- \nmarized. The  philosophy  of  external  ear  dynamics  has  been  touched \non  but  there  still  remain  difficult  problems  both  theoretical  and  ex- \nperimental. A  start  has  been  made  on  a  sound  basis  in  the  explana- \ntion of  the  action  of  the  cochlea  by  Roaf,  \"  Analysis  of  Sound  Waves \nby  the  Cochlea,\"  Philosophical  Magazine,  February  1922.  Nothing \ndependable  has  as  yet  been  published  on  the  action  of  the  middle \near  for  audio  frequencies.  It  is  usually  assumed  that  the  various  parts \nundergo  relative  displacements  at  audio  frequencies  in  the  same  way \nas  they  react  to  static  forces  but  this  is  very  likely  far  from  the  truth.\n\nTHE  Theory  of  Probabilities  lends  itself  to  the  solution  of  many \nimportant  telephone  problems.  These  problems  arise  not  only \nin  connection  with  the  trunking  of  calls  but  also  in  statistical \nstudies  which  underlie  the  making  of  fundamental  plans,  in  studies \ncarried  on  in  physical  research  and  in  the  manufacturing  of  telephone \napparatus.\n\nThe  purpose  of  the  present  paper  is  to  discuss  certain  simple  types \nof  trunking  problems  which  can  readily  be  handled  to  a  sufificient \ndegree  of  approximation  by  well-known  probability  methods.  It \nwould  be  quite  impossible,  within  the  scope  of  a  single  paper  to  give \na  complete  discussion  of  trunking  problems  in  general.  For  years  ^ \nit  has  been  known  that  light  could  be  shed  on  these  problems  by  the \napplication  of  probabilities  and  many  articles  ^  have  appeared  on \nthis  subject;  however  the  treatment  to  be  found  in  the  literature  is, \nas  yet,  by  no  means  comprehensive.\n\nAbout  1905,  the  development  of  machine  switching  systems  arrived \nat  a  stage  where  the  relative  efficiencies  of  different  sizes  of  trunk \ngroups  became  of  prime  importance.\n\nIn  designing  and  engineering  machine  switching  systems,  it  is \nnecessary  to  compare  the  costs  of  various  plans  using  trunk  groups \nof  widely  different  sizes,  in  order  to  choose  the  cheapest  arrangement. \nSome  plans  use  trunk  groups  as  small  as  5  and  others  groups  as  large \nas  90.\n\nMachine  switching  development,  therefore,  gave  a  great  impetus \nto  the  application  of  the  Theory  of  Probabilities  to  telephone  engineer- \ning and  in  the  Bell  System  work  along  this  line  has  been  in  progress, \nsystematically,  for  many  years.  This  work  has  included  not  only \nthe  theoretical  solutions  of  various  trunking  problems,  but  has  also \ninvolved  the  computation  of  special  probability  tables  and  collec- \ntion of  data  by  means  of  which  theoretical  results  have  been  closely \nchecked.\n\nIn  the  articles  which  have  hitherto  appeared,  little  or  no  effort \nhas  been  made  to  present  the  mathematical  theory  of  trunking  in  a\n\n'  G.  T.  Blood  of  the  A.  T.  &  T.  Co.  in  1898  found  a  close  agreement  between  the \nterms  of  a  binomial  expansion  and  the  results  of  observations  on  the  distribution \nof  busy  calls.  The  first  comprehensive  paper  was  one  written  by  M.  C.  Rorty  in \nOctober  1903  and  was  quite  widely  circulated  within  the  Bell  System.\n\nmanner  that  can  be  understood  by  those  who  are  not  experts  on  the \nsubject.  It  is  hoped  that  this  article  will  assist  the  reader  in  under- \nstanding both  what  has  been  and  will  be  written  on  the  subject.  As \nPoisson  ^  has  said  \"a  problem  relative  to  games  of  chance  and  proposed \nto  an  austere  Jansenist  by  a  man  of  the  world,  was  the  origin  of  the \ncalculus  of  probabilities,\"  and  today  the  reader  will  find  that  in  the \nmajority  of  text  books  the  subject  is  introduced  by  the  solution  of \ngames  of  chance  and  particularly  of  dice  problems.  This  established \ncustom  will  be  followed  by  the  present  writer  who,  in  the  course  of \nthis  article,  will  show  how  various  fundamental  trunking  problems \ncan  be  transformed  into  equivalent  dice  problems.  This  being  done, \nsolutions  will  be  found  to  be  at  hand.\n\nThree  trunking  problems,  each  one  step  more  complicated  than \nthe  preceding,  will  be  dealt  with.  In  order  to  facilitate  the  trans- \nformation to  the  three  equivalent  dice  problems  it  is  desirable  that \nthe  basic  assumptions  made  be  as  simple  as  possible.  The  asump- \ntions  made  in  all  three  problems  are:\n\nConditions  substantially  approximating  this  assumption  frequently \noccur  in  practice.\n\nB \u2014 If  a  call  when  initiated  obtains  a  trunk  immediately  it  retains \npossession  of  that  trunk  for  exactly  two  minutes.  In  other  words,  a \nconstant  holding  time  of  two  minutes  duration  will  be  assumed.\n\nIn  practice,  holding  times,  of  course,  vary  from  a  few  seconds  to \nmany  minutes  and  it  may  at  first  sight  seem  that  the  assumption  of \na  constant  holding  time  might  lead  to  results  deviating  too  much \nfrom  practice  to  be  of  value.  On  this  point,  the  theory  of  probabilities \nitself  sheds  some  interesting  light.  As  will  be  pointed  out  in  the \nfollowing  problems,  the  assumption  of  a  constant  holding  time  is \nthe  equivalent  of  a  dice  problem  in  which  a  single  die,  or  several  iden- \ntical dice  are  considered.  The  telephone  problem  with  variable \nholding  times  may  be  reduced  to  the  consideration  of  many  dice, \neach  with  a  different  number  of  faces.  Suppose  600  throws  are  made \nwith  a  die  having  6  faces  so  that  on  the  average  3^  of  600  or  100  aces \nwould  be  expected.  With  Bernoulli's  formula  it  is  easy  to  find  the \nprobability  that  the  number  of  aces  which  turn  up  shall  lie  between \n75  and  125,  that  is  to  say,  within  25  on  each  side  of  the  average.  Now \nsuppose  200  throws  are  made  with  a  die  having  20  faces,  200  with  a\n\n10  face  die,  100  with  a  5  fac-e  die  and  finally  100  with  a  2  face  die. \nThese  600  throws  would  also  give  on  the  average  100  aces.  Using \nPoisson's  generalization  of  the  Bernoulli  formula'  we  can  calculate \nthe  probability  that  these  (iOO  throws  with  various  kinds  of  dice  shall \ngi\\e  a  number  of  aces  lying  between  75  and  125.  This  probability \nwill  be  greater  than  in  the  case  of  the  600  throws  with  the  die \nwith  a  constant  number  of  faces,  i.e.,  the  chance  that  the  result  will \ncome  outside  the  range  75  to  125  is  less.\n\nThe  thought  is  at  once  suggested  that  for  the  same  total  volume  of \ntraffic  and  a\\-erage  holding  time,  fewer  calls  would  be  lost  when  the \nholding  time  is  not  constant.'*  The  above  theory  was  tested  in  practice \na  few  \\ears  ago  by  the  engineers  of  the  American  Telephone  and  Tele- \ngraph Company,  who  made  pen  register  records  of  hundreds  of  thou- \nsands of  actual  calls  as  handled  by  groups  of  machine  switching \ntrunks  at  Newark,  New  Jersey.  A  pen  register  was  made  which \noperated  as  follows:  Each  trunk  in  the  group  was  represented  by  a \npen.  These  pens  were  mounted  side  by  side  and  each  was  controlled \nby  a  magnet  in  such  a  manner  that  when  the  trunk  was  busy  the  pen \nmade  a  mark  on  a  wide  strip  of  paper  driven  at  constant  speed  under \nthe  pens.  There  was  thus  obtained  a  record  showing  when  each  call \noriginated  and  when  it  was  concluded.  An  artificial  record  was  now \nmade  showing  what  would  have  happened  if  each  call  had  lasted  for \nthe  average  holding  time  as  determined  from  the  original  record. \nSome  100,000  calls  w^ere  analyzed  in  this  manner  and  it  w^as  found  that \nwith  a  group  of  trunks  of  a  size  to  carry  the  calls  of  the  original  record \nwith  only  a  small  loss,  30  per  cent,  more  calls  would  have  been  lost  if \nthe  traffic  had  been  as  shown  by  the  artificial  record.  It  should  be \nborne  in  mind,  however,  that  a  30  per  cent,  change  in  a  probability \nof  the  order  of  one  in  one  hundred,  considering  the  values  we  are  deal- \ning with,  is  practically  negligible.\n\nThis  assumption,  although  artificial,  simplifies  materially  the  analy- \nsis of  the  problems.     Just  what  happens  in  practice  to  every  call\n\nwhich  fails  to  get  a  trunk  immediately  is  unknown.  It  is  obvious, \nhowever,  that  when  the  number  of  trunks  is  such  that  the  liability  of \nthe  call  failing  to  get  a  trunk  immediately  is  very  small \u2014 for  example: \nof  the  order  of  one  in  one  hundred \u2014 the  reaction  of  these  calls  on  other \ncalls  must  be  negligible  independently  of  whatever  assumption  ^  is \nmade  in  place  of  C.\n\nit  comes  to.  The  20  points  of  all  switches  are  multipled  together  so \nthat  a  single  group  of  20  trunks  must  handle  the  calls  originating  from \nthese  269  lines.\n\nReferring  to  Fig.  2  let  point  P  represent  the  unknown  instant \nwithin  the  hour  at  which  X  calls.  Consider  the  two  minutes  imme- \ndiately preceding  the  instant  P.  Evidently,  by  assumption  C,  calls \nfalling  outside  of  this  particular  two-minute  interval  can  not  prevent \nX  from  obtaining  a  trunk.\n\n*  It  is  well  known  that  the  Erlang  formula  which  is  based  on  an  assumption  dia- \nmetrically opposed  to  assumption  C,  namely  that  calls  which  find  all  trunks  busy \ndo  not  wait  for  a  trunk  to  become  idle,  gives  essentially  the  same  results  (for  small \nprobabilities,  which  are  the  only  ones  of  interest  in  practice)  as  the  Poisson  furmula \nwhich  assumes  C.\n\nIt  is  evident  then,  that  the  probability  that  X  fails  to  get  a  trunk \nimmediately  is  the  same  as  the  probability  of  throwing  at  least  20  aces \nif  268  throws  are  made  with  a  30-face  die.  To  facilitate  the  de- \ntermination of  this  probability  and  the  solution  of  similar  problems, \nprobability  tables  of  a  type  shown  in  Table  I  have  been  computed.*^ \nIn  the  table,  the  average  number  of  times  an  event  may  be  expected \nis  represented  by  a.  The  probability  that  the  event  occurs  at  least \na  greater  number  of  times  c  =  a  -[-  d  \\s  represented  by  P.  In  the \nproblem  under  consideration,  the  average  number  of  aces  expected  is\n\nthe  table,  we  find  that  corresponding  to  c  =  20  and  a  =  8.96,  the \nvalue  of  the  probability  P  is  .OOL  In  the  particular  telephone  prob- \nlem under  consideration  this  means  that  once  in  a  thousand  times\n\nor  choices.  The  reader  unfamiliar  with  automatic  systems  may \nconsider  a  10  level  selector  as  one  from  which  calls  may  be  sent  in  10 \ndifTcrent  directions.  Assume  that  each  level  is  equipped  with  8 \ntrunks  to  second  selectors.  The  591  line  switches  are  multipled \ntogether  so  that  one  group  of  35  first  selectors  must  handle  the  calls \noriginating  from  these  591  lines.  The  35  first  selectors  are  multi- \npled  together  so  that  one  group  of  8  second  selectors  must  handle \nthe  calls  originating  from  the  591  lines  for  a  particular  level.  It  is \nassumed  that  the  591  calls  are  distributed  at  random  with  reference \nto  the  10  levels  of  the  first  selectors.\n\nThe  probability  of  Y  fulfilling  the  first  condition  is  equal  to  the \nprobability  of  throwing  the  ace  with  a  30  face  die.  The  probability  of \nY  fulfilling  the  second  condition  is  equal  to  the  probability  of  throw- \ning the  ace  with  a  10  face  die.\n\nThe  question  may  then  be  stated  in  the  form  of  a  dice  problem  as \nfollows:  591  throws  are  made  with  a  30  face  die  giving  C  aces.  C \nthrows  are  made  with  a  10  face  die  giving  D  aces,  and  the  question \nis  the  probability  that  D  is  not  less  than  8.  Assuming  no  restric- \ntion ^  on  the  value  of  Cthis  probability  is  the  same  as  that  of  throwing \nat  least  8  aces  in  591  throws  with  a  die  having  (30)  (10)  =  300  faces.\n\nThe  average  number  of  aces  to  be  expected  is  (591/300)  =  1.97  and \nwith  this  average  the  tables  tell  us  that  once  in  a  thousand  times \nwe  may  expect  at  least  8  aces.\n\n'  Since  it  is  assumed  that  X  obtained  a  first  selector  it  follows  that  in  the  2  minutes \npreceding  the  instant  when  X  called  the  number  of  calls  must  have  been  less  than \nthe  number  of  first  selectors  and  we  should,  therefore,  not  count  the  throws  giving \nvalues  of  C  which  are  not  less  than  the  total  number  of  first  selectors.  This  res- \ntriction becomes  of  practical  importance  only  where  a  large  proportion  of  the  calls \nfrom  the  first  selectors  go  to  one  level.  To  take  an  extreme  case,  assume  that  all \nthe  calls  went  to  one  level,  and  that  therefore  each  10  first  selectors  would  require \n10  second  selectors  to  handle  the  traffic.  Placing  no  restriction  on  the  value  of  C, \nsince  C  exceeds  the  number  of  first  selectors  occasionally,  we  would  get  the  result \nthat  10  second  selectors  were  not  enough  to  handle  all  the  calls  from  10  first  selectors, \nwhich  is  of  course  absurd.  Where,  however,  the  values  of  C  e.xceeding  the  number \nof  first  selectors  are  assumed  to  be  distributed  over  all  10  levels  of  the  first  selectors \ntheir  effect  on  the  number  of  second  selectors  is  negligible.\n\nIn  practice,  a  modification  of  Problem  II  frequently  arises.  As- \nsume an  arrangement  similar  to  that  of  Problem  II  except  that  the \nnumber  of  lines  is  multiplied  by  a  factor  of  perhaps  3  or  more,  each \nline  switch,  however,  still  having  access  to  all  first  selectors.  The \nrequired  number  of  first  selectors  will  also  be  larger  but  not  in  ex- \nactly the  same  ratio  because  the  margin  of  idle  selectors  need  not  be \nrelatively  as  great  in  the  large  system  as  in  the  small.  An  enlarged \ngroup  of  trunks  running  from  the  first  selectors  to  the  second  selectors \nwill  now  be  required,  and  it  will  be  assumed  that  there  are  four  times\n\nas  many  trunks  coming  from  each  level  of  the  first  selectors  as  there \nare  points  of  contact  on  each  level.  To  meet  this  situation,  the \nfirst  selectors  and  their  outgoing  trunks  are  divided  into  four  sub- \ngroups as  shown  in  Fig.  4.  The  corresponding  sub-groups  of  second \nselectors  are  designated  by  d,  G2,  G3,  G4,  the  number  of  trunks  to \neach  sub-group  being  10.  The  solution  of  this  problem  depends \nprimarily  on  the  manner  in  which  the  line  switches  distribute  calls \nto  the  first  selectors.     Three  cases  will  be  considered.\n\nReferring  to  Fig.  4  consider  a  group  of  n  =  1486  lines,  and  let  the \ntraffic  be  divided  among  the  ten  levels  or  directions  available  in \nsuch  a  way  that  on  the  average  V3  of  the  calls  are  made  for  a  particular \ndirection  or  level.  Let  us  suppose  the  circuit  connections  between \nthe  line  switches  and  first  selectors  to  be  such  that  the  calls  are  dis- \ntributed individually  at  random.  By  this  is  meant  that  the  first \nselector  seized  by  a  calling  line  is  as  likely  to  be  one  having  access \nto  sub-group  Gi  as  to  sub-group  G2,  G3  or  G4.  Note  carefully  that \nthis  distribution  is  assumed  whether  or  not  the  calling  line  wants \nthe  particular  level  under  consideration.  One  way  of  securing  this \nrandom  distribution  by  sub-groups  would  be  to  allow  the  line  switches \nfirst  to  choose  by  chance  one  of  the  four  sub-groups  of  first  selectors \nand  then  to  choose  an  idle  first  selector  in  the  sub-group.\n\nAs  before  we  are  interested  in  the  calls  made  during  the  two  minutes \npreceding  the  instant  at  which  X  calls.  Let  the  number  of  these  calls  be \nC.  Of  these  C  calls  a  certain  number  D  want  the  level  for  which  X \nhas  called.  If  at  least  10  of  these  D  calls  were  distributed  by  the \nline  switches  to  first  selectors  having  access  to  the  same  sub-group \nas  the  one  to  which  the  first  selector  seized  by  X  has  access,  then \nthere  will  be  no  idle  trunk  in  the  sub-group  for  X.  Our  telephone \ntrunking  problem  evidently  transforms  to  the  following  series  of \ndice  problems.\n\n1st.       1486  throws  are  made  with  a  30  face  die  giving  C  aces. \n2nd.     C  throws  are  made  with  a  3  face  die  giving  D  aces. \n3rd.      D  throws  are  made  with  a  4  face  die  giving  x  aces,  and  the \nquestion  is  the  probability  that  x  is  not  less  than  10.\n\nBy  the  theory  of  dice  (assuming  no  restriction  ^  on  the  value  of \nC)  the  probability  is  the  same  as  that  of  throwing  at  least  10  aces  in \n1486  throws  with  a  die  having  30  X  3  X  4  =  360  faces.  The  average \nnumber  of  aces  to  be  expected  being  1486/360  =  4.13  the  probability \ntables  give  .01  as  the  answer.\n\nAs  in  Case  1  assume  that  on  the  average  3^  of  the  calls  are  for \nthe  level  under  consideration,  but  take  n  =  1725  for  the  number  of \nlines.     Now  suppose  the  circuit  connections  between  line  switches\n\nand  first  selectors  to  be  such  that  the  calls  are  distributed  uniformly \nto  the  first  selectors,  meaning  that  if  at  any  instant  C  calls  exist,  C/4 \nof  them  are  on  first  selectors  having  access  to  the  10  trunks  of  sub- \ngroup Gi,  C/4  are  on  first  selectors  having  access  to  the  10  trunks \nof  sub-group  G2  and  so  on.  With  a  constant  holding  time  such  as \nassumed  this  result  could  be  secured  by  a  device  common  to  all  line \nswitches  which  would  route  the  first  call  to  the  first  sub-group,  the \nsecond  call  to  the  second  sub-group,  etc.\n\nX  will,  as  before,  be  interested  in  the  calls  falling  in  the  two  minutes \npreceding  him.  By  hypothesis  y^  of  these  will  have  been  distributed \nto  first  selectors  having  access  to  the  same  sub-group  of  second  selec- \ntors as  the  first  selector  seized  by  X.  Finally,  the  probability  is  % \nthat  one  of  these  calls  wants  the  level  in  which  X  is  interested.  The \nequivalent  dice  problem  is  therefore:\n\naces  which  turn  up  are  noted.     Let  this  number  be  C. \n2nd.     C/4  throws  are  made  with  a  3  face  die.\n\nWhat  is  the  probability  that  this  sequence  of  throws  results  in  at \nleast  10  aces?  This  probability  is  not  that  of  getting  at  least  10  aces \nif  1725  throws  are  made  with  a  die  having  30  X  3  =  90  faces.  We \nmust  write  separately  the  formula  for  each  of  the  two  steps  of  the \nproblem,  then  multiply  them  together  and  finally  sum  the  product \nfor  all  values  of  C/4  from  10  up.  If  this  is  done,  again  ignoring  the \nrestriction  on  the  upper  limit  of  C,  the  answer  will  come  out  0.01. \nNote  that  whereas  in  Case  1  the  average  volume  of  traffic  carried \nby  a  sub-group  of  10  trunks  was  4.13,  in  this  case,  with  the  same \nprobability  of  failure,  it  is  1725  (1/30)  (1/4)  (1/3)  -  4.79.\n\nNumber  each  first  selector  and  a  corresponding  card;  shufifle  the \ncards  and  deal  out,  for  example,  37  of  them.  The  distribution  under \nconsideration  is  such  that  when  37  calls  exist  the  probability  that \nthey  occupy  a  specified  set  of  37  selectors  is  equal  to  the  probability \nthat  the  cards  dealt  have  the  corresponding  numbers.  This  dis- \ntribution of  calls  would  be  measurably  secured  by  arranging  the  line \nswitch  multiple  so  that  the  trunks  to  the  first  selectors  appear  so  far \nas  possible  in  a  different  order  before  every  line  switch.  This  case  of \ndistribution  differs  from  that  of  Case  1.     In  Case  1,  if  the  first  call\n\nfalls  on  a  first  selector  liaxing  access  to  sub-group  d,  for  example, \nthe  second  call  still  has  the  same  chance  of  falling  on  a  first  selector \nha\\ing  access  to  sub-group  Gi  as  on  one  having  access  to  any  one  of  the \nother  three  sub-groups.  In  Case  3,  however,  the  busy  first  selectors \ntend  to  be  distributed  uniformly  between  the  4  sub-groups,  so  that \nif  any  sub-group  should  have  a  preponderance  of  busy  first  selectors \nthe  probability  of  its  receiving  another  call  is  less  than  the  probability \nthat  one  of  the  other  sub-groups,  with  more  idle  first  selectors,  should \nreceive  it.     The  full  discussion  of  this  case  is  reserved  for  the  future.\n\nIf  it  is  known  that  one  of  two  events  must  occur  in  any  trial  or \ninstance,  and  that  the  first  can  occur  in  u  ways  and  the  second  in  v \nways,  all  of  which  are  equally  likeh-  to  happen,  then  the  probability \nthat  the  first  will  happen  is  mathematically  expressed  by  the  fraction\n\nthe  last  equation  following  from  the  first  two,  and  being  the  mathe- \nmatical expression  for  the  certainty  that  one  of  the  two  events  must \nhappen.\n\nIf  the  probabilities  of  two  independent  events  are  pi  and  p2  re- \nspectively, the  probability  of  their  concurrence  in  any  single  instance\n\nof  several  independent  events,  and  P  the  probability  of  their  con- \ncurrence, then\n\nConsider,  now,  what  may  happen  in  n  trials  of  an  event,  for  which \nthe  probability  is  p  and  against  which   the  probability  is  q.     The\n\nwhere  the  factor  p  appears  n  times;  that  is  the  probability  is  p\". \nThe  probability  that  the  event  will  occur  (\u00ab  \u2014  1)  times  in  succession \nand  then  fail  is  p\"\"'^  q.\n\nBut  if  the  order  of  occurrence  is  disregarded,  this  last  combination \nmay  arrive  in  n  different  ways;  so  that  the  probability  that  the  event \nwill  occur  (\u00ab  \u2014  1)  times  and  fail  once  is  n^\"~^  q.  Similarly,  the  prob- \nability that  the  event  will  happen  {n  \u2014  2)  times  and  fail  twice  is \n^\"~2  g2  multiplied  by  n{n  \u2014  l)/2,  etc.  That  is,  the  probabilities \nof  the  several  possible  occurrences  are  given  by  the  corresponding  terms \nof  the  binominal  expansion  of  {p  +  qV-     Let\n\nThen  P  =  probability  that  the  event  happens  exactly  n  times,  plus  the \nprobability  that  it  happens  exactly  {n  \u2014  1)  times  .  .  .  plus\n\nIf  the  series  for  P  contains  few  terms  it  may  be  computed  easily. \nIn  general,  however,  it  is  impracticable  to  compute  P  by  means  of \nthe  above  binomial  expansion.  Other  forms  for  the  value  of  P \nmust,  therefore,  be  developed.\n\nOne  of  the  most  convenient  approximations  for  P  when  p  is  small \nhas  been  developed  by  Poisson.  It  is  known  as  Poisson's  Exponen- \ntial Binomial  Limit  and  gives  the  value  of  P  by  the  following  ex- \npansion\n\nThe  following  Table  gives  corresponding  values  of  P,  a,  c  satisfying \nequation  (2).\n\nAverages  (a)    Corresponding  to  Deviation   (d)   plus  Average  (a)   to  be  Expected  with\n\nThe  Relation  Between  Rents  and  Incomes,  and \nthe  Distribution  of  Rental  Values\n\nSynopsis:  Many  parts  of  telephone  plant,  such  as  central  office  buildings \nand  equipment,  conduits,  underground  and  aerial  cable  at  the  time  of \ninstallation  must  have  the  capacity  to  handle  not  only  the  immediate \ndemand  for  telephone  service,  but  also  to  take  care  of  growth  for  a  number \nof  years  to  come.  In  order  to  engineer  such  items  of  telephone  plant \neconomically  it  is  necessary  to  know  in  advance  as  accurately  as  possible \nwhat  the  demand  for  telephone  service  will  be  five,  ten,  or  twenty  years \nin  the  future.  Forecasts  of  the  future  market  are  very  necessary \nfor  plant  engineering,  operating  plans,  rate  treatment,  and  other  purposes, \nin  multi-central  office  cities.  In  such  cities  detailed  estimates  are  made \nof  the  market  some  twenty  years  ahead  and  of  its  telephone  development \nunder  stated  rate  conditions.  Such  estimates  are  called  commercial  surveys, \nand  they  involve  a  study  of  the  various  factors  which,  in  the  course  of  events, \nwill  be  likely  to  control  the  industrial,  commercial  and  residential  develop- \nment of  the  city  concerned.\n\nIn  the  course  of  such  a  survey,  a  rental  classification  of  all  families  is \nobtained  and  at  the  same  time  a  record  is  made  of  existing  telephone  service \nin  each  rental  class.  The  rent  data  of  this  article  have  been  gathered  in \nrepresentative  large  cities  throughout  the  country  and  the  results  as  here \nset  forth  are  being  used  together  with  many  other  kinds  of  data  to  guide \nthe  engineering  of  future  additions  to  the  plant  of  the  Bell  Telephone  System.\n\nIn  general  the  income  of  a  family  is  an  index  of  the  market  it  creates  for \nvarious  commodities  including  telephone  service.  Rental  values  may  also \nbe  considered  as  such  an  index  and  the  present  study  seeks  to  correlate  rents \nwith  incomes.  Rents  can  be  readily  recorded  and  classified,  whereas  it  is \nnot  feasible  to  determine  the  money  incomes  of  large  numbers  of  families. \nWhile  it  may  be  ideally  possible,  by  a  study  of  rent  data,  to  compare  the \ninherent  markets  for  telephone  service  and  also  the  strength  of  the  tele- \nphone habit  in  various  cities,  there  are  many  practical  limitations  to  such \na  procedure.  Comparison  of  the  residence  market  for  telephone  service \nin  different  cities,  as  determined  by  rent  values,  is  made  difficult  by  the \nfact  that  the  variation  between  cities  in  rentals  paid  for  substantially  similar \ndwellings  is  considerably  greater  than  the  variation  in  prices  for  food  or \nclothing.  Further,  there  is  considerable  variation  in  rent  levels  even  in \ndifferent  sections  of  any  one  city.  Attempts  to  compare  rent  distribu- \ntions by  application  of  the  usual  statistical  measures  of  dispersion  and \nskewness  have  proved  unsatisfactory.  However,  a  method  of  charting \nhas  been  found  by  which  rent  distributions  may  be  readily  compared \nwith  one  another  and  an  index  of  spread  or  dispersion  determined.  It \nhas  been  found  that  cumulative  curves  of  rent  distribution  may  be  plotted \non  logarithmic  probability  paper  to  yield  straight  lines  for  a  large  number \nof  cities.  These  are  called  logarithmic  skew  distributions.  Although  it \nhas  not  been  found  possible  to  assign  any  special  significance  to  the  par- \nticular value  of  the  index  of  rent  dispersion  in  any  city,  this  index  appears \nto  remain  practically  constant  for  that  city  regardless  of  changes  in  the \nlevel  of  prices.  In  the  appendix  the  mathematical  features  of  the  loga- \nrithmic skew  curve  are  discussed. \u2014 Editor.\n\nTT  is  a  well  recognized  fact  that  the  better  class  families,  i.e.,  those \n-\u25a0-  with  higher  incomes,  are  a  better  market  for  telephone  service \nthan  the  poorer  families.  For  purposes  of  market  analysis  in  com- \nmercial surveys  it  is  not  feasible  to  determine  the  money  incomes \nreceived  by   families  but  the   rental  values  of  dwellings,   which,  as\n\nwill  be  shown,  are  a  measure  of  the  incomes  of  their  occupants,  are \ncomparatively  readily  collected  and  classified.  Rent  data  obtained \nin  the  course  of  a  commercial  survey  show  the  \"character\"  of  a  city \nand  are  used  as  a  basis  for  estimates  of  the  future  residence  telephone \nmarket.\n\nIn  view  of  the  importance  of  these  rent  data,  it  seems  desirable \nto  study  them  in  some  detail  to  find  out  just  what  their  limitations \nare.\n\nThere  may  be  set  down  in  advance  certain  things  which  it  is  desirable \nto  know,  as,  for  instance,  the  relationship  betw-een  money  incomes \nand  house  rents  and  methods  and  limitations  of  comparison  of  different \ncities  on  the  basis  of  rental  values.  On  the  first  point,  as  applied  to \nany  particular  city,  a  knowledge  of  the  relation  between  incomes  and \nrents  is  desirable  in  a  general  way,  although  there  is  no  necessity  to \ntranslate  the  telephone  market  expressed  in  terms  of  rent  types  into \na  scale  of  incomes.  On  the  second  point,  the  comparison  of  different \ncities,  it  should  be  ideally  possible  by  a  study  of  rent  data  to  compare \nthe  inherent  economic  markets  for  telephone  service,  and  also  to \nmeasure  differences  in  the  strength  of  the  telephone  habit,  but  in \npractice  only  rough  approximations  may  be  made.\n\nCertain  limitations  to  work  of  this  kind  are  fairly  evident.  The \nmost  obvious  difficulty  is  the  fact  that  rent  levels  have  changed \nalong  with  the  general  price  level.  Rent  lev^els  in  various  cities  differ \naccording  to  the  varying  degrees  of  housing  congestion  and  the \nvarying  social  standards  of  the  population.  Furthermore,  the  varia- \ntion in  rent  levels  extends  to  different  sections  of  any  one  city.  The \nmere  fact  that  a  given  family  paid  say  $30  rent  is  not  an  indication \nof  that  family's  economic  condition  or  its  value  as  a  telephone  pros- \npect, unless  there  is  also  known  the  city  and  the  part  of  the  city  in \nwhich  that  family  lived,  and  the  time  when  the  given  rent  was  paid. \nTherefore  rent  data  from  different  cities  and  of  various  dates  are \nnot  directly  comparable  at  their  face  value.  To  adjust  the  money \nvalues  of  house  rents  for  an  accurate  comparison  of  the  telephone \nmarkets  in  different  cities  w'ould  require  a  knowledge  of  the  relative \nproportions  of  income  spent  for  rent  in  the  different  cities,  of  the \nrelative  levels  of  incomes  and  rents  at  the  time  of  the  surveys  as \ncompared  w^ith  their  levels  in  some  base  year,  and  perhaps  of  other \nfactors  equally  difficult  to  estimate.\n\nVarious  rent  tabulations  can  not  be  compared  one  with  another \nwithout  knowing  something  of  the  way  in  which  rents  are  distributed \nabout  their  average.  The  nature  of  the  distribution  is  determined \nby  the  house  count  data,  but  from  those  data  in  their  usual  form  it\n\nis  not  easy,  when  making  comparisons,  to  make  proper  allowances  for \ndifferences  in  rent  levels  and  in  the  schedule  of  rent  classes.  In  the \nfollowing  pages  there  is  discussed  a  method  of  charting  by  which \nrent  distributions  may  be  readily  compared,  and  their  spread  or \ndispersion  determined.\n\nRents  as  a  Market  Index.  The  relation  between  rents  and  incomes \nis  concerned  with  the  use  of  rental  values  both  as  an  index  of  tele- \nphone market  in  a  given  city,  and  in  comparing  the  markets  in  differ- \nent cities.  In  what  follows  it  is  not  always  possible  to  separate  these \ntwo  views,  but  the  distinction  should  be  borne  in  mind  by  the  reader.\n\nThe  goal  of  an  analysis  of  residence  telephone  market  is  to  determine \nthe  future  sales  possibilities.  In  theory  either  incomes  or  rents  may \nbe  considered  as  an  index  of  the  telephone  market.  The  market \nindex  adopted  in  commercial  surveys  is  the  rental  of  dwellings. \nThis  may  be  considered  either  as  a  direct  measure  of  the  ability \nand  desire  of  families  to  subscribe  to  telephone  service,  or  as  an \nindirect  index,  if  incomes  are  considered  the  real  measure  of  the \nmarket.  If  the  first  viewpoint  is  accepted,  it  may  be  logically  con- \ncluded, although  not  proved,  that  rents  are  a  better  index  of  tele- \nphone market  than  are  incomes.  Incomes,  as  measured  in  money, \nare  the  nearest  approach  which  may  be  made  to  a  measure  of  the \nposition  of  families  on  an  imaginary  scale  of  economic  welfare.  An \nattempt  to  translate  rent  data  to  an  income  basis,  as  a  working \nmethod  in  commercial  surveys  would  introduce  errors  with  no  com- \npensating advantage,  but  a  translation  of  this  kind  is  more  or  less \nunconsciously  made  in  making  comparisons.\n\nSources  of  Information.  Much  of  the  literature  on  the  question  of \nhouse  rents  versus  incomes  is  generalization  based  on  limited  or \nantiquated  data.  Such  careless  statements  as  \"rent  approximates \nabout  one-third  of  the  average  worker's  income,\"  may  be  found  in \nthe  literature  of  the  subject.  Adam  Smith,  the  father  of  Political \nEconomy,  \"made  the  assertion,  surprising  to  us  in  these  days,  that \nthe  proportion  of  income  spent  in  house  rent  is  highest  among  the \nrich.\"  Frederick  Engels  concluded  in  1857  that  rent  was  12  per \ncent  and  heat  and  light  5  per  cent  of  the  workingman's  expenditure, \nregardless  of  the  amount  of  his  income.\n\nInvestigation  of  budgets  in  recent  years  has  been  confined  almost \nentirely  to  the  field  of  the  wage  earning  class.  The  first  really  com- \nprehensive study  was  made  by  the  United  States  Bureau  of  Labor\n\nStatistics  in  several  states  in  1901  and  1902,  and  is  detailed  in  the \n1903  annual  report  of  that  organization.  It  consisted  chiefly  of  a \nstudy  of  11,156  so-called  \"normal\"  families,  each  including  a  husband \nat  work,  a  wife,  not  more  than  5  children  all  under  14  years  and  no \nlodgers  or  servants.  The  average  income  of  these  families  was  S651. \nOriginal  work  on  a  smaller  scale  has  been  done  by  R.  C.  Chapin \n(1908)  and  by  the  Philadelphia  Bureau  of  Municipal  Research  (1918). \nThe  Bureau  of  Labor  Statistics  collected  a  large  amount  of  data  in \n1918  and  1919^  concerning  the  incomes  and  expenses  of  12,837  families \nin  92  towns  having  an  average  income  of  $1491.  This  investigation \nincluded  families  of  wage  earners  and  low  salaried  men,  but  none  of \nthe  slum  or  recent  immigrant  classes.  Families  of  the  lowest  type \nare  automatically  excluded  from  such  studies  as  this  by  their  inability \nto  supply  the  desired  information  from  accounts  or  from  intelligent \nestimates.\n\nDistribution  of  Family  Expenses.  Representative  distributions  of \nfamily  expenditures  are  given  in  Table  I.  The  National  Industrial \nConference  Board  has  adopted  for  use  in  computing  their  cost  of \nliving  index  and  representative  budgets  a  list  of  standard  weights \nmade  by  combining  the  results  of  a  number  of  studies  made  from \n1901  to  1917.  Most  importance  was  assigned  to  the  first  Bureau  of \nLabor  Statistics  study,  the  results  of  which  it  closely  resembles. \nThe  standard  weights  used  by  the  Bureau  of  Labor  Statisics  are  the \nresult  of  surveys  made  in  22  cities  from  July  31  to  November  30,  1918, \ncovering  families  whose  average  income  was  $1,434.\n\nFrom  Table  I  it  may  be  inferred  that  the  per  cent  spent  for  rent \nis  reduced  in  a  period  of  inflated  prices,  at  least  during  the  first  part \n^  Monthly  Labor  Review,  May-December,  incl.,  1919.\n\nof  that  period.  This  is  reasonable,  since  rents  respond  less  rapidly \nthan  most  other  prices  to  fluctuations  in  the  general  price  level.  An \nextreme  example  of  this  type  is  found  in  Germany  where  rents, \nwhich  are  to  some  extent  under  government  regulation,  \"at  the \npresent  time  absorb  not  more  than  3>^  per  cent  of  total  expenditure \nas  against  20  per  cent  before  the  war.\"  ^\n\nThe  percentage  distribution  of  total  expenses  depends  on  the  size \nof  the  family,  the  income  received  and  the  city  lived  in.  Of  course, \nit  must  be  understood  that  any  particular  family  may  differ  widely \nfrom  general  averages.  Other  things  being  equal,  large  families  spend \nmore  for  food  and  clothing  and  less  for  rent  and  sundries  than  do \nsmall  families.  Large  families  of  the  lower  middle  class  accommodate \nthemselves  to  whatever  housing  accommodations  they  can  afford \nafter  the  more  inflexible  demands  for  other  things  have  been  provided \nfor.  Less  than  one  room  per  person  is  considered  over-crowding \nand  the  recent  Bureau  of  Labor  Statistics  investigation  found  this \ncondition  to  exist  rarely,  except  in  families  having  more  than  three \nchildren.  Families  with  one  to  three  children  were  found  to  have \nLO  to  L3  rooms  per  person  in  almost  all  cities.\n\nAmount  of  Income  vs.  Per  Cent  Spent  for  Rent.  The  extent  to  which \nthe  distribution  of  expenses  is  modified  by  the  amount  of  income \nreceived  is  known  only  within  the  very  limited  range  for  which  data \nare  available.  The  best  recent  figures  are  those  of  the  1918-1919 \nstudy  of  the  Bureau  of  Labor  Statistics.  These  are  given  here  for \n12,096  white  families  in  92  cities  and  towns:\n\nWhen  the  original  data  are  examined  in  detail,  it  appears  that  in \nalmost  every  city  as  incomes  increase  the  per  cent  spent  for  rent \n^  M.  Elsas.  Economic  Journal,  September,  1921,  p.  332.\n\nand  fcHxl  decreases  and  the  per  cent  for  clothing  increases.  The \ndecrease  in  the  per  cent  for  food  as  incomes  increase  is  slight  and  the \nincrease  in  the  per  cent  for  clothing  is  especially  marked  in  the  higher \nincomes  within  the  range  covered.  Thus,  it  appears  that  among \nfamilies  of  moderate  incomes  as  incomes  rise  the  increase  is  spent \nby  preference  for  clothing  rather  than  for  food  or  rent.  The  relative \ndecrease  in  expenditure  for  rent  as  incomes  increase  is  significant  in \nrental  analysis.  This  means  that  while  a  10  per  cent  difference  in \nrents  among  the  lower  rents  in  a  city  indicates  an  average  difference \nin  income  of  about  10  per  cent,  a  similar  difTerence  among  the  higher \nrents  indicates  a  difference  in  income  of  much  more  than  10  per  cent.\n\nRent  Levels  in  Various  Cities.  As  nearly  as  may  be  determined \nfrom  the  Bureau  of  Labor  Statistics  data,  there  is  no  regular  tendency \nfor  Eastern,  Western  or  Southern  cities  to  differ  from  the  average \nof  all  cities,  either  in  the  amount  of  wage-earners'  incomes  or  in \namounts  spent  for  food  or  clothing.  In  Southern  cities  somewhat \nless  is  paid  for  rent  than  in  other  cities.  This  refers  only  to  w^hite \nfamilies.  Negro  families  have  smaller  average  incomes  than  white \nfamilies  and  at  any  given  income  they  spend  less  for  rent,  more  for \nfood  and,  to  a  less  degree,  more  for  clothing  than  white  families. \nThe  size  of  a  city,  so  far  as  may  be  told  from  these  data,  does  not \ndetermine  either  total  incomes  or  expense  for  food,  rent  or  clothing.\n\nIn  different  cities  the  difference  in  rent  levels,  that  is,  variation  in \nrentals  paid  for  substantially  similar  dwellings,  is  considerably  greater \nthan  the  difference  in  levels  of  prices  for  food  or  clothing.  The  varia- \ntion in  price  levels  is  about  twice  as  great  for  rents  as  for  food  or \nclothing,  reckoned  as  percentages  of  the  amounts  spent  for  each  class. \nThe  expenditure  for  food  by  the  lower  middle  class  families  included \nin  this  investigation  is  more  nearly  the  same  in  different  cities  than \nis  the  expenditure  for  rent  or  for  clothing.  Food  expense  is  the  only \none  of  these  classes  in  which  all  cities  are  as  closely  grouped  as  in \ntotal  expenses,  considering  deviations  from  the  averages  on  a  per- \ncentage basis.  The  amounts  spent  for  rent  show  relatively  wide \nvariation  between  cities.  It  appears  that  if  a  workman  moves  from \none  city  to  another  to  secure  increased  wages  a  large  proportion  of \nthe  increase  in  income  goes  for  increased  rent.  This  is  to  be  expected \nsince  land  rents  and,  to  a  less  extent,  construction  costs  are  peculiar \nto  each  individual  city,  much  more  than  food  or  clothing  costs.\n\nA  comparison  of  rent  data  from  the  1918-1919  investigation  of  the \nBureau  of  Labor  Statistics  and  data  from  Commercial  Surveys  leads \nto  the  conclusion  that  differences  in  average  rents  in  various  cities \nare  due  at  least  as  much  to  differences  in  the  level  of  prices  for  rents\n\nas  to  differences  in  the  grade  of  the  population.  Wage-earners  and \nlow  salaried  people  of  the  types  studied  by  the  Bureau  of  Labor \nStatistics  occupy  about  the  same  position  in  the  community  in  a \nlarge  number  of  cities.  As  a  rule  they  pay  about  80  to  90  per  cent \nof  the  median^  rent  in  any  city.  Exception  must  be  made  in  the \ncase  of  cities  having  an  unusually  large  proportion  of  negro  or  very \nlow  grade  white  population.  It  is  interesting  in  this  connection  to \ncompare  wage  rates  for  different  classes  of  labor  in  various  cities. \nThe  variation  between  cities  in  wage  rates  for  common  labor  is  pro- \nportionately much  greater  than  the  variation  in  wages  for  work \nrequiring  some  skill,  such  as  bricklaying  and  structural  iron  work.\n\nAs  examples  of  the  impossibility  of  accurately  rating  the  grade \nof  a  city's  population  by  its  median  rent  alone,  we  may  take  four \ncities  where  surveys  were  made  in  1921.  Spokane  and  Houston  had \npractically  identical  median  rents  of  $23.00  and  $23.40  respectively, \nbut  Houston  is  not  as  good  a  telephone  market  as  Spokane.  In \nCleveland  and  Minneapolis  the  median  rents  were  found  to  be  $35.50 \nand  $31.00  respectively,  but  this  is  no  measure  of  the  grades  of  the \ntwo  cities.\n\nRent  Data  from  Various  Sources,  Including  England.  Some  addi- \ntional rent  data  is  presented  here  without  extended  comment.  The \ntwo  following  tables  show  the  proportion  which  rent  bears  to  total \nexpense  in  different  communities.\n\nThe  first  four  of  these  \"standard\"  budgets  are  by  the  National \nIndustrial  Conference  Board,  and  the  others  in  order  b\\-  the  Bureau \nof  Municipal  Research  (Philadelphia),  Professor  W.  F.  Ogburn, \nand  the  U.  S.  Bureau  of  Labor  Statistics.\n\nIt  has  already  been  mentioned  that  the  per  cent  spent  for  rent \nshows  a  tendency  to  decrease  with  increasing  incomes.  This  trend \nis  confined  by  data  from  other  sources,  as  follows:\n\nAlthough  no  data  are  av'ailable  for  families  above  the  lower  middle \nclass,  the  relationship  may  be  extended  by  conjecture  into  the  higher \nincome  levels.  That  this  is  reasonable  is  brought  out  in  subsequent \npages  in  comparing  distribution  curves  of  rental  values  and  incomes.\n\nSome  interesting  conclusions  from  English  experience  are  given \nby  Sir  J.  C.  Stamp. ^  The  rent  corresponding  to  an  income  of  \u00a3160 \naverages  at  least  \u00a35  greater  in  London  than  outside  that  city.  Among \nthe  lower  incomes,  say  up  to  \u00a31000,  the  variation  or  dispersion  of  the \npercentages  paid  for  rent  becomes  less  as  the  amount  of  the  income \nrises.  Owner-occupants  live  in  larger  houses  than  tenants  with  the \nsame  total  income.  It  was  not  found,  as  is  generally  supposed,  that \nprofessional  men  pay  relati\\'ely  more  for  rent  than  business  men.  The \nfollowing  table  is  from  the  work  abo\\e  mentioned ;\n\nThe   second   of   these   two   tables   represents  average   conditions   for \nGreat  Britain.\n\nIn  making  comparisons  of  survey  rent  data  it  is  desirable  to  dis- \ntinguish differences  in  price  levels  as  they  afTect  rents,  difTerences \nin  economic  grade  of  the  population,  and  differences  in  the  distri- \nbution of  families  about  their  average  grade.  Failure  to  take  account \nof  these  three  factors  will  result  in  misleading  impressions,  which  may \nbe  illustrated  by  summaries  from  successive  surveys  in  Atlanta. \nThe  following  table  shows  composites  of  private  residences,  flats  and \napartments:\n\nIt  might  be  inferred  from  this  set-up  that  the  condition  of  the \npoorer  families  had  been  very  much  improved  or  that  the  average \nfamily  had  attained  a  higher  condition  of  well-being.  It  will  be \nshown  later  that  there  was  no  material  change  in  the  distribution  of \nrental  values  about  an  average  rent  when  rents  are  considered  as \npercentages  of  that  average,  and   it  is  probable  that  the  principal.\n\nif  not  the  sole  cause  of  the  changes  shown  in  the  table  above,  is  the \ngeneral  rise  in  the  le\\'el  of  prices.\n\nMethods  of  Study \u2014 Graphic  Representation.  The  most  convenient \nand  practical  method  of  studying  rent  distributions  is  by  the  use \nof  graphs  and  charts.  The  distribution  of  values  of  renjs  or  of  other \nvariables  may  be  charted  in  either  a  detail  or  a  cumulative  form.  A \ndetail  curve  shows  at  any  value  of  the  independent  variable  the \nfrequency  of  occurrence  of  items  of  that  value.  A  cumulative  curve \nshows  at  any  value  of  the  variable  the  number  (or  better,  the  per \ncent)  of  all  cases  which  have  values  below  (or  above)  that  value. \nCumulative  curves  are  better  than  detail  curves  for  presenting  rent \ndata  since  the  number  of  classes  into  which  the  data  are  divided  is \nsmall  and   the  class  widths  are  non-uniform,   resulting  in   uncertain\n\ncurves  of  the  detail  type.'^  Attempts  to  compare  rent  distributions \nby  application  of  the  usual  statistical  measures  of  dispersion  and \nskewness  have  proved  unsatisfactory.\n\nTypical  rent  distributions  plotted  in  the  cumulative  manner  on \nordinary  coordinate  paper  are  shown  in  Fig.  1.  Diagrams  of  this \ntype  may  be  used  to  determine  the  rent  paid  by  families  of  correspond- \ning position  in  the  rent  scale  at  the  dates  of  successive  surveys,  but \nthey  do  not  give  a  very  clear  picture  of  changes  in  the  distribution  of \nrents  and  from  them  it  is  not  readily  apparent  whether  rents  are \nclosely  concentrated  or  widely  distributed  in  any  given  case.\n\n^  Detail  curves  for  rent  and  income  distributions  are  most  easily  drawn  on  paper \nwith  a  logarithmic  scale  both  ways.\n\nLogarithmic  Probability  Charts.  Cumulative  curves  for  rent  dis- \ntributions may  be  plotted  on  logarithmic  probability  paper\u00ae  in  which \ncase  the  resulting  graph  is  a  straight  line  for  a  large  number  of  cities. \nSuch  a  graph  will  be  said  to  represent  a  logarithmic  skew  distribution. \nIn  the  appendix  there  is  given  a  discussion  of  frequency  curves, \nwith  special  reference  to  curves  of  this  type.  The  essential  point \nin  reading  charts  on  logarithmic  probability  paper  is  that  the  slope \nof  the  line  determines  both  the  spread  or  dispersion  of  the  data  and \nthe  skewness  or  lack  of  symmetry  of  distribution.  Since  the  hori- \nzontal scale  is  logarithmic  it  follows  that  the  dispersion  is  represented \non  a  percentage  and  not  a  linear  basis.  A  steep  slope  indicates  a \nclose  concentration  of  the  data,  a  less  steep  slope  indicates  a  wider \ndistribution,  and  parallel  lines  indicate  distributions  which  are  identical \non  a  percentage  basis.  As  explained  in  the  appendix,  the  most  con- \nvenient index  or  coefficient  for  expressing  the  spread  or  dispersion \nof  a  distribution  is  the  ratio  of  the  upper  quartile'^  to  the  median  rent. \nIf  the  curves  for  a  given  city  are  closely  parallel  for  successiv'e  surveys \nit  follows  that  there  has  been  no  material  change  in  the  character  of \nthe  distribution.  In  other  words,  rents  have  increased  approximately \nproportionately  at  all  points  of  the  scale.\n\nExamples  of  charts  of  this  kind  (Figs.  2-4)  are  shown  for  twelve \ncities  for  which  successive  surveys  are  available.  Curves  for  suc- \ncessive surveys  are  nearly  parallel  in  eight  of  the  twelve  cities.  For \nCleveland,  Dallas  and  Houston  there  are  distinct  differences  in  the \ncurves  for  the  two  dates,  indicating  changes  in  the  distribution  of \nrents,  which  changes  may  be  measured  since  horizontal  distances \nbetween  points  on  the  curves  for  two  dates  represent  the  percentage \nincreases  in  rents.\n\nWhen  a  rent  distribution  is  plotted  on  logarithmic  probability \npaper  the  points  do  not  always  lie  on  a  straight  line,  but  a  straight \nline  of  best  fit  may  be  chosen  by  eye,  giving  greatest  weight  to  points \nnear  the  middle  of  the  scale  of  ordinates.  Of  the  rent  distributions \nfor  large  cities  which  have  been  plotted  on  this  paper,  nearly  one- \nthird  are  very  closely  represented  by  straight  lines,  an  equal  number \nare  slightly  concave  upward,  and  the  remainder  are  more  or  less \nconcave  downward.  Most  of  the  deviations  from  straight  lines  are \nslight.  The  examples  submitted  herewith  (cities  in  which  successive \nsurveys  have  been  made)  are  rather  poorer  than  the  average  in  this\n\n\"See  an  article  by  G.  C.  Whipple  in  the  Journal  of  the  Franklin  Institute  for  July \najid  August,  1916,  for  a  description  of  this  paper  and  some  examples  of  its  use  in  the \nfield  of  sanitation.\n\nrespect.  Concavity  upwards  represents  a  distribution  which  is  less \nskew  than  the  theoretical  logarithmic  skew  curve,  and  concavity \ndownwards  a  distribution  of  greater  skewness.  No  great  importance \ncan  be  assigned  to  small  differences  of  this  sort,  as  they  are  not  perma- \nnent betw^een  successive  surveys,  whereas  the  general  type  of  the \ndistribution  is  quite  constant  for  a  given  city.\n\nThe  justification  for  assuming  that  rents  follow  the  logarithmic \nskew  curve  is  made  stronger  by  certain  data  from  Volume  19  of  the \nreport  of  U.  S.  Immigration  Commission  made  in  1912.  This  com- \nmission collected  a  large  mass  of  data  concerning  the  living  conditions \nof  families  of  the  immigrant  type.  The  data  are  classified  by  nation- \nality of  head  of  family,  by  income,  etc.  Distributions  of  amounts \npaid  for  house  rent  per  apartment,  per  room  and  per  person  for  certain \nnationalities  are  shown  in  Fig.  5.  The  data  shown  were  chosen  from \nthose  classes  which  were  made  up  of  the  largest  numbers,  and  the \ndeviations  from  straight  lines  shown  by  data  for  other  groups  are  in \nboth  directions,  so  the  straight  line  relation  may  be  considered  fairly \nrepresentative.  It  may  be  noted  that  the  rents  per  month  per  person \nshow  a  greater  dispersion  than  the  rents  per  room  or  per  apartment. \nThese  latter  moreover  show  as  small  a  spread  as  do  rents  for  any  of \nthe  cities  studied  as  a  whole.\n\nFig.  5  also  shows  the  distribution  of  British  house  rents  at  intervals \nduring  the  period  1890-1913.  There  has  been  a  gradual  but  steady \ndecrease  in  the  dispersion  of  rents  during  the  period  covered.  Unless \nthe  relation  between  rents  and  incomes  has  radically  changed,  this \nmeans  that  the  inequality  of  distribution  of  wealth  has  been  decreased, \nand  that  the  condition  of  the  poor  has  been  improved  as  compared \nto  that  of  the  rich.  Data  for  1830  indicate  that  the  inequality  of \ndistribution  was  distinctly  greater  at  that  date  than  in  1890.  Changes \nin  the  relative  condition  of  the  rich  and  poor  may  be  readily  demon- \nstrated by  charts  of  this  kind,  but  of  course  conclusions  regarding \nabsolute  degrees  of  well-being  must  be  reached  by  other  means.\n\nDistribution  of  Rents  Compared  with  that  of  Incomes.  Significant \nconclusions  regarding  the  relation  between  rents  and  incomes  may  be \ndrawn  from  a  comparison  of  their  respective  distributions.  Fig.  6 \nshows  a  detail  curve  for  income  distribution  in  the  United  States \nbased  on  preliminary  data  of  the  National  Bureau  of  Economic \nResearch.  These  data  are  subject  to  revision  but  are  the  best  avail- \nable and  are  sufficiently  accurate  for  comparative  purposes.  The \nusual  way  to  chart  income  distribution  assumes  conformity  with \nPareto's  law  which  says  that  the  frequency  curve  of  incomes  may \nbe  plotted  as  a  straight  line  on  double  logarithmic  paper,  either  on  a\n\ndetail  or  cumulative  basis.  This  law  does  not  hold  for  the  lower \nincome  levels  which  may  be  best  represented  by  a  curve  of  approxi- \nmately hyperbolic  form,  as  shown  in  Fig.  6.  The  same  income  data \nare  shown  plotted  on  logarithmic  probability  paper  in  an  insert  on\n\nBased  on  Preliminary  Data  from  the \nNational  Bureau  of  Economic  Resrarch\n\nFig.  6.  From  the  form  of  this  curve  it  may  be  concluded  that  the \ndistribution  of  the  lower  two-thirds  of  both  incomes  and  rents  is \nsimilar  but  that  the  spread  of  the  higher  incomes  is  much  greater \nthan  the  spread  of  the  corresponding  rents.\n\nAnother  comparison  of  incomes  and  rents  may  be  made  from  a \nsecond  insert  on  the  same  chart.  It  may  be  readily  demonstrated \nthat  the  normal  curve  of  error  plots  as  a  parabola  on  semi-logarithmic \npaper  and  the  logarithmic  skew  curve  as  a  parabola  on  double  logar- \nithmic paper.  Two  parabolas  which  represent  extreme  conditions  of \nspread  and  of  concentration  of  rents  in  large  cities  are  shown.  If \nthe  degree  of  dispersion  remains  fixed  a  change  in  the  rent  level \nmerely  shifts  the  parabola  on  the  chart  without  changing  its  shape. \nThe  parabolic  shape  of  rent  curves  and  the  hyperbolic  shape  of  the \nincome  curve  indicate  that  rents  are  somewhat  less  concentrated \nlocally  about  their  mode,^  but  are  more  concentrated  as  an  entire \ngroup  than  are  incomes.  These  curves  can  not  well  be  superposed \nfor  comparison  since  areas  are  not  equivalent  on  different  parts  of \nthe  chart.  Those  incomes  which  are  closely  grouped  around  the \nmode  represent  wage-earners  of  such  a  type  that  several  may  come \nfrom  a  single  family.  Conclusions  regarding  comparison  of  incomes \nand  rents  must  be  made  with  caution  since  rents  are  on  a  family \nbasis  and  incomes  on  an  individual  basis.  No  satisfactory  data \nare  available  to  show  the  variation  in  income  distribution  between \nsmall  subdivisions  of  the  United  States,  as  cities,  but  it  is  reasonable \nto  assume  that  there  is  some  such  variation  for  incomes  as  well  as \nfor  rents,  although  perhaps  not  of  so  great  a  range.\n\nA  third  comparison  of  incomes  and  rents  is  made  possible  by  the \nuse  of  Lorenz  curves  illustrated  in  Fig.  7.     On  this  form  of  chart  a \n*  See  Appendix.\n\ndiagonal  line  at  45  represents  a  uniform  distribution  and  the  further \na  given  curve  falls  to  the  right  of  and  below  that  line  the  more  unequal \nthe  distribution  represented.  If  incomes  were  plotted  on  a  family \nbasis,  the  resulting  curve  would  lie  somewhat  closer  to  the  diagonal \nline  than  the  one  shown,  but  it  is  fairly  evident  that  incomes  are \nmore  unequally  distributed  than  rents.  For  instance,  the  top  10 \nper  cent  of  incomes  are  on  the  average  about  42  per  cent  of  the  total \nincome,  while  the  top  10  per  cent  of  rents  are  from  22  to  32  per  cent  of \nthe  aggregate  rent  in  most  cities.  These  three  comparisons  confirm \nthe  idea  discussed  in  the  first  part  of  this  paper,  that  the  proportion \nof  income  spent  for  rent  is  less  among  the  larger  incomes.\n\nOf  the  extensive  data  on  income  distributions  few  can  well  be  used \nfor  comparison  with  rent  data.  In  order  that  a  cumulative  curve \nreally  mean  anything,  it  must  represent  an  entire  group,  not  merely \nitems  from  one  end  or  the  other  of  the  complete  scale.  Therefore,  the \nvarious  tables  of  earnings  of  working  class  families  and  individuals \nare  of  doubtful  use  here,  although  they  do  show,  plotted  on  double \nlogarithmic  or  logarithmic  probability  paper,  that  the  type  of  dis- \ntribution of  earnings  about  an  average  value  is  practically  identical  for \nvarious  nationalities  in  similar  industries,  or  for  men,  women  and \nchildren  in  all  industries.  However,  the  average  earnings  of  the \nvarious  classes  are  widely  different.  A  few  examples  are  shown  in \nFig.  5.\n\nIncome  tax  returns  are  of  some  interest  although  they  are  defective \nin  several  respects:  they  only  include  the  upper  part  of  society,  a \nlarge  number  of  persons  fail  to  make  returns  and  large  amounts  of \nincome  are  tax  exempt.  Federal  income  tax  data,  which  are  available \non  a  uniform  basis  for  the  years  1917-1920,  may  best  be  studied  by \nplotting  on  double  logarithmic  paper,  preferably  after  reducing  the \nfigures  for  the  various  states  to  a  basis  of  returns  per  1000  popula- \ntion. There  are  small  changes  from  year  to  year  in  the  position  of \nthe  curve  for  any  given  state,  which  are  not  significant,  since  they \nmay  be  due  either  to  changes  in  the  average  income,  or  to  increased \nefficiency  of  tax  collection.  Changes  from  year  to  year  in  the  slope \nof  the  curve  for  any  one  state  are  small,  indicating  that  there  exists \nin  each  state  a  definite  type  of  distribution  of  wealth  and  earning \npower.  Differences  in  the  position  and  slope  of  the  curves  for  dif- \nferent, states  are  conspicuous,  indicating  that  both  the  per  capita \nincome  and  the  distribution  of  the  total  income  among  individuals \nare  different  in  different  states.  New  York,  for  instance,  shows  a \nwide  spread,  i.e.,  a  relatively  large  number  of  very  high  incomes,  and \nIowa  shows  a  narrow  spread,  i.e.,  a  large  number  of  incomes  around\n\n$2000-5000,  and  comparatively  few  incomes  over  $20,000.  New \nJersey  occupies  an  intermediate  position.  Alabama,  more  or  less \ntypical  of  Southern  states,  shows  a  much  smaller  number  of  returns \nin  proportion  to  population  than  any  of  these  states,  and  a  distribu- \ntion nearly,  but  not  quite,  as  closely  grouped  as  Iowa.  The  fact \nthat  a  particular  shape  of  curve  is  typical  of  a  given  state,  and  that \nthe  curves  are  different  for  different  states,  corresponds  to  similar \ncharacteristics  of  rent  curves  for  cities.  British  income  statistics \nshow  about  the  same  degree  of  dispersion  as  do  returns  for  the  United \nStates  as  a  whole.\n\nDistributions  of  the  logarithmic  skew  type  may  be  found  in  other \nfields  than  those  of  incomes  and  house  rents.  The  theory  has  been \nadvanced  by  some  statisticians  that  while  the  normal  curve  of  error \nis  characteristic  of  observational  errors,  errors  of  estimate  agree  with \nthat  law  if  logarithms  rather  than  actual  estimates  be  considered. \nPrice  fluctuations,  corporation  earnings,  and  the  profits  of  farmers \nare  distributed  in  a  similar  manner.  The  lengths  of  life  of  telephone \ncontracts  agree  quite  closely  with  this  type  of  distribution,  if  we \nallow  for  the  fact  that  very  long  lives  are  relatively  few  in  number \nbecause  they  started  when  the  telephone  business  was  comparatively \nsmall.  A  peculiarity  of  rent  distribution  is  that  if  we  choose  only \nfamilies  having  telephone  service,  or  families  having  any  one  class \nof  service,  we  obtain  a  logarithmic  skew  distribution  about  as  closely \nas  though  we  plotted  all  families  in  a  city.\n\nApplication  to  Survey  Data.  The  charting  method  described  above \nwas  applied  to  rent  data  for  57  cities,  both  for  composites  of  private \nresidences,  flats  and  apartments  and  for  private  residences  alone. \nTable  VIII  gives  the  median  rents  and  values  of  the  rent  dispersion \nindex  Q^  for  the  composite  data.  Results  for  private  residences \ndiffer  in  most  cases  only  very  slightly  from  the  results  given;  there \nis  no  dominant  tendency  for  the  spread  of  private  residence  rents  to \nbe  greater  or  less  than  that  of  all  rents  in  a  city,  but  the  median  rent \nin  private  residences  is  usually  somewhat  greater  than  that  for  the \ncomposite.\n\nAn  effort  was  made  to  determine  the  significance  of  the  various \nvalues  of  the  index  Q,  but  the  results  are  chiefly  negative.  There \nis  some  tendency  for  the  smaller  cities  to  have  a  wide  spread  of  rent \nvalues;  i.e.,  a  high  value  for  Q,  but  there  is  considerable  scattering \nof  the  data.  This  tendency  is  most  apparent  in  the  South,  where  the \nsmaller  cities  have  extremely  high  values  for  Q.  The  relationship \nbetween  the  index  Q  and  the  per  cent  of  families  with  telephone \n\u2022  See  Appendix  for  a  quantitative  definition  of  Q.\n\nservice  is  not  very  well  defined.  Cities  with  a  very  high  residence \ndevelopment  have  low  values  for  the  index  and  Southern  cities  with \npoor  residence  development  have  high  values  for  Q,  but  the  inter- \nmediate scattering  of  data  is  quite  wide.  It  might  be  supposed \nthat  cities  with  high  values  for  Q,  which  indicate  a  wide  spread  of \nsocial  strata,  would  have  a  relatively  large  number  of  business  firms, \neither  total  or  retail,  to  meet  the  widely  divergent  needs  of  the  popula- \ntion. As  a  matteg-  of  fact,  no  such  relationship  is  apparent.  There  is, \nhowever,  positive  correlation  between  Q  and  the  proportion  of  insti- \ntutions to  population.  This  may  be  due  in  part  to  the  fact  that  high \nvalues  of  Q  are  found  in  Southern  cities  which  have  separate  churches \nand  schools  for  whites  and  negroes.\n\nAlthough  no  special  significance  has  been  found  for  the  particular \ndegree  of  rent  dispersion  found  in  any  city,  some  interest  attaches  to \nthe  fact  that  this  index  remains  practically  constant  in  a  given  city, \nregardless  of  changes  in  the  level  of  prices.  The  diagrams  illustrating \nthis  point  have  already  been  discussed.  If  the  type  of  distribution \nis  not  found  constant  in  a  particular  city,  it  would  seem  probable \nthat  a  change  in  character  of  the  population  is  taking  place,  but  a \nchange  in  the  average  economic  grade  might  occur  without  any \nchange  in  the  type  of  distribution.  When  two  distributions,  each  of \nwhich  agrees  with  a  logarithmic  skew  curve,  are  added  together, \nthe  new  combined  distribution  may  be  represented  by  another  loga- \nrithmic skew  curve  only  in  case  both  the  medians  and  coefficients  of \ndispersion  for  the  two  original  curves  are  identical.  It  follows  that \nif  the  index  of  rent  dispersion  in  a  city  is  found  to  be  the  same  in  suc- \ncessive surveys  and  if  it  may  be  assumed  to  have  remained  constant \nduring  the  interval,  then  the  new  families  which  have  come  into  a  city \nat  any  time  comprise  a  group  having  substantially  the  same  coefficient \nof  dispersion  and  median  rent  as  the  families  which  made  up  the  original \npopulation.  The  apparent  permanence  of  the  type  of  rent  distri- \nbution in  a  city  may  be  considered,  along  with  the  telephone  habit,  as \na  reasonable  explanation  of  the  rather  high  degree  of  stability  of \nstation  distribution  by  classes  of  service  among  residence  subscribers.\n\nIn  commercial  survey  work  a  city  is  divided  into  market  areas, \nknown  also  as  homogeneous  sections,  which  are  so  laid  out  that  in  any \none  section  the  families  at  any  stated  rent  are  similar  telephone \nprospects.  A  study  of  rent  distributions  in  market  areas  was  carried \nout  in  a  number  of  cities,  considering  only  those  market  areas  in  each \ncity  which  had  fairly  large  populations.  There  appears  to  be  no \nrelationship  between  the  index  of  rent  dispersion  and  either  the  median \nrent,  the  per  cent  of  families  in  private  residences,  or  the  per  cent\n\nof  families  with  telcphoiu\"  service,  in  the  various  areas  in  any  one \ncity.  Whether  a  particular  area  is  suburban  or  downtown,  likewise \nhas  no  apparent  effect  on  the  value  oi  Q.  It  was  found  in  Atlanta, \nwhere  the  di\\ision  of  the  city  into  market  areas  was  substantially \nthe  same  in  successiv'e  surveys,  that  the  distribution  index  which  had \npreviously  been  found  to  be  stable  for  cities  as  a  whole,  behaved  in \nthe  same  way  in  separate  sections  of  the  city.  It  was  found  that  the \nrent  distribution  index  for  any  single  market  area  is  smaller,  usually \nmuch  smaller,  than  the  index  for  the  entire  city  in  which  the  area \nis  located.  One  section  in  Atlanta  is  the  only  exception  found  to \nthis  rule.  In  market  areas  it  was  noted  that  a  considerable  number \nof  the  graphs  on  logarithmic  probability  paper  were  formed  of  two \nintersecting  straight  lines.  This  indicates  that  the  sections  are  not \nreally  homogeneous,  but  contain  elements  of  population  radically \ndifferent  in  character.  This  condition  can  not  be  obviated  by  the \nmost  careful  laying  out  of  section  boundaries  in  case  there  exists  a \nmixture  of  families  oi  essentially  different  types,  as  when  negro \nresidences  are  scattered  among  a  predominantly  white  population.\n\nIndices  of  Rent  Distribution  in  Large  Cities \nComposites   of   Private    Residences,    Fiats   and   Apartments\n\nFrequency  curves  may  be  symmetrical  or  skew.  The  particular \nsymmetrical  distribution  known  as  the  normal  curve  of  error  is  typical \nof  distributions  of  observational  errors  and  in  general  of  all  phenomena \nobeying  the  laws  of  chance.  It  is  approximated  by  a  number  of  other \ndistributions  which  have  not  obviously  originated  in  the  same  way, \nwhich  implies  that  \"the  variable  is  the  sum  of  a  large  number  of \nelements  each  of  which  can  take  the  values  0  and  1,  these  values \noccurring  independently  and  with  equal  frequency.\"  Skew  distri- \nbutions may  take  a  variety  of  forms  but  the  type  shown  in  the  dia- \ngram is  closely  approached  by  a  large  number  of  rent  distributions. \nThe  essential  characteristic  of  this  curve,  which  may  be  called  the \nlogarithmic  skew  curve,  is  that  logarithms  of  the  values  of  the  variable \nare  distributed  according  to  the  normal  curve  of  error.  This  skew \ncurve  is  of  course  not  the  only  one  which  might  be  selected  to  repre- \nsent rent  data,  but  it  presents  the  fewest  mathematical  difficulties \nand  gives  a  sufficiently  close  approximation  for  all  practical  purposes.\n\nwhere  e  is  2.7183,  the  base  of  natural  logarithms  and  a  is  a  measure \nof  dispersion,  known  as  the  standard  deviation.  An  ordinate  of \nthe  curve  is  called  the  frequency,  and  expresses  the  fraction  of  the \nwhole  number  of  items  which  occurs  per  unit  interval  of  the  variable  x.\n\nSubstituting  log  X  for  x,  a  new  equation  may  be  obtained,  in  which \ny  is  the  frequency  per  unit  of  x  (or  log  X).  An  expression  for  the \nfrequency  per  unit  of  X  is  desired,  which  may  be  called  F.  It  may  be \nshown  that  the  desired  equation  is\n\nThis  is  the  equation  of  what  we  have  called  above  the  logarithmic \nskew  curve,  which  is  really  not  a  curve  of  error  in  the  same  sense  as \nequation  (1)  is.\n\nIn  the  course  of  this  discussion  it  will  be  convenient  to  refer  to \ncertain  features  of  the  frequency  curves  by  the  accustomed  terminol- \nogy of  statistics.  The  median  item  of  a  group  is  such  that  one-half \nof  all  the  items  are  larger,  and  one-half  are  smaller,  and  is  the  central \nitem  when  they  are  arrayed  in  order  of  size.  The  quartiles,  upper  and \nlower,  together  with  the  median,  divide  the  array  into  four  parts, \neach  containing  one-fourth  of  the  items.  The  percentiles  divide  the \narray  into  100  equal  parts.  The  mode  is  that  value  of  the  variable \nwhich   is  of  most   frequent   occurrence.\n\nIn  the  normal  curve  of  error  a,  the  standard  deviation,  is  tech- \nnically defined  as  the  square  root  of  the  mean  of  the  squares  of  the \ndeviations  of  the  items  from  their  mean.  For  present  purposes  it \nmay  be  regarded  as  a  measure  of  dispersion  approximately  equal  to \nthe  difference  between  the  values  of  x  at  the  84th  percentile  and  the \nmedian.  In  the  logarithmic  skew  curve  a  is  the  difference  between \nthe  corresponding  logarithms.\n\nThe  origin  of  x  in  the  normal  curve  of  error  is  the  arithmetic  mean, \nmedian,  or  mode,  which  are  coincident.  When  a  logarithmic  scale \nof  abscissas  is  introduced,  the  median  value  of  x  (or  log  X)  corre- \nsponds to  the  median  value  of  X,  which  is  smaller  than  the  mean \nvalue  of  X,  and  larger  than  the  mode.  In  a  logarithmic  skew  curve \nthe  median  may  be  considered  the  origin,  and  at  this  point  x  (or \nlog  X)  is  equal  to  zero,  and  X  is  equal  to  unity.  When  this  curve \nis  applied  to  house  rents  the  median  rent  occurs  at  this  point.  The \nrelation  between  rents  and  values  of  X  is  a  simple  one.  If  rents  be \ndenoted  by  R,  and  if  M  be  the  median  rent,  then\n\nThe  relationship  of  the  various  scales  is  presented  in  Fig.  9.  The \nscales  for  X  and   log  X  may  be  considered   fixed,   and    the  scale  for\n\nR  a  movable  one,  as  on  a  slide  rule,  corresponding  values  always \nbeing  opposite  each  other.\n\nEquation  (2)  above  for  the  logarithmic  skew  curve  gives  the  fre- \nquency per  unit  of  X.  The  frequency,  per  unit  of  rent  when  expressed \nin  dollars,  is  1/M  times  that  value,  and  substituting  for  X  from \nequation  (3),  there  results\n\nIf  it  is  desired  to  make  computations  from  this  equation,  it  is  best \nto  use  the  base  10  for  logarithms  rather  than  the  natural  base.  For \nthis  purpose  the  equation  becomes  approximately\n\nalthough  it  will  very  rarely  be  necessary  to  make  such  computations. \nIt  has  been  stated  above  that  the  median  value  of  X  (or  rents \nexpressed  in  dollars)  may  be  logically  regarded  as  the  origin  of  the \nlogarithmic  skew  curve,  although  X  is  not  equal  to  zero  at  this  point, \nbut  is  equal  to  1.  If  some  of  the  other  forms  of  statistical  averages  are \nalso  known,  the  properties  of  the  curve  may  be  better  understood. \nTo  determine  the  mode,  the  first  derivative  of  equation  (2)  of  the \ncurve  is  equated  to  zero,  and  there  results\n\nThe  arithmetic  mean  for  a  distribution  agreeing  with  the  logarith- \nmic skew  curve  probably  can  not  be  defined  by  any  mathematical \nexpression  sufficiently  simple  for  practical  use.  It  is  a  function  of \n<r,  but  its  exact  position  on  the  curve  has  not  been  determined.  As \napplied  to  rent  data,  the  mean  may  be  computed  direct  from  a  house \ncount  summary  with  an  error  of  two  or  three  per  cent.  Thus  found, \nits  position  on  the  curve  is  in  the  neighborhood  of  the  65th  to  70th \npercentile.\n\nThe  geometric  mean  coincides  with  the  median  for  a  logarithmic \nskew  distribution.  This  follows  from  the  fact  that  the  median  value  of \nX  corresponds  to  the  median,  which  is  also  the  mean,  value  of  log  X.\n\nThe  measure  of  dispersion  for  a  logarithmic  skew  curve  is  also  a \nmeasure  of  skewness.  Up  to  this  point  a  has  been  used  as  the  measure \nof  dispersion,  in  agreement  with  conventional  usage.  For  practical \npurposes  another  measure  may  be  substituted,  which  has  a  more \nreadily  understood  meaning.  This  is  the  quartile  deviation,  known \nalso  by  the  misleading  term  probable  error.  The  quartile  deviation \nfor  a  logarithmic  skew  curve  is  that  deviation  either  above  or  below \nthe  median  which  includes  one-fourth  of  all  the  items  in  the  array. \nIt  may,  like  o-,  be  measured  in  logarithms,  and\n\nPerhaps  the  easiest  mathematical  conception  of  a  measure  of  skew- \nness and  dispersion  is  that  of  the  ratio  of  the  upper  quartile  to  the \nmedian.  This  is  identical  with  the  ratio  of  the  median  to  the  lower \nquartile,  and  is  the  number  whose  logarithm  is  the  quartile  deviation \nas  defined  above.    We  shall  let  this  ratio  be  denoted  by  Q.\n\nEither  a  or  the  quartile  deviation  for  a  given  set  of  data  may  be \nbest  determined  from  a  straight  line  graph  on  logarithmic  probability \npaper.  The  following  table  gives  the  positions  in  the  array  for  certain \nconvenient  multiples  of  a  and  the  quartile  deviation,  when  measured \nin  logarithms.\n\nTo  obtain  the  value  of  the  ratio  Q,  take  the  antilogarithm  of  the \nquartile  deviation  determined  in  this  manner.  For  ordinary  purposes \nit  is  sufficiently  accurate  to  obtain  Q  as  the  ratio  of  the  upper  quartile \nto  the  median,  read  from  the  75th  and  50th  per  cent  lines  on  the \ngraph.\n\nSynopsis:  It  is  shown  that  a  satisfactory  measure  of  power  loss  in  a \ndielectric  is  the  product  of  phase  angle  and  dielectric  constant.  Although \nthe  dielectric  constant  need  not  be  explicitly  considered  in  the  design  of \ncondensers,  it  i?  important  in  such  cases  as  the  design  of  apparatus  panels, \nand  vacuum  tube  bases.  The  method  used  in  measuring  phase  angle  and \ndielectric  constant  is  discussed. \u2014 Editor.\n\nIN  working  with  electrical  circuits  operating  at  very  high  frequen- \ncies and  moderately  high  voltages,  such  as  radio  transmitting \ncircuits,  it  is  found  that  failure  in  the  insulation  is  seldom  due  to \npuncture  or  flashover  as  is  usually  the  case  at  power  frequencies, \nbut  is  generally  due  to  excessive  heating  which,  in  turn,  causes  both \nmechanical  and  chemical  disintegration.  As  this  heating  is  due \nalmost  entirely  to  the  energy  losses  occurring  in  the  dielectric  itself, \nit  is  essential  that  the  factors  involved  in  the  calculation  of  these \nlosses  be  well  understood.\n\nIn  the  past,  various  indices  have  been  used  as  a  measure  of  power \nlosses  for  the  purpose  of  comparing  different  dielectrics.  Of  these, \npower-factor,  phase  difference  and  watts  per  cubic  centimeter  prob- \nably are  the  most  common.  None  of  these,  however,  is  very  satis- \nfactory for  this  purpose  since  the  first  two  give  only  part  of  the  de- \nsired information,  and  the  last  is  not  in  any  sense  a  property  of  the \nmaterial,  as  it  is  dependent  on  both  the  voltage  gradient  and  the \nfrequency.\n\nHowever,  it  can  be  shown  that  the  product  of  the  phase  difference \nand  the  dielectric  constant  of  a  material  is  to  a  sufficient  approxima- \ntion an  index  of  its  power  losses.  Let  us  consider  for  a  moment  the \ncomplete  expression  for  dielectric  loss.     In  any  condenser  the  capacity\n\nwhere  a  is  a  constant  depending  on  the  geometrical  dimensions,  and \nK  is  the  dielectric  constant.  If  a  voltage,  E,  is  applied  to  the  con- \ndenser the  power  loss\n\nwhere  /  is  the  current  through  the  condenser  and  ^  is  the  phase \ndifference  of  the  dielectric;    sin  ^  being  the  power  factor.     (Plate\n\nwhere  w  is  a  constant  depending  on  the  units  used,  A  the  area  of  one \nplate,  and  d  the  thickness  of  the  dielectric.\n\nwhere  m'  =  2  -k  m,  and  m'  K  ^  =  loss  per  unit  volume  at  unit \nfrequency  and  potential  gradiant.\n\nThus  it  is  seen  that  while  no  single  factor  of  the  expression  can  be \nused  to  represent  the  losses,  the  product  of  phase  difference  and  di- \nelectric constant  ^  can  be  used  in  this  way.  Furthermore,  for  most \ngood  insulators,  this  product  remains  fairly  constant  throughout  a \nconsiderable  range  of  voltage  and  frequency.  For  example,  we  have \nfound  that  for  such  materials  as  wood,  phenol  fibre,  and  hard  rubber, \nthe  change  of  this  product  with  frequency  is  of  the  order  of  20  per  cent \nfrom  200,000  cycles  to  1,000,000  cycles.  Hence  it  is  possible  to  com- \npare directly  the  losses  in  different  materials  even  though  the  measure- \nments were  not  made  at  exactly  the  same  frequency.\n\nIf  ^  is  taken  in  degrees.  Eg  in  volts  per  centimeter,  and/  in  cycles \nper  second,  the  constant  w'  reduces  to  0.97X10.\"^*.  Hence,  for  a \nfrequency  of  1,000,000  cycles  per  second  and  a  potential  gradient  of \n10,000  volts  per  centimeter,  the  product  of  K  and  \"^  (in  degrees)  is \nwithin  3  per  cent  of  being  numerically  equal  to  the  dielectric  loss  in \nwatts  per  cubic  centimeter.\n\n1  The  substitution  of  the  angle  for  its  sine  is  correct  to  better  than  5  per  cent \nfor  angles  as  large  as  30\u00b0.\n\n-  This  relation  has  been  brought  to  the  attention  of  the  Committee  on  Electrical \nInsulating  Materials  of  the  American  Society  for  Testing  Materials  and  is  included \nin  their  \"Tentative  Method  of  Test  for  Phase  Difference  (Power  Factor)  and  Di- \nelectric Constant  of  Molded  Electrical  Insulating  Materials  at  Radio  Frequencies.\"\n\nData  showing  the  variations  with  frequency  and  temperature  of \nthe  phase  difference,  dielectric  constant,  and  their  product,  for  several \nmaterials  are  given  in  Tables  I.  and  II.  below.\n\nDielectric  Constant,  Phase  Difference  and  Their  Product  for  Several  Commercial \nInsulating  Materials  ^\n\n'All  of  the  samples  had  been  in  the  laboratory  for  some  time  during  summer \nweather  without  artificial  drying  or  other  special  preparation.\n\nVariation  with  Temperature  of  Dielectric  Constant,  Phase  Difference  and  Their  Product \nfor  Some  Commercial  Insulating  Materials  {Frequency  500,000  C.  P.  S.)  *\n\nThe  above  data  were  obtained  by  the  resistance  variation  method,* \nFig.  1.  Each  value  of  phase  difference  and  of  dielectric  constant \nrepresents  the  average  of  at  least  five  readings  on  a  single  sample \nusing  not  less  than  three  different  values  of  the  known  resistance  R. \nA  condenser,  the  dielectric  of  which  consists  of  the  material  to  be \ntested,  is  connected  in  series  with  a  suitable  inductance,  a  known  re-\n\n*  The  measurements  on  each  sample  were  made  in  the  order  in  which  they  are \ngiven  in  the  table.\n\nsistance  which  can  be  varied,  and  a  radio  frequency  ammeter.  An \noscillator  is  coupled  loosely  to  the  inductance  and  its  frequency  varied \nuntil  resonance  is  obtained  as  indicated  by  maximum  current  through \nthe  meter.     Without  changing  the  tuning,  the  resistance  is  changed\n\nand  a  second  reading  of  current  is  obtained.  Then,  since  the  e.m.f. \ninduced  in  the  measuring  circuit  is  the  same  in  both  cases,  if  Ri  and \nRi  are  the  known  resistances,  I\\  and  I2  the  corresponding  currents, \nand  r  the  resistance  of  the  remainder  of  the  circuit,\n\nA  standardized  variable  air  condenser  having  negligible  resistance \nis  then  substituted  for  the  condenser  under  test  and  the  process  re- \npeated except  that  the  circuit  is  tuned  to  resonance  by  varying  the \ncapacity  instead  of  the  frequency.  In  this  way  the  resistance  of  the \ncircuit  exclusive  of  the  test  condenser  may  be  determined.  The \ndifference  between  these  two  circuit  resistances  is  the  resistance  of \nthe  test  condenser  from  which  the  phase  difference  may  be  computed. \nThe  capacity  of  the  test  condenser  is  equal  to  the  capacity  of  the \nstandard  condenser  which  produces  resonance.  From  this  and  the \ndimensions  of  the  sample,  its  dielectric  constant  may  be  computed.\n\nIn  addition  to  the  general  precautions  mentioned  in  the  Bureau  of \nStandards  Bulletin,  two  others  should  be  observed  in  the  measure- \nment of  dielectrics.  First,  the  electrodes  must  be  in  intimate  con- \ntact at  all  points  with  the  surface  of  the  sample,  as  a  very  small  air \nspace  will  cause  a  large  error  in  the  values  of  phase  difference  and  di-\n\nelectric  constant  obtained  for  the  sample.  For  this  reason  only \nmercury  electrodes  have  been  found  suitable,  the  sample  being  floated \non  a  pool  of  mercury  forming  the  lower  electrode  and  the  upper  elec- \ntrode being  formed  by  pouring  a  pool  of  mercury  inside  a  metal  ring \non  the  upper  surface  of  the  sample.  This  introduces  the  second \ndifficulty.  The  lower  electrode  being,  of  necessity,  larger  than  the \nupper  one,  the  electric  field  spreads  out  considerably  beyond  the \nedges  of  the  upper  electrode  and  increases  the  effective  area  by  an \nunknown  amount.\n\nTo  determine  the  magnitude  of  this  error,  measurements  were \nmade  on  samples  prepared  as  shown  in  Fig.  2  ^  is  the  upper  electrode, \nB  is  the  sample  under  test  and  C  is  a  tinfoil  guard  ring  shellaced \nto  the  lower  surface  of  the  sample  but  separated  from  the  lower  elec- \ntrode Z)  by  a  sheet  of  paper  shellaced  over  the  guard  ring.  The \nguard  ring  covers  all  of  the  lower  surface  of  the  sample  except  the \narea  equal  to  the  upper  electrode  and  directly  under  it.  The  direct \ncapacity  ^  between  A  and  D  is  measured  using  the  guard  ring  as  a \nshield  to  intercept  the  flux  which  spreads  out  around  the  upper \nelectrode,  diverting  it  away  from  the  measuring  circuit.  This  measure- \nment is  made  at  audio  frequency  on  a  completely  shielded  substitu- \ntion bridge.  The  difference  between  this  capacity  and  the  capacity \no{  A  to  D  without  the  guard  ring  is  approximately  the  correction  due \nto  this  edge  effect.  Using  samples  6  inches  square  with  an  upper \nelectrode  5  inches  square,  this  correction  was  found  to  be  about  7 \nper  cent  for  samples  }4  inch  thick  and  14  per  cent  for  samples  ^ \ninch  thick.  All  values  of  dielectric  constant  given  above  have  been \ncorrected.  The  phase  difference  is  not  affected  appreciably  since  it  is \ndependent  only  on  the  ratio  of  resistance  to  reactance  and  does  not \ninvolve  the  area  of  the  sample.\n\n^  Direct  Capacity  Measurement,  George  A.  Campbell,  The  Bell  System  Technical \nJournal,  July,  1922.\n\ncare  was  taken  to  avoid  capacity  couplings.  The  measuring  circuit \nwas  shielded  from  the  observer  by  a  metal  screen.  In  spite  of  these \nprecautions  the  results  obtained  on  the  same  sample  at  different \ntimes  do  not  agree  as  well  as  might  be  desired  although  the  individual \nreadings  taken  at  the  same  time  agree  in  most  cases  to  within  5  per \ncent.  Other  observers  have  found  that  measurements  made  on  the \nsame  sample  at  intervals  of  a  few  hours  often  differ  by  more  than \nthe  apparent  error  of  the  measurements  and  have  attributed  it  to \nactual  changes  in  the  properties  of  the  material.''  Hence  it  is  possible \nthat  at  least  part  of  the  apparent  variation  with  frequency  shown \nabove  is  due  to  unknown  errors  in  the  measurements  or  unknown \nchanges  in  the  samples  or  both.\n\nAs  an  illustration  of  the  error  involved  in  taking  only  phase  differ- \nence as  a  measure  of  power  loss,  suppose  we  wish  to  compare  hard \nrubber  having  a  phase  difference  of  about  0.5  degree  and  a  dielectric \nconstant  of  3.,  with  a  certain  grade  of  glass  having  a  phase  difference \nof  about  0.3  degree  and  a  dielectric  constant  of  7.  On  the  basis  of \nphase  difference  alone  the  hard  rubber  appears  very  much  worse  than \nthe  glass,  but  when  the  dielectric  constant  is  taken  into  account,  the \nglass  is  found  to  give  a  40  per  cent  higher  power  loss.  Similarly,  some \nuntreated  woods  were  found  to  have  considerably  lower  losses  than \nthe  phenol  fibres  although  their  phase  differences  are  nearly  the  same.\n\nIn  the  case  of  ordinary  insulation  where  the  object  is  to  provide  a \nmechanical  separator  or  support,  the  product  of  phase  difference  and \ndielectric  constant  is  a  true  measure  of  the  energy  loss  per  unit  volume \nas  shown  by  the  equation  (1).  In  the  case  of  a  condenser  where  the \nobject  is  to  obtain  a  given  capacity,  the  phase  difference  alone  de- \ntermines the  power  loss  since  in  this  case  the  effect  of  the  increased \ndielectric  constant  is  exactly  balanced  by  the  smaller  volume  of \ndielectric  required.\n\n^  R.  Mesing \u2014 L'Onde  Electrique,  April,  1922,  p.  235  and  Augustin  Frigon \u2014 Comptes \nRenins,  May  22,  1922,  p.  1339.\n\nSynopsis:  This  article  points  out  that  radio  and  wire  communication \nsystems  are  subject,  fundamentally,  to  the  same  general  requirements, \nand  its  purpose  is  to  develop  for  radio,  points  of  view  which  are  familiar \nto  wire  transmission  engineers.  The  transmission  characteristics  over  a \nwide  range  of  distances  are  compared.  For  short  distances  the  comparison \nis  favorable  to  wires.  Although  over  great  distances,  the  attenuation  of \nelectric  waves,  guided  by  wires,  may  be  greater  than  the  unguided  waves \nof  radio,  it  is  pointed  out  that  at  the  present  time  intermediate  amplifiers \ncan  be  more  economically  applied  in  wire  transmission  than  in  radio  to \nboost  the  message  energy.  Economy  of  transmission  requires  the  handling \nof  messages  at  as  low  an  energy  level  as  possible  and,  as  the  author  points \nout,  wire  transmission  satisfies  this  requirement  much  better  than  radio. \nReferring  to  the  transcontinental  line  with  radio  extensions,  which  was \nused  recently  to  talk  from  Catalina  Island  in  the  Pacific  Ocean  to  a  ship \nin  the  Atlantic  Ocean,  it  is  stated  that  had  all  of  the  necessary  energy  been \nintroduced  at  one  end  of  the  circuit,  there  being  no  intermediate  amplifica- \ntion, the  total  power  required  would  have  been  1.8  x  10'^  kilowatts,  an \namount  unavailable  in  the  world.  In  the  actual  system,  distributing \nthe  amplification  along  the  transmission  line,  the  power  required  sums \nup  to  something  less  than  1  kilowatt.\n\nInterference  between  messages  and  extraneous  disturbances  is  discussed, \nand  the  requirements  involved  in  keeping  message  energy  well  above  the \nenergy  level  of  the  disturbances  in  both  systems  are  pointed  out.  The \nlimitations  on  two-way  operation  resulting  from  \"singing\"  of  the  entire \nsystem  are  considered  for  both  cases  and  for  combination  wire  and  radio \ncircuits  as  well.  The  method  of  improving  the  efficiency  of  transmission \nby  suppressing  the  carrier  and  one  side  band  is  discussed.  Finally  the \nfactors  involved  in  obtaining  high  grade  quality  of  transmission  are  enumer- \nated.\u2014 Editor.\n\nONE  of  the  most  interesting  aspects  of  the  development  of  radio \nduring  the  last  few  years,  and  particularly  of  radio  telephony, \nis  the  obvious  convergence  of  its  technique  with  that  of  wire  trans- \nmission. It  is,  of  course,  the  advent  into  both  of  these  arts  of  that \nremarkable  device,  the  electron  tube,  which  is  responsible  for  the \nclose  technical  relations  which  now  exist  between  them.\n\nThis  community  of  interest,  however,  altho  thus  greatly  stimu- \nlated by  a  device  of  such  range  of  utility  as  to  find  important  applica- \ntions in  both  arts,  is  not  due  primarily  to  any  device  per  se,  but  rather \nto  the  fact  that  both  type  of  systems  are  subject  fundamentally,  as \ncommunication  systems,  to  the  same  general  requirements  and  design \nconsiderations  concerning  their  intelligence-carrying  capabilities. \nThese  underlying  communication  requirements  lead  to  similar  con- \nsiderations in  both  types  of  systems  as  to  the  efficiency  and  fidelity \nwith  which  the  transmission  of  intelligence  is  effected  and  give  rise\n\n\u2022  Presented  before  The  Institute  of  Radio  Engineers,  New  York,  January  23, \n1922.  Received  by  the  Editor  April  17,  1922,  Also  printed  in  the  Procd.  Radio \nInstitute  for  October,  1922.\n\nThe  engineering  handhng  of  the  transmission  problems  which  arise \nfrom  these  fundamental  communication  requirements  has  been  quite \nhighly  developed  in  the  older  of  the  two  arts \u2014 ^wire  transmission \u2014 in \nconnection  with  telephone  repeaters  and  carrier  telephone  and  tele- \ngraph systems.  It  should  be,  therefore,  interesting  and  profitable \nto  apply  some  of  the  transmission  technique  thus  developed  in  the \nwire  art  to  several  of  the  more  important  radio  problems.  In  so \ndoing,  we  obtain  rather  new  viewpoints  of  radio  transmission  and  a \nuseful  correlation  of  it  with  the  better  established  wire  methods.  It \nis  hoped,  therefore,  that  the  picture  which  is  presented  of  radio  and \nwire  transmission,  treated  from  a  common  standpoint,  may  con- \ntribute to  a  better  appreciation  of  both  arts  by  radio  and  wire  engineers \nalike  and  may  make  clear  the  underlying  transmission  principles \nwhich  are  common  to  them.\n\nPrincipal  among  the  problems  of  electric  communication  is  the  one \nof  delivering  at  the  receiving  end  the  required  volume  of  signal  with \nthe  necessary  freedom  from  interference.  The  delivering  of  the \nrequired  volume  is  a  matter  of  overcoming  the  transmission  losses \nof  the  system  by  amplification;  while  the  obviating  of  interference  is, \nof  course,  concerned  with  the  reduction  of  the  ratio  of  the  interfering \nto   the   signaling  energy.\n\nIn  considering  these  factors  we  will  take  up  first  the  primary  one \nof  the  losses  which  are  suffered  by  the  carrier  waves  as  they  are \npropagated  thru  the  transmission  medium.  In  both  wire  and  radio \ntransmission,  of  course,  the  actual  propagation  of  the  electromagnetic \nwave  energy  occurs  in  the  \"ether,\"  the  difference  being  that  in  the \nwire  case,  the  waves  are  bound  to  a  guiding  path,  whereas  in  the  radio \ncase  they  are  transmitted  freely  in  all  directions  and  bound  merely \nto  the  earth's  surface.  This  difference  in  the  mechanism  of  trans- \nmission gives  rise  to  an  important  difference  in  the  transmission \nlosses  occurring  in  the  two  cases.  In  order  to  assist  in  visualizing  the \ntwo  cases  they  are  indicated  diagrammatically  in  Fig.  1.\n\nReferring  first  to  the  wire  case,  the  law  in  accordance  with  which \nthe  current  and  voltage  strength  decrease  as  the  transmission  wave \ntravels  along  the  wire,  is  the  familiar  one  of  attenuation.\n\nwhich  simply  expresses  the  fact  that,  as  the  wave  proceeds  along \nthe  wire,  the  losses  in  the  resistance  of  the  conductor  and  in  the  insu- \nlation, extract  for  each  mile  a  certain  definite  proportion  of  the  volt- \nage and  current  which  arrives  at  that  point.  After  traveling  (/) \nmiles  the  original  current  /i  is  attenuated  down  to  a  value  /i\u00ab~\"' \nwhich  represents  the  received  current  h-  This  is  the  same  general \nlaw  of  damping  as  applies  to  the  dying  down  of  the  voltage  and  cur- \nrent in  an  oscillation  circuit,  except  that  here  the  damping  is  with \nrespect  to  distance  along  the  line  rather  than  time.  We  are  assuming, \nof  course,  that  the  circuit  is  so  terminated  as  to  avoid  reflection \neffects  at  the  terminals \u2014 a  condition  readily  met,  by  making  the \nterminal  impedance  equal  to  the  characteristic  line  impedance.  This \nis  indicated  in  the  figure  by  the  designations,  Z  (internal)  equals  Z \n(line).  A  similar  relation  is  taken  for  the  radio  case.  The  \"  line  \" \nimpedance  is  here  the  antenna  radiation  resistance  while  the  \"  termi-\n\nnal  \"  impedance  is  the  resistance  internal  to  the  antenna  and  the \napparatus,  assuming  resonance;  thus  R  (internal)  equals  R  (radiation).\n\nWe  know  that  in  radio  there  are  tw^o  distinct  causes  of  the  trans- \nmission loss:  (1)  that,  due  to  the  spreading  out  of  the  waves,  which \nis  characteristic  of  non-guided  wave  transmission;  and  (2)  that  due \nto  absorption  in  the  air  and  earth's  surface,  which  extracts  a  definite \npercentage  loss  for  each  mile  of  the  radio  circuit  and  which  conforms, \ntherefore,  to  an  exponential  law  similar  in  its  general  nature  to  that \nof  wire  attenuation.\n\nformula,  is  given  in  the  Appendix. ^  In  order  to  express  the  radio \ntransmission  loss  in  some  general  manner  which  will  be  comparable \nto  the  expression  of  wire  transmission  loss,  these  conditions  have  been \ntaken  for  the  radio  case:\n\n(1)  That  we  will  use  as  the  measure  of  the  transmission  loss  the \nratio  of  the  square  root  of  the  power  radiated  from  the  sending \nantenna  to  the  square  root  of  the  power  delivered  in  the  re- \nceiving antenna.  This  is,  of  course,  the  same  criterion  as  is \nused  for  wire  transmission.\n\n(2)  The  radiation  resistance  of  the  two  antennas,  sending  and \nreceiving,  are  made  equal,  analogous  to  the  equality  of  line \nimpedance  at  the  two  ends  of  the  wire  system.\n\n(3)  Also  the  internal  antenna  resistance,  which  corresponds  to \nthe  terminating  impedance  in  the  wire  case,  is  made  equal  to \nthe  radiation  resistance  for  both  ends.  This  is  the  condition \nof  maximum  power  transfer  between  the  \"  line  \"  and  the \nterminal.\n\nThese  assumptions  set  up  the  two  cases,  radio  and  wire,  on  a  com- \nparable basis  and  facilitate  a  comparison  of  them.  They  are  favor- \nable to  radio  in  that  they  do  not  take  account  of  practical  limitations \nwhich  obtain  in  antennas.  The  radio  curves  should  be  read,  there- \nfore, as  giving  the  minimum  possible  losses  for  daylight  transmission \nover  water.\n\nThese  curves  show  the  manner  in  which  the  transmission  loss \nvaries  with  distance,  for  various  frequencies,  for  both  radio  and  wire. \nThe  ordinates  are  plotted  in  terms  of  the  logarithm  of  the  ratio  of  the \nsent  to  the  received  currents,  or  voltages,  in  circuits  of  equal  im- \npedances. In  so  doing  we  are  plotting  the  losses  on  the  straight \nattenuation  basis  upon  which  they  are  usually  plotted  in  the  wire \nart;  that  is,  the  ordinates  represent  the  exponent  (\u00ab/)  of  the  wire \nattenuation  law,  and  may  be  directly  interpreted  in  terms  of  miles \nof  standard  cable^  by  multiplying  by  21.  approximately.  The  ad- \nvantage of  dealing  with  the  exponent  rather  than  the  current  ratio\n\n*  Measurements  on  ship-shore  transmission  made  since  the  above  was  written, \nindicate  that  the  Austin-Cohen  law  holds  quite  well  for  frequencies  as  high  as  about \n1,000,000  cycles.\n\n'For  the  mile  of  standard  cable  the  attenuation  a  (at  800  cycles)  equals  0.109. \nTherefore  the  equation  for  current  ratio,  in  terms  of  miles  of  standard  cable,  becomes\n\nitself  is  the  very  considerable  one  which  is  characteristic  of  logarithms, \nnamely,  that  when  thus  expressed  the  individual  losses  and  gains \nthruout  a  system  may  be  summed  up  algebraically,  and  the  over- \nall transmission  equivalent  of  the  system   thus  readily  determined.\n\nIt  should  be  noted  that  the  transmission  loss  given  in  the  radio \ncurves  is  that  obtaining  between  the  point  at  which  power  is  delivered \nto  the  ether  at  the  sending  end  and  that  at  which  it  is  delivered  to  the \ndissipative  load  of  the  receiving  antenna  circuit.  In  Fig.  1  these \npoints  are  represented  by  Ry  at  the  transmitter  and  /?,  at  the  receiver. \nIf  at  the  sending  end,  we  start  with  the  power  developed  within  the \ngenerator,  meaning  in  Ri  instead  of  R^,  then  the  power  ratio  is  sim- \nply doubled,  for  the  conditions  assumed,  and  the  attenuation  is \n0.15  units  or  about  3  miles  greater  than  given  in  the  curves.  The \ncurves  can  be  used  for  obtaining  the  loss  in  any  practical  case  simply \nby  taking  the  minimum  loss  as  given  by  the  curves  and  adding  thereto \nthe  additional  loss  obtaining  in  the  actual  antenna.\n\nReferring  now  to  Fig.  2 \u2014 the  transmission  losses  in  the  two  cases \nare  given  for  distances  up  to  200  miles  (320  km.).  The  straight \nlines  represent  the  wire  losses,  the  bending-over  curves  the  radio \nlosses.  Of  the  radio  curves,  the  dash  lines  give  the  spreading-out \nlosses  alone,  while  the  full  lines  give  the  total  losses,  including  ab- \nsorption.\n\nThe  first  thing  one  observes  is  the  difference  in  the  nature  of  the \ntwo  sets  of  curves \u2014 the  wire  losses  being  represented  by  straight \nlines,  because  of  their  exponential  law  and  the  fact  that  it  is  the \nlogarithm  or  the  exponent  itself  which  is  being  plotted,  while  the  radio \ncurves  jump  up  rapidly  at  first  and  then  straighten  out,  in  accordance \nwith  the  \"  inverse-with-distance  \"  law.\n\nThe  second  thing  one  notes  is  the  fact  that  as  a  result  of  the  large \ninitial  (or  \"  jump  off  \")  loss,  the  radio  values  run  on  the  whole  higher \nthan  do  the  wire  for  the  more  usable  wire  frequencies,  and  very  much \ngreater  than  the  wire  losses  at  telephone  frequencies  (1  k.c.).'*  For \nthe  wire  case  the  number  8  Birmingham  wire  gauge  open  wire  circuit  is \ntaken.^  This  is  the  standard  long  distance  telephone  circuit  of  the \nUnited  States.     The  constants  are  given  in  the  appendix.\n\nA  third  characteristic  which  one  notes  in  the  radio  curves  is  that \nthe  losses  are  greater  for  the  higher  frequencies  or,  conversely,  lower \nfor  the  lower  frequencies.  This  is  because  the  efficiency  of  the  an- \ntenna has  been  kept  constant  for  all  frequencies.     In  practice  the\n\ntransmission   losses   at   the   lower   frequencies   are   higher   than   here \nindicated  because  of  limitations  in  antenna  heights.\n\nWere  we  to  take  the  ideal  condition  as  regards  the  transmission \nmedium  itself,  where  for  wires  there  is  no  conductor  or  dielectric  loss, \nand  for  radio  there  is,  likewise,  no  earth  or  air  absorption  loss,  we \nwould  note:  (1)  that,  for  wires,  there  would  be  no  attenuation  what-\n\nRadio  Dispersion  and  Attenuation \nDashed  Curves-Loss  Due  to  Dispersion  Only.\n\never,  the  curve  following  along  the  X  axis;  (2)  for  radio,  there  would \nremain  the  loss  due  to  dispersion,  inherent  in  the  unguided  method \nof  transmission,  the  magnitude  of  which  loss  is,  of  course,  very  sub- \nstantial. The  dash-line  radio  curves  show  the  radio  losses  without \nattenuation,  the  full  line  curves  with  attenuation.\n\nin  the  transmitting:;  medium,  we  fmd  that  for  moderate  distances, \nup  to  200  miles  (320  km.),  as  plotted  in  Fig.  2,  the  wire  losses  are  in \ngeneral  less,  and  at  telephone  frequencies  very  much  less,  than  the \nradio  losses.  The  low  wire  attenuation  at  telephone  frequencies  is, \nof  course,  in  keeping  with  experience  and  accounts  for  the  economical \nterminal  apparatus  which  is  employed  in  telephone  practice.  Like- \nwise the  relatively  high  losses  for  radio  accounts  for  the  large  amplifica- \ntion at  either  the  sending  or  receiving  end  or  both,  which  experience \nhas   proven    to    be    necessary.     This   brings   in   an   interesting   side-\n\nlight,  namely,  that  altho  in  radio  the  transmission  medium  is  provided \nby  nature,  the  effective  use  of  this  medium  is  not  as  economical  as \nmight  be  expected  because  it  requires  considerable  equipment,  ampli- \nfiers at  both  ends  for  overcoming  the  large  attenuations,  selective \nmeans  for  dividing-up  the  frequency  range  and  thereby  \"multiplex- \ning\" the  ether,  and  antennas  for  getting  into  the  medium  and  out \nagain.\n\nFor   the   higher   frequencies,   the  wire  attenuations   increase   rela- \ntively more  rapidly  than  the  radio,  thus  limiting  the  frequency  range\n\nwhich  can  be  employed  on  wires  without  Hkewise  running  into  large \namplification  requirements.  For  example,  the  loss  at  100,000  cycles \nfor  a  distance  of  200  miles  (320  km.)  is  about  as  great  over  wires  as \nthe  minimum  loss  which  it  is  theoretically  possible  to  obtain  over \nradio.\n\nReferring  now  to  the  attenuations  for  longer  distances,  as  given  in \nFig.  3,  it  is  of  interest  to  note  that  for  distances  of  the  order  of  2,000 \nto  3,000  miles  (3,200  to  4,800  km.)  the  lower  radio  frequency  curves \ncross  the  1,000  cycle  wire  curve,  meaning  that  for  these  distances \nit  is  possible  for  radio  transmission  to  be  as  efficient  as  straight  tele- \nphone transmission.  The  wires  present  to  carrier  frequencies  for \nthese  long  distances  losses  which  are  generally  greater  than  prevail \nfor  radio.\n\nThese  attenuation  relations  cannot  be  directly  converted  into  an \neconomic  comparison,  however,  for  the  economies  depend  not  upon \nthe  attenuation  itself  but  upon,  among  other  factors,  the  cost  in- \nvolved in  overcoming  the  losses  by  means  of  amplification;  and  this \ncost  in  turn  depends  largely  upon  the  extent  to  which  the  amplifica- \ntion can  be  applied  at  weak  powers,  as  by  the  frequent  application \nof  telephone  repeaters.  By  applying  repeaters  every  few  hundred \nmiles  in  the  wire  case,  the  attenuation  is  prevented  from  piling  up \nand  the  amplification  is  handled  at  relatively  weak  and  therefore \neconomical  power  levels.  This  brings  us  to  the  point  of  requiring \nthat  the  attenuation  values  given  above,  be  considered  in  reference \nto  the  amplification  and  power  required  to  overcome  them  and  yield \nthe  necessary  volume  of  transmission  at  the  receiving  end  over  and \nabove  interference.\n\nIn  both  the  radio  and  wire  cases  there  is  always  present  in  the \ntransmission  medium  a  certain  amount  of  stray  wave  energy  which \ntends  to  interfere  with  the  proper  reception  of  the  message-carrying \nwaves.  It  is  necessary  that  the  communication  waves  arrive  at  the \nreceiving  end  of  the  system  with  such  power  as  to  be  large  compared \nwith  the  interfering  waves \u2014 by  a  factor  determined  by  the  type  and \ngrade  of  communication  involved.  Inasmuch  as  the  stray  energy \nalways  has  some  finite  value,  this  requirement  of  freedom  from  inter- \nference will  determine  in  the  radio  case  as  well  as  in  some  types  of \nwire  transmission  the  minimum  wave  power  required  at  the  receiving \nend  of  the  transmission  system.\n\nIn  the  wire  system  the  minuinuin  power  requirement  may  be \nexpressed  directly  in  terms  of  a  power,  or \u2014 as  it  is  usually \u2014 a  current, \nin  the  recei\\ing  apparatus,  the  \"transfer\"  between  the  line  and  the \nreceiver  being  a  constant  and  efficient  relation.  In  radio  the  power \ndelivered  out  of  the  ether  or  \"line\"  into  the  receiving  antenna  is  so\n\nlargely  a  function  of  the  antenna  dimensions  that  it  is  necessary  to \nexpress  the  necessary  received  power  in  terms  of  the  received  field, \nusually,  simply  as  a  field  intensity \u2014 and  not  as  a  certain  current  in \nthe  terminal  apparatus.  Of  course,  the  transmission  loss  occasioned \nby  getting  from  the  \"ether\"  into  the  receiving  circuit  does  not  affect\n\nthe  interference  relation  but  merely  the  absolute  amount  of  terminal \namplification  required.  On  the  other  hand  the  transmission  loss \noccasioned  by  getting  into  the  \"ether\"  at  the  sending  end  aflfects \nthe  interference  relation  vitally,  as  we  shall  see.\n\nThis  necessity  of  having  to  keep  the  power  of  the  received  waves \nabove  the  interference  level  may  be  visualized  by  reference  to  Fig.  4. \nHere  we  have  what  in  wire  practice  is  called  a  \"transmission  level\" \ndiagram.  Such  a  diagram  is  useful  in  showing  what  goes  on  in  the \nsystem  from  the  power  and  interference  standpoints.  The  vertical \nscale  is  plotted  in  terms  of  the  transmission  level  expressed  as  the \nlogarithm  of  the  current  or  field  intensity  ratios,  and  the  horizontal \nscale  represents  progression  along  the  system.  For  illustration \npurposes,  the  presence  of  interference  is  indicated  at  the  bottom  of \nthe  transmission-level  scale  by  the  shading.\n\nThe  point  of  \"zero \"level  is  taken  roughly  as  that  corresponding  to \nthe  power  delivered  into  a  telephone  circuit  by  a  certain  telephone \ntransmitter  when  spoken  into  by  the  average  talker,  and  is  here  taken \nto  be  0.01  watt.  As  the  voice  currents  are  amplified  to  power  pro- \nportions in  the  transmitting  station,  at  the  left,  the  transmission  level \nis  greatly  increased,  as  illustrated  by  the  vertical  jump  in  the  curve. \nThe  amplified  voice  currents  are  assumed  to  be  converted  by  modula- \ntion into  high  frequency  currents  at  this  high  power  level  and  put  into \nthe  antenna.  The  high  frequency  loss  in  the  antenna  system  is  in- \ndicated by  the  perpendicular  jog  in  the  curve.  The  drooping-off  curve \nthen  commences,  starting  with  a  point  which  represents  the  power \nusefully  applied  to  the  ether  in  accordance  with  the  expression  PR, \nwhere  /  is  the  antenna  current  and  R  is  the  radiation  resistance. \nThe  level  curve  falls  off  in  accordance  with  the  transmission  loss \ncurves  previously  discussed,  as  it  extends  across  the  transmitting \nmedium  to  the  receiving  station.  It  will  not  do  to  permit  the  trans- \nmission level  to  fall  as  low  as  that  of  the  interference,  so  I  have  shown \nthat  the  transmission  reaches  the  receiving  station  before  dropping \ndown  very  far  into  the  interference  level.  At  the  receiving  point  a \nfurther  transmission  loss  occurs  in  getting  into  the  receiving  antenna \ncircuit,  shown  by  the  drop  in  the  curve,  but  this  loss  obtains  for  the \ninterference  as  well  as  the  desired  signals  and  does  not  affect  the \ninterference  ratio.  The  terminal  amplification  brings  the  level  up \nto   that  required   for  suitable  audition   and   the  difference  between\n\nwhere  this  level  leaves  off  and  the  original  zero  level,  measures  the \nover-all  transmission  equivalent  of  the  circuit,  shown  in  this  case  as\n\n\u25a0I2 \nresponds  to  a  current  ratio  of  about  4,  a  value  ample  for  good  \"  talk.\" \nOf  course,  in  a  one-way  circuit  the  terminal  amplification  can  be \nraised  to  any  value  desired.  In  a  two-way  circuit,  however,  a  limit \nin  the  terminal  amplification  is  imposed  by  interference  between  the \ntwo  transmissions,  as  will  be  understood  subsequently.\n\n1.  The  net  transmission  equivalent  represents  the  difference  be- \ntween the  over-all  loss  and  the  over-all  gain.\n\n2.  The  over-all  gain  is  divided  betw^een  the  transmitting  and  re- \nceiving ends.  We  should  like  to  throw  as  much  of  this  amplifica- \ntion as  possible  to  the  receiving  end  because  of  the  economy \nwith  which  amplification  can  be  provided  at  low  powers.\n\n3.  The  extent  to  which  we  can  do  this,  however,  is  distinctly  limited \nby  the  fact  that  the  transmission  level  obtaining  at  the  receiving \nend  in  the  transmission  medium  must  be  held  above  a  certain \namount  in  order  to  overcome  interference.\n\n4.  It  is,  then,  the  absolute  intensity  of  the  interference  which \ndetermines  the  receiving  power  level  required,  and  in  turn  this \ntogether  with  the  attenuation  back  to  the  transmitting  station \nwhich  determines  the  transmitting  power  required.\n\nThus  the  two  transmission  features  most  fundamentally  important \nin  a  radio  communication  system  are  (1)  the  interference  level  and \n(2)  the  transmission  loss  thru  the  medium.  These  once  given,  the \nother  engineering  considerations  follow  naturally.  There  are  an- \nalogously fundamental  factors  in  wire  communication  systems.  In \nthe  latter  case,  however,  the  art  has  advanced  to  a  point  where  means \nof  controlling  the  interference  level  are  available,  so  that  the  ratio  of \ninterference  to  transmitted  power  may  be  made  small  by  decreasing \nthe  former  rather  than  increasing  the  latter.\n\nThe  working  value  w^hich  should  be  assumed  for  the  ratio  between \nthe  transmission  level  of  the  received  signals  and  the  interference, \ndepend  upon  the  type  of  communication  involved,  whether  it  is \ntelephone  or  telegraph,  for  example,  and  upon  the  grade  of  service \nto  be  given.  There  is  a  wide  difference  between  the  transmission \nlevel  which  will  enable  telephonic  signals  to  be  barely  discerned  by  an\n\nexpert  ear  and  that  which  is  required  for  a  pubUc  service  communica- \ntion system  which  must  provide  sufificient  operating  margin  to  enable \nthe  average  person  to  converse  with  ease  and  certainty  under  all \nordinary  conditions.  Under  favorable  static  conditions,  the  trans- \nmission level  can  be  permitted  to  fall  to  extraordinarily  low  values. \nWhen  this  condition  is  accompanied  by  a  substantial  reduction  in  the \neffective  attenuation,  which  sometimes  occurs  at  short  wave  lengths \nespecially  at  night,  apparently  due  to  the  effective  absence  of  either \nabsorption,  then  it  becomes  possible  to  \"get  thru\"  over  relatively \nlong  distances  with  powers  diminutive  as  compared  with  those  re- \nquired for  giving  a  regular  service.  With  these  exceptional  trans- \nmission conditions  we  are,  of  course,  familiar.  They  are  exemplified \nby  the  long  distances  reached  at  night  by  the  amateurs,  as  across  the \nAtlantic,  and  by  the  hearing  of  the  normally  30-mile  (48  km.)  Catalina \nIsland  system  in  Australian  waters.  The  transmission  curves  of \nFig.  3  account  for  these  unusual  long  distance  transmissions  if  we \nassume  that  the  attenuation  due  to  absorption  is  eliminated  on  these \noccasions  by  some  natural  cause.  Thus,  at  3,000  miles  (4,800  km.) \nthe  curves  for  1,000  kilocycles  (300  meters),  for  example,  show  that \nwere  the  absorption  eliminated,  the  transmission  equivalent  would  be \nimproved   by   the  difference   between   about    10.8  and   5.2   for  logio\n\nstandard  cable.  The  remaining  or  purely  spreading-out  loss  of  about \n5  units,  or  100  miles  of  standard  cable,  is  then  taken  care  of  by  the \nsending  and  receiving  amplification.\n\nInterference  may  occur  in  either  or  both  of  two  ways \u2014 by  the \ninterference  level  rising  to  a  point  comparable  with  the  normal  trans- \nmission level  at  the  receiving  end  of  the  ether  circuit,  or  by  the \ntransmission  level  of  the  waves  themselves  dropping  so  low,  due  to \nexcessive  atmospheric  absorption,  as  to  fall  below  that  of  the  atmos- \npheric disturbances.  For  reliable  transmission  it  is  necessary,  there- \nfore, to  deliver  normally  at  the  receiving  end,  a  wave  intensity  suffi- \ncient to  allow  for  the  fluctuations  which  occur  in  atmospheric  absorp- \ntion and  in  the  intensity  level  of  the  atmospherics.  The  importance \nof  working  to  transmission  level  standards  which  give  an  adequate \noperating  margin  against  interference,  for  the  types  of  service  re- \nquired, will  be  appreciated  from  the  foregoing.  The  following  values \nof  minimum  transmission  levels  will  be  of  value  to  know:\n\n(a)   For  carrier  wire  telephone  transmission  at  frequencies  in  the \ntens  of  thousandths  of  cycles,  the  limiting  interference  may  be\n\n(b)  While  for  radio  telephone  transmission  the  available  data  are \nas  yet  very  meagre,  we  have  obtained  a  few  order-of-magnitude \nfigures  which  should  be  of  interest.  For  the  Catalina  Island \nradiophone  system,  for  example,  the  minimum  field  intensity \nis  estimated  at  roughly  1,000  microvolts  per  meter.  The \ncircuit  is  sometimes  quite  noisy  during  the  summer  months \naltho  not  prohibitively  so.  In  our  ship-to-shore  radio  tele- \nphone experiments  along  the  Atlantic  coast,  we  have  on  occa- \nsions worked  with  lower  field  intensities,  as  low  as  100  micro- \nvolts per  meter.  The  latter  figure,  however,  gives  a  grade  of \nservice  far  below  wire  standards.\n\n(c)  The  best  data  on  the  minimum  permissible  transmission  level \nfor  radio  telegraphy  are  those  obtained  from  the  experience  in \ntrans-Atlantic  telegraph  operation.  The  figures  prevailing \nfor  present  trans-oceanic  radio-telegraph  operation  are  under- \nstood to  lie  in  the  order  of  10  to  100  microvolts  per  meter, \ndepending  upon  individual  cases  and  the  time  of  the  year.\n\nThe  net  over-all  transmission  equivalent  of  the  system  is  measured \nby  the  ratio  of  the  transmitted  to  the  received  signaling  power,  and  is \nshown  in  Fig.  4  as  the  difference  between  the  transmission  levels  at \nthe  two  ends.  This  relatively  small  loss  represents  the  difference \nbetween  two  large  values,  the  transmission  loss  and  the  transmission \ngain  thruout  the  system.  Relatively  small  changes  in  either  the  at- \ntenuation or  amplification  may,  therefore,  cause  large  changes  in  the \nnet  equivalent  of  the  circuit,  thus  tending  to  give  rise  to  instability \nin  the  transmission  performance  of  the  circuit.\n\nThis  problem  of  fluctuation  becomes  very  serious  with  the  use  of \nvery  high  frequencies,  whether  transmitted  by  wires  or  by  radio. \nWere  we  to  attempt  to  employ,  for  example,  a  million  cycles  for  wire \ncarrier  transmission  over  considerable  distances,  as  has  been  proposed, \nnot  only  would  the  losses  be  very  large,  but  they  would  be  unstable, \nchanging   with   weather   conditions,    so   that   the   maintenance   of   a\n\nconstant  volume  of  transmission  would  become  extremely  difficult. \nSimilarly  in  radio  transmission,  the  fluctuations  in  the  ether  attenu- \nation, particularly  at  short  wave  lengths  where  over  long  distances \nwe  experience  the  well  known  \"swinging\"  or  fading  efTects,  render  the \nmaintenance  of  a  satisfactory  volume  of  transmission  a  difficult \nproblem.  As  noted  above  these  fluctuations,  particularly  as  between \nday  and  night  transmission  with  very  high  frequencies,  may  be \nenormous.\n\nsiderably  larger  transmission  equivalents  can  be  talked  over.  The \nlong  distance  toll  lines  themselves  are  usually  designed  for \ntransmission  equivalents  of  0.5  to  0.75  or  10  to  15  miles  of  standard \ncable.  These  figures  will  serve  as  a  general  guide  for  the  transmission \nequivalents  which  radio  telephone  circuits  should  provide.  Where  a \nradio  circuit  forms  a  link  in  a  direct  wire  circuit  as,  for  example,  in \nthe  case  of  Catalina  Island,  it  is  desirable  to  work  the  radio  link  as \nclose  to  a  zero  equivalent  as  possible,  that  is,  to  give  out  at  the  receiv- \ning end  a  volume  nearly  equal  to  that  fed  in  at  the  transmitting  end.\n\nWhen  the  two  one-way  radio  channels  are  merged  at  their  two \nends  into  a  regular  telephone  circuit  for  connection  to  the  wire  net- \nwork, as  illustrated  in  Fig.  5,  then  there  is  a  limit  in  the  transmission \nequivalent  which  can  be  given  over  the  radio  part  of  the  circuit.\n\nThis  limit  will  be  appreciated  by  reference  to  Fig.  5.  It  is  im- \nposed by  the  tendency  of  the  two  one-way  channels  to  form  a  round- \ntrip  circuit  by  \"feeding-back\"  from  one  to  the  other  via  the  voice \nfrequency  connecting  circuit.  If  the  total  amplification  around  the \ncircuit  including  the  voice-frequency  line,  exceeds  the  total  losses  in \nthe  circuit,  \"singing\"  will  result.  Were  no  line  balance  provided  at \nthe  voice  frequency  terminals,  then  it  would  be  impossible  to  operate \nthe  circuit  at  a  zero  equivalent.  By  setting  up  a  balancing  circuit  at \neach  end  in  the  manner  illustrated,  a  transmission  loss  is,  in  effect, \ninserted  between  the  sending  and  receiving  sides  of  the  voice  circuit \nwhich  tends  to  prevent  this  sing-around  action.  Actually,  there  is  a \nlimitation  in  the  degree  of  balance  which  can  be  realized  between  the\n\ntelephone  line  and  the  balancing  network,  especially  if  tlie  telephone \nline  is  to  be  switched  at  a  nearby  central  office,  and  this  factor,  to- \ngether with  the  margin  of  safety  which  is  required  between  the  oper- \nating condition  and  the  singing  condition,  prevents  the  radio  channels \nfrom  being  operated  much  better  than  the  zero  equivalent.  This \nwhole  matter  of  realizing  in  practice  an  adequate  transmission  equiva- \nlent, will  be  appreciated  to  be  an  especially  difficult  problem  in  the \ncase  of  marine  radio  telephony,  where  the  connection  is  switched \nfrom  one  vessel  to  another  at  varying  distances.\n\nIt  should  be  noted  further,  with  reference  to  two-way  operation, \nthat  the  difficulty  of  effecting  simultaneous  sending  and  receiving \nat  a  station  arises  primarily  from  the  large  attenuation  which  must  be\n\novercome  and  the  resulting  large  ratio  between  the  energies  trans- \nmitted into  and  received  from  the  ether.  The  receiver  must  be  pre- \nvented from  being  overloaded  by  the  home  transmitter  and  this, \nin  general,  requires  that  there  be  provided  between  the  high  frequency \nside  of  the  transmitter  and  that  of  the  receiver,  a  transmission  loss \ncomparable  in  size  to  that  obtaining  over  the  radio  circuit  itself.  This \n\"separating\"  transmission  loss  is  ordinarily  provided  (a)  by  fre- \nquency-selecting circuits  (tuned  circuits  and  filters),  the  sending  and \nreceiving  transmissions  being  placed  on  different  frequencies;  (b)  by \nbalance,  as  w'hen  using  the  blind  spot  of  a  loop-antenna  receiver, \nand  (c)  by  spatial  separation  between  sending  and  receiving  points, \nwhere  the  large  step-off  loss  is  used  to  advantage.\n\nIt  will  be  of  interest  to  trace  thru  the  approximate  transmission \nlevels  which  obtain  for  a  radio  system  linked  up  with  a  long  repeat- \nered  land  line  system.\n\nIn  Fig.  6,  there  is  taken  a  rather  striking  example  of  this  case  in \nthe  transcontinental  telephone  line  as  connected  up  to  radio  ex- \ntensions at  its  termini \u2014 to  Catalina  Island  on  the  Pacific  and  to  a \nvessel  at  sea  on  the  Atlantic.  The  transmission  illustrated  is  that \noccurring  from  east  to  west.  The  voice  currents  start  out  from  the \nvessel  at  zero  level,  are  amplified  to  a  relatively  high  level  and  upon \nbeing  transmitted  to  the  shore  150  miles  (240  km.)  away,  drop  to  a \nvery  low  level.  At  the  shore  radio  station  they  are  boosted  up,  at \nNew  York  amplified  again,  and  put  upon  the  transcontinental  circuit. \nRegularly  at  about  300  miles  (480  km.)  the  telephone  repeaters  pull \nback  the  transmission  level  to  about  its  original  value.  In  the  radio \nlink  at  the  w-estern  end  the  currents  are  again  amplified  to  a  high  level \nat  the  transmitting  station,  drop  down  to  a  very  low  level  at  the  re- \nceiver and  are  brought  back  to  a  level  at  which  they  can  be  heard. \nActually  in  the  receiving  telephone  the  transmission  is  about  logio\n\n\"  down.\"  The  total  loss  and  the  total  gain  in  the  circuit  is  enormous, \nas  is  shown  by  the  figures  given  in  the  diagram.  This  is  a  rather \nstriking  illustration  of  the  extent  to  which  amplification  properly  dis- \ntributed and  maintained  can  be  used  to  overcome  attenuations  enor- \nmous in  the  aggregate.  Just  to  give  a  better  idea  of  what  these  values \nof  attenuation  and  amplification  mean,  it  may  be  noted  that  were  it \nnecessary  to  supply  at  the  transmitting  end  all  of  the  amplification  re- \nquired for  delivering  this  volume  of  transmission  to  the  receiving  end \nthru  the  combination  circuit,  the  kilowatts  required  would  be  measured \nby  a  twenty-nine  place  figure,  an  amount  of  power  unavailable  in \nthe  world.  The  importance  of  correctly  distributing  the  amplifica- \ntion along  the  system  is  well  illustrated  by  this  figure  by  comparing \nit  with  the  signaling  power  actually  represented  in  the  system,  which \nsums  up  to  something  less  than  1  kilowatt.  The  difference  is  simply \na  question  of  the  transmission  level  at  which  the  amplification  is \nworked.\n\nFig.  7  gives  a  view  of  the  interior  of  one  of  these  radio  telephone \nstations  of  the  American  Telephone  and  Telegraph  Company  and \nWestern  Electric  Company.    It  is  located  at  Deal  Beach,  New  Jersey.\n\nIn  the  foreground  is  the  switchboard  for  enabling  the  operator  to \ncontrol  the  radio-wire  circuit  at  the  connecting  point.  In  the  back- \nground are  the  transmitter  units \u2014 four  of  them.  These,  together \nwith  the  four  antennas  with  which  the  station  is  equipped,  \"multiplex\" \nthe  ether,  in  effect,  and  permit  four  channels  to  be  established  to  as \nmany  distant  stations.  It  is  intended  that  three  of  these  be  tele- \nphone talking  channels  and  the  fourth  a  signaling  or  a  reserve  talking \nchannel.  The  receiving  station  is  located  at  another  point.  It  is \nnot  desired  to  describe  this  station  in  any  detail  but  merely  to  illus- \ntrate it  as  an  example  of  a  radio  repeating  station  functioning  to\n\nconnect  the  wire  system  with  ships  at  sea  and  capable  of  effecting \nsimultaneously  three  different  connections.  It  is  hoped  that  this \nship-to-shore  development  may  be  itself  the  subject  of  an  Institute \npaper.\n\nThe  transcontinental  line  with  radio  extensions  as  shown  in  Fig.  6 \nis  a  good  illustration  of  the  use  of  intermediate  repeaters  generally. \nTwo  types  of  repeaters  are  represented,  the  straight  wire  telephone\n\nrepeaters  and  the  shore  radio  stations  which  are  in  effect  huge  re- \npeaters rehaying  between  the  land  line  and  the  radio  circuits.\n\nBecause  of  the  moderate  attenuation  obtaining  in  the  wire  trans- \nmission system,  we  can  work  with  fairly  long  repeater  spacings,  about \n300  miles  in  this  case,  and  with  moderate  amounts  of  power  and  yet \nkeep   the   transmission  levels  at   the  receiving  end   relatively   much\n\nTRANSMISSION  LEVEL  DIAGRAM \nWire  Transmission f=l  K.C \nRad  lolransmission  f  \u25a0- 1000  K  C\n\nhigher  than  is  usual  in  radio  systems.  Due  to  the  large  attenuations \nobtaining  over  the  radio  extensions,  the  radio  repeaters  must  put  out  a \nhigh  transmission  level,  making  them  costly,  and  even  with  this  rela- \ntively high  output,  the  level  drops  to  very  low  values  at  the  receiving\n\nend.  This  falling  off  occurs  largely  in  the  get-away  loss  at  the  trans- \nmitting end  of  the  radio  circuit,  as  the  diagram  indicates.\n\nFig.  8  depicts  an  all-radio  system  provided  with  intermediate \nrepeaters  and  compares  for  illustration  purposes  the  transmission \nlevels  obtaining  therein  with  those  for  a  wire  system.  The  solid \nlines  are  for  radio  and  the  dash  for  wire.  It  will  be  seen  that  the \nradio  system  courses  thru  wide  transmission  level  variations  as  com- \npared to  the  ordinary  wire  system  due  to  the  large  attenuations \nobtaining  and  particularly  to  the  large  step-off  loss  near  the  sending \nstation.\n\nThe  figure  illustrates  the  same  spacing  for  both  radio  and  wire \nrepeating  and  gives  a  measure  of  the  difference  in  amplification \nrequired  in  the  two  cases.  Altho  in  the  radio  repeaters  the  level \ncan  be  permitted  to  drop  to  low  values,  nevertheless  a  large  part  of \nthe  total  amplification  has  to  be  supplied  at  relatively  high  power \nlevels  and  it  is  this  fact,  together  with  the  antenna  structures  required \nat  each  point  to  \"get  into\"  the  ether  transmission  medium  anew, \nthat  militates  against  the  economics  of  radio  repeaters  as  compared \nwith  straight-away  radio  transmission.  The  tendency  will  be  to \n\"stretch  out\"  the  straight-away  transmission  due  to  the  fact  that \nfor  the  longer  distances  the  transmission  loss  increases  relatively \nslowly.  While  we  may  look  for  some  important  uses  of  radio  re- \npeaters in  special  cases,  we  should  not,  in  general,  expect  them  to  be \nas  important  to  the  radio  art  as  are  wire  repeaters  in  wire  operation.\n\nIn  dealing  with  the  subject  of  power  levels  in  radio  transmission, \nit  is  important  to  recognize  that  a  modulated  radio  telephone  wave \nconsists  of  two  components,  one,  the  carrier  frequency  itself  and  the \nother,  the  so-called  side  bands,  which  are  the  actual  modulated \ncomponents.  This  resolution  of  the  modulated  carrier  into  two \nor,  rather,  three  components,  the  carrier  and  two  side-bands,  has \nbeen  given  mathematically  a  number  of  times  and  need  not  be  re- \npeated. It  is  physically  analogous  to  the  resolution  of  the  uni- \ndirectional current  of  a  microphone  transmitter  into  direct  current \nand  alternating  current  components,  the  direct  current  corresponding \nto  the  carrier  and  the  alternating  current  to  the  modulated  components.\n\nNow,  the  important  thing  about  this  matter  of  side  bands  and  the \nunmodulated  carrier  component,  with  reference  to  transmission \nconsiderations,  is  this,  that  it  is  the  side  bands  alone,  and  not  the \ncarrier,  which  convey  the  actual  intelligence.     The  function  of  the\n\ncarrier  comes  in  merely  at  the  rerei\\inp:  end,  in  the  detector,  as  a \nmeans  for  translating  the  side  band  from  radio  frequency  back  to \naudio  frequency.\n\nThis  will  be  made  clearer  by  reference  to  Fig.  9.  At  the  bottom \nof  the  figure  is  shown  schematically  a  one-way  radio  system.  Above \nit  is  depicted  the  voice-frequency  band,  showing  the  manner  in  which \nit  is  shifted  by  modulation  up  to  the  carrier  frequency  range,  and  at \nthe  receiving  end,  by  detection,  back  to  the  voice  frequency  range. \nThe  voice  frequency  band,  as  it  comes  out  of  the  ordinary  telephone \ntransmitter,  is  shown  at  (61)  at  its  normal  telephone-frequency  posi-\n\ntion.  Upon  modulation  with  the  carrier  the  reference  point  of  the \nvoice  frequency  band  is  shifted  from  zero  frequency  (direct  current) \nup  to  the  carrier  frequency  as  shown  at  b^  where  the  two  side  bands \nappear.  The  effect  of  modulation  is,  therefore,  simply  to  shift  the \nband  of  signaling  frequencies  upward  in  the  frequency  range  and \nrefer  it  in  a  double  relation  to  the  carrier  frequency.\n\nLocated  between  the  upper  and  lower  side  band  in  the  figure, \nthere  is  indicated  the  unmodulated  component  of  the  carrier.  The \nfact  that  this  component  is  unnecessary  so  far  as  the  actual  intelli- \ngence-carrying energy  is  concerned,  is  proven  by  the  fact  that  it \nneed  not  be  transmitted  to  the  receiver.     The  carrier  may  be  sup-\n\npressed  as  shown  at  63.  A  means  for  doing  this  is  the  Carson  balanced \ntube  modulator  circuit  illustrated  below  in  the  figure.\n\nA  reduction  of  the  total  band  can  be  effected  by  filtering  out  one \nof  them  as  shown  at  ^4-  The  remaining  single  band  is  the  simplest \ncomponent  of  the  modulation  process  with  which  intelligence  can \nbe  transmitted  to  the  distant  end.  Upon  arriving  at  the  receiving \nend  at  b^,  this  side  band  is  fed  into  the  detector  along  with  a  carrier \nof  the  same  frequency  as  that  employed  at  the  sending  end;  these \ntwo  components  demodulate  one  another,  with  the  result  that  the \nside  band  is  shifted  down  to  its  original  audio-frequency  position  in \nthe  scale,  as  indicated  at  be,.\n\nActually  this  general  method  of  transmission,  involving  both \ncarrier  suppression  and  side  band  elimination,  is  being  employed  in \nwire  carrier  systems  in  the  Bell  Telephone  Plant. ^  It  is  briefly  ex- \nplained here  because  it  represents  a  valuable  improvement  in  wire \ntransmission  which  should  have  important  application  in  radio.\n\nFrom  the  standpoint  of  transmission  levels  its  application  is  in \nshowing  that  the  real  intelligence-carrying  component  of  a  radio  wave \nis  the  side-band  and  not  the  carrier  itself.  In  considering  transmission \nlevels  accurately  care  should  be  taken,  therefore,  to  deal  in  terms  of \nthe  level  or  wave-intensity  of  the  side-band  component  and  not  the \ncarrier.  It  is  because  of  this  that  as  nearly  complete  modulation \nas  possible  is  desired  at  the  transmitting  station.\n\nIt  follows  that  the  power  resident  in  the  carrier  is  a  pure  waste  in \nso  far  as  overcoming  interference  is  concerned.  An  important  power \nsaving  can  be  effected  in  the  transmitting  station  by  providing  some \nsuch  means  as  is  illustrated  whereby  the  carrier  power  is  held  back \nin  the  circuit.  The  two  side-bands  together  can  never  be  greater  in \ncurrent  and  voltage  value  than  the  carrier,  and  each  side-band  alone \ncannot  be  greater  than  half  the  carrier.  The  power  of  the  carrier  is \ntherefore  always  at  least  four  times  the  power  of  one  side-band  or \ntwice  that  of  both  together.  Thus  by  \"  holding-back  \"  the  carrier \nat  the  transmitter  we  can  transmit  with  but  one-third  the  power \nordinarily  required.  Actually  the  power  saving  is  much  greater \nthan  this  because  of  the  necessity  of  normally  working  with  larger \nratios  between  carrier  and  side-band  in  order  to  accommodate  the \npeaks  of  the  telephone  waves  and  thereby  preserve  the  quality  of \ntransmission.  The  power  saving  is,  of  course,  especially  important \nin  long  distance  work.\n\n\u2022\"Carrier  Current  Telephony  and  Telegraphy,\"  by  Colpitis  and  Blackwell, \nJournal  of  the  American  Institute  of  Electrical  Engineers,  February  16-18,  1921.\n\nband  required  for  transmission  and  would  double  the  message-carrying \ncapacity  of  the  ether  were  no  frequency  range  required  to  space  the \nchannels  apart.  This  advantage  of  the  present  method  is  likewise  of \nspecial  importance  in  long-distance-long-wave  transmission.\n\nWe  have  spoken  above  of  factors  concerned  with  the  volume  of \ntransmission  and  only  incidentally  of  that  other  requisite  of  trans- \nmission, namely,  good  quality.  Without  going  into  this  matter  in \nmuch  detail,  it  will  be  well  to  make  note  of  the  several  factors  in- \nvolved in  obtaining  good  quality,  as  follows:\n\n1.  It  is  important  that  a  substantial  band  of  voice  frequencies  be \ntransmitted.  Of  course,  distorted  talk  can  be  transmitted  on  a \nrelatively  narrow  band,  but  commercial  transmission  has  been  found \nto  require  a  single  side-band  width  of  the  order  of  2,000  or  more \ncycles,  the  band  width  increasing  with  the  quality  desired,  up  to \nabout  5,000  cycles.\n\n2.  It  is  necessary  that  the  distortion  which  is  due  to  non-linearity \nof  transmission  with  respect  to  amplitude,  be  avoided.  This  is \nequivalent  to  saying  that  there  should  not  be  permitted  to  take  place \nself-modulation  between  the  components  of  the  side-band,  nor  the \ntoo  close  cutting-ofT  of  the  peaks  of  the  telephone  waves  due  to  satura- \ntion effects.\n\n3.  The  transmission  must  be  kept  free  from  interfering  noises. \nThe  ratio  between  interfering  noise  current  and  voice  currents  of  the \norder  of  0.1  is  regarded  as  large  in  wire  practice.  While  this  amount  of \ninterfering  current  will  not  prohibit  service,  it  does  seriously  impair  the \neffectiveness  of  transmission  and  annoys  the  listener.  In  radio  the \nratio  of  static  noise  to  signal  strength  is  very  often  much  greater  than \nthis  value.  As  the  radio  art  progresses  it  will  be  necessary  to  work \ntoward  standards  more  nearly  in  keeping  with  those  which  have  been \nfound  necessary  for  wire  service.\n\nThe  writer  wishes  to  express  his  indebtedness  to  the  following  of \nhis  associates  for  helpful  suggestions  and  assistance \u2014 Messrs.  J.  R.  Car- \nson, Ralph  Bown,  and  D.  K.  Martin.\n\nAs  regards  antenna  resistance  we  assume  symmetry  as  between \nthe  two  ends,  and  that  the  external  (radiation)  resistance  of  the  an- \ntenna equals  the  resistance  within  the  antenna  (which  resistance \nwould  be  internal  apparatus  resistance  in  the  case  of  a  perfect  an- \ntenna). This  makes  R^  (radiation  resistance)  =  i?,-  (ohmic  resistance) ; \nand  R  of  (1)  becomes  =  R^  -^  Ri  where:\n\nExpressing  in  terms  of  current  ratio  and  substituting  values  of  R, \nequation  (1)  becomes,\n\nIn  order  to  plot  this  equation  on  the  same  basis  as  we  usually  plot \nwire  attenuation,  the  logarithm  of  the  ratio  is  used,  thus;\n\nThe  ratio  of  the  currents  in  the  two  antennas  is  in  this  case  a  true \nmeasure  of  the  transmission  because  they  are  in  circuits  of  equal  im- \npedances, by  the  assumption  of  antenna  symmetry.\n\nFor  number  8  Birmingham  wire  gauge  open  wire  (diam.  =  0.165 \nin.  =  4.19mm.  wire  spacing  =  12  in.  =  30.5  cm.  40  poles  per  mile) \ndry  weather,  the  constants  per  mile  are;\n\nSynopsis:  A  sensitive  cathode  ray  oscillograph  is  described  which \noperates  at  the  low  potential  of  from  300  to  400  volts.  The  electron  stream \ncomes  from  a  thermionic  cathode  and  is  focused  by  the  action  of  ionized \ngas  in  the  tube.  This  gas,  at  a  pressure  of  a  few  thousandths  of  a  milli- \nmeter, serves  to  reduce  to  1mm.  diameter  a  spot  which  would  be  1  cm. \nacross  in  a  high  vacuum  tube.  The  sensitivity  of  the  tube  is  such  that \nthe  deflection  of  the  spot  is  about  1  mm.  per  volt  applied  between  deflector \nplates.  When  using  magnetic  deflection,  a  pair  of  coils  4  cm.  in  diameter, \nplaced  on  the  sides  of  the  tube  produces  a  deflection  of  approximately  1  mm. \nper  ampere-turn  of  the  coils. \u2014 Editor.\n\nIn  the  older  types  of  Braun  tubes  the  electron  stream  is  produced \nby  a  high  voltage  discharge  through  the  residual  gas  in  the  tube. \nThis  requires  a  source  of  steady  potential  of  from  10,000  to  50,000 \nvolts,  an  installation  which  is  expensive,  non-portable,  and  dangerous. \nIn  the  new  type  of  tube  the  low  voltage  operation  has  been  obtained \nby  the  use  of  a  Wehnelt  cathode  as  the  source  of  electrons,  so  that  the \nlower  limit  of  voltage  is  set  by  the  effect  of  the  electrons  on  the  fluores- \ncent screen  and  not  by  the  voltage  needed  to  obtain  the  electrons. \nAt  300  volts  the  electrons  produce  quite  bright  fluorescence  on  the \nscreen  and  the  tubes  are  therefore  designed  to  operate  at  300  to  400 \nvolts.\n\nThe  external  appearance  of  the  tube  is  shown  in  Fig.  1.  The \nelectrodes  are  located  at  one  end  of  the  pear-shaped  bulb,  and  the \nfluorescent  material  is  deposited  on  the  inside  of  the  larger,  flattened \nend.  The  tube  is  provided  with  a  base  which  fits  into  a  bayonet \nsocket  such  as  is  used  for  vacuum  tubes,  and  all  the  connections  are \nmade  through  the  base.  There  are  two  orthogonal  pairs  of  deflector \nplates  inside  the  tube  for  electrostatic  deflection,  while  magnetic \ndeflection  is  produced  by  applying  a  field  from  the  outside.\n\n'Also  published  in  the\"  Journal  of  the  Optical  Society  of  America  and  Review  oj \nScientific  Instruments,  September,  1922. \n*Phys.  Rev.  (2),  Vol.  17,  p.  420,  1921.\n\nIn  some  forms  of  Braun  tube  a  sharp  spot  has  been  secured  by  using \na  very  high  voltage,  and  therefore  high  electron  velocity,  so  that \nafter  the  electrons  have  passed  through  one  or  two  fine  apertures \nto  make  the  beam  parallel  there  is  not  time  enough  for  the  mutual \nrepulsion  to  spread  the  beam  again  appreciably  before  the  electrons \nstrike  the  screen.  With  other  tubes  an  external  \"striction\"  coil  has \nbeen  used  which  maintains  a  strong  longitudinal  magnetic  field  in  the \nregion  between  the  anode  and  the  cathode  and  which  brings  the \nelectrons  to  a  focus  on  the  screen.  In  the  low  voltage  tube  the  spread- \ning of  the  electron  stream  is  greater  than  in  high  voltage  tubes  because \nof  the  greater  time  during  which  the  mutual  repulsion  of  the  electrons\n\nacts,  so  that  some  means  of  focusing  must  be  used.  The  electrons \ncan  be  brought  to  a  focus  by  a  longitudinal  magnetic  field  so  adjusted \nthat  each  divergent  electron  makes  very  nearly  one  complete  turn  of \na  spiral  and  in  travelling  the  length  of  the  tube  returns  to  the  axis  at \nthe  screen.  In  this  way  a  very  sharp  spot  can  be  produced,  but  the \nsensitivity  of  the  beam  to  deflection  is  reduced  very  much  by  the \ndirecting  magnetic  field.\n\nThe  method  of  focusing  that  is  used  in  the  present  tubes  grew  out \nof  the  suggestion  by  Dr.  H.  J.  van  der  Bijl,  that  a  small  amount  of \ngas  be  introduced  into  the  tube.  This  gas,  at  a  pressure  of  a  few \nthousandths  of  a  millimeter  of  mercury,  serves  to  reduce  to  1  mm. \ndiameter  a  spot  which  would  be  1  cm.  across  in  a  high  vacuum  tube. \nThe  sharpness  of  the  spot  depends  also  upon  the  current  in  the  electron \nstream  so  that  the  focus  may  be  controlled  by  the  cathode  temper- \nature.    The  mechanism  of  this  focusing  action  will  be  explained  later.\n\nThe  presence  of  this  slightly  ionized  gas  also  serves  the  purpose  of \npreventing  the  accumulation  of  charges  on  the  glass,  and  it  provides\n\nfor  the  discharging  of  the  fluorescent  screen  so  that  the  electrons  can \ndrift  back  to  the  metallic  circuit.\n\nWith  gas  present  in  the  tube,  steps  have  to  be  taken  to  guard  against \narcing  and  the  injurious  effects  of  positive  ion  bombardment  on  the \ncathode.  This  is  done  by  making  the  volume  of  gas  surrounding  the \nelectrodes  very  small.  For  this  purpose  the  cathode  and  anode, \nthemselves  small,  are  sealed  into  a  short  and  narrow  glass  tube  so \nthat  the  volume  exposed  to  both  electrodes  in  common  is  less  than \n1  cu.  cm.  All  paths  between  the  electrodes  are  then  so  short  that  at \nthis  low  pressure  there  is  not  sufficient  ionization  to  build  up  an  arc.\n\nThe  structure  of  this  unit,  or  \"electron  gun\"  is  shown  in  Fig.  2. \nThe  cathode,  /,  is  an  oxide  coated  platinum  ribbon  of  the  same  kind\n\nas  the  filament  in  our  audion  tubes.  The  anode,  a,  is  a  thin  platinum \ntube  1  cm.  long  and  1  mm.  in  diameter,  one  end  of  which  is  about \n1  mm.  from  the  top  of  the  filament  loop,  the  other  end  opening  into \nthe  main  tube  towards  the  fluorescent  screen.  Between  the  cathode \nand  the  anode  and  connected  to  the  cathode  is  a  metal  shield,  s,  with \na  small  aperture  through  which  the  electrons  must  pass  in  going  to  the \nanode.  Nearly  all  of  the  electrons  must  then  go  to  the  inside  of  the \ntubular  anode,  and  a  small  fraction  of  them  pass  through  the  whole\n\nlength  of  the  anode  and  form  the  beam  in  the  main  part  of  the  tube. \nThe  deflector  plates,  p,  are  also  mounted  rigidly  on  this  unit.  In \norder  to  avoid  large  differences  of  potential  in  the  tube,  one  plate  from \neach  pair  is  permanently  connected  to  the  anode,  the  variable  poten- \ntials being  applied  to  the  other  plates.  The  complete  unit  is  mounted \nat  the  small  end  of  the  tube  with  the  anode  and  deflector  plates  toward \nthe  fluorescent  screen.\n\nIn  some  earl>'  forms  the  filament  was  bent  into  a  simple  hair  pin \nloop  which  was  placed  close  to  the  aperture  in  the  shield.  It  was  then \nfound  that  the  positive  ions  striking  the  filament  from  the  direction \nof  the  anode  soon  destroyed  the  oxide  coating  and  left  the  filament \ninactive.  This  trouble  was  largely  overcome  by  placing  the  filament \nout  of  the  direct  path  of  the  positive  ions.  The  flat  filament  is  now \nshaped  into  a  ring  as  shown  in  P'ig.  3,  slightly  larger  in  diameter  than\n\nthe  aperture  in  the  shield  and  is  placed  coaxial  with  the  anode.  The \nmomentum  of  the  positive  ions  then  carries  them  past  the  active \npart  of  the  filament  and  they  strike  where  little  damage  can  be  done. \nThe  length  of  service  of  the  tube  is  still  limited  by  the  filament  life, \nbut  this  has  been  increased  by  the  above  artifice  so  that  the  tube  now \ngives  around  200  hours  of  actual  operation.\n\nThe  deflector  plates  are  made  of  German  silver,  which  is  non- \nmagnetic and  which  has  a  high  specific  resistance  that  diminishes  the \neffect  of  eddy  currents  when  magnetic  deflection  is  used.  The  plates \nare  13.7  mm.  long  in  the  direction  of  the  tube  axis  and  the  separation \nbetween  them  is  4.7  mm.\n\nThe  sensitivity  of  the  tube  is  such  that  the  deflection  of  the  spot \nis  about  one  mm.  per  volt  applied  between  the  deflector  plates.    When\n\nusing  magnetic  deflection,  a  pair  of  coils  4  cm.  in  diameter  placed  on \nthe  sides  of  the  tube  at  the  level  of  the  deflector  plates  produces  a \ndeflection  of  approximately  1  mm.  per  ampere-turn  flowing  in  the  coils. \nThe  electrons  striking  the  screen  drift  back  to  the  anode  structure, \nand  most  of  them  are  collected  by  the  deflector  plates.  There  is  also \na  small  ionization  current  flowing  to  the  plates.  The  tube  is  therefore \nnot  strictly  an  electrostatic  device,  and  this  must  be  kept  in  mind \nwhen  using  it.  Fig.  4  shows  the  current  flowing  to  the  two  free  plates \nat  various  voltages  with  respect  to  the  anode.     With  the  large  posi-\n\ntive  values  of  plate  voltage  the  current  to  the  plates  is  practically \nequal  to  the  current  in  the  electron  stream  and  consists  largely  of  the \nreturning  electrons.  The  small  current  in  the  other  direction  when \nthe  plate  voltage  is  negative  is  a  measure  of  the  ionization  in  the  tube.\n\nThe  screen  is  spread  on  the  inner  surface  of  the  large  end  of  the \ntube,  using  pure  water  glass  for  binder.  The  active  material  consists \nof  equal  parts  of  calcium  tungstate  and  zinc  silicate,  both  specially \nprepared  for  fluorescence.  This  mixture  produces  a  generally  more \nuseful  screen  than  either  constituent  alone.  The  pure  tungstate \ngives  a  deep  blue  light  which  is  about  30  times  as  active  on  the  photo- \ngraphic plate  as  the  yellow-green  light  of  the  silicate,  while  the  silicate \ngives  a  light  which  is  many  times  brighter  visually  than  that  from  the \ntungstate.  By  mixing  the  two  materials  in  equal  parts  a  screen  is \nproduced  which  is  more  than  half  as  bright  visually  as  pure  zinc \nsilicate  and  more  than  half  as  active  photographically  as  pure  calcium \ntungstate.\n\nFor  mechanical  strength  the  end  of  the  bulb  which  carries  the  screen \nis  rounded  outwards  so  that  the  screen  is  not  a  plane  surface.  This \nintroduces  a  distortion  of  the  fluorescent  pattern  which  in  most  in- \nstances is  negligible.  If  the  pattern  is  recorded  by  a  camera  whose \nlens  is  D  cm.  from  the  end  of  the  tube,  then  the  apparent  reduction \nof  the  deflection  produced  by  the  curvature  of  the  bulb  is  given  in \nterms  of  the  deflection  y  approximately  by\n\nThe  part  which  the  gas  plays  in  focusing  the  beam  of  electrons \nis  an  interesting  phenomenon  which  depends  upon  the  difference  in \nthe  mobilities  of  electrons  and  positive  ions.  The  electrons  of  the \nbeam  are  pulled  toward  the  common  axis  by  a  radial  electric  field \nproduced  by  an  excess  of  positive  electricity  in  the  electron  stream \nand  an  excess  of  negative  electricity  in  the  space  outside  the  beam. \nThis  distribution  is  produced  as  follows:  Some  of  the  electrons  of \nthe  stream,  in  passing  through  the  gas,  collide  with  gas  molecules \nand  ionize  them.  Both  the  colliding  electrons  and  the  secondary \nelectrons  leave  the  beam  but  the  heavy  positive  ions  receive  very \nlittle  velocity  from  the  impact  and  drift  out  of  the  beam  with  only \ntheir  comparatively  low  thermal  velocity.     Positive  ions,  therefore,\n\naccumulate  down  the  length  of  the  stream  and  may  exceed  in  number \nthe  negative  charges  passing  along.  At  the  same  time,  electrons  are \nmoving  at  random  outside  the  stream,  producing  negative  electrifica- \ntion. There  is  then  a  field  surrounding  the  stream  which  tends  to \npull  the  electrons  inward.  If  there  were  only  the  mutual  repulsion \nbetween  the  electrons  to  compensate  for,  this  would  be  done  when \nthe  number  of  positive  ions  in  the  beam  equals  the  number  of  electrons. \nThere  is  in  addition  an  original  divergence  of  the  beam  which  must  be \novercome.  If  this  divergence  is  assumed  to  be  one  degree  from  the \naxis  and  the  electron  current  2  x  10~^  amp.,  then  a  simple  calculation \nshows  that  the  radial  field  required  to  pull  the  beam  to  a  focus  at  the \nusual  distance  is  about  one  volt  per  cm.  This  field  strength  is  pro- \nduced, with  beams  of  the  ordinary  intensity,  if  there  are  four  positive \nions  for  each  electron  in  the  stream,  a  Condition  which  seems  not \nunreasonable.\n\nThe  number  of  ions  per  electron  in  the  stream  is  probably  constant \nas  the  current  in  the  stream  is  varied,  since  the  conditions  of  collision \nand  recombination  are  not  altered.  When  the  current  is  increased, \ntherefore,  the  total  positive  ionization  of  the  beam  increases,  the  field \naround  the  beam  becomes  stronger,  and  the  electrons  are  brought  to \na  focus  in  a  shorter  distance.\n\nThese  deductions  have  been  confirmed  experimentally.  That  the \nfocusing  of  the  stream  depends  upon  the  current  flowing  was  one  of \nthe  earliest  observations  made  in  developing  the  tube  and  this  method \nhas  been  used  ever  since  to  obtain  a  sharp  spot.  The  point  of  con- \nvergence can  be  seen  moving  in  the  manner  expected  when  the  current \nis  changed,  and  the  effect  has  been  further  verified  by  using  a  tube  with \na  movable  fluorescent  screen  so  that  the  length  of  the  electron  beam \ncould  be  varied.  The  presence  of  the  electric  field  around  the  beam \nwas  shown  by  the  effect  of  two  beams  on  each  other,  in  a  tube  in \nwhich  there  were  two  electron  streams  crossing  each  other  at  right \nangles  at  their  mid-points,  each  falling  on  a  fluorescent  screen.  When \none  beam  was  moved  away  from  the  other  by  a  field  between  the \ndeflector  plates,  the  second  beam  moved  as  if  attracted  by  the  first. \nThe  directed  electrons  in  each  beam  were  attracted  toward  the  posi- \ntive ionization  in  the  other,  and  for  one  particular  adjustment  of  the \ntube  the  displacement  was  such  as  would  have  been  caused  by  a \nfield  ot  about  3  volts  per  cm.,  a  result  not  far  different  from  that \npreviously  calculated.\n\nSince  the  beam  must  produce  its  own  positive  ionization  some  time \nmust  elapse  before  it  can  produce  by  collisions  the  required  number \nof  positive  ions.     Calculation  shows  this  time  to  be  of  the  order  of\n\n10~*  second.  When  the  beam  moves  it  has  to  build  up  the  ionization \nas  it  goes  along,  and  we  should  expect  that  when  deflected  very \nrapidly  it  might  no  longer  be  focused,  due  to  lack  of  positive  ions  in \nits  path.  A  test  was  made  of  this  by  applying  a  high  frequency \npotential  on  the  deflector  plates  so  that  the  spot  described  an  elliptic \npattern.  At  a  frequency  of  10*  cycles  per  second  the  line  was  still \nsharp,  but  at  10*  cycles  there  was  a  noticeable  widening  of  the  line \nwhich  is  probably  to  be  ascribed  to  imperfect  focusing  at  this  high \nspeed .\n\nIn  these  experiments  the  evidence  all  points  to  the  view  that  the \nfocusing  of  the  electrons  is  caused  by  an  excess  of  positive  charge  in \nthe  beam  itself,  produced  by  ionizing  collisions  of  the  electrons  with \nthe  gas  molecules.  Further  confirmation  is  found  in  the  fact  that  a \nfocus  is  much  more  readily  obtained  in  the  heavier  gases  having  slow \nmolecules,  such  as  nitrogen,  argon  or  mercury  vapor,  than  in  hydrogen \nand  helium  where  the  mean  velocity  of  the  molecules  is  greater.  The \ntubes  are  therefore  filled  with  argon,  the  heaviest  available  permanent \ngas  which  does  not  attack  the  electrodes.  The  best  pressure  for  the \nlength  of  tube  adopted  and  for  the  current  which  can  be  obtained \nin  the  beam  is  5  to  10  microns,  and  this  leaves  considerable  latitude \nfor  the  adjustment  of  the  electron  current  to  get  a  sharp  focus.\n\nBecause  of  the  small  amount  of  auxiliary  apparatus  required  with \nthis  form  of  Braun  tube  it  has  proved  to  be  a  very  convenient  labora- \ntory instrument.  It  has  found  application  in  studying  the  behavior \nof  vacuum  tubes  and  amplifier  and  oscillator  circuits,  of  gas  discharge \ntubes,  of  relays,  and  of  numerous  other  kinds  of  apparatus,  both  at \nlow  and  at  high  frequencies.  Some  reproductions  of  photographs  of \nvarious  types  of  curves  are  given  below  to  illustrate  the  kind  of  results \nwhich  are  possible  with  this  oscillograph.\n\nFig.  5  shows  the  hysteresis  curve  of  a  sample  of  iron  wire.  The \nwire  was  placed  in  a  small  solenoid  with  one  end  toward  the  side  of \nthe  tube.  The  magnetizing  current  passed  through  a  resistance, \nthe  voltage  drop  of  which  was  applied  to  one  pair  of  deflector  plates \nso  as  to  give  a  deflection  proportional  to  the  magnetizing  field.  The \nstray  magnetic  field  from  the  iron  itself  produced  the  deflection \nproportional  to  the  induction.  Alternating  current  was  used,  and \nthe  exposure  was  20  seconds  with  lens  opening  /  6.3  and  speed  roll \nfilm.\n\noscillating  vacuum  tube.     The  axes  were  obtained  by  grounding  one \nor  the  other  deflector  element.\n\nbeen  reduced  to  a  fairly  simple  process  by  means  of  the  cathode  ray \ntube.     The    low    frequency    modulating   voltage,    controlled    by    the\n\nvoice,  is  applied  to  one  pair  of  deflector  plates,  while  the  radio  fre- \nquency output,  with  amplitude  varying  according  to  the  low  frequency \nvoltage,  is  applied  to  the  other  pair  of  deflector  plates.  The  resulting \npattern  on  the  screen  is  a  quadrilateral  of  solid  fluorescence,  since  the\n\ntwo  frequencies  are  not  commensurate.  The  two  vertical  sides \nindicate  the  greatest  and  the  least  amplitude  of  the  high  frequency, \nwhile  the  other  two  sides  show  the  current-voltage  characteristic  of \nthe  transmitter.  Fig.  7  shows  such  a  pattern  (retouched) ,  the  edges \nbeing  much  brighter  than  the  centre.  The  exposure  was  two  minutes \nusing  a  Seed  23  plate  and  /  6.8  lens  opening.\n\nGeorge  A.  Campbell,  B.S.,  Massachusetts  Institute  of  Technology, \n1891;  A.B.,  Harvard,  1892;  Ph.D.,  1901;  Gottingen,  Vienna  and \nParis,  1893-96.  Mechanical  Department,  American  Bell  Telephone \nCompany,  1897;  Engineering  Department,  American  Telephone  and \nTelegraph  Company,  1903-1919;  Department  of  Development  and \nResearch,  1919 \u2014 ;  Research  Engineer,  1908 \u2014 .  Dr.  Campbell  has \npublished  papers  on  loading  and  the  theory  of  electric  circuits  and  is \nalso  well-known  to  telephone  engineers  for  his  contributions  to  re- \npeater and  substation  circuits.  The  electric  filter  which  is  one  of  his \ninventions  plays  a  fundamental  role  in  telephone  repeater,  carrier \ncurrent  and  radio  systems.\n\nRalph  V.  L.  Hartley,  A.B.,  Utah,  1909;  B.A.,  Oxford,  1912; \nB.Sc,  1913;  instructor  in  physics,  Nevada,  1909-10;  Engineering \nDepartment,  Western  Electric  Company,  1913 \u2014 .  For  some  time \nMr.  Hartley  has  been  closely  connected  with  the  development  of \ncarrier  current,  telephone  repeater,  and  telegraph  systems.\n\nThornton  C.  Fry,  A.B.,  Findlay,  1912;  A.M.,  University  of  Wis- \nconsin, 1913;  Ph.D.,  1920;  instructor  of  mathematics,  Wisconsin, \n1912-16;  Engineering  Department,  Western  Electric  Company, \n1916 \u2014 .  Mr.  Fry  has  written  several  papers  on  the  theory  of  electric \ncircuits  and  other  subjects  allied  to  telephony.\n\nJohn  R.  Carson,  B.S.,  Princeton,  1907;  E.E.,  1909;  M.S.,  1912; \nResearch  Department,  Westinghouse  Electric  and  Manufacturing \nCompany;  1910-12;  instructor  of  physics  and  electrical  engineering, \nPrinceton,  1912-14;  American  Telephone  and  Telegraph  Company, \nEngineering  Department,  1914-15;  Patent  Department,  1916-17; \nEngineering  Department,  1918;  Department  of  Development  and \nResearch,  1919 \u2014 .  Mr.  Carson's  work  has  been  along  theoretical  lines \nand  he  has  published  several  papers  on  theory  of  electric  circuits  and \nelectric  wave  propagation.\n\nR.  L.  Wegel,  A.B.,  Ripon  College,  1910;  assistant  in  physics, \nUniversity  of  Wisconsin,  1910-12;  physicist  with  T.  A.  Edison,  1912- \n13;  Engineering  Department  of  Western  Electric  Company,  1914 \u2014 . \nMr.  Wegel  has  been. closely  associated  with  the  development  of  tele- \nphone transmitters  and  receivers,  and  has  made  important  contribu- \ntions to  the  theory  of  receivers.\n\nEdward  C.  Molina,  Engineering  Department  of  the  American \nTelephone  and  Telegraph  Company,  11)01-19,  as  engineering  assist- \nant; transferred  to  the  Circuits  Design  Department  to  work  on  ma- \nchine switching  systems,  1905;  Department  of  Development  and \nResearch,  1919 \u2014 .  Mr.  Molina  has  been  closely  associated  with \nthe  application  of  the  mathematical  theory  of  probabilities  to  trunking \nproblems  and  has  taken  out  several  important  patents  relating  to \nmachine  switching.\n\nWilliam  C.  Helmle,  B.  S.,  University  of  Wisconsin,  1917;  Uni- \nversity of  Chicago,  1919-20;  Commercial  Engineer's  Ofifice,  American \nTelephone  and  Telegraph  Company  1920 \u2014 .\n\nE.  T.  HocH,  B.S.,  in  Electrical  Engineering,  Case  School  of  Ap- \nplied Science,  1914;  Western  Electric  Company,  Manufacturing  and \nInstallation  Departments,  1914-15;  Engineering  Department,  1915 \u2014 .\n\nLloyd  Espenschied,  Pratt  Institute,  1909;  United  Wireless  Tele- \ngraph Company  as  radio  operator,  summers,  1907-08;  Telefunken \nWireless  Telegraph  Company  of  America  assistant  engineer,  1909- \n10;  American  Telephone  and  Telegraph  Company,  Engineering \nDepartment  and  Department  of  Development  and  Research,  1910 \u2014 . \nTook  part  in  long  distance  radio  telephone  experiments  from  Wash- \nington to  Hawaii  and  Paris,  1915;  since  then  his  work  has  been  con- \nnected with  the  development  of  radio  and  carrier  systems.\n\nJ.  B.  Johnson,  B.S.,  University  of  North  Dakota,  1913;  M.S., \n1914;  Ph.D.,  Yale,  1917;  Engineering  Department,  Western  Elec- \ntric Company,  1917 \u2014 .  Since  coming  to  the  Western  Electric  Com- \npany, Mr.  Johnson  has  devoted  much  time  to  high  vacua  and  ioniza- \ntion in  gases.\n\nApplication  of  Carrier  Telephone  and  Telegraph  in  the  Bell  Sj-stem,  Arthur  F.\n\nRose,  Vol.  II,  No.  2,  page  41. \nApplication  to  Radio  of  Wire  Transmission  Engineering,  Lloyd  Espenschicd,  Vol.\n\nI,  No.  2,  page  117. \nArnold,  H.  D.,  Permalloy,  A  New  Magnetic  Material  of  Very  High  Magnetic\n\nTransatlantic  Radio  Telephony,  Vol.  II,  No.  4,  page  116. \nAudition,    Physical   Characteristics   of,   R.   L.    Wegel,   Vol.    I,    No.  2,   page   56. \nAudition,  Physical  Measurements  of,  Harvey  Fletcher,  Vol.  II,  No.  4,  page  145.\n\nBateman,  Helene  C,  A  Method  of  Graphical  Analysis,  Vol.  II,  No.  3,  page  77. \nBinaural  Location  of  Complex  Sounds,  R.  V.  L.  Hartley  and  Thornton  C.  Fry, \nVol.  I,  No.  2,  page  33.\n\nCable,  Philadelphia-Pittsburgh  Section  of  the  New  York-Chicago,  J.  J.  Filliod,\n\nVol.  I,  No.  1,  page  60. \nCable    Circuits,   Telephone   Transmission   Over,   A.   B.    Clark,   Vol.    II,    No.    1,\n\npage.  67. \nCable  Circuits,  Telephone  Equipment  for,  Charles  S.  Demarest,  Vol.  II,  No.  3,\n\nProbability   Curves   Showing    Poisson's   Exponential    Summation,   Vol.    II, \nNo.  1,  page  95. \nCarrier  and  Side  Bands  in  Radio  Transmission,  R.  V.  L.  Hartley,  Vol.  II,  No.  2,\n\npage  90. \nCarrier  Telephone  and  Telegraph  in  the  Bell  System,  Practical  Application  of,\n\nArthur  F.  Rose,  Vol.  II,  No.  2,  page  41. \nCarson,  John  R.,  Heaviside  Operational  Calculus,  Vol.  I,  No.  2,  page  43.\n\nTransmission  Characteristics  of  Submarine  Cable,  Vol.  I,  No.  1,  page  88. \nCharlesworth,  H.  P.,  Machine  Switching  Telephone  System  for  Large  Metro- \npolitan Areas,  Vol.  II,  No.  2,  page  53. \nClark,  A.  B.,  Telephone  Transmission  Over  Long  Cable  Circuits,  Vol.  II,  No.  1, \npage  67.\n\nUse   of    Public   Address    System   with    Telephone   Lines,    Vol.    II,    No.   2, \npage  143. \nComplex  Sounds,  the  Binaural  Location  of,  R.  V.  L.  Hartley  and  Thornton  C.\n\nFry,  Vol.  I.,  No.  2,  page  33. \nCraft,    E.    B.,   Machine    Switching  Telephone    System    for   Large    Metropolitan\n\nAreas,  Vol.  II,  No.  2,  page  53. \nCrandall,  I.  B.,  Analysis  of  the  Energy  Distribution  in  Speech,  Vol.  I,  No.  1,\n\npage  101. \nDemarest,  Charles   S.,  Telephone   Equipment   for  Long  Cable  Circuits,  Vol.   II,\n\nPermeability,  Vol.  II,  No.  3,  page  101. \nEnergy  Distribution,  Analysis  of   in   Speech,  /.  B.   Crandall  and  D.  Mackenzie,\n\nVol.  I,  No.  1,  page  116. \nEspenschied,   Lloyd,   Application   to   Radio   of   Wire  Transmission   Engineering, \nVol.  I,  No.  2,  page  117.\n\nTransatlantic  Radio  Telephony,  Vol.  II,  No.  4,  page  116. \nExternal  Ear,  Dynamical  Analysis  of,  R.  L.  Wegel,  Vol.  I,  No.  2,  page  56.\n\nFletcher,  Harvey,  The  Nature  of  Speech  and  Its  Interpretation,  Vol.  I,  No.  1, \npage  129. \nPhysical  Measurements  of  Audition  and  Their  Bearing  on  the  Theory  of \nHearing,  Vol.   II,  No.  4,  page  145. \nFry,  Thornton  C,  The  Binaural  Location  of   Complex  Sounds,  Vol.  I,  No.  2, \npage  33.\n\nGases  Evolved  from  Glasses  of  Knov^^n  Chemical  Composition,  Measurements  of, \nJ.  E.  Harris  and  E.  E.  Schumacher,  Vol.  II,  No.   1,  page  122.\n\nGilbert,  J.  J.,  Transmission  Characteristics  of  Submarine  Cable,  Vol.  I,  No.  1, \npage  88.\n\nHartley,  R.  V.  L.,  Relation  of  Carrier  and  Side  Bands  in  Radio  Transmission, \nVol.  II,  No.  2,  page  90. \nThe  Binaural  Location  of  Complex  Sounds,  Vol.  I,  No.  2,  page  ii.\n\nHarris,  J.  E.,  Measurements  of  the  Gases  Evolved  from  Glasses  of  Known  Chem- \nical Composition,  Vol.  II,  No.  1,  page  122.\n\nHelmle,  W.  C,  The  Relation  Between  Rents  and  Incomes  and  the  Distribution  of \nRental  Values,  Vol.  I,  No.  2,  page  82.\n\nHoyt,  Ray  S.,  Impedance  of  Smooth  Lines  and  Design  of  Simulating  Networks, \nVol.  II,  No.  2,  page  1.\n\nImpedances  of  Smooth  Lines  and  Design  of  Simulating  Networks,  Ray  S.  Hoyt, \nVol.  II,  No.  2,  page  1.\n\nInductive  Eflfects  in  Neighboring  Communication  Circuits,  the  Relation  of  Peter- \nson System  of  Grounding  Power  Networks  to,  //.  M.  Truchlaod,  Vol.  I, \nNo.  1,  page  39.\n\nKing,  R.  \\\\'.,  Thermionic  Vacuum  Tubes  and  Their  Uses,  Vol.  II,  No.  4,  page  31. \nKirk,  J.  N.,  Bell  System  Sleet  Storm  Map,  Vol.  II,  No.  1,  page  114.\n\nSpecializing  Transportation  Equipment  in  Order  to  Adapt  it  Most  Econom- \nically to  Telephone  Construction  and  Maintenance  Work,  Vol.  II,  No.  1, \npage  47. \nUse  of  Labor-Saving  Apparatus  in  Outside  Plant  Construction  Work,  Vol. \nII,  No.  3,  page  53.\n\nLabor-Saving  Apparatus  in  Outside  Plant  Construction  Work,  /.  A'^.  Kirk,  Vol. \nII,  No.  3,  page  53.\n\nMachine  Switching  Telephone  System  for  Large  jMetropolitan  Areas,  E.  B. \nCraft,  L.  F.  Morehouse  and  H.  P.  Charlesivorth,  Vol.  II,  No.  2,  page  53.\n\nMartin,  W.  H.,  Use  of  Public  Address  System  with  Telephone  Lines,  Vol.  II, \nNo.  2,  page  143.\n\nMeasurements  of  the  Gases  Evolved  from  Glasses  of  Known  Chemical  Com- \nposition, /.  E.  Harris  and  E.  E.  Schumacher,  Vol.  II,  No.  1,  page  122.\n\nMolina,  Edward  C,  The  Theory  of  Probabilities  Applied  to  Telephone  Trunking \nProblems,  Vol.  I,  No.  2,  page  69.\n\nMorehouse,  L.  F.,  Machine  Switching  Telephone  System  for  Large  Metropolitan \nAreas,  Vol.  II,  No.  2,  page  53.\n\nNew  York-Chicago  Cable,  Philadelphia-Pittsburgh  Section  of,  /.  /.  Pilliod,  Vol.\n\nOperational  Calculus,  Heaviside,  John  R.  Carson,  Vol.  I,  No.  2,  page  43. \nOscillations,  Transient,  in  Electric  Wave-Filters,  /.  R.  Carson  and  0.  J.  Zohel,\n\nVol.  II,  No.  3,  page  1. \nOscillograph,  A  Low  Voltage  Cathode  Ray,  J .  B.  Johnson,  Vol.  I,  No.  2,  page\n\nPermalloy,  A  New  Magnetic  Alaterial  of  Very  High  Magnetic  Permeability,  H. \nD.  Arnold  and  G.  W.  Elmen,  Vol.  II,  No.  3,  page, 101.\n\nPeterson  System  of  Grounding  Power  Networks,  the  Relation  to  Inductive  Ef- \nfects in  Neighboring  Communication  Circuits,  H.  M.  Trueblood,  Vol.  I, \nNo.  1,  page  39.\n\nPhiladelphia-Pittsburgh  Section  of  the  New  York-Chicago  Cable,  /.  /.  Pilliod, \nVol.  I,  No.  1,  page  60.\n\nPhysical  Characteristics  of  Audition  and  Dynamical  Analysis  of  the  External \nEar,  R.  L.  Wegel,  Vol.  I,  No.  2,  page  56.\n\nPhysical  Measurements  of  Audition  and  Their  Bearing  on  the  Theory  of  Hear- \ning, Harvey  Fletcher,  Vol.  II,  No.  4,  page  145.\n\nPilliod,  J.  J.,  Philadelphia-Pittsburgh  Section  of  the  New  York-Chicago  Cable, \nVol.  I,  No.  1,  page  60.\n\nPoisson's  Exponential  Summation,  Probability  Curves  Showing,  G.  A.  Campbell, \nVol.  II,  No.  1,  page  95.\n\nProbabilities  Applied  to  Telephone  Trunking  Problems,  The  Theory  of,  Edzvard \nC.  Molina,  Vol.  I,  No.  2,  page  69.\n\nProbability  Curves  Showing  Poisson's  Exponential  Summation,  G.  A.  Campbell, \nVol.  II,  No.  1,  page  95.\n\nPublic  Address  System,  Use  of.  The  Telephone  Lines,  W.  H.  Martin  and  A.  B. \nClark,  Vol.  II,  No.  2,  page  143.\n\nRadio  Extension  of  the  Telephone  System  to  Ships  at  Sea,  H.  W.  Nichols  and \nLloyd  Espenschied,  Vol.  II,  No.  3,  page  141.\n\nRadio  Telephony,  Transatlantic,  H.  D.  Arnold  and  Lloyd  Espenschied,  Vol.  II, \nNo.  4,  page  116.\n\nRents  and  Incomes  and  the  Distribution  of  Rental  Values,  The  Relation  Be- \ntween, W.  C.  Hchnlc,  Vol.  I,  No.  2,  page  82.\n\nRose,  Arthur  P.,  Practical  Application  of  Telephone  and  Telegraph  in  the  Bell \nSystem,  Vol.  II,  No.  2,  page  41.\n\nChemical  Composition,  Vol.  II,  No.  1,  page  122. \nSleet  Storm  Map,  Bell  System,  /.  A^.  Kirk,  Vol.  II,  No.  1,  page  114.\n\nSome  Con  temporal'  Advances  in  Physics,  A.'.  K.  Darroiv,  Vol.  II,  No.  4,  page  101. \nSpeech,  Analysis  of  Energy  Distribution  in,  /.  B.  Crandall  and  D.  Mackenzie,\n\nVol.  I,  No.  1,  page  116. \nSpeech,  The  Nature  of  and  Its  Interpretation,  H.  Fletcher,  Vol.  I,  No.  1,  page  129. \nSubmarine  Cable,  Transmission  Characteristics  of,  /.  R.  Carson  and  /.  /.  Gilbert,\n\nNo.  3,  page  112. \nTheory  and  Design  of  Uniform  and  Composite  Electric  Wave  Filters,  Otto  J.\n\nZobel,  Vol.  II,  No.  1,  page  1. \nTheory,  Electric  Wave  Filter,  G.  A.  Campbell,  Vol.  I,  No.  2,  page  1. \nThermionic  Vacuum  Tubes  and  Their  Uses,  R.  IV.  King,  Vol.  II,  No.  4,  page \nTransatlantic  Radio  Telephony,  H.  D.  Arnold  and  Lloyd  Espenschied,  Vol.  II,\n\nNo.  4,  page  116. \nTransient  Oscillations  in  Electric  Wave-Filters,  /.  R.  Carson  and  0.  /.  Zobel,\n\nVol.  II,  No.  3,  Page  1. \nTransmission  Characteristics  of  the   Submarine  Cable,  /.  R.   Carson  and  /.  /.\n\nGilbert,  Vol.  I,  No.  1,  page  88. \nTransmission  Over  Long  Cable  Circuits,  A.  B.  Clark,  Vol  II,  No.  1,  page  67. \nTransportation  Equipment,  Specializing,  In  Order  to  Adapt  It  Most  Economically\n\n1,  page  47. \nTrueblood,  H.  M.,  The  Relation  of  the  Peterson  System  of  Grounding  Power\n\nNetworks  to  Inductive  Effects  in  Neighboring  Communication  Circuits,  Vol.\n\nI,  No.  1,  page  39. \nTrunking  Problems,  The  Theory  of  Probabilities  Applied  to,  Edward  C.  Molina,\n\nVacuum  Tube,  a  New  Type  of  High-Power,  W.  Wilson,  Vol.  I,  No.  1,  page  4. \nVacuum  Tube,  Thermionic,  R.  W.  King,  Vol.  II,  No.  4,  page  31.\n\nNo.  3,  page  1. \nWegel,  R.  L.,  Physical  Characteristics  of  Audition  and  Dynamical  Analysis  of\n\nthe  External  Ear,  Vol.  I,  No.  2,  page  56. \nWilson,  W.,  A  New  Type  of  High-Power  Vacuum  Tube,  Vol.  I,  No.  1,  page  4. \nWire  Transmission  Engineering,  Application  to  Radio,  Lloyd  Espenschied,  Vol.\n\nZobel,  Otto  J.,  Theory  and  Design  of  Uniform  and  Composite  Electric  Wave \nFilters,  Vol.  II,  No.  1,  page  1. \nTransient  Oscillations  in   Electric  Wave-Filters,  Vol.  II,  No.  3,  page  1.", "title": "The Bell System Technical Journal, Volume 1", "trim_reasons": [], "year": 1922}
{"archive_ref": "pat-us4135240", "canonical_url": "https://patents.google.com/patent/US4135240A/en", "char_count": 17169, "collection": "archive-org-bell-labs", "doc_id": 105, "document_type": "patent", "id": "bella-qwen-pretrain-doc105", "record_count": 1, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": "pdftotext", "selected_extraction_score": 1.0, "source_family": "google_patents", "source_url": "https://patents.google.com/patent/US4135240A/en", "split": "test", "text": "[54] PROTECTION OF DATA FILE CONTENTS\n[75] Inventor: Dennis M. Ritchie, Summit, N.J.\n\n[73] Assignee: Bell Telephone Laboratories,\nIncorporated, Murray Hill, N.J.\n\n[21] Appl. No.: 377,591\n[22] Filed: Jul. 9, 1973\n\n[SU] Unt, C02 oonceccceeccsssescoeees GOG6F 11/10; GOGF 13/00\n\n52] U.S. CL ccccseccssccssssecssscccssssecsssecsssuecossesesnee 364/200\n\n[58] Field Of Search o..scccccccsssssssssssscscssssonee 340/172.5;\n364/200 MS File, 900 MS File\n\n[56] References Cited\nU.S. PATENT DOCUMENTS\n\nLU/1971 \u00ab9 Ullrich .u....0............eccceneseeee 340/172.5\n12/1971 Amdahl et al. ................... 340/172.5\n2/1968  Beausoleil et al. ................ 340/172.5\n4/1968 Nelson et al. ...............c0000. 340/ 172.5\n9/1969 Richmond ...................... 340/172.5\n4/1971 Cordero et al. .........0.....0... 340/172.5\n8/1971 Creech et all. ..........cccc:ceee 340/172.5\nW2/1971 \u2014 Hofh oon. c eee ccee cece cccepsecene ences 364/200\n\n- GB ,\n\nRe. 27,239\nRe. 27,251\n3,368,207\n3,377,624\n3,469,239\n3,576,544\n3,599,159\n3,631,405\n\nSTORED DATA e02\n\n+\n\n205 -~> a\nR ID\n\n204\n\nCOMPUTER\n\n7\n207\nACTUAL\nUSER ID\n\n[45] Jan. 16, 1979\n\n3,683,418 \u00ab= 8/1972 Martin 2.00... sseccosescceoneseee 340/172.5\n3,735,364 S/197S =\u2014\u00a7-_ attr once cccccseersenes 340/172.5\n3,742,458 6/1973 Inoue et al... cee 340/172.5\n3,761,883 9/1973 A] VALET ..........ceessecersreccennrcen 364/200\n\nPrimary Examiner\u2014James D. Thomas\nAttorney, Agent, or Firm\u2014Stephen J. Phillips\n\n[57] ABSTRACT\n\nAn improved arrangement for controlling access to\ndata files by computer users. Access permission bits are\nused in the prior art to separately indicate permissions\nfor the file owner and nonowners to read, write and\nexecute the file contents. An additional access control\nbit is added to each executable file. When this bit 1s set\nto one, the identification of the current user is changed\nto that of the owner of the executable file. The program\nin the executable file then has access to all data files\nowned by the same owner. This change is temporary,\nthe proper identification being restored when the pro-\ngram is terminated.\n\n4 Claims, 2 Drawing Figures\n\nMEMORY\n\n201 ;\n\nAC\nAE\n\nEXECUTE\nSEQUENCE\n\nU.S. Patent 4,135,240\n\nSheet 1 of 2\n\nFIG. [At\n\nUSER PASSWORD a D PROGRAM 2\n\nJan. 16, 1979\nBOB LXR2\nTED FRTE\n\nJIM STPA\n\n10\nlot SUID BIT 07\n102 OWNER ID 18\n\nOWNER\n03 RD WR EX\nre ee\n\nNON- OWNERS\nWR EX\n\n\u201coto\n'PROGI\" FILE\nCONTENTS\n\nI\nSUID BIT 0.\nOWNER ID  \u00a7 |\n\nRD WR\npot | tt\n\nNON-OWNERS\nEX\n\n104\n\n105\n\nRD WR\n\"EDIT\" FILE\nCONTENTS\n\n12\nSUID BIT a\nOWNER ID\n\nOWNER\nWR\n\nRD EX\ntt tt\n\nNON-OWNERS\nWR EX\n\n\"PROLL\" FILE\nCONTENTS\n|\n\nPASSWORD FILE\n\nFILE STORAGE\n\nPROLL\n\n8 PROG|\n6 EDIT\n\n13\n\nSUID BIT\n\nOWNER ID\n\nOWNER\nRD WR EX\n\naE :\n\u00bb NON DRNERS\n\nCONTENTS\n\nE\nSUID BIT. 0\u201d\nOWNER ID | 6.\n\nOWNER\nRD WR EX\n\nNON-OWNERS\nRD WR EX\n\n\"BFILE\" FILE\nCONTENTS\n|\n\naT BIT 2.\nOWNER ID\nOWNER\nRD WR EX\ntit {| ji [| 0\nNON-OWNERS\nRD wR EX\no | o | 0\nCFILE\" FILE\n\nCONTENTS\n\nTd\n\nU.S. Patent Jan. 16, 1979 Sheet 2 of 2 4,135,240\n\nSTORED DATA 2\u00b0 col\nUSER ID\nOWNER ID STORED\nSUID BIT \\NSTRUCTIONS\n\nOWNER RD WR EX\nNON-OWNERS RD WR EX\n\n218 | INSTRUCTION\n\nLOCATION\nCOUNTER\n\nACCESS\nDENIED\n\nINSTRUCT ION\nDECODER\n\nEXECUTE FILE\n\nACCESS\nREQUEST\n| 216~} COMPARATOR ACCESS\nOWNER ID\nUID SIT | a REQUEST\n| JOWNER_ID EXECUTE\n220 SEQUENCE\n1 2084 EFFECTIVE\nUSER ID\n| 2005) (225K\nluseR 1p _| LOGIN\n204\u2014Sts\u00ab* SEQUENCE\n\n207 3\nACTUAL 4\nUSER ID | .\nFILE\nCOMPUTER CONTROL\n\n4,135,240\n\n1\n\nPROTECTION OF DATA FILE CONTENTS\n\nBACKGROUND OF THE INVENTION\n\n1. Field of the Invention\n\nThis invention relates to computer systems, and more\nparticularly, to computer systems having multiple users\nand multiple data files.\n\n2. Description of the Prior Art\n\nComputer systems are more efficiently operated\nwhen there are multiple users, and file storage devices\nare more efficiently used when many users share stor-\nage space on the same physical device. Each user then\nhas the potential for accessing files belonging to other\nusers. Free access is not generally permissable since files\nmay contain programs or data of sensitive nature.\n\nVirtually all computer systems provide means for\nprotecting sensitive files against access by users legiti-\nmately present in the computer system but not autho-\nrized to use all files. Hardware or software control\nmechanisms are provided to decide at the time of a user\nrequest for file access whether access permission is to be\ngranted or denied. In general the information necessary\nfor this decision are (1) file identity (2) user identity and\n(3) access purpose.\n\nComputer systems have been designed which include\nelaborate lists identifying which users are permitted to\naccess which files for which purposes. The result is a\ncomplex internal bookeeping task. As users share pro-\ngrams and data, the lists of permitted functions must be\ninterchanged. See the article \u201cDynamic Protection\nStructures\u201d by B. W. Lampson, AFIPS Fall Joint Com-\nputer Conference, 1969, pp. 27-38. The scheme de-\nscribed by Lampson solves the access permission prob-\nlem in a general way, but the result is so complex that it\nhas not found wide acceptance in the computer field.\n\nThis improvement is addressed to the simpler\nschemes which are in wide use. Each user of the com-\nputer system is preassigned an identification number\n(user ID). Whenever a user creates a file by reserving\nfile space for his own use, his user ID ts stored along\nwith the file to identify the file owner (owner ID). In\ncreating the file, the owner also specifies certain permis-\nsions which are to be granted or denied to himself as\nowner, and to everyone else as nonowners. Generally,\nthese permissions are for reading and for writing the\nfile. This information may be contained in as few as four\nbinary digits or \u201cpermission bits,\u201d a modest addition to\neach stored file. Also, in systems having a common\nformat for files containing programs and files contain-\ning data, it is usual to have permission information to\n\n2\n\ncomputer usage by the various users of the system. The\naccounting programs and the accounting files are\nowned by the same user who has permission to read and\nwrite the accounting file to permit regular updates.\nSuppose now that it is desired to permit each user to\nread from the accounting file the information associated\nwith that user\u2019s own computer usage. This 1s certainly a\nlegitimate access purpose so long as the user does not\nattempt to read other accounting information which ts\n\n10 considered private as far as he is concerned.\n\n|)\n\n20\n\n25\n\n30\n\n33\n\n45\n\nindicate that the file contents may or may not be loaded \u2014\n\ninto the computer and executed as a program. This may\ncomprise an additional execute permission bit, or an\nadditional two bits, separate permissions for owners and\nnonowners.\n\nThe described scheme takes into account file identity\nbecause access control information ts stored in associa-\ntion with each file individually. User identity is taken\ninto account in a gross but useful distinction between\nowner and nonowner. Access purpose is also a factor\nbecause of the coarse selection between reading, writ-\ning and execution permissions.\n\nA shortcoming with this scheme is its lack of ability\nto include fine distinctions of access purpose. Consider,\nfor example, the problem of accessing a computer time\nusage accounting data file. Such a file is used by com-\nputer time accounting programs to store elapsed time of\n\n35\n\n65\n\nUnder the described scheme there is no simple way to\npermit this kind of special purpose data file access. A\ngeneral user wishing to read the accounting file cannot\ndo so directly because he will not have nonowner per-\nmission to read. He cannot execute the general account-\ning programs to read for him and return the information\nbecause he will not have nonowner permission to exe-\ncute the general accounting programs. Such permis-\nsions must generally be denied to nonowners to assure\nprivacy of the accounting file contents. This problem is\nfurther described in the article \u201cMOO in Multics\u201d by J.\nM. Grochow, Software - Practice and Experience, Vol.\n2, pp. 303-308 (1972).\n\nSUMMARY OF THE INVENTION\n\nThe present invention adds a facility to the basic\nprotection scheme just described which permits com-\nputer users to access a data file for any specific purpose.\nThis is done by providing for the execution of a com-\nputer program to access the file, which program is sup-\nplied by the file owner and thus can impose any degree\nof control which the file owner wishes to include. This\nnew facility uses an additional file access control bit for\neach stored file of executable program. This additional\nbit is termed the \u201cset user identification bit\u201d (SUID bit).\n\nThe user ID which is stored by the computer and ts\neffective to control subsequent file access is changed\nwhenever a stored file containing an executable pro-\ngram (executable file) is loaded into computer memory\nfor execution and whenever the associated SUID bit is\nset to one. The effective user ID is changed from that of\nthe actual user to that of the owner of the executable\nfile. During the execution of the program, therefore, the\ncurrent user appears to be the owner of the executable\nfile and all of the data files accessible to the owner of the\nexecutable file are available to the program. The user\nmay request the program to access those data files, and\nthe program will operate to satisfy that access request in\nthe manner it was designed to do, making whatever\ntests and restricting access in any manner intended by\nthe program designer, the actual owner of the execut-\nable file and the data files. For the duration of the pro-\ngram execution the change in user ID ts effective. When\nthe program is terminated, as for example the attempted\nexecution of a new program, the user ID of the actual\nuser is restored.\n\nUnder this improved scheme, the problem of ac-\ncounting file access 1s easily solved. The computer user\nwho owns the accounting programs and accounting file\nprovides a special program for nonaccounting users\nwhich reads the accounting file. This special program\nreads the user ID of the actual current user and com-\npares this with the user ID for the accounting file re-\ncord sought to be read. If they match, the information\nconcerns the requesting user and can therefore be re-\nturned to him. This special program is stored in a file\nwhich has nonowner permission for execution, and\nwhich has the SUID bit set to one.\n\n4,135,240\n\n3\n\nWhen the general user executes the special program,\nthe SUID bit causes the effective user ID to be changed\nto the owner ID of the special program, the accounting\nuser ID. Thus, during the execution of the special pro-\ngram, access to the accounting files is allowed by the\nowner permission bits of the accounting file. Now the\nuser requests the special program to read the account-\ning file. The special program has the proper permission,\nbut the action of the special program ts determined by\nthe accounting user who designed the special program.\nThe special program therefore reads the actual user [ID\nof the requesting user and only returns to him the ac-\ncounting information from the accounting file which\nrelates to himself. The general user can therefore access\nthe accounting files only for the specific bona fide pur-\npose for which the special program was provided by the\naccounting user. After the execution of the special pro-\ngram is terminated, the effective user ID is restored to\nthe user ID of the actual user.\n\n10\n\n15\n\nThe accounting file access problem ts exemplary of 20\n\nthe type of problem this new facility alleviates. Other\napplications will become apparent from the following\ndescription of one embodiment of the invention.\n\nBRIEF DESCRIPTION OF THE DRAWING\n\nTaken together,\n\nFIGS. 1a and 15 comprise a single Figure showing a\ncomputer system embodying the present invention.\n\nFIG. 1a illustrates a plurality of files stored in a com-\nputer storage device, having access control information\n\nFIG. 10 illustrates a digital computer and its memory\nwhich operate in conjunction with the files stored in the\nprevious Figure to embody the present invention.\n\nDETAILED DESCRIPTION\n\nThe drawing shows in a single Figure (comprising\nFIGS. 1\u00a2 and 16 together) a computer system compris-\ning computer 1 which accesses file storage Z by means\n\n25\n\n30\n\n35\n\nof file control 4 and accesses memory 3 by means of 40\n\nmemory control 5. Files 10, 11, and 12 contain stored\nprogram information and are read from file storage 2\ninto memory 3 for execution by computer 1. Files 15,\n14, 15 and 16 contain stored data information and are\nread from file storage 2 into memory 3 in order that the\nstored data contents may be accessed. Computer 1 is\ncontrolled, for the most part, by instructions read from\nmemory 3 and executed by instruction decoder 6. In-\nstruction location counter 7 controls the location within\nmemory 3 of the stored instruction to be next executed\nby computer 1.\n\nIn computer systems, it is common practice to refer\nto files of programs or data by means of arbitrarily\nchosen symbolic reference names. In keeping with this\npractice files 10 through 16 will hereinafter be referred\nto by such symbolic names as they appear in the Figure,\ne.g., PROG1, EDIT, PROLL, AFILE, BFILE,\nCFILE and PASSWORD, respectively. For conve-\nnience, these names will also be used to denote program\nor data contents of the respective files as well as the files\nthemselves. Thus PROG1 will be used to refer to file 10\nas it appears in file storage 2 and also to the program\ncontained in file PROGI after being read into memory\n3 for execution by computer 1.\n\nAs will become apparent the program PROGI regu-\nlarly accesses the data AFILE, the program EDIT\nregularly accesses data BFILE and the program\nPROLL regularly accesses data CFILE. Each of these\n\n45\n\n55\n\n65\n\n4\n\nsix files has associated with them various access control\ninformation including: set user identification bit 101\n(SUID bit), owner identification number 102 (owner\nID) owner permission bits 103 and nonowner permis-\nsion bits 104. This information controls access to stored\nfile contents 105 in a manner to be described.\n\nEach user of the computer system 1s identified by a\nunique preassigned user identification number 106 (user\nID) which is retained in the PASSWORD file and\nwhich is retrieved when the user begins requesting\ncomputer services. A user may create a new file to\ncontain data or program by reserving space in file stor-\nage 2 for that purpose. The owner ID of the new file is\nthen set to be equal to the user ID of the creating user.\nThus, the creating user is identified as the owner of the\nfile. When the new file is thereafter to be accessed, the\nowner ID is compared with the user ID of the request-\ning user. If they match, owner permission bits 103 con-\ntrol file usage; if they do not match, nonowner permis-\nsion bits 104 are used. Permission bits 103 and 104 are\nset to values prescribed by the owner when the file is\ncreated.\n\nThere are three permission bits each for owners and\nnonowners labeled RD, WK and EX in the Figure\ncorresponding to read, write and execute permission,\nrespectively. When a permission bit is set to 1, the asso-\nciated function is permitted; when set to 0, the function\nis denied.\n\nIn the Figure read, write and execute permission, are\ngranted for the respective owners of PROGI, EDIT,\nand PROLL. Thus user \u201cTED\u201d with user ID equal to\n18 in the PASSWORD file is permitted to read from the\ncontents of the PROGI file, write into the PROGI file,\nand load the contents of the PROG! file for execution\nas a program. Similarly, AFTLE, BFILE and CFILE\nhave permission for reading and writing by their respec-\ntive owners. Execution permission is denied to both\nowners and nonowners of AFILE, BFILE and CFILE\nsince these files contain data and now executable pro-\ngram instructions.\n\nIn the Figure al] nonowner permissions are denied for\nPROGI and AFILE. Thus only user \u201cTED\u201d, the\nowner of PROG1 and AFILE, may access them. If user\n\u201cTED\u201d executes PROGI, and if PROG1 contains ap-\npropriate read and write instructions, PROG1 would be\ncapable of reading and writing AFILE. PROGI could\ntherefore represent a program written by user \u201cTED\u201d\nfor maintaining AFILE as a file of private data.\n\nNonowners are permitted to read and execute EDIT,\nbut not write into the EDIT file. EDIT could therefore\nrepresent a program provided by user \u201cJIM\u201d its owner,\nfor public use but with a restriction upon its alteration\nby any user other than its owner. This prevents unau-\nthorized changes from being made in the EDIT pro-\ngram. Nonowners of BFILE are granted both read and\nwrite permission making it universally available.\nBFILE may be a temporary storage file available to any\nuser.\n\nPROLL has nonowner permission bits similar to\nEDIT, making PROLL similarly publicly usable but\nprivately alterable only by user \u201cBOB,\u201d its owner.\nCFILE has no permissions granted for nonowners.\n\nIn the Figure, each file has associated with it an addi-\ntional file access control bit, the set user identification\nbit (SUID bit). When the SUID bit is set to zero for a\ngiven file, the effect of the various permission bits ts\nexactly that which has been so far described; owners\nand nonowners are identified by reference to their user", "title": "Operating system for a data processing system", "trim_reasons": ["leading_ocr_noise"], "year": 1979}
{"archive_ref": "pat-us2129711", "canonical_url": "https://patents.google.com/patent/US2129711A/en", "char_count": 3062, "collection": "archive-org-bell-labs", "doc_id": 108, "document_type": "patent", "id": "bella-qwen-pretrain-doc108", "record_count": 1, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": "pdftotext", "selected_extraction_score": 1.0, "source_family": "google_patents", "source_url": "https://patents.google.com/patent/US2129711A/en", "split": "test", "text": "GUIDED TRANSMISSION OF ULTRA HIGH FREQUENCY WAVES |\n\nFiled March 16, 1953\n\nE\n2 SAN SE LE EMEAAAAN #8 AXAAA\nF, \" :\n\nHisulaiion\n\nsp\n\nee ee\n\nsan\n\nFtg.10\na\n\nnectbhets\n\nS\n\nLo\n\nS\n\nT Sheets-\u2014oheet 1:\n\n[peas\nTS\n\n2b TE\n=19.69\na ~\n2b TO DG 2b 30\nAa.\n4 287\n20 3\n\\\n\\\nQO\n2b } 3b b 26 ey,\n? INVENTOR\nG. C. Southworth\n\nOC wane\n\n2,129,711\n\nG. C. SOUTHWORTH\n\nSept. 13, 1938.\n\nATTORNEY\n\nS\n~\n3 ,\n\u00bb OD | |\n\\ t.\n* G > % | \u00b0 \u2018 ow\n\u00ae l 7 wy} \u00a9\nra }-7 AN AY Ll\nOp: A RA Ke Welty y) J, } 4 E am |\nEe S27) SN : - SS\n| <=\n> 2 \u00a5 MSKSLES V % SS >\n= 0 My PESXST= . i cn\n\u00e9 SSS .\n\u2122~\nce \u00bb\ng g\n2 } VD\nfs ro\nBa | vi on Hatiny\nB\u00ae Sy\noc * a re (rn fen\\ (rn)\nco ss (IS Clay hy\nre \" = Ws IS 7\naS uO\n-  g B78\nkK, 2\noO\nz % ,\n+6\nma ah\nt) jy 26 tL-\ni Nn\na OZ 2/4\n8 $2'Ol -\nca\n: 83\nam | .\n= 3\nwD DH it\n: ! L [ ~~\nA \u2122\nRB \u00a9 N \u00a9 H F S.\n\nAPIONIA _\nIAIN D 2104 Wx NY. SH\n\nSept. 13, 1938.\n\nGUIDED TRANSMISSION OF ULTRA HIGH FREQUENCY WAVES\n\nxr.\n\n_ Veloctty of Light\nVeloctty of Wowe tn Guide\n\nS\n\n[a\n\na Ny\n\nSd\n\nG. C. SOUTHWORTH\n\nFiled March 15,\n\n2,129,\n\n711\n\n1933 Yt Sheets\u2014-sheet 3\n\n=e\n\n| l sul .\n&O J7O =\u00a7=\u00a7=\u2014 G0 200 240 280 520\n\nWawe Lengt!z ~ Crs.\n\n\u2014_ eee \u2014\n\n1 hn.\n\nA\n\\ \u2014 a\n\nee Poe ae Jy, OTCEeE oF Co ed Td \u00e9 red\n\na) wgrals Modulator \u00a3Fittlter mit?\n\nWave\nGuide\n\n\u00a5\n\nGF || SF [+153\n\nOsciidator\n\nCoupling\n\nCh2tt\u201d Demodulator Amplifier\n\n7\npt bs 5g OF\n\n| Hig. 13\n\n5\n\n|\n\naS OF || GF |\u2014>\n\n60\n\n59 KE G2\n\nG7\n\nGscitiater\n\" INVENTOR\n\nLay 7 4-\n\nG.C. Scuthwortt\nBY :\n\nfi tee\nATTORNEY\n\nSept. 13, 1938. | G. C. SOUTHWORTH 2,129,711\nGUIDED TRANSMISSION OF ULTRA HIGH FREQUENCY WAVES\n\nFiled March 16, 1933 7 Sheets-Sheet 4\n\n1\n|\n\n3 Parallel wires of\naadzustable length\n\nSSeS ee\n\n_ = = ee\n\nme i oe\n\n\u20147\n\nCoup (tmg AL}\n\nQOO A ) \u2018 , 73\n\\ | G .\nModulation ~ 000 wk =) Oscullatory nf Lita. 10\n\n\u2014 000\naf | Co | 412 | ib -\n\nL Ny\nZ, z, Ey fower SUBILY\n\nMeLt\n+ <*ys *\n\n4 way td\n\ny are eed\n\nOF Ge aot Ped ae\n\nf\nbAtelded.\nConductors \u2014\nJor Lower_and\nSignal input\n\n=\n* a *\nCe i oe 2 bala |\nCc! *\" art |\npov | y ' a oun np\n\n-\n\u00a5F\nqf a\n\nSPiielded\n\nShielded Llug Oscillator fA LY- 1S\n\n\u2018Copper Plate 25 Bayonet Connecttowm\n\na A\n\nSe te Yat\nen) ; .\nees r \u201cPa*\n\n= s \u00ae - .\n' * Tal a\u201c, % *\n\nphates geal reriflaant eileen a, ih ok\u201d ak a a a eel ae a ar ae\nF\n\noy \u201cSD. ( :\n\nafar\nYe eid\n\n}\n\n+\n-\n\na\n_\n\na\n\ni\n\n|\na1\na\nay\n4 .\n+\n\ni\n\n=\n\n1\noe\n\n/ vi) oo on agon8 6\n\nShielded 7\n\nPug ANA Leaas 4 JR veeeees 4\n\nJor Lower ang NSS 65 =\ni ae\n\nSigal 1a a at\n6G Glass -cogoer Seals\n\nINVENTOR\nLAGI 37 G.C. Southworth,\nY\n\nPEAR\n\nATTORNEY\n\n4\n-\na\naoig\n\nete te\n\na Fa ar\na\n\nvm,\na =\n\u2014 weeks wea ety a+\n\n*\nste\n\n'\n\nTeer\n\na4ye a iy al\n\na!\n\nFIP PS Oe 2S P.O Pee\n\n44% de\nI\n\n'\n\nS\n\nSept. 13, 1938. = 3 ~\u2014 6. \u00a9 SOUTHWORTH 2,129,711\nGUIDED TRANSMISSION OF ULTRA HIGH FREQUENCY \u2018WAVES\nFiled March 16, 1933 7 Sheets-Sheet 5\n\nDetector Amplifier,\n4 \\ Chokes +\n\nWi Fig. 20\nTE AMajustable coupling ~\na a Almypalsfter\nYy Conductors of - : \u2014_\nag estabile ERG. | |\nLAG) 22\n\nCOU\npang - Detector\n\u2014, eee\n\nog 4 estab tris\n\nDielectric wie\n\nLAG LF | | Leo\n\nOscriiltator\n\nCoupling Tuned\nmet Detector Amplifier Detector Amplifier\n\n(2 EKO fF FE\n\n|\n\nror\n\n.  Oscvator\n\nINVENTOR\n\nG.C. Southworth\nBY\n\n\u00a2 ATTORNEY", "title": "Guided transmission of ultra high frequency waves", "trim_reasons": ["leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise"], "year": 1938}
{"archive_ref": "sim_record-at-t-bell-laboratories_1929-02_7_6", "canonical_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1929-02_7_6", "char_count": 86596, "collection": "archive-org-bell-labs", "doc_id": 151, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc151", "record_count": 102, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1929-02_7_6", "split": "test", "text": "With the publication of \u201cSpeech and Hearing\u2019 by \nHarvey Fletcher, Acoustical Research Director, the \nfifth book of the Bell Telephone Laboratories series \ncomes into being. An introduction to this volume, \ncontributed by Dr. Arnold, is here reprinted.\n\nHE atmosphere of sounds in \nwhich we live ministers so \nconstantly to our knowledge \nand enjoyment of our surroundings \nthat through long familiarity we have \ncome to feel, if not contempt, at least \nindifference toward the marvelous \nmechanism through which it works. \nHearing, we are inclined to consider \nas little a matter for concern as \nbreathing; and so long as our own \nfaculty remains unimpaired we feel \nlittle curiosity concerning the provi- \nsions of nature either for ourselves \nor for others. When we hear too \nfaintly or indistinctly we know we \nneed only trace the sound to its source \nto hear its perfect form, for that is \nthe method we have used from child- \nhood in investigating the sounds of \nour immediate neighborhood. \nNow with one broad sweep the \nbarriers of time and space are gone \nand all the world becomes our vocal\n\nneighborhood. No longer can we \ntransport ourselves to the origin of a \nsound and thus become convinced that \nwe are hearing it aright, for that \norigin may be thousands of miles \naway or may have vanished years \nbefore; and so we must establish a \nnew method to measure the accuracy \nof the copy which reaches our ears. \nWe must also find a clearer index to \nour satisfaction in it, for we are no \nlonger concerned with the immutable \nprovisions of nature but may ap- \nproach at corresponding expense \nwhatever perfection we may demand \nin our instruments of translation and \nreproduction. Thus the telephone \nand the phonograph should excite a \nkeener interest in how we hear and \nin what measures our satisfaction in \nthe speech and music they provide. \nOur ears are only machines to \ntranslate air waves into a form suit- \ned to stimulate the auditory nerve;\n\nand as machines we may measure and \ndescribe them in the same terms that \napply to devices we ourselves con- \nstruct. We may compare them as to \nperformance, and may accommodate \nour devices to their requirements. But, \nto understand the mechanism of the \near is by no means to understand the \nact of hearing, for we have not heard \nuntil the brain has perceived the mes- \nsage sent by the auditory nerve. We \ncannot explain in precise mechanical \nterms how this is done, nor indeed \nhave we any very clear comprehen- \nsion of the process at present. Some \nimportant factors relating to the pro- \ncess of hearing we can, however, de- \ntermine by measuring the least \nchanges in sound which can be detect- \ned under a variety of conditions of \npitch, loudness, and accompanying \nnoise. Thus we may obtain a quanti- \ntative means of comparing individuals \nin this respect, and establish a stand- \nard of average hearing.\n\nThere is a most important factor \nin hearing, however, which is much \nmore difficult of analysis and measure- \nment. This is the individual\u2019s ability \nto recognize small defects in those \nsounds with which he has become es- \npecially familiar. We all know how \nquickly we note a slight change in a \nfriend\u2019s voice, and with what uncanny \nskill a trained musician will detect \nminute imperfections in very complex \nsounds. Our approach to a quantita- \ntive understanding of the importance \nof this must be by an indirect method. \nFirst we must construct devices so \nperfect that even the keenest ear can- \nnot find a flaw in their rendition, and \nthen step by step we may introduce \nmeasured imperfections until an ob- \nserver can detect a fault. In the re- \nsponse of individuals to this test there \nwill, of course, be great differences;\n\nbut when we have collated opinions \nfrom a wide variety of observers we \nmay forecast in a reasonable way the \ndegree of mechanical perfection that \nmay be demanded of our instruments.\n\nThis, then, has been the philosophy \nof the investigation of hearing which \nhas been carried on in Bell Telephone \nLaboratories during the past fifteen \nyears: to get an accurate physical \ndescription and a measure of the me- \nchanical operation of human ears in \nsuch terms that we may relate them \ndirectly to our electrical and acousti- \ncal instruments; to test the keenness \nof the sound-discriminating sense and \nfind what is the smallest distortion \nwhich the mind can perceive and how \nit reacts to somewhat larger distor- \ntions; and thus to reach a reasonable \nbasis of design both for separate in- \nstruments and for systems, as a \nwhole, to give a proper balance be- \ntween cost and performance.\n\nWith hearing, speech and music \nare linked inseparably for they only \nbring a meaning through our aural \nsense. It is an instinctive first thought \nthat they must be heard to be criti- \ncized. They can, nevertheless, be in- \nvestigated by mechanical means and \nbe described in the same physical \nterms that we use in describing hear- \ning; and thus to an extent we may \nconsider them both objectively. But \nif we attempt to divide the study be- \ntween speech and music we come at \nonce upon the difficulty that speech \nconveys information by intonation as \nwell as by articulate syllables; and this \nmakes it infeasible to set a definite \nboundary between them. A division, \nhowever, between vocal sounds and \ninstrumental sounds proves more use- \nful, for in the one case we are limited \nby our vocal organs which we must \ntake as they are, while in the other\n\nwe have a definite control and can \nadapt the nature and complexity of \nthe sounds produced to conform to \nour sense of hearing and our musical \nappreciation. The investigation of \nspeech and music has been governed \nby these general considerations. An \nattempt has been made to establish \nin definite terms the performance and \nlimitation of the voice and, although \nso far in considerably less detail, to \nfind the corresponding factors in in- \nstrumental music.\n\nWith a clear knowledge of the na- \nture of the sounds that we must pro- \nduce and the accuracy with which we \nmust maintain their form, there re- \nmains the problem of securing instru- \nments which are sufficiently refined \nfor the purpose. Instruments of re- \nmarkable precision are required in the \nconduct of the investigation, since if \nwe are to measure the smallest detect- \nable variations in sounds we must ob- \nviously use equipment which is cap- \nable of a degree of exactness beyond \nthese small quantities. Such instru- \nments would appear at first sight not \nto have much utility outside the labo- \nratory, since they are costly and often \ncomplicated and difficult to adjust.\n\nIt is interesting to note, however, \nthat some of the instruments, in es- \nsentially their original laboratory \nform, have found other important \nuses. Indeed, a surprising number of \nmodern acoustical accomplishments \nhave come about through the use of \nslightly modified forms of the appa- \nratus which was originally developed \nfor these investigations. Modern\n\nphonographic records are produced \nwith an electrical transmitter which \nwas developed in the very early stages \nof these studies; and radio broad- \ncasting has grown up around this \nsame \u201cmicrophone.\u201d\n\ning equipment of the modern phono- \ngraph and of the radio were predi- \ncated directly upon these investiga- \ntions; and talking motion-pictures \nowe their success and much of their \napparatus to this same source. \nAlthough the results which relate \nto normal speech and hearing are nat- \nurally the most familiar and widely \nknown, there have also been import- \nant outgrowths in the way of aids \nto those handicapped in one or the \nother of these faculties. In establish- \ning the functioning of the average ear \nit was obviously necessary to investi- \ngate a large number of cases and \namong them some which departed \nrather widely from the average. For \nthis study an instrument was devised, \nnow known as the audiometer, which \nhas put within the reach of all who \nneed it the possibility of an accurate \nmeasure of their hearing. In quite \nanalogous fashion there grew out of \nthe investigation of the limits of hear- \ning a better knowledge of ways to \nprovide aids for those partially deaf; \nand it has even become possible to \nprovide means of speech for some \npersons whose vocal cords are gone. \nValuable as these results are, eco- \nnomically the most important out- \ncome of the work has been the in- \ncrease of exact knowledge as to the \nrequirements and limitations to be \nplaced upon the transmission of \nspeech in the telephone system. As \ntime goes on there must be an evolu- \ntion toward even greater perfection \nin those particular elements which are \nmost important to intelligibility. The \nsystem is so large that the cost of \nsuch an evolution is immense and \nchanges undertaken without an accu- \nrate knowledge of their value might \nlead to burdensome expenditures for \ndisproportionate results; but, with the\n\nfacts established by this investigation \nin hand, we can weigh any contem- \nplated change and judge whether it \nis the one that offers most improve- \nment at the moment and what its ulti- \nmate effect will be in its operation \nwith other elements of the system.\n\nCrandall and himself, for which since \nDoctor Crandall\u2019s death he has had \nthe full responsibility. No one can \nspeak with better knowledge of the \nfacts or with more complete author- \nity for the opinions he expresses.\n\nThe work is not complete\u2014indeed \nsome parts of it are hardly more \nthan started; yet its results have been \nso great, both for the original pur- \npose which was planned and for the \nmany issues which have since arisen, \nthat it presents a unique exemplifica- \ntion of the worth of systematic and \nsustained research; and Doctor \nFletcher is to be congratulated that \nhe has seen it through with such clear \nvision as permits its presentation in \nits present form.\n\nBELL TELEPHONE LABORATORIES SERIES \nof books published by \nD. Van Nostrand and Company\n\nbrought the telephone sys- \ntem to its present state of usefulness \nand efficiency, none, perhaps, is more \ntypical of steady improvement than \nis lead-covered cable. The growth \nfrom the fifty-pair of 1888 to the \n1200-pair cable of 1912 has already \nbeen described in the REcorpD.* As \nis commonly the case, however, difh- \nculties become greater the further a \ndevelopment is carried. The step \nfrom 1200 to 1800 pairs, taken by \nthe new development, although it \nrepresents no greater percentage in- \ncrease in size than many of the pre- \nceding steps, and is even smaller than \nsome, is a noteworthy \nachievement because \nwith each increase in \nthe number of pairs, \nand each decrease in \nsize of wire, the difh- \nculties of manufacture \nbecome greater.\n\nWith the increasing \ncongestion in metro- \npolitan areas, avail- \nable space beneath the streets \u2014 \nwhere cables must be placed \u2014 is rap- \nidly being used up. Added demands \nfor power, light, water, transporta- \ntion, and other utilities all require \ntheir share of this space. Obtaining \na fifty per cent increase in the number\n\nof telephone wires that may be run \nthrough a duct is thus a real contribu- \ntion to the growth and development \nof densely populated districts, and \npromises further economies in this and \nalso other situations.\n\nThe largest cable existing previous- \nly used twenty-four gauge conductors \nwrapped with paper one-quarter inch \nwide and two and one-half mils thick, \nand its 1212 pairs, so insulated, were \ncontained within a lead sheath of \nabout two and five-eighths inches out- \nside diameter. Cables of this diameter \nare the largest that can be pulled into \nmany existing ducts, and so this out- \nside diameter is one of the limiting \nfactors of exchange-cable design. The\n\nproblem, therefore, is not merely to \nmake a cable with a greater number \nof conductors but to change the de- \nsign and methods of manufacture so \nthat the larger number of wires can \nbe contained within the same size of \nsheath.\n\nObviously there are two possibil- \nities; either the size of the conductors\n\nor the thickness of the paper insula- \ntion may be reduced. There are dif- \nficulties involved in such reductions, \nhowever, both in the manufacture of \nthe cable and in handling it after its \ncompletion. A cable must be pulled\n\nFig. 2\u2014In this cross-section of the 1200- \npair cable also the layered construction is \nclearly evident\n\ninto a duct from one manhole to the \nnext, and possibly in the course of its \nlife may be pulled out of one duct \nand into another. This pulling sub- \njects the cable to strains and there \nis a possibility of breaks resulting in \nthe conductors if they are too weak. \nBesides this the lengths of cable must \nbe spliced in each manhole and because \nof the difficulty of handling smaller \nwires they cannot be reduced too far \nin size. In addition the factor of re- \nsistance can never be overlooked. The \nsmaller the conductor the higher is \nthe resistance and the greater, there- \nfore, are the transmission losses.\n\nIn the course of development, pro- \njected to overcome these difficulties, \nmany experiments were tried. Wires \nof different materials were experi- \nmented with to find out whether \ntougher substances, such as hard cop- \nper or bronze, would not add enough\n\nstrength to offset a decrease in size. \nResults were not very satisfactory. \nThe harder wire was springy and dif- \nficult to handle at the splices. Cables \nof soft copper but of smaller sizes \nwere also experimented with and it \nwas found that wire as small as twen- \nty-seven gauge could be used.\n\nThen, attention had to be given to \nthe matter of paper. The thickness \nused with the 1200-pair cable was \nchosen principally for reasons of \nstrength. In building up the cable the \nwires must be crowded together with \nconsiderable pressure and in handling \nthe cable after completion, particu- \nlarly in bending it, similar pressures \nare developed. If the paper is too\n\nFig. 3-\u2014\u2014A_ cross-section of the 1800- \npair cable shows how the eighteen units \nare formed into the final cable\n\ntempting to use paper thinner than \nused in the 1200-pair cable another \ndifficulty arises. In wrapping the \npaper strip around the conductor in \nthe manufacturing process, certain \ntension must be maintained and with \ntoo fragile a paper tearing is apt to\n\nresult. This difficulty was partly over- \ncome by the Western Electric Com- \npany through improving its insulating \nmachinery so that thinner paper could \nbe satisfactorily used. It was thus \npossible to employ paper which was \nappreciably thinner than the two-and- \none-half-mil paper of the earlier cable.\n\nWhile all this work was going on \nanother development had been made \nwhich changed the arrangement of \nthe pairs of conductors inside the \nsheath. The basic unit of all tele- \nphone cables of this type is the pair: \ntwo conductors, each insulated with \npaper, twisted together. In _ practi- \ncally all the earlier cables these pairs \nwere laid up in concentric layers. \nEach layer was spiraled around the \nlayer beneath it, and the direction of \nthe spiral of successive layers was \nreversed. This gave a firm body that \ncould be easily handled.\n\nThe new development built up \nsmall unit cables, each of 101 pairs \nfor the 1818-pair cable, and then as- \nsembled these in a long spiral to \nform the complete cable. This new \nmethod brought about certain small \nsavings due to a reduction in the take- \nup\u2014or shortening of the effective \nlength of the conductors due to \nstranding\u2014and considerably simpli- \nfied the manufacturing of the larger \nsizes of cable. It resulted in a more \nflexible core which is also important \nwith the larger cables.\n\nAs a result of these various devel- \nopments and improvements, an ex- \nperimental cable was constructed with \n1818 pairs of twenty-seven-gauge wire. \nExperiments in pulling this cable in \nand out of ducts indicated that a \nsomewhat firmer core would be desir- \nable. A cable of the same number of \npairs but with twenty-six-gauge wire\n\nwas then tried and found satisfactory. \nIt is this cable that has now been \nstandardized and will be used in the \nfuture in many of the large metro- \npolitan areas and other places where \nconditions make it desirable.\n\nIn building up the new cable, two \nof the hundred-pair units are used as \na center and around them are spiraled \nsix other hundred-pair units. Around \nthese are laid the remaining ten units \nto complete the full 1800 pairs. This \nmay be seen in Figure 4, where the \ntwo inner groups point vertically up-\n\nward, the next layer of six units \npoints out at an angle of about forty- \nfive degrees, and the outer layer of \nten units is bent down.\n\nFor convenience in identifying pairs \nfor splicing, it is customary in all the \nlarger cables to use several colors of \npaper to insulate different groups of \npairs. In the 1200-pair cable, for ex- \nample, the pairs in each group all have \ninsulation of the same colors, which \nare different for adjacent groups. In\n\nall, six colors are used. Each group is green and the other blue-white Paper. \ndistributed over the whole or part The next layer has one unit of red.\n\nN \n\u2014 ORANGE BINDER ON EACH \ngl UNIT IN OUTER LAYER. \nRED- GREEN \u00bb \n\u201cWHITE BLUE-WHITE \nORANGE-GREEN BINDER \nRED-GREEN ON EACH UNIT IN FIRST \nLAYER. \nRED- / BLUE- BLUE-\\ \nwuHite / WHITE WHITE \\ wHiTe \nRED-GREEN \nORANGE BINDER ON \nBLUE- WHITE v4 EACH UNIT IN CENTER \nBLUE- RED- ED- /BLUE- \nWHITE \\ WHITE WHITE / WHITE [7 \nUE-WHITE \nRED-WHITE RED-wuire \nBLUE-WHITE LY \nYo\n\nthree red-green groups makes identification of \ngroups comparatively simple\n\ngreen, and on each side of it a \nblue-white unit, followed by a \nred-white unit, and so on. The \nouter layer is similarly built \nup with one red-green unit and \nthen alternately blue-white and \nred-white units. As a result \nthe cable is symmetrical in ap- \npearance when looked at from \neach end so that locating cor- \nresponding units in making a \nsplice in a manhole is a rela- \ntively simple matter.\n\nAt the termination of these \ncables in central offices it is \ncustomary, for convenience in \ndistributing the pairs to the \nmain frame, to splice them\n\nof one or more of the annular layers, into several small cables with double \nand the same color appears only in \u2014 silk and single cotton insulation. In\n\neach group is bunched to- \ngether and joined to the \ncorresponding group of the \ncontinuing cable.\n\nThis method is facili- \ntated by the new construc- \ntion, as the hundred pairs \nof a unit are all of the same \ncolor and already bunched. \nEach unit in addition is \nloosely bound with cotton \nthreads, those of adjacent \nlayers being of different \ncolors, which makes it easier \nto identify corresponding \nunits. A simplification of \nthe color scheme has been \ndeveloped which is indi- \ncated in Figure 5. The two \nwires of a pair are always \ninsulated with differently \ncolored paper and one of\n\nFig. 6\u2014The fourth cable from the bottom shows \nan 1800-pair cable spliced to nine 200-pair cables\n\nFigure 6 is shown such a splice \nwhere nine 200-pair are con- \nnected to one of the new 1800- \ncables. The 200-pair cables \nrun under the floor for a few \nfeet and then up the vertical \nside of the frame where they \nare fanned out and connected \nto terminal protector blocks. \nThe amount of space required \nfor fanning out one of the new \ncables can be seen in Figure 7. \nEach 200-pair cable requires \nabout five feet of terminal \nstrip so that nine rows of \nblocks, each about five feet \nhigh, are required in all. As \na contrast a small piece of the \n1800-pair cable is placed be- \nside the terminal blocks. \nThese last two views were \ntaken in the Cumberland Of- \nfice of the New York Tele- \nphone Company in Brooklyn \nand show the termination of \nthe first commercial installa- \ntion of 1800-pair cable, which \nwas completed in the spring \nof 1927. Since that time a \nlimited amount of this cable\n\nFig. 7\u2014Nine rows of terminal blocks\u2014each \nabout five feet high\u2014are required to fan out \nan 1800-pair cable\n\nhas been manufactured on temporar- tion new and improved machinery in \nily modified machinery. The Western order that it may be in a position to \nElectric Company, however, has de- take care of the demand that is ex- \n\u2018signed and is about to put into opera- pected for the new product.\n\nCENTRAL office, function- \nally considered, is that part \nof a telephone system which\n\nthis rapidly growing part of the tele- \nphone plant. In some of the larger \ncities P. B. X. attendants outnumber \ncentral-ofice operators; in Manhat-\n\ntan, for example, there are about \n9500 central-ofhice operators while the \nP. B. X. attendants number approxi- \nmately twenty thousand.\n\nIn the very early days a single tele- \nphone in any one residence or place \nof business was ample. The telephone \nwas so superior to former methods \nof communication, by mail or mes- \nsenger, that its easy accessibility to \neach individual that might have occa- \nsion to use it never occurred to any \none as being at all necessary. Nor \ndid it seem necessary to have tele- \nphonic means of communication be- \ntween persons within a single estab- \nlishment \u2014 much smaller at that time \nthan are some of our gigantic cor- \nporations now. As the telephone \ncame to be used more and more, how- \never, and as greater numbers of em- \nployees in every office or factory \nfound occasion to take advantage of \nits time-saving possibilities, a demand \nnaturally arose for more than one \nstation in the larger establishments. \nThere arose also a need for the tele- \nphone for intercommunication; a \nbookkeeper in the office found much \ntime and energy could be saved if he \ncould talk directly with a clerk in the \nshipping room. Out of these de- \nmands arose the private branch ex- \nchange, or P. B- X. as it is colloquially \ntermed, a private central office serv- \ning directly its own stations.\n\nonly solution possible. There could \nhave been the same multiplicity of \nstations within the establishment and \nthe same possibility of intercommuni- \ncation, had each station had its own \nline to the main central office and its \nown number. Any station could have \ncalled another by passing through the \ncentral office operator, and so far as \nthe members of the organization were \nconcerned there would have been little \ndifference except that the station \ndesignation numbers might have been \nlonger.\n\nFrom the standpoint of some one \ncalling in from the outside, however, \nthere is a considerable difference. The \noutsider often does not know the \nname of the person he wants to talk \nto; he knows what he wants to find \nout or what knowledge he wants to \nimpart but that is all. If for any es- \ntablishment there were only a long \nlist of names and numbers in the di- \nrectory he would be helpless. Obvi- \nously the only satisfactory solution\n\nis to have but a single number for \nany one establishment and a single \nperson\u2014or, for the larger concerns, \na single group of persons\u2014to an- \nswer all incoming calls. This person, \nthe P. B. X. attendant, with intimate \nknowledge of the organization could \nconnect the incoming call to the sta- \ntion that could most effectively deal \nwith it. Thus the need for an at- \ntendant was paramount in calling the \nP.B.X. into existance and the at- \ntendant is still of primary importance \nto the modern P. B. X. system.\n\nIf there are fifty telephone stations \nin the local establishment they will \nnever all want to talk with the outside \nsimultaneously. At any one time some \nstations will not be in use at all and \nsome may be talking to other local \nstations, leaving only a small re- \nmainder making demands for outside \nservice. The ratio of central-office \ntrunks to stations, therefore, is -al- \nways less than unity although it varies \nover a considerable range depending\n\non the type of business. The at- \ntendant and the P.B.X., of course, \nalways partially, and occasionally en- \ntirely, offset the savings due to the \nreduction in number of trunks. The \nattendant, however, performs so \nmany useful services that she is gen- \nerally regarded as an asset rather \nthan an expense.\n\nDifferences in types of business as \nwell as differences in size naturally \naffect the P. B. X. equipment fur- \nnished for the subscriber. In some \norganizations most of the calling is \nbetween members within the organi- \nzation, whereas in others practically \nno local calls are made. In some \nplaces conditions favor dial service \nwhile in others the manual system \noffers certain advantages. Differences \nin the number of stations to be served \nnecessarily make a difference in the \ntype of P. B. X. furnished. Thirty- \nfive varieties of P. B. X. have been \nmade in the past and probably more\n\nwill be built in the future as methods \nof doing business change and as tele- \nphone apparatus and equipment im- \nprove. \u2018There has never been this \nnumber in standard use, however, at \nany one time. At present only six \ntypes are standard and stocked and \nthis number probably fairly repre- \nsents the number standardized at any \none time in the past. |\n\nIn the manual class the range of \nsize varies from the small 505 type, \nwith a maximum of seven stations \nand three central-oflice trunks, to the \nlarge 604-C which, as used by the \nConsolidated Gas Company of New \nYork, has 1650 station lines, 221 \ntrunks to central offices, and 175 tie \nlines to other branch exchanges, and \nrequires 42 attendants during the \nbusy part of the day.\n\nThe small 505-type board uses no \ncords or plugs but all connections are \nmade by keys on the front of the \nboard. It is so small that it can be\n\nplaced on an ordinary desk, and may \nbe operated by a clerk who may also \nhave other duties. Contrasting with \nthis the larger exchanges, such as the \none in our own Laboratories, have all \nthe appearance of a central office. \nBetween these extremes are other \ntypes and sizes such as the 550* and \nthe 600-C.\n\nEach type of board is made neces- \nsary by the conditions of service that \nit has to meet and by the number of \nlines that it must serve. The cord- \nless 505 type is compact and seems \nvery simple but actually would become \ndificult to operate if the number of \nlines were increased to any great ex- \ntent. To locate an operated key \nfrom the large number all exactly \nalike becomes somewhat trying. A \ncord plugged into a jack, on the other \nhand, is easy to see and so as the \nnumber of lines increases better re- \nsults are secured with cords.\n\nThe No. 1 and No. 2 intercom- \nmunicating sets, which do not require \nattendants for local calls, are interest- \ning because of their small size. Lines \nrun from each station to every other, \nand each station is in reality its own \nP. B. X. as a key is depressed corre- \nsponding to the line wanted and no \nother switching equipment is required. \nExcept for this very small unit, pri- \nvate branch exchanges not requiring \nan attendant for local calls have been \ndesignated for the larger installations \nonly. The present standards are the \n700 and the 740** dial-types, the for- \nmer of which may have any capacity \nup to one approaching that of a cen- \ntral office. Both of these boards may \ndial an outside number without the\n\nThere is a large and increasing de- \nmand for tie-line intercommunication\n\nFig. 4\u2014Dial-type private branch ex- \nchanges have usually been made in the \nlarger sizes as indicated by this section\n\nbetween P. B. X.\u2019s._ Large organiza- \ntions, such as public service companies, \noften have anumber of private branch \nexchanges located at different points \nthroughout the city. The tie-lines may \nbe arranged for dial or manual serv- \nice, depending upon the type of P. B. \nX. and the requirements of the sub- \nscribers. When tie-lines are provided \nbetween dial P. B. X.\u2019s the circuits can \nbe arranged, if desired, so that a sta- \ntion in one P. B. X. can dial any sta- \ntion in another. At Bell Laboratories \nthere are direct tie-lines to and from \nnine other Bell System P. B. X.\u2019s in \nand around New York City.\n\nFig. 5\u2014A P. B. X. of the dial type as arule has a group of manual positions for the \nanswering of incoming calls. The above is a part of a 700 P. B. X.\n\nOf the 115,000 P. B. X.\u2019s manufac- \ntured by the Western Electric Com- \npany during the period from 1910 to \n1926, 96% were of the small sizes. \nOf these 9% were of the No. 2 and \nNo. 4 types, the latter of which is now \nno longer standard, the other 87% \nbeing about equally divided between \nthe 505 and the 550 types. Natural- \nly, there are fewer of the very large \nboards so it is not surprising that in\n\nthis same period only a score or more \nof these have been manufactured, al- \nthough additional sections are being \nmade continually to take care of the \nexpansion of the existing boards. \nThese figures make it easier to under- \nstand that in some central offices in \nthe business section of New York \nCity approximately seventy-five per \ncent of the total working lines termi- \nnate in private branch exchanges.\n\npears to be a similarity between \nsome of the phenomena observed in \natomic physics and those found in cer- \ntain non-linear electrical circuits. This \nidea is interesting and very important \nsince it suggests the possibility of ac- \ncounting on purely classical grounds \nfor the methods used in calculating \nquantum phenomena. The non-linear \ncircuits of interest here are of the \nvariable reactance type such as the \nmagnetic modulator for variable in- \nductance, and the condenser transmit- \nter for variable capacitance.\n\nA number of years ago some of us \nin the Research Laboratory were \nworking on magnetic modulator cir- \ncuits to check theoretical speculations \nwhich were of considerable interest \nat that time. All modulators have this \nfeature in common, that they distort \nthe wave shape impressed upon them: \na process which yields new frequen- \ncies related to those originally im- \npressed on the circuit. If for example \na pure sine-wave generator is con- \nnected to the modulator, then the new \nfrequencies produced are harmonics, \nintegral multiples, of the generator \nfrequency. When two of these gen- \nerators of different frequency are con- \nnected to the modulator (they may \nbe the carrier and voice waves of the \nordinary radio broadcast transmitter ) \nwe get, in addition to the harmonics \nof each of the generator frequencies,\n\ncombinations between them, among \nwhich are the sum and the difference \nof the two original frequencies. The \nlatter are known as side frequencies; \nthey are evidently produced by the \nwave-distorting or non-linear prop- \nerties of the modulator, since these \nnew frequencies do not arise when or- \ndinary non-distorting elements make \nup the circuit. In the magnetic modu- \nlator the non-linearity is provided by \nthe magnetization characteristic when \nsufficiently large magnetizing forces \nare used.\n\nThe problems arising in modulator \ncircuits are those concerning the re- \nlationship of these new current com- \nponents to the original ones. Two of \nthe properties of the modulator which \nare of interest in this connection are \nthe stability, and the ratio of the side- \nfrequency power developed to the sig- \nnal power supplied; usually the latter \nis termed the modulating gain. In \nthe first magnetic modulator circuit \nset up, we had three filters connected \nto the modulating coil: one to supply \nthe carrier, the next to supply the sig- \nnal, and the last to take off one of \nthe side-frequencies to a load circuit. \nEven at the most favorable value of \ncarrier voltage, the modulating gain \nwas found to be too low with our \nhastily improvised arrangement and \nwe set about the construction of a \nmore efficient circuit. To do this the \nmodulator coil was built up with very \nthin laminations of permalloy, and\n\nthe number of turns and the size of \nthe coil were adjusted to draw maxi- \nmum power from the two connected \ngenerators. When these changes \nwere gradually incorporated in the \ncircuit, the modulating gain rose un- \ntil finally a curious phenomenon was \nnoticed by E. T. Burton: the circuit \nbecame unstable, and a number of \nnew frequencies were produced which \napparently bore no relation to the\n\nCircuit for the production of low- \nfrequency oscillations. Current from the \ncarrier frequency (p) generator, applied \nto the toroidal coil yields simultaneously a \nlow frequency (q) current and the side- \nfrequencies. A tuned circuit is provided \nfor only one side-frequency \u2014 the lower\n\noriginally impressed frequency. These \nwere made evident by a mushy note \nwhich was heard when a telephone \nreceiver was connected in circuit. This \neffect obviously set a limit to the gain \nwhich could be usefully obtained and \nit seemed desirable to establish the \nreasons for its appearance.\n\nThe effects are more easily follow- \ned when we use simple tuned circuits \ninstead of filters. In that case single \nfrequencies are generated rather than \nthe large number of frequencies which \nappeared in the filter circuit and gave \nrise to the mushy note observed. An \napproximate mathematical analysis \nof this simplified circuit was made. \nAssuming a magnetic core free from \nhysteresis and working under simple \ncircuit conditions, it was found that \nthe flow of the two side-frequencies \nchanged the coil resistance to the two\n\ninput frequencies. In the path of the \nhigher impressed frequency p this re- \nsistance was positive; in the path of \nthe lower impressed frequency q the \nresistance introduced by the flow of \nthe upper side-frequency was positive \nbut the resistance introduced by the \nflow of the lower side-frequency was \nnegative. These conclusions were in \ncomplete agreement with those which \nhad been found theoretically by Mr. \nHartley in 1917, and which he had \nalso obtained somewhat later in the \ncase of a condenser transmitter. These \nconclusions are carried over un- \nchanged to the higher order side- \nfrequencies such as 2p+q and 3p+q.\n\nThis idea of negative resistance \nenabled us to account for the experi- \nmental observation in a qualitative \nway at least, and justified the initial \nassumption that the non-linear mag- \nnetic characteristic, and not the \nhysteresis loop, was responsible for \nthe effects. For if the upper side-fre- \nquency meets a high circuit imped- \nance, so that the resulting current is \nsmall, and if the lower side-frequency \nmeets a low circuit impedance, so that \nthe resulting current is large, then the \nnet effect of these two currents on the \nresistance to q will be negative. When \nthis net effect, which depends upon \nthe carrier amplitude, is great enough \nto make the total circuit resistance \nnegative, and when the circuit re- \nactances are nearly annulled, then \noscillations will result and if the gen- \nerator of frequency q be withdrawn, \nthis frequency, together with its side- \nfrequencies, will be sustained. It was \nwith the simpler circuit shown that \nthe theoretical conclusions were more \ndefinitely tested and found to be in \ngeneral agreement. In order that \nsustained frequencies might arise the \ncarrier was required to exceed a defi-\n\nnite amplitude, the associated circuits \nwere required to be fairly well tuned \nto the two frequencies involved and \nthe resistances of the tuned circuits \nwere required to be sufficiently small. \nIn a circuit using filters these condi- \ntions are obeyed at a number of fre- \nquencies so that several currents are \nsustained.\n\nTo sum up these effects we may em- \nploy terms ordinarily used of regen- \nerative vacuum-tube circuits, in which \noscillation is produced by feeding \ncurrent from the plate to the grid \ncircuit in proper phase. The source \nof energy here is the plate battery, \nwhereas in the magnetic modulator it \nis the carrier generator. For each, \nthere is a critical value below which \nsustained oscillations cannot exist. \nFurther, in the magnetic modulator \ncase there is not merely one sustained \noscillation, but at least two, which we \nhave designated q and p-qg. Neither \ncan be sustained without the other, \nand if we introduce sufficient resis- \ntance to either g or p-q, or if we in- \ntroduce sufficient reactance to either \ncircuit both oscillations cease.\n\nMr. Hartley has suggested in a \npaper presented before the American \nPhysical Society on December 29, \n1928 that there is a striking resem- \nblance between the phenomenon just \ndescribed and the Raman effect in \natomic physics. Raman found that \nin passing light of a definite spectral \ncomposition through certain liquids, \nthe spectral composition of the emerg- \ning light was changed. New lines \nappeared, sometimes as pairs equally \nspaced above and below a line in the \nspectrum of the incident light, with \nthe higher frequency line always \nweaker than the lower. In some cases \nonly the lower frequency line was\n\nstrong enough to be observed. Fur- \nther, the interval between the new \nlines and the original lines from which \nthey were displaced has been asso- \nciated with a line in the absorp- \ntion spectrum of the liquid; that is, \nthe interval has been associated with \na molecular resonance. To press the \nanalogy with the modulator circuit, \nthe frequency of the incident light \ncorresponds to the supply frequency \np, the molecular resonance frequency \nis q, the stronger observed line is the \nlower side-frequency p-g, and the \nweaker is the upper side-frequency \np+q. The parallel is even more strik- \ning in the case of the condenser trans- \nmitter, in which the same type of ef- \nfect occurs, and in which, moreover, \nthe power involved is proportional \nto the frequency. This is suggestive \nof the quantum relation.\n\nThe importance of Mr. Hartley\u2019s \nsuggestion lies not only in its applica- \ntion to the Raman effect but also in \nits application to the general field of \natomic physics. The present effort in \natomic physics is devoted to discover- \ning methods such as matrix mechanics \nby which the observed effects may be \ncomputed, and the results are in many \ncases in excellent agreement with the \nfacts. However, there does exist \nsome question as to the physical sig- \nnificance of the quantities employed \nin these non-classical methods of com- \nputation. Following Mr. Hartley\u2019s \nsuggestion it may be possible to com- \npute these effects by treating non- \nlinear mechanisms by ordinary mathe- \nmatical means. If this should prove \nfeasible a physical interpretation \nwould be provided for quantum me- \nchanics, and we should then have a \nclassical mechanism capable of repro- \nducing quantum phenomena.\n\nHARRY PRESCOTT CHARLESWORTH \nEntered American Telephone and Telegraph Company, Engineering Department, \n1905; Equipment and Transmission Engineer, 1919; Plant Engineer, 19203 \nVice-President Bell Telephone Laboratories, 1928\n\nwas elected Vice-President \nof the Laboratories. He will be in \ncharge of its operations during Mr. \nCraft\u2019s absence.\n\nAfter graduation by the Massa- \nchusetts Institute of Technology in \n1905, Mr. Charlesworth entered the \nEngineering Department of the \nAmerican Telephone and Telegraph \nCompany, which was then located in \nBoston, and in connection with which \none of the development and research \nlaboratories of the Bell System was \noperated.\n\nHis first assignment was to the de- \nvelopment of local and toll circuits \nand associated apparatus. In this \nwork he was engaged for two years, \nuntil in 1907 the American Company \nmoved to New York City.\n\nDuring the next ten years Mr. \nCharlesworth was active in the de- \nvelopment of toll operating methods \nand the general related engineering \nproblems involved in extending and \nimproving telephone service.\n\nWith the opening of the war he \nwas specially assigned, as a result of \nhis broad knowledge of telephone \nplant and its operations, to handle \nproblems wherein the Bell System \ncould be of assistance to the Govern- \nment in the national emergency. In \nthis capacity he was throughout the \nwar active on communication facili- \nties for army camps, naval bases, \nsupply depots and particularly for\n\nthe Government Departments at \nWashington, where he also rendered \ncontinuous assistance to the Tele- \nphone Company on general equip- \nment and trafic engineering matters.\n\nAt the close of the war he took up \ndial-system development work and \nwith the creation of the Department \nof Operation and Engineering, was \nappointed Equipment: and Transmis- \nsion [ingineer under Vice-President \nGherardi. A year later he became \nPlant Engineer of that Department, \nin which position he has been con- \ncerned with all phases of the engi- \nneering of the telephone plant and \nwith relations with other wire-using \ncompanies.\n\nThus, throughout his telephone \ncareer of almost a quarter of a cen- \ntury, Mr. Charlesworth has taken an \nactive interest in the development of \ntelephone apparatus and equipment \nto meet telephone needs, and in the \norigination of proper operating meth- \nods for handling telephone traffic. \nHis background of scientific training \nand long contact with the develop- \nment and operating problems, and his \nproven ability to coordinate and ad- \nvance the work of others, have re- \nsulted in constantly increasing respon- \nsibilities in the Bell System. And now \nwith an experience well rounded by \nhis supervision of plant engineering \nin general, he returns to his original \ninterest in development by assuming \nleadership of the Bell Telephone \nLaboratories.\n\nsubscriber\u2019s line. It made the \nfirst of a series of successive selections \nresulting in a connection to the line \ncalled. Selectors, however, are com- \nparatively expensive mechanisms, and \nas they are in use only when the line \nis making an outward call a much \nsmaller number of them would meet \nthe requirements could some method \nbe devised of connecting them to a \nline only while an outward call was \nbeing made. To the search for the \nmost efficient method of doing this, \nmuch thought has been given and \nmany studies made which, embodying \nthemselves in laboratory development, \nhave resulted in the step-by-step line\n\ntroduction of the Keith or plunger. \ntype line switch in 1907. This is a \ncomparatively inexpensive switch one \nof which is connected to each subscrib- \u2014 \ner\u2019s line. It starts operating as the \nsubscriber lifts the receiver from the \nhook and completes a connection to \none of a group of ten trunks leading \nto selectors before dialing is begun. \nLine switches of a large number of \nlines (the exact number being depen- \ndent on traffic conditions) have ac- \ncess to the same group of ten trunks \nwith the result that the number of \nselectors is much smaller than the \ntotal number of lines.\n\nThe possible reduction in selectors \ndepends for the most part on the num- \nber that may be reached by any one \nline, or conversely, on \nthe number of lines\n\nFig. 1\u2014Diagrammatic representation of the advantages \nfrom greater mutuality between selectors and lines\n\nfinder. Although the mechanism it- \nself is of the step-by-step type, the \nfunction it performs owes its incep- \ntion to the panel system, a later de- \nvelopment of the dial-system method \nof telephone switching.\n\nvice is fixed by the \nnumber of calls be- \ning placed at one \ntime but only if each \nselector had access to \nall the lines could this lower limit be \nreached. With a lesser degree of \naccessibility the chance distribution of \ncalls enters to require a greater \nnumber.\n\nConsider for example a small cen- \ntral-ofice of one thousand lines over \nwhich a maximum of seventy outgo-\n\ning calls are made at any one time. \nAs an initial assumption let each se- \nlector have access to twenty lines so \nthat the thousand lines in the office \nmay be divided into fifty groups of \ntwenty lines, each group being served \nby a certain number of selectors. The \nseventy outgoing calls being made sim- \nultaneously will not, of course, be \nequally divided among these groups. \nChance affects the distribution so that \nsome groups will have more and \nothers less than the average number. \nA possible distribution is indicated on \nFigure 1 in numbers written under \neach group. With this distribution and \nassuming that the number of selec- \ntors is to be so chosen that all outgo- \ning calls can be handled immediately, \nthere must be\u2014-in the case given\u2014 \nfive selectors serving each group as \nthat is the maximum number of calls \noccurring at the same time in any one \ngroup. [he minimum number of se- \nlectors possible\u2014seventy for the \nthousand lines if there were complete \naccessibility \u2014 is seven per cent of the \nnumber of lines while with the ar- \nrangement indicated, where each se- \nlector has access to only twenty lines, \nthe selectors are twenty-five per cent \nof the number of lines.\n\nIn grouping lines an effort is always \nmade to obtain a balance by so sorting \nthe lines with lower or higher calling \nrates among the groups, that the aver- \nage and peak calling rate for all \ngroups is the same. Over a period of \ntime, therefore, all groups will have \nhad at one time or another the same \nnumber of simultaneous calls. The \ndiversity indicated on Figure 1 is only \nthat existing at any one moment. A \nfew minutes later the grouping prob- \nably would be entirely different, with \na group now having no calls perhaps \nthen having five, and this shifting of\n\nIncreasing the accessibility by mak- \ning each set include one hundred lines, \nindicated by the brackets on the dia-\n\ngram, reduces the ratio of selectors \nto lines to twelve per cent. If the in- \ncrease of accessibility is carried to two \nhundred lines the ratio of selectors to \nlines may be lowered to ten per cent. \nThe actual possible decrease in selec-\n\nderived from somewhat intricate \nprobability calculations based on such \nassumptions as the average duration \nof a call, and the calling rate, and \nprobably would not, in any particular \ncase, fit the figures given in the illus- \ntration. These are arbitrarily picked \nmerely to bring out the possible reduc-\n\ncent, but the use of a line switch hunt- \ning over five trunks to selectors would \nreduce this figure, for example given \nin Figure 1, to twenty-five per cent. \nIn this case a group of twenty lines \nwould be arranged to be served by \nfive selectors.\n\ntion in selectors due to the effect of \nchance distribution and greater acces- \nsibility.\n\nIn this example only the matter of \nmutual accessibility has been men- \ntioned; the method of obtaining this \naccessibility in practice has not been \nbrought in. Actually there are two \nmethods in general use as has already \nbeen indicated: by line switches, or by \nline finders. It was natural that the \nline switch should have been devel- \noped first because it accomplished a \nconsiderable reduction in the number \nof selectors with a simple mechanism \nand a minimum number of terminals \nover which hunting would have to be \ndone. Without either line switches \nor line finders the ratio of selectors to \nlines is, of course, one hundred per\n\ning the line switches so that they can \nhunt over a greater number of trunks. \nBased on the data of Figure 1, a line \nswitch hunting over nineteen trunks \nwould reduce the ratio of selectors to \nlines to nine and one-half per cent. \nThe line switch method of obtaining \naccessibility, however, becomes ex- \npensive when it is attempted to obtain \na large degree of accessibility. A \nswitch is required for each line and \nevery increase made in the number of \ntrunks over which the switch will hunt \nmeans an increased cost which is mul- \ntiplied by the number of lines. De- \nvelopment in the Bell System has been \nalong the line of the alternative meth- \nod of obtaining accessibility by use of \na line finder. With this method the \nfinding process is reversed; instead of\n\n\u2018 | > \nj | \nSWITCHING | SWITCHING SWITCHING | SWITCHING SWITCHING | SWITCHIN | SWITCHING __|_ SWITCHING \nRELAY RELAY RELAY RELAY RELAY RELAY\n\nhaving the lines hunt for an idle selec- \ntor, the selectors are made to search \nfor the calling line. A line finder is \nprovided for each selector so that \neach reduction in the number of selec- \ntors brings a like reduction in the \nnumber of line finders. This proves \nto be a more satisfactory way, from \nan operating standpoint, of solving \nthe problem.\n\nTo obtain the same reduction in se- \nlectors, however, the wipers of a line \nfinder must have access to many more \ncontacts than those of the line \nswitches. When the ratio of selectors \nto lines is twenty-five per cent, for ex- \nample, each of the 250 line finders \nwould require access to twenty lines \nwhereas had line switches been used \neach of the thousand would have re- \nquired access to only five trunks to \nselectors. It is desirable to have the \nhunting done between the instant the \nreceiver is lifted from the hook and \nthe time the subscriber is ready to \ndial so that the difficulty is not so \nmuch in actually hunting over a larger \nnumber of contacts as in accomplish- \ning it in this short time. The manner \nin which this rather difficult feat is\n\nAs a basis for the step-by-step line \nfinder there was available the stand- \nard step-by-step mechanism out of \nwhich the selectors and connectors are \nmade. Its contacts are arranged in \ntwo banks, each having ten levels with \nten contact positions arranged around \na section of circular arc. A pair of \nwipers for each bank, by suitable re- \nlays and mechanisms, can be moved \nup one level at a time and then ro- \ntated around the arc to any one of \nthe ten positions. The movement is \nalways step-by-step, that is, up one \nlevel at a time or around one contact \nposition at a time, which gives the \nname to the complete system of \nswitching.\n\nAs it was originally used this switch \nhad access to one hundred lines. Four \ncontacts were available for each line, \nthe two banks each of one hundred \npairs of contacts just making up this \nnumber. In the more recent systems \neach line requires only three contacts \nso that by adding a third bank to the \nswitch six hundred contacts, or two \nhundred lines, become available. An\n\nCOMMUT) COMMUTA\u2019 COMMUTATOR COMMUTATOR \n; \n~~ \n\u201cOo \n<a <a < <a \n| LINE FINDER 2 LINE FINDER | LINE FINDER | LINE 2 o INE FINDER | LINE FINDER 2 \nWIRE SUB-GROU |suB-GRouP\u2019 \n| RELAY . | RELAY ' | | RELAY \nSTART \nADVANCE RELAYS ADVANCE RELAYS \u2014- ADVANCE RELAYS ADVANCE RELAYS | \nTO SuUB- \nSTART WIRE GROUP 9\n\nTHE GROUP RELAYS ARE CONTROLLED BY THE LINE CIRCUHS OF THE HOME LINES WHICH ARE SHOWN IN THE FIRST\n\nFig. 4\u2014Schematic arrangement of sub-group relays, commutators, and start wires\n\nFig. 5\u2014Line-finder frames in a central office. Each group of two hundred lines \nhas only sixteen line finders associated with it\n\naccessibility of two hundred lines was \ndeemed sufficient to reduce the num- \nber of selectors to a satisfactory total \nand the switch thus developed is \nshown in Figure 2.\n\nAlthough this switch gives access \nto two hundred lines there are only \none hundred stopping positions of the \nbrush, ten rotational positions on each \nof the ten levels. Which line is chosen \nof the two that terminate in each po- \nsition depends on the action of relays \nassociated with the switch. If one of \nthe pair of lines running to a brush \nposition were in use and a call should \ncome in on the other, another of the \nfinders serving that group would \n\u201chunt\u201d for it, as the two hundred lines \nrunning to any finder are multipled \nto each of the other finders serving \nthat particular two hundred lines.\n\neach group of two hundred lines is \ndivided into ten sub-groups of twenty \nlines, and each of these sub-groups \nappears on the lowest level of one or \nmore of the line-finder banks, and \nconstitutes the \u201chome lines\u201d for these \nfinders. \u2018This same sub-group of lines \nappears in the second level of another \nset of line finders, in the third level \nof another set, and so on. The line \nfinder \u201cstart\u201d circuit is so arranged \nthat when a call originates, one of the \nline finders that has the calling line \namong its \u201chome lines\u2019\u2019 begins to hunt \nfor it. If these particular finders were \nbusy the start lead would be trans-\n\ning line in their second level, and if \nthese in turn were busy, to the finders \nwith the line in their third level, and \nsoon. In this way minimum hunting \ntime is required for each call. The \nmanner in which the lines are mul-\n\ntipled to the different finders and the \nway in which the start lead is trans- \nferred from one to another may be \nunderstood by a study of Figures 3 \nand 4.\n\nThese two-hundred-point line find- \ners, a typical installation of which is \nshown in Figure 5, have now been in \nuse for some time. In the installa- \ntion illustrated, three groups of find- \ners, each group serving two hundred \nlines, are mounted in a vertical frame.\n\nNE of the telegraph systems \nbeing used to a steadily in- \ncreasing degree on Bell Sys-\n\nessary apparatus is now available, re- \ntaining the electrical characteristics \nof the older assembly but introducing \nan improvement in compactness and \nconvenience.\n\nBy the use of carrier currents in a \nfrequency range approximating that \nof the voice currents of telephone con- \nversations, the system has been made \nsuitable for a toll plant designed \nprimarily for telephone transmission. \nAt intermediate points the complex \ncurrent conveying all the telegraph \nmessages is amplified by ordinary\n\nCarrier-telegraph apparatus for a single wo-way channel, connected to a toll line\n\nfour-wire telephone repeaters. The \nessential elements of the system are \na generator for producing the carrier \ncurrents; relays for controlling their \ntransmission in accordance with a \ntelegraph code; filters for separating \nthe different frequencies, combined \nfor transmission over the line; and \nvacuum-tubes for rectifying each of \nthe separated alternating currents \ninto direct current for operating the \ntelegraph receiving apparatus. Ter- \nminal apparatus embodying these\n\nunits, connected at each end of a toll \nline, make it a path for telegraph \nmessages rather than for conversa- \ntions.\n\nIn the schematic the circuit ar- \nrangement is shown. Operation of a \nsubscriber\u2019s key, or automatic trans- \nmitter, controls direct current through \nthe windings of the sending relay, \nwhich in turn controls the alternating \ncurrent passing onto the line. During \nspacing signals this relay short-cir- \ncuits the alternating current, and dur- \ning marking signals allows it to flow \nthrough the sending filter to the line. \nAt the incoming end, the receiving fil- \nter of each channel terminal selects \nthe proper current, which is amplified \nand rectified by the vacuum tubes of \nthe detector circuit. The rectified \ncurrent operates the receiving relay, \nwhich in turn controls the receiving \nsubscriber\u2019s sounder or printer.\n\nbe sent as desired, but not simultane- \nously, only one circuit is required be- \ntween the subscriber\u2019s premises and \nthe toll office, but a one-way carrier \nchannel is needed in each direction. \nAn important operating feature of \nthe terminal equipment is the provi- \nsion made for monitoring. An atten- \ndant at the central office must be able \nto observe operation of the circuits to \nbe sure that everything is in good \norder, and must communicate with \nthe subscribers\u2019 operators from time \nto time, in emergencies as well as \nabout routine matters. The circuit \nsketch shows the sounders and meters \nby which operation can be observed, \nand the telegraph key by which the \nattendant can take control of the cir- \ncuit. It is important that the monitor- \ning apparatus be as convenient as pos- \nsible, since many of the telegraph \nchannels are used by financial houses \nin whose service int\u00e9rruptions or de- \nlays might be costly. Following then- \ncurrent telegraph practice, the orig- \ninal voice-frequency equipment had \nmonitoring apparatus permanently \nconnected to each channel terminal, \nalthough it was not possible for the \nattendants to give their oversight to \nmore than a few repeaters at a time. \nSince experience has shown that ade- \nquate supervision can be maintained \nwith fewer sets of monitoring ap- \nparatus, in the new form of equip- \nment only one set has been provided \nfor each four channel terminals. \nThis use of common monitoring \nequipment, connected into any circuit \nat will by jacks, plugs and cords, has \nbeen greatly facilitated by the con- \nvenient location of the apparatus in- \nvolved. It will be noted from the \nphotographs that like pieces of ap- \nparatus have been grouped on mount- \ning plates especially designed for\n\nthem. At the top of each bay are \nplates bearing retardation coils, con- \ndensers, resistances and other minor \nelements normally requiring no atten- \ntion; the telegraph relays, which need \noccasional maintenance, are located \nnear the bottom; and the central por- \ntion of the bay is left clear for the \napparatus used in controlling the re- \npeaters. An improvement is thus real- \nized over the method of mounting \nboth fixed and adjustable apparatus \nfor a particular circuit on one panel,\n\nFig. 2\u2014Two bays of panel-mounted \nterminal equipment, equivalent to a single \nbay new equipment\n\nwhich of course could not be located \nat the most convenient part of the bay \nfor all the circuits.\n\nMounting most of the apparatus \non standard mounting plates has \nmade the bay rather than the panel\n\ngraph relays and the receiving po. \ntentiometers, there are exactly four \nunits of apparatus, one for each of \nthe channel terminals. Where there \nare several units for each circuit on \none of the plates, as in the case of the\n\nthe unit of equipment. Apparatus for \nfour channel terminals and one mon- \nitoring panel are normally assembled \non a channel-iron rack, and are wired \nin a single local cable. In contrast, \nthe older panel-mounted equipment \nrequired a complete bay for two chan- \nnel terminals and the two monitor- \ning panels associated with them. On \nmany of the mounting plates in the \nnew bays, such as those for the tele-\n\nresistances and keys, these have been \narranged in four separate groups. \nThe only exception to the sub-group- \ning on the mounting plates, aside \nfrom the monitoring panel, is the fil- \nters; each of these is assembled com- \nplete on an individual plate. As a re- \nsult the bay is divided in a general \nway into four vertical groups, each \nconstituting a channel terminal. Or-\n\ngroups of three, the number required \nfor twelve channels in each direction.\n\nThe advantages of this type of \nmounting are apparent in Figure 1. \nAny particular piece of apparatus is \nlocated in the same position in the \nbay for every repeater, a feature es- \npecially important for the circuit ele- \nments requiring attention\u2014the jacks, \nkeys and potentiometers. Convenience \nalso follows on the wiring side of the \nbays from the arrangement, as well \nas from the accessibility of the ter- \nminals of all the apparatus.\n\nAlong with the improved assembly, \nsimpler types of apparatus have been \nintroduced. In the older terminal re- \npeater, an expensive forty-step rheo- \nstat was used for the variable resis- \ntance in series with the subscriber\u2019s \nloop. This has been replaced by a \nseries of flat-type resistances having \nvalues of fifty ohms, one hundred \nohms, two hundred ohms, and so on \nup to eight hundred ohms, each of \nwhich can be short-circuited by a \npush-button key. Any value of resis- \ntance from fifty ohms to the total can \nbe secured, in steps of fifty ohms, by \noperation of the proper combination \nof keys, with five or six resistances \ntaking the place of the forty used in\n\nIn the Bell System plant, voice-fre- \nquency telegraph systems are in use \nalong the main toll cables, with a to- \ntal of about 170,000 miles of two- \nway channels in operation on October \nI, 1928. There were at that time \nthirty-three systems in service, the \nlongest being from New York to Chi- \ncago, a distance of 884 miles. These \nsystems form a network together with \nthe open-wire high-frequency tele- \ngraph systems, which touches almost \nevery important city in the country, \nand is an important factor in main- \ntaining the highest grade of telegraph \nservice now offered \u2014 telephone type- \nwriter service at sixty words per min- \nute in each direction. Combining the \nadvantages of reliability and simplic- \nity with savings in first cost and floor \nspace, the new equipment will pro- \nmote further the use of such facilities, \nthereby helping to secure the most \nproductive use of the toll cables.\n\nproblem of an econom- \nical source of ringing current \nfor the smaller common-bat-\n\nringing machine of such a capacity \nthat it will take its place between the \ntwo types already in use. In one \ntype of these present machines, used \nprincipally in the smallest offices and \nin private branch exchanges, battery \ncurrent is converted into alternating \ncurrent for ringing by a vibrating- \nreed interrupter and a transformer. \nAlthough this apparatus is satisfac- \ntory at the switchboards for which it \nis intended, by its nature it does not \nregulate voltage closely enough for \noffices presenting a greater variety of \nringing conditions. It is intended for \nswitchboards at which ringing is con- \ntrolled manually and at which busy \nsignals and other information are \ngiven by voice, rather than by signal \ntone. No provision is made, there- \nfore, for intermittent current for ma- \nchine ringing, or for signal tones. \nWhere requirements call for inter- \nrupted currents and tones for sig- \nnalling, and for ringing current con- \ntrolled by the machine rather than \nby the operators, a ringing machine \nof the other type \u2014 Type P \u2014 has \nbeen used. In this, a motor driven \nalternator furnishes twenty- -cycle volt- \nage and rotating drums interrupt the \nringing and signalling currents as re- \nquired. The capacities of these ma- \nchines, ranging from one to eight am-\n\nperes, make them adequate for the \nlargest offices, and their output volt- \nage and frequency, which are con- \ntrolled by regulators, are sufficiently \nclose for the most severe conditions. \nFor the smaller offices using machine \nringing, however, certain of the prop- \nerties of these ringing generators \nhave not been needed. Regulation is \nconsiderably closer than is necessary \nin many such offices, and the current \noutput of the smallest Type P ma- \nchine exceeds requirements of maxi- \nmum traffic by a substantial margin.\n\nThe new machines have therefore \nbeen designed to embody only those \nfeatures useful in offices with a busy- \nhour trafic of three thousand calls \nor less. A small twenty-cycle motor- \ngenerator set was selected, and high- \nspeed interrupters were added for \nbusy and other tones, a commutator \nfor pulsating current, and low-speed \ninterrupters to make possible the con- \nnections needed for machine ringing. \nThe generator has a capacity of a \nhalf ampere, delivered at 72 to 88 \nvolts. No voltage regulator on the \noutput has been provided. In the case \nof the battery-driven sets for which \nthe input voltage may vary between \ntwenty and twenty-eight, a centri- \nfugal regulator has been provided to \nhold the speed between 1100 and \n1200 revolutions per minute. The \ngenerator voltage is held within suf- \nficiently close limits by additional \nfield windings in series with the mo-\n\ntor armature. Neither the regulator \nnor the compound winding will be \nrequired for machines to be driven \nby central-station power.\n\nOn the generator shaft are two \nslip-rings and a ring for delivering \npulsating current. Tones of 480 and \n160 interruptions per second are se- \ncured by a brass drum on an extension \nof the generator shaft; \none brush feeds direct \ncurrent to the drum, \nand two pairs of col- \nlecting brushes pick up \nthe interrupted cur- \nrents. The low-speed \ninterrupter, at the end \nof the machine, is \ndriven by a worm-gear. \nInstead of another \ndrum with intermittent \ninsulated areas in the \npath of the brushes, \nthere is a group of ro- \ntating discs which bear \nrollers near their circumferences. \nDuring rotation each set of rollers \noperates a group of flat contact \nsprings, similar to the spring pileup \nof a key or a relay, which make and \nbreak contact 60 or 120 times per \nminute for signalling current, or con- \nnect and disconnect ringing current at \nthe standard ringing intervals. No \nprovision has been made for current \nto collect and return coins, but that \ncan be supplied readily in these small \noffices, by a bank of dry cells or other- \nwise.\n\nis also used when the voltage is near \nthe lower limit to step it up for test- \ning the tripping relays. When the \nvoltage rises, a relay switches the \ntesting circuit to the generator ter- \nminals.\n\nThe whole set, except for the trans- \nformer, is mounted on a slate slab \nwhich in turn is screwed onto a\n\nThe new ringing machine, mounted on its slate and wood \nbase. The motor-speed regulator is at the extreme right\n\nwooden base. One machine is ade- \nquate in capacity, but two will ordi- \nnarily be provided to insure continu- \nous service. The light weight and \ncompactness of the sets make it pos- \nsible to mount them on a pipe frame- \nwork, one above the other; the floor \nspace needed is thus reduced to less \nthan half of that occupied by two \ntype P ringing machines.\n\nEconomy is not restricted to floor \nspace, however; two of the sets cost \nless than half as much as two of the \ntype P sets which would otherwise be \nneeded for an office. The resulting \nreduction in the power-plant cost will \nhave a tendency to extend the appli- \ncation of machine ringing in the \nsmaller offices.\n\nMembers of the Laboratories are \nnaturally attentive to civic enterprises \naffecting the convenience of trans- \nportation to and from 463 West \nStreet and the appearance of its en- \nvirons. West-side improvement pro- \njects have been under discussion for \nsome time and have taken many \nforms. The Board of Estimate of the \nCity of New York has recently ap- \nproved plans for an elevated express \nmotor highway along the west side \nof the city.\n\nThe highway will extend from \nCanal Street to Seventy-second Street. \nConsisting of two roadways, each \nthirty feet in width, it will provide \nfor three lines of traffic in each direc- \ntion. Sets of ramps about a mile \napart will furnish access to the high- \nway at points other than its termini. \nFour ramps with ten-foot roadways \nwill compose each set, affording as- \ncent and descent to uptown and down- \ntown trafic. From Canal Street to \nFifty-ninth Street the highway will be\n\nsupported on sets of three columns. \nThe sets will be about sixty feet \napart, except where the highway \nspans cross-streets, and the columns \nof each will be thirty-four feet apart.\n\nProceeding along West Street, the \nhighway will pass about seventy feet \nwest of the front of our building. \nThe nearest set of ramps will be \nsomewhat south of Twenty-third \nStreet; the picture above shows the \nanticipated highway at that point. \nEnding conveniently near the Holland \nTunnel, and connecting with the Hud- \nson River Bridge, the highway will \nyet further facilitate automobile pas- \nsage between the Laboratories and \nNew Jersey.\n\n-The section from Canal Street to \nFifty-ninth Street will cost about $16,- \n000,000 and will be financed by assess- \nment on the taxpayers of the Bor- \nough of Manhattan. It is expected \nthat work on this section will begin \nin a month or two and will continue \nfor about three years.\n\nService Honors for 1928 . \nMORE THAN THIRTY YEARS a \nArthur E. Peterson Wilton L. Richards Arnold S. Bertels \nResearch Department Consulting Historian Systems Development a \nThirty-five years Fifty years Forty years =) \n| \nConrad Schaul John F. Toomey a \nDevelopment Shop Systems Development - ; \nThirty-five years Thirty-five years - \n{249} \n} 4 \n|\n\nAdolph Bregartner Frederick E, Anderson \nSystems Development Systems Development\n\nGoodwin Rosenblum Albert J. Ketschke William McA. Smith \nDevelopment Shop Research Department Systems Development\n\nEdward J. Harrington George Thompson \nDevelopment Shop Systems Development\n\nAdolph Sulzmann Charles Borgmann George Hatcher \nResearch Department Manual Equipment Engineer Plant Operation\n\nGeorge Hess Sergius P. Grace \nPlant Construction Assistant Vice-Presidert\n\nArthur K. Forsyth George Dodd \nApparatus Development Apparatus Development\n\nGeorge F. Morrison Charles H. Klein \nPower Engineer Apparatus Development \nPlant Operation\n\nDr. JEWETT attended an assembly \nluncheon of the Boston Chamber of \nCommerce at Boston on January 10 \nand spoke on \u201cThe Romance of Re- \nsearch in the Telephone Industry.\u201d\n\nON DECEMBER 29, Dr. Jewett con- \ntributed to a symposium on the \u201c\u2018Or- \nganization of Scientific Research in \nIndustry\u201d, at a meeting of the Ameri- \ncan Association for the Advancement \nof Science. His subject was \u2018Finding \nand Encouragement of Competent \nMen.\u201d\n\nJ. E. Moravec has been appointed \nAssistant Vice-President in charge of \nStaff Department, effective January \n14. His organization is as follows: \nJ. W. Farrell, Secretary and General \nAttorney, in charge of the Legal De- \npartment; W. B. Wallace, Treasurer, \nin charge of the Financial Depart- \nment; E. J. Santry, General Audi- \ntor, in charge of General Accounting \nDepartment; B. B. Webb, Commer- \ncial Relations Manager, in charge of \nCommercial Relations Department; \nJ. S. Hartnett, General Service Man- \nager, in charge of the General Ser- \nvice Department; W. B. Sanford, \nPlant and Shops Manager, in charge \nof Plant and Shops Department.\n\nR. B. Bonney, Educational Direc- \ntor of the Mountain States Telephone \nand Telegraph Company, was a guest \nof the Laboratories January 2 to 16 \nfor an extended inspection of the fac- \nilities here and for observation of \ncurrent developments. On account of \nhis interest in college relations work, \nhe wanted, among other objectives,\n\na general picture of the Laboratories \nas a research organization, and in- \nformation on studies now in progress.\n\nAT THE JANUARY 7 MEETING of \nthe Colloquium, J. A. Becker spoke \nScience of Coated Fila-\n\nments\u201d. On January 21, J. E. Harris \nspoke on \u2018\u201c\u2018Application of the Phase \nRule to Metallurgical Problems\u2019. \nTue AcousTiCAL SOCIETY OF \nAMERICA was formed at a meeting \nheld here on December 27, for the \npurpose of bringing together workers \nin all branches of pure and applied \nacoustics. Among its activities will \nbe the provision of a medium of pub- \nlication for papers on acoustics, for\n\nwhich there is acute need; such papers \nhave hitherto been widely scattered.\n\nElected to temporary office were: \nPresident\u2014Harvey Fletcher of the \nLaboratories; Vice-President\u2014Pro- \nfessor V. O. Knudsen of the Univer- \nsity of California; Secretary\u2014 Wal- \nlace Waterfall of the Celotex Com- \npany; Treasurer\u2014C. F. Stoddard of \nthe American Piano Company. A \ncommittee was appointed by Dr. \nFletcher to consider the details of or- \nganization, and the first regular meet- \ning will take place here some time \nin April. There were present forty\n\nof the leading authorities on acous. \ntics, among whom were Professor D. \nC. Miller of the Case School of Ap- \nplied Science, Professor F. A. Saun. \nders of Harvard University, J. P. \nMaxfield of the Victor Talking Ma. \nchine Company, R. V. Parsons of the \nJohns Manville Company, and Pro. \nfessor F. R. Watson of the Univer. \nsity of Illinois. After the meeting, \nDr. Arnold entertained the members \nat luncheon.\n\nMembers of the Acoustical Society of America: left to right: back row: W. P. \nMason, B.T.L.; J. C. Steinberg, B.T.L.; V. L. Chrisler, Bureau of Standards; \nE. J. Schroeter, Macoustic Eng. Co.; E. C. Wente, B.T.L.; W. C. Jones, B.T.L.; \nsecond row from back: L. J. Sivian, B.T.L.; E. L. Norton, B.T.L.; W. A. \nMacNair, Victor Co.; R. F. Mallina, Victor Co.; Lonsdale Green, Jr., Johns- \nManville Corp.; R. H. Schroeter, Macoustic Eng. Co.; H. W. Lamson, Gen. Radio \nCo.; C. N. Hickman, American Piano Co.; D. G. Blattner, B.T.L.; second \nrow from front: H. A. Erf, Northern Ohio Tel. Co.; H. C. Harrison, B.T.L.; \nJ. B. Kelly, B.T.L.; R. L. Wegel, B.7.L.; H. A. Frederick, B.T.L.; N. &. \nFrench, A. T. & T.; C. W. Hewlett, G. E. Co.; A. T. Jones, Smith College; \nIrving Wolff, R.C.A.; J. B. Taylor, G. E. Co.; front row: F. A. Saunders, \nHarvard U.; R. V. Parsons, Johns-Manville Corp.; D.C. Miller, Case School of \nApplied Science; Wallace Waterfall, Celotex Co.; V. O. Knudsen, U. of Cali- \nfornia; H. Fletcher, B.T.L.; C. F. Stoddard, American Piano Co.; J. P. Maxfield, \nVictor Co.; F. R. Watson, U. of Illinois; F. K. Richtmeyer, Cornell U.; \nG. R. Anderson, U. of Toronto\n\nField Engineer, visited Seattle, Ta- \ncoma, Portland and other cities in \nWashington and Oregon during the \nsecond week in December in connec- \ntion with Field Engineering duties.\n\nA. F. GILSON AND P. S. OLMSTEAD \nattended a regular survey conference \non telephone receivers at Hawthorne \nduring the latter part of December.\n\nW. A. Boyp, in company with a \ncommittee for the study of protective \nmetal finishes, visited Norfolk, At- \nlanta, Birmingham and New Orleans \nduring the early part of January.\n\nA. G. DALTON AND R. J. Nossa- \nMAN visited Philadelphia during the \nlatter part of December in connection \nwith Field Engineering work.\n\nH. F. DopceE was the author of an \narticle \u2018Using Inspection Data to \nControl Quality\u201d, which appeared in \nMANUFACTURING INDUSTRIES for \nNovember and December.\n\nE. F. Kincspury addressed the \nWomen\u2019s Club of Rutherford, New \nJersey, on television on January 7.\n\nAMONG PAPERS presented before \nthe American Physical Society at its \nmidwinter meeting at Columbia Uni- \nversity were \u201cThe Maximum Excur- \nsion of the Photoelectric Long Wave \nLimit of the Alkali Metals\u201d, by H. \nE. Ives, read on December 27, and \n\u201cA Wave Mechanism of Quantum \nPhenomena\u201d by R. V. L. Hartley, on \nDecember 29. On the latter date, a \npaper \u201cAnomalous Dispersion of \nElectron Waves by Nickel\u201d, by C. J. \nDavisson and L. H. Germer, was also \ngiven.\n\nT. C. Fry was one of the presiding \noficers of a joint meeting of the \nMathematical and Physical Societies.\n\nResearch Council on \u201cSubstitutes for \nSound\u201d at St. Louis on December 8. \nGEORGE KNaAP?P gave an illustrated \ntalk on television at the Columbian \nClub at Elizabeth, New Jersey.\n\nG. R. YENZER was at Hawthorne \non December 10 in connection with \nresistance requirements for condenser \ntransmitters.\n\nR. M. Burns inspected some in- \nstances of corrosion in central offices \nat Atlantic City on December 3 and 4.\n\nB. L. CLARKE made a survey of \nlubrication problems at Hawthorne \nand visited centers of lubrication re- \nsearch at Washington, Minneapolis, \nMadison and Pittsburgh.\n\nC. L. HipPpeNsTEEL C. W. \nBORGMANN discussed corrosion prob- \nlems with engineers of the New Jer- \nsey Zinc Company at Palmerton, \nPennsylvania.\n\nA. R. Kemp visited the Columbian \nRope Company at Auburn, New \nYork, in regard to cable develop- \nments.\n\nR. L. PEEK attended the Plasticity \nSymposium held at Lafayette College \nin Easton, Pennsylvania.\n\nR. M. Burns anv D. J. SALLEY \nattended the Soil Corrosion Confer- \nence at the Bureau of Standards at \nWashington and made experimental \ninvestigation of protective coatings at \nHawthorne, from December 9 to 29.\n\nHARVEY FLETCHER attended com- \nmittee meetings on research for the \ndeaf at Washington, D. C., in connec- \ntion with the Anthropology and Psy- \nchology Division of the National Re- \nsearch Council.\n\nR. O. MERCNER visited the Vibra- \ntion Specialty Company at Phila- \ndelphia on December 15 to observe \nthe balancing of high-speed machines.\n\nE. C. WENTE\u2019s article, \u2018Speech \nInterpretation in Auditoriums\u201d\u2019, which \nappeared in the RecorpD for October, \nhas been reprinted in the December \nissue of the Gamma Alpha Record.\n\nH. EckuaArpT made phonograph \nrecords at Houlton, Maine, of trans- \natlantic conversations between Austin \nBailey of the Long Lines Department \nat 24 Walker Street, and Colonel \nShreeve, at London, on January 7 and \n8. Simultaneously, E. Peterson re- \ncorded these conversations at the La- \nboratories.\n\nC. S. GorDoN visited the New \nYork Insulated Wire Company at \nWallingford, Connecticut, on Decem- \nber 6 in connection with the prepara- \ntion of trial lots of drop wire.\n\nE. M. Honan was in Boston to \ndiscuss with members of the Simplex \nWire and Cable Company problems \ndealing with the manufacture of rub- \nber-insulated wire.\n\nC. D. Hocker ann F. F. FARNs- \nWORTH visited several of the asso- \nciated telephone companies in the \nmiddle west, as well as the plants of \nseveral automobile body manufactur- \ners and paint companies, to investi- \ngate paint finishes for automotive \nequipment.\n\nC. Q. LUMSDEN was engaged in stu- \ndies of chestnut poles at Shipman, \nNatural Bridge and Waynesboro, \nVirginia, from December 20 to 22.\n\nM. B. Lone addressed the Phila- \ndelphia Section of the A. I. E. E. on \nJanuary 14, on the subject of the pho- \ntoelectric cell and its use in com- \nmunication. Prior to the meeting, \nMr. Long was the guest of the En- \ngineers Club at dinner.\n\nEyes and Their Use in Communica. \ntion\u201d before the Engineers Club of \nSt. Louis on January 9, the South. \nwestern Bell \u2018Telephone Company em. \nployees on January 10, and on Janu- \nary 11, he addressed the engineering \nstudents of Washington University, \nThe following day, Mr. Mills spoke \non the same subject before the Round \nTable of the University.\n\nS. P. GRACE addressed the staff of \nthe Carson Peck Memorial Hospital \nin Brooklyn on January 10, on recent \ndevelopments of the Laboratories, \nMr. Grace also lectured before the \nEngineers Club of Cincinnati.\n\nT. W. CLarkeE has been elected to \nmembership in the Edward J. Hall \nChapter of the Telephone Pioneers.\n\nHELEN O. LUEDTKE, a stenog- \nrapher in the Transcription Depart- \nment, died on Friday, January 4. Miss \nLuedtke had been a member of the \nLaboratories since February, 1926.\n\nH. M. ANDERSON, a shop mechanic \nin the Plant and Shops Department, \ndied on December 14. Mr. Anderson \nhad been associated with the Labo- \nratories since March 6, 1918.\n\nMEMBERS OF THE PATENT Dr- \nPARTMENT who went to Washington \nduring the period from December 11 \nto January 10, in connection with the \nprosecution of patents were: E. W. \nAdams, W. C. Kiesel, I. MacDonald, \nT. P. Neville, J. W. Schmeid and G. \nH. Stevenson.\n\nF. F. Lucas spoke on metallog- \nraphy before the American Society \nfor Steel Treating at Buffalo.\n\nS. J. ZAMMATARO\u2019S article, \u201c\u2018Devel- \nopment of the Impedance Bridge,\u201d \nwhich appeared in the Recorp for \nDecember, has been reprinted in full\n\nJ. R. TowNnsenp\u2019s article, \u201cNew \nSpecifications for Raw Materials,\u201d \nwhich appeared in the Recorp for \nFebruary, 1928, has been reprinted in \nthe December issue of \u201c\u2018Instruments.\u201d\n\nO. L. WALTER discussed suggested \nchanges in the 202-B reproducer set \nat Hawthorne.\n\nJ. C. Herser\u2019s article, \u201cFrequency \nControl for Broadcasting\u201d, which ap- \npeared in the ReEcorD for September, \nhas been reprinted in the November \nissue of the Bent of Tau Beta Pi.\n\nS. J. ZAMMATARO AND T. SLONC- \nZEWSKI visited the Leeds and North- \nrup Company at Philadelphia to in- \nspect shielded-type precision bridges.\n\nH. BROADWELL attended confer- \nences at Hawthorne on the manufac- \nture of precision interrupters.\n\nF. H. visited Chat- \ntanooga, Tennessee, and Atlanta, \nGeorgia, in connection with the revi- \nsion of adjustment requirements on \nstep-by-step relays.\n\nW. L. HEArpD, as a member of the \nStandards Committee of the A. I. E. \nE., has participated in the prepara- \ntion of graphical symbols for tele- \nphone and telegraph use. This work \nhas now been completed, the symbols \nand abbreviations have been approv- \ned as an Institute standard and they \nare being transmitted to the Ameri- \ncan Standards Association.\n\nE. H. SmirH spent a week at Chi- \ncago discussing step-by-step central- \noffice equipment.\n\nJ. R. STONE AND H. L. MUELLER \ndiscussed exhaust equipment for pneu- \nmatic tube systems and various power\n\nmachine problems with representa- \ntives of the General Electric factory \nin West Lynn, Massachusetts.\n\nW. O. FULLERTON inspected at \nRochester the new 506-A cordless \nP. B. X. being built by Stromberg- \nCarlson.\n\nV. T. CALLAHAN discussed gas en- \ngine mufflers with members of the \nMaxim Silencer Company at Hart- \nford, Connecticut.\n\nE. J. KANE ann J. M. Ducutn in- \nvestigated busy-tone wiring in the \nstep-by-step office at Hartford.\n\nG. E. Battey R. E. NosBie \nrecently inspected a new type of super- \nstructure for frames and racks in- \nstalled at Laurel Springs, New Jersey.\n\nL. P. BARTHELD discussed cabling \nproblems in the new Chicago Toll \nOffice with Illinois Bell engineers.\n\nC. ECKBERG is arranging for in- \nstallation of improved radio broad- \ncasting circuits at various repeater \nstations on the Pittsburgh-New York \ncable.\n\nC. BORGMANN visited Boston, Chi- \ncago and Rochester in connection with \nequipment problems.\n\nW. J. LAcerTE visited Hartford \nand Atlantic City in connection with \nimproved circuit arrangements for \nstep-by-step central offices.\n\nOn page 186, last paragraph, the \nnumber of regulators in the Long \nLines plant should have been given \nas 41.\n\nIn the second paragraph on page \n189 the second sentence should read \n\u201cUnder these conditions on a full \nsize cable there is an IR drop along\n\nDuring the month of March the \nClub will hold an indoor handball \ntournament at Labor Temple, Four- \nteenth Street and Second Avenue, \nManhattan. Play will commence \nTuesday evening, March 5, and \nmatches will be held on Tuesday and\n\nThursday evenings from 5:30 to \n7:30, for four weeks. \u2018The single- \nwall form of play will be used, and \nall games will be singles. The tourna- \nment will be conducted on an elimina- \ntion basis, with all matches up to the \nsemi-finals consisting of a single game. \nSemi-finals and finals will be decided \nby two games out of three. The win- \nner will receive an order good for ten \ndollars\u2019 worth of merchandise at Alex. \nTaylor and Company, the second \nman, an order worth five dollars, and \nthe two men eliminated in the semi- \nfinals will each get an order worth \ntwo dollars and a half.\n\nEntry blanks may be obtained from \nD. D. Haggerty, Room 164, and must \nbe returned to him with the entry fee \nof twenty-five cents not later than \nMonday, February 25. Before each \nround, players will be advised of date, \ntime and opponents with whom they \nhave been matched.\n\nPlans have been completed for the \nfinal tournament of the indoor golf \nseason, which will be played on Wed- \nnesday evening, February 27, at the \nMiniature Golf Course of America, \n41 East 42nd Street, Manhattan.\n\nAs in previous tournaments all con- \ntestants will play thirty-six holes of \nmedal play in the qualifying round \nafter which they will be divided into \ntwo groups of four flights each with \nprizes for the winner in each of the \neight flights. Sixty-four players will \nbe qualified for the finals.\n\nEntries should be forwarded to D. \nD. Haggerty, and accompanied by an \nentrance fee of $1.50.\n\nby the Club, a valuable prize will be \ndonated by the management of the \ngolf course.\n\nThe annual Photographic Contest, \nopen to all the members of the La- \nboratories, is now under way. Entries\n\nshould reach D. D. Haggerty before \nMarch 1, after which date they will \nbe exhibited and judged. Prints are \nto be divided into four classes: por- \ntraits, landscapes, still life and col- \nored transparency. For further in- \nformation regarding rules and prizes, \ncall Margaret Horne, Extension 786, \nor D. D. Haggerty, Extension 542.\n\nIt is the wish of the music commit- \ntee that every department of the La- \nboratories be represented in the Glee \nClub. The active membership may \nbe divided as follows: Systems De- \nvelopment, 12; Patent, 11; Appara- \ntus Development, 6; Research, 5; Ac- \ncounting, 2; Inspection, 1.\n\nTrained voices are not necessary, \nbut a fairly good voice and some abil- \nity to read music are desirable. The \nGlee Club is now rehearsing for a \nformal recital to be held in April as \na part of the Spring Dance and Enter- \ntainment of the Club. For that rea- \nson, new members will not be ac- \ncepted after March 1. Meetings are \nheld every Wednesday evening from \n5:10 P. M. to 6:10 P. M. in the \nWomen\u2019s Rest Room on the eleventh \nfloor. For further information call\n\nThe men\u2019s bridge club recently \ncompleted its ten-week tournament \nand on Monday, January 7, and Mon- \nday, January 14, sixty men took part \nin an elimination tournament to de- \ntermine who would represent the \nClub in the annual tournament with \nthe Western Electric Bridge Club. \nThis match will be held on Monday \nevening, March 4, at 195 Broadway, \nand forty players from each company \nwill compete for the trophy, now held\n\nA second tournament of ten weeks\u2019 \nduration was started by the members \nof the Club on Monday evening,\n\nJanuary 28. This group meets each \nMonday evening in Rooms 275 and \n277 and the committee in charge is \nalways glad to welcome new players.\n\nThe women\u2019s bridge club played \nthe first game of the mid-winter tour- \nnament on Friday, January 5th. The \ntournament will continue for ten \nweeks and in order to be eligible for \ntournament prizes players must take \npart in at least seven out of the ten \ngames.\n\nThe bridge club conducts a meeting \nevery week, at which prizes are given \nfor high scores. Those interested \nmay apply to Mary V. Lynch.\n\nUntil further notice the orchestra, \nwhich formerly met on Thursday \nevenings, will meet on Tuesdays at \n5:05 in the Women\u2019s Rest Room. A \nbuffet supper will be served by the \nrestaurant management in the Rest \nRoom at 5:45.\n\nThe orchestra, for a time under \nthe leadership of Mr. Egon Elbert, \na professional conductor, is again be- \ning conducted by L. E. Melhuish. \nMr. Melhuish stresses the fact that, \nalthough regular attendance at re- \nhearsals is desirable, the program for \nthe remainder of the season has been \nso arranged as to make it interesting \nfor those who cannot attend each of\n\nA class of music suitable to the \nnumber attending rehearsals will be \nundertaken. Members are encouraged \nto suggest selections for rehearsals, \nwhich the club in turn will be glad to \nprocure. The repertoire now consists \nof standard overtures, operatic selec- \ntions, symphonies and popular selec- \ntions from musical comedies.\n\nseason begins with a reorganization \nof the women\u2019s bowling group under \nthe supervision of a committee of men, \nappointed by the General Chairman \nof the men\u2019s bowling league. This \ncommittee is looking after the inter- \nests of the women bowlers, with a\n\nview to promoting an enthusiastic and \nsuccessful season. \nArrangements may be made\n\nthrough Antoinette Kelly or Marian \nKane to have one\u2019s name placed on \nthe substitute list.\n\nWith the close of the bowling sea- \nson, a dinner and theatre party will \nbe given to the bowlers. At dinner, \nprizes will be awarded to the winners.", "title": "Bell Laboratories Record 1929-02: Vol 7 Iss 6", "trim_reasons": [], "year": 1929}
{"archive_ref": "sim_record-at-t-bell-laboratories_1929-01_7_5", "canonical_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1929-01_7_5", "char_count": 95122, "collection": "archive-org-bell-labs", "doc_id": 158, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc158", "record_count": 97, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1929-01_7_5", "split": "test", "text": "A year ago I ventured to look into the future of 1928 and to see there \na picture of achievement of interest to us all. The accomplishments of the \ntwelve months just past have fully justified that picture. As we come once\n\nmore to the turning of the page to another new year, one and all we are \ncurious as to what we shall find there.\n\nViewing the vast array of problems which confront us \u2014 problems \nwhich arise out of the irresistible pressure in the Bell System for enlarged \nservice or from those great new industries which are building on the basis \nof our work \u2014 it is clear that we have a job to do in 1929.\n\nFor twenty-five years I have been a part of the research and develop- \nment organization of the Bell System. In all that period there has never \nbeen a time when the work immediately ahead bulked so large either \nrelatively or absolutely. The problems which now confront us are numer- \nous, vast and intricate almost beyond belief. Their solution will tax to the \nlimit our ability, ingenuity, organizng capacity and energy.\n\nNone of us is afraid of such a challenge; and since the substantial \nsatisfactions of life are the memories of constructive work well done, we can \none and all welcome the opportunity to show what manner of men and \nwomen Wwe are.\n\nMy greetings of the Season are tempered by the sadness which we all \nfeel at the sudden and untimely loss of our friend and associate, Vice- \nPresident Clifford.\n\nHEN we read in the morn- \ning paper that it rained yes- \nterday in Chicago and the\n\nday before in North Dakota, and \npresently look out of the window and \nsee clouds gathering in the Western \nsky, we think it natural to say \u201cthe \nrain is arriving from the West.\u201d \nThis, however, is loose language; but \nto say precisely what it is that is ar- \nriving from the West would require \nmore knowledge than most of us pos- \nsess, more perhaps than the greatest \nweather experts can justly claim. \nCertainly the water which is about \nto descend into the streets of Man- \nhattan was not over Illinois yester- \nday. Probably it was suspended not \nmany miles away from New York as \ninvisible vapor in the sky, when the \nair suddenly grew cooler and it be- \ngan to condense into visible cloud. \nWhat came from the West at the \nspeed of an express train was not the \nsubstance of the cloud, but the fall in \nthe temperature of the air, or some \nother state of the atmosphere, which \ncreated the cloud in its footsteps as \nit raced along. Rain, of course, is \nmore intelligible. What goes from \nthe clouds to the pavements is mat- \nter in motion, individual corpuscles \nof water which go as units through \nthe air until they splatter upon the \nground or are swallowed up into the \nrivers. When a raindrop falls into \nwater, its identity is gone forever; \nbut for a time there are ripples \nspreading in widening circles over the\n\nwater-mirror from the place where \nthe drop ceased to exist. The ripples \nseem to have inherited the motion of \nthe raindrop; but their onward march \nis not motion of substance. The wa- \nter is not flowing away from the place \nwhere the raindrop perished; the \nwater is only heaving up and down; \nwhat moves away is a mere form or \nmolding of its surface, even less pal- \npable than the travelling fall in tem- \nperature which brought about the \ncoming of the rain.\n\nNow this little history of the rain- \nstorm exhibits three kinds of motion \nonly one of which is strictly motion \nof matter; and since the study and \nthe forecasting of motion is one of \nthe major objects of science, we find \nhere three problems to set before the \narchitects of physics. Only the sec- \nond of the three was solved by New- \nton in his masterpiece, the Principia, \nin which he founded the \u2018classical\u2019 \nmechanics. The \u201cLaws of Motion\u201d \nwhich Newton gave as the foundation \nfor mechanics were laws of the mo- \ntion of matter. He solved, for in- \nstance, the problem of the projectile, \nand traced the path for the cannon- \nball to follow after it leaves the can- \nnon\u2019s mouth. Were it not for forces, \nNewton said, the ball would continue \nonward in a straight line forever; but \nthe earth exerts a force upon it, and \nso it bends its course along an arc \nwhich brings it back to earth again \nin the predicted spot. The roar of \nthe gun, however, is not a piece of\n\nExamining, however, the journey \nof the roar of the gun, one finds that \nit is not altogether unlike the jour- \nney of the ball. We know\u2014though \nthe knowledge was not available to \nNewton\u2014that the sound of the shot, \nhaving risen into the sky, may bend \nover and come down again to earth \nat a great distance, so that beyond a \nbroad zone of silence the gun is \naudible. Could we make visible the \ntrack of a particular section of the \nwavefront of the explosion-wave, a \nsection starting nearly vertically up- \nward, we should occasionally see it \nas an arch in the high air, like though \nnot exactly the same as the arched \ntrajectory of the cannonball. Short \nradio waves have at times the same \nhabit of rising up from the broad- \ncasting aerial, overarching wide \ntracts of land where it is vain to try \nto detect them, and descending to be \nreceived by listeners far away.\n\nNow it is said that the cannonball \ndescends because the earth pulls it \ndownward out of the straight course \nwhich it otherwise would follow into \nthe depths of space. Are the waves \nalso drawn downward by some force \nout of a straight path in which other- \nwise they would continue? It is not \nnecessary to imagine such a force; it \nis sufficient to recall that the speed of \nthe waves varies from point to point \nin the air. At times the air is warmer \nseveral miles overhead than near the \nground. When a section of a wave- \nfront ascends obliquely into this \nwarmer stratum, its higher part ad- \nvances faster than its lower part, and \nthe wavefront is tilted gradually over \n\u2014one might say that at the start it is \nleanine backward, and bends forward, \neventually standing vertical, then \nleaning forward and finally plunging\n\ndownward to earth. The radio waves \nwhich \u201cskip\u201d long distances behave \nin the same way: their wavefronts \nare gradually tilted over and down \nbecause they move faster in the upper \nair than in the lower, though the \ncause of the greater speed in the \nhigher atmosphere is not increase of \nwarmth but an abundance of free elec- \ntrons. For the same reason light \nis refracted when entering water \nobliquely from air; the part of the \nwavefront which is\u2018-first to dive is \nalso first to be slowed down, and the \nwhole wave is wheeled around until \nit moves in a direction more nearly \nperpendicular to the interface of air \nand water.\n\nDifferent reasons, therefore, are \ngiven to explain why waves on the \none hand, projectiles on the other, \ndeviate from the straight course: the \n\u201ccause\u201d\u2019 in the one case is described \nas a variation of the wave-speed from \npoint to point in the medium, in the \nother case as an external field of \nforce prevailing at every point of \nspace. Now when we come to express \nthese two explanations in mathe- \nmatical form we find (as not infre- \nquently happens in mathematical \nphysics) that the two sets of equa- \ntions display strong resemblances, \nwhich are not in the least evident to \na mere casual contemplation of the \nseemingly very different pictures \nwhich they describe. The picture of \nan external force-field and the pic- \nture of a variable wave-speed are in \nessence much more alike than they \nseem. Indeed it is permissible to say \nthat the cannonball, travelling along \nits parabolic arc from the cannon\u2019s \nmouth into the sky and back to the \nground, is riding upon an invisible \nwave which carries it along like a \nparcel; the motion of the ball is con- \ntrolled by the motion of the wave,\n\npoint. It is easy to formulate the \nway in which the speed of this imagi- \nnary wave must depend upon height \nabove the ground-level, in order to \nexplain the motion of the projectile. \nOne of the Bernoulli family did it \nmore than two centuries ago. One \nmight imagine the space around the \nearth to be filled with an aether, able \nto carry waves with a wave-speed \nprescribed in the proper way; the \nfiring of the gun would be supposed \nto create a mighty wave, which car- \nried the projectile on its back. There \nwould have to be a different aether \nfor every different choice of initial \nvelocity for the cannonball, which \nmakes the picture rather clumsy; yet \nit might present advantages great \nenough to make even this clumsiness \nendurable.\n\nThe waves of the foregoing pic- \nture are not precisely the waves \nwhich de Broglie invented and Schroe- \ndinger adapted, thus founding the \nnew \u2018\u2018undulatory mechanics\u2019\u2019 or \n\u2018\u201c\u2018wave-mechanics\u201d\u2019; but the two kinds \nare sufhciently alike for either to \nserve excellently as an introduction \nto the other. As a matter of fact, the \n\u201cde Broglie waves\u201d for the cannon- \nball trace the proper parabolic arc, \nbut move much faster than the ball, \nwhich therefore cannot be thought of \nas riding upon them. (The speed of \nthe ball is the \u201cgroup-speed\u201d of the \nde Broglie waves\u2014a concept which \nhowever is a little too intricate to be \nexplained here, although it is familiar \nin optics). The waves of de Brog- \nlie and Schroedinger are endowed, \nhowever, with another feature, which \nI will now illustrate by an example \nfrom acoustics.\n\ncertain point, or preferably set into \ncontinuous simple-harmonic vibration \nby a tuning-fork pressed against it. \nA wave of distortion starts from the \nplace where the sidewise push or \npull is applied, and travels indefinite. \nly along the wire with a constant \nspeed. Suppose now that the length \nof the wire is limited, say by cuttin \nit in two places and attaching the \nends firmly to walls. We will suppose \nfor convenience that the tuning-fork \nis pressed against the midpoint of the \nwire between the walls. The waves . \ntravel from the fork along the wire \nto the walls, are reflected, return, \nmeet new waves and combine with \nthem, are again reflected from the \nopposite walls, meet additional waves \nand coalesce with these, and so forth; \nand eventually a \u201csteady state\u201d\u2019 is at- \ntained, in which the motion of the \nwire is in general very complex. It is, \nhowever, a matter of common knowl- \nedge that if a certain condition is sat- \nisfied\u2014if the frequency of the tuning- \nfork is properly adjusted to the \nlength of the wire and to the speed \nof the waves, which last is deter- \nmined by the tension of the wire\u2014 \nthen the motion of the wire becomes \nboth simple and vigorous. If, for in- \nstance, the frequency of the fork is \nmade equal to the \u201c\u2018fundamental fre- \nquency\u201d of the wire, which can easily \nbe computed from its length and from \nthe wave-speed, then the particles of \nthe wire vibrate with an amplitude \nwhich increases steadily from both \nends to the centre. A loud musical \ntone is heard from the wire, and to \nthe eye it seems to be spread into a \nribbon of which the breadth varies \nas a sine-curve, or rather would so \nvary if the force were properly ap- \nplied. This \u201cribbon\u201d is a wave-pat- \ntern; and when the frequency of the \nfork, which is the frequency of the\n\nwaves, is adjusted to agree with any \none of the natural frequencies of the \nwire\u2014its fundamental or any over- \ntone\u2014-a wave-pattern characteristic \nof that frequency appears. This is \ncalled the phenomenon of resonance. \nA like effect would occur if the wire \nwere bent into a closed ring. Here \nresonance would take place when the \nfrequency of the waves was so ad- \njusted that each wave on completing \nthe circuit found another just start- \ning on the course.\n\nFrequency is the new feature which \nde Broglie introduced into the waves \nwhich I earlier mentioned, the \n\u201cnhase-waves.\u201d Frequency implies \nthe possibility of resonance; fre- \nquency implies diffraction, interfer- \nence, perhaps polarization, a variety \nof properties of light and sound \nwhich could be looked for and might \nbe found. Suppose, for instance, that \nthe Bernoulli waves which I described \nat first, which may be conceived to \ncarry the cannonball in its flight, were \nendowed with a special frequency and \nwave-length; might it be possible to \ngenerate them inside a box of dimen- \nsions so chosen that they would enter \ninto resonance, and the resonance be \ndetected in some peculiar way? The \nidea seems a wild one; but the atom- \nmodel of de Broglie and Schroedinger \nis developed from one like it.\n\nFor twenty years it has been ac- \ncepted that a hydrogen atom is com- \nposed of a positively-charged nucleus \nand an electron revolving around it, \nsometimes in a circular orbit. In \norder to explain the properties of \nthis atom, it is necessary to assume \nthat out of all the infinity of con- \nceivable circular orbits of various \nradii, only a few are \u201cpermitted\u201d to \nthe electron; we know which these \npermitted ones must be. Now let us \nfollow de Broglie and correlate with\n\nthe electron a stream of waves. Like \nthe Bernoulli waves they shall rush \naround and around the orbit of the \nelectron, but unlike the Bernoulli \nwaves they shall not have its speed. \nTheir wave-speed shall depend upon \ndistance from the nucleus of the atom, \nas the speed of the Bernoulli waves \nassociated with the cannonball de- \npended on height above the ground, \nalthough not according to the same \nlaw. The choice of the law of de- \npendence of the wave-speed upon the \ndistance is the chief step in the de- \nsign of the atom-model; one might \nenvisage it as the invention of a spe- \ncial kind of aether filling the space in \nand around the hydrogen atom, de- \nsigned to carry waves with a wave- \nspeed varying from point to point ac- \ncording to a law specially chosen. \nFinally, these electron-waves shall \nhave a special frequency, proportional \nto the energy of the system composed \nof electron and nucleus, which is the \natom itself.\n\nWhen all these assumptions are \nmade, and the law of the wave-speed \nproperly chosen, it turns out that the \npermitted circular orbits are precisely \nthose on which the electron-waves so \noverlap one another that they form \na stationary wave-pattern, and reso- \nnance occurs. Were we to take a cir- \ncular orbit at random we should prob- \nably find that the overlapping was \nsuch as to produce a very confused \nand intricate vibration, like the mo- \ntion of a piano-wire forced to vibrate \nby a fork not tuned to any of its nat- \nural frequencies. But for the per- \nmitted circular orbits the wave-mo- \ntion in the imaginary aether pervad- \ning the hydrogen atom is of the sim- \nple character of the motion of the \nwire stimulated by a fork having one \nof its own natural frequencies.\n\nthe atom becomes the same as the \ngreat problem of acoustics: to find \nthe natural frequencies of a body \ncapable of vibration. The problem \nof the oscillating electron becomes \nlike that of the piano-wire; the prob- \nlem of the revolving molecule like \nthat of the telephone-diaphragm; the \nproblem of the hydrogen atom like \nthat of the ball of fluid. They are \nmore intricate than their acoustic \nanalogues, for in the wire or the \nsphere of fluid the wave-speed of the \nsound-waves is everywhere the same, \nwhile in the \u201cimaginary string\u2019\u2019 used \nas symbol for the oscillating electron \nand in the \u201cimaginary fluid\u2019\u2019 used as \nmodel for the hydrogen atom, the \nwave-speed must be supposed to vary \nfrom point to point in a peculiar and \ncharacteristic way\u2014the selection of \nthis way being, as I said before, the \nchief step in the building of the \ntheory. En revanche, this variability \nof wave-speed is precisely what makes \nthe problems soluble. When we are \ndealing with an actual wire, mem- \nbrane or fluid, we get resonance only \nwhen they are bounded\u2014when the \nwire is cut off and clamped at the \nends, the membrane cut into a circuit \nand clamped around the edge, the \nfluid contained within a rigid hollow \nsphere. But the imaginary strings \nand membranes and fluids of the \natomic theory have resonances even \nthough they continue to infinity; and \nthese resonances correspond to the \nStationary States of the atoms, from \nwhich most of their qualities are de- \nrived.\n\nOne more example: if an electron \nis attended by waves moving along \nthe same path, and a beam of elec- \ntrons moving along parallel lines is \ntherefore escorted by a beam of \nparallel waves, may not these waves\n\nbe diffracted by a grating as light. \nwaves are? and may they not carry \nelectrons into the diffraction-fringes \nas Bernoulli\u2019s waves were thought to \ncarry the cannonball (though here \nagain the speeds are not equal), so \nthat in effect the electrons will be dif- \nfracted? On computing the wave- \nlengths which, in order to make the \nmodel for the hydrogen atom valid, \nmust be assigned to electrons of given \nspeeds, it is found that those asso- \nciated with electrons of the speeds \nordinarily occurring in discharge- \ntubes are of the magnitude of the \nwavelengths of X-rays. We must, \ntherefore, use natural crystal grat- \nings such as are used with X-rays. \nCarrying this idea (which first occur- \nred to a young German student named \nElsasser) into practice, Davisson and \nGermer discovered that electrons are \nin fact diffracted as are waves of the \ncalculated wavelength.\n\nWe are therefore confronted with \nthe amazing fact that the phase- \nwaves, conceived purely as a mathe- \nmatical artifice, seemingly differing \nfrom matter and electricity as the \nvisions of the mind differ from the \ncommon-sense actuality of the solid \nphysical world, have nevertheless ac- \nquired a reality ranking with the real- \nity of things visible and things tan- \ngible. Indeed theirs may be even the \nstronger; for when the equations de- \nscribing the resonances of the hydro- \ngen atom are actually derived and \nwritten down, the electron and the \nnucleus are no longer obvious in them, \nand nothing remains but the wave- \npattern. It is as though at a concert \nthe air which carries the music and \nthe instruments of the orchestra and \nthe players themselves should be pro- \nclaimed unreal, and nothing remain \nbut the music.\n\ncable varies in temperature \nover a wide range. Changes result \nin resistance of the conductors and, \nto a lesser degree, in dielectric \ncapacity, with accompanying changes \nin attenuation of the speech currents. \nFor the cables in the exchange plant, \nand for the shorter toll cables, these \nchanges may be disregarded, but for \nthe long toll cables they make neces- \nsary frequent readjustment of re- \npeater gains to keep the net loss of \neach circuit within the rather close \nlimits specified. The readjustment is \neffected by an automatic regulator, \nwhich derives its information as to \nconditions in the cable from the \nchange in resistance of a metallic cir-\n\ncuit. The regulator is essentially an \nautomatic Wheatstone bridge whose \nrebalancing mechanism controls also \nthe telephone repeaters.\n\nIn applying the regulating system \nthe cable is divided into sections usu- \nally from one to two hundred miles \nin length. Generally, all circuits pass- \ning through each section are auto- \nmatically compensated by one \u2018\u2018master \nregulator\u201d placed at a repeater sta- \ntion near the center of the section.\n\nTwo cable-pairs, extending in op- \nposite directions, are short-circuited \nat their distant ends, and resistances \nare connected in series with the short- \ner pair. Thus approximately equal- \nized, the pairs are connected in paral- \nlel to form one arm of the bridge.\n\nFig. 1\u2014Why automatic regulation is necessary\u2014daily variations of temperature\n\nquently the resistance, of the pilot \nwire changes, the Wheatstone bridge \nis thrown out of balance, and the gal- \nvanometer deflects. This galvanom-\n\neter is part of a Leeds and North- \nrup motor-driven recorder. At inter- \nvals of a few seconds the driving\n\ntion; control keys; test jacks, dial-switches, \nand miscellaneous apparatus\n\nmotor, through a cross-bar, lifts the \npointer of the galvanometer so that, \nif deflected, the pointer in turn raises \none or the other of two levers. If not\n\ndeflected, that is, when the bridge js \nin balance, the pointer passes between \nthe ends of the two levers. Through \nan ingenious mechanical system, the \nraising of either lever causes a shaft \nto be rotated through a small angle, \nwhose amount depends on the gal- \nvanometer deflection and whose di- \nrection depends on which lever is \nraised. \u2018The shaft in turn operates \na slide-wire resistance to restore the \nbalance of the bridge, and also moves \nthe arm of a dial switch over a series \nof contact-points connected to a group \nof relays, which in turn readjust the \ngain of the associated repeaters. In \nthe case of 19 gauge H-174* circuits \nwhere the changes in transmission are \nsubstantially the same at different \nfrequencies throughout the voice \nrange, simple resistance potentio- \nmeters can be used to regulate the \ngains of the repeaters. In the case \nof 19 gauge H-44-25** circuits, how- \never, difficulty was encountered in the \nfact that changes in transmission at \nlow frequencies produced by a given \nresistance change are less than those \nat high frequencies. More complicat- \ned networks were therefore required \nto vary the gain differently at differ- \nent frequencies.\n\nThere are ten steps on each side of \nthe \u2018\u2018zero\u201d\u2019 position of the dial switch; \nthat one which is in contact with the \narm is indicated by a lighted lamp on \nthe control panel. The central \u2018\u201c\u2018zero\u201d \nstep is symbolized by a green lamp; \nthe extremities by red lamps and the \nremaining steps by white lamps. \nShould the regulator stop with the \narm bridging two contacts, the relays \nchange the gain by one-half the \namount of a full step. A permanent \nrecord is made by a pen travelling\n\n* Sometimes called \u201cmedium heavy loaded.\u201d \n** Sometimes called \u201cextra light loaded.\u201d\n\nFig. 3\u2014A close-up of the Leeds and Northrup \nrecorder which is the heart of the system\n\nacross a chart which in turn is ad- \nvanced by the motor at one and one- \nhalf inches an hour.\n\nTrouble on the pilot wire, such as \nan open or short circuit, looks to the \nregulator like a sudden large change \nin the attenuation of the cable. Doing \nits best to meet the emergency, the \nregulator would change the repeater \ngains quickly through a considerable \nrange, perhaps even to the limit of \nits ability, with disastrous results on \ntelephone transmission. By an in- \ngenious mechanical modification of \nthe mechanism, the galvanometer is \nprevented from operating the bridge \nin case of a large deflection, but in- \nstead intermittently lights a lamp and \nsounds an alarm buzzer. Still greater \nunbalance of the bridge might send \nso much current through the galvano- \nmeter as to endanger its winding. A \nrelay with differential windings in \ntwo of the bridge arms is therefore\n\nprovided; its operation short- \ncircuits the galvanometer and \noperates the lamp and buzzer \nsignals continuously. In both \nthese cases, disappearance of \nthe trouble restores the appa- \nratus to its normal operation.\n\nUnder extreme conditions \nof temperature, or for other \nreasons, the resistance of the \npilot wire may change beyond \nthe ability of the regulator to \nfollow it. On the tenth step \nof the dial switch, a contact \nis closed which stops the driv- \ning motor, disconnects the bat- \ntery supply from the bridge \nand operates the buzzer. As \nsoon as circuit readjustments \nhave been made, the attend- \nant sets the dial switch on the \nninth step and places the ap- \nparatus again in operation. \nShould the trouble persist, the regu- \nlator would advance to the tenth step \nand the alarm would again be given.\n\nan ungrounded battery. Since the cur- \nrent drain is only about twenty-five \nmilliamperes, either dry or storage \ncells may be used. Failure of this cur- \nrent-supply lights a lamp and operates \nthe buzzer. Current for the motor is \ndrawn from the 24-volt battery. \nIdeally, the pilot-wire should be \nloaded in the same way as the tele- \nphone circuits which it is to control.\n\ncable at the time of test will be at a \ndifferent temperature, so the slide \nwire must be set at some other point. \nKnowing the theoretical resistance of \nthe pilot wire at fifty-five degrees, \nand adding to this value the nominal \nresistance of the composite coils, if \nany, it is possible by comparison with \nthe measured resistance of the cable \nto determine where the slide wire\n\nFig. 5\u2014How regulators and repeaters are spaced on the New York-Chicago Cable\n\nThe regulator has been designed for \nuse with 19 gauge H-44-25 pilot \nwires, since these circuits require the \nmost accurate regulation, and field \ntrials have shown that other types \nof circuits are then controlled with \nsuficient accuracy. Non-loaded 19 \ngauge pairs may also be used. Spare \npairs may be used as pilot wires, or \na metallic (ungrounded) circuit may \nbe obtained by use of two composited \ntelegraph circuits connected at their \ndistant end. To balance out the com- \nposite coils, whose temperature- \nchange is that of central-office air, a \nnumber of resistances wound with \ncopper wire are included in the regu- \nlator circuit.\n\nSetting a transmission regulator in \noperation requires, beside the installa- \ntion and mechanical test of the ap- \nparatus, some careful tests of the \nlines involved. The zero position of \nthe slide-wire corresponds to an aver- \nage temperature of the pilot wire of \nfifty-five degrees F. In general, the\n\nThe regulator motor having been \nstarted, the battery circuit is closed. \nIf necessary, the bridge is balanced \nby adjusting resistances T and S. Final- \nly, transmission measurements with \n1000-cycle current are made on one \nor more of the 19 gauge H-44-25 cir- \ncuits to be regulated. With individual \nrepeaters adjusted to give their speci- \nfied gains, the overall transmission \nequivalent of the circuit must be that \nspecified to within 0.2 db; otherwise \nthe resistances T and S are varied to \nsecure the required agreement.\n\nRegulators are installed along all \nlong-distance cables; for example, be- \ntween New York and Chicago there \nare regulators at Allentown, Bedford, \nNew Castle, Bogart, West Unity and \nSouth Bend. In the area covered by \nthe Long Lines cable plant, about 4 \nof these devices are constantly on \nguard to help maintain satisfactory \ntelephone transmission.\n\nmetal to elements of its en- \nvironment such as water, oxygen, \ncarbon dioxide, gaseous sulfur com- \npounds, acids, alkalies and salts. All \nmetals corrode but in many cases the \nproduct of corrosion is a protective \nfilm which retards or prevents con- \ntinuation of the action. A film is pro- \ntective only if it is adherent to the \nmetal, continuous, and insoluble in the \nsurrounding medium. Films which \nare not continuous may modify or \neven accelerate the rate of corrosion. \nLead forms protective films under \ncertain conditions. For instance when \nexposed to the atmosphere it tar- \nnishes and becomes inert owing to the \nformation of an oxide film which in- \ncreases in thickness during a period \nof about ten days. In the presence \nof moisture and carbon dioxide, lead \nbecomes coated with a film of lead \ncarbonate. While this film consider- \nably reduces the rate of corrosion it \ndoes not prevent corrosion. The car- \nbonate film forms when a_ small \namount of carbon dioxide is present, \nand the rate of corrosion is then gov- \nerned by the rate at which oxygen \npermeates or diffuses through the \nfilm. A third kind of film is formed \non lead by small amounts of soluble \nsilicates. The existence of this film \non the surface of lead prevents its \ncorrosion in slightly acid, neutral or \nslightly alkaline solutions. Under \nhigh conditions of alkalinity, how-\n\nCorrosion, according to the mod- \nern viewpoint, is an electrolytic pro- \ncess. It depends upon the existence \nof small potential differences between \ndifferent points on the metal surface, \nmaking tiny batteries, and upon the \ncharacter of the reactions which oc- \ncur at the electrodes as a result of the \nflow of small electrical currents. The \nfundamental reaction for the corro- \nsion of lead may be represented by \nthe equation:\n\nThis is merely a statement of the \nfact that metallic lead dissolves at \nanodic areas and that hydrogen \n\u201cplates out\u201d as atoms on cathodic \nareas. There is formed in this way \na film of atomic hydrogen which must \nbe removed if corrosion is to pro- \nceed, the rate of corrosion being de- \npendent upon the rate of removal. \nThe process of hydrogen removal con- \nsists either in evolution as hydrogen \ngas or in oxidation. From a lead sur- \nface there is little tendency to evolve \nhydrogen even in highly acidic solu- \ntions. If however there are present \nother metals or substances such as \ncopper, iron, carbon, etc. which \nmore readily discharge hydrogen as \na gas, corrosion is accelerated. More \ncommonly hydrogen is removed from \nthese cathodic areas by oxygen or \noxidizing substances with the forma- \ntion of water, and for this reason the\n\nrate of corrosion is determined gen- \nerally by the rate at which oxygen \nreaches the lead surface undergoing \ncorrosion.\n\nCertain secondary reactions occur- \nring at the electrodes often influence \ncorrosion. For instance, the products \nof electrolysis may offer resistance to \nthe flow of current or may by film \nformation on anode areas reduce or \nprevent dissolution of the metal. In \nthe latter case current flow may con- \ntinue, the anode process consisting of \ngas evolution (usually oxygen). On \nthe other hand, electrolysis may re- \nsult in the formation of a substance \nwhich is corrosive to the metal. For \nexample, alkali and lime salts are con- \nverted into caustic alkali and free \nlime at the cathode and these attack \nlead.\n\nA knowledge of the relationship \nof anode and cathode potentials to \ncurrent density, make it possible to \npredict under given conditions whether \nor not a metal is likely to corrode. \nThe influence of surrounding condi- \ntions may be illustrated by the be- \nhavior of lead in solutions of acetic \nacid, in which it dissolves readily, and \nin solutions of sulfuric acid, in which \nit is but slightly soluble. In the \nformer case small potentials yield \nhigh current densities (correspond- \ning to high rates of corrosion) while \nin the latter case relatively high po- \ntentials are required to maintain even \nmoderate current densities. More- \nover in this latter case formation of \na lead sulfate film on the anode pre- \nvents corrosion at current densities \ncorresponding to appreciable rates of \ncorrosion.\n\nThe sheathing used for telephone \ncables consists of lead or more gener- \nally an alloy of lead and one per cent \nof antimony. While the alloy is some- \nwhat more resistant to corrosion than\n\nis lead, the corrosion reactions in the \ntwo cases are similar and for the pur- \nposes of this paper may be considered \nas identical.\n\nThere are from a practical stand- \npoint four kinds of corrosion to \nwhich cable sheath is subject. These \nare characterized as (1) soil or soil \nwater, (2) acetic acid or white lead, \n(3) stray current anodic or ordinary \nelectrolytic, and (4) stray current \ncathodic or negative corrosion. \nFundamentally, these types of corro- \nsion are all electrolytic and involve \nthe same mechanism, the distinction \nbetween them having to do mainly \nwith the source of potentials respon- \nsible for corrosion. In the first two \nclasses the potentials are galvanic in \ntype and arise from a lack of homo- \ngeneity either in the metal or in the \nmedium adjacent to it; the third type, \nstray current anodic, is due to po- \ntentials picked up from improperly \ninsulated power circuits, i.e., street \nrailway, light and power lines; and \nthe last class, stray current cathodic, \nis a combination of the galvanic and \nstray current types or in the ordinary \nsense a chemical reaction between \nthe cable sheath and caustic alkali or \nfree lime generated by electrolysis \nwhen the sheath is cathodic or nega- \ntive to earth.\n\nWhether a metal is corroded uni- \nformly or pitted locally depends upon \nthe area, position and number of \nanodes in a small area. If the anodes \nare large in area or numerous the at- \ntack tends to be uniform in appear- \nance. This is illustrated in the cor- \nrosion of lead in acetic acid. If the \nanodes are widely separated or of \nsmall area with respect to the area of \nthe lead surface, then the attack is \nof a pitting nature. In the case of \ncable sheath pitting action is the more \ndangerous since it produces perfora-\n\nWhile soil corrosion is not a seri- \nous problem in cable maintenance, oc- \ncasional cases are reported. It is due \ngenerally to exposure of different \nparts of the sheath to different con- \ncentrations of oxygen, salts, acids or \nalkalies. Pitting action initiated by \nstray current electrolysis may con- \ntinue after elimination of the positive \npotential conditions owing to the fact \nthat the pits, being less accessible to \noxygen, become the anodes of gal- \nvanic cells which eventually cause per- \nforation of the sheath. This is doubt- \nless the explanation of many cases of \n\u201cold action.\u201d Soil corrosion of cable \nsheath sometimes occurs in regions \nwhere there is an abrupt change in the \ncharacter of the soil. This might \nhave been predicted from laboratory \nstudies which have shown potential \ndifferences of more than 0.3 volt be- \ntween lead electrodes immersed in \ntwo different soil waters.\n\nIt is not uncommon for the sheath- \ning of underground telephone cables \nto carry as much as 100 amperes of \ncurrent derived from leaky power \ncircuits. Under these conditions, on \na full size cable there is an IR drop \nalong the cable of about 0.1 volt per \nhundred feet which may create a \nlarge-scale galvanic cell provided \nthere is a low-resistance path through \nthe earth between two points on the \ncable at some distance apart. Pre- \ncaution against corrosion in such a \ncase consists in bonding the two \npoints with a metallic conductor, \nthereby completing a metallic circuit \nand preventing current flowing from \nthe sheath to earth.\n\nThe corrosion of cable sheath by \nacetic acid vapor occurs sometimes \nin creosoted wood conduit. It de- \npends upon the maintenance of a suit- \nably high concentration of hydrogen\n\nions in the presence of oxygen and \ncarbon dioxide and the formation of \na non-protective film of basic lead \ncarbonate. Early cases of cable \nsheath corrosion in creosoted wood \nducts resulted from the use of wood \ncreosote containing acetic acid. Re- \ncent extensive corrosion of this type \non the Pacific Coast has been found \nto be due to the use of Douglas fir \nducts incompletely creosoted. This \nwood is markedly more acidic than \nsouthern pine, which is used largely in \nother parts of the country for wood \nconduit, and moreover fir conduit is \ndificult to penetrate with creosote. \nThe corrosive attack appears to have \nbeen stopped by neutralizing the acid \nwith ammonia gas forced through \nthe ducts in a mixture with air. \nThe most dangerous and most com- \nmon kind of corrosion to which cables \nare subject is the ordinary electrolytic \nor stray current corrosion which may \noccur when the sheath becomes posi- \ntive to earth. Here moisture condi- \ntions and soluble salts are of consid- \nerable importance since these affect \nthe earth resistance and the anodic \nareas and hence influence the current \ndensities at the point where current \nflows from the sheath to earth. Pre- \ncautions against this type of corro- \nsion consist in periodic electrolysis \nsurveys, the bonding of cables to the \nnegative returns of power circuits, \nand sometimes the shielding of the \ncable network against stray currents. \nFinally there is cathodic or nega- \ntive corrosion to which reference has \nalready been made. This is charac- \nterized by the formation of red crys- \ntalline lead monoxide which crystal- \nlizes from saturated solutions of lime \nor alkali plumbites. Lead is readily \ndissolved by strong alkalies with the \nformation of plumbites which are \nstable only in solution. The alkalies\n\nresponsible for corrosion may be pro- \nduced by the electrolysis of soluble \nlime or alkali salts. Sodium  chlor- \nide or ordinary table salt is used \nsometimes to thaw out street car \nswitches in winter and occasionally \nfinds its way into cable ducts where it \nis converted into caustic soda at the \nsurface of the cable sheath as a result \nof current flow from the earth to the \nsheath. This type of corrosion also \noccurs in newly manufactured con- \ncrete conduits owing to the presence \nof free lime in the concrete.\n\nbefall a lead sheath, that under. \nground cables are a rather vulnerable \npoint in the Bell System plant. As a \nmatter of fact, the losses of cable due \nto corrosion are relatively small, \nthanks largely to care in the selection \nof conduit material, and to the skill \nand vigilance of the Plant Depart. \nments\u2019 forces. But when it is recalled \nthat the Bell System has in service \nsome 55,000 miles of underground \ncable, containing nearly thirty-eight \nmillion miles of wire, it is evident \nthat no pains are too great to find the \ncauses and so prevent corrosion of \nthe protecting lead sheath.\n\ncreasing extent taken the place \nof steel, brass, bronze and other more \ncommon metals in many fields of in- \ndustry. Used in the form of thin \nsheets, one group of the alloys has \nbrought marked advantages to cer- \ntain of the newer developments in \ntelephone transmitters and receivers. \nThere, used as diaphragms, they con- \ntribute prominently to high operating \nefficiency.\n\nThis group presents such a range \nof mechanical properties that those \ndesired for almost any use can be se- \ncured by choice of an alloy of suitable \ncomposition, coupled with appropri- \nate metallurgical treatment. Since\n\nstrength, elongation and ductility \ncan be controlled within wide limits, \nthe necessary values in these respects \ncan be secured in any cases but the \nmost exacting. The greatest advan- \ntage, of course, is the low density, \nabout a third that of steel. Another \nmajor advantage with most of the \nalloys, not otherwise found in non- \nferrous metals, is that parts can be \nincreased in strength by heat treat- \nment when they are in final manufac- \ntured form. Less important, manu- \nfacturing operations are facilitated \nby the low modulus of elasticity. It \nis about a third that of steel, so that \nthe force needed for bending and \nother forming operations is about a \nthird that which would be required\n\nFormed diaphragm for a WE-555 Receiver, made of 17ST Alloy 0.002 inch thick. \nIt is used principally with sound-picture apparatus\n\nwith steel parts of the same strength. \nThe alloys are not immune to corro- \nsion, but are much less subject to it \nthan many of the materials which \nmight be used in their stead. A quite \ndifferent consideration makes them \navailable for use in large quantity: \nthe cost, although commonly consid- \nered high, is about the same as that \nof steel, when taken on the basis of \nvolume rather than of weight. These \nfacts, therefore, mean that there are \navailable materials higher in ultimate \ntensile strength and in elastic limit \nthan mild steel, with approximately \none third the weight for a given part, \nand costing about the same per part.\n\nMany different compositions are \navailable, but discussion will be con- \nfined to the two alloys commonly used \nin thin sheet form for diaphragms. \nThese are the 17S or duralumin alloy \nand the 3S alloy. They are known \nby these numbers in accordance with \nthe code adopted by the principal sup-\n\nplier, wherein each alloy is identified \nby a number followed by the letter S. \nThat code reveals the chemical com- \nposition only; the temper and metal- \nlurgical treatment are given by a sup- \nplementary code consisting of one or \nmore letters. These symbols do not \ntell the complete history, but only the \nprocesses revealing the current condi- \ntion. Annealing is represented by O, \nand cold-working by H. W repre-\n\nquenching only, T those processes fol- \nlowed by aging, and RT the more \ncommon sequence of heat-treating \nand quenching followed successively \nby aging and cold-working. The ma- \nterial for a particular diaphragm may \ntherefore be 17SO at one stage of \nmanufacture, after it has been an- \nnealed to facilitate forming, and at \nthe end, when it has been hardened \nfor use, be 17ST.\n\n0.5% each of manganese and mag- \nnesium, and the remainder commer- \ncially pure aluminum. Although the \ntotal percentage of alloying ingredi- \nents is small, this material, whose \ndensity is very little higher \nthan that of pure aluminum, \ncan be brought by means of a \ntempering process to a tensile \nstrength of 55,000 pounds per \nsquare inch with a substantial \ndegree of ductility. The ten- \nsile strength may be raised by \nsubsequent cold-working to at \nleast 70,000 pounds per square \ninch; at the same time the \nproperty of ductility is almost \nentirely lost. Tempering may \nmay be produced in various \nways, but the usual method consists \nof heating the material to approxi- \nmately 510\u00b0 C., quenching it in water \nand allowing it to age at room tem- \nperature. Heating at this tempera- \nture for seven to ten minutes is usu- \nally sufficient for thin stock. The \nhardening proceeds very rapidly at \nfirst and then at a gradually diminish- \ning rate until the maximum is reached \nat the end of five days. After heating \nand quenching, hardening may be car- \nried out more quickly by maintaining\n\nReproducer for an Orthophonic talking machine, \nof 3SH Alloy; the spider is of hard commercially\n\nthe parts at 100\u00b0C. for a period of \nforty hours, a process known as arti- \nficial aging. When a part is to be \nmade from heat-treated duralumin, \nthere must be no delay between heat-\n\ntreating and forming else the material \nwill become so hard that it may fail \nin the course of the forming opera- \ntions.\n\nFor use in several pieces of appa- \nratus it has been necessary to raise \nthe ultimate tensile strength of the \n175 alloy to 75,000 pounds per \nsquare inch, and in special cases to \nas much as 85,000-95,000 pounds. \nThis was accomplished by reducing \nthe cross-sectional area about 90% \nthrough operations known as cold- \nworking \u2014 swaging, drawing, \nrolling, and other cold pro- \ncesses. Material so handled, \ncharacterized by high tensile \nstrength and elastic limit and \nby very small elongation, is \nspecified as 17SRT. For other \noperations, annealed material \nis reduced 80% in thickness by \nsimilar cold operations. The \nmaterial thus produced, 17SH, \nhas an _ ultimate _ tensile \nstrength of 45,000 pounds per \nsquare inch and_ negligibly \nsmall elongation.\n\nThe other alloy extensively used, \n38, consists of commercially pure \naluminum to which has been added \napproximately 114% manganese. In \nthe annealed state it is about 25% \nstronger than annealed pure alu- \nminum, and of slightly lower elonga- \ntion. By cold-working its tensile \nstrength may be brought up to about \nthree times that of annealed alu- \nminum, with a high ratio of elastic \nlimit to ultimate strength. This alloy \nis therefore an extremely useful ma- \nterial in cases where severe forming \nis required. It does not, however, re- \nspond to heat treatment.\n\nBoth of these alloys may be ren- \ndered soft and ductile by annealing \nat approximately 343\u00b0C. at any time \nin the manufacture of parts if the \nhardness was caused by manufactur- \ning operations carried out on formerly \nannealed material. Where the \nhardened condition was pro- \nduced by heat treatment, as \nmay be the case with duralu- \nmin, annealing requires heat- \ning for a longer time and at a \nhigher temperature, followed \nby very slow cooling. Some of \nthe deeply formed diaphragms \nare annealed a number of \ntimes in the course of produc- \ntion to remove the hardness \nresulting from the stresses of \nthe forming operations.\n\nThe alloys are received at \nthe Hawthorne plant of the \nWestern Electric company \nfrom the manufacturer in \ncoils 0.020 inch thick. There \nthey are rolled into ribbons of the \nthickness wanted, and generally just \nwide enough to make the blank for a \nsingle diaphragm. In all cases the \nmaterial is buffed as it comes from \nthe rolls. By this operation the re- \nsistance to corrosion is raised ma-\n\n387-W Transmitter, for broadcasting. \nphragm, which passes between the two carbon but- \ntons, is a sheet of 17SRT Alloy, 0.0017 inch thick, \nstretched by the clamping ring\n\nterially by the highly polished surface \nobtained. The thinnest material in \nuse at present is a duralumin alloy for \nspacing washers of condenser trans- \nmitters; its thickness is only 0.00075 \ninch, about a sixth that of the paper \non which the REcorD is printed. The \nmost common thickness is 0.003 inch, \nand the thickest material commonly \nused is 0.005 inch.\n\nAluminum and its alloys are gen- \nerally considered resistant to corro- \nsion, and for thicknesses greater than \n0.020 inch that is the case. In sev- \neral pieces of apparatus the alloys are \nused in sheets 0.001 inch to 0.003 inch \nthick; obviously a very small degree \nof corrosion would be enough to re- \nduce the cross-section of those parts \nby an extent sufficient to cause failure. \nCorrosion has therefore been studied \nintensively on pieces of apparatus\n\nand on the sheet stock as well. Since \ninterest was mainly in objects 0.0017 \ninch and 0.003 inch thick, the tests \nwere carried out on tensile samples \ncut from materials of these thick- \nnesses. Several methods of corrod- \ning the alloys artificially have been\n\ntried, principally spraying with a salt \nsolution, outdoor exposure tests on \nthe roof and intermittent immersion \nin corroding solutions.\n\nOur most dependable corrosion \ntest makes use of a device resembling, \non a small scale, the Ferris wheels of\n\namusement parks. The device was \ndeveloped by the Chemical Research \nDepartment. The tensile samples \nafter measuring and cleaning are \nclamped into position at the circum- \nference of the wheel, which is then \nrotated slowly and evenly for eigh- \nteen hours. They dip successively \ninto the corroding solution at the \nbottom of each revolution and then \ndrain and dry while completing the \ncycle; the corrosion occurs during \ndrying. At the end of the corroding \nprocess the samples are removed and \nare broken in a tensile testing ma- \nchine. Uncorroded samples from the \nsame stock are also broken, for com- \nparison. Protective finishes are ap-\n\npraised with the same piece of \ntesting apparatus; coated and un- \nprotected samples are exposed to- \ngether, and then the ultimate tensile \nstrengths of both groups are deter- \nmined.\n\nAs rated by this machine chem- \nically pure aluminum was highest in \nresistance to corrosion, and commer- \ncially pure aluminum second. Then \nfollowed a number of alloys, depend- \ning for their order on the treatment \nthey had undergone as well as on \ntheir composition. The corrosion re- \nsistance is also determined in part by \nthe surface. As might be expected, \nwhen it is smooth and highly polished \nit affords a considerably smaller op- \nportunity for corrosion to start than \ndoes the somewhat rough gray sur- \nface commonly seen on aluminum \ncooking utensils.\n\nChoice of the alloy for a particu- \nlar use cannot be made entirely on \nthe basis of corrosion tests, of course; \nmechanical requirements at times are \nsuch that a particular alloy and tem- \nper must be used regardless of its low \nresistance to corrosion. For such \ncases the problem of increasing the \nresistance to corrosion, in order to \ninsure that the part will last through \nthe life of the apparatus, has been \nstudied in the Chemical Laboratories. \nEither one or two coats of varnish, \nsprayed onto a finished part, affords \nsatisfactory protection but adds ap- \npreciably to the weight. Thin Bake- \nlite varnish is also satisfactory for \nalloys whose mechanical properties \nare not changed by the baking oper- \nation.\n\nAn interesting effort to combine \nmoderate mechanical strength with \nmaximum resistance to corrosion is \nthe Alclad alloy, developed and manu- \nfactured by the Aluminum Company \nof America. Sheets are made by roll-\n\ning a slab made of duralumin or one \nof the other alloys in the center and \naluminum on the outside. Thick- \nnesses of the layers are chosen to \ngive a finished sheet of duralumin \n0.002 inch thick covered on each side \nwith 0.0005 inch of pure aluminum. \nUsers can secure Alclad 35, Alclad \n17S, or sheets made with other alloys \nin the center, to suit their needs. Par- \nticular care is needed, however, in \nheat-treating Alclad sheets, since \nthere is a tendency, while the ma- \nterial is hot, for the copper to diffuse \nfrom the center layer into the alumi- \nnum and so to reduce the corrosion \nresistance of the protective layers. \nThe material presents interesting pos- \nsibilities where there is danger of \ncorrosion and where the necessity for \nlightness makes it inadvisable to in- \ncrease the weight with a protective \nvarnish.\n\nAlthough aluminum and its alloys \nare used elsewhere in the telephone \nplant, these high strength alloys in \nsheet form find their most important \nuse in the solution of a basic prob- \nlem, the transfer of vibrations be- \ntween the air and the vibrating mem-\n\nber of a transmitter or receiver. For \nthe most efficient and faithful trans- \nfer of energy the difference in im- \npedance of the two media must be re- \nduced to a minimum and for that end \nthe moment of inertia of the dia- \nphragm, and hence its mass, must be \nmade as low as possible. In high- \nquality transmitters, where the flat- \nness of the frequency-response curve \ncomes in part from the fact that the \nrange of resonance of the diaphragms \nis brought near the upper end of the \nvoice frequencies by the tension under \nwhich the diaphragms are held, the \nmaterial must be of minimum density, \nyet strong enough to withstand \nstretching to the desired tension. \nLikewise in the diaphragms of dy- \nnamic receivers stiffness is as much \nan essential as light weight, and \nstrength is needed on account of the \nhandling during assembly. For these. \nuses, and for many others, the sheet \nalloys by their combination of high \nstrength and light weight, and by \ntheir ability to withstand difficult \nmanufacturing processes, make pos- \nsible results that could otherwise not \nbe attained.\n\nmunication\u201d. This medal is awarded annually by a committee \nof twenty-four members of the A. I. E. E., for \u201cmeritorious \nachievement in electrical science, electrical engineering, or the \nelectrical arts\u201d. Among previous Edison Medallists are Alex- \nander Graham Bell, John J. Carty and Michael I. Pupin.\n\nWestern Electric Company beginning about 1895, I \nthink of a period of intense and continuous activity. We \nwere going through a revolution. We had been accustomed \nto growth but the kind of growth in which building fac- \ntories becomes a routine was just beginning. We were \nfinding that our methods in the shop and office had to be \nreconstructed to adjust them to the faster pace and that our \norganization had to be modified and expanded. But we \nhad men.\n\nNow, thirty years after, the specific difficulties do not \nlinger in my mind. The picture I see is of a group of eager, \nresourceful, loyal, young men grappling the difficulties \nwith the enthusiasm of a foot-ball squad going into the \nfield. One of those young men, and not the least, was our \nfriend Mr. Clifford. When you read the story of his prog- \nress in the Company, it does not mean stepping from one \nestablished, well organized position on to other better and \nsimilarly organized positions. It meant reconstructing a \ndepartment and getting it to run smoothly and then repeat- \ning the process in a larger field. So, when during the \nwar, the Laboratory found itself, almost over night, a \ncommercial manufacturing concern, making in large quan- \ntities the apparatus needed by the Army and Navy, but \nwithout the necessary commercial and manufacturing or- \nganization or methods, it was Mr. Clifford who went into \nthe Laboratory organization to supply what was lacking. \nThere was no question then as to whether or not it was a \npromotion. There was his kind of a job to be done and he \ndid it and he made it a promotion. We shall all remember \nhim with great affection and respect. I shall also remember \nhim as one of those who helped over the hard places.\n\nOccasionally it happens that a sub- \nscriber unintentionally sends what is \nknown in dial parlance as a prelimi- \nnary pulse before he dials the number \nof the subscriber whom he wishes to \ncall. This preliminary pulse is gen- \nerally caused by an accidental jiggling \nof the switchhook before the number \nis dialed. When the switchhook is \nmomentarily pushed down after hay- \ning been released, it opens and then \nshort-circuits the two wires of the line \nas though the digit 1 had been dialed. \nThis causes the selector in the cen- \ntral office to take one vertical step \nand prepare the circuits for the next \ndigit.\n\nIn order to show how a step-by- \nstep dial office is arranged to take \ncare of this condition, it will be of \naid to review briefly the progress of a \ncorrectly dialed call through such an \nofice. Figure 1 gives a schematic \nshowing the manner in which the or- \ndinary line finders, selectors, and con- \nnectors are arranged in a four-digit \nstep-by-step office. When a calling \nsubscriber takes the receiver from its \nhook, a line finder immediately steps \nvertically until it reaches the level on \nwhich the terminals of the calling \nsubscriber\u2019s line appear, and then \nhunts across the level until these ter- \nminals are found. The first selector, \nwhich is permanently connected to \nthe line finder, then sends back a tone \nto the subscriber as a signal for him \nto dial. In returning to normal on the \nfirst digit the dial opens and closes \nthe line loop, causing the first selector \nto step vertically on2 step each time \nthe line is opened. As a concrete \nexample let it be assumed that the \nnumber 9284 is dialed. When the \nnumber g is dialed, the first selector\n\ntakes nine vertical steps and then \nhunts across the ninth level for an \n\u2018dle trunk leading to a second selector. \nOn the next digit, 2, the second se- \nlector takes two vertical steps and \nhunts across the second level for an\n\nthe regular second selectors insofar \nas circuit operation is concerned. \nHowever, the trunks from all levels \nof the special second selectors except \nthe first are multipled to the trunks \nof corresponding levels of the first\n\nidle trunk to a connector. This con- \nnector has been chosen by the first \ntwo digits as the one having the ter- \nminals of the called subscriber\u2019s line \nappearing among the terminals acces- \nsible to it. The calling subscriber next \ndials 8. The connector steps ver- \ntically just as does a selector, but \nsince it is now required to seize the \nterminals of a particular line on the \neighth level rather than to choose an \nidle trunk from a group, it does not \nhunt over this level. As the subscriber \ndials the last digit, 4, the connector \ntakes four steps in a rotary direction, \nseizes the line, rings on it, and con- \nnects the two subscribers together \nwhen the called subscriber answers. \nIt may be seen that if the subscriber \nhad given a preliminary pulse, in ef- \nfect dialing the digit 1, the call would \nhave been routed to a connector \nwhich did not contain the terminals \nof the called subscriber\u2019s line.\n\nTo prevent errors of this sort the \narrangement shown in Figure 2 is \nused. The trunks from the first level \nof the first selectors, instead of being \nrun to regular second selectors, are \nterminated in special second selectors \nwhich are in no way different from\n\nselectors. If the calling subscriber \ndials 9284 preceded by a preliminary \npulse, or in other words dials 19284, \nthe first selector will take one vertical \nstep and find an idle trunk to a special \nsecond selector. When the digit 9g is \ndialed, the special second selector will \nstep vertically nine levels and find an \nidle trunk. In effect, however, the \nninth level of the special second se- \nlector is the same as the ninth level \nof the first selector because the trunks \nfrom these levels are multipled. \nHence the call is now at the same \npoint as it would have been if the \npreliminary pulse had been omitted. \nFurther digits will direct it through \nthe third selectors and the connectors \nto the same line as that to which the \ncorrectly dialed number would have \nrouted it; the preliminary pulse has \ntherefore been absorbed.\n\nformation desk, the toll switchboard, \nthe trouble clerk, and the reverting\n\ncall selectors. In the interests of \neconomy one level, the first of the \nselectors, is used both in counteract- \ning preliminary pulses, and in routing \nthe special calls just mentioned.\n\ncomparatively small number of spe. \ncial second and auxiliary third selec. \ntors. Selectors equivalent in function \nto the special third selectors would \nbe needed for the special calls in any \ncase, even if no effort were made to \ncompensate for the preliminary \npulses.\n\nFig. 2\u2014 Path of a call preceded by a preliminary pulse, showing the supplementary \nswitches for special conditions\n\nalso be used for the special calls, \ntrunks from the first level of the spe- \ncial second selectors are carried to \nspecial third selectors, from whose \nlevels, except the first, trunks go to \njacks and other terminals for com- \npleting the calls. From the first level, \ntrunks run to auxiliary third selec- \ntors, whose banks are multipled with \nthose of the special third selectors to \ncare for preliminary pulses on special \ncalls. With this arrangement, sub- \nscribers are able to make toll, revert- \ning and other special calls by dial- \ning the number 11 followed by an- \nother digit. It may be seen from Fig- \nure 2 that the only additional appa- \nratus required by this method of \ncounteracting preliminary pulses is a\n\nThe handling of service and other \nspecial calls may be seen by tracing \ncalls with and without a preliminary \npulse to the toll switchboard, to \nwhich the number IIo is usually as- \nsigned. Without the preliminary \npulse the call will be directed from \nthe line-finder to the first level of the \nfirst selectors, to the first level of the \nspecial second selectors, to the tenth \nlevel of the special third selectors, and \nthence to the toll operator. When \npreceded by a preliminary pulse, the \ncall will advance from the line finder \nto the first level of the first selectors, \nto the first level of the special second \nselectors, to the first level of the spe- \ncial third selectors, and then to the \ntenth level of the auxiliary third se-\n\nlectors. Since this level is multipled \nto the tenth level of the special third \nselectors, calls routed to the tenth \nlevel of either group of selectors will \ngo to the toll operator.\n\nHaving considered the method by \nwhich calls preceded by preliminary \npulses are routed to their proper des- \ntination let us turn our attention to \ncalls which are routed to vacant se- \nlector levels. The zero (tenth) level \nof the first selector is used for calls \nto the operator, and the first level is \nused as just described. Unless the \ncentral-ofice unit is fully equipped \nfor the 8000 lines possible with the \nremaining levels no numbers will be \nassigned in certain of the hundreds \nand possibly in certain of the thou- \nsands, and the corresponding levels \nof the second or first selectors will be \nvacant. If a call were directed by the \nsubscriber to one of these vacant \nlevels, its progress would be immedi- \nately stopped, and, if no provision \nwere made to care for such a condi- \ntion, the subscriber would receive no \nindication to tell him what is wrong. \nConsequently two different methods \nof handling such calls have been de- \nveloped, the method to be used for a \nparticular selector level depending \nupon whether the calls which must be \nhandled from this level are due to \nerrors in the current directory or to \nsubscribers\u2019 errors.\n\nWhen there is a directory error \ninvolving a vacant selector level, all \nof the trunks of the level are made \nartificially busy except those which \nare carried through intercepting \ntrunk circuits to jacks in front of an \noperator. If a selector starts hunt- \ning over a vacant level equipped in \nthis manner, it will hunt over the busy \ntrunks and seize one of the idle in- \ntercepting trunks, causing a lamp as-\n\nUnless a directory error is in- \nvolved, the calls which come to a \nvacant selector level are nearly al- \nways chargeable to subscribers\u2019 errors \nin dialing. Such levels have all trunks \nmade busy except those which are \nconnected to tone trunk circuits \nwhich, when seized by a selector, re- \nturn a modified busy tone to the sub- \nscriber. The modified busy tone is \nthe same as the ordinary tone, ex- \ncept that every third spurt is omitted. \nSince the subscriber has instructions \nthat such a tone is received only be- \ncause an incorrect number has been \ndialed, he has been informed that he \nhas made some error and may act \naccordingly.\n\nWith either of these methods, the \nnecessary number of intercepting \ntrunk circuits or tone trunks is re- \nduced by multipling the trunks from \nthe several selector levels to the same \nintercepting trunk or tone trunk. \nWhen any one of the trunks becomes \nbusy, it will cause all of the other \ntrunks in the same multiple to test \nbusy. Hence the number of selector \nlevel trunks which may be connected \nto the same intercepting or tone trunk \nis limited by traffic conditions.\n\nnumber as published in the telephone \ndirectory. Likewise a subscriber call- \ning a published number can reason- \nably expect to have the call trans- \nferred to a more recent number, or \nto be informed of any condition pre- \nventing completion of the call. Ac- \ncordingly, the terminals of lines in- \nvolved in number changes, discontinu- \nation of service and similar circum- \nstances are carried from the con- \nnector multiple to trunk circuits \nwhich terminate in jacks in front of \nan operator. When a call is directed\n\nThe number of trunks required \nfrom the unequipped connector ter- \nminals is reduced by multipling sey- \neral connector terminals to the same \ntrunk. Here again, the number which \nmay be multipled to the same trunk \nis limited by traffic conditions.\n\nI thank you for this award; and I wish to express my ap- \npreciation of the honor which it conveys. It is interesting to \nrecall that it was here, in the City of Philadelphia, the vacuum \ntube mentioned in the citation found its first commercial tele- \nphone use. In September, 1913 vacuum tubes were installed \nhere to relay conversations through the telephone cable be- \ntween New York and Washington, and from that small begin- \nning has come its present widespread use. In accepting this \npresent recognition, Mr. Secretary, I feel doubly honored; I \nam proud to stand as the representative of that large body of \nscientists and engineers whose efforts have brought this de- \nvelopment to its present state; and I am profoundly grateful \nto Fortune for allowing me to have a part in the early begin- \nning of this important phase of telephone communication.\n\nITH the institution of the \nfirst telephone central office \nthere arose a need for\n\ntransmitters and receivers which \nwere suitably arranged for oper- \nators\u2019 use. The requirements for \nthese instruments in central office ser- \nvice were so different from those of \nuse at the homes and offices of sub- \nscribers that instruments for the two \ntypes of service necessarily were de- \nveloped along separate lines. The \ndevelopment of operators\u2019 instru- \nments has profited by the funda- \nmental studies made upon _instru- \nments in general, but at the same time \nhas involved differentiations in phys- \nical form to facilitate the work of the \noperators in handling subscribers\u2019 \ncalls.\n\nSucceeding the original \u201cbutter \nstamp\u201d hand telephone the instru- \nment, used by the operators in 1878, \nwas a device serving both as a trans- \nmitter and receiver. In appearance \nslightly resembling a desk stand re- \nceiver of today, it is shown in use in \nFigure 1 of the historical pictures. \nDeveloped by Bell and known there- \nfore as Bell\u2019s hand telephone, it was \na single-pole, permanent magnet in- \nstrument. The magnet was a straight \nbar mounted at right angles to the \ndiaphragm, and bore a coil placed \naround a soft-iron extension at the \nend. The earliest of these instru- \nments were enclosed in cases of wood. \nWhen one was used as a transmitter, \nsound waves impinging upon the dia-\n\nphragm caused it to vibrate and so \nvaried the magnetic field, inducing in \nthe coil varying currents which were \nelectric counterparts of the sound \nwaves. When this instrument was \nused as a receiver the incoming voice \ncurrent, by its variations, changed the \nmagnetic field of the coil through \nwhich it flowed, thereby vibrating the \ndiaphragm and producing a close ap- \nproximation to the original sounds. \nThe following year, 1879, there \ncame the instrument shown in Figure \n2. The transmitter and receiver units \nwere separate, but were arranged to \nbe held in one hand. The transmit- \nter, of the type developed by Edison, \nwas changed in function; it no longer \ngenerated the voice-current, but by \nchanging its resistance in response to \nthe sound waves it varied the current \nsupplied by a battery. A toothed \nmetal ring in contact with the back \nof the diaphragm pressed against the \nfront electrode, a hard carbon disc. \nBy that construction pressure of the \ndisc against the back electrode, a disc \nof soft carbon, was varied, and the \ncurrent was changed correspondingly. \nThe receiver operated on the same \nprinciple as the older instrument and \nwas made up of equivalent parts. In- \nstead of being at right angles to the \ndiaphragm however the bar magnet \nformed the handle on which the trans- \nmitter and receiver were mounted. \nThat same year there was intro- \nduced the operator\u2019s set of Figure 3, \nwith the transmitter and receiver en-\n\n1, Bell\u2019s hand telephone in use as a transmitter. The same instrument acts as a\n\nreceiver with incoming voice-currents; 2, Transmitter and receiver were mounted\n\nBlake transmitter and associated receiver, mounted to leave one of the operator\u2019s\n\nhands free; 4, Instruments the same as those of Figure 3, supported in a \u201cGilliland\n\nharness\u201d to free both hands; 5, Another mounting arrangement keeping both hands \nfree\n\n6, 4 White Transmitter, and a Rich- \nards receiver mounted on a headband. \nThe set was brought out in 1891; 7, \nAnother supported transmitter, used \nwith a completely enclosed receiver. \nCompactness of the receiver was ob- \ntained by use of a spiral magnet; 8, In \nthis set the transmitter is essentially the \nsame as those of today, and the receiver \nts of bipolar type, with a horseshoe mag- \nnet; 9, Today\u2019s set, with receiver held \nby a swivel yoke to a wire headband\n\ntirely separate. The transmitter de- \nrived its name from Francis Blake, \nJr., who carried out the original de- \nvelopment on which it was based; its \noperation was by variation in resist- \nance at the contact between the disc \nof hard carbon and a metal ball sup- \nported against the back of the dia- \nphragm by a light spring. A heavier \nflat spring supporting the disc kept it \npressed against the ball, so that vi-\n\nresting on her shoulders and strap- \nped around her waist, supporting the \ntransmitter a short distance in front \nof her mouth and the receiver at her \near. On account of its weight, almost \n61% pounds when fitted with instru- \nments, this structure was not put into \ngeneral use. Figure 5 shows another \nmounting arrangement brought out \nthat year which held the transmitter \nat a convenient position by a vertical\n\nThough requiring that \nshe hold her head in \na particular position \nfor speaking or listen-\n\nSection of Bell's hand telephone which was used as both standpoint of comfort.\n\nbrations of the diaphragm produced \npressure differences at the contact \narea. To secure the proper range of \npressures, adjustment was provided \nby supporting the spring upon a hinge- \nlike member whose position was con- \ntrolled by a screw. The transmitter \nwas enclosed in a wooden box having \nan opening in front of the diaphragm, \nand was supported before the op- \nerator by a vertical rod. The re- \nceiver, held in the operator\u2019s hand \nduring use, was unchanged in prin- \nciple, but resembled in appearance the \ndesk stand receivers with external \nbinding-posts; the case was of hard \nrubber.\n\nThe next year the same instruments \nwere mounted in a structure which \nfreed both the operator\u2019s hands. This \nwas the \u201cGilliland Harness\u201d of \nFigure 4, an adjustable framework\n\nembodied in operator\u2019s transmitters \nproduced in 1888. In accordance \nwith a somewhat earlier development, \nboth electrodes were of carbon and \nthe space between was filled with car- \nbon granules. Change in resistance \nwith vibration of the diaphragm came \nfrom variations in pressure at the \nlarge number of contact surfaces be- \ntween the granules, rather than at a \nsingle contact surface as before. In \n1890 came the \u201c\u2018solid back\u201d design of \nAnthony C. White, not incorporated \nin operator\u2019s instruments however un- \ntil the following year. The rear elec- \ntrode was no longer mounted on a \nspring, but was attached rigidly \nthrough a small intermediate part to \nthe back of the case, and the front \nelectrode was held by a metal stud \nwhich passed through and clamped \nthe center of the diaphragm. In 1891\n\n4 similar transmitter was made a part \nof the set shown in Figure 6. It is of \ninterest to note that the mouthpece \nused resembles that of today\u2019s desk \nstands in size and shape.\n\nThe receiver used was designed in \n1884 by W. L. Richards, now Con- \nsulting Historian of the Laboratories. \nIt was one of the earliest of the thin or \nwatch-case type, suitable for use with \na headband. Essential parts corre- \nsponded to those of previous designs, \nbut compactness was secured by plac- \ning the bar magnet on the outside of \nthe receiver case to form the bracket \nby which the headband was attached. \nAn extension, consisting of a soft-iron \ncore at right angles to the end of the \nmagnet, ran through the center of the \ncoil inside the case. Small screws near\n\nIn Blake\u2019s transmitter a carbon disc was \npressed by a flat spring against a contact \npoint touching the diaphragm\n\nthe middle of the bar magnet fast- \nened the receiver to a headband suf- \nficiently large in area to distribute the \nweight well over the operator\u2019s head.\n\ndiscs, gave a large number of contact sur- \nfaces for variation in resistance\n\nAbout 1900 came the transmitter \nwhich is similar in form to that now \nused by operators throughout the Bell \nSystem. It was mounted on a breast- \nplate supported and held in place by \na tape extending around the opera- \ntor\u2019s neck. Adjustment of the tape \nand movement of the mouthpiece in \nits ball and socket mounting bring the\n\nopening directly in front of the op- \nerator\u2019s mouth, and cleaning is facil- \nitated by the ease with which the \nmouthpiece can be removed. The re- \nceiver used with this transmitter was \nof bipolar type, and a horseshoe mag- \nnet was substituted for the spiral \nmagnet. Transmitter and receiver are \nshown in Figure 8.\n\nmethod of mounting is changed. On \nthe outer periphery of the case are \ntwo small lugs, 180 degrees apart, \nwith a hole in each for attachment \nof the headband. The band is made \nfrom a loop spring whose ends are \nclamped in a bracket; to the bracket \nis attached a swivel yoke with points \nfor engaging the lugs. By the rota- \ntion and the free movement intro- \nduced, closer fitting is secured and ex- \ntraneous noises are thereby excluded. \nThese instruments, as is apparent, \nsignalize a notable improvement over \nthe early types of operators\u2019 sets, and \nembody the necessary electrical func- \ntions in a form which facilitates great- \nly the speedy completion of calls.\n\nEdward Preston Clifford, administrative officer of the \nBell Telephone Laboratories, whose untimely death was re- \ncorded in Tuesday\u2019s newspapers, was not a scientific man. Yet \nthe death of few New Yorkers could leave a wider gap in the \nhuman machinery of scientific research or one more difficult to \nfill. It was Mr. Clifford\u2019s lifework to help on other people\u2019s \njobs; to hold an important place in the great army of those \nwho smooth the pathway of science along which other men ad-\n\nplishments of his more strictly scientific colleagues Mr. Clifford \nand the skilled but unobtrusive administrators whom he had \ngathered deserve no small share of the praise.\n\nDr. JEWETT was a member of the \nAdvisory Board of The Second Inter- \nnational Conference on Bituminous \nCoal held at Pittsburgh from Novem- \nber 19 to November 24, and presided \nover the morning session of the Con- \nference on November 21. He was \nalso a delegate to represent the Na- \ntional Academy of Sciences.\n\nOn November 23, Dr. Jewett at- \ntended the ceremonies at the inau- \nguration of Harvey Nathaniel Davis \nas President of The Stevens Institute \nof Technology. On this occasion he \nalso represented The University of \nChicago, attending a dinner in honor \nof the delegates at the Hotel Astor \non the evening of November 22.\n\nDr. Jewett gave a short talk be- \nfore the Transmission Conference \nwhich met on November 16 at 195 \nBroadway.\n\nAT THE DECEMBER 3 MEETING of \nthe Colloquium, C. H. Prescott, Jr. \nspoke on \u201cChemical Equilibria at In- \ncandescent Temperatures\u201d. R. V. L. \nHartley spoke on \u201cA Wave Mechanism \nof Quantum Phenomena\u201d on Decem- \nber 17, prior to his presentation of a \npaper on that subject at the Christ- \nmas meeting of the American Physi- \ncal Society.\n\nTHERE ARE NOW AVAILABLE for \nmembers of the Laboratories sets of \npersonal financial records in booklet \nform. These records supplement the \nA. T. & T. income and expense rec- \nord which has been available to mem- \nbers of the organization for some \nyears. They are especially intended \nfor those who do not keep a detailed \nexpense account, but who wish to de-\n\ntermine their financial status at inter- \nvals. Copies of these records may be \nobtained by applying to the Employee \nService Department, Section 1-H, Ex- \ntension 435.\n\nH. Van Deusen, J. R. TowNseEND \nAND C. H. GREENALL were at Haw- \nthorne during the week of November \n19 in connection with the formula- \ntion of specifications for non-ferrous \nsheet and rod stock.\n\nJ. R. TowNnsENpD presented a paper \non telephone apparatus springs be- \nfore the Annual Meeting of the \nAmerican Society of Mechanical En- \ngineers in New York on December 6.\n\nW. T. PriTcHARD visited \u201cMa- \nrine\u2019 central office in Atlantic City \nto make studies of burnishing tools \nfor step-by-step switches.\n\nJ. N. REYNOLDs was at Hawthorne \nduring the week of November 12 in \nconnection with new dial apparatus \ndevelopments.\n\nJ. W. Beyer attended a meeting \nof the High Temperature and Elec- \ntrical Resistance Alloys Committee \nof the American Society for Testing \nMaterials at the Bureau of Standards \nin Washington.\n\nF. F. Lucas spoke on \u2018Micro- \nscopic Structure of Metallic Alloys\u201d \non December 28 before the American \nAssociation for the Advancement of \nScience at the Museum of Natural \nHistory, in this city.\n\nE. W. GENT visited Hawthorne in \nconnection with the production of a \nmodulator unit to be used in equip- \nment for a news reel camera.\n\nR. NorpENSWAN was at Haw- \nthorne in regard to the manufacture \nof a new loud speaker.\n\nN. BisHop recently supervised the \nconversion to provide crystal control \nand increased modulation in the five- \nkilowatt broadcasting equipment of \nthe Churchill Evangelistic Association, \nnow known as the Buffalo Broadcast- \ning Corporation. He performed the \nsame service for the Bankers Life \nCompany of Des Moines; the Strom- \nberg-Carlson Telephone Manufactur- \ning Company of Rochester, and the \none-kilowatt equipment of the Wood- \nmen of the World at Omaha. He \nthen inspected the five-kilowatt sta- \ntion of the National Life and Acci- \ndent Insurance Company at Nash- \nville, Tennessee.\n\nB. R. Coe recently made a sur- \nvey for Memphis Commercial Ap- \npeal, Incorporated, at Memphis, \nTennessee, which recently purchased \na one-kilowatt, crystal-controlled \nbroadcasting equipment.\n\nJ. C. HeRBER supervised the con- \nversion to crystal control of the five- \nkilowatt equipments of the Moody \nBible Institute and Sears, Roebuck \nand Company of Chicago, and the \nConsolidated Gas, Electric Light and \nPower Company of Baltimore; also \nthe one-kilowatt equipments of Larus \nand Brother Company, Richmond, \nVirginia and the Detroit News and \nthe Milwaukee Journal.\n\nF. E. NIMMCKE supervised the con- \nversion to crystal control of the one- \nkilowatt broadcasting equipments of \nthe Universal Broadcasting Company \nof Philadelphia and the Peoples Pul- \npit Association at Rossville, Staten \nIsland.\n\nH. S. Price made surveys for the \none-kilowatt equipment purchased by \nthe Puget Sound Broadcasting Com- \npany, Seattle, and the five-kilowatt\n\nequipment purchased by the South- \nwestern Sales Corporation of Tulsa. \nEn route he inspected the one-kilo- \nwatt station of Louis Wasmer, Inc. at \nSpokane, and visited Fort Worth to \nconfer on the relocation of the one- \nkilowatt equipment taken over from \nthe Searchlight Publishing Company \nby the Texas Air Transport Broad- \ncasting Company.\n\nW. L. TieRNEy has completed the \ninstallations of a one-kilowatt equip- \nment for the Oregon State Agricul- \nture College at Corvallis, Oregon, \nand a five-kilowatt equipment at Sher- \nman Oaks, California, for the West- \nern Broadcasting Company. He also \nmade a survey for the five-kilowatt \nequipment purchased by Hale Brothers, \nSan Francisco, and supervised the \ncrystal control conversions of the one- \nkilowatt equipments of Don Lee, In- \ncorporated, at Los Angeles and San \nFrancisco, and of the Fisher Blend \nStation at Seattle, Warner Brothers \nat Los Angeles, and Nichols and \nWarinner at Long Beach, California.\n\nO. W. Towner supervised the con- \nversion to crystal control of the five- \nkilowatt equipment of the Voice of \nSt. Louis, Inc., and the one-kilowatt \nequipment of the Jenny Wren Con- \npany, Lawrence, Kansas, the Gurney \nSeed Company of Yankton, South \nDakota, and the Kansas City Star.\n\nR. M. Pease has, during the past \nfew months, assisted S. P. Grace in \ndemonstrations given in connection \nwith Mr. Grace\u2019s talks. He was pres- \nent at the latter\u2019s lectures before the \nRotary Club of Brooklyn on Decem- \nber 6; Sigma Xi Chapter of McGill \nUniversity at Montreal on December \n19, and before the employees of The \nBell Telephone Company of Canada \nat Montreal on December 20. He \nalso aided in setting up the equip- \nment used by John Mills in his Tor-\n\nJ. R. P. GOLLER AND R. L. Luns- \nFORD inspected the new unit type toll \nrepeater power installation at Greens- \nboro, North Carolina. This instal- \nlation is part of the Washington- \nAtlanta Cable.\n\nJ. R. Stone anp C. BoRGMANN \ntested and inspected high speed cen- \ntrifugal exhauster equipment for \npneumatic tube installations in West \nLynn, Massachusetts.\n\nV. T. CALLAHAN tested a BA gas- \noline engine set, recently installed in \nSavannah, Georgia.\n\nC. E. BoMAN spent several days in \nChicago in connection with problems \nof machine switching equipment.\n\nI. W. Brown visited Milwaukee \nand Detroit, preliminary to the in- \nstallation of new toll facilities there.\n\nJ. H. Sore visited the General \nElectric Company at Pittsfield, Mas- \nsachusetts, to confer on the develop- \nment of a transformer to provide \ncurrent for lamp signals in toll \nboards.\n\nJ. IrisH visited Chicago in con- \nnection with circuit drawings and \npractices.\n\nD. M. Terry discussed the 1-A \ncarrier pilot channel equipment with \nmembers of the staff at Hawthorne.\n\nJ. P. Kinzer visited Morristown, \nNew Jersey, and Allentown, Reading \nand Harrisburg, Pennsylvania, in \nconnection with modifications of 2- \nwire repeaters in cable circuits.\n\nA. E. BACHELET conducted tests \nof half-ampere ballast lamps used \nwith 2- and 4-wire repeaters at the \nA.T.&T. repeater station at Juliet, \nIllinois.\n\nJ. A. Krecek spent several weeks \nin El Paso, Texas, testing 2-wire echo \nsuppressors to be used in conjunction\n\nkK. P. BANCROFT is engaged in the \ninstallation of telephone and carrier \ntelegraph equipment for the Cana- \ndian Pacific Railways.\n\nC. B. Suriirr recently assisted \nwith the demonstration of an oscillo- \ngraph before the American Optical \nSociety in Washington.\n\nwere present at a meeting of the \nAmerican Society for Testing Ma- \nterials at Philadelphia.\n\nH. H. Lowry attended a meeting \nof the Committee on Insulation Ma- \nterials of the National Research \nCouncil at Johns Hopkins University \nin Baltimore.\n\nA. R. Kemp visited the plant of \nthe General Electric Company at \nBridgeport on November 20 to con- \nfer on the development of rubber cov- \nered wire and cable.\n\nJ. H. INGMANSON viewed a dem- \nonstration of the impregnation of \njute for the protection of lead cable \nsheath, at the Schlichter Jute Cord- \nage Company, Philadelphia.\n\nW. A. Marrison demonstrated \noscillographs of polar and motion- \npicture types at the apparatus exhibit \nheld in connection with the meeting \nof the Optical Society of America in \nWashington.\n\nC. H. Haynes conferred with rep- \nresentatives of the Standard Ma- \nchine Company, Providence, on the \nconstruction of special rolls for the \nmanufacture of platinum filament.\n\nC. A. KoTTERMAN made an inspec- \ntion trip in connection with the Labo- \nratories\u2019 exhibit at the National Acad- \nemy of Sciences, Washington.\n\nT. C. Fry spoke on the use of con- \ntinued fractions in the design of elec- \ntrical networks before the American\n\nC. J. Davisson AND L. H. Ger- \nMER presented a paper on a test of \nthe state of polarization of reflected \nelectron waves before the same So- \nciety on December 1.\n\nJ. C. ScHELLING addressed the \nPhysics Seminar of Cornell Univer- \nsity on November 19 on \u201cSome Theo- \nretical and Experimental Phases of \nRadio Transmission.\u201d\n\nE. F. Kincssury spoke on televi- \nsion before the Men\u2019s Club of Ruth- \nerford on December 14.\n\nE. G. D. PATERSON visited the \nHighway Trailer Company at Edger- \nton, Wisconsin, on November 12, in \nconnection with studies of inspection \nmethods for automotive equipment.\n\nW. A. Boyp attended a regular \nsurvey conference on step-by-step ap- \nparatus at Hawthorne.\n\nH. F. KorTHever anp H. C. \nCUNNINGHAM attended a regular \nsurvey conference in connection with \ncarrier equipment at Kearny.\n\nG. D. Epwarps visited Cleveland, \nChicago and Boston during the latter \npart of November and early Decem- \nber, in connection with Field Inspec- \ntion Engineering work. While in \nBoston Mr. Edwards attended the \ndecennial anniversary celebration of \nthe \u201cBinaural Sons of the C\u2019\u2019, an or-\n\nganization composed of representa. \ntives of the United States Navy De. \npartment, the Submarine Signal Com. \npany, the General Electric Company, \nthe Western Electric Company and \nof various educational institutions \nwhich cooperated during the World \nWar in developing methods and ap- \nparatus for the detection and loca- \ntion of enemy submarines.\n\nW. A. SHEWHART visited Philadel- \nphia during the latter part of Oc. \ntober to attend conferences held by \nthe American Society for Steel Treat- \ning and the American Society for \nTesting Materials.\n\nA. W. Drinc investigated several \nspecial cable terminal installations in \nPhiladelphia on November 13.\n\nJ. M. Harpesty made tests of re- \ninforced concrete poles in Charlotte, \nNorth Carolina.\n\nR. C. EGGLESTON investigated pre- \nservative treatments and made \nstrength tests of jack-pine poles in \nMinneapolis, from October 20 to \nNovember 12.\n\nE. M. Honan observed field trial \ninstallations of drop wire in the ter- \nritories of the New York Telephone \nCompany, the New Jersey Bell Tele- \nphone Company and The Southern \nNew England Telephone Company.\n\nE. P. Cxiirrorp, Vice-President of \nthe Laboratories, died Sunday, De- \ncember 16. He entered the Western \nElectric Company in 1892 as an of- \nfice boy, and became Cashier of the \nNew York office some six years later. \nAfter experiences as Chief Clerk at \nChicago, Philadelphia and New York, \nhe became Assistant Manager of the \nNew York House, and in 1911 its \nManager, with supervision of the\n\nBoston, Philadelphia and Pittsburgh \nHouses. In 1917 he was appointed \nEastern District Manager of the \nWestern Electric Company. On July \n1, 1918, Mr. Clifford became Office \nManager of the Engineering Depart- \nment. A year later he became Com- \nmercial Manager and his responsibil- \nities were gradually extended to em- \nbrace all non-technical activities. In \n1925 the Engineering Department \nbecame Bell Telephone Laboratories, \nIncorporated, and Mr. Clifford, as \nVice-President, was placed in charge \nof the General Staff Department.\n\nG. B. THoMAs attended a dinner \nof electrical engineering students at \nMassachusetts Institute of Technol- \nogy, at which the Bell System cooper- \native course was discussed. Mr. \nThomas also addressed a group of \nsenior students of electrical and me- \nchanical engineering at Cornell Uni- \nversity on December 14.\n\nJoun Mitts addressed the Cen- \ntral Association of Science and \nMathematical Teachers at Chicago \non December 1, on \u201cThrough Elec- \ntrical Eyes.\u201d On December 13, Mr. \nMills spoke before the Engineering \nstudents of Toronto University, and \non the following day, before em- \nployees of The Bell Telephone Com- \npany of Canada in Toronto, on re- \ncent developments of these Labo- \nratories.\n\nS. P. GRrAceE included in his lec- \nture program for December talks \nbefore the Rotary Club of Brooklyn, \non December 6; Sigma Xi Chapter \nof McGill University, Montreal, on \nDecember 19, and employees of The \nBell Telephone Company of Canada,\n\nR. A. DELLER spoke on television \nbefore a group of Student Engineers \nat the New York Telephone Com- \npany Headquarters. \u2018Through Elec- \ntrical Eyes\u201d was the subject of a talk \ngiven by him on December 4 before \nthe A. S. M. E., A. I. E. E. and the \nProvidence Engineering Society at \nBrown University.\n\nEpwarpD AIKEN, a porter in the \nPlant and Shops Department, died \non November 17. He had been with \nthe Laboratories since March, 1927.\n\nL. S. O\u2019RoarkK spoke on the elec- \ntrical transmission of personality be- \nfore the Buffalo Section of the A.I. \nE.E. on December 14.\n\nW. F. JOHNSON AND B. B. WEBB \nwere in Trenton, Lawrenceville and \nPrinceton, New Jersey, in connection \nwith commercial matters involved in \nthe installation of four short wave \nradio transmitters which the Lab- \noratories are manufacturing for the \nLong Lines Department, American \nTelephone and Telegraph Company.\n\nP. May has been assigned to duties \nat the Radio Station at Lawrenceville \nwhere he will handle the commercial \nphases incident to the installation of \u00a9 \nthe short wave transmitting sets.\n\nH. W. Drppet recently visited the \nCorning Glass Works, Corning, New \nYork, to arrange for the purchase of \nglass needed for the enlarged tube \nshop program.\n\nW. B. WELLs visited Schenectady, \nand J. W. Schmied and P. C. Smith \nwent to Washington, in connection \nwith the prosecution of patents.\n\nThe next dance of the Club will \ntake place on Wednesday, February \n6, 1929. The music will be Herbert \nHood\u2019s, and D. R. McCormack is \narranging something very special by \nway of surprise. Tickets are $1.00, \nand are obtainable from D. D. Hag- \ngerty or the departmental representa- \ntives.\n\nUrged by importunate Brooklyn- \nites, the dance will be held at the Ho- \ntel St. George. Contrary to the stub- \nborn preconceptions of out-of-town- \ners, Brooklyn Heights is quite as ac- \ncessible for the average Club member \nas Manhattan. The I. R. T. thought- \nfully built the Clark Street Station in \na sub-cellar of the St. George \u2014a \nfact to be remembered by Jerseyites, \nwho will find that the trip from Cort- \nlandt Street to Clark Street is a mat- \nter of only fifteen minutes.\n\nDuring the first half of the present \nseason the amazing record of one hun- \ndred per cent attendance was estab- \nlished. One hundred and sixty men \nhave bowled every Friday evening for \nfourteen weeks, and four additional \nalleys have been reserved for those \nwho are not members of the regular \nleague. These men, though not elig- \nible for league prizes, will substitute \nin the order of their averages, for \npossible absentees of the league \nteams.\n\nClub records show a decided im- \nprovement over last year\u2019s scores. \nThere is keen competition for the \nvarious prizes to be awarded. Spe- \ncial interest is shown in the D. D.\n\nHaggerty Trophy, which goes to the \nbowler rolling the highest score for \nthe entire season.\n\nOn November 23, the league offi- \ncers expressed the season\u2019s greetings \nby donating two turkeys, to be \nawarded as prizes for the evening\u2019s \nhighest scores. T. C. Rice and N. \nScribner, each, to borrow a phrase \nfrom another sport, \u201cgot a birdie\u2019.\n\nThe final event of the hiking sea- \nson, on Sunday, November 25, nar- \nrowly escaped the blight of supersti- \ntion. Of thirteen who entrained for \nTuxedo only twelve were in evidence \nat Tuxedo station. But the satisfac- \ntion of those who had advocated \ndrawing lots to decide which member \nof the party should be thrown over- \nboard was short-lived. The  thir- \nteenth hiker, after buying a bar of \nchocolate to round out his luncheon, \nrejoin the group and the hikers gaily \nhit the trail for Mt. Ivy.\n\nSnow had not yet fallen in the city, \nand the party was unprepared to find \nit underfoot (as much as one-quarter \ninch of it, in spots). No casualties \nresulted from this cause, however; \nthe one snowball thrown, missed.\n\nBy way of the T-MI (Tuxedo-Mt. \nIvy) trail and voluntary detours, the \nhikers wound their way through the \nRamapo hills to Ladentown. Since \nthe party was still full of vigor, \nthough the scheduled hike had been \npractically completed, plans were re- \nvised, and the hikers swung down the \nroad to Suffern. Such energy was \nbound to be rewarded, and it was,\n\nwhen the party came upon a cider mill \nand stopped to sample its product. \nMuch refreshed, the group stepped \nout with a will and reached Suffern \nnot merely in time for the train, but \nwith ample leisure for a sumptuous \nrepast in a nearby lunch wagon. And \nso to home!\n\nThough the official hiking season \nis over, the committee is always glad \nto get in touch with Club members \nwho are interested in that form of \nrecreation. If you have not already \nmade yourself known to them, call \nMiss Barton, extension 857, or Alex- \nander Grendon, extension 637, and \nmake sure to be informed of the hik- \ning Club\u2019s further activities.\n\nThe Laboratories team is tied for \nfirst place in the Bell System League, \nin spite of a handicap incurred at the \nstart, when it lost its first game to \nthe Installation Department of West- \nern Electric. A number of ardent \nrooters have been witnessing the \ngames, and Manager Granville Mat- \nthews is confident that our team will \nrepeat the splendid showing it made \nlast year. On Monday and Friday \nevenings, games are played at Stuy- \nvesant High School; on Wednesday \nevening, at Erasmus Hall High \nSchool. The games begin at 8:30, \nand are followed by dancing.\n\nThe squad: Gittenberger, C. W. \nMaurer, Steinmetz, Christ, Cahill, \nKirsch, Keogh and West.\n\nFirst prize in the December indoor \ngolf tournament was won by J. G. \nDusheck, who defeated R. J. Nos- \nsaman two up. As a result, Mr. Dush- \neck, who obviously needs it least, \nhas a season membership in the In- \ndoor Golf Course of America.\n\nWith memories of Christmas win- \ndow-displays, the entrants rather ex- \npected to see a toy elephant in the \nbushes of the thirteenth hole, or to \nhear the rattle of an electric train \namong the green valleys of the course. \nSo realistic was the scenery that play- \ners were seen trying to open the swing- \ning door of a supposed locker room.\n\nSeventy-two entrants played twice \naround and G. A. Brodley won the \nmedal offered for low medal score. \nThe sixty-four who qualified were \ndivided into two groups of four \nflights each. Group 1, Flight A was \nwon by R. J. Nossaman, who early \nestablished a lead over N. Thorne, \nand finished at the Fifteenth, four up. \nOther winners in this group were R. \nKoernig, I. W. Whiteside, and J. A. \nHall. J. G. Dusheck took the final \nmatch of Group 2, Class A from T. \nC. Rice by six holes. Scoring six \n\u2018\u201ctholes in one,\u201d\u2019 and his usual cool- \nheaded putting contributed to this \nvictory. E. H. Clark, J. M. Hey- \nwood, and E. K. Ebehardt were \nthe other winners of this group. Each \nwinner in the two groups received a \nprize and the respective winners of \nClass A played off for the special \nprizes donated by the management. \nIn this contest, Dusheck took the \nlead on the third hole. Nossaman \ncame up even, but lost his lead at the \nninth. He was two down at the six- \nteenth, and by a twenty-inch putt lost \nthe seventeenth hole and the match.\n\nThose elected as Club officers for \nthe coming year are: O. M. Glunt, \npresident; H. F. Dodge, first vice- \npresident; Marian Kane, second vice-\n\nrepresentative; J. W. Hultquist, Re- \nsearch; P. B. Fairlamb, Systems De- \nvelopment; and J. C. Kennelty, Com- \nmercial.\n\nTwo swimming classes will again \nbe conducted: one on Mondays from \n7:00 to 7:30, starting January 7; the \nother on Wednesdays, from 5:30 to \n6:00, commencing on January 9. \n\u2018Lhe courses will consist of eight les- \nsons covering a period of ten weeks \nat a charge of $2.50. They will be \ngiven at the Carroll Club, Madison \nAvenue and 30th Street, under the \ntutelage of Miss K. Spranger.\n\nThe Bell System Basketball Lea- \ngue, which this year has included the \ngirls\u2019 teams, has provided a definite \nschedule of games. Two have been \ncompleted, with three more to be \nplayed on the evenings of January \n7 at St. Luke\u2019s court, and January \n16 and 30 at Erasmus Hall High \nSchool. Games commence at 7:30 \nand dancing follows.\n\nThe Laboratories team played two \ngames outside its regular schedule, \nwinning on each occasion. The first \nengagement was with the Federal Re- \nserve team; the other, an exhibition\n\ngame with the Long Island Divi- \nsion of the New York Telephone \nCompany. Further outside games \nare to be booked. On our squad are \nAnn Barioni, Marie Boman, Mari- \nanne Grimm, Lilliam Kaempffe,. \nDorothy Murtha, Marie O\u2019Neill and \nNatalie Skinner.\n\nFour women\u2019s teams have been \nbowling regularly every Friday since \nOctober 19, 1928 at Dwyer\u2019s. Though \nthe women\u2019s bowling club is a new \ndeparture, it has become an unusually \npopular activity. Novices desiring an \nopportunity to play may arrange \nthrough Marion ane or Leona Feil \nto act as substitutes for members of \nthe regular teams. Prizes will be \nawarded at the close of the season.\n\nAfter careful consideration, the \ncommittee in charge of the Christ- \nmas Poster Contest selected fifteen \nof the best posters, submitting these \nfor final judging to two well-known \ncommercial artists. The winning \nposter was reproduced and displayed \non the bulletin boards during the \nChristmas season. It was made by E. \nAlisch, of the Equipment Drafting \ngroup who was awarded a prize of a \nten-dollar gold piece.", "title": "Bell Laboratories Record 1929-01: Vol 7 Iss 5", "trim_reasons": [], "year": 1929}
{"archive_ref": "sim_record-at-t-bell-laboratories_1934-10_13_2", "canonical_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1934-10_13_2", "char_count": 90891, "collection": "archive-org-bell-labs", "doc_id": 213, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc213", "record_count": 100, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1934-10_13_2", "split": "test", "text": "transoceanic telephony in 1928 \nand 1929, attention was directed by \nBell Telephone Laboratories toward \ndetermining the properties and use- \nfulness of ultra-short waves. The \nshort-wave transoceanic circuits are \noperated at frequencies between 5 \nand 21 megacycles while the ultra- \nshort waves are at frequencies above \n30 megacycles, which is generally \ntaken as the upper limit of the short \nwave range. It had previously been \ndiscovered that these higher frequen- \ncies were not in general reflected from \nthe Kennelly-Heaviside layer. They \nwere, therefore, considered primarily\n\ncation, where the waves followed es- \nsentially an optical or straight-line \npath from the transmitter to the re- \nceiver. In the telephone plant there \nare instances where natural barriers \nso separate points only a short dis- \ntance apart that it is difficult and ex- \npensive to construct ordinary tele- \nphone lines or submarine cables to \nconnect them. It seemed that for \nsuch conditions ultra-short wave radio \nextensions might be a satisfactory \nmeans of giving telephone communi- \ncation. To be economically feasible, \nhowever, such radio circuits must be \ninexpensive both in first cost and in \noperation.\n\nDuring the last few years, the Lab- \noratories have been experimenting \nwith an ultra-short wave circuit be- \ntween Deal and Holmdel, New Jersey, \nwith the thought of developing equip- \nment capable of unattended operation. \nSome time ago this development \nreached the stage where it seemed de- \nsirable to carry out a trial of a two- \nway circuit under conditions approx- \nimating commercial use to gain ex- \nperience with the problems involved \nin regular operation. In particular, \nit was desired to design and install the \nradio stations for operation without \ndirect attendance, so that the appara- \ntus could be located remotely from a \ncentral! office.\n\nAfter a study of possible locations, \nit was decided in cooperation with the \nNew England Telephone and Tele- \ngraph Company to carry out the trial \ninstallation across Cape Cod Bay, be- \ntween Green Harbor and Province- \ntown, Mass. The coastal station of \nthe New England Telephone and Tele- \ngraph Company, already existing at \nGreen Harbor, made a convenient \nplace in which to install one end of the \nsystem. The physical conditions are \nalso favorable for an ultra-short wave\n\nlink between that point and Province- \ntown, 25 miles away. The sand dunes \nnear Provincetown, rising about 100 \nfeet in height, make it possible to \nsecure an optical path across the bay. \nFurthermore, Provincetown is fairly \naccessible by motor car around the \nCape and is already provided with \nwire circuits, so that the radio link \nneed not be completely depended \nupon. This location is thus a good \nproving ground for this new type of \ntelephone circuit.\n\nAccordingly, the radio link has been \nestablished across the bay, as indicated \non the map, and extended at Green \nHarbor by wire to Boston, to form \na direct Boston-Provincetown toll \ncircuit. At Boston and at Province- \ntown the circuit appears at a jack \nin the switchboard alongside the jacks \nof other toll circuits. The insertion of \na cord into the jack starts the radio \ntransmitter at that end of the radio \nlink. The receivers at both ends are \nkept in constant operation while the \ncircuit is available for trafic. Ringing \nis accomplished by sending a 1000 \ncycle tone interrupted at twenty \ncycles over the radio circuit. Since \nthe radio transmitter requires less\n\nCRYSTAL FIRST SECOND FINAL FREQUENCY \nOSCILLATOR HARMONIC HARMONIC AMPLIFIERS \n8.125 MC GENERATOR GENERATOR NO. | NO. 2 \n32.5 MC 65 MC 65 MC ___OUTPUT \n16.25 MC \nFIRST STAGE \nAUDIO AMPLIFIER \nAUDIO \nINPUT \n| SECOND 310 VOLTS 1000 VOLTS \nSTAGE RECTIFIERS \nAUDIO \nAMPLIFIER \nRELAY TO MAKE \nSTAGE OSCILLATE \nAT 1500 CYCLES \nFOR TEST \nFig. 1\u2014Block schematic of ultra-short-wave transmitter \nOctober 1934 35\n\nthan one second to start, the operator \nmay ring immediately after inserting \nthe cord. Privacy equipment, similar \nto that used on the transatlantic short \nwave channels, is installed.\n\nThe receiver is started and stopped \nby the operation of a key at the local \ntest board. The power supply is \narranged so that when the receiver is \nin operation, current is also applied \nto some of the filaments of the trans- \nmitter. Provision is also made for \ntesting the overall operation of the \ntransmitter and receiver at each end \nfrom the local test board. A tone is \ngenerated which modulates the trans- \nmitter, and if both transmitter and \nreceiver are operating properly, a side- \ntone will be produced in the local \nreceiver which can be heard by the \ntest board operator.\n\nThe transmitter, developed by R. \nW. Friis and L. M. Klenk under the \nsupervision of N. F. Schlaack, is \ncrystal controlled, and is capable of \ndelivering 15 watts of carrier power \nwhich can be completely modulated. \nA block schematic for the Green Har- \nbor transmitter is shown in Figure 1. \nThe Provincetown transmitter is the \nsame except that the output frequency \nis 63 megacycles. A quartz crystal\n\noscillator is followed by two harmonic \ngenerators, a push-pull modulating \namplifier, and a push-pull power am- \nplifier. Modulation is accomplished \nby supplying audio-frequency modu- \nlating power to the plate and screen of \nthe modulating amplifier and to the \nscreens of the second harmonic gener- \nator and the power amplifier.\n\nTwo rectifiers employing hot-cath- \node mercury-vapor tubes supply plate \nand screen potentials for all tubes. \nGrid bias potentials are obtained from \ncathode resistors and grid leaks. Grid \nand plate circuits of each stage are \nshielded from each other to prevent \nextraneous coupling and interstage \nfeedback. The transmitter operates \nentirely on standard commercial 1 1o- \nvolt, 60-cycle current.\n\nThe radio receivers, developed by \nG. Rodwin and C. H. Swannack under \nthe direction of F. A. Polkinghorn, \nalso operate from a 110-volt, 60-cycle \ncircuit, and are of the double detection \ntype. A block schematic is shown in \nFigure 2. To make unattended opera- \ntion possible, a crystal oscillator is \nused as a source of beating frequency. \nA single-stage harmonic generator \nproduces sufficient voltage of the \neighth harmonic of the crystal fre-\n\nCRYSTAL \nOSCILLATOR \n8325 KC \n16,650 KC \nHARMONIC \n3 STAGES OF \nINTERMEDIATE \nFREQUENCY AMPLIFIERS \n66,600 KC AT 1600 KC SECOND AUDIO \nBAND WIDTH= 50 KC DEMODULATOR AMPLIFIER \nRECEIVED INE \nSIGNAL \n65,000 KC \nFIRST \nDEMODULATOR \nAUTOMATIC \nVOLUME CONTROL \nFig. 2\u2014Block schematic of ultra-short-wave receiver \n36 October 1934\n\nFig. 3\u2014The ultra-short-wave transmitter is \nmounted in a metal container suitable for \npole mounting\n\nquency for satisfactory operation of \nthe detector. The intermediate fre- \nquency amplifier consists of three \nstages of amplification at 1600 kilo- \ncycles, and has a band width of \napproximately 50 kilocycles. A small \namount of automatic volume control \nis provided to compensate for slight \nvariations in received voltage caused \nby variation in humidity and other \nfactors.\n\nThe receiving and transmitting an- \ntennas are identical and are mounted \non 100 foot poles fifty feet apart. \nHorizontal exciter and reflector ele-\n\nments are supported on standard \ncross-arms. Four pairs of half-wave \nexciter elements, each comprising two \nhalf-wavelength conductors, are spaced \none-half wavelength apart in a vertical \nplane on one side of the pole. Re- \nflector elements are similarly arranged \non the opposite side of the pole, the \nspacing between exciters and reflectors \nbeing one-quarter wavelength. The \ntransmitter and receiver are each \nmounted in a metal container suitable \nfor mounting on the antenna poles at \na later date. At the present time they \nare installed in a small building lo-\n\nSions to prevent maintenance men from com- \ning in contact with high potentials\n\ncated between the transmitting and \nreceiving antennas. The mechanical \ndesign of the transmitters and the \nstation layout were made by M. E. \nFultz and J. L. Mathison.\n\nunder the direction of A. A. Oswald. \nThe system was put into trial service \nearly in July. It was found to yield a \nhigh-grade, two-way telephone cir- \ncuit and is adding to our knowledge of \nthe practical problems and capabili- \nties of such a system.\n\nThis picture shows all that remains in evidence today at Montauk, Long Island, of the \nexperimental radiotelephone station which was erected there by the Bell System in 1915 \nand shown on page 512 of the Recorn for Fuly 1931. This was the first radiotelephone \neffort of the Bell System and, indeed, was the first vacuum-tube radiotelephone trans- \nmitter ever built. Its successful transmission to Wilmington and to St. Simon\u2019s Island, \nGeorgia, were preliminaries to the famous experiments later that year, in which the voice \nwas first projected across the oceans. All that remains is this box on a stub pole. It \nhoused the switch which threw the antenna lead-in either to ground or to the apparatus \nwithin a building which siood nearby\n\nthe receiver is lifted from the \nhook to the instant it is returned\u2014 \nmagnetic forces play a very important \nrole. The relay which closes the lamp \ncircuit and thus attracts the attention \nof the operator; the transformer, called \na repeating coil, through which the \nspeech current flows from one sub- \nscriber\u2019s circuit to another or to a \ntrunk; the ringer which summons the \ncalled subscriber and the receiver it- \nself which transforms the speech cur- \nrents into sound waves\u2014all depend \nupon magnetic materials for their \noperation. What material is the best \nfor each purpose has been the subject \nof continuous inquiry and develop- \nment in the Laboratories.\n\nIn a very general way magnetic \nmaterials divide into two classes: \nthose which can be easily magnetized \nbut will lose their magnetism when \nsubjected to small reverse magnetiz- \ning forces; and those which, although \nmore difficult to magnetize, retain a \nlarge part of their magnetism even \nunder the action of considerable de- \nmagnetizing forces. The first class \ncomprises the so-called magnetically \nsoft materials used for electromagnets, \namong which are magnetic iron, silicon \nsteel, the permalloys, cobalt iron, \nperminvar and permalloy: dust. In \nthe second class, of magnetically hard \nmaterials for permanent magnets, are \nfound low chrome steel, high chrome \nsteel, tungsten steel, and cobalt steel.\n\nterial there are five properties which \nserve as criteria of its quality. The \nfirst is the ease with which it may be \nmagnetized, a quality known as \u201c per- \nmeability\u201d, which is measured by the \nratio of the flux density to the mag- \nnetizing force which produces it. Not \nonly do materials differ in permea- \nbility, but the same material itself has \ndifferent values of permeability under \ndifferent magnetic conditions. The \nease of magnetization, in other words, \ndepends upon the intensity of mag- \nnetization. \u2018One of the permalloys, \nfor example, has a very high initial \npermeability which increases rapidly \nuntil a flux density of 5,000 or 6,000 \ngausses is reached. From this point \nthe permeability falls rapidly to a \nflux density of about 11,000 gausses \nwhen it becomes practically saturated. \nFor moderately high values of mag- \nnetizing force and, correspondingly, \nof flux density, magnetic iron on the \nother hand has a relatively high per- \nmeability, although for small mag- \nnetizing forces its permeability is only \na small fraction of that of permalloy.\n\nAnother property of importance in \na material is the continuance of the \nmagnetized condition when the mag- \nnetizing force has been removed. This \nquality is measured by the residual \ninduction, that is, that value of flux \ndensity remaining in a completely \nclosed core of the material when a \ngiven magnetizing force is removed. \nAnother very important quality is \nthe coercive force\u2014the magnetizing \nforce which must be applied in the\n\n\u201c frictional resistance. This \nphenomenon, known as \n\u201chysteresis\u201d gives rise to \nthe term \u201chysteresis loss\u201d \nwhich represents the ener- \ngy used up in such a rear- \nrangement whenever the \nmaterial undergoes _peri-\n\nzation, as in an alternating \ncurrent transformer. \nAnother loss under the \nsame conditions arises from \nthe existence of eddy cur- \nrents which are induced \nwithin a magnetic core \nwhen it is energized by an \nalternating current. This \nloss, like the hysteresis loss,\n\nhence its designation \u201c\u2018hysteresis loop.\u201d Dotted lines are \nfor lesser degrees of magnetization\n\nopposite direction to remove the \nresidual induction: that is, to de- \nmagnetize the material. These last \ntwo qualities are of course most im- \nportant in the selection of materials \nfor the production of permanent mag- \nnets such as are used in polarized \nrelays, in telephone receivers and in \nmagneto-generators. The existence of \nthe qualities of residual induction and \ncoercive force means that the flux \ndensity within the magnetized core \ndepends at any instant not only upon \nthe magnetizing force then acting, but \nalso upon the previous magnetic his- \ntory of the core.\n\nIn attaining its successive values \nthe flux density lags behind a changing \nmagnetizing force much as if the mag- \nnetic elements of the core were being \nrearranged against the opposition of\n\nmanifests itself by heat. \nSectionalizing the core by \nlaminations insulated from \none another reduces this \nloss. Similarly, using a \nmaterial of a high elec- \ntrical resistivity means fee- \nbler eddy currents and \nsmaller: losses. Hysteresis loss and \nelectrical resistivity are therefore two \nqualities of importance in the design \nof apparatus for alternating currents.\n\nThe fact, for example, that the \nelectrical resistance of magnetic iron \nis relatively low, hence, that its eddy \ncurrent loss is large, practically limits \nits use in the design of telephonic \nequipment to cores for electromagnets \nwhich operate on direct currents. The \nfact that its coercive force is fairly \nlow and its permeability quite high, \neven at relatively high flux densities, \nindicates that it may be used for re- \nlays in which the armature is attracted \nwhen a current is sent through the \nwindings and falls away again under \ngravity or spring pressure when the \ncurrent ceases. \u201cMagnetic iron\u2019 is a \nterm applied to the material which is\n\nIn silicon steel, which may be con- \nsidered a 4% silicon alloy of magnetic \niron, the resistivity is about five times \nthat of magnetic iron. Eddy currents \nare greatly decreased in consequence, \nand the material is therefore adapted \nfor use in cores of apparatus operating \nwithin the audio frequency range. \nThis steel also has the advantage of a \nconsiderably higher permeability than \nmagnetic iron at the low flux densities \nwhich are usually produced in tele- \nphone apparatus operated by speech \ncurrents. It is used largely in cores \nof transformers, and induction and \nretardation coils.\n\nA material physically superior to \nsilicon steel for the cores of audio fre- \nquency apparatus is \u2018 45 permalloy\u201d. \nThe name \u201cpermalloy\u201d\u2019 covers certain \nalloys of nickel and iron developed in \nthese Laboratories and possessing re- \nmarkable magnetic properties. Differ- \nent combinations of these materials \nare designated by the amount of \nnickel which they contain. The \u201c45 \npermalloy\u201d has a resistivity nearly as \nhigh as silicon steel and a permeability \nat low or moderately high flux den- \nsities two or three times greater. Its \ncoercive force is also considerably \nlower. Its cost however is considera- \nbly greater and in designing apparatus \nthis factor must be balanced against \nits better magnetic properties. It is \nused largely in transformers and cer- \ntain types of relays.\n\nThe \u201c78 permalloy\u201d is a material \nhaving remarkably high permeabil- \nities. Its initial and maximum per- \nmeabilities are more than ten times \nthose of silicon steel while its coercive \nforce is only about one-tenth. Al- \nthough i its hysteresis loss is very low, \nits resistivity which is only about one- \nthird that of silicon steel makes it\n\nless suited than some of the other . \npermalloys for audio frequency appa- \nratus. It is used mainly for the cores \nand armatures of sensitive relays.\n\nTo adapt permalloy for high fre- \nquency use about 4% of the iron con- \ntent of the 78 permalloy was replaced \nby a like amount of chromium. This \ngave us \u201c3.8-78 chrome permalloy\u201d \nwhich has been used for general pur- \nposes. This material has a resistivity \nsomewhat higher than 4% silicon steel \nand an initial permeability about ten \ntimes as great while its coercive force \nis only about one-tenth as large. While \nit is relatively expensive, it is largely \nused, in audio frequency transformers \nand in some carrier frequency coils. \nIn the form of thin tape a chromium \npermalloy has been used for the con- \ntinuous loading of a number of sub- \nmarine telegraph cables.\n\nIn distinction from the permalloys \nthere has been developed in the Lab- \noratories a material which, unlike the \npermalloys that saturate at relatively \nlow flux densities, does not reach \npractical saturation until the flux \ndensity reaches about 24,000 gausses. \nThis is an alloy of iron and cobalt in \napproximately equal proportions. To \naid in its fabrication 2% of vanadium \nhas been added. Its initial permea- \nbility is three or four times that of \nmagnetic iron while its maximum per- \nmeability is only about one-third as \nhigh; but above a flux density of \nabout 12,000 gausses, its permeability \nis about four times as great as that of \nmagnetic iron. The high cost of this \nmaterial has greatly limited its appli- \ncation, but it is finding a field in the \npole tips of electromagnets producing \nhighly concentrated magnetic forces.\n\nAnother development of the Lab- \noratories is \u2018\u2018perminvar\u201d, whose re- \nmarkable magnetic qualities bid fair \nto extend the usefulness of magnetic\n\nmaterials. This name has been applied \nto the alloys of iron, nickel and cobalt \nwhich when properly heat-treated are \ndistinguished by their constancy of \npermeability over a region of flux \ndensity from zero to about one thou- \nsand gausses. Since their permeability \nis substantially constant and their \nhysteresis loss small in this region, \nthe perminvars are superior to the \npermalloys, and to the other magnetic \nmaterials previously discussed, in re- \nspect to the amount of distortion pro- \nduced in transmitted speech currents. \nHence the use of coils with perminvar \ncores will result in distinctly superior \ntransmission circuit characteristics. \nThis is of especial importance in \ncarrier systems. A perminvar which, \nover the region mentioned, exhibits\n\ngreat constancy in permeability, con- \ntains 45% nickel, 25% cobalt, and \n30% iron. When properly heat-treated \nthis perminvar has a permeability of \nabout 400; this value is sensibly con- \nstant up to a flux density of 1000\n\ngausses. Other compositions have \nbeen found more desirable for specific \npurposes. For example, the perminvar \ncontaining 70% nickel, 7.5% cobalt, \nand 22. 5% | iron and designated er \n70 perminvar\u201d has a permeability of \nabout 800 and its constancy is limited \nby specification so that its permea- \nbility increase must be less than 1% \nover a range of flux density from zero \nto 500 gausses. One of the greatest \ndrawbacks at present to the wide ex- \ntension of the use of perminvar is its \npermanent loss of the perminvar \ncharacteristics if carried even mo- \nmentarily to flux densities of only a \nfew thousand gausses. Ordinary meth- \nods of demagnetization are only par- \ntially successful. Due to this char- \nacteristic and its high cost it has had \napplication to only a few coils in which \ndrastic reduction of modulation effects \nwas demanded.\n\nFor apparatus such as the loading \ncoils which are inserted in long tele- \nphone lines to improve transmission, \na core material is required which will \nintroduce very small transmission \nlosses and wave-form distortion. More \nspecifically, the hysteresis and eddy\n\ncurrent losses as well as the distorting \nmodulation effects must be small over \na wide range of magnetizing force. \nTo accomplish this it can be demon- \nstrated that the effective permeability \nof the loading coil core must be kept \nlow relative to the intrinsic permea- \nbility of the materials discussed in \nother sections of the paper. By finely \ndividing the core material and using \n80 permalloy which intrinsically has a \nlow hysteresis loss, coils peculiarly \nsuited to the loading of transmission \ncircuits may be produced. The fine \ndust particles are coated with a thin \ninsulating film and are compressed \ninto suitable ring cores under a pres- \nsure of about 200,000 pounds per \nsquare inch. The coils made from these \npermalloy dust cores have extremely \nsmall losses and their characteristics \ncontinue very constant in service. \nThese cores are also used in audio \nfrequency filter coils.\n\nBy the addition of certain elements \nto ordinary carbon tool steel, mate- \nrials of varying degrees of excellence \nfor permanent magnets have been pro- \nduced. One per cent of chromium \nadded to ordinary carbon steel in- \ncreases somewhat its coercive force \nand permits it to be hardened by \nquenching in oil instead of water. This \nis of considerable advantage in the \nelimination of quenching cracks. The \ncoercive force of this steel is usually \nbetween 40 and 50 oersteds. It is the \nleast expensive of the magnet steels \nand is largely used whenever weight \nis not of such importance as to pre- \nclude a magnet of sufficient length to \nretain its strength. This steel is large- \nly used in magneto-generators, coin \ncollectors and ringers.\n\nBy the addition of approximately \n4% instead of 1% of chromium a \nmagnet steel of still higher coercive \nforce is produced. Its coercive force\n\nis in the neighborhood of 70 oersteds.. \nThis is also an oil hardening steel and \nbecause of its freedom from quenching \ncracks and its lower cost it has largely \nreplaced the earlier 5% tungsten steel \nwhich is a water quenching steel and \nwhich has similar characteristics. This \nsteel is also largely used where the \nslightly higher cost is not a determin- \ning factor or where a somewhat \nshorter and consequently lighter mag- \nnet is demanded, for example, in \npolarized relays and some of the more \nrecently designed ringers.\n\nCobalt steel, invented by Professor \nK. Honda of Japan, marked a decided \nadvance in permanent magnet steels. \nIt contains about 35% cobalt, 7% or \n8% tungsten and 3% chromium. The \ncarbon content is about the same as \nthat in other magnet steels. The \ncoercive force of this steel is more than \nthree times that of the 4% chrome \nsteel and consequently it is very re- \nsistant to demagnetization. Magnets \nmade of it may be very short and still \nbe of satisfactory strength. It is em- \nployed extensively in receivers and \nother high class instruments. The \nonly drawback to its more extended \nuse is its cost which is some twelve or \nfifteen times that of the 4% chrome \nsteel.\n\nIn the processes of rolling or draw- \ning soft magnetic materials their \nstructure is strained and distorted, \nand this greatly decreases their per- \nmeability and increases their coercive \nforce. This condition may be relieved \nand the normal magnetic qualities \nobtained by a proper heat treatment \nwhich differs somewhat with different \nmaterials. In general, the process con- \nsists in heating them at the proper \ntemperature for a given time, sealed in \nmetal boxes which are then allowed to \ncool slowly. As might be expected, \nthe magnet steels require hardening\n\nby quenching in water or in oil from a \ndefinite temperature. \nIn the investigation of all these \nmagnetic materials there has been in \nthe past an immense amount of work \nwhich is the basis of our present de- \nsign information on magnetic appara- \ntus. All of the materials have been \nimproved in their properties, but the \nmost remarkable achievements have \nbeen in the development of the perm- \nalloys for high permeability and of\n\nA new day has thus come in design \nand a new field for improvement has \nopened before the designer with each \ndevelopment of magnetic materials, \nThe continuation of present investiga- \ntions which are relating more com- \npletely the physical condition and the \nchemical constituents of a material \nto its magnetic properties gives prorn- \nise of even greater developments in \nthe future.\n\nsubstance is known to be \nentirely impervious to it. The life of \nplants and animals may be said to \ndepend on the diffusion of water \nthrough their cell walls, and it is a \nfamiliar fact that the water which \npermeates the human body forms its \nlargest single constituent. Water \npasses through materials often con- \nsidered water-proof such as rubber, \nasphalts and waxes.\n\nIn the Bell System a number of \norganic materials are used as insula- \ntions which often are exposed to mois- \nture. Although the actual quantity \nof water which diffuses through most \norganic substances is small, this small \namount becomes important in con- \ndensers, coils and terminal boxes and \nother apparatus which must be pro- \ntected over a period of years.\n\ninsulations which are to be exposed to \nmoisture is the measurement of the \nquantity of water absorbed. Such \nmeasurements do not always give a \ncomplete picture of water-proofing \ncharacteristics, since a considerable \namount of water may diffuse through \na layer of material into some interior \nspace and only a small amount re- \nmain absorbed in the outer layer. \nDifferences in molecular structure, \nphase relationships, and crystalline \nstate may play important roles in \ndiffusion, and often the absorption of \nwater does not appear to be the prin- \ncipal controlling factor.\n\nOf the general methods for de- \ntermining the rate of diffusion of water \nthrough a layer of a substance, prob- \nably the most simple and direct is to \nestablish on one side a concentration \nof water vapor, maintain on the other \nside a lower concentration, and mea- \nsure the amount of water which passes\n\nFig. 1\u2014Water vapor passes from the interior of the aluminum \ncell through the rubber sample into a drying agent outside\n\nthrough the material in a given time. \nA convenient way to do this is to seal \na sheet of the material across a cup or \ncell in such a manner that moisture \npasses from the cell through the sam- \nple into a drying agent outside. For \nsuch measurements several types of \ncells are used in these \nLaboratories depend- \ning on the nature of \nthe material.\n\nFor rigid and semi- \nrigid substances like \nsoft and hard rubber, \nphenol fibre, cellulose \nderivatives, and many \nplastics, an aluminum \ncell is used in which \nthe sample is held in \nplace by an aluminum \npressure ring, the seal \nbeing completed by \nsuitable washers coated \nwith petrolatum (Fig- \nure I). The cell is \nplaced in a dessicator \ncontaining a drying \nagent and maintained \nat 25\u00b0 C. by immersion \nin a water bath (Fig- \nure 2). The loss in \nweight of the entire \ncell combination is \ntaken as the amount of \nwater which has dif- \nfused through. Mois-\n\nFor substances hav- \ning a tendency to cold \nflow, such as asphalts, \nthe cell shown in Fig- \nure 3 is used. The as- \nphalt is supported by \na thin alundum disc, \nand the drying agent \nis supported on a cali- \nbrated quartz spring whose elongation \nindicates the amount of water which \nhas passed through.\n\nA third type of cell is used for mate- \nrials having a very low diffusivity \nand which require an extremely tight \nseal. Such a cell may be constructed\n\nFig. 2\u2014The temperature is held constant during a diffusion \ntest, often by keeping the cells in a desiccator immersed in a\n\nby making a thick-walled wax cup \nacross the mouth of which a thin disc \nof the material to be measured may be \nsealed. The sealing is accomplished \nby melting together the edges of the \ncup and the disc, if the disc is of wax, \nor by melting the cup around the edge \nof the disc if the latter cannot easily \nbe melted. The water which passes \nthrough is collected by a drying agent \nattached to a calibrated quartz spring, \nas shown in the picture at the head of \nthis article. The temperature is kept \nconstant by placing the apparatus in an \nair bath or constant temperature room.\n\nMost materials obey quite closely \nthe linear diffusion law established by \nFick, and since the factors which enter \ninto linear diffusion are definite and \nmeasurable, the law is particularly \napplicable from an engineering stand- \npoint. It states substantially that the \namount of vapor diffusing through a \nunit area of a material in unit time is \nproportional to the difference in vapor \npressures on the two sides and in- \nversely proportional to the thickness. \nThe constant of proportionality, or dif- \nfusivity, constant depends principally \non the nature of the material. The \nlaw is somewhat analogous to Ohm\u2019s \nlaw; the differential vapor pressure \ncorresponds to the electromotive force, \nand the flow of water molecules driven \nby it is checked by the \u201cresistance\u201d \nof the material. Diffusivity is thus \nanalogous to conductance.\n\nFig. 3\u2014When a material with a tendency to \ncold-flow is placed in a diffusion cell, it is \nusually supported by an alundum disc\n\ncellulose acetate, if plotted on this chart would \nappear as a bar extending to 160\n\nThe diffusivity constant of a mat- \nerial is taken as the number of grams \nof water which pass through a one \ncentimeter cube in one hour under a \nvapor-pressure difference of one milli- \nmeter of mercury at a definite tem- \nperature.\n\nOrganic materials whose diffusivity \nconstantsat 25\u00b0C. have been measured \nin the Laboratories include soft and \nhard rubber compounds, submarine \ncable insulations, phenol fibre and \nsimilar materials, cellulose acetate and \nrelated substances, asphalts, and pure \nhydrocarbons such as_ polystyrene. \nThe amount of water which diffuses \nthrough a centimeter cube of material \nat 25\u00b0C. ranges between 10\u00b0 and 10\u00b0 \ngrams per hour when the difference in \nvapor pressures is one millimeter of \nmercury (Figure 4).\n\nMeasurements of this sort have \nproved especially valuable in dealing \nwith rubber, asphalts, and waxes,\n\nwhich are used extensively in the Bell \nSystem as insulating materials, pri- \nmarily to provide protection from \nwater. The amount of material \nneeded for protection has hitherto \nbeen determined largely empirically, \nbut a knowledge of the rate at which \nwater passes through an insulation \nnow makes it possible to estimate the \nthickness needed to protect a piece of \napparatus for a specified length of \ntime under known conditions of tem- \nperature and humidity. The mini- \nmum thickness of rubber sheath neces- \nsary to protect a dry-core paper cable \ncan be calculated, for example, and \nthe amount of material required in \nsealing apparatus with an organic \nsubstance can be estimated.\n\nThe mechanism by which water \ndiffuses through solid materials is not \ndefinitely known. There are some \ndata on the diffusion of gases through \ninorganic substances, however, which \nhelp to explain similar processes in \norganic materials.\n\nThere are indications that sorption* \nplays an important part in some cases \nof the diffusion of gases through \ninorganic materials. Hydrogen dif- \nfuses into copper only after it has first \nbeen adsorbed on the copper surface, \nand palladium sorbs hydrogen so \nstrongly as to indicate that they com- \nbine chemically. On the other hand, \ndiffusion independent of adsorption\n\nseems to be exemplified by the Passage \nof inert gases through fused silica \nwhere the gas molecules are believed \nto enter extremely minute cracks in \nthe film directly from the gas phase. \nThe diffusion of moisture through \ncertain organic substances such as \npolystyrene and hydrocarbon wax, is \nassociated with very little absorption. \nThe amount of water absorbed by \npolystyrene is extremely small, al- \nthough the diffusion of water through \nit is comparatively large. The water \nabsorption of the wax is also small, \nyet its diffusivity is only about one \nfortieth of that of the polystyrene. \nBoth materials are homogeneous, non- \npolar compounds. The difference be- \ntween their diffusivities appears to be \ndue to differences in their physical and \nchemical structure of some other sort \nthan those determining sorption. In \nsharp contrast is the case of cellulose \nacetate, through which moisture pas- \nses 1600 times as fast as through \nhydrocarbon wax. Here the sorption \nof water is strong and probably bor- \nders on chemical union, as does the \ninteraction of palladium with hydro- \ngen. It is very likely that the higher \nsorption affects the diffusivity, and \nthat both are largely accounted for by \nthe same features of chemical structure. \nTo arrive at a true picture of \nthe mechanics of moisture diffusion \nthrough organic substances much ad- \nditional work is required. The mea- \nsurements made in these Laboratories \nform only a beginning in this direction.\n\nPresident, American Telephone and Telegraph Company in charge of develop- \nment and research, and President of Bell Telephone Laboratories.\n\nIn Honor oF Dr. JEWETT, upon his \ncompletion of thirty years of service in the \nBell System, a luncheon was arranged in \nthe twelfth floor conference room on \nSeptember 13th, and attended by engi- \nneers who had been closely associated with \nhim on the projects for which he was \npersonally responsible in the early years \nof his career, and by executives and de- \npartment heads of the Laboratories.\n\nThose present were Messrs. Colpitts, \nCharlesworth, Norton, E. W. Adams, \nAnderegg, Bascom, J. H. Bell, Blackwell, \nBuckley, G. A. Campbell, Carson, A. B. \nClark, Curtis, Dixon, Englund, G. D. \nEdwards, Farrell, Ferris, Fletcher, Fondil-\n\nAlso Messrs. Haislip, Hartley, Ives, R. \nL. Jones, M. J. Kelly, Lowry, Mathes, \nMills, Moravec, Morehouse, W. H. Mar- \ntin, Molina, Parker, Quarles, Sanford, \nT. Shaw, Sheppard, G. B. Thomas, \nToomey, Warren, Webb, R. R. Williams, \nR. H. Wilson and W. Wilson.\n\nAbsence from New York prevented the \nattendance of Messrs. Espenschied, Heis- \ning, R. S. Hoyt, K. S. Johnson, Kendall, \nMatthies and J. G. Roberts.\n\nAt the conclusion of the luncheon a \npicture was taken of Dr. Jewett and those \nmembers of the present Bell Telephone \nLaboratories who had been associated\n\nDr. Fewett, surrounded by some of his early Boston associates, examines a cable loading\n\nE. H. Colpitts; Standing, Left to Right\u2014}. F. Toomey, E. C. Molina, Thomas Shaw, \nO. B. Blackwell, H. S. Warren and H. P. Charlesworth\n\nwith him in the Boston laboratories of the \nAmerican Bell Telephone Company.\n\nL. EspenscHIeD and R. A. as \nrepresentatives of the Bell System, sailed \nto attend the International Technical \nConsulting Committee on Radio Com- \nmunications (C.C.I.R.) in Estoril, Portu- \ngal, September 22 to October 10. This \ncommittee meets every two years to con- \nsider the progress of the radio art, both \nfrom the commercial and broadcasting \nstandpoints. It drafts technical recom- \nmendations for the international Tele- \ncommunications Conference which meets \nevery five years. The next international \nconference which will have treaty-making \npowers will be at Cairo in 1937. Among \nthe major questions to be considered at \nPortugal will be the allocation of fre- \nquencies to the different types of radio \nservices and means of reducing inter- \nnational interference.\n\nMessrs. Espenschied and Heising trav- \neled by way of London where they at- \ntended a Technical Session of the Inter- \nnational Scientific Radio Union (U.R.S.I.) \nheld September 11 to 19.\n\nDr. F. B. JeEwetr and Dr. W. A. \nSHEWHART were guests of Lt. Col. Stuart, \ncommandant of the Picatinny Arsenal, \nrecently. After an inspection of the re- \nsearch laboratories and manufacturing \nactivities of the Arsenal, Dr. Shewhart \ndiscussed with Col. Stuart the application \nof quality control techniques to the manu- \nfacture of munitions.\n\nA THREE-WAY discussion on new aids \nto education was recently broadcast over \nthe NBC network by Dr. Jewett, Dr. \nDearborn of New York University and \nWilliam D. Boutwell of the Office of Edu- \ncation of the United States Department \nof the Interior.\n\nPRoBLEMS concerning radio equipment \nhave taken F. M. Ryan, D. K. Martin, \nW. M. Knott, A. D. Liguori, J. G. Nordahl \nand W. A. Woods to Washington at\n\nvarious times to confer with members of \nthe Bureau of Engineering of the Navy \nDepartment.\n\nH. C. Atkinson, R. S. Barr and G. N. \nTHAYER accompanied the U.S. Fleet dur- \ning maneuvers off Newport, Rhode Island.\n\nW. L. Tierney visited radio stations \nKHJ and KFRC of the Don Lee Broad- \ncasting Company in Los Angeles and San\n\nR. S. Hoyt, one of the members of the present \nLaboratories who started in the Boston Lab- \noratories\n\nJ. C. Bain visited the Engineering and \nResearch Laboratories of the British Post \nOffice at Dollis Hill and \u201cBroadcasting \nHouse\u201d of the British Broadcasting Com- \npany, Ltd. in London.\n\nB. Leuve.ink visited the Northern \nElectric Company at Montreal to take \npart in conferences on radio broadcasting \ntransmitter design.\n\nG. Martejxa visited Wright Field at \nDayton and the Western Electric Com- \npany at Hawthorne in connection with \nradio equipment for the Army Air Corps.\n\nJ. F. Morrison completed a series of \nradio field intensity studies of Station \nWOR at Newark for the Bamberger \nBroadcasting Service, Inc.\n\nO. W. Towner supervised the instal- \nlation of an ultra-high frequency radio \nsystem for the Department of Public \nSafety of the City of Newark, New Jersey.\n\nJ. E. Rances was at Hawthorne on \nloading coil cases for use on loading pro- \ngram transmission circuits. He also in- \nspected other types of coil cases under \nconstruction.\n\nA. J. CorIsTOPHER visited Hawthorne to \ndiscuss the manufacture of induction coils.\n\nOn THE monp\u00e9Ay following the tragic \nburning of the Morro Castle of the Ward \nLine off the Jersey coast on September 8, \nmembers of the Laboratories learned with \nregret that among those lost was Miss M. \nF. Moran of the Apparatus Development \nDepartment. Miss Moran was returning \nfrom a vacation cruise with her mother \nwho was also lost. She came to the Lab- \noratories on August 29, 1917. In 1918 \nshe was made a laboratory assistant and \nin 1927 a technical assistant. Shortly \nafter coming with the Laboratories she \nwas concerned in the construction and \nadjustment of built-up condenser net- \nworks used in the Transcontinental Line \nand became very skillful in the operation \nof the capacity bridge. She continued\n\nthis type of work until 1926 when she \nengaged in measurements, computing and \nanalysis of test data in connection with \nlaboratory investigations of cords, wire \ninsulation and switchboard cable. Miss \nMoran took an active part in the Women\u2019s \nBridge Club.\n\nJoun J. Kuxyn, Manual Apparatus \nEngineer, received a five-star button on \nthe twenty-third of last month, emblem-\n\natic of twenty-five years of service in the \nBell System. Mr. Kuhn joined these Lab. \noratories in 1909 as a member of the \ndrafting division of the Apparatus Design \nDepartment, later transferring to the\n\nspecial order division. When the Western \nUnion Company was for a time affiliated \nwith the American Telephone and Tele- \ngraph Company, he became concerned \nwith the design of telegraph apparatus. \nThe advent of the war supplanted this \nactivity with the mechanical design of \ncommunication and signalling equipment \nfor the United States Government. At \nthe close of the war he retained super- \nvision of this work and also took charge of \nthe mechanical design of repeaters, rack \nmounted apparatus, telegraph equipment, \nticket distributing systems, and the like. \nIn 1927, when the Special Products De- \npartment was organized, he took charge \nof several groups responsible for the \nmechanical design of amplifiers, audi- \nphones, and government apparatus, and \ntwo years later his responsibilities wid- \nened to include the development of \nmechanisms employed in recording and \nreproducing sound pictures. Last year \nhe transferred to the Telephone Appara- \ntus Development Department, taking \ncharge of the group which had been super- \nvised by the late H. T. Martin. As \nManual Apparatus Engineer he is now \nresponsible for the design of coin col- \nlectors, telephone booths, desk and wall\n\nsets, plugs, jacks, keys, protectors, and a \nnumber of different types of maintenance \nand testing apparatus.\n\nA QUALITY SURVEY on filters and net- \nworks took H. Whittle, A. N. Jeffries and \nH. O. Wright to Kearny.\n\nB. O. TEMPLETON discussed problems \nconnected with the manufacture of coin \ncollectors with engineers at Hawthorne.\n\nA. K. Smit visited the New Jersey \nBell Telephone Company\u2019s toll office in \nNewark, New Jersey, in connection with \na service trial of apparatus for the pneu- \nmatic ticket distributing system.\n\nJ. R. TownseEnp inspected the experi- \nmental installations of lead alloy cable \nsheaths at Chester, Pennsylvania.\n\nC. H. Amapon spent nearly a month \ninvestigating the occurrence and extent of \ndecay in creosoted pine pole lines in \nTennessee and North Carolina.\n\nA. W. Drinc spent a week in Phila- \ndelphia and Point Breeze in connection \nwith a new protector for the Bell Tele- \nphone Company of Pennsylvania.\n\nA. W. CLeMeEnT presented a paper en- \ntitled Line Filter for Wide Band Open \nWire Program System at the Pacific Coast \nConvention of the American Institute of \nElectrical Engineers held in Salt Lake \nCity. D. H. Keyes also attended this \nconvention and presented a paper Foint \nUse of Poles with 6,900-Volt Lines of \nwhich he was joint author with W. R. \nBullard.\n\nC, Dorcan, who completed a \nquarter century of service in the Bell Sys- \ntem on the twenty-eighth of last month, \ngot his early telephone experience as in- \nspector and as repairman of private \nbranch exchanges for the Southern New \nEngland Telephone Company. At the \noutbreak of the war, after nine years of \nservice with that company, he left his \nposition of wire chief to engage with an \noutside concern in manufacturing and \ntesting telephone and signal apparatus for \nthe Army. In 1918 he returned to the \nBell System, this time joining the Western \nElectric Company, in its Engineering De-\n\nH. A. Larlee (left) talking over handset problems with F. L. Alden, development \nengineer, at Hawthorne\n\npartment which is now these Laboratories. \nHere he first took charge of installing \nradio telephones on the patrol boats pro- \ntecting New York harbor. On the com- \npletion of the war, he joined a newly\n\nformed group engaged in writing the \nmethod of operation of the early panel \ndial circuits. Later he engaged in relay \ntime studies, analyzing and testing the \ncircuits used in interconnecting the man- \nual and panel systems. Most recently he \nhas been engaged in laboratory work with \nthe carrier telephone group of the Toll \nSystems Department.\n\nH. S. Warren led a discussion on \ngrounding practices as embodied in Ar- \nticle 9 of the National Electrical Code \nat the Eastern Section of the Interna- \ntional Association of Electrical Inspectors \nheld at Washington.\n\nAN INVESTIGATION of shielding against \nlow frequency induction in cable circuits \nhas been undertaken by H. R. Moore, B. \nC. Griffith and J. H. Harding at Green- \nville, South Carolina. It is expected that \nthe investigation will take about three \nmonths.\n\nJ. C. Davenport, Jr., has been in \nWisconsin for several weeks in connection \nwith the work of Project Committee 2J of \nthe Joint subcommittee on development \nand research of the Edison Electric Insti- \ntute and the Bell System. In this work he \nis primarily interested in checking up on \nthe coupling between power and telephone\n\ncircuits to compare the estimated with \nthe observed values of induced voltage,\n\nJoun A. McIntyre received a seven. \nstar service emblem on the twenty-eighth \nof August, emblematic of thirty-five years \nof continuous service with the Bell Sys. \ntem. Mr. McIntyre joined the Ameri. \ncan Bell Telephone Company in 1899, \nentering its Boston Laboratory, a parent \nof these Laboratories. Four years later \nhe came to New York to take part in the \ninspection activities of the American Bell \nCompany at West Street, and continued \nin this work after it was taken over by the \nWestern Electric Company in 1907, \nWhen the Manufacturing Department \nand its associated inspection activities \nmoved to Hawthorne in October 1913, \nMr. McIntyre remained at West Street \nwith the Engineering Department, engag- \ning in the construction of transmitter and \nreceiver standards conducted by the Re- \nsearch Department. Five years ago he \ntransferred to the Apparatus Develop- \nment Department where he has since\n\nH. PFANNENSTIEHL was in Rochester in \nconnection with sound picture equipment.\n\nwhere he is engaged in inspecting the \nannouncing system for the U.S.S. Nevada.\n\nR. A. Mitter and W. Herriortr visited \nNela Park to discuss exciting lamps for \nsound picture machines.\n\nR. A. Mitier and H. C. Rusty discus- \nsed the design of a special synchronous \nclock system for the New York Telephone \nCompany with engineers of the Telechron \nCompany at Ashland, Massachusetts.\n\nH. C. Curt and R. L. Hanson were in \nNorfolk, where they made a sound survey \non the U.S.S. Ranger.\n\nD. G. BLaTrner was aboard the U.S.S. \nRanger in Norfolk, Virginia, for the in- \nstallation of new loud speaking apparatus.\n\nR. M. Burns visited the Naval Re- \nsearch Laboratory near Washington in \nconnection with storage battery develop- \nments.\n\nATTENUATION MEASUREMENTS on over- \nhead cables have been made at Ticon- \nderoga by H. H. Benning and M. S. \nBurgess. Recently S. Brand, A. G. Jen- \nsen, P. Mertz and T. Odarenko visited \nTiconderoga to confer with the engineers \ndoing the work.\n\nIn 1905 Cart H. Hircucock left the \nUniversity of Chicago, and the following \nyear joined the Western Electric Company \nat Clinton Street, beginning a period of \nservice in the Bell System which reached \nthe quarter-century mark on the second \nof August. After short periods with its \nInspection Department, and with the \nNational Packing Company, he came to \nNew York to take the student course of \nthe New York Telephone Company. In \n1910 he transferred to the Engineering \nDepartment of the Western Electric \nCompany, now these Laboratories, where \nhe engaged in inspection work at West \nStreet and at the plant of the Simplex \nWire and Cable Company in Boston. \nFollowing this he joined the lead covered \ncable development group at Hawthorne,\n\nlater shifting with a part of this group to \nthe Kearny plant, and has participated in \nmany of the cable development projects \nof this period, including that of the unit- \ntype cable which is now standardized.\n\nA sTupy OF interference at high fre- \nquencies took E. I. Green, F. A. Liebe, \nR. G. McCurdy, T. Odarenko, R. A. \nShetzline and M. E. Strieby to Phoenix- \nville, Pa., where Mr. Odarenko has been \ninvestigating and measuring radio inter- \nference and static on open wire lines and \ncarrier cables.\n\nM. E. Srriesy visited the Point Breeze \nWorks to inspect high frequency cables.\n\nJ. R. Witson and V. L. Ronct visited \nthe Corning Glass Company, on prob- \nlems of glass development.\n\nC. H. Prescotr and J. O. McNa.ty \nwere in New Haven for the field trial of a \nnew repeater tube.\n\nJ. A. Becker visited the Westinghouse \nElectric and Manufacturing Company \nand the Union Switch and Signal Com- \npany in Pittsburgh, to discuss with the \nengineers of these companies the prob- \nlems arising in the manufacture of copper \noxide rectifiers.\n\nALTHOUGH RoceEr G. RamsDELL, who \nreceived a five-star service emblem on \nthe first of August, has been with the \nBell System for twenty-five years, his \nmembership in these Laboratories is only \na few months old. Mr. Ramsdell received \nfrom the University of Vermont the B.S.\n\ndegree in Electrical Engineering in 1909 \nand immediately came to New York to \njoin the Engineering Department of the \nAmerican Telephone and Telegraph Com- \npany, then headed by the late John J. \nCarty and located at 15 Dey Street. In \n1919, when the Department of Develop- \nment and Research was formed, he trans- \nferred to its equipment development sec- \ntion with which he remained connected \nuntil the consolidation of the D. & R. \nDepartment with these Laboratories last \nspring. He has been principally concerned \nwith the development of central office \napparatus, more particularly of switch- \nboard jacks, plugs and cords, and the \ntools and gauges required in the field to\n\nmaintain this apparatus. He is con- \ntinuing this work as a member of the \nLocal Facilities group of the Systems De- \nvelopment Department.\n\nAvotpuH R. Swosopa received the B.S. \ndegree from the University of Nebraska \nin 1903. He immediately joined the \nStudent Course of the Western Electric \nCompany at Clinton Street and at its \nconclusion came to West Street as a \nswitchboard inspector, soon taking up the \ndesign of power plants for central offices. \nReturning to the University he obtained \nthe E.E. degree in 1907 and after a four \nyear period with the Kellogg Switchboard \nand Supply Company rejoined the West- \nern Electric Engineering Department at \nWest Street. The bridging of this gap in\n\nservice made Mr. Swoboda eligible for a \nfive-star service emblem on the fifth of \nAugust. An early connection with the \nPhysical Laboratory was followed by \nwork with groups developing public\n\nK. K. Darrow presented a paper en- \ntitled Nuclear Chemistry before the Chem- \nical Education Division of the American \nChemical Society at Cleveland.\n\nJ. A. Manoney returned on September \n4 from Australia, where he cooperated in \nfurnishing technical information to the \nAustralian Post Office Department on a \nproposed Australia-Tasmania submarine \ncable to provide telephone, telegraph and \nradio broadcast program channels from \nMelbourne, Australia to Launceston and \nHobart, Tasmania.\n\nV. T. CALLAHAN visited the Sturtevant \nCompany at Boston where he conferred \nwith engineers on tests of ventilating equip- \nment for an exchange power installation.\n\nF. S. Enrz spent a week in Princeton, \nand Stroudsburg, Pennsylvania, inspect- \ning the pilot-wire regulating system.\n\n1. G. Witson visited Atlanta, Green- \nville, and Charlotte arranging for the \nreturn from these stations to the Lab- \noratories of the experimental models of the \nnew carrier telephone repeater using a \ntwo-way single amplifier. A service trial \nduring the past year was recently suc- \ncessfully completed.\n\nE. C. Mouina attended the summer \nmeeting of the American Mathematical \nSociety at Williams College the week of \nSeptember 3.\n\nC. S. Demarest and W. J. Heirsmiru \nwere present at the cutover of the step- \nby-step and toll central office in Roanoke, \nVirginia. The power plant for this \ncentral office is the initial installation of \nthe full-automatic type.\n\nO. H. Loynes recently completed the \nlaw course of the LaSalle Extension Uni- \nversity and was awarded its LL.B. degree.\n\nWuitney visited Columbus and \nDes Moines for the purpose of discussing \ncommunity dial terminations in toll \noffices.\n\nWuen R. Barney was grad- \nuated from Cornell University thirty \nyears ago, he immediately entered the \nClinton Street Works of the Western \nElectric Company, and joined the Stu- \ndent Course. After the survey of the \nvarious branches of the Company\u2019s activ- \nity afforded by that course, he was asso- \nciated with the Engineering Department, \ntaking part first in the equipment branch \nand then in the inspection of manufac- \ntured products. After the depression of \n1907 he joined the sales organization to \nhelp dispose of overstocks by converting \nthem into other types of apparatus or \nequipment. Later he served as the Haw- \nthorne representative on matters per- \ntaining to new and changed apparatus, \nfirst for the Sales Department and later \nfor the Merchandise Department. In \n1913 he came to New York as a member \nof the Advertising Department to take\n\ncharge of the work on the card catalog \nwhich still occupies him. When the Ad- \nvertising Department was transferred to \n195 Broadway, he remained at West \nStreet where as a member of the Lab- \noratories he continues the task of main- \ntaining a card-catalog file of up-to-date \ninformation regarding all coded apparatus \nwhich the Western Electric Company \nmanufactures for sale to its customers.\n\nOn Avucust 11 Alfred C. Merriam, \nbetter known among his many friends as \n\u201cJerry\u201d, joined the little band of those \nwho have served the Bell System for forty \nyears or more. Mr. Merriam entered \nthe toll rate department of the American \nTelephone and Telegraph Company in \n1894. Six years later he went to the \nPurchasing Department of the Southern \nBell Telephone and Telegraph Company, \nand when its headquarters moved further \nsouth in the fall of 1901, he transferred to \nthe Cost Department of the Western \nElectric Company. In 1913 he became \nsupervisor of the Payroll Department at \nWest Street. The following year the \nmanufacturing organization moved to \nHawthorne, and Mr. Merriam moved \nwith it, as a supervisor in the Cost De- \npartment.\n\nIn the Winter of 1917 he returned to \nWest Street, to take charge of a cost group \nin the Engineering Department, now these \nLaboratories. Later he supervised the\n\nShop, and in 1929 rejoined the Account- \ning Department at West Street. His \nfamiliarity with Tube Shop activities \nbrought him again to Hudson Street to \nhandle accounting work there, until the \ntransfer of tube manufacture to the West-\n\nern Electric Company last spring, when \nhe returned to the expense accounting \ngroup here.\n\nTHE TRIAL installation of a new grid- \nbattery supply unit for voice-frequency \ncarrier telegraph systems took V. P. \nThorp to Pittsburgh.\n\nA. H. Incuis has been in Chicago two \nweeks studying the effect of reducing \nnoise in the operating rooms of Chicago \ncentral offices.\n\nE. S. Witcox has completed the check- \ning of transpositions on the Fort-Scott- \nWichita line and is now making cross- \ntalk measurements at 100 kilocycles.\n\nE. D. Guernsey recently completed \nnoise and balance measurements on the \nJoplin-Tulsa line and is preparing to make \nsimilar tests on the Fort Scott-Wichita \nline.\n\nA. D. Fow er has been in Cleveland \nfor several weeks supervising and testing \nequipment in the toll office associated with \nstandard frequency service which is being \ninitiated on a trial basis for the Cleve- \nland Electric Illuminating Company.\n\nA FIELD TRIAL of narrow open wire \nspacing which is being conducted on a \ntwenty mile section of the New York-\n\nScranton line took R. S. Alford, J. A, \nCarr, R. C. Edson and J. W. Kennard to \nMount Pocono, Pennsylvania.\n\nE. P. FELcn and H. S. WIinBIGLER have \ngone to Kansas City to prepare for delay \nmeasurements on telephotograph circuits \nin cooperation with engineers of the Long \nLines Department. Similar work is being \ndone in New York City by F. R. Dennis, \nR. S. Hawkins and P. F. Jones with an \nengineer of the Long Lines Department. \nK. W. Pfleger is coordinating the work of \nboth groups.\n\nSHORT RADIO WAVES, used in inter. \nnational transoceanic communication, are \nvariable in their effects and greatly in- \nfluenced by magnetic storms, which in \nturn appear to be due to solar activity. \nThe Azores have been chosen as a con- \nvenient point from which to make ob- \nservations on the effect of such storms, \nparticularly their influence on the short \nwaves in transit between America and \nEurope. C. F. Edwards left August 11 on \nthe S. S. Saturnia, and took with him \nsuitable measuring apparatus for the pur- \npose of making such observations over a \nthree or four months\u2019 period. This is \nbeing done with the knowledge and con- \nsent of the Portugese administration \nwhich is, of course, interested in radio \ntransmission.\n\nANDREW ALLAN, a retired member of \nthe Laboratories, died on September 8, \n1934 following an automobile accident. \nMr. Allan, whose service dated from April \n18, 1910, was formerly in the Plant De- \npartment and was retired on September \nI, 1932.\n\nS. B. Kent attended a hearing before \nthe Examiner of Interferences in Wash- \nington.\n\nR, J. FLuskey received an E.E. degree \nfrom New York University, his alma \nmater, last June.\n\nTue Lasorarories learned with regret \nof the death of Lewis Kubic on August 26. \nOn April 26 Mr. Kubic had completed \ntwenty-four years\u2019 service with the Lab- \noratories. During that time he was\n\nemployed as an instrument maker in the \ndevelopment shop of the Plant Depart- \nment. During the last few years he had \nspent considerable time in making dials \nand other machine switching apparatus.\n\nAn EcHo of the Laboratories\u2019 exhibit at \nCambridge last Winter* was an editorial \ndescription of the magnetic speech re- \ncorder in the Boston \u2018\u201c\u2018Record\u201d recently. \nUnder the caption \u201cPark Your Voice\u201d the \neditor imagined several forms of domestic \nand public discourse which might well be \n\u201cparked\u201d indefinitely. Developing that \nidea, he suggested to angry people, \u201cPark \nyour voice within yourself\u2014until you are \nconvinced that it should be released\u201d\u2019.\n\nPRomPTED BY reading H. A. Frederick\u2019s \narticle, \u201cEarly Handsets\u201d, in the REcorp \nfor last July, Robert G. Brown, inventor \nof the first commercial handset, paid the \nLaboratories a visit last month. Mr. \nBrown was especially interested in the \nhistorical museum, through which its \ncurator, W. C. F. Farnell, guided him, \nrecognizing there not only the handset \nmodels directly following his own but \nmany types of central office equipment \nwith which he was concerned as chief \noperator in the 1870\u2019s of the first exchange \nof the Gold and Stock Telegraph Company.\n\nB. Simon completed twenty \nyears of service in the Bell System on the \nseventeenth of last month.\n\nfive years of service with the Western \nElectric Company and these Laboratories. \nMr. Broadwell went to Hawthorne im- \nmediately on receiving the degree of B.S. \nin Electrical Engineering from the Uni- \nversity of Pennsylvania. After a year in \nthe Student Course there, he entered a \nbranch of the apparatus design depart- \nment then located at Hawthorne, and in \n1912 this work brought him to New York. \nIn 1915 he transferred to the dial appa- \nratus group. During the war he was \nconcerned with vacuum tube develop- \nment, but later he reengaged in the \ndesign of various units for dial and toll \ntelephone systems. Among these were a \nseries of motor-driven interrupters for test \nand service in dial circuits, and motor- \ndriven transmission regulating equipment \nfor controlling the effects of temperature\n\nTHE FALL SEASON of the Men\u2019s Bridge \nClub is being opened with a Mitchell \ntournament for pairs, similar tournaments \nbeing held weekly until Christmas. The \nclub has also entered a twelve-man team \nin the newly organized Bell System \nBridge League consisting of teams from \nthe A. T. & T. Company, Electrical Re- \nsearch Products, Inc., Western Electric \nCompany and the New York Telephone\n\nCompany. The Laboratories has again \nentered the Metropolitan Commercial \nBridge League with another twelve-man \nteam to play matches bi-weekly.\n\nThe Women\u2019s Bridge Club is also open- \ning its fall season this month with pro- \ngressive contract and duplicate contract \ntournaments held on alternate Mondays.\n\nDuRING THE MONTH of August patents \nwere issued to the following members of\n\nH. F. Smith at the land transmitter testing the effectiveness \nof ultra-short waves for communication with vessels in \nNew York Harbor\n\nATISFACTORY transmission \nrequires, among other things, \nthat the overall loss of the tele- \nphone circuit must remain reasonably \nconstant over long intervals of time. \nThis constancy of transmission loss, \nunfortunately, is not an inherent prop- \nerty of most transmission lines that \nare in use, and so it has been found \ndesirable on the longer lines to provide \nmeans for the regulation of transmis- \nsion loss within suitable limits.\n\nThe attenuation of open-wire line \nconductors increases with frequency. \nThe rate of increase is proportional to \nthe a-c resistance of the pair and to the \nconductance. Both of these factors \nincrease with frequency and as a re- \nsult the transmission loss does so also. \nAt the higher frequencies, which are \nused for carrier transmission, the loss \nis considerably greater than at voice \nfrequencies. Of more importance, \nhowever, is the fact that both the re- \nsistance of the pair and the conduc- \ntance between wires do not remain \nconstant but vary considerably from \ntime to time with the prevailing \nweather conditions, and the variation \nin attenuation is of sufficient magni- \ntude to require correction.\n\nThis variation in transmission loss \nis due to two causes. First\u2019 and most \nimportant is the change of leakage at \nthe insulators. Of secondary impor- \ntance is the change of conductor resis- \ntance with temperature. The former \nis peculiar to the open-wire form of\n\nline construction in which the insula- \ntors are fully exposed to weather, \nwhile the latter effect is present in all \ncopper line conductors. Already im- \nportant strides have been taken to \nreduce the change of insulator leakage \nwith weather by the design of im- \nproved types of insulators, but con- \nsiderable remains to be done before \nthis source of variation can be elim- \ninated.\n\nOccasionally, the insulator problem \nis aggravated by unusual weather con- \nditions which are sometimes encoun- \ntered in certain parts of the country. \nThese conditions may be caused by \nsleet building up a wet ice coating \nover the insulators, thus forming a \nwet path between the wires at each \ncrossarm. Another serious condition \nencountered in small areas in the far \nwest is the building up of a frost coat- \ning on wires and insulators. This \ncoating sometimes becomes several \ninches thick and increases the line loss \nper unit length tremendously. Still \nanother is the effect of sudden fogs in \nregions such as the part of the central \ntranscontinental line normally subject \nto salt dust deposits on the surfaces of \ninsulators. The dampening of this \nsalt dust by the fog on the insulator \nsurfaces causes a large increase of \nleakage and consequently of trans- \nmission loss.\n\nWhile these unusual effects are \ninteresting and now and then quite \nserious, the variations of transmission\n\nthe attenuation change \n0.20 from the wet to dry \nas iw condition is much more \n2 ~ wa important than the va- \n0.16 riation due to tempera- \nj \nture. Figure 2 shows \n| a record of attenuation \nx . \n\u00a9 0.12 sect oft taken at a single fre- \n4 . \n. ne] quency on a line 150 \n\u00b0 = miles long, equipped \n0.08 with S.P. insulators, \nBas and gives a continuous \n| ween wet AND ORY | measurement for a 36- \n< 004 ERENCE hour period. The \nweather conditions at \nDIFFERENCE FOR TEMPERATURE CHANGE OF 25F one end of the line are \nT I I I I\n\nloss even with ordinary change of \nweather are sufficient to present a real \nproblem, especially in the longer sys- \ntems. To illustrate the amount of \nthese variations, Figure 1 shows the \nattenuation of an open-wire pair in \ndb per mile at different frequencies in \nthe range employed for carrier trans- \nmission. One curve, the highest in the \nfigure, shows the average attenuation \nduring wet weather. The next lower \ncurve is the same characteristic during \naverage dry weather. It will be noted \nthat not only the transmission loss of \nthe line increases with frequency but \nalso the change in loss with the change \nfrom wet to dry weather increases \nwith frequency. This differential is \nplotted below the dry-weather atten- \nuation curve. At the bottom oi the \nillustration is shown the change in \nattenuation with temperature for a \nrepresentative day. The maximum \nchange per day throughout a period \nof a year might amount to twice this \nin some localities.\n\nnoted along the time \nscale. Evidently the \nattenuation varies al- \nmost continuously and \nsometimes quite rapid- \nly. In this case the total range of varia- \ntion is nearly 4 db, which occurs within \na 12-hour interval. To show how this \nrange of variation compares with the \naverage curves of Figure 1, the broad \nblack vertical line at 25 kilocycles be- \ntween the wet and dry curves has been \ndrawn in after being reduced to the \nvariation per mile. During the period \nshown in Figure 2, the line was not \nentirely dry nor did it show as large \nan increase in attenuation during wet \nweather as may frequently occur. The \nS.P. insulators used on this line are an \nold type and show about three times \nthe normal variation with weather \nchanges exhibited by the modern C.S. \ntype which are now standard for new \nconstruction.\n\nThe length of line measured to pro- \nduce Figure 2 was about average for \nthe spacing of repeaters on a type-C \ncarrier system. Evidently for a carrier \nsystem operating over many sections \nof line in tandem, averaging 150 miles \nper section, the problem of varying \nline attenuation is important. There\n\nare a number of carrier systems em- \nploying 10 or more of such sections. \nIf the variations shown in Figure 2 \nwere multiplied by 10, it would \namount to 39 db for the whole line. \nFortunately, these line sections, when \nconnected together, extend over such \nwide areas that it is very unlikely the \nsame weather conditions will prevail \nalong the entire line and this reduces \nthe probability of obtaining large \nvariations of total line attenuation. It \nis the reason, however, why a more \ncontinuously and rapidly varying at- \ntenuation can be expected on long \ncarrier systems, and makes more im- \nportant the matter of providing suit- \nable level indicators for the transmis- \nsion maintenance forces so that proper \nlevel corrections can be made.\n\nOf course, there are other sources of \ntransmission variations which may \noccur in the repeaters and terminals, \nbut these can and have been made a \nsmall part of the total by suitable \ndesign and maintenance.\n\nsand miles or more in length have been \nplaced in service, an indicator of level \nfor maintenance purposes has been \nemployed. Nearly all type-C carrier \nsystems have been equipped with this \nindicator, the so-called 1A pilot chan- \nnel. A single-frequency pilot current \nis located between channel bands and \nas near the central part of the trans- \nmitted range of carrier channels as \npossible. To measure this pilot fre- \nquency properly, tuned-rectifier me- \nters are provided at all intermediate \ncarrier repeaters and at the receiving \nterminals. These meters are cali- \nbrated in db and are adjusted to \nindicate \u2018\u2018 Normal\u201d level when normal \nlevels are transmitted on all channels \nincluding the pilot. Deviations of \nlevel can be noted and corrected man- \nually by the amplifier gain-control \npotentiometers. Periodical observa- \ntions of the meter readings at each \noffice are required as part of the reg- \nular maintenance procedure. At re- \nceiving terminals a sensitive marginal \nrelay is connected in series with the\n\nFig. 2\u2014Variation of transmission loss of a section of open-wire line at 25 \nkilocycles due to weather changes\n\nindicator meter to give an alarm when \nthe pilot deviates too far from normal. \nBy this means the maintenance per- \nsonnel at the receiving terminal is \nnotified as soon as a major deviation \nof the pilot level has occurred. Ina \nlong system, after an alarm has been \nsounded, it is not known where the \nfault is located, and to avoid making \nthe gain correction at the wrong \npoint, it is necessary to get in touch by \nan order wire with each repeater office \nto determine where a gain correction \nshould be made, and then to authorize \nthe proper adjustment. This does not \ncomplete the adjustment. It merely \ninsures that the repeater gains are \ncorrect at the pilot-channel frequency.\n\nReferring back to Figure 1, one may \nnote that the variation of line loss \nis different at different frequencies. \nThe gain correction by means of the \nrepeater-amplifier potentiometer, on \nthe other hand, is uniform at all fre- \nquencies. This gain adjustment, there- \nfore, is correct only at the pilot fre- \nquency and leaves small differentials \nto be added or subtracted at adjacent \nchannel frequencies.\n\nWhen the change of line loss is \nlarge and requires a number of con- \nsecutive steps from time to time, the \nerror at adjacent channel frequencies \nwill also become large. To avoid ac-\n\ncumulation of excessive errors from \nthis cause, the receiving terminal \nmaintenance forces are required to \nkeep accurate account of the steps of \ngain correction which have to be made \nat all repeaters in the system, and \nwhen these have accumulated suff_ \nciently to justify correction of the \nequivalents of the individual channels, \nthe corrections are made. This can \nbe done by manipulating potentio- \nmeters at the individual channel re- \nceiving terminals. <A table is em- \nployed for the corrections of this \nkind, which is based on the attenua- \ntion differential curve given in Fig- \nure I.\n\nVery fortunately, this curve has \npractically the same shape for all \ngauges of line conductors and types \nof insulators, so the same basic table \ncan be applied to all open-wire sys- \ntems of the same frequency allocation \nregardless of the facilities over which \nthe systems operate. Of course, the \ncorrection in the individual channel \ndepends upon the frequency of the \npilot and the frequency of that chan- \nnel, so the use of different frequency \nallocations calls for different tables.\n\nThis transmission maintenance pro- \ncedure is open to the objections that \nit is complex and requires much time \nto carry out, and that the smaller \nvariations of trans-\n\n8 \nCHANNEL NO.2 \n\\ lected. Under cer- \ntain conditions these \n5 prot {NORMAL small variations can add \nthan is sometimes. al- \n\u201c9 lowable. Difficulties \nwith the older system \nsuch as this have led to\n\npanion article.* In this system, des- \nignated as the 2A pilot channel, the \ntable of corrections has been replaced \nby networks having the basic char- \nacteristic shown in Figure 1.\n\nAutomatic control has the advan- \ntage that it makes adjustmentsin much \nsmaller steps and makes communica- \ntion between offices for the mainte- \nnance of levels unnecessary. It also \nprovides level correction at all fre- \nquencies at all repeater offices as well \nas terminals and this is an important \nadvantage for the longer systems. \nRecords of the operation of systems \nusing 1A manual pilot-channel control \nindicate that the maintenance of chan- \nnel equivalents is ordinarily within the \n*On the following page.\n\nrange of about 314 db either way from \nnormal for any channel of a system. \nWith the automatic control, this range \nis reduced to 2 db, and generally a \nmuch better figure even than this can \nbe realized.\n\nFigure 3 shows the variation in loss \non two channels of the Washington- \nWest Palm Beach type-C system on \nwhich the trial of the automatic pilot \nchannel was made. The recorded net \nloss of channels 2 and 3 are shown to- \ngether with the pilot level which was \nautomatically adjusted as soon as that \nlevel departed more than 44 db from \nnormal. Evidently a much better \ngrade of transmission maintenance \ncan be obtained with the use of the \nautomatic pilot regulator.\n\nHysteresis Loop with superposed incremental loops taken on a ring specimen of 45 per- \nmalloy. Curves were made with Haworth Magnetic Curve Tracer. At intervals in the \nprocess of tracing the hysteresis loop the rate of change of the magnetizing force was re- \nversed for a predetermined increment and then the forward tracing of the loop continued\n\nfrom day to day for a variety \nof reasons, as discussed in an accom- \npanying article*. Since the variations \nat carrier frequencies if uncorrected \nmay be of sufficient magnitude to \nseriously impair transmission, manual \nadjustments\u2014by aid of a device called \nthe 1A pilot channel\u2014have generally \nbeen required for the longer type-C \ncarrier systems. Although manual \nadjustment has given satisfactory re- \n*Page 49.\n\nsults, considerable improvement, as \nwell as some economies in operation, \nseemed to be possible if an automatic \npilot-controlled system were devel- \noped. As a result the 2A pilot system \nhas been made available, which auto- \nmatically adjusts the gain at each \ncarrier repeater station and at the \nterminal at fifteen second intervals.\n\nThe system is like the 1A in that it \ndepends on the transmission of a \nsmall amount of high-frequency pilot \ncurrent over the line from terminal to \nterminal. At each station a portion of\n\nthis pilot current is selected by a \ntuned circuit, and its level is measured \nby a control meter calibrated in db \nbelow and above normal level. If the \nlevel departs as much as 14 db from \nnormal, a contact is made and a group \nof relays and selectors act to adjust \nthe gain. Every fifteen seconds the\n\na bell when there is any abnormal \nchange in level of as much as 114 db. \nA delay is incorporated in the circuit \nso that the alarm is given only if the \nabnormal change persists for thirty \nseconds. This avoids having the alarm \noperate on momentary transients.\n\nFREQUENCY IN KILOCYCLES PER SECOND \nFig. 1\u2014Position of bands, carriers, and pilot currents for the CS carrier system\n\ncontrol circuit operates the necessary \napparatus to insert in or remove from \nthe circuit attenuating networks in \nquarter db steps if either of the con- \ntacts in the indicating controller has \nclosed.\n\nBesides the control circuit that \nchanges the gain whenever it deviates \nas much as 14 db from normal, there \nis at the receiving terminal an alarm \ncircuit that lights a lamp and rings\n\none-way carrier telephone channels \nare provided; the three lower fre- \nquency channels carry speech in one \ndirection, east to west, and the three \nhigher ones, in the other direction, \nwest to east. There is one pilot fre- \nquency for each of the two bands \nlocated about fifty cycles from the \ncarrier of the middle channel. The loca- \ntion of the six channels, their carriers, \nand the two pilot frequencies for the\n\nL | \nBAND | TRANSMITTING \nFILTER | AMPLIFIER \nHIGH-PASS \nFILTER \n= \nBAND | | RECEIVING \nFILTER | AMPLIFIER \u2014 \nREGULATING \nNETWORK DIRECTIONAL \nDEMODULATOR | FILTER \nTO\"DEMODULATOR | \nBAND FILTERS OF | PILOT\u201c conrroL | EQUALIZER \nOTHER CHANNELS | --__] INDICATOR EQuiPMENT \n| VOICE CURRENT --\n\nFig. 2\u2014Schematic of terminal showing location of 2A pilot channel equipment\n\nCS system are diagrammed in Figure 1. \nThe equipment required for the two \npilot channels, and its location in the \ncircuit, both at a terminal and at a \nrepeater station, is shown in the block \nschematics of Figures 2 and 3. At the \nterminal station the pilot oscillator \nis located in the transmitting side \njust ahead of the transmitting ampli- \nfier. In the receiving side are the pilot \nindicator, the pilot alarm, the control \nequipment, and the regulating net- \nwork. At intermediate stations there \nare in each branch a pilot indicator, \nthe control equipment, and the regu- \nlating network. The pilot alarm at \nintermediate stations is optional.\n\nThe oscillator, a schematic of which \nis shown in Figure 4, must be very \nstable in frequency and of constant \noutput. It may be tuned to the de- \nsired location between channels, and \nthe output is set to the proper value \nby an adjustable resistance network. \nThe pilot indicator circuit, shown \nschematically in Figure 5, has a high- \nimpedance input circuit which is \nbridged across the output of the re- \nceiving amplifier. Since the pilot\n\ncurrent is between adjacent channel \nbands, the indicator circuit must be \nsufficiently selective to prevent serious \ninterference into the pilot channel\n\nfrom speech on the adjacent telephone \nchannels. This is accomplished by \ntwo loosely coupled tuned circuits be- \ntween the transformer and vacuum \ntube. As shown in Figures 2 and 3, \nthe indicator circuit is tapped on the \noutput side of the main amplifier \nwhile the regulating network is con- \nnected in the circuit on the input side \nof the amplifier. Any change made\n\nPILOT | \nINDICATOR | \n| \nCONTROL | | \nEQUIPMENT | \n| \n| | \nHIGH-PASS HIGH-PASS \nEQUALIZER | WEST-EAST FILTER \nDIRECTIONAL |REGULATING| AMPLIFIER DIRECTIONAL \nFILTERS EAST-WEST NETWORKS FILTERS \nAMPLIFIER EQUALIZER | \n| | \n| | \n| CONTROL \n| EQUIPMENT \nLOW-PASS | LOW-PASS \nFILTER | FILTER \nt PILOT \n|INDICATOR\n\nFig. 3\u2014Schematic of intermediate station showing location of 2A pilot channel equipment\n\nin the network as a result of the action \nof the indicating controller will there- \nfore act immediately to bring the in- \ndicator back to the proper reading.\n\nThe regulating network is composed \nof eight separate units. Four of them \neach produce a nominal loss of four \ndb, one of two db, one of 1 db, one of \n14 db, and one of 4 db. They are \nconnected in the circuit in series and \nthe control equipment acts to remove \nfrom the circuit those that are not \nneeded. The attenuation units, with \nthe exception of the two smaller sizes, \nintroduce losses that simulate an \nactual length of open line\u2014the loss \nincreasing with frequency. The regu-\n\nlating network thus builds out the \nline attenuation to a fixed value for \nany weather condition. This fixed \nattenuation (line plus regulating net- \nwork), increasing with frequency, is \nthen closely equalized to an overall \nloss-frequency characteristic that is \nuniform with frequency by a base \nequalizer network which is normally a \npart of the repeater. This attenuation \ncorrection with frequency equaliza- \ntion is illustrated in Figure 6.\n\nThe control equipment consists of \nthree rotary type selectors, designated \nA, B, and C, a group of eight relays \nto cut in or out the units of the regu- \nlating network, an indicating con- \ntroller, and certain other auxiliary\n\nrelays. A simplified schematic show- \ning the regulating network, the in- \ndicator, and control equipment in \ntheir relation to the amplifier and line \ncircuits is given in Figure 7. In the \nindicating controller on the right are \nshown the high and low contacts, one \nor the other of which is closed by the \nmeter every 15 seconds if the level \ndiffers from normal by as much as 4% \ndb. A third switch, marked trigger, \nis also closed every fifteen seconds, \n10 seconds after the operation of the \nlevel meter, and completes the change \nin the attenuation that has _ been \nshown necessary by the previous oper- \nation of either the high or low contact. \nIf neither the high nor low contact \nhas closed, the subsequent trigger \noperation produces no change in at- \ntenuation, but if one of them has \nclosed, the circuit acts to increase or \ndecrease the loss introduced by the \nnetwork by 14 db. This sequence is \nrepeated every fifteen seconds and the \ncorrection, if required, is always made \nin 14 db steps.\n\nThe B selector controls the opera- \ntion of the relays that cut in or out \nthe 1, %, and 14 db units. As the \nselector is moved from positions I to 8\n\ninclusive, with intervening trigger con- \ntacts, the total loss introduced by \nthese units increases from 0 to 134 db \nin quarter db steps. The C selector \non the other hand, controls the relays \nthat cut in or out the 2 and 4 db units, \nand as it is moved from steps I to Io \ninclusive, with intervening trigger \ncontacts, the loss inserted increases \nfrom o to 18 db in 2 db steps. By \nselecting the proper combination of \nsteps of these two selectors, therefore, \nany loss from 0 to 1934 db may be \nobtained. The A selector is employed \nonly when effective backward stepping \nof the B or C selector is required in \nreducing attenuation by a 4 db step. \nSince the selectors will operate in only \none direction, the A selector, making \none revolution, acts as a master \nselector to give the right number of \npulses or steps to the B or C selector \nto step it forward one-half revolution \nminus one step. Then when the trig- \nger contact operates, the attenuation \nwill be reduced by 14 db.\n\ntime there is 8 db inserted and the \nhigh contact of the controller closses, \nthe B selector will move forward one \nstep. Then at the end of the 15 second \ninterval when the trigger switch oper- \nates, an additional 14 db loss will be \ninserted in the circuit. Had the low \ninstead of the high contact closed, the \nB selector would step under self-inter- \nruptions from the o db position to the \n134 db position, while the C selector, \nreceiving pulses from the A selector, \nwould have been moved to reduce the \nloss by 2 db. As a result a net change \nof\u201414 db would be brought about \nwhen the trigger switch operates.\n\nThe control circuit is also arranged \nto give an alarm when the pilot cur- \nrent suddenly becomes excessively low, \nas would happen from an open or a \nshort on the line, and also on failure of \nthe automatic control. Provision is \nalso made for manual operation in \nemergencies. With the 2A carrier \npilot control it is possible to hold the \noverall loss between terminals to with- \nin 2 db of the normal value.\n\nF a beam of white light is passed \nthrough a prism it will spread out \ninto a continuous colored band\n\nspectrum, shading from red to violet. \nThis shows that white light is com- \nposed of a great number of colors, and \nthat the wave length for each color is \ndifferent. An instrument called a \nspectrometer is used to separate any \nkind of light into its different colors or \nwave lengths, and the process is \nknown as spectrum analysis.\n\nIn like manner the various com- \nplex sounds which we hear are com- \nposed of a number of simple tones of \ndifferent wave lengths, but each wave \nlength has a different pitch or fre- \nquency. The acoustical research de- \npartment of Bell Telephone Lab- \noratories has recently developed an \ninstrument which separates complex \nsounds into their simple component \ntones and at the same time indicates \ntheir approximate amplitudes. By \nanalogy it is called an acoustic spec- \ntrometer.\n\nThe separation of complex tones \ninto their components is not new in \nitself. With one method employed,\n\nthe component frequencies were de- \ntected one at a time. Such a method \nof analysis is applicable only to com- \nplex tones that remain constant in \nevery respect while tests for the \ndesired frequencies are being made. \nAnother method, and the oldest one, \nwas to obtain a record of the wave \nform, such as a tracing on smoked \nglass or an oscillogram, and then by \nmathematical analysis or by mechan- \nical means, determine the combination \nof simple tones that would produce \nthe complex wave. This latter method \nhas the advantage that it is possible \nto analyze tones of short duration. \nThe former method, while applicable \nonly to tones of relative long dura- \ntion, is much faster.\n\nThe acoustical spectrometer re- \ncently developed, is applicable to \ntones of very short duration, and \nshows all the components at the same \ntime. Instead of having a single \nselective element which must be suc- \ncessively adjusted to all the desired \nfrequencies, it has a large number of \nselective elements each responding to \nfrequencies within a narrow range,\n\nFig. 1\u2014Small spherical mirrors mounted on vibrating reeds reflect a ray of light \nto indicate on a screen the component frequencies and their magnitudes\n\nand all acting simultaneously. With \nthis new apparatus it is therefore pos- \nsible to read or record at any moment \nall the components present in any\n\ncomplex tone. The analysis is ac- \ncomplished by first converting the \ncomplex sound waves into correspond- \ning complex electric currents. Tuned \nreeds, electromagnetically driven, are \nthen caused to vibrate\u2014each reed re- \nsponding to components in a different \npart of the frequency range.\n\nFigure 1 shows a schematic arrange- \nment of the spectrometer. Light from \na lamp filament passes through a slit \nand falls on concave mirrors which are \nmounted on the tuned reeds. An \nimage of the slit is brought to a focus \non a screen by each\n\nmicrophone, passed through an ampli- \nfier and then on to the six magnets\u2014 \none for each octave. All the fre- \nquency components that were in the \noriginal sound will be represented in \nthe magnetic flux, and the reeds will \nbe set in vibration when there is a fre- \nquency component that corresponds \nor is close to their resonance fre- \nquency. As the machine is now ar- \nranged, the spectrum may be watched, \nor the glass screen may be replaced by \nsensitized paper and _ photographic \nrecords made. A motion picture \ncamera might also be employed to \nphotograph a changing spectrum. \nThe appearance of the record made \nis illustrated by Figure 2, which shows\n\ncal line. The length of \nthe line is a measure \nof the amplitude of the \ncomponent, and the \nposition of the line \nalong the screen indi- \ncates the frequency.\n\nsecond by quarter tone \nintervals. The reeds \ncomprising each octave \nare driven by a single \nlong narrow electro-\n\nFig. 2\u2014Spectrograms of various sounds indicate the dif- \nferent magnitudes of the various harmonics\n\na group of spectra for sounds of var- \nious types and pitches. In each case \nthere is a fundamental of low fre- \nquency and a series of harmonics. The \namplitudes of the harmonics vary, and \nit is this variation in the relative mag- \nnitudes of the harmonics that gives \nthe difference in quality between tones \nof the same fundamental pitch.\n\nIn the design of the apparatus, pro- \nvision has been made for tuning each \nreed accurately to its correct fre- \nquency, and also for tilting the mir- \nrors so that the dots of light from suc- \ncessive reeds fall at equally spaced \nintervals along a straight line. Their \namplitudes of vibration are also ad- \njusted so as to be the same for the\n\nThe method and design that has \nproved so satisfactory in this demon. \nstration apparatus is easily applicable \nto wider frequency ranges and smaller \nfrequency ratios between adjacent \nWith 48 reeds per octave, \nwhich is readily obtainable, it is pos- \nsible to detect components with fre- \nquency differences of only *4 per cent,\n\nThis development provides a long- \nwanted tool for studying the fre \nquency components of sound of all \ntypes. The rapidity with which re. \nsults can be obtained by its use will, it \nis expected, make it useful in the \nacoustic studies carried on in these \nLaboratories.\n\nCc. N. Hickman received the A.M. \ndegree at Clark University in 1917, and \nthen engaged in ballistic research for the \nU. S. Government. At the close of the \nwar he was employed at the Bureau of \nStandards in the Inductance and Capaci- \ntance Laboratory. He then returned to \nClark University and received the Ph.D. \ndegree in 1922 under Dr. A. G. Webster. \nFollowing a short period of service with \nthe Navy Department designing sub- \nmarine mines, he joined the Department \nof Development and Research of the \nA. T. and T. Company in 1924. In 1930 \nhe joined the Technical Staff of the Lab- \noratories where he has been making a \nstudy of magnetic recording.\n\nF. F. Merrtam left Georgia Tech dur- \ning the war to engage in aircraft radio \nwork as a second lieutenant in the Army \nAir Service. After the armistice he re- \nturned to his studies, and completed the \ncourse in electrical engineering in 1919. \nComing directly to the Laboratories he \nworked on various electrical tests in the \nApparatus Development Department un- \ntil December 1919. He then transferred\n\nto the Research Department and assisted \nin the first ship-to-shore radiotelephone \ndevelopment during 1920 and 1921. The \nfollowing year he left the Laboratories to \nengage in radio merchandising in Atlanta, \nGa. and later in geophysical research in \nOklahoma and Texas. In 1927 he re- \nturned to the Research Department of the \nLaboratories, where he has since been en- \ngaged in short wave and ultra-short wave \nantenna development. He is at present \nin charge of the group which is engaged \nin this work.\n\nD.M. Terry received the B.E.E. degree \nfrom Ohio State University in 1920, and \nat once joined the Technical Staff of the \nLaboratories. Here he was first associated \nwith the Research Department, where he \nworked on fundamental carrier research \nand on the development of picture trans- \nmission. He was in charge of the trans- \nmitting apparatus in Cleveland for the \nfirst public demonstration of this system \nin 1924. Two years later he transferred to \nthe toll group of the Systems Department. \nHis work here has been chiefly on the de- \nvelopment of automatic control of trans-\n\nD. B. HERRMANN entered the Labora- \ntories in 1923. After a short time in the \nmessenger service, he transferred to the \nPatent Department and then to the \nChemical Laboratories as a laboratory \nassistant. In 1925 he left to attend the \nCollege of the City of New York, return- \ning to the Laboratories for two summers. \nHe joined the staff of the Laboratories in \n1928, and continuing his college work \nreceived the B.S. degree in 1930. Here \nhe has been concerned with general prob- \nlems of organic chemistry, particularly \nthose relating to rubber, balata, gutta \npercha and like ingredients of submarine \ncable insulation, and with measurements \nof the dielectric properties of this and \nother types of insulation. Recently he \nhas been occupied with studies of the \ndiffusion of water through organic insulat- \ning materials, and of the deterioration of \nvarious substances by light.\n\nAFTER GRADUATION by Cornell in 1903, \nI. C. Pettit was for the next several years \nan instructor in engineering for his alma \nmater. On entering the Engineering De- \npartment of the Western Electric Com- \npany in 1910, he at once began the work \non magnetic materials which has occupied \nhim almost continuously since. Among \nhis activities have been early measure-\n\nments on magnetic alloys; the application \nof silicon steel to repeating coils; and \ndesign of permanent magnets for tele- \nphone receivers. He also participated in \nthe development work preceding the in- \ntroduction of the porcelain protector block. \nWithin recent years in the Materials De- \nvelopment group Mr. Pettit has been \nresponsible for tests and specifications of \nall magnetic materials; for studies of the \nlarge number of new magnetic materials \nwhich have been invented in the Labora- \ntories and elsewhere, and for their recom- \nmendation to design groups for incorpora- \ntion into telephone apparatus.\n\nL. M. ILcenrrirz received a B.S. de- \ngree in electrical engineering from the \nUniversity of Michigan in 1920, and at \nonce joined the Department of Develop- \nment and Research of the American Tele- \nphone and Telegraph Company. Here, \nwith the Transmission Development \nGroup, he worked on the development of \nall types of carrier telephone and telegraph \nsystems, and of gain control systems em- \nploying carrier pilot channels. With the \nrecent merging of the D. and R. and Bell \nTelephone Laboratories, Mr. Ilgenfritz \nbecomes a member of the Technical Staff \nof the Laboratories with the Transmission \nDevelopment Department, where he con- \ntinues his studies in carrier problems.", "title": "Bell Laboratories Record 1934-10: Vol 13 Iss 2", "trim_reasons": [], "year": 1934}
{"archive_ref": "sim_record-at-t-bell-laboratories_1937-04_15_8", "canonical_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1937-04_15_8", "char_count": 89170, "collection": "archive-org-bell-labs", "doc_id": 250, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc250", "record_count": 116, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1937-04_15_8", "split": "test", "text": "single pair of conductors by using \nfrequencies above the voice range. \nThe greater the frequency range the \nmore telephone channels can be oper- \nated on the same conductors. Carrier \ntransmission is therefore attractive, \nbut there are, of course, serious tech- \nnical barriers, due to the tendency, in \ngeneral, for the circuit losses and in- \nterference effects to increase as the \nfrequency is raised, and to problems \nincident to the stepping up and down, \nand to separating the high frequencies. \nBy 1918, these problems had been \nsolved to a degree sufficient to enable \nthe first multi-channel carrier-tele- \nphone system to be designed and put \ninto commercial service. A_ later \nstandardized system\u2014the Type-C\u2014 \nwhich by using frequencies up to \n30,000 cycles, provides three tele- \nphone circuits in addition to the regu- \nlar voice circuit on a single pair of \nopen wires, has been extensively ap- \nplied all over the country.\n\nIn the meantime, development en- \ngineers have been working to widen \nthe frequency range to obtain more \ncircuits on a single pair of conductors, \nand to extend the usefulness of the \nmethod by applying it to the fine-wire \npairs in cables as well as to open-wire \nlines. They have also been developing \na new type of transmitting structure, \nespecially suitable- for transmitting \nvery high frequencies, this being \nknown as the coaxial circuit.* \n~ *Record, July, 1935, 322.\n\nFrom this work three new carrier \nsystems are resulting. Because of the \nwide band of frequencies they trans- \nmit, these are all classed as broad-band \nsystems. One system, for cables, whose \ndevelopment is fairly well completed, \nwill give 12 telephone channels on \neach pair, using frequencies up to \n60,000 cycles. Another, the coaxial \nsystem, in its experimental form, could \nbe used to give 240 circuits, using a \nfrequency range up to one million \ncycles. The third system under de- \nvelopment is for open-wire lines, and \nwill provide 12 telephone circuits in \naddition to the three circuits of the \nexisting T'ype-C system and the voice \ncircuit, 16 circuits in all on a single \npair of conductors. The top frequency \nis 140,000 cycles.\n\nOn the open-wire lines the large \ngauge and wide spacing of the con- \nductors are favorable to high-fre- \nquency transmission, but this exposed \ntype of structure is difficult to keep \nfree from unwanted currents induced \nby other nearby telephone circuits, by \npower lines, lightning, or other outside \ndisturbances. The limiting factor in \nusing higher frequencies on open-wire\n\nlines has generally been induction or \ncrosstalk between nearby circuits.\n\nUntil recently 30,000 cycles, the top \nfrequency of the Type-C system, repre- \nsented about the upper limit of trans- \nmission from the standpoint of keep- \ning similar systems on different pairs \nfrom crosstalking unduly one into an- \nother. Crosstalk on open-wire lines is \nreduced by applying transposition \nsystems; that is, the wires of a pair are \ncrossed over at predetermined inter- \nvals so that the induc- \ning potentials and cur- \nrents tend to balance \nout. For high frequen- \ncies these transposi- \ntions have to be made \nvery frequently and \nprecisely. There have \nbeen certain important \ndevelopments in the \nart of transposition de- \nsigning during the last \nfew years; also the \nspacing between the \ntwo wires of a pair is \nbeing reduced from 12 \ninches down to 6 or 8 \ninches so that there is \na greater distance be- \ntween the pairs and \nless direct pickup by \neach pair. These im- \nprovements will make \nit possible to raise the \nfrequency to about 140 \nkilocycles, thus paving \nthe way for the 12- \nchannel carrier systems \nwhich are being de- \nsigned.\n\nThe fine-gauge pa- \nper-insulated pairs in \nexisting cables do not \ninherently make good \nhigh-frequency con- \nductors; the attenua-\n\ntion is too great. This is offset to some \nextent by the shielding effect of the lead \nsheath, which largely excludes external \ndisturbances, and thus permits greater \namplification at repeater points than \ncan be used with open-wire circuits. \nEven so, repeaters will be required at \nmuch more frequent intervals than for \nthe open-wire line for a corresponding \nfrequency of transmission. Here again, \ncrosstalk between the pairs, which in \ncables are packed close together, is\n\nthe practical limiting factor in deter- \nmining the upper frequency and the \nnumber of channels which can be ap- \nplied. Since the pairs in a cable are \ntwisted, or in a sense transposed with \nrespect to each other, in such a way \nthat they can not be changed after \nhaving once been installed, there is \nno way of improving the crosstalk \nsituation comparable to changing the \ntransposition system as with open- \nwire lines.\n\nIt has been discovered recently, \nhowever, that it is possible to balance \nout, to a large extent, the crosstalk \nbetween the different pairs on which \ncarrier systems are to be applied, by \nusing small adjustable inductances or \ncondensers. These balancing units will \nbe connected at intervals between all \nthe pairs involved. A separate balanc- \ning unit is required between a given \npair and each of the other pairs in- \nvolved. In this way the practical fre- \nquency range of operation on the cable \npairs is being extended from a few \nthousand cycles to about 60,000 cycles.\n\nIn the coaxial conductor, the engi- \nneers set out to design a new structure \nwhich would combine the favorable \nlow attenuation characteristic of the \nopen-wire lines with the interference- \nfree characteristic of shielded cable \ncircuits. With such a structure a large \nnumber of circuits can be obtained on \nbut one set of conductors, using a \nfrequency range up to a million cycles \nor more if necessary.\n\nIn a coaxial unit, the outside tube is \na good metallic shield at the high fre- \nquencies. If currents are induced from \nexternal fields, such as power circuits, \nstatic, or lightning, they tend, because \nof skin effect, to confine themselves to \nthe outside surface of the tube, while \nthe useful currents for the same reason \nare on the inside of the tube. The \nspace between the inside conductor\n\nand the outside tube is largely air, \nwhich is a good dielectric having no \nlosses. The spacing insulators of hard \nrubber fill up only a relatively small \nproportion of the space, and they are \nof low-loss material. Also, the tube \nconductor is particularly well propor- \ntioned to offset skin effect and thus \nreduce the series losses. It should be \nnoted, however, that a very satis- \nfactory high-frequency conductor can \nbe made of a pair of wires separated \nby spacers within a metallic shield. \nSuch balanced pair structures may \nalso play a part in future high fre- \nquency transmission systems.\n\nWhile these various broad-band \nsystems have different frequency \nranges, they have many common \nfundamental problems. Probably the \nmost important is that of the re- \npeater or amplifier. In all cases the \nrepeaters are spaced much more fre- \nquently than is necessary for lower \nfrequencies. The approximate inter- \nvals are: for the cable system, 16 \nmiles; for the million-cycle coaxial \nsystem, 10 miles; for the open-wire \nsystem, 50 to 100 miles. This means \nusing many amplifiers in tandem, \nwhich in turn means that each ampli- \nfier must be quite perfect. Elaborate \nregulating means must also be pro- \nvided to make the amplifiers compen- \nsate for the varying attenuation of the \nline circuits with temperature or \nweather changes. The regenerative or \nnegative-feedback type of amplifier, \nwhich has already been described in \nthese pages,* is employed in the dif- \nferent broad-band systems. It has a \ntremendous advantage over older \ntypes from the standpoint of stability, \nfreedom from battery variations, and \nin linearity or perfection of amplifi- \ncation. The improvement over other \namplifiers in these respects is a hun- \n~ *REcorD, June, 1934, P. 290.\n\ndredfold or more, and thus puts the \nfeedback amplifier in a class by itself \nfor purposes of this kind.\n\nAmplifiers require power, and the \npower plant designers have been busy \nproducing new types of battery re- \nserve plants for the auxiliary repeater \nstations which will be added between \nexisting stations. For the coaxial sys- \ntem, means have been provided for \ntransmitting the power for the re- \npeaters over the structure itself.\n\nThe highest repeater amplifications, \nsome 60 or 70 db, will be employed \nwith the well-shielded conductors, \nthat is, the cable pairs and coaxial \nsystems. This is about 20 db, or about \na hundredfold power ratio greater \nthan amplifiers employed in the tele- \nphone plant in the past. It is interest- \ning to note that here, for the first time, \nrepeaters will be employed in which \nthe degree of amplification is set, not \nby external interference, but by inter- \nference which arises within the con- \nductors themselves \u2014 the so-called \nthermal noise or resistance noise*. \nThe motion of the free electrons within \nthe conductors sets up tiny electro- \nmotive forces which have a spread-out \nfrequency distribution covering the \nentire frequency spectrum. This in \neffect is Nature\u2019s own limitation to \nthe amount of amplification which can \nbe employed in any system.\n\nIt should not be judged, however, \nthat other interference problems are \nabsent in the case of the sheathed \nconductors. For example, in the case \nof the cable-carrier system, there has \nbeen a particularly severe problem in \nreducing the man-made interference, \nwhich arises within telephone offices \nwhen battery circuits are broken by \nthe operations of switches and relays. \nEach current break creates a tiny \nhigh-frequency oscillation, the effects \n~ *REcorD, February, 1937, p. 185.\n\nof which are readily picked up by in- \nduction in the various cable pairs as \nthey pass through the building. In \napplying the cable-carrier system, it is \nnecessary, therefore, not only to shield \nthe carrier pairs and apparatus in such \noffices but to keep this artificially \ncreated interfering current from trav- \neling out along the other wires in the \ncable and crosstalking into the pairs \nused for carrier systems. High fre- \nquency filtering circuits will therefore \nbe applied to all the non-carrier pairs \nwhere a cable leaves such an office.\n\nA common part of all of the broad- \nband systems is a group of modulating \nand filtering equipment that provides \ntwelve voice channels over the fre- \nquency range from 60 to 108 kilocycles. \nIn the system for cable circuits one of \nthese groups is transmitted over a \nsingle pair of wires, but by a second \nstep of modulation it is placed in the \nfrequency band from 12 to 60 kilo- \ncycles. A similar group of channels on \nanother cable pair is used for the re- \nturn transmission.\n\nThe open-wire systems will use two \nof the 12-channel groups on each pair, \none for each direction of transmission, \nthe first placed by a step of modula- \ntion above 30 kilocycles, which is the \nupper limit of the present three- \nchannel system, and the second, for \nreturn transmission, above it, with a \ntop frequency of about 140 kilocycles.\n\nThe coaxial system, as at present \nbeing used experimentally, can trans- \nmit 20 groups in the frequency range \nfrom 60 to 1020 kilocycles, employing \na second modulation for 19 of the \ngroups. This, as noted, gives 240 \nchannels. A similar coaxial provides \nthe 240 opposite directional channels. \nUltimately such systems will probably \nbe arranged to transmit a _ con- \nsiderably greater number of channels.\n\nthe coaxial and cable-carrier system it \nis found better to use different me- \ntallic circuits for opposite directional \ntransmission than to use different fre- \nquencies on the same pair as has been \na common practice on open-wire lines. \nOne reason is that in these systems, \nrepeaters will be needed so often, that \nthe directional filters which would be \nnecessary at each repeater point to \nseparate the opposite directions of \ntransmission would unduly increase \nthe costs. In the open-wire systems, \nthe attenuation is lower, as noted \npreviously, repeaters will be 50 to 100 \nmiles apart; and it is feasible to use \ndirectional filters. Furthermore, on \nopen-wire lines there are no possibili- \nties of getting the necessary high de- \ngree of opposite directional crosstalk \nseparation except by using different \nfrequencies. Separate pole lines are \nobviously impractical.\n\nTo summarize\u2014broad-band sys- \ntems will use much greater frequency \nranges than has been practicable in \nthe past. For the most part these \nbroad-frequency ranges will be \nchopped up into narrow bands for \ntelephone or telegraph purposes. \nBands of several million cycles in co- \naxial conductors can if needed also be \nmade available for television pur- \nposes. Each of the new systems will\n\nfill a useful place in the Bell System \nplant\u2014the cable and open-wire sys. \ntems, particularly where these lines \nalready exist\u2014the coaxial system \nwhere new structures are needed, \nand, in particular, on the heavy traffic \nroutes. Of course, considerable com- \nplicated equipment is required and \nfor the immediate future these various \nsystems will probably be found eco- \nnomical chiefly for transmission over \nthe longer distances.\n\nOne not obvious but important ad- \nvantage deserves mention in closing. \nBroad-band systems are all being ap- \nplied on non-loaded conductors. This \nmeans that they are high-speed sys- \ntems whose wave velocity approaches \nthat of light. As a matter of contrast, \nthe older type, long-haul, four-wire \nloaded cable circuits have a velocity \nof only 20,000 miles per second as \ncompared with 100,000 miles and up- \nwards for the broad-band systems. \nOn 20,000-mile-per-second circuits, \nproblems of echoes are exaggerated \nand even two-way talking may be \nawkward on circuits several thousand \nmiles long, because of the time taken \nfor a talker at one end of a circuit to \nbe heard at the other end, and to re- \nceive a reply. Broad-band high-speed \ncircuits will practically relieve us of \nconcern about matters of this kind.\n\nPPARATUS for measuring at- \ntenuation loss at high frequen- \ncies has received considerable\n\nattention by the Laboratories during \nthe past few years. This has been found \nnecessary because of new require- \nments imposed by broad-band carrier \ntelephone systems. These systems in- \nclude a large number of filters and \nequalizers, the design and production \nof which depend in part on having \navailable accurate means of measur- \ning their loss-frequency characteris- \ntics. The measurements are made by \napplying an electromotive force of \nknown frequency to the equipment \nunder test and to a standard attenu- \nator and adjusting the attenuator \nuntil the loss is the same in both cir-\n\ncuits. The attenuation of the equip- \nment under test can then be read di- \nrectly on the standard attenuator.\n\nA fundamental circuit of the type \nused is shown in Figure 1. The output \nof an oscillator, which can be varied \ncontinuously in frequency from 50 to \n5000 kc, is applied simultaneously to \nthe input of the apparatus under test \nand the standard attenuator through \nterminating resistances. The appa- \nratus and attenuator terminate at the \nother ends in resistances which are \nequal to the characteristic impedances \nof the apparatus and attenuator re- \nspectively. The detector is a\u2019 super- \nheterodyne and a microammeter indi- \ncates the point of balance because the \near cannot detect small enough dif-\n\nferences in sound level to make head \nphones satisfactory for precision work.\n\nThe errors due to the inherent i1n- \nductance and capacitance of the cir- \ncuit increase with frequency. If the \ncircuit impedance is made high, the \nerror due to the shunt capacitance be- \ncomes predominant, and conversely if \nlow the error due to the series induc- \ntance predominates. For the frequency \nrangeofthiscircuit, animpedanceof 100 \nohms was chosen as the best compro- \nmise. These limitations also prohibit \nthe use of decade resistances for termi- \nnations and small plug-in resistors \nwere, therefore, developed for the pur- \npose. These resistors have a smaller \nphase angle than the decade type and \nallow the set to be kept more compact.\n\nLosses of at least 80 db have to be \nmeasured to determine the attenua- \ntion peaks of filters. In making bal- \nanced-to-ground measurements of \nlosses of this magnitude, longitudinal \ncurrents, which flow down both sides \nof the circuit and return through \nground or by some other path, are the \nprincipal cause of error. Since these \ncurrents are not in general the same \nin either phase or magnitude in the \ntwo branches of the circuit, they re-\n\nsult in a false balance if they flow into \nthe detector. To reduce this error, the \ncircuit is isolated at input and output \nwith carefully shielded and balanced \nrepeating coils. A grounded center tap \non the measuring set side of the out- \nput coil gives a low impedance path to \nground for the longitudinal currents. \nIn addition to this it was found neces- \nsary to use wire having a very tightly \nwoven copper-braid shield for all leads \nbetween the component parts of the \nset-up and to make sure that these \nwire shields are continuous with the \nexternal shields of the apparatus. \nThis was done in the latest set con- \nstructed by using recently developed \ncoaxial plugs and jacks for making all \nconnections between the necessary \nleads and the apparatus.\n\nIn attempting to make high loss un- \nbalanced measurements it was found \nnecessary to construct the input and \noutput sections of the measuring set in \nseparate shielded compartments which \nwere insulated from each other. This \nwas required to eliminate multiple \nground paths.\n\nThe absolute accuracy of the meas- \nurements made with a circuit of this \nkind is dependent upon the accuracy\n\nFig. 1\u2014Attenuation at radio frequencies is measured by comparing the loss in the \napparatus under test with that of a standard attenuator\n\nStandard attenuator constructed with parallel-opposed resistance units wound\n\nof the attenuator used as a standard. \nIt has, therefore, been necessary to \ndevelop attenuators having negligible \nfrequency errors along with suitable \nmeasuring circuits. These attenu- \nators are made of constant-impedance \nresistance networks provided with \nsuitable means for switching. Any re- \nactance in these resistances or in the \nassociated switches and wiring results \nin increasing departure from the theo- \nretical value of the attenuator as the \nfrequency is increased. This undesir- \nable effect has been reduced by taking \nadvantage of recent improvements in \nwinding low-phase-angle resistances. \nTwo types are being used at present; \none with a parallel-opposed winding \non a small glass tube and the other, \nwhich may be wound to higher resist- \nance values, of the woven wire type. \nIn both forms the phase angle is \nnegligible except for very low or very \nhigh resistance values.\n\nTwo types of both balanced and un- \nbalanced attenuators which use these \nresistors have been developed. In one \ntype, shown in Figures 2 and 3, the \nnetworks are cut in or out by means of \ndouble-pole double-throw keys. These \nare relatively inexpensive but have \nthe disadvantage that the reactance \nof the keys and wiring of the networks \nnot in use is still in the circuit. Rotary \nswitches are used in the other type \nwhich represents a distinct improve- \nment over previous switching methods.\n\nAn unbalanced attenuator of this \ntype is shown in Figure 4. As may be \nseen the networks are connected di- \nrectly between the switch contacts; \nthis allows the wiring to be kept at a \nminimum. The construction of the \nswitch is such that the spurious ca- \npacitances are extremely small; other- \nwise they would be very troublesome \nat the higher frequencies. The switch \nwhich has 10 db steps is double decked\n\nFig. 3\u2014The resistances of this standard attenuator are mounted in two rows with keys \nbetween to cut the units in and out\n\nwith a network in each deck for losses \nabove 30 db. Splitting the higher \nlosses into two networks in this man- \nner keeps the series resistances from \nbecoming unduly high and also helps \nto decrease the direct capacitance be- \ntween input and output. A shield \nplate is inserted between the two \ndecks and the complete switch is en- \nclosed in a shielding box.\n\nTo reduce still further the direct ca- \npacitance between input and output \nof the switch it was \nnecessary to include \nadditional brushes to \nground the adjacent \nnetwork on either side \nof the one in use. Much \nthought and _ experi- \nmentation was re- \nquired to work out the \nproper method of mak- \ning the various ground \nconnections on this at- \ntenuator, since the \nground impedance \nmust be properly \nplaced in the circuit if \naccuracy is to be ob-\n\ntenuators of this type it is now pos- \nsible to measure losses of go db at 5000 \nke with an accuracy of 1 db. This is a \nconsiderable improvement over equip- \nment previously available.\n\nThe oscillator must be extremely \nwell shielded and must have high \nstability in amplitude and frequency \nif precision measurements are to be \nobtained. The detector must likewise \nbe well shielded and must have high \nsensitivity and also stability. A tuned\n\ntained at the higher Fig. 4\u2014Standard attenuator of the unbalanced type with\n\nFig. s\u2014The input and output sections of the measuring set are in separate shielded \ncompartments with space for the apparatus under test and the standard attenuator\n\ndetector is necessary to prevent errors \ndue to oscillator harmonics for meas- \nurements in the attenuating range of \nnetworks. A superheterodyne type of \ndetector seems to meet all of the re- \nquirements in the simplest manner. \nThe intermediate frequency is modu- \nlated by a built-in oscillator so that a \nfrequency of approximately 1000 \ncycles is obtained in the output. In \nthe latest form the indicator consists \nof a diode-triode rectifier tube with a \nmicroammeter connected in the plate \ncircuit of the triode section. By proper \nchoice of circuit constants a 3 db \nlinear scale can be obtained on this \nmeter which makes it possible to read \na balance with a precision of 0.05 db. \nThis meter can be calibrated at 1000 \ncycles and then serves as a check on \nthe accuracy of the attenuator in the \nhigh-frequency circuit for small in- \ncreases in attenuation. The use of a \ncalibrated meter makes steps smaller \nthan one or two db unnecessary on the \nattenuator and thus saves the expense \nof several additional keys. or one \ndecade. A further advantage is that\n\nthe tube cuts off on overload so that it \nis impossible to damage the meter.\n\nAn attenuation measuring set em- \nbodying the latest developments is \nshown in Figure 5. The repeating coils \nare provided with plugs and are in- \nserted in different jacks for balanced \nand unbalanced measurements. Two \nsets of coils are provided, one covering \na nominal frequency range of 50 to \n1500 ke and the other from 1500 to \n10,000 ke. By the lower key the out- \nput coil is switched from the attenu- \nator to the apparatus under test; the \nupper key gives a 180-degree phase \nshift in the input to the detector and \nthus serves as a check on the presence \nof longitudinal currents when meas- \nuring balanced apparatus. The equip- \nment to be measured is connected be- \ntween the jacks in the lower corners \nand those above either to a balanced \nattenuator or an unbalanced one.\n\nThe increased accuracy attainable \nwith the new equipment will make it \nuseful in development work where \nprecise measurements of insertion loss \nat high frequencies are required.\n\ntaining and improving the ser- \nvice have required some method of \nmeasuring noise on telephone circuits. \nAt first a simple form of listening test \nwas used, and the amount of noise \ntolerated depended largely on the \njudgment of those operating the tele- \nphone plant. This led to different \nstandards in different localities, which \nnaturally was not a satisfactory ar- \nrangement, and as a result the method \nwas supplanted by the use of the 1A \nnoise measuring set. This set in- \ncorporates a buzzer to produce a noise, \nand a potentiometer with which the \nnoise volume may be varied over a \nwide range. By listening alternately \nto the noise on the circuit and to the \noutput of the measuring set as the \nlatter is reduced in volume by the po- \ntentiometer, it is possible to find the \npoint at which the two may be judged \nto produce the same interfering effect, \nand the reading of the potentiometer\n\nFig. 1\u2014The 2A noise measuring set, with cover removed, \nshowing arrangement of apparatus on the front panel\n\nAlthough the 1A noise measuring set \nwas a great improvement over pre- \nviously existing devices, the results \nobtained with it did not attain the \ndegree of objectivity desirable, since \nthey rested in the ultimate analysis \non a personal subjective judgment. \nThis judgment required the balancing \nof an adjustable amount of standard \nbuzzer noise against the unknown \namount of circuit noise, not on the \nbasis of equal loudness but on the \nbasis of the interfering effect of the \nnoises on the interpretation of speech. \nIf the buzzer and circuit noises differed \ngreatly in quality the balance was \ndifficult. Unless care and good judg- \nment were exercised, two noises that \nwere judged to be equivalent by use of \nthe 1A set might be found to have \ndifferent interfering effects on a trans- \nmitted conversation.\n\nThe defects of the 1A set were \nlargely remedied in the 1A Noise \nAmplifier, brought out \nin 1929 for measuring \nnoise in toll offices. This \namplifier employed fre- \nquency weighting based \non the then available \ndata on the relative in- \nterfering effects of \nsingle-frequency tones, \nand used the 6A Trans- \nmission Measuring Set \nas an indicating device. \nBy eliminating the ear \nas the basis for deter-\n\nmining the noise magnitude, a distinct \ngain in objectivity of measurement \nwas made, but the set fell short of the \nideal in other ways, such as porta- \nbility, dynamic characteristics of the \nindicating instruments, and additional \nfrequency weightings.\n\nRecently a new noise measuring set \nhas been brought out that goes much \nfarther toward realizing the primary \nobjectives of noise measurements. This \nset, shown in Figure 1, is known as the \n2A noise measuring set, and incor- \nporates the results of many years\u2019 re- \nsearch on the problem of noise meas- \nurement. It carries its own power \nsupply in the form of dry cells, and is \ndesigned to be used as a portable \ninstrument although brackets may be \nobtained for mounting it on central \noffice frames when desired.\n\nThe interfering effect of noise de- \npends, in general, on its frequency \ncomposition, on the relative magni- \ntude and spacing of the various fre- \nquencies, on the duration of the noise, \non the type of circuit on which it \noccurs, and on the person who is listen- \ning to it. This large number and \nvariety of factors very greatly compli- \ncates the design of a satisfactory noise \nmeasuring set. The subjective element \nof noise is largely eliminated by em- \nploying a meter to give the indication \nof noise; but before such a method can \nbe satisfactorily used, much research \nmust be done to determine the inter- \nfering effects of different noises on \nspeech as heard by the ear, and then \ncircuits must be incorporated in the \nset between the noise input and the \nmeter to simulate the action of the ear.\n\nSince the effect of noise depends to a \nvery large extent on its component fre- \nquencies, a weighting network is in- \ncorporated in the set to attenuate the \nnoise frequencies in inverse ratio to \ntheir interfering effect. The impor-\n\ntance of various frequencies depends, \nhowever, on the type of circuit and on \nthe position in the circuit where the \nnoise is being measured. On an ordinary \ntelephone circuit, for example, which \ndoes not efficiently transmit fre- \nquencies above about 3000 cycles, a \nnoise frequency of 5000 cycles would\n\nFig. 2\u2014Circuit arrangement of noise set \nwhen connected through \u201c\u2018line\u201d jacks, \u201cre- \nceiver\u201d jacks, and \u201c\u201cprogram\u2019\u201d\u2019 jacks\n\nIf the noise were of a single fre- \nquency, these networks would insure \nthat its measured effect would be that \ndetermined by the average ear for\n\nrepresentative amplitudes of both \nspeech and noise; but when, as is \nusually the case, the noise comprises \nmany frequencies, the action of the \near in combining different frequencies \nmust also be considered. The measur- \ning set includes a rectifier for convert- \ning the alternating noise current to a \ndirect current for actuating the meter, \nand this rectifier is designed to com- \nbine the weighted noise currents in \nsuch a way that the sum is approxi- \nmately equal to the square root of \nthe sum of the squares of the weighted \ncomponents. Tests have been made \nwhich show that this rule of combina- \ntion is similar to that in the ear for a \nnumber of common noises.\n\nAfter all these adjustments of the \nnoise have been made, the period of \nduration of the noise must still be con- \nsidered. As a result of considerable \nresearch, it has been found that for \nthe usual noises found on telephone \ncircuits a time of about 0.2 second is \nrequired for the noise to be fully ap- \npreciated by the ear. To incorporate \nthis factor into the measuring set, the \nindicating meter has been given such \ndynamic characteristics that a noise\n\nenduring for 0.2 second will produce \nthe same deflection that a continuous \nnoise would, while noises of shorter \nduration produce proportionally \nsmaller deflections. For intermittent \nnoise, such as static, this dynamic \ncharacteristic makes the meter some- \nwhat more difficult to read, but by \nestablishing rules for reading the meter \non intermittent noises, approximately \nequal meter readings can be obtained \nfor both steady and intermittent noises\n\nhaving equal interfering effects. \nBesides these elements already dis- \ncussed, the noise set must also include \na coupling or connecting circuit so as \nto present a suitable impedance to the \ncircuit to which it is to be connected. \nThe complete measuring set thus in- \ncludes an input circuit, a weighting \nnetwork, an attenuator, an amplifier, \na rectifier, and an indicating meter. \nThe last four elements will remain the \nsame regardless of the type of circuit \nor the point of measurement, but pro- \nvision must be made for suiting the \nweighting and input networks to the \nparticular measurement being made. \nThis is accomplished by incorporating \nin the set the various networks re- \nquired, and by provid-\n\ntransmission between the line and the receiver, which is the \ndifference between curves A and C\n\n2500 various types of meas- \nurement. Connection \nfrom the line under test \nis usually made to a \npair of binding posts, \nand a plug\u2014connected \nto these binding posts\n\nFig. 4\u2014Weighting curve for program circuits, curve \nA, and for sound and volume measurements, curve B\n\nThe bottom pair of jacks is marked \n\u201cline,\u201d and is used for measuring \nnoise on a telephone circuit at some \nother place than the subscriber\u2019s re- \nceiver\u2014for example, at the toll office. \nIts input circuit presents an impedance \nof 600 ohms so as to match the line \nimpedance. The circuit arrangement \nis shown at the top of Figure 2. The \nweighting network used for this con- \nnection has the characteristics shown \nby curve A of Figure 3. This curve \nrepresents the sum of two weightings, \nviz., a frequency weighting for the \nrelative interfering effects of single \nfrequency voltages in the telephone \nreceiver (curve C) and the frequency \ncharacteristic of the circuit between \nthe line and receiver (curve B).\n\nWhen measurement is to be made \nat the subscriber receiver, the set is \nconnected across the receiver and has \na high input impedance. The circuit \nemployed, shown in the center of \nFigure 2, has an impedance of about \n2000 ohms. A different weighting net- \nwork is required for measurements at \nthe subscriber\u2019s receiver because of \nthe frequency characteristic of the \nattenuation between the line and \nreceiver terminals of the telephone \nset, which is indicated by curve B of \nFigure 3. The weighting characteristic \nprovided is therefore that shown by\n\nshown in the lower part of \nFigure 2, but a different weight- \ning characteristic is required \nbecause of the wider trans- \nmitted band of frequencies. The char- \nacteristic employed is shown as curve \nA on Figure 4.\n\nWhile these three types of measure- \nments are those that will be most \ncommonly employed, the usefulness of\n\nFig. s\u2014Circuit arrangements for measure- \nments of noise to ground, above; of sound, in \ncenter; and of volume, below\n\nthe set has been widened by providing \njacks for four other types that prove \nadvantageous under various condi- \ntions. It is sometimes desirable, for \nexample, to measure noise on a tele- \nphone circuit without breaking the \ncircuit. To make this possible a \n\u201cbridging\u201d input circuit is provided, \nwhich presents an impedance of about\n\n6000 ohms. The output of this circuit \nis connected to the input of the \u201c\u2018line\u201d\u2019 \ncircuit, and it thus uses the same \nweighting network.\n\nIn making any of these measure- \nments the potentiometer is turned \nuntil the meter pointer is on the scale, \nand the noise magnitude is then the \nsum of the readings of the potentiom- \neter and the meter\u2014the potentiom- \neter having a range of 60 db, and the \nmeter, of 18 db. Reference noise in all \ncases is the meter reading that would \nbe obtained with 10-!? watts of 1000- \ncycle power dissipated at the point of \nmeasurement.\n\nAnother input circuit, shown at the \ntop of Figure 5, makes it possible to \nmeasure noise voltage to ground. This \nmeasurement is often desirable when \nit is expected that large longitudinal \nvoltages are operating through circuit \nunbalances to cause metallic-circuit \nnoise. This circuit consists of a 100,000- \nohm input circuit and the \u201cline\u201d input \ncircuit in series. Reference noise to\n\nA \u201csound\u201d input circuit, shown \nschematically in the middle of Figure \n5, permits the set to be used, with the \naid of a microphone and matching \ntransformer, for measuring sound \nlevels. Minimum measurable level \ndepends on the sensitivity of the mi- \ncrophone and the efficiency of the \ntransformer, but with the 630A micro- \nphone and the 115A repeating coil, \nsound levels as low as about 55 db \nabove reference sound level can be\n\nmeasured. The weighting employed is \nthat shown by curve B of Figure 4. \nReference sound level corresponds to \n10-!6 watts per square centimeter at \n1000 cycles in a free progressive sound \nwave in the open air.\n\nBy use of the \u201cvolume\u201d input, \nshown schematically in the lower part \nof Figure 5, volume measurements \nmay be made on message circuits, but \nnot on program circuits, because the \nresponse of the volume circuit is not \nflat, the weighting being the same as \nfor the \u201csound\u201d circuit. The rela- \ntively low response in the volume \ncircuit at low frequencies is not of \ngreat Importance in measuring message \ncircuit volumes, but causes serious \nerrors in measuring volumes on pro- \ngram circuits.\n\nThe upper pair of jacks and the \ntelephone set dial are used for cali- \nbrating the set. Operation of the dial \nsets up a train of waves which impress \na predetermined input across the set. \nThe gain of the amplifier is then ad- \njusted until the peak swings of the \nmeter pointer (ignoring the first deflec- \ntion) are on a red line marked on the \nscale. This method gives a calibration \nfor all circuits by a single operation.\n\nAlthough the 2A Noise Measuring \nSet is not a perfect instrument for all \nconditions, the results of a wide range \nof tests indicate that it agrees better \nwith tests of the effects of noise than \ndoes any other apparatus available. \nIts applicability to a wide variety of \nconditions should make it very useful \nin maintaining satisfactory noise con- \nditions in the telephone plant.\n\nF. B. Llewellyn spoke on The Inside \nStory of the Vacuum Tube at the February \n15 meeting of the Colloquium. According \nto Dr. Llewellyn there has been recently \na renewed interest in finding out what \nhappens to electrons on the inside of \nvacuum tubes. Although this was brought \nabout largely by the requirements of \nultra-high-frequency operation, the re- \nsults have thrown additional light on the \nperformance of tubes at much lower fre- \nquencies. Several funda- \nmental properties of elec- \ntron streams were dis- \ncussed, and their influ- \nence on tube operation \ndescribed in terms of \nequivalent circuits.\n\nAt the meeting held on \nMarch 1, G. K. Teal dis- \ncussed The Application of \nRaman Spectra to Physi- \ncal and Chemical Prob- \nlems. In 1928, C. V. \nRaman found that the \nlight scattered by an or- \nganic liquid illumined \nwith visible radiation of \na single frequency con- \ntained not only the origi- \nnal frequency but also \ncertain other frequencies \ncharacteristic of the liq- \n' uid being studied. It soon \nbecame evident that the \nRaman Effect was a new \nphenomenon giving in- \nformation concerning the \nrotation and vibrations \nof molecules and that it \nconstituted a potent tool \nfor the chemist. Mr. Teal \ndiscussed the main fea- \ntures of the Raman spec-\n\ntra of some simple molecules and the \napplications of the data, and then re- \nviewed the results which have been ob- \ntained during the course of studies of the \nmore complicated molecules.\n\nNatTionaL Apvisory CouNCIL oN \nAPPLIED Puysics \nDuring the past year several members \nof the Laboratories have taken an active \npart in the meetings and deliberations of \nthe National Advisory Council on Ap-\n\nW. H. Harrison, Assistant Vice President of the American \nTelephone and Telegraph Company, has been elected a \nDirector of Bell Telephone Laboratories\n\nK. W. Waterson, Assistant Vice President of the American \nTelephone and Telegraph Company, has been elected a \nDirector of Bell Telephone Laboratories\n\nplied Physics. The Council was formed, \nlate in 1935, by the American Institute of \nPhysics to stimulate the application of \nphysics to other sciences and to industry. \nO. E. Buckley, M. J. Kelly and Harvey \nFletcher are three of the thirty-nine men \nforming the council. Dr. Kelly is a mem- \nber of the Executive Committee and \nChairman of the Committee on Applied \nPhysics. He also aided in the recasting of \nthe journal Physics to the Fournal of Ap- \nplied Physics which appeared in the new \nform with the January, 1937, issue. R. M. \nBozorth is an associate editor of this \njournal. G. A. Campbell and W. Wilson \nare members of a special committee which \nhas been appointed to assist those respon- \nsible for setting up university curricula.\n\nAT A MEETING of the \nstockholders of the Lab- \noratories held on March \n3, three new directors \nwere elected: O. E. Buck- \nley, W. H. Harrison and \nKk. W. Waterson to suc- \nceed E. H. Colpitts, Ban- \ncroft Gherardi and E. F, \nCarter. The other direc- \ntors are E. S. Bloom, \nC. G. Stoll, W. F. Hos- \nford and F. B. Jewett.\n\nDr. Jewett addressed \nthe Bond Club of New \nYork at its regular lunch- \neon held February 19 on \nthe subject Science and \nResearch in Electrical \nCommunication.\n\nE. L. Netson and H. \nB. FisHER visited Wash- \nington to attend the Air \nLine Safety Conference \ncalled by the Depart- \nment of Commerce. Mr. \nNelson and D. K. Martin \nalso attended a meeting \nof the Radio Technical \nCommittee for Aeronau- \ntics held in connection \nwith this conference.\n\nH. PrannenstIEHL and E. T. Morr- \nRAM visited Hawthorne to discuss the \nmanufacture of new recording equipment.\n\nF. G. Bunrenporr was in Hawthorne \non testing equipment to be used in the \nproduction of steel tape for magnetic \nrecording.\n\nH. C. Curt and W. L. Berrs visited \nDahlgren, Virginia, to conduct outdoor \ntests on experimental apparatus. Mr. \nCurl made several trips to Washington to \ndiscuss matters of interest with various \nGovernment organizations.\n\nR. E. Coram spoke on Radio Telephony \nin the Bell System before the Dover \nRotary Club on March 8.\n\nT. E. Suea visited Washington to dis- \ncuss communication problems with of- \nficials of the U. S. Navy.\n\nE. O. Scriven and D. T. BELL spent a \nday in Washington conferring with mem- \nbers of the Acoustics Division of the \nBureau of Standards.\n\nH. H. Gienn with W. V. THompson \nwere at Point Breeze on the development \nof rubber insulated cords.\n\nW. R. Netsser at Hawthorne conferred \nwith engineers of the Western Electric \nCompany on the manufacture and pro- \nduction of the 101A induction coil which \nis used in the combined telephone set.\n\nE. W. Apams, appointed General \nPatent Attorney on March 1 to suc- \nceed J. G. Roberts who retires on \nMay 1, graduated from \nArmour Institute of \nTechnology in 1908 with \nthe Degree of B.S.E.E. \nHe then studied law at \nthe National University \nLaw School and obtained \nhis LL.B. twoyears later. \nContinuing these studies, \nhe received the Degree of \nM.P.L. at Washington \nUniversity the next year. \nFor the four years fol- \nlowing his graduation \nfrom Armour | Institute, \nhe was an Assistant Ex- \naminer in the United \nStates Patent Office. He \nis a member of the Bar \nof the District of Colum- \nbia and of the State of \nNew York.\n\nIn 1912 Mr. Adams \nentered the Patent De- \npartment of the Western \nElectric Company, and \nfrom 1913 to 1916 was \nin Antwerp and London \nfor the International \nWestern Electric Com- \npany. On his return from\n\nEurope he again joined the Western \nElectric Patent Department. In 1925 and \n1926 another foreign assignment took \nhim to London and the Continent where \nhe prosecuted and maintained European \npatents owned by the Western Electric \nCompany. Since then Mr. Adams has \nbeen Assistant General Patent Attorney, \nhandling all phases of the Laboratories\u2019 \npatent problems dealing with telephone \nsystems and general equipment.\n\nA. C. WALKER, speaking on Moisture \nin Textiles, took part in a Symposium on \nApplied Physics at the North Carolina \nmeeting of the American Physical Society \non February 20. While there, he visited \nthe laboratories of the University of \nNorth Carolina at Chapel Hill and of \nDuke University at Durham. He also\n\nE. W. Adams, appointed General Patent Attorney of Bell \nTelephone Laboratories\n\nspent a day at the Charlotte branch of \nthe Kendall Mills observing the manu- \nfacture of cotton cloth. On March Io, Dr. \nWalker talked on Location and Distribu- \ntion of Moisture in Cotton before an \nA.S.T.M. committee meeting at Provi- \ndence, Rhode Island.\n\nP. NEILL discussed problems relating \nto the manufacture of switchboard plugs \nat Hawthorne.\n\nENGINEERING ROUTINES of designing \nfor manufacture were discussed by R. L. \nJones, O. M. Glunt, H. J. Delchamps and \nH. C. Atkinson with engineers of the \nGeneral Radio Company at Cambridge, \nMassachusetts.\n\nW. Fonp1Lier while visiting the Uni- \nversity of Florida at Gainesville spoke to \nthe senior engineering students on various \nproblems arising in engineering practice.\n\nTHE INSTALLATION of a trial length of \ncorrosion-protected cable at North Bergen, \nNew Jersey, was witnessed by L. S. Ford, \nJ. H. Gray and R. Pope.\n\nF. E. Nrmmcke supervised the installa- \ntion of a quarter-wave shunt-excited\n\nvertical radiator at Station WSAZ, \nHuntington, West Virginia. He also \nsupervised the installation of high fidelity \nconversion parts in the 304A Radin \nTransmitting Equipment at Station \nWPRO, Providence.\n\nP. C. Paguetre visited Kansas and \nColorado in connection with the trial in- \nstallation of the type-J carrier system. He \nalso visited Dallas, Texas, to discuss \ncable-terminal maintenance problems.\n\nA. L. Fox spoke on Non-Electrical \nProblems in an Electrical Industry before \nthe Columbus, Ohio, section of the \nA.I.E.E. While in Ohio he visited the of.- \nfices of the Ohio Bell Telephone Company \nat Columbus and Cleveland.\n\nR. C. and J. W. Kennarp \ndiscussed insulator development prob- \nlems at the Whitall Tatum Company\u2019s \nglass plants at Millville, New Jersey. Mr. \nDehmel later attended the meeting of \nCommittee D-9 on Electrical Insulating \nMaterials of the A.S.T.M. which was held \nat Pittsburgh.\n\nAmelia Earhart and W. C. Tinus discussing the Western Electric radio equipment \nwhich is installed in her \u201claboratory\u201d plane\n\nE. B. Smirx of the Systems Development \nDepartment completed thirty years of service \nin the Bell System on the third of March\n\nJersey, to inspect the ultra-high-frequency \nradio equipment being installed for the \nMunicipal Police Department.\n\nF. A. Hoyt visited the Chicago offices \nof the Illinois Bell Telephone Company to \ndiscuss coin collector problems. While in \nChicago he made a trip to the Hawthorne \nWorks of the Western Electric Company.\n\nH. I. Bearps.ey was at Hawthorne for \nten days on problems relating to the \nmanufacture of the combined handset \nand of other station apparatus.\n\nR. V. Terry visited the Warren Tele- \nchron Clock Company at Ashland, Massa- \nchusetts, to discuss the design of clocks \nfor use in telephone time bureaus.\n\nTHE MANUFACTURE and production of \nnew station apparatus were discussed by \nH. A. Frederick, H. O. Siegmund and \nJ. R. Townsend with engineers at Haw- \nthorne during the week of February 8.\n\nTELEPHONE OFFICES in Floral Park, \nLong Island, and Bayonne, New Jersey, \nwere visited by F. C. Kuch, R. B. Bauer \nand W. J. Lacerte where they observed \nthe performance of several relays of \nspecial design.\n\nAt Kearny, a demonstration of manu- \nfacturing processes for polarized relays \nwas witnessed by E. J. Pratt, D. E. \nBranson and M. R. Purvis together with \nR. J. Anspach and J. W. Harrison of the\n\nW. Y. Lance has been with the Teletype \nCorporation in Chicago discussing de- \nvelopment work relating to teletypewriter \napparatus.\n\nR. H. Linpsay and J. F. Morrison \nvisited Stations WEAN and WPRO at \nProvidence to supervise the tuning of the \ndirectional antenna arrays. Mr. Morrison \nalso took care of the tuning of the direc- \ntional antenna arrays at radio station \nBoston.\n\nA. C. Garrecut, with Mr. Scheff of the \nNew Jersey Telephone Company, visited \nthe East Orange, New Jersey, telephone \noffice to obtain information relating to \nprojected vibration tests of panel-dial \nequipment.\n\nCERAMIC INSULATING materials were \ndiscussed by W. A. Evans, V. L. Ronci \nand K. G. Coutlee with engineers of the \nAmerican Lava Corporation at Chatta- \nnooga on February 1. On the next two \ndays Mr. Evans and Mr. Coutlee visited \nthe Point Breeze plant on matters per- \ntaining to the manufacture of the 98A \nprotector block.\n\nO. W. Towner supervised the installa- \ntion of a 353E-1 (1-kilowatt) radio trans- \nmitting equipment and a direct-connected \nshunt-excited vertical radiator for Station \nWSAN, Inc., Allentown, Pennsylvania.\n\nT. C. Rice of the Inspection Engineering \nDepartment completed thirty years of service \nin the Bell System on the nineteenth of March\n\nHe also spent nine days at Station WHAS \nof the Louisville Courier Fournal assisting \nin the operation of the 306A (s0-kilowatt) \nradio transmitting equipment during the \nperiod of the recent flood emergency.\n\nMr. Evans and R. Burns attended \ncommittee meetings of the A.S.T.M. \nwhich were held at Pittsburgh. While \nthere Mr. Evans also visited the Stupa- \nkoff Laboratories.\n\nOn Fesruary 24, J. R. Townsend dis- \ncussed ceramic materials with scientists \nof the General Electric Company at \nSchenectady.\n\nW. Herriotr visited the Bausch and \nLomb Optical Company and the Folmer- \nGraphlex Company, Rochester, New \nYork, on various optical and photographic \nproblems.\n\nC. H. Greena.t took part in the dis- \ncussion of the paper The Fatigue of Copper \nAlloys which was presented at the meet- \ning of the American Institute of Mining \nand Metallurgical Engineers held in New \nYork. With Mr. Townsend, he attended \nvarious committee meetings of the \nA.S.T.M. held in Chicago in March.\n\nA GROUP LUNCHEON of the supervisors \nof the Apparatus Development Depart- \nment was held at the Fifth Avenue Hotel\n\non March to. Following the luncheon, \nR.L. Jones introduced O. E. Buckley who \ngave a most interesting talk on Executiye \nHeadaches. H. C. Atkinson responded and \nexpressed the appreciation of the audience, \nDr. Buckley discussed some of the prob- \nlems intimately associated with the func- \ntioning of the Laboratories, about which \nthe Executive Vice President has to \nworry and which are of vital interest to \nthose in supervisory capacity. About 180 \nmen attended the luncheon. The com- \nmittee in charge of arrangements was as \nfollows: W. F. Brown, H. C. Curl, P. \u00a7, \nDarnell, J. B. Dixon, R. C. Koernig, \nW. H. Sellew, O. A. Shann, H. D. Wilson, \nand J. M. Wilson.\n\nF. P. Wicur is at Norfolk, Virginia, \nin connection with the installation of \nteletypewriter line concentration unit and \nassociated equipment.\n\nJ. C. Crow supervised the instal- \nlation of a 310B (250-watt) radio trans- \nmitting equipment at Station WBAX for \nJ. H. Stenger, Jr., located at Wilkes- \nBarre, Pennsylvania.\n\nW. H. LicHTENBERGER attended the \ncoordination conference at Hawthorne on \nthe New St. Paul and Minneapolis toll \nand dial project.\n\nMEMBERS OF THE LABORATORIES TO WHom Patents WERE \nIssuED DurinGc JANUARY AND FEBRUARY\n\nJ. W. Beyer D. W. Grant \nW. M. Bishop A. J. Grossman \nB. G. Bjornson C. W. Halligan \nN. W. Bryant H. S. Hamilton \nW. W. Carpenter A. E. Harper \nR. S. Caruthers (2) H. C. Harrison \nH. R. Clarke J. M. Hayward \nI. E. Cole R. A. Heising (2) \nW. W. Cramer L. B. Hilton \nG. C. Cummings (2) _ V. L. Holdaway\n\nTo measure internal \nnoise, a transmitter \nis carefully shielded \nfrom vibration in \nthis spring-sus- \npended chamber\n\nassignments in the Mississippi and Ohio \nValley flood area, where they were co- \noperating with the telephone companies \nin applying emergency power equipment. \nMr. Callahan was also in Buffalo, New \nYork, and West Chester, Pennsylvania, \nin connection with high-speed gasoline \nengine-driven generators and Mr. Sole \nwas in Fort Wayne, Indiana, on motor \nand generator development and voltage \nregulation.\n\nH. H. Spencer visited various points \non the Toledo-South Bend cable to in- \nspect the cable-carrier power plants de- \nsigned for this cable.\n\nR. P. Jurson has been checking the \noperation of the power equipment on the \nNew York to Philadelphia coaxial cable.\n\nDurinc Fesruary the flood area in \nPaducah, Kentucky, was visited by \nE. W. Hancock to inspect the damage to \nthe Paducah central office equipment and \nto assist the telephone company in plan-\n\nW. J. Lacerre visited Hartford in \nconnection with a trial installation of \n248-type relays for use in step-by-step \nsystems.\n\nIN CONNECTION with tests on the New \nYork to Philadelphia coaxial circuit, \nJ. P. Radcliff has gone to Philadelphia \nand R. L. Tambling has returned to \nNew York.\n\nL. F. Porter, accompanied by F. S. \nJames and M. J. Stigers of the Operation \nand Engineering Department of the \nAmerican Telephone and Telegraph Com- \npany, visited Pittsburgh to discuss with \nengineers of the Bell Telephone Company \nof Pennsylvania the proposed new toll- \ntandem facilities for the Pittsburgh toll \noffice.\n\nAN INVESTIGATION of crosstalk at car- \nrier frequencies in lead-covered cables \nwas made at Point Breeze by F. W. Am- \nberg, L. Hochgraf and M. A. Weaver.\n\nTen members of the Laboratories were \nretired with pensions during 1936, making \na total of 89 receiving service or disability \npensions at the end of the year. One re- \ntired member died in 1936.\n\nPayments of accident disability bene- \nfits and expenses resulting from acci- \ndental injuries were lower in 1936 per \n$1000 of total Laboratories\u2019 payroll than \nin any previous year. Sickness benefit \npayments showed a marked increase; \nmore cases of over a week\u2019s duration oc-\n\nFifteen active members of the Lab- \noratories died in 1936 and, in accordance \nwith the Plan, where eligibility to Death \nBenefits existed, payments were author- \nized to their qualified beneficiaries.\n\nIn addition to benefits under the Plan, \nsupplementary relief payments totalling \n$5,499.25 were paid to active and retired \nmembers of the Laboratories who were in \nneed of special assistance during the year.\n\nThe members of the Employees\u2019 Bene- \nfit Committee are A. F. Dixon, Chairman, \nE. W. Adams, A. B. Clark, J. W. Farrell, \nR. L. Jones, M. J. Kelly, L. Montamat, \nG. B. Thomas and W. Wilson. J. S. \nEdwards is Secretary of the Committee \nand D. W. Eitner is Assistant Secretary.\n\nJ. S. Epwarps, Secretary, \nEmployees\u2019 Benefit Committee. \nThe above statement of payments \naudited and found correct. \nE. J. Santry, General Auditor.\n\nRecent y E. S. Witcox, M. V.HunTER \nand M. Aruck moved from Fort Worth \nto San Antonio, Texas, where they will \ncontinue high-frequency crosstalk tests \non open-wire circuits.\n\nR. L. Kaytor has been at Phoenixville \nmaking induction tests in connection with \nbroad-band systems.\n\nR. S. Hoyt graduated from the Uni- \nversity of Wisconsin in 1905 with the \ndegree of B.S. in Electrical Engineering. \nHe then spent a year at the Massachu- \nsetts Institute of Technology as a gradu- \nate student and assistant instructor. In \nJune, 1906, he joined the Engineering \nDepartment of the American Telephone \nand Telegraph Company in Boston and \nthere he became concerned with trans- \nmission development and research. In \nthe 1907 move to New York, Mr. Hoyt \ncame to West Street to be with the Engi- \nneering Department of the Western Elec- \ntric Company. Two years later he con- \ntinued his education at Princeton where \nhe received his M.S. in 1gto. Returning \nto the Western Electric Company that \nSummer, he worked for a year in connec- \ntion with the development of various re- \npeaters and on loaded lines, and then \ntransferred to the American Company \nwhere he was first with the Engineering \nDepartment and then with the Depart- \nment of Development and Research. \nWhen the latter was consolidated with\n\nthe Laboratories in 1934, Mr. Hoyt re. \nturned to West Street with the Trans- \nmission Development Department.\n\nDuring the thirty years of service which \nhe completed on February 22, Mr. Hoyt \nhas contributed much to the theory of \ntelephone transmission lines and associ- \nated apparatus, to the theory of crosstalk \nand other interferences and to probability \ntheory with particular regard to its ap- \nplication in telephone transmission engi- \nneering. He has been issued twenty-one \npatents and has been the author of several \npapers published in the Bell System \nTechnical Journal.\n\nON THE ELEVENTH of last month Herb- \nert Vaderson completed a quarter century \nof service with the Western Electric Com- \npany and the Laboratories. Mr. Vaderson \njoined the Engineering Department of the \nWestern Electric Company in 1912 asa \ndraftsman in the apparatus design group. \nFour years later he transferred to the \ngroup designing and developing radio \nequipment for Army and Navy purposes, \nparticularly for use on aircraft and on \nsubmarine chasers.\n\nFollowing the War he spent six months \nat Deal Beach in the mechanical design \nof the apparatus used in the first practical \ndemonstration of ship-to-shore radio tele- \nphone. Since then Mr. Vaderson has been \nprincipally engaged in the mechanical \ndesign of radio broadcasting apparatus\n\nand radio communication apparatus for \naircraft. In 1927 he participated in the \ndevelopment of the radio equipment used \nin the New York to Washington television \ndemonstration.\n\nTHE PRESENTATION of a five-star serv- \nice emblem to W. L. Heard on March 4 \nmarks his completion of a quarter century \nof service in the Western Electric Com- \npany and the Laboratories. Following his\n\n1g11 with the degree of B.S. in Electrical \nEngineering he immediately went with \nthe Automatic Electric Company in \nChicago. The following March he joined \nthe student course of the Western Elec- \ntric Company at Hawthorne and then \nentered the Engineering Department on \nEquipment Development. In 1918 he \ntransferred to New York and became \nintimately associated with the develop- \nment of machine switching equipment, \nparticularly that used in the first panel \ninstallations such as the New York \nMetropolitan Toll-Tandem and the Kan- \nsas City office. He also assisted in the de- \nvelopment of the No. 11 manual switch- \nboard and various other equipment for \ncentral office use. He also held various \nsupervisory positions in the equipment \ngroup during this period.\n\nSince 1924, as Equipment Methods En- \ngineer of the Systems Development De- \npartment, Mr. Heard has had charge of \nall drafting work and of handling, index- \ning and distributing the tracings, specifi- \ncations and drawings prepared in the \nSystems group as well as quantities of \nsimilar technical information received \nfrom the telephone company and the \nWestern Electric Company for reference \npurposes. Methods of handling this work \nefficiently and departmental costs and \nclerical work are also under the super- \nvision of Mr. Heard.\n\nR. M. Bozortu spoke on Ferromagnetic \nTheory before the Communication group \nof the New York Section of the American \nInstitute of Electrical Engineers.\n\nOPEN-WIRE CIRCUIT crosstalk tests on \ncircuits involved in the trial of the \nType-J carrier system were made by John \nMallet, C. H. Gorman, Jr. and J. L. \nLindner at Stafford, Kansas.\n\nDwRING THE MONTH of February, G. C. \nSouthworth spoke on the subject of Wave \nGuide Transmission at McGill University, \nMontreal, at the Research Laboratory of \nthe General Electric Company, Schenec- \ntady, before the Montreal Electric Club \nat luncheon and at meetings of the Engi- \nneering Institute of Canada and the \nAmerican Institute of Electrical Engi- \nneers that were held in Montreal and \nToronto respectively.\n\nC. R. Burrows spoke on The Surface \nWave in Radio Propagation Over Plane \nEarth before the Radio Club of America \nin New York on February 11.\n\nE. E. ScouMACHER has been elected to \nserve on the Executive Committee of the \nInstitute of Metals Division of the Ameri- \ncan Institute of Mining and Metallurgical \nEngineers.\n\nJ. R. Witson and H. A. PipGeon were \nin Princeton, Philadelphia, Harrisburg, \nand Elkton, Pennsylvania, to inspect \nvacuum tubes in repeater stations.\n\nAr Scuenectapy, F. R. Lack, H. E. \nMendenhall and S. B. Ingram discussed \nthe design and manufacture of various \nvacuum tubes with engineers of the \nGeneral Electric Company.\n\nnow held by members of the Laboratories: Wetherill, by E. C. Wente and \nF. F. Lucas; Howe, by F. F. Lucas; Longstreth, by E. Bruce; Edison, \nby F. B. Fewett; Elliott Cresson, by G. W. Elmen, C. F. Davisson and \nL. H. Germer; Faraday, by F. B. Fewett; Hughes, by C. F. Davisson; \nFohn Scott, by G. W. Elmen and H. E. Ives; Royal Photographic \nSociety, by F. F. Lucas; Progress (S.M.P.E.), by E. C. Wente; Dudley, \nby Ff. R. Townsend. Others are the Franklin Medal, by F. B. fewett; and \nthat recently awarded to the Laboratories as a whole by the American \nInstitute of the City of New York, shown in the last issue\n\nHarvey FLETCHER attended the meet- \ning of the American Physical Society at \nDuke University in Durham, North \nCarolina. Later, at Troy, New York, he \nspoke on The Nature of Hearing at a \nconvocation of Russell Sage College.\n\nR. R. WitutaMs spoke on The Quest of \nVitamin B, before the Franklin Institute \nin Philadelphia on February 17.\n\nW. J. Ciarke attended the A.S.T.M. \nCommittee Dg meeting, held in Pitts- \nburgh, to discuss cure tests on various \nmolded plastics.\n\nR. M. Burns and S. O. Morcan \nvisited the New York College of Ceramics \nat Alfred, New York. They also visited the \nResearch Laboratory of the General Elec- \ntric Company in Schenectady, New York. \nMr. Morgan has been made vice-chair- \nman of the Subcommittee on Mono- \ngraphs of the Engineering and Industrial \nResearch Division of the National Re- \nsearch Council.\n\nTHE INSTALLATION of testing equip- \nment for the combined handset took \nF. West, J. T. L. Brown and L. E. Krebs \nto Hawthorne.\n\nG. A. Wau. was at Hawthorne on a \nstudy of various molded parts which are \nto be used in the combined handset.\n\nR. E. Waterman (left) examining small wood speci- \nmen at Gulfport test plot while G. 9. Lumsden\n\nH. A. Larter visited several telephone \ncompanies and distributing house shops \nin the Southern and Western Areas.\n\nH. A. Pipceon, on March 3, lectured \non Multi-Electrode Vacuum Tubes and \nBeam Power Tubes in the A.1.E.E. Elec- \ntronics Course.\n\nAT THE CONVENTION of the Op- \ntical Society of America held at \nCorning, New York, on March 4 \nto 6, H. E. Ives presented a \npaper entitled Graphical Exposi- \ntion of the Michelson-Morley Ex- \nperiment. He also presented a \npaper entitled The Optical Con- \nstants of Sodium with H. B. \nBriggs as co-author.\n\nOn January 27, M. E. Strieby \ngave an illustrated talk on co- \naxial cable systems before the \nTreasury Luncheon Club com- \nposed of members of the financial \ndepartments of the Bell System \ncompanies in the Metropolitan \ndistrict. He spoke on Quantity \nProduction of Toll Circuits.\n\nKart K. Darrow has been elected to the \nCouncil of the French Physical Society.\n\nL. H. Germer, on March 12, spoke be- \nfore the Metropolitan Section of the \nAmerican Physical Society on Structure \nof Organic Crystals.\n\nPLATING PROBLEMS made it necessary \nfor C. L. Hippensteel to visit the Western \nElectric Company at Hawthorne. Later \nhe presented R. M. Burns\u2019 discussion of \nthe paper entitled Principles of Corrosion \nTesting presented before the American \nSociety for Testing Materials in Chicago.\n\nF. M. Ryan spoke on Radio Reflections \nbefore the March meeting of the Collo- \nquium held at the Deal Radio Laboratory.\n\nA. M. SKELLeTT addressed the Ama- \nteur Astronomers Association at the \nAmerican Museum of Natural History in \nNew York, March 3, on The Solar Corona.\n\nFouR-STAR SERVICE emblems signifying \nthe completion of twenty years of service \nwith the Bell System were awarded to \nnine members of the Laboratories during \nthe month of March. In the Apparatus De- \nvelopment Department, E. H. Chatterton \ncompleted twenty years\u2019 service on the \nsixteenth, L. P. Collins on the twenty- \nsixth and Michael O\u2019Connell on the \ntwelfth; in the Research Department, \nJ. J. Curley on the twenty-eighth; in the \nSystems Development Department, J. W. \nWoodard on the second; in the General \nService Department, Angeline McDer- \nmott on the twenty-first; in the Plant De- \npartment, N. R. Zucconi on the twenty- \nthird; and in the Patent Department \nM. R. McKenney on the twelfth and \nC. A. Sprague on the fourteenth.\n\nbilled on a message basis. To \nrecord this usage as many as 10,000 \nindividual message registers may be \nmounted on racks in each central of- \nfice. The monthly job of recording the \nreadings of these registers requires \nconsiderable time and effort. For \nmany years it was accomplished uni- \nversally by sight-reading routines \nwhich varied in details but usually in- \nvolved sending teams of two girls \neach into the terminal rooms during \nthe night hours.\n\nIn an effort to develop other and \nbetter means for doing the job, cam- \neras of special design have been made \nto photograph the registers. The one \nnow used in a few of the largest cities, \nFigure 1, is a manually operated \ncamera which has a special hood in \nfront to fit against the registers photo- \ngraphed. Inside of the hood \nare four 50 c.p. 24-volt lamps \nto provide illumination. The \nrecords are made on sensitized \npaper of high photographic \nspeed and contrast which is \nwound on a_wooden-cored \nspool with metal ends and pro- \ntected by a heavy opaque \nleader and trailer. The spacing \nof the sensitized paper and set- \nting of the shutter are done \nsimultaneously by operating \nthe winding handle and the \nexposure is made by pressing \na finger release located at the \nside of the camera. Ten regis-\n\nters are recorded at a time and since \neach spool of sensitized paper takes \n100 exposures, space for photographing \n1000 registers per loading is available.\n\nWhen the recordings are made the \ncamera hood is adjusted against the \nupper left group of ten registers. After \nthe first exposure, turning the hand \ncrank resets the camera and then a \nrecord is made of the adjacent block of \nten registers. This operation is re- \npeated one hundred times after which \nthe camera is unloaded and the spool \nsealed and sent to the laboratory for \ndevelopment. It is then ready for the \nAccounting Department. Figure 2\n\nIt was found that this camera could \nbe improved from the standpoint of \nease of handling and speed of opera- \ntion since much time is lost in the \nmanual operation of hand cranking\n\nFig. 1\u2014Photographic records of message register \nreadings are now made in some exchanges with a\n\nFig. 2\u2014The present message register camera photographs ten registers at a time\n\nbetween each exposure. Accordingly, \na new camera shown in Figure 3 has \nbeen developed in which a small uni- \nversal motor is used instead of the \nhand to move the sensitized paper, \ncontrol the illuminating lamps, and \nset the shutter. Twenty-five registers \nare photographed at a time, thus re- \nquiring only 40 exposures per 1000 \nregisters. A single switch controls the \nentire action of the camera and it \noperates on either 110 volts ac or dc, \nwhich is supplied through a cord with \nterminal plug mounted flush in the side \nof the camera. The body of the camera \nis made of sheet duralumin and the \nexterior surfaces are lacquered black.\n\nA projecting hood rests \nagainst the register cover as in \nthe previous design. I]lumina- \ntion is provided by four s0- \ncandlepower double-filament \nautomobile lamps mounted \ntwo on each side of the camera \nunder removable protecting \ncovers. Condensing lenses are \nplaced immediately in front of \nthe lamps to increase the il- \nlumination at the plane of the \nregister and a vignetting de- \nvice serves to blend the illumi- \nnation from the two sides of \nthe camera over the central\n\nsection of the area photographed. \nFour mirrors of chromium-plated \nmetal line the interior of the project- \ning hood. These serve to build up \nillumination at the sides of the pic- \nture area and to insure proper light- \ning of numbers located near the edges \nof the picture. A mirror inclined \nat 45\u00b0 near the lens deflects the \nlight toward the base of the camera \nwhere the sensitized paper is located. \nA large section of the back cover ex- \ntending from the loading chamber to \nthe top of the camera is removable to \ngive access to the lens, shutter and \ninclined mirror.\n\nthe sensitized paper, controls the illuminating \nlamps and sets the shutter\n\nFigure 5, is covered by a light-tight \ndoor for loading and unloading. The \nroll of unexposed sensitized paper is \nmounted at the left-hand side of the \nchamber. It passes under an idler \nroller and around the drum at the left \nin front. This drum has a toothed \nsprocket mounted immediately below \nthe spool to measure the advance of \nthe sensitized paper for each exposure. \nIt is then drawn across the exposure \naperture which is parallel to the base \nof the camera and wound onto the \ntake-up roller shown in the right-hand \nsection of the loading chamber. The \noperator knows when the sensitized \npaper is in position for the first ex- \nposure and again when the last frame \nis in place by the lighting of a minia- \nture bulb behind a small red window \nin the camera. Cut-outs in the edge of \nthe sensitized paper actuate the cir- \ncuit which controls this lamp.\n\nThe mechanism which transmits \npower to the sensitized paper spools \nand to the measuring roller also con- \ntrols the current to the lamps used to \nilluminate the registers. These lamps \nare lighted a very short time before\n\nthe actual exposure is made and \nturned. off immediately afterward. \nThis decreases the duration of ex- \nposure to approximately a third of \nthe time required for the full cycle of \ncamera operations and materially re- \nduces the effect of heat from the \nlamps. The torque-voltage charac- \nteristics of the motor are such that \nany variation in line voltage auto- \nmatically changes the exposure. If the \nline voltage is high, the lamps burn \nmore brilliantly but the motor speed \nwill then increase so that the lengths\n\nof the exposure portion of the cycle \nwill be correspondingly shortened and \nover-exposure avoided. On the other \nhand, if the line voltage is low, the \nlessened intensity of illumination is \ncompensated for by a lower motor \nspeed which prolongs the exposure. \nRecords made with the new camera \nare shown in Figure 6.\n\nThe operating mechanism is \nmounted as a removable self-con- \ntained unit located behind the plate \nto which the motor and gear box are \nattached. Power from the drive motor\n\nis transmitted through reduction gears \nto the pawl which controls the spacing \nroller, to the shutter, the toggle \nswitches used to operate the lamps \nand motor, and to a precision clutch \nmember which extends into the load- \ning chamber to control the take-up \nspool. All four of the lamps in the \ncamera are operated in series and a \nfixed resistance is added to reduce the\n\nThis new camera is of relatively \nsimple construction and is ruggedly \nbuilt. Endurance tests, involving \nthousands of operations, have proven \nits ability to perform satisfactorily \nunder practical conditions of use and \nwith considerable saving in time over \nthe previous design.\n\nFig. 6\u2014The new message register camera photographs twenty-five registers at a time\n\nand higher frequencies has made \nit necessary to design impedance- \nmeasuring equipment for a precision \nover this extended range, comparable \nto that obtainable in the audio and \nlower-carrier ranges. For this purpose, \nthe familiar alternating-current \nWheatstone bridge* was refined to \nmeet the severe requirements imposed \nby the higher frequencies. While es- \nsentially the same as other a-c im- \npedance bridges in circuit arrange- \nment, it differs from them in the care \ntaken to secure constancy of all resid- \nual impedances over a very wide \nfrequency range, thorough shielding, \nand accurate circuit balances. An in- \n~ *REcoRD, Fanuary, 1932, p. 173.\n\nteresting innevation is the use of \ncoaxial plugs and jacks for connecting \nto the standard impedances, to the un- \nknown impedance and to the oscil- \nlator and detector.\n\nThe circuit for the new bridge is \nshown schematically in Figure 1. It \nconsists primarily of a pair of double- \nshielded ratio arms, repeating coils for \nthe oscillator and detector, and a set of \nsix coaxial jacks by which the de- \ntector, oscillator, and the standard \nand test impedances are connected. \nThese components are assembled on a \nseven by sixteen inch panel, and en- \nclosed in an aluminum housing, which \nserves as a shield.\n\nA photograph of this balance unit \nfrom the rear, with the covers re- \nmoved from the double shielded ratio\n\narms, is shown in Figure 2. The ratio \narms, each of 75 ohms, employ the \nwoven-wire* resistor element, which \nis ideally suited for the purpose be- \ncause of its very small distributed \ninductance and capacitance. They \nmaintain their impedance balance or \nratio to within one one-hundredth of\n\nunit in mechanical layout, in wiring, \nand in the disposition of the ground \nadmittance of the elements. The elec. \ntrical symmetry obviates many of the \nerrors that are peculiar to high-fre- \nquency measurements.\n\nTwo impedance standards have \nbeen designed for use with the new \nbalance unit. One is a six-dial \nresistance standard and the \nother an adjustable capaci- \ntance standard. Both are \nequipped with coaxial jacks to \npermit them to be readily con- \nnected to the balance unit in\n\none per cent for resistance and to a \nthousandth of a microhenry for in- \nductance, over the entire frequency \nrange from Io to 5000 kilocycles.\n\nThe double-shielded repeating coils, \nwhich serve to isolate the bridge cir- \ncuit electrostatically from the power \nsource and the detector circuits, are \nscarcely less important than the ratio \narms in their contribution to the satis- \nfactory performance of the bridge. \nWith an impedance ratio of one to one, \nthey are designed and constructed with \nspecial care to insure a low value of \nintershield capacitance, and to mini- \nmize the capacitance between the ele- \nments separated by the ground shield.\n\nAs is partially evident in Figure 2, \na very high degree of symmetry is \nmaintained throughout the balance \n~ *RECORD, January, 1935, p- 136.\n\nThe resistance standard, \nFigure 3, is similar to the six- \ndial rotor-decade already de- \nscribed in the Recorp.* The \nresistors are mounted in drums, \nwhich are rotated between fixed \nbrushes so that only one re- \nsistor unit per decade is in the \ncircuit at a time. The three \nlarger decades employ resistors \nof the woven-wire type. In each of the \nthree smaller decades, the resistor ele- \nments are made to have as nearly as \npossible identical inductance residuals, \nso that the total inductance of the \nbridge is maintained substantially \nconstant for any setting.\n\nThe capacitance standard has a \nrotor-decade of mica condensers with \nstep values of 1000 micro-microfarads \nand a special two-dial air condenser. \nOne of the component air condensers \nhas step values of 100 wuf and the \nother is continuously variable over a \nrange of 110 wuf. The mica condenser \nis similar to the resistance decades in \ninserting only one unit in the circuit at \na time. This arrangement not only \nminimizes stray internal capacitances, \nbut keeps the series inductance of the \n~ *RECORD, January, 1935, p. 136.\n\nFig. 2\u2014Thorough double-shielding and symmetrical arrangement are featured in the \nbalance unit of the five-megacycle impedance bridge\n\nleads and wiring to a small constant \nvalue which may be compensated by \nadding an equal inductance in the op- \nposite arm of the bridge. The complete \ncapacitance standard removed from \nits case is shown in Figure 4, where \nthe mica decade is at the left.\n\nThe air condenser unit, which is of \nexceptionally rugged design, has a \nsingle stator that is common to two \nrotors. Self-adjusting bearings are used \non the rotors to compensate for wear\n\nFig. 3\u2014The resistance standard differs from others of the \nsame type chiefly in the provision of coaxial jacks and in the \ncare taken to reduce and equalize-the inductance and capact-\n\nand also to assure a low electrical con- \ntact resistance in the connections to \nthe rotors. The stator is supported on \nthree ceramic spheres, which minimize \nleakage by providing what are essen- \ntially point contacts with the frame, \nand which permit the stator plates of \nthe condenser unit to be accurately \naligned with those of the rotor.\n\nThe larger dial has a click detent \nmechanism arranged to stop the rotor \nat 100 uyuf intervals as the dial is \nturned. The detents \nare individually ad- \njustable so that the re- \nsulting steps may be \nmade to agree with the \nnominal value to within \n+o.1 pwuf. Settings \nare reproducible to \nwithin + .03 uuf. The \ncontinuously variable \ndial, at the right of \nFigure 4, carries an en- \ngine-engraved scale \nwith a vernier that al- \nlows direct reading to \nwithin 0.1 pyf.\n\nof the condenser to better than \u00ab1 \nmmf, it is essential when measuring \nhigh impedances having a large ratio \nof reactance to resistance to be able \nto adjust the condenser with much \ngreater precision. For this purpose a\n\nslow motion device operating on the \nmain dial of the condenser has been \nprovided which permits easy adjust- \nment to within .co2 Rapid pre- \nliminary adjustment to within 1 yyuf \nmay be made by means of the large \nknob on the condenser shaft in the \nusual manner, and as smoothly as \nthough the slow motion \ndevice were absent.\n\nA novel feature of \nthe bridge is the use of \nequal-length coaxial \ncords for connecting \nthe standard and un- \nknown impedances to hi \nthe balance unit. In | \naddition a small ter- | \nminal unit, fitted with - I \ncoaxial jacks, is plugged \nto the extremity of the \nleads for the test im-\n\npedance to balance the residual im- \npedances of the standards. In this way \nthe symmetry of the bridge network \nis extended to the unknown imped- \nance, and the uncertainty and the in- \nconvenience of corrections due to the \nuse of random leads are \navoided. Moreover, \nsince the potential drop \nin the sheath conductor \nof the coaxial leads is \npractically nil, the \nshields of the standards \nand of the unknown \nimpedance are brought \nto the same potentiai \nas that of the bridge \nground.\n\nThe balance unit is \narranged for both \ngrounded and bal- \nanced-to-ground meas- \nurements; the switch \njust above the bottom \ncorner of the bridge in \nFigure 1 is closed for \ngrounded measurements and left open \nfor balanced-to-ground measurements. \nFor grounded measurements only the \nupper of the two jacks on each side \nand a single cord are used to connect \nthe bridge to the standards and to the \nunknown. Such an arrangement is \nshown schematically in Figure 5. A\n\nphotograph of such a \nset-up is given in Fig- \nure 6. For balanced-to- \nground measurements \ntwo cords are used from \neach side of the balance \nunit, and the standards \nused, although similar \nto those used for \ngrounded measure- \nments, have elements \nthat are symmetrical \nwith respect to ground.\n\nFig. 6\u2014The terminating unit is in the form of a small \ncylinder with binding posts to which the unknown is attached\n\nabove, the widest range is obtained \nwith the parallel connection, where \ncapacitances from practically zero up \nto .o1 microfarad, with a wide range \nof power factor, may be measured. \nInductances that can be resonated \nwithin this capacitance range may be \nmeasured also, provided the detector \nhas adequate discrimination against \nharmonics originating in the power \nsource. An inductive standard of suit- \nable range may be substituted for the \ncapacitance where it is undesirable to \nemploy resonance methods.\n\nBell System, are attached to \ncords having usually three conductors, \nas shown in the accompanying photo- \ngraph. One conductor runs to the tip \nof the plug, another to the sleeve or \nmain body of the plug, and the third \nto the ring\u2014a small metallic band \nseparated from the dead collar and \nsleeve by narrow segments of insula- \ntion. In assembling these cords and \nplugs in the Western Electric Shops, \nsmall metal cord tips are crimped onto \nthe tip and ring conductors as al- \nready described in the Recorp.* The\n\nsleeve conductor is turned back over \nthe main body of the cord which is \nthreaded into the cord end of the plug \nthus making contact with the metallic \nbody of the plug. Small screws fasten \nthe cord tips to the terminals in the \nplug, and the outer shell is slipped \nover the assembled plug and fastened \nin place.\n\nIn service, wear on cords first be- \ncomes noticeable on the braid just \nback of the plug, where the greatest \nand most frequent bending occurs. \nContinued use of the cord would re- \nsult ultimately in serious damage to \nthe insulation and possibly breakage \nof the conductors, but the warning\n\ngiven by the fraying of the outer \nbraiding permits repair long before \nactual failure. Such frayed cords are \nrepaired by cutting the cord a short \ndistance back of the plug, beyond the \nplace where the braiding is damaged,\n\nand reconnecting it to \nthe plug. These repairs \nare often made in the \ncentral office where \nconditions do not war- \nrant the use of the \nheavy crimping ma- \nchine used for apply- \ning the solderless cord \ntips at the manufac- \nturing plant. Until re- \ncently it was the prac- \ntice to pierce the insu- \nlation with a special \nhand tool like a pair of \npliers, and push the \nterminal screw through the hole. This \nmethod had several shortcomings and \ninvestigations were started to provide \neconomical ways and means of placing \nsolderless cord tips on repaired cords \nas a maintenance operation by the \nplant forces. The outcome of this in- \nvestigation was the development of \nthe No. 444-A tool which is mounted \non the cord repair table, as shown in \nthe photograph at the top of the \nopposite page.\n\ntool employs a lever operated cam. \nThis moves the dies rapidly at the \nbeginning of the operation, when little \npressure is required, and more slowly \nas the pressure builds up. The lever is \nfairly long and massive with a heavy\n\nFig. 1\u2014Bottom: an assembled cord and plug with insulating \nsleeve or shell removed to show method of fastening conductors; \ntop and center: plug without cord and with cord end cut in\n\nhandle, which gives it a flywheel ac- \ntion as it is swung rapidly through one \ncomplete revolution. Pressures esti- \nmated to be in the neighborhood of \n1000 pounds are required to apply a \nsolderless cord tip satisfactorily, and \nthis portable tool so balances opposing \nforces that only small unbalanced \ncomponents react on the light port- \nable table itself.\n\nIn operation the outer braid and \ntinsel conductor are cut back and the \ntip and ring conductors are placed in \nthe gauge block on top of the machine. \nWith cutting pliers each is \nthen cut to its proper length. \nA tip is placed in one of the \ncavities of the die and a cord \nconductor is placed in it. The \nhandle, which has been hang- \ning vertically downward, is \nthen brought up with an ac- \ncelerated motion and turned \nrapidly through one complete \nrevolution. During the down- \nward part of the stroke, par-\n\nticularly during the last quarter when \nthe greatest pressure is applied, the \nflywheel action of the handle adds a \nvery appreciable amount to the actual \nforce applied by hand. At the bottom \nof the stroke the cam is cut back to re- \nlease the die, which is pushed back \ninto its normal open position by a\n\ngauge block are shown in Figure 2, \nWith this new apparatus solderless \ncord tips may be rapidly applied with- \nout removing the cord from the \nswitchboard. The repaired cord will \nbe essentially the same in its terminal \nconnections as a new one. The cords \nare long enough to allow several repairs \nto be made before being discarded,\n\nAssembling the elements of an experimental vacuum tube in the \nTube Development Laboratory\n\nE distributing frame has been, \nin one form or another, a part \nof the telephone system since \nthe early days. It carries terminals for \nthe incoming lines on one side and \nterminals connecting to the central \nofice apparatus on the other side. \nCross-connecting wires between term- \ninals on the two sides of the frame \nallow flexibility in associating cable \npairs with the office apparatus so that, \nfor example, a subscriber may move \nwithin the office area without having \nhis number changed. Where the wires \nchange direction, their side pull is \ncarried by insulated rings.\n\nThe old No. 1 and No. 2 rings, which \ndiffer from each other only in the size \nof the rings, consist of an iron rod three- \neighths inch in diameter, threaded on \none end and bent to form an open \ncircular ring. The threaded portion is \nused to mount the ring on the frame,\n\nwhile the ring portion with which the \ncross-connection wire comes in contact \nis insulated by a hard rubber covering. \nRings of this type have been in use for \nforty years. Old and new rings, \nmounted side by side in a central office \nduring the field trial, are shown above.\n\nThe new 9A ring, which was de- \nveloped to replace both the No. 1 and \nNo. 2 rings in new equipment, consists \nof a closed cast iron ring intermediate \nin size between the No. 1 and No. 2 \nrings. The whole ring is insulated with \na coating of vitreous enamel. A shoulder \nengages the frame so that the ring is \nheld in exact position by a single bolt.\n\nThe 9A ring offers the advantages \nof greater strength and rigidity than \nthe old ring, fixed alignment in the \nframe and a hard, smooth insulating \nfinish which will be practically im- \npervious to wear and will not abrade \nthe cross-connection wire.\n\nF. R. Dennis received the B.S. degree \nin Electrical Engineering from the State \nCollege of Washington in 1929 and joined \nthe Apparatus Development Department \nof the Laboratories in September of that \nyear. A considerable part of his time for \nthe first two years was spent making im- \npedance and attenuation measurements \non cables for carrier telephony. It became \nevident that the attenuation measuring \nsets then available were inadequate at the \nhigher frequencies where measurements \nwere beginning to be required. He there- \nfore undertook the development of at- \ntenuators and attenuation measuring cir- \ncuits which would maintain high accuracy \nover the radio frequency range. For the \npast year and a half the group with which \nhe is connected has been charged with the \ndevelopment of oscillators, detectors and \nother vacuum tube apparatus. His present\n\nwork is partly of this nature and partly \ncontinuation of work on attenuation \nmeasuring circuits and attenuation stand- \nards suitable for checking the accuracy of \nshop sets which operate in the radio fre- \nquency range.\n\nand joined the Engineering Department \nof the Western Electric Company in 1917, \nDuring the next year he was in the \ndrafting group working on various de- \nvices for detecting airplanes and sub- \nmarines. He then transferred to the \nApparatus Development Department, \nwhere, in addition to developing ap- \nparatus for the above uses, he designed \nradio apparatus for the Signal Corps. \nSince then he has remained with the \nsame group and has been in charge of \nthe development of a wide variety of \napparatus for use in the telephone plant.\n\nC. H. Younc received the degree of \nB.S. in Electrical Engineering from the \nUniversity of Michigan in 1927. He at \nonce joined the Technical Staff of the \nLaboratories where, with the electrical \nmeasurements group of the Apparatus \nDevelopment Department, he has en- \ngaged in the development of precise im- \npedance measuring equipment and resist- \nance standards with low time constants.\n\nJ. M. Barstow was graduated from \nWashburn College in 1923 with the B.S. \ndegree. He then took up graduate work at \nthe University of Kansas and received his \nM.S. degree the following year. Following\n\nthis he was employed for three years as an \ninstructor in Physics at the Kansas State \nAgricultural College. In June, 1927, he be- \ncame a member of the Department of De- \nvelopment and Research of the American \nTelephone and Telegraph Company, his \nmajor work being on noise problems. As a \nmember of Project Committee 1B of the \nJoint Subcommittee on Development and \nResearch, Edison Electric Institute and \nBell System he has conducted judgment \nand articulation tests on the interfering \neffects of various sorts of noise and has \nbeen instrumental in devising noise \nmeasuring devices. As secretary of the \nTechnical Committee on Sound Levels \nand Sound Level Meters, he took part in \nsetting up the American Tentative Stand-\n\nards for Sound Level Meters approved \nby the American Standards Association \nin February, 1936.\n\nAFTER GRADUATION by Massachusetts \nInstitute of Technology in 1914, H. A. \nAffel returned for two years as a research \nassistant in electrical engineering. In 1916 \nhe joined the engineering staff of the \nAmerican Telephone and Telegraph Com- \npany, and with the organization of the \nDepartment of Development and Re- \nsearch in 1919 became associated with its \nTransmission Development group, where \nhe engaged in studies of carrier and radio \ntransmission. In 1929 he transferred to \nthe Toll Transmission group where his \nwork included among other things the \ndesign of repeaters, program circuits, and \ncarrier systems. With the transfer of the \nDevelopment and Research Department \nto the Laboratories in 1934, Mr. Affel re- \nmained with the same group. Now as \nToll Transmission Development Direc- \ntor, he has been placed in charge of the \nentire department.\n\nern Electric Engineering Department \nwhere he worked on electrolytic con- \ndensers and later on household appliances \nand farm equipment, at that time a \nmerchandise activity of Western Electric.\n\nparatus Development Department he be- \ncame a member of the group responsible \nfor inside wires and switchboard cables. \nHere he has been concerned with a num- \nber of related problems such as the design \nof apparatus for humidity studies and for \ninvestigation of central office ventilation. \nOn account of its close association with \ndistributing frame wire, the 9A ring,\n\nWHILE AN undergraduate at the Uni- \nversity of Pittsburgh from 1914 to 1917, \nW. Herriott engaged in astronomical re- \nsearch at the Allegheny Observatory. \nThis was followed from 1917 until 1920 by \nresearch in astronomical and aerial pho- \ntography at the Eastman Kodak Com-\n\npany. Between 1920 and 1925 he was en- \ngaged in the development of optical \napparatus for microscopy, photography \nand motion pictures with the Bausch and \nLomb Optical Company. During the fol- \nlowing three years he was in charge of \nthe Scientific Department of the Fair- \nchild Aerial Camera Corporation. In 1928 \nhe joined the Engineering Department of \nthe Electrical Research Products, Inc., to \nwork on optical and photographic prob- \nlems associated with sound pictures. Mr. \nHerriott came to the Laboratories in 1929 \nand continued work in sound pictures \nuntil October, 1936. Then he transferred \nto the Materials Group of the Electrome- \nchanical Division of the Telephone Appa- \nratus Development Department.", "title": "Bell Laboratories Record 1937-04: Vol 15 Iss 8", "trim_reasons": [], "year": 1937}
{"archive_ref": "sim_record-at-t-bell-laboratories_1939-07_17_11", "canonical_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1939-07_17_11", "char_count": 85145, "collection": "archive-org-bell-labs", "doc_id": 279, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc279", "record_count": 99, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1939-07_17_11", "split": "test", "text": "it is often important to \npick up sounds coming from a par- \nticular direction to the exclusion of \nother sounds such as reflections, \naudience noise, camera click or feed- \nback from loudspeakers. For these \nsituations the microphone should be \nmore sensitive to sounds coming from \nthe desired direction. In a model \nthat has recently been developed in \nthese Laboratories, there is a possible \ndiscrimination of 20 db against un- \nwanted sounds.\n\nThe new microphone, known as the \nWestern Electric No. 639AA, uses a \ndiaphragm transmitter and a ribbon \ntransmitter in combination. The rib- \nbon type operates on the difference in \npressure developed by the sound wave \nbetween the front and back of a\n\nribbon which moves only in the direc. \ntion perpendicular to its surface. The \neffective force on the ribbon, for \nsound approaching at some angle to \nthis direction, is proportional to the \ncosine of that angle. When sound \nstrikes the ribbon on edge the pres. \nsure is the same on both sides; there \nis no motivating force and no re. \nsponse. This gives the directivity a \ntwo-looped characteristic which jis \nshown by the dotted lines in Figure 1, \nThe motion of the ribbon is in phase \nwith the motion or velocity of the air \nparticles, hence the term \u201cvelocity\u201d \nmicrophone is frequently applied to \nthe ribbon type. The control obtain- \nable with this type alone is limited, \nhowever, because unwanted sounds \nvery often come from the opposite \ndirection from the desired sounds and \nnot at right angles to them.\n\nA pressure-type microphone con- \nsists essentially of a movable dia- \nphragm closed on one side by a hollow \nhousing. When the sound pressure in- \ncreases outside of the diaphragm it \nmoves inward. This motion depends \non the pressure at the outer surface \nand not on the direction from which \nthe sound comes, since the sound \nflows around the housing. The pres- \nsure type is essentially non-directional \nand diaphragm\u2019s motion is in phase \nwith the pressure in the sound field.\n\npressure microphone is 180 degrees \nout of phase with one of the loops of \nthe ribbon microphone, that loop can \nbe cancelled. The response of the \nother loop will be increased then, be- \ncause the ribbon reverses its direction \nof motion as the source moves through \nthe plane of zero response, while the \noutput of the pressure microphone \ndoes not change. The output of the \ntwo together has the shape shown by \nthe solid line in Figure 1. Due to the \nfact that it is heart shaped, this curve \nhas been called the \u2018\u2018cardioid\u201d and \nthis characteristic is a satisfactory \napproximation to that desired for a \ndirectional microphone.\n\nCombining a ribbon and dynamic \ninstrument is complicated in practice \nbecause neither type follows com- \npletely the simple conditions as- \nsumed above. The size of the elements \nrequired to give an adequate output \nlevel is so large that the dimensions \nare not small compared with \nthe wave length of the sound \nat the higher frequencies, and \nthese sounds will not flow \naround the microphone. The \nmagnitude and phase of the \noutputs is not the same over \nthe frequency range, hence \ncancellation of one response \nloop of the ribbon microphone\n\nhighly efficient alloy steel. The hous- \ning of the dynamic unit was re- \nshaped and streamlined so that its \npresence does not affect the opera- \ntion of the ribbon. This housing en- \ncloses also the ribbon transformer and \na three-way switch as illustrated in \nFigures 2 and 6. It supports the \nwind-screen housing for the ribbon \nand provides the mounting and termi- \nnal facilities in the form of a pro- \njecting plug. The result is the micro- \nphone shown in the headpiece, a \ncompact instrument smaller than the \nusual ribbon microphone of com- \nparable efficiency. Its overall height, \nincluding the plug terminal mounting, \nis 7% inches; and it weighs approxi- \nmately 314 pounds. The housing of \nthe microphone is made of cast \naluminum finished grey with the \nhorizontal lines in polished metal. \nThe cardioid microphone can be \nmounted either on a floor or a desk\n\ncannot be achieved in simple \nfashion. By incorporating new \ndesign features and careful \nequalization, however, a satis- \nfactory solution of these diffi- \nculties was attained.\n\nA compact pressure type \nunit was already available in \nthe Western Electric 630A* or \n633Af microphones. A special\n\ndotted lines; the characteristic of the cardioid micro- \nphone by the solid line\n\nthe cardioid directional \ncharacteristic. Thus \nthe cardioid is three mi- \ncrophones in one: non- \ndirectional dynamic, bi-directional \nribbon and cardioid-directional when \nthe two elements are used together.\n\nWIND \nSCREEN MAGNET, \n> \nRIBBON-> \nPLATE \nAND \u2014> \nBAFFLE \nMAGNET, \nACOUSTIC \nYA DYNAMIC RESISTANCE \nHOUSING \n| \nDIAPHRAGM \nRIBBON \nTRANSFORMER] = \nINN = \nTERMINAL \nPLUG \nAND \nMOUNTING\n\nFig. 2\u2014Simplified cross-sectional view of \nthe cardioid directional microphone\n\nFig. 3\u2014Typical response characteristic of a combined \nunequalized-dynamic and ribbon microphone\n\nTo increase the stability of the rib- \nbon a new design was developed for \nthe velocity microphone by W. R. \nHarry of the Laboratories Staff. This \nribbon is thicker than usual and has a \ncylindrical curvature over most of its \nlength. It is corrugated at each end \nso that it acts like a bar with spring \nsupports and by this means other \nmodes of vibration than the funda- \nmental that may be encountered are \neffectively suppressed.\n\nBesides achieving a smooth re- \nsponse, which matches that of the \ndynamic both in magnitude and \nphase, the stiff ribbon reduces wind \nnoise to a level approximately ten db \nlower than that of other ribbon micro- \nphones. This permits using the micro- \nphone more freely outdoors where \nbreezes are often unavoidable. Dam- \naging the ribbon, by exposure to a \nsudden gust of wind, is also unlikely \nespecially since mechanical stops have \nbeen provided in order to prevent its \nmotion beyond the elastic limit of \nthe material.\n\nIn Figure 3 is illustrated the re- \nsponse for 0\u00b0 and 180\u00b0 incidence of a \ntypical combination of dynamic and \nribbon elements when connected with-\n\n15 over the frequency \n-80 ssamaaens range from 40 to 10,000 \ncycles and there is \n4 hard] ib] \nBan's. ardly any perceptible \n| quality change for any \nWN, angle of incidence up \nw-100 W to 120 degrees. The \nquality of the response \nat angles greater than \n120 degrees is of little \n\u2019 importance because the \n500 1000 2000 5000 19900 Sensitivity in this re- \nFREQUENCY IN CYCLES PER SECOND\n\nout equalization. The curves cross at \nabout 7000 cycles and at low fre- \nquencies they come close together, be- \ncause the electrical outputs of the \ntwo elements are out of phase at these \nfrequencies. To correct this limita- \ntion and extend the useful range to \n10,000 cycles a network was designed \nby Mr. Harry. The improvement is \nshown in the response curves of the \nequalized microphone, Figure 4. \nThe output level of the cardioid \nmicrophone is unusually high, eighty- \nfour db below one volt \nper dyne per square \ncentimeter. This level \nis four to six db higher \nthan that of the West- \nern Electric 630A or \n633A dynamic types \nand only two db lower \nthan that of the highly \nefficient Western Elec- \ntric 618A dynamic mi- \ncrophone. The new mi- \ncrophone is designed \nto operate into an im- \npedance of from thirty \nto fifty ohms. The \nnormal-incidence re- \nsponse of the cardioid \ncombination is smooth\n\ngion is low. The per- \nformance of the dy- \nnamic and ribbon mi- \ncrophones when used \nalone is similar to that of other \ncommercial instruments of their re- \nspective types.\n\nA visual indication of the variation \nof sensitivity with angle is given in \nthe chart of Figure 5 by shading from \nlight to dark. The \u201cwide pick-up\u201d \nzone of 120 degrees represents the \nregion in front of the microphone \nwhere there is practically no variation \nin quality or sensitivity. In the \n\u201cfading\u201d zone from 60 degrees to \n150 degrees on either side the sensi-\n\nFig. s\u2014Chart illustrating the directional characteristics of \nthe 639A cardioid microphone\n\ntivity changes rapidly with angle and \ncare must be exercised to keep within \nrange of the microphone. Sounds are \ndiscriminated against by approxi- \nmately twenty db in the dead zone of \n60 degrees behind the microphone. \nIn addition to these three principal \nzones, the sector from 60 degrees to \ngo degrees on either side has been \nselected as the \u201cannouncer\u2019s\u201d\u2019 or close \ntalking zone. Since the ribbon micro- \nphone contributes very little in this \nsector, the characteristic \u201c\u201cboomy\u201d \nreproduction of unequalized ribbon \nmicrophones has been avoided.\n\nIn sound studios and public addresg \ninstallations the directional charac. \nteristics of the cardioid microphone \nare particularly useful. When located \nnear a studio wall it suppresses sound \nreflected from behind; and in q \ntheatre it automatically divides the \naudience from the stage, thus elimi- \nnating noise and reducing feedback \nso that more sound reinforcement can \nbe used. The microphone has been \nthoroughly tested under practical \nconditions which have demonstrated \nits superiority in pick-up situations \nlong baffling to sound engineers.\n\nOME metals, notably sodium, \npotassium, cesium and others \nof the alkali group, liberate elec- \ntrons freely when light falls on them. \nThis behavior is the basis of photo- \nelectric cell activity; it has been \nstudied in detail not only on this \naccount but also because it bears on \nrecent theories of the metallic state. \nBetween the photoelectric and the \noptical properties of a metal there \nexists so close a relation that the \nphotoelectric behavior of a metal can \nbe predicted from its index of re- \nfraction and its absorbing power for \nlight. These optical constants can be \nfound by allowing a beam of plane \npolarized light to fall on a polished \nsurface and determining how much \nthe incident light has been elliptically \npolarized by reflection. Previous meas- \nurements of these constants for the\n\nalkali metals have been confined to \nthe visible spectrum but the Labora- \ntories has extended them into the \nultraviolet region where the maxi- \nmum photoelectric emission occurs.\n\nElliptically polarized light may be \nthought of as consisting of two beams \nof light plane polarized at right \nangles and differing both in phase \nand amplitude. With an optical de- \nvice, called a Babinet compensator, \nwhich retards one of these beams more \nthan the other, elliptically polarized \nlight can be restored to plane po- \nlarized light. The compensation re- \nquired to do this measures the phase \nshift on which the ellipticity depends. \nThe azimuth of the plane of vibration \nof the restored light differs from that \nof the incident light and the difference \ncan be measured in the visible range \nwith an analyzer of the Nicol prism\n\ntype. This prism extinguishes plane \npolarized light entirely when the plane \nof vibration is oriented at right \nangles to certain crystallographic di- \nrections in the prism.\n\nBy substituting values of the phase \nshift and azimuth in equations as \ndeveloped by Drude or by T. C. Fry \nthere may be found the index of re- \nfraction N of a metal, and the ex- \ntinction coefficient K, which measures \nthe damping the light experiences on \nentering the metal. With the values \nof n and xk, for a particular wave- \nlength there can be calculated the re-\n\nFig. 1\u2014Dispersion measurement curves of \nthe index of refraction, N, for sodium, \npotassium, rubidium and cesium\n\nflecting power, transmission for va- \nrious thicknesses, and the absorbing \npower of the metal either in bulk con- \ndition or as a thin film supported by \nanother metal. The last is of special \ninterest because the photosensitive \nsurfaces of ordinary photoelectric \ncells are of this type. Experiment \nshows that the rate of emission of \nphotoelectrons from such a surface \nis proportional to the rate at which it \nabsorbs light.\n\nIn the ultraviolet region the method \nof measurement must be modified so \nthat phase and azimuth changes are\n\naccomplished by a pair of optical \nwedges, one of right and the other of \nleft rotary quartz, which are intro. \nduced between the Babinet compen. \nsator and the Nicol analyzer. The \nresult is a rectangular pattern of dark \nspots on a bright background which \nmay be recorded photographically, 4 \ncomparison of the spot patterns for \nlight before and after reflection from \nthe metal surface shows that each \nspot of the reflection pattern has \nundergone a shift from the position it \noccupied before reflection. By analy- \nsis the x-component of the shift may \nbe identified with the phase change \nintroduced by reflection, and the \ny-component with the azimuth change, \nor the rotation of the plane of vibra- \ntion. These two measurable quan- \ntities when substituted in the Drude\n\nFig. 2\u2014Dispersion measurement curves of \nthe extinction coefficient, K,, for sodium, \npotassium, rubidium and cesium\n\nMirror surfaces of alkali metal \nhave to be prepared in vacuum and \nrotected from the air to prevent \noxidation, which dulls them. For \nmeasurementson potassium and rubid- \nium, a heavy deposit of the metal \nwas collected on a quartz plate in an \nexhausted bulb. The metal surface \nwas protected by covering it with a \nsecond plate while still in the ex- \nhausted bulb, by raising the second \nplate by magnetic action on the iron\n\nFig. 3\u2014Reflecting power at normal inci- \ndence for sodium, potassium, rubidium \nand cesium\n\ntip at the lower end of its holder. A \nstrain-free 60-degree quartz prism \nwas then sealed to the upper plate \nwith paraffin oil; and the edges were \nprotected with parafin wax, where \nthe alkali metal was exposed. Meas- \nurements could then be made in air.\n\nIn a second method, used for \nsodium and cesium, the metal was de- \nposited on an optically flat surface \nenclosed in an evacuated bulb which\n\nFig. 4\u2014Rate of absorption of energy for \nequal thickness (ca. 10\u00b0 cms.) of sodium, \npotassium, rubidium and cesium on plati- \nnum-iridium. Light incident at 60 degrees, \npolarized parallel to the plane of incidence\n\nwas provided with quartz windows. \nThe arrangement had the advantage \nthat reflection occurred at the free \nsurface of the metal under investiga- \ntion but special precautions were \nnecessary to provide that sealing and \nevacuation of the tube should not \nproduce strains in the windows.\n\nThe values of N and kK, computed \nfrom measurements of the phase shift \nand the azimuth of the restored beam \nof plane polarized light are shown in \nFigures 1 and 2. In the visible range \nsodium has the lowest index of re- \nfraction and the highest extinction \ncoefficient of the group, potassium is \nnext, followed by rubidium and \ncesium. The relative values of N and \nK, in the ultraviolet are reversed, the \nextinction coefficients being low and \nthe refractive indices higher.\n\nFigures 3 and 4 show properties of \nthe alkali metals, calculated from the \nvalues of their optical constants.\n\nFigure 3 gives the reflecting powers \nfor normal incidence. The wave length \nfor which nN =k, divides the range of \nhigh reflecting powers from that of \nlow reflecting powers.\n\nThe rate of absorption of energy by \nthin films of the alkali metals de- \nposited on platinum-iridium is illus- \ntrated in Figure 4. The absorption is \nmoderately selective in cesium; be- \ncomes increasingly so for rubidium \nand potassium, and is extremely high \nin the extrapolated range for sodium. \nThe wave length for which n =k, here \ngives the peak absorption.\n\nA comparison between the rate of \nabsorption of energy and the rate of \nemission of photoelectrons is of par- \nticular interest. Rubidium illustrates \nthis correlation. Curve 1 of Figure 5\n\nshows the photoelectric emission of a \nthin film of rubidium on platinum. \niridium and Curve 2 gives the calcy. \nlated rate of absorption of energy at \nthe surface of such a film, 10~ ems. \nthick. Part of the energy absorbeq \nat the surface of the metal is ey. \npended in effecting the escape of the \nphotoelectrons. The amount may be \ncalculated approximately by multi. \nplying the ordinates of Curve 2 by a \nfactor which depends on the longest \nwave length at which rubidium emits \nphotoelectrons. Curve 3, in which this \ncorrection has been carefully incor. \nporated, fits Curve 1 closely and go \nshows conclusively that the photo- \nelectric emission of thin films may be \naccurately predicted from a knowl. \nedge of their optical constants.\n\nFig. s\u2014Comparison of photoelectric emission and calcu- \nlated energy absorption of rubidium on platinum-iridium\n\nbetween the line and both the \ntransmitter and receiver. A function \nof the two-winding transformer, which \nhas been widely used in this set, is to \nimprove the efficiency of these con- \nnections. With such an arrangement, \nthe subscriber hears not only the \nspeech coming over the line from the \ndistant end, but also his own speech, \nsince the voltages generated in the \ntransmitter by his voice affect his \nown receiver. The sound picked up by \na transmitter and heard through its \nassociated receiver is called \u201c\u2018side- \ntone.\u201d It is possible, however, to \ndesign the circuit of a two-way tele- \nphone set so that under ideal condi- \ntions the sound picked up by the \ntransmitter is not heard through the \nassociated receiver. Such an arrange- \nment, known as an anti-sidetone cir- \ncuit, involves some complications \nover the two-winding transformer cir- \ncuit. In recent years an anti-sidetone \nsubscriber set employing a three- \nwinding transformer has come into \nwide use in the Bell System.\n\nSidetone has several detrimental \neffects on telephone transmission. A \nperson naturally regulates the volume \nof his speech by the apparent loud- \nness of his own voice in his ear. \nThe presence of sidetone increases \nthe apparent loudness, and there is a \ntendency to decrease his talking \nvolume, which reduces the level of \nthe outgoing voice signals. Sidetone \nalso interferes with incoming trans-\n\nmission, because it allows room noise, \npicked up by the transmitter, to \nmingle with and partially mask the \nincoming speech.\n\nThe first anti-sidetone circuit in \ncommercial use was invented by \nC. E. Scribner in 1893, and was used \nfor some of the operator telephone \nsets. In 1906, G. A. Campbell showed \nthat there were a large number of \npossible anti-sidetone circuits em- \nploying a single transformer and a \nsingle balancing network. To be prac- \nticable for general use, the circuit \nmust be applicable to existing condi- \ntions and apparatus, and should re- \nquire a minimum of additional ex- \npense. Considerable development \nwork was necessary to secure an ar- \nrangement which provided the de- \nsired transmission improvements, and \nwhich, at the same time, could be \neconomically incorporated in new or \nexisting telephone apparatus. The \ncircuit adopted meets these various \nrequirements and _ necessitates\u2014be- \nsides the equipment of the sidetone \nsubscriber sets\u2014an additional con- \ndenser, an additional winding on the \ntransformer, and, where the com- \nbined station set is not used, one \nextra conductor in the cord to the \ntelephone set.\n\ncurrent to the path through the \ntransmitter when the set is in use. \nOmitting the bell and the switchhook, \nwhich play no essential part under \ntalking conditions, this circuit can be \narranged schematically as shown in\n\nthe middle sketch. Since the receiver \nand the B winding of the induction \ncoil are in series, they can be inter- \nchanged in position without affecting \nthe operation of the circuit; and since \nthe condenser, under ideal conditions, \nwould be of infinite capacitance, it \nwould have no effect on the talking \ncircuit and may thus be omitted in \nthis explanation of the operation of \nthe circuit. With these two changes \nthe circuit would be as shown in the \nlower sketch.\n\nUnder talking conditions, a voltage \n\u2014marked Eo on the diagram\u2014is \ngenerated across the transmitter. Cur- \nrent flows through the transmitter \nfrom I to 2, where it divides\u2014part,\n\nmarked 1, passing out over the line. \nand part, marked Is, passing around \nthe circuit including the receiver, |; \nis the latter current that causes the \nsidetone. Received speech current \nalso will pass through both the trang. \nmitter and the receiver.\n\nFor many years a modified ar. \nrangement of this circuit, known as \nthe sidetone-reduction connection and \ninvolving only a simple interchange of \nconnections, was frequently used on \nthe shorter station loops to obtain a \nreduction of sidetone, although it js \nnot strictly what is termed an antj- \nsidetone circuit. The arrangement js \nshown in Figure 2. It gives a reduc. \ntion in sidetone of about 7 db, and a \ngain in receiving efficiency of about \n1 db, but it causes a loss of about 5 db \nin transmitting efficiency.\n\nIn the anti-sidetone set now being \nemployed, the sidetone circuit is \nmodified by shunting another path \naround the receiver, as shown in \nFigure 3. In this anti-sidetone circuit \nalso, under transmitting conditions \nshown in the lower diagram, the cur- \nrent from the transmitter divides\u2014 \npart going over the line and part \naround the local circuit. If the addi- \ntional circuit element, consisting of \nwinding c of the transformer and the \nnetwork n, can be made to carry all \nof 1s, however, there will be none left \nto pass through the receiver, and \nthere will thus be no sidetone. This is \nexactly what this additional circuit \nelement is designed to do; and al-\n\nFig. 2\u2014Schematic of sidetone reduction \ncircuit used for short station loops\n\nthough under commercial conditions \nthe action is not perfect, the sidetone \nis greatly reduced.\n\nThis condition, and the further con- \ndition that there shall be no current \nin the network during receiving, is \nbrought about by a proper design of \nthe windings A, B, and c and the \nnetwork N in relation to the im- \npedances of the receiver, transmitter, \nand line. The calculations, and the \nrelationships that must exist, are \nsomewhat complicated, but the quali- \ntative behavior of the circuit is not \ndificult to see.\n\nIn the lower diagram of Figure 3, \nwhich shows the transmitting condi- \ntion, arrows indicate the relative \ndirections of current flow at an ar- \nbitrary instant. The windings are \nso poled\u2014a and B in the same sense \nand B in the opposite sense\u2014that the \nvoltage E, tends to make current flow \nthrough the network to an amount \njust equal to the current I; flowing \nthrough winding B. Hence no current \nflows to or from the junction (4) in the \nbranch which contains the receiver. \nAnother way to visualize the situation \nis to consider that the circuit is de- \nsigned so that the voltage drop from \npoint 2 to point 4 is just equal to the \ndrop from point 2 to points I or 3. \nAs a result there is no voltage drop \nfrom point 4 to point 3 and no current \nflows through the receiver. As already \npointed out, these ideal conditions are \nnot completely met, but the sidetone \ncurrent that passes through the re- \nceiver is greatly reduced.\n\nFor received speech, the relative \ndirections of the currents in the cir- \ncuit are shown by the arrows in the \ncentral diagram. The currents in \nwindings a and B, as in the trans- \nmitting case, induce opposing volt- \nages in winding c, the voltage from \nwinding a being the greater. The net\n\nvoltage in winding c is just sufficient \nto counteract the tendency of the volt- \nage drop across the receiver, due to \nthe current Is, to send a current \nthrough the network. As a result \nthere is no current in the network \nduring receiving, a condition which \nmust be fulfilled if the circuit is to be \nideally efficient.\n\nIn the actual set, the network Nn \nis a simple resistance, and is incorpo- \nrated in the winding c, so that there\n\nis no separate physical element in the \ncircuit corresponding to N. The actual \ncircuit is shown in the upper diagram. \nInstead of using the same condenser \nfor both ringing and talking, as is \ndone with the sidetone set, two con- \ndensers are used\u2014one of small capaci- \ntance in series with the ringer and one \nof larger capacitance for the talking \ncircuit. This gives better transmission,\n\ndialing, and ringing performance, and \npermits fewer switch-hook contacts.\n\nThe new anti-sidetone set is slightly \nlower in transmitting and receiving \nvolume efficiency than is the side- \ntone set, since the ideal requirements \ncan only be partially fulfilled; but its \nlarge reduction in sidetone, amount- \ning to some 10 db, results in an effec- \ntive overall gain in transmission of\n\nWhen, on Fuly 29, 1914, Theodore N. Vail, then President of \nthe American Telephone and Telegraph Company, took his tele- \nphone receiver from its hook in New York and spoke to G. E. \nMcFarland, President of The Pacific Telephone and Telegraph \nCompany, at San Francisco, another epoch-making achievement \nin telephone communication became a fact.\n\nPrevious to 1911, dy using as large wires as were economically \npracticable, the telephone would talk reasonably well over a \ndistance about as long as New York to Chicago. With the develop- \nment and practical application of loading coils this distance was \ndoubled in 1911 and New York and Denver were interconnected. \nThe desirability, however, of a transcontinental telephone system \nhad long been apparent. It was accomplished in 1914 through \nextensive research work under the \u2018direction of F. B. Fewett. \nThere was extensive research on transmission lines, on repeater \ncircuits and on amplifiers for repeaters, all of which culminated \nin a transcontinental system.\n\nIn the early days of the New York-San Francisco line three \ntypes of repeaters were successfully demonstrated: a mechanical \nrepeater, a mercury arc repeater and the vacuum tube repeater. \nThe latter proved to be one of the most powerful and flexible \ntools in the communication art; and in succeeding years tt has \nbeen vital in carrier current transmission, transatlantic radio \ntelephony, radio broadcasting and sound pictures.\n\nHE development of modern \nbroad-band carrier systems has \nrequired extensive measure- \nments of impedance at transmission \nfrequencies much higher than those \nused for earlier carrier systems. Over \nthe usual range of impedance, these \nmeasurements are made on the five- \nmegacycle bridge recently described \nin the Recorp.* With this bridge, the \ninductance component of the im- \npedance is measured by a capacitance \nstandard, but for inductances less \nthan ten microhenrys this method is \nunsatisfactory because of the very \nlarge size of capacitance required. To \ncover this low range, therefore, a \nspecial inductometer has been de- \nveloped for use with the bridge. \nThe conventional inductometers are \nnot satisfactory at these high fre- \n~ *REcorp, April, 1937, p. 261.\n\nquencies because of the relatively \nlarge distributed capacitance of the \ncoils. In addition, the large residual \ninductance, the limitations of range, \nand the low ratio of inductance to \neffective resistance at low settings \nfurther restrict their usefulness. The \nnew inductometer avoids these diffi- \nculties by employing a helical coil \nwhose self inductance is changed by a \ntrolley contactor that varies the num- \nber of active turns between the \nterminals.\n\nAs actually constructed, two such \ncoils are employed to provide elec- \ntrical symmetry with respect to \nground, which is necessary in measur- \ning balanced circuits. A partition \nthrough the center of the cover, as \nshown in Figure 1, completes the \nshielding provided by the mounting \nplate and cover, so that there is no\n\nFig. 1\u2014A crank on the front of the inducto- \nmeter drives the two coils through non-back \nlash gears and the sprocket wheel\n\ncapacitive or inductive coupling be- \ntween the two coils or between the \ncoils and any adjacent apparatus.\n\nEach coil consists of a number of \nwidely spaced turns wound helically \non an insulating spool, and a trolley \nslides along the turns as the coil is \nrotated. The arrangement is clearly \nshown in Figure 2. Collector rings, not \nevident in the photographs, bring out \na connection from each end of each \ncoil, and the trolleys are connected to \nthe terminals connected to one end of \nthe coils. All of the contact parts are \nsilver plated to give a low and stable \ncontact resistance, and the four termi- \nnals are brought to coaxial jacks for \nconnection to the standard high- \nfrequency bridge.\n\nThe coils are rotated in unison \nthrough a crank and non-back-lash \ngears, and as the coils rotate the \ntrolleys slide along the turns of the \ncoils, giving a continuous adjustment. \nAnother set of gears on the front of\n\nthe mounting plate drives two spools \nthat carry the film on which the cali. \nbration is marked. This film jg \ncoupled to the coils by a sprocket \nwheel mounted on the same shaft as \nthe crank. The range of inductance jg \nfrom one to twelve microhenrys, and \nthe film employed is twenty feet long \nthus permitting very small changes \nto be accurately read. The smallest \ndivisions of the scale are two-tenths of \nan inch apart, and represent a change \nof 0.01 microhenry.\n\nSince with the construction em. \nployed only that part of the coil neces. \nsary for the inductance being meas. \nured is in the circuit, it is possible to \nsecure a high ratio of induction to \neffective resistance. Measurements in- \ndicate that settings are reproducible \nto within a hundredth of a micro- \nhenry and to a ten-thousandth of an \nohm of effective resistance.\n\nFig. 2\u2014A twenty-foot film scale provides \neasy reading over the entire range\n\n144th commencement on \u2018Fune 12 at which time the honorary degree of Doctor \nof Civil Law was conferred upon him\n\nA PROGRAM OF OBSERVATION and tests of the \nperformance and characteristics of the two- \nmegacycle coaxial installation between New York \nand Princeton has been under way for some time. \nThis has included measurements of modulation, \nload rating, noise and crosstalk, and studies of \ntransmission variations, pilot-channel charac- \nteristics, amplifier-switching arrangements, and \nprogram transmission. As a result of this work, \na number of modifications have been introduced \nto improve the system performance. Tests on \nthe present two-megacycle repeaters between \nNew York and Princeton will be continued \nuntil about the end of August, at which time \nthis equipment will be removed.\n\nPlans are under way for a trial, to extend the \nentire distance between New York and Phila- \ndelphia, of new repeater equipment designed to \nhandle a top frequency of either about 2 mega- \ncycles for telephone or about 3 megacycles for \ntelevision. This new equipment will be of the\n\nGeorge A. Campbell (right) and Fohn R. Car- \nson at the Medal Day exercises of The Franklin \nInstitute, held in Philadelphia on May 17, \nat which they received Elliott Cresson Medals\n\ntype which it is planned to install for commercial \nservice between Stevens Point and Minneapolis, \nThe new repeaters will be spaced approximately \nfive miles apart. Construction of this new repeater \nequipment is already under way and installation \nbetween New York and Philadelphia will be \nstarted early in July. Later, new terminal equi \nment will be tried out in conjunction with the \n3-megacycle repeater equipment.\n\nTHE HONORARY DEGREE of LL.D. was con- \nferred on W. H. Harrison, vice-president and \nchief engineer of the American Telephone and \nTelegraph Company and a director of the Lab- \noratories, by the University of Notre Dame \nduring the graduating ceremonies held on June \n5. Mr. Harrison delivered the commencement \naddress in which he gave a general discussion of \nengineering.\n\nMr. Harrison has also been appointed a direc- \ntor of The Bell Telephone Company of Canada.\n\nThe National Committee has received the \nreports of nineteen cases in which local Com- \nmittees of Award in the various Operating \nCompanies have awarded twenty-two Theodore \nN. Vail bronze medals. These make a total of \n1093 medals awarded since the establishment of \nthe Fund in 1920. The National Committee has \nselected from these cases the following for \nfurther award:\n\nA silver medal, with a cash award of $250, to \nJerry Frank Kincannon, exchange repairman, \nSouthwestern Bell Telephone Company, Miami, \nOklahoma, \u201cfor courage, initiative and good \njudgment in effecting a rescue and for unusual \ncompetence in the use of first aid under trying \nand hazardous conditions.\u201d\n\nA similar award to George F. Wilson, installer- \nrepairman, The Pacific Telephone and Tele- \ngraph Company, Grass Valley, California, \u201cfor \nconspicuous courage, persistence and fortitude \nin an extreme effort to save the life of a fellow- \nemployee under very hazardous conditions \nduring a blizzard, while engaged in the restora- \ntion of telephone service.\u201d\n\nBronze plaques were awarded, in commemora- \ntion of these acts, to both Companies.\n\nReading clockwise around the table from the extreme left: H. E. Marting, H.C. Spryer, A. W. \nBates, L. W. Tucker, I. H. Moore, H. H. Lowry, H. H. Glenn, L. H. Fohnson, F. R. Shea, \nC. G. Stoll (guest), F. M. Williams, O. E. Buckley (guest), D. D. Miller, A. G. Hall, R. L. \nQuass, L. T. Marks, G. Dobson, C. W. Lowe, F. R. MacGregor, E. H. Schroeder, F. C. Wright\n\nA special bronze plaque has been awarded to \nthe employees of the Bell System in recognition \nof loyalty, devotion to duty and achievement \nduring and after the hurricane last September.\n\nOn JuNE 9, nineteen members of the Haw- \nthorne student group of 1909 lunched together \nat West Street with O. E. Buckley and C. G. \nStoll attending as guests. This class was one of \nthe first large groups to take the students\u2019 \ntraining course then offered young college gradu- \nates who elected to enter the telephone field. \nThe original group numbered 65 men of whom 31 \nare still in the Bell System. Most of the members \nof the group were hired by G. A. Anderegg who \nretired from active service in the Laboratories \nlast year. The Laboratories\u2019 men shown in the \naccompanying photograph, taken during the \nluncheon, are:\n\nG. Dobson H. H. Lowry \nH. H. Glenn H. E. Marting \nA. G. Hall D. D. Miller \nL. H. Johnson R. L. Quass \nC. W. Lowe H. C. Spryer \nJ.C. Wright \nCoLLoguiuM\n\nR. Bown spoke at the last meeting of the \n1938-1939 season of the Colloquium on May 22. \nHe reviewed some of the most recent develop-\n\nments in short-wave telephony, such as the \nmusa and single-sideband transmission and dis- \ncussed current problems, including horizontally \ndeviated wave propagation and the development \nof a truly multiplex system.\n\nTue ResearcH Department of the Labora- \ntories is making a statistical study of hearing in \nthe United States by securing photographic \ncopies of the musical tone test blanks at the Bell \nSystem exhibit at the World\u2019s Fairs. Members of \nthe Laboratories who have taken the test may \nassist materially in this study. Simply write \nyour name and room number on your test blank \nand mail it to H. C. Montgomery, Room 857, \n463 West Street. This will be returned to you \npromptly and you will be sent a brief question- \nnaire for additional information to aid in the \ninterpretation of the data. If you are interested \nin following your test at the Fair with a more \ndetailed laboratory test, Mr. Montgomery will \narrange for such a test insofar as facilities are \navailable. This request applies only to the test \nfor musical tones and not for spoken words.\n\nOn June 5, Dr. Jewett gave the commence- \nment address, entitled Forty-One Years After \nGraduation, at the University of Nebraska.\n\nA. F. Drxon and H. M. Bascom visited Johns- \ntown, Pennsylvania, on May 22 to inspect an \nall-relay automatic exchange which was manu- \nfactured and installed by the North Electric \nCompany for the non-Bell telephone company \nthat gives service in Johnstown. This is the \nlargest exchange of this type which has ever \nbeen furnished.\n\nR. H. Witson has been appointed General \nService Manager, succeeding J. S. Hartnett \nwho died on May 1. Reporting to Mr. Wilson \nare D. R. McCormack, Central Service Man- \nager; E. V. Mace, Local Service Manager; and \nR. E. Merrifield, Merchandise Manager. In the \nPlant Department, C. W. Lowe succeeds E. V. \nMace as Building Layout Engineer. At the time \nthese appointments became effective, May 29, \nthe Library was transferred from the General \nService Department to the Bureau of Publication, \nwith Miss L. E. Smith, Librarian, reporting to \nJohn Mills, Director of Publication.\n\nA GROUP LUNCHEON of the supervisors of the \nApparatus Development Department was held \nat the Hotel Abbey on May 25. Following the \nluncheon, R. L. Jones, presiding at the meeting, \nintroduced the speaker, W. H. Harrison, who \nfirst summarized progress made in the telephone \nbusiness during the past ten years and then \ndiscussed the problems of the future. The com- \nmittee in charge of arrangements consisted of \nP. S. Darnell, J. B. Dixon, E. C. Edwards, \nR. C. Koernig, G. Puller, W. H. Sellew, O. A. \nShann, H. D. Wilson, Jr., and J. M. Wilson.\n\nstrated during the Conference of Presidents of \nthe Associated Companies by A. B. Bailey, \nW. K. Caughey and H. A. Hay.\n\nH. T. BupEnzBom was in Washington on radio \nequipment problems of the Navy Department.\n\nG. N. Tuayer and W. H. Donerry recently \nvisited the Hartford plant of the General \nElectric Company.\n\nT. E. Lenican supervised the installation of \nthe s00-watt ultra-high-frequency broadcast \nstation and the conversion of a 6B radio trans- \nmitter for high-fidelity operation at Brooklyn \nTechnical High School.\n\nF. E. Nimmcke visited Topeka, Kansas, where \nhe inspected the installation and adjusted the \n355E-1 radio transmitting equipment (5 kilo- \nwatts) at Station WIBW.\n\nE. L. Owens visited the Raytheon Manv- \nfacturing Company at Waltham, Massachusetts, \nto discuss speech input equipment. _\n\nR. H. Miter, on the twenty-seventh of May, \ncompleted twenty-five years of service in the \nBell System. Mr. Miller joined the Western \nElectric Company at West Street in 1913 and \nalmost immediately was sent to the Installation \nDepartment at Baltimore on the installation and \ninspection of manual central-office equipment. \nFor the next few years he worked for the Instal- \nlation Department along the entire Atlantic \nseaboard. Late in 1916, following a ten-month \nleave of absence, he transferred to the Providence \nTelephone Company on central-office engineering \nand inspection. During the World War he was\n\nwith the 4o1st Telegraph Battalion, formed by \nthe Bell System, and spent fourteen of his \ntwenty-two months of service in France.\n\nUpon his return to this country, Mr. Miller \nspent a short time with the Holmes Electric \nCompany, a subsidiary of the New York Tele- \nphone Company, and then entered what is now \nthe Equipment Development Department of the \nLaboratories. Following a period of special \ntraining both at New York and Hawthorne, he \norganized routines for handling relations between \nthe engineering staff and the clerical force. After \nthis he spent considerable time with various \noperating companies aiding them in preparing \njob specifications for newly developed equipment \nto be purchased from the Western Electric Com- \npany. From 1925 to 1931 he was the Labora- \ntories\u2019 cobrdination engineer handling the instal- \nlation of No. 3 toll offices at Cleveland, Detroit \nand New York. Since 1931, Mr. Miller has been \nin charge of the toll current-engineering group of \nthe Equipment Development Department.\n\nA FIVE-STAR SERVICE EMBLEM signifying the \ncompletion of twenty-five years of service in the \nWestern Electric Company and the Labora- \ntories was awarded to Frank Wallenius on the \ntwenty-fifth of May. Mr. Wallenius joined the \nWestern Electric Company at New York in 1912 \nand worked on printing telegraph machines. \nWhen the manufacturing organization trans- \nferred to Hawthorne a short time thereafter, he \nalso went as a group supervisor on the production \nof printing telegraph machines. He left the com- \npany in 1916 and for the next two years worked \nas a foreman for the Sperry Gyroscope Company \nin Brooklyn.\n\nMr. Wallenius then came to the Development \nShop of the Western Electric Company where he\n\nhas since spent most of his time on the making \nof accurate instruments and intricate tools and \ndies. For the past ten years he has been in the \nprecision room. Some of the special projects on \nwhich he has worked include the making of light \nvalves used for the recording of sound pictures \nand picture transmission, metal graphs for the \noptical-tone generator, and the contour machine \nfor grinding styli to a radius of 0.001 of an inch.\n\nFrank Howarp GranamM, who completed \ntwenty-five years of Bell System service on the \ntwenty-seventh of May, graduated from Penn- \nsylvania State College in 1914 with the B.S. in \nE.E. degree. During the summer of 1913 he \nworked for The Bell Telephone Company of \nPennsylvania on a field survey of plant equipment \nwhich included a study of property rights. Fol- \nlowing his graduation he joined the student \ncourse of the Western Electric Company at Haw- \nthorne and after his shop training gained his in- \nstallation experience at Philadelphia, and at \nWilmington where he worked on the first call- \ndistributing central office. He then came to the \nPhysical Laboratory at West Street and late in \n1915 transferred to the Research Department \nwhere he engaged in the development of vacuum \ntubes, working with the tubes made for the \nArlington transatlantic tests.\n\nThe next year Mr. Graham joined the Trans- \nmission Department but when the World War \nbroke out he was sent to Boston to work on the \ncable system for detecting submarines, par- \nticularly on balancing problems. Towards the \nend of 1917 he returned to West Street to aid in \nthe production of vacuum tubes for the govern- \nment. Following the war he joined the acoustical \nresearch group where, until 1929, he was con- \ncerned with the development of audiphones,\n\naudiometers and hearing aids and with general \nresearch in hearing. From 1929 to 1937 he was \nwith Electrical Research Products, Incorporated, \non general transmission and engineering of \nstudio recording and theater reproducing equip- \nment for sound pictures. Since his return to the \nLaboratories in November, 1937, he has been \nengaged in field studies to determine the optimum \ntransmission requirements for operators\u2019 tele- \nphone sets and associated circuits.\n\nR. A. Mitter conferred with engineers of the \nYankee Network at Boston on June 8 on ampli- \nfier systems for radio broadcasting.\n\nT. H. Crastree visited the Pittsburgh Coal \nCompany at Pittsburgh on June 1 in connection \nwith field studies of the 10A magnetic telephone \nin mine telephone systems.\n\nH. C. Rusty visited the plant of the Art Metal \nConstruction Company at Jamestown, New \nYork, on projects involving metal furniture.\n\nA. C. WaLKER visited the National Bureau \nof Standards in Washington in connection with \ntextile drying research.\n\nE. B. WHEELER discussed lamp problems at \nthe Nela Park Laboratories of the General Elec- \ntric Company, Cleveland. He also visited Haw- \nthorne on matters pertaining to lamp problems.\n\nD. A. Quartes, with D. M. Taggart of the \nWestern Electric Company, visited telephone \nadministrations and various manufacturing plants \nin England, Belgium, Holland, Denmark, \nSweden, Germany and France, studying outside \nplant designs with particular attention to cables \nand cable manufacture.\n\nW. S. Hayrorp and C. SHAFER, JR., visited the \nNew England and Southern New England Tele- \nphone Companies in connection with the intro- \nduction of improved joints in outside distributing \nwires and station wiring.\n\nJ. B. Dixon, with representatives of the Long \nLines Department, studied the corrosion of line \nwire on the Petersburg-Georgetown line near \nCovington, Virginia.\n\nR. J. Kent, at the Bond Electric Company in \nNew Haven on May 22, discussed design features \nof flashlights.\n\nC. H. Amapon, in Toronto, discussed methods \nof pole-line inspection and maintenance. Later \nhe went to Denver to conduct experiments on \nthe full-length creosoting of lodgepole pine poles. \nHe also inspected some of the first creosoted \npoles of this type that have been in service in the \nterritories of the Northwestern and Mountain \nStates Companies.\n\nG. Q. Lumspen, with Messrs. Walseth and \nStratton of the Chesapeake and Potomac Tele- \nphone Company, examined experimental salt- \ntreated southern-pine poles on a recent trip to \nthe Norfolk, Virginia, area.\n\nJ. G. SEGELKEN attended the meeting of the \nTransmission and _ Distribution Committee of \nthe Edison Electric Institute at Chica\n\nMadison, Wisconsin, to discuss methods of \nmeasuring the electrical resistance of poles, \nW. H. S. Youry recently conferred with repre- \nsentatives of the Food and Drug Administration \nin Washington on the labeling and packaging of \nfirst aid and central-office emergency supplies\n\nA. H. Hearn, at Montreal and Ottawa, i. \nserved the effect of various types of creosote treat. \nments on the surface condition of red pine poles\n\nC. D. Hocker and R. H. Cottey, with repre \nsentatives of The Bell Telephone Company of \nPennsylvania and the Philadelphia Electric \nCompany, inspected certain salt-treated poles in \nPhiladelphia.\n\nR. L. Jones and H. A. FReperick visited the \nUnited Shoe Machinery Corporation at Beverly \nMassachusetts.\n\nB. B. Grace of the International Standard \nTelephone and Cables, Limited, London, visited \nthe Laboratories to discuss station apparatus.\n\nL. Devaux of Le Materiel Telephonique, \nParis, discussed station and other apparatus \nwhile visiting the Laboratories.\n\nAt Greenwicx, Connecticut, C. H. Greenall, \nG. H. Downes and P. T. Higgins observed step- \nby-step wipers and banks under field-test con- \nditions. Mr. Higgins also visited the telephone \noffice at Roanoke, Virginia, to study main- \ntenance problems on step-by-step banks.\n\nC. H. GrEENALL attended a meeting of the \nMetallurgical Advisory Committee of the Na- \ntional Bureau of Standards held at Washington.\n\nSWITCHBOARD PLUGS were investigated by \nT. S. Huxham, P. Neill, R. G. Ramsdell and \nM. Fritts in a Jersey City central office.\n\nOn May 5, W. Fondiller, A. R. Kemp, H. H. \nGlenn, D. R. Brobst and C. S. Fuller, together \nwith G. R. Brown, W. S. McGill, J. S. Little and \nT. G. Stover of the Western Electric Company, \ndiscussed insulating materials with engineers of \nthe du Pont Company in Wilmington.\n\nAMONG MEMBERS of the Laboratories who have \nrecently made trips to the Hawthorne plant of \nthe Western Electric Company were H. A. \nFrederick, H. O. Siegmund, J. R. Townsend and \nJ. J. Kuhn on the development of new materials \nand apparatus; B. O. Templeton, on manufac- \nturing problems of coin collectors; N. R. Stryker, \nacoustical testing equipment for the combined \ntelephone sets; W. G. Laskey, tool design for \nelectric meters; R. Burns, handset molding prob- \nlems; D. H. Gleason and C. R. Moore, new dial \ndevelopments; J. E. Shafer, the manufacture of \nan improved relay; A. F. Bennett, W. L. Tuffnell \nand D. T. Eighmey, handset and instrument \nproblems; and H. W. Heimbach, problems that\n\nMEMBERS OF THE LABORA- \nwho presented papers \nat the annual meeting of the \nAcoustical Society of Amer- \nica, held in New York on \nMay 15 and 16, were Harvey \nFletcher who spoke on Audi- \ntory Patterns, Homer Dudley, \nRemaking Speech, and R. N. \nMarshall and W. R. Harry, \nNew Microphone Providing \nUniform Directivity Over an \nExtended Frequency Range.\n\nJ. A. Becker spoke on \nElectron Microscopes and \nSome of Their Uses at the \nfinal meeting of the Radio \nColloquium, held at the \nHolmdel laboratory. He dis- \ncussed various forms of electron microscopes and \ntheir use in varied fields and demonstrated a \nsimple microscope with a magnification of about \n300,000. At this meeting the following officers \nwere elected for the 1939-1940 year: A. E. Bowen, \npresident, A. E. Kerwien, secretary, and J. P. \nSchafer, chairman of the program committee.\n\nAr THE Partin, New Jersey, plant of the \ndu Pont Company, B. L. Clarke, H. G. Arlt \nand C. C. Hipkins discussed finishes.\n\nJ. G. Cuarree presented a paper, The Appli- \ncation of Negative Feedback to Frequency-Modu- \nlation Systems, at the May 3 meeting of the \nNew York Section of the I.R.E.\n\nA. J. AHEARN presented a paper, Electron \nMicroscopes and Some of Their Uses, before the \nAlpha Chapter of Epsilon Chi at Columbia.\n\nFrancis WiLtiaMs, for many years a watch- \nman at the Section K entrance on Bank Street, \ndied on May 27. Mr. Williams was born in \nIreland and came to this country in 1909. He \njoined the Western Electric Company on \nOctober 7, 1919, as a night cleaner and night \nwatchman. In 1929 he was transferred to the \nday shift where he had since been a uniformed \nwatchman assigned to building entrances.\n\nApotpH BREGARTNER, a technical assistant \nin the Central Office Switching Development \nDepartment with over forty-one years of service \nin the Bell System, died on June 4. Mr. Bre- \ngartner joined the Western Electric Company \nin New York in 1898, and after a few months in \nthe machine shop, entered the group handling \nthe assembly and adjustment of relays. Later he \nworked on special apparatus, such as the early \nequipment for train dispatching, and then be-\n\ncame associated with the development of semi- \nmechanical central office apparatus. For nearly \nfour years he took part in the early installations \nof this equipment in Newark. Since then he had \nbeen handling the adjustment and assembly of \nrelays in the Systems Circuit Laboratory.\n\nOn May 8, K. K. Darrow spoke on The Origin \nof the Cosmic Rays before the student body at \nWellesley College.\n\nV. J. AtBano and J. H. Gray visited Pitts- \nburgh to discuss with engineers of The Bell \nTelephone Company of Pennsylvania various \nmethods of duct sealing and to inspect duct- \nsealing materials.\n\nR. M. Burns spoke on The Tarnishing of \nMetals at the meeting of the Metallurgy Collo- \nquium held at M.LT.\n\nH. W. Hermance and T. F. Ecan visited \nPittsburgh, Cincinnati, Cleveland, Detroit and \nSt. Louis where they made chemical studies of \nconditions affecting performance of contacts of \npanel bank terminals. A portable chemical lab- \noratory was used in these studies and a photo- \nmicrographic record was made of the terminals.\n\nR. L. Lunsrorp and J. Zouter were at Atlanta \nand J. G. Ferguson at Harrisburg, Cleveland \nand Detroit to discuss installation problems of \nthe 355A community dial office.\n\nJ. H. SoLe went to the General Electric Com- \npany in Fort Wayne, Indiana, to discuss alter- \nnators and control equipment. He also visited \nthe telephone company and the manufacturers of \ndiverter pole generators in Cleveland.\n\nSTUDIES OF GASOLINE and Diesel engines took \nA. E. Petrie to Portland, Boston, Cleveland and \nLansing, F. F. Siebert to Baldwin, Pennsylvania,\n\nand V. T. Callahan to Portland, Boston, Lan- \nsing, Cleveland, Canton and Salem (Ohio).\n\nL. M. Datutwore, of the International Stand- \nard Electric Company, London, is at the Lab- \noratories in connection with the engineering of \ncable-carrier installations in Finland and Sweden.\n\nA. G. Lane spent several days at Cleveland in \nconnection with the initial installation of the \n3B toll switchboard.\n\nH. D. Canttu was in Albany testing circuits \nmodified to meet new requirements.\n\nW. H. ScuEeER spent several weeks at Jones- \nville, New York, testing the first installation of \nthe 380A unattended crossbar dial office.\n\nT. L. Dimonp was at Batavia, Ohio, for a week \nmaking tests on the first installation of the 355 \nstep-by-step community-dial office.\n\nR. W. Harper, at Cincinnati, inspected the \ninstallation of 355A dial-office equipment.\n\nAt Cuicaco, W. J. Lacerte, C. V. Taplin and \nE. L. Getz spent several days discussing with \nengineers of the Illinois Bell Telephone Company \nplans for making through tests on the first \ncrossbar offices to be installed in Chicago.\n\nF. A. Brooks has been at Toledo in con- \nnection with the field trial of a pilot-channel \ndeviation regulator for type-K carrier systems.\n\nS. Rosen and J. B. Irwin were at Princeton \nfor tests of a line-switching circuit for the coaxial \ncable between New York and Philadelphia.\n\nA. J. AIKEns is in Winnsboro, South Carolina, \ntesting crosstalk in type-J systems.\n\nL. Y. Lacy made several trips to Princeton in \nconnection with noise studies on coaxial cables.\n\nR. M. Hawexortre has been in Morristown \nwhere he took part in a co\u00e9perative inductive \nco\u00e9rdination study involving telephone circuits \nof the New Jersey Bell Telephone Company.\n\nJ. H. SHunart and C. H. Gorman, Jr., have \ngone to Eau Gallie, Florida, where they are \nmaking open-wire crosstalk tests on type-J \ncarrier-system circuits.\n\nE. S. Witcox and J. L. Linpner at Amarillo, \nTexas, made J-frequency crosstalk tests on \nopen-wire circuits of the Fourth Transconti- \nnental Line.\n\nO. H. Coo.ipce participated in tests on a \ncable being installed near Danielson, Connecti- \ncut, to determine comparative results for voice- \nfrequency cable when installed in 750-foot \nversus 1500-foot lengths.\n\nL. C. Rosperts spent a week in Charlotte \nmaking tests preliminary to starting a field trial \non a ground-potential compensator circuit. \nH. W. Reikel of the Long Lines Engineering \nDepartment accompanied Mr. Roberts.\n\nWitH THE compLeETIon of high-frequency \ntransmission tests on open-wire lines at Mount\n\nH. S. Winsicter and J. C. Lozier were in \nReading, Pennsylvania, during the week of Ma \n27 in connection with tests of volcas as a aa \nwire echo suppressor.\n\nTHE TRIAL of negative coefficient thermistors \non non-loaded cable circuits was observed at \nLong Valley by L. G. Abraham, J. A. Becker \nP. G. Edwards, and R. D. Fracassi. ;\n\nH. A. Erueripce, Jr., and O. H. \nof the Laboratories, and A. C. MircuE t of the \nLong Lines Department, were at Danielson \nConnecticut, on tests of combined capacitance \nunbalance and capacitance deviation.\n\nAt Gens Fatus, New York, A. E. Bachelet \nE. B. Mechling, and J. T. Schott tested the \nefficacy of proposed means for suppressing clicks \ncaused in program circuits by operation of relays \nin the office, particularly those in transmission \nregulating equipment.\n\nOn May 19, L. E. Coon made a tour of the \nDeal and Holmdel Radio Stations and discussed \nplant equipment and procedures from the stand- \npoint of safety engineering and health.\n\nA. R. THompson has been elected a director of \nthe American Institute of Graphic Arts, for a \nterm of three years. Mr. Thompson is chairman \nof the Institute\u2019s Textbook Clinic, a group of \nsome two hundred publishers, authors, de- \nsigners, and printers organized to study the \ndesign and manufacture of educational books.\n\nDurinc May, E. W. Adams visited the Haw- \nthorne plant with T. Brooke Price, F. T. Wood- \nward, R. R. Adams and E. J. Driscoll of the \nWestern Electric Company and J. R. C. Palmer \nof Electrical Research Products, Incorporated.\n\nTue LaBoraToRIEs were represented in inter- \nference proceedings by S. B. Kent attending the \ntaking of testimony in Camden, New Jersey; \nby R. J. Guenther at a hearing before the Board \nof Appeals; and by H. S. Wertz at a hearing \nbefore the Primary Examiner at the Patent \nOffice in Washington.\n\nSINGAPORE came within reach of Bell and Bell- \nconnecting telephones on June 1 when radio \ntelephone service was extended to Malaya. The \nnew service is provided by means of two short- \nwave radio circuits\u2014one 9,100-mile channel \nbetween San Francisco and Bandoeng in Java, \nover which service is maintained with the \nNetherlands Indies, and another, 720 miles long, \nwhich reaches from Bandoeng to Kuala Lumpur \nin the Federated Malay States. Wire lines extend \nfrom Kuala Lumpur to the Straits Settlement, \nof which Singapore is the capital, and to most of \nthe Malay Peninsula.\n\nN the early manual systems, all \nringing was under the control of \nan operator. She would operate a \nkey to start it, and restore the key \nto stop it. Manual ringing has long \nbeen replaced by machine ringing on \nall but small, light-traffic boards. \nWith this system the operator starts \nthe ringing, as before, but it is cut \noff automatically when the  sub- \nscriber answers by a \u201ctripping\u201d relay. \nThe ringing current passes through its \nwinding, but the relay must not oper- \nate on it under even the most unfavor- \nable conditions, such as a very short \nline and two ringers bridged across it. \nWhen the subscriber answers, how- \never, the relay must operate to stop \nthe ringing, again under the most un- \nfavorable conditions of a line that is \nextremely long.\n\nA ringing circuit is shown in simpli- \nfied form in Figure 1. Under ringing \nconditions, relay A is operated and \nringing current passes through the \nwinding of the tripping relay, a front \ncontact on relay a, the \u201cring\u201d side of\n\nWhen the receiver is lifted from the \nhook, the condenser and ringer are \npractically short-circuited by the \ntransmitter, and the lower impedance \nresulting should allow enough current \nto pass to operate the tripping relay. \nWith a maximum loop resistance of \n750 ohms, including the station re- \nsistance, however, and some 250 ohms \nin the relay winding and the rest of \nthe circuit, the current is only about \n100 milliamperes, which as noted\n\nabove is the value at which it should \nnot operate. To insure satisfactory \ntripping, therefore, two things are \ndone. One consists in putting a bat- \ntery in series with the ringing ma- \nchine; and the other, in making the \ntripping relay more sensitive to di- \nrect than to alternating current. This \ndoes not affect the ringing, since the \nd-c component is blocked by the con- \ndenser, but it increases the current \nthat flows when the call is answered, \nsince the path through the trans- \nmitter has no condenser in series. \nRinging is supplied through an \ninterrupter that gives a short interval \nof ringing followed by a longer inter-\n\nval of silence. To make it possible for \nthe tripping relay to operate during \nthe silent as well as during the ringing \ninterval, battery is also connected \nduring this silent interval. A simpli- \nfied schematic diagram of the ar- \nrangement is shown in Figure \n2. To make the tripping relay \nless sensitive to alternating \nthan to direct current, a cop- \nper sleeve is placed over the \ncore. With alternating current \nin the winding, an opposing \ncurrent is induced in this \nsleeve, so that the net effect of \nthe current is less, while with \ndirect current in the winding, \nthe sleeve has little effect. As a\n\nresult, the relay is from three to five \ntimes more sensitive to direct current \nthan it is to alternating. The lowest \noperate current\u2014or a 750-ohm ex. \nternal circuit loop during the silent \ninterval\u2014is about 50 milliamperes d-c; \nthe highest non-operate current\u2014for \na minimum length of loop and two \nparallel ringers\u2014is some 100 milli- \namperes a-c. Since the relay is at least \nthree times more sensitive to d-c than \nto a-c, however, it can readily be made \nto operate on 50 milliamperes d-c and \nnot to operate on 100 milliamperes a-c.\n\nIn recent years there has been a \ntrend toward the use of smaller con- \nductors in cables, with the result that \nthe resistance of subscriber loops has \ntended to increase. Because of this \ntendency, the new crossbar system \nhas been designed to give service over \n1500-ohm external circuit loops. As \npart of the crossbar development, \ntherefore, it has been necessary to \nprovide a tripping system for this in- \ncreased range. With a 1500-ohm loop, \nthe operate current under the most \nunfavorable conditions will be only \nabout 25 milliamperes. The former \nrelay, if adjusted to operate on this \nvery small current, would have un- \nstable contact, and so it was necessary \nto redesign its structure to provide the \nnecessary contact stability at 25 \nmilliamperes. Since this requires that \nthe relay be more sensitive to direct\n\ncurrent, something \nalso had to be done to \nmake it less sensitive \nto alternating current.\n\nThis was accom- \nplished by a design \nmodification due to H. \nN. Wagar. Its major \nfeatures are shown in \ncomparison with those \nof the former relay in \nFigure 3. Both relays \nare alike in having a \ncentral core on which \nthe winding is placed, \na cylindrical iron shell \nsurrounding the wind- \ning, and a circular armature that \nbridges the annular space between the \ncore and the outer shell. In the earlier \nform a copper sleeve was placed over\n\nIn the new relay, known as the \n114KA, the sensitivity is increased by \nan enlarged pole piece at the end of\n\nFig. 4\u2014The 114KA tripping relay with armature removed \nto show the copper ring around the pole piece\n\nthe core nearest the armature. This \nlarger area reduces the reluctance of \nthe air gap, and thus permits a larger \nflux to flow for the same magnetizing \nforce. This larger flux, in turn, results \nin a greater pull. To prevent this in- \ncreased sensitivity from being effec- \ntive on alternating current, a large \ncopper ring is placed adjacent to the \nadded pole piece to allow a larger \nopposing current to flow. This added \ncopper ring is evident in Figure 4.\n\nThis modified construction has not \nchanged the general appearance of the \nrelay, so that the new relay\u2014shown \nin the photograph at the head of this \narticle\u2014looks like the former one \nexcept for a few minor changes in \nspring and contact arrangement. A \nfront contact has been provided in \naddition to the back contact, but it \nis used in only a few cases. With this \nnew relay, the ratio of a-c to d-c cur- \nrent necessary to operate is from \nseven to fifteen\u2014a substantial in- \ncrease over the former ratio.\n\nS has been explained in a pre- \nvious article, the functions of \nterminating markers are in\n\ngeneral similar to those of originating \nmarkers. The differences arise chiefly \nfrom the difference in type of circuits \nwhich they control. While the origi- \nnating markers must find an idle \ntrunk out of several hundred groups of \ntrunks, the terminating marker must \nbe able to find one particular line out \nof a possible ten thousand. It must \nfirst test this line for busy, and, if the \nline is found idle, must then establish \na connection to the crossbar switch to \nwhich that line is run, so that it may \nfind an idle path to it from the in- \ncoming trunk, and in this manner \nestablish the talking connection.\n\never, may be that of a private branch \nexchange, instead of an individual or \nparty line. In such a case it is not \nsufficient to test only one line; since \nthere will a group of trunks to the \nPBX, the marker\u2019s task is to select the \nfirst one in the group that is idle. One \nof the fundamental features of the \ncrossbar system is to test a group of \npaths and trunks simultaneously so as \nto reduce the holding time of the \nmarkers to a minimum. To carry out \nthis principle, the terminating marker \nis arranged to test twenty lines at a \ntime where the number called is that \nof a PBX; while when it is that of \nan individual line, only the desired \nline will be tested, although the entire \ngroup of twenty consecutively num- \nbered lines, including the desired one, \nwill be brought to the marker.\n\nThe possible 10,000 subscriber num- \nbers of the office are therefore ar- \nranged in blocks of twenty, and a \nblock relay is provided for each such \ngroup. In function the block relays \nare similar to the route relays of the \noriginating marker, but they are con- \ntrolled and operated in a different \nmanner. The route relays are part of \nthe originating marker, and one or \nmore for each office code is provided in \neach marker. The block relays, on the \nother hand, are not part of the termi- \nnating marker, and there is only one \nfor each group of twenty subscriber \nnumbers in the office. They are multi- \ncontact relays assembled ten in a row,\n\nand four such rows are mounted on \nthe upper part of a frame as shown in \nthe photograph at the head of this \narticle. When operated, each block \nrelay closes sixty contacts, three for \neach of twenty line numbers. Twenty \nof these, one for each line, are in the \nleads used for the busy test. There \nmay be as many as 500 of these relays \nin an office, and, to simplify the selec- \ntion of the desired one, they are \ngrouped into sets of five, and operated \nthrough hundred-block relays. Each \nof the latter has five contacts, one for \nthe winding of each of the twenty- \nblock relays in that hundred-block. \nThese hundred-block relays \nare mounted under one of the \nlong cans that are mounted \nabove the four rows of twenty- \nblock relays.\n\nSeveral of the hundred-block \nrelays are formed into a \u201c\u2018num- \nber group\u201d and by means of a \nnumber-group connector can \nbe temporarily connected to \nany one of the terminating \nmarkers in the office. The \nnumber-group connector re- \nlays are of the same type as \nthe twenty-block relays, and \nare also assembled in rows of \nten, eight rows being mounted \non a frame as shown in Figure \n1. Four of these relays, oper- \nated simultaneously, are re- \nquired to connect a number \ngroup to one marker\u2014the four \nbeing mounted one above the \nother. Since each number \ngroup must have access to all \nof the terminating markers, of \nwhich there may be as many \nas ten, ten sets of four relays \nmay be furnished for each \nnumber group. Each frame of \nFigure 1 has connector relays Fi \nfor two number groups, the\n\nThis division of lines into number \ngroups permits a number of markers \nto be locating lines at the same time. \nIf two markers should receive calls \nfor numbers in the same number \ngroup at the same time, one marker \nmust wait for the other to set up its \nconnection. This delay is ordinarily a \nsmall fraction of a second. To pre- \nvent unreasonable delays, however, \nthe hundred-blocks formed into a \nnumber group usually represent only \nabout one thousand terminating calls \nduring the busy hour. For simplicity\n\n'g.. 1\u2014Number-group connector frames in the \nMurray Hill-6 Office on East Thirtieth Street\n\nWhen a marker is seized by a \nsender, it must determine\u2014from the \nnumber that has been dialed\u2014the\n\nFig. 2\u2014Synoptic diagram of the decoding \nfeatures of the terminating marker and \nnumber-group circuit\n\ncorrect number-group connector to \noperate, and also the correct hundred- \nblock and _ twenty-block relays to \noperate in this number group. This \nprocess is known as decoding, because\n\nFig. 3\u2014Combinations of operated register \nrelays to give the various digits\n\nthe decimal directory number it re- \nceived, 2845 to take a concrete ex- \nample, must be translated or decoded \ninto terms designating a number- \ngroup, a hundred-block relay, and a \ntwenty-block relay.\n\nA synoptic diagram of the circuits \ninvolved in this part of the marker\u2019s \nwork is shown in Figure 2, where \nreference is made to two following \nillustrations giving sections of the \ncircuit in greater detail. As a basis for \nits action, the marker has recorded \non four groups of register relays the \nnumber that has been dialed by the \nsubscriber. The proper register relays \nin each group were operated by the \nterminating sender when the marker \nwas seized, as described in the pre- \nvious article. These relays are similar \nin arrangement and function to those \nof the originating markers used for re- \ncording the office code, except that \nfour sets of relays are required instead \nof three because the subscriber num- \nber consists of four numerals. The re-\n\nlays of the group for the first, or \nthousands digit, also differ in being \nnumbered 1, 2, 4, and 8, instead of \n1, 2, 4, and 5, as are the other groups \nin the terminating marker and all the \ngroups in the originating marker. The \ncombinations of the relays within a \ngroup to give the various digits are\n\nSince a number group \nis made up of hundred- \nblocks, a determination \nof the hundred-block in- \ndicated by the registered \nnumber is sufficient to \ndesignate the number \ngroup desired. This hun- \ndred-block designation \nwill also, of course, indi- \ncate which hundred-block \nrelay in the number group \nto operate. The marker \nis designed to determine \nthe hundred-block by \nfirst picking a group of \ns00 numbers \u2014 or five- \nhundred group \u2014 and \nthen narrowing its selec- \ntion to one of the five, \nalthough the two opera- \ntions are carried out si- \nmultaneously, not suc- \ncessively. Due to the \nfact that there are twenty \ngroups of 500 sub- \nscriber numbers in a \n10,000 number office, \nthe marker is equipped \nwith twenty 500-group \nrelays, one for each of \nthe twenty groups of \nfive hundred numbers. \nIt also has five hun- \ndred-block relays, one \nfor each of the hun- \ndred-blocks in a 500- \ngroup, and leads from \ncontacts of hundred-\n\nblock relays are multipled to a set of \ncorresponding contacts in each five- \nhundred group relay. The arrange- \nment is shown in greater detail in \nFigure 4. These hundred-block relays \nin the marker should not be confused \nwith those in the number-groups. \nThe latter are individual to a set of\n\nFig. 5\u2014Register relays in the crossbar system operated for \nthe various five-hundred groups\n\n100 directory numbers, while those in \nthe markers merely designate the five \nsets of 100 numbers into which any \ngroup of 500 may be divided.\n\nThe five-hundred group relay to be \noperated is determined from the \nthousands-group and the No. 5 relay \nof the hundreds-group of the register \nrelays, as shown in Figure 5. From this\n\ntable it will be noted \nthat the number of the \ns00-group relay is de- \ntermined by the thou- \nsands digit of the num- \nber dialed, but since \nthere are two 500- \ngroups in each thou- \nsand numbers, there \nare two $00-group re- \nlays for each of the ten \n\u201cthousand\u201d digits from \n\u00a9 to 9, the number for\n\none group being primed \nto make the distinc. \ntion. In each case, the \nprimed relay is for the \n500-group whose hun- \ndreds digit is \u00a7 of \ngreater. For number \n2845, for example, the \nNo. 2\u2019 FH relay will be \noperated, the 2 indicat- \ning that the thousands \ndigitis2,and the prime, \nthat the hundreds digit \nis five or above. \nWhich hundred- \nblock relay is operated \ndepends on the first \nthree relays of the \nhundreds register re- \nlays as shown in Fig- \nure 6. It will be noticed \nthat each combination \nof the H relays corre- \nsponds to two hun- \ndred-blocks, one below \nand one above 500. For\n\nthe number 2845, the No. 3 HB relay \nwould be operated, in conjunction with \nthe No. 2\u2019 rx relay. This determines \nthe number as between 2800 and 2899, \nbut had the same uB relay been oper- \nated in conjunction with the No. 2 Fx \nrelay, the number would have been \nbetween 2300 and 2399. In this way a \nsingle set of five HB relays, in conjunc-\n\nFig. 6\u2014Register relays that are operated for the various \nhundred-block groups\n\ntion with twenty FH relays, \ncan be made to designate any \nhundred-block in the office.\n\nThe operation of one HB and \none FH relay in the marker \ncloses paths to two sets of \nterminal strips\u2014one marked \nst, and the other us. These \nterminal strips form part of \none of the marker bays as \nshown in Figure 7. Since there \nare twenty FH relays, each \ncarrying a pair of contacts for \neach of the five HB relays, \nthere are one hundred termi- \nnals in each of the two termi- \nnal sets, one corresponding to \neach possible hundred-block \nin the office. Adjacent to the \none hundred st terminals is \na group of terminal strips, \neach strip representing a num- \nber group. Each of the one \nhundred st terminals is cross- \nconnected to the strip repre- \nsenting the number group con- \ntaining that hundred-block of \nsubscriber numbers, and clo- \nsure of this path in the marker \nwill cause the number group \nto be connected to the marker. \nAdjacent to the one hundred \nHB terminals is a similar group \nof terminal strips, each strip \nrepresenting one of the hundred-blocks \nof subscriber numbers in the selected \nnumber group. Each of the one \nhundred HB terminals is cross-con- \nnected to the strip which will oper- \nate the hundred-block relay in the \nselected number group that gives ac- \ncess to the corresponding hundred \nsubscriber numbers.\n\nSo far, a particular number-group \nconnector has been selected, and a \nhundred-block relay in that group \nhas been operated. Which particular \ntwenty-block relay is operated is de-\n\nFig. 7\u2014Each terminating marker consists of two \nbays, one consisting chiefly of three cabinets of relays, \nand the other of one relay cabinet and the cross-\n\ntermined by the combinations of \noperated relays of the tens-group of \nregister relays in the marker. Five \nleads are carried from this set of \nregister relays through the number- \ngroup connector to the number group, \nwhere they are multipled to the five \ncontacts of all the HB relays as- \nsembled in that number group, as \nshown in Figure 8. From this relay \nthey run to the windings of the \ntwenty-block relays\u2014the contacts of \neach hundred-block relay being con- \nnected to the five twenty-block relays\n\nit controls. The circuits through the \ntens register relays are so arranged \nthat when no relay or the No. 1 relay \nis operated, ground will be placed on \nthe lead that runs to the first twenty- \nblock relay\u2014indicating that the last \ntwo digits are between 00-19, and \nwhen the No. 2 relay, or the No. 1 \nand No. 2 relays, are operated,\n\nBy this decoding process, a par. \nticular block relay has been operated, \nand three contacts have been closed \nfor each number of the block that in. \ncludes the one called. One of these \ncontacts for each line will be used for a \nbusy test, and the other two will be\n\nground will appear on the lead run-__ used in the process of establishing the\n\nPRIMARY LINE-SWITCH\u00ae \nCROSS POINT \u00a5%z\u00a5 \nTO SECONDARY ns \nTBO SwitcH \nHBO LINE HOLD 935 \nMAGNET ata \nHB! Pict: \nTBI \nLINE \nNSO RELAY \n- NFO \nuse \nHER HB MULT. TO OTH NC IO} \nRELAYS IN TB RELAYS NC 19] = \nSAME NUMBER Y Y Y Ysame NUMBER | \nGROUP GROUP VVVyy 1.0000 \n\u2014 \u00a9 0001 \n\u201ca TB2 \u00a9 0002 \n: \nif = \n\u2014 ns$26tS \nHB3 LL, \n\u2014 \n09998 \nLO 9999 \nDISTRIBUT \nIBUTING \nFRAME \nHB6 \nHB 24 \nTB3 \nMARKER \nATING MARKER) T| veaps|2 9 Pre TF, HF, RF \n[ [ fe [ [ [ [ [ [ [ [ \nTO MARKER\n\nN large central offices a variety of \ntone signals are required to expe- \ndite the operating procedure, and \nthese as well as ringing current are \nsupplied by motor-generator sets as \nalready described.* For private branch \nexchanges and for the smaller central \noffices ringing and tone supply are \nalso required, but the load is much \nlighter so that the large motor- \ngenerators are not economical to use. \nTo meet the requirements of these \nsmaller offices a number of types of \nequipment have been employed, de- \npending on the size and requirements \nof the office, and for the most part \nhave given excellent service. None of \nthem, however, have sufficient tone \nand interrupter equipment for some \nof the new community dial offices; \nand for the most part they permit a \ngreater variation in output voltages \nthan is desir- \nable for some \napplications. A \nnew small ring- \ning machine \nthus seemed \nnecessary. \nWhat was \ndesired was an \nassortment of \ntone and inter- \nrupter circuits \ncomparable to \nthat of the \nlarger motor- \ngenerator sets\n\nbut without the large size and cost of \nthe more extensive installations. To \nmeet this need, the KS-5510 ringing \nmachine, shown in Figure 1, was de- \nveloped. Instead of being a motor- \ngenerator as the larger sets are, it is a \nrotary converter like the gp design, \none of the smaller machines. A rotary \nconverter combines a d-c motor and \nan a-c generator in the same unit, \nhaving a common set of field coils \nand a single armature, with a com- \nmutator on the motor end and slip \nrings on the generator end. The con- \nverter runs on current from the office \nbattery, and 20-cycle a-c is taken \nfrom the slip rings, through a step-up \ntransformer, for the ringing supply.\n\nOn a shaft extension at the commu- \ntator end are four rings for supplying \nthe high and low tones,* from which \n~ *REcorD, April, 1931, p. 385.\n\nthe various tone signals are derived \nby interrupting the continuous tones \nat various rates. Each basic tone is \nproduced by a slip ring and a seg- \nmented ring with its alternate seg- \nments connected to the slip ring. The \nnumber and width of the segments are \nproportioned so that battery current \npassing in through the slip ring is\n\nthe advantage of being a smaller ang \nless expensive machine than a motor. \ngenerator, speed regulation with jt \nbecomes more difficult. With a motor. \ngenerator set, there are two separate \nmachines, each with its own field, and \nan adjustment of the field of the \nmotor\u2014made to hold a constant \nspeed\u2014has no effect on the voltage\n\ninterrupted at the proper rate. The \nchanges in current are converted by \nmeans of repeating coils into alter- \nnating voltages or tones having fre- \nquencies equal to the revolutions per \nsecond of the machine times the num- \nber of interrupting segments in each \nring. Beyond the tone rings is the \ncentrifugal speed regulator, which \nmaintains the speed of the set within \nthe required limits.\n\nAt the other end of the converter, \nbeyond the main slip rings, is a worm \ndriving a gear on a counter shaft, \nwhich rotates at 10 rpm. This is the \ninterrupter shaft. It is arranged to \nmount as many as six disks, each \ndrilled around its periphery to take on \neach side one or more pins with \nsleeves over them to act as rollers. \nEach such disk has a set of contact \nsprings that the pins raise and lower \nas the shaft rotates, and each set of \nsprings acts as an interrupter for \nringing current or for one type of \ntone. As many are mounted as needed \nup to a maximum of twelve.\n\n20-CYCLE \noutput sistance in the motor field cir. \ncuit. As the motor slows down, \n-- the period of short circuit is re-\n\nduced and that of open circuit \nincreased, so that there is a \nnet decrease in average field \ncurrent and a corresponding \nincrease in the speed of the motor. \nA simplified schematic of such a set \nis shown in Figure 2.\n\nWith a rotary converter, however, \nthe same field serves for both motor \nand generator, and if the field current \nis decreased to offset a falling speed, \nthe generated voltage will decrease. \nTo avoid this, a resistance is inserted \nin the armature circuit, as shown in \nFigure 3, and the regulator contact is \nallowed to short circuit this resistance \nfor longer or shorter periods as the \nmotor slows down or speeds up. Both \nspeed and voltage vary directly with \nthe voltage across the armature, and \nthus increase as the armature resist- \nance is shorted out for longer inter- \nvals. A series field is also used with\n\nA modified form of this machine, \ncalled the KS-5546 code-ringing ma- \nchine, is arranged to give the code \nringing commonly used on rural lines. \nSince as many as twenty parties may \nbe connected on such a line more \ninterrupter springs are required. For \nthis reason the interrupter shaft is \nmade long enough to accommodate \neighteen spring pile-ups, eleven of \nwhich are used to furnish the codes \nalone. Since cost is particularly im- \nportant in these small offices, the\n\nspeed regulator has been omitted. \nThis results in wider voltage varia- \ntions in the ringing output, and the \nrating of the machine has been slightly \ndecreased, which is permissible be- \ncause of the light load under which \nthese offices operate. For the same \nreasons it has been possible to use \nsmaller brushes on the tone rings, and \nthus make another small saving. As a \nresult of the accumulated savings \nthis code-ringing machine will give a \ngreater variety of signaling interrup- \ntions than the KS-5510 machine, and \nat the same time will cost less.\n\nModern business management should be, and I believe \nfor the most part 1s, imbued with an interest in the public \nwelfare. It provides the basis of satisfaction to educated \nmen, for industry is the basis of the well-being of the \nnation and commerce the chief hope of an economy in \nwhich the nations of the world can live in peace. Business \nis not a simple calling. It requires skill of a high order, \ncapacity and a sense of responsibility. Business today is \nnot based on the conception of a world of a limited amount \nof goods in which, if one man gets more, another man must \nget less. Its objective, whether conscious or not, is to create \nmore for all. And in doing so it must reconcile the interests \nof the workers, the owners and the consumers. Especially \ntoday it must carry on with sympathetic understanding of \nthe necessary restrictions to its complete freedom that \ngrow out of what is called the \u201cpublic interest\u2019\u2019\u2014the \ninterest of the general public, whether or not workers, \nowners or customers of the particular industry.\n\n\u2014From an address by Walter \u00a7. Gifford at the Commence- \nment Exercises of Union College, Fune 12, 1939.\n\nIncreased power and better quality of \nreception is provided for police cars and \nother mobile units by the new Western \nElectric 228A radio telephone equipment. \nIts major units are a short-wave receiver \nand 15-watt transmitter, each with its \nown high-voltage power supply operated \nfrom the car battery. In addition, a small \ncontrol unit, a loudspeaker, and a tele- \nphone handset are mounted in front of the \ndriver. Besides its threefold increase in \npower, this new Laboratories apparatus \nincludes a greatly improved codan to dis- \nable the receiver when no signal is being \nreceived, anda device to reduce noise peaks.\n\nJ. W. Dexn joined the Engineering \nDepartment of the Western Electric \nCompany in the fall of 1919, and at first \nengaged in laboratory testing and analy- \nsis of manual and panel telephone cir- \ncuits. Subsequently he worked on the \ndesign of manual circuits and later of \nautomatic testing circuits for panel \noffices. Since 1933 he has been engaged in \nthe design of automatic testing circuits \nand marker circuits for the crossbar sys- \ntem. During this period, \nhe has attended classes \nat the Polytechnic Insti- \ntute of Brooklyn from \nwhich he received the de- \ngree of Electrical Engi- \nneer in 1932.\n\nJ. W. Fotey graduated \nfrom the University of \nIllinois in Ig91I, receiving \nthe degree of B.S. in Elec- \ntrical Engineering, and \nimmediately joined the \nWestern Electric Com- \npany at Hawthorne. Upon \nthe completion of the \nstudent course, he entered\n\nthe Transmission Laboratory at New \nYork, where he was concerned with \ngeneral transmission studies. His work \nsoon became centered upon telephone \nsets and their associated circuits. He has \nbeen intimately associated with the de- \nvelopment of the anti-sidetone telephone \nsets now standard for both subscribers \nand operators as well as many sets for \nspecial purposes such as train dispatch- \ning, amplifier sets for the hard of hearing, \nand loudspeaker systems \nfor both intercommunicat- \ning and regular tele- \nphones. More recently he \nhas participated in the \ndevelopment of closed- \ncore induction coils for \noperator and subscriber \nsets, of acoustic shock pre- \nvention devices, and in \nfurther studies for the im- \nprovement of subscriber \ntelephone sets and their \nassociated apparatus.\n\nTwo years prior to his graduation, he had \njoined the Technical Staff of the Lab- \noratories, and until 1928 was associated \nwith the Research Department in con- \nnection with the development of terminal \nequipment for high-speed submarine \ncables. At the completion of this work he \ntransferred to the Apparatus Develop- \nment Department where he has been en- \ngaged in the development of precise im- \npedance-measuring equipment.\n\nH. B. Briccs received the B.A. and \nM.A. degrees from the University of \nWisconsin in 1919 and 1921, and the \nPh.D. degree in Physics from the Uni- \nversity of Chicago in 1925. His under- \ngraduate work was interrupted by two \nyears spent in France as a Lieutenant of \nInfantry. From 1921 to 1923, and again \nfrom 1925 to the time he joined the \nLaboratories in 1929, he was a member of \nthe teaching staff at the State College of \nWashington. Until the past year and a \nhalf he has carried on fundamental re- \nsearch in the fields of photoelectricity and \nthe optical constants of metals. At pres- \nent he is working on electron optical \nproblems associated with television.\n\nR. D. pe Kay was graduated from the \nUnited States Naval Academy in 1918. \nHe served in the war as engineer officer on \ndestroyers, and after the war as com- \nmanding officer. In 1922 he left the Navy \nand joined the power development group\n\nof the Laboratories, where he supervises _ \nrectifier and machine development.\n\nL. J. Sracy received the A.B. degree \nfrom St. Lawrence University in 1go9. \nAfter six years of school teaching, he \nentered the University of Chicago for \ntwo years of graduate study in physics \nand mathematics. In the radio division of \nthe Signal Corps in 1918 he rose to a \nsecond lieutenancy, then returned to \nChicago to receive the Ph.D. degree in \n1919, and later that year entered the \nSystems Development Department of \nthese Laboratories. He has been con- \ncerned with ringing and tone studies and \nother special technical problems in the \nlocal central office laboratory, and later \nwas placed in charge of a group devoted \nto this work. Since early in 1937 he has \nalso been in charge of the step-by-step \nand manual testing groups.", "title": "Bell Laboratories Record 1939-07: Vol 17 Iss 11", "trim_reasons": [], "year": 1939}
{"archive_ref": "bell-tele-1929-02", "canonical_url": "https://archive.org/details/bell-tele-1929-02", "char_count": 135821, "collection": "archive-org-bell-labs", "doc_id": 295, "document_type": "book", "id": "bella-qwen-pretrain-doc295", "record_count": 151, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bell-tele-1929-02", "split": "test", "text": "A GROUP OF SIX PAPERS \nPRESENTED AT THE LAKE PLACID CONFERENCE \nOF THE SOCIETY OF MOTION PICTURE ENGINEERS \nOUTLINING THE FUNDAMENTAL FACTORS INVOLVED \n- IN THE PRODUCTION OF SOUND PICTURES\n\nFALL MEETING OF THE \nSOCIETY OF MOTION PICTURE ENGINEERS \nLAKE PLACID, N. Y. SEPTEMBER, 1928\n\nVOL. XII, PP. 633-643; 657-741 SEPTEMBER, 1928 \nTHE BELL SYSTEM TECHNICAL JOURNAL \nVOL. VIII, PP. 159-208 (Papers 3, 4, 5, 6-1) JANUARY, 1929\n\n. The Quality of Speech and Music..... \n_ General Principles of Sound Recording \n_ Recent Advances in Wax Recording.\n\nsounds probably played equally important parts. Although all \nthree have survived, the voice has become the one most universally \nused. With the invention of the telephone and the phonograph \nit became possible for man to communicate at a distance and to \nperpetuate his speech. Now man may communicate at a distance \nand perpetuate, not only his speech, but his grimaces and gestures \nas well.\n\nThis progress has been accompanied by extensive investigations \ninto the physical nature of speech and hearing. It is the aim of the \npresent paper to consider the bearing of various data on the sensory \naspects of speech and music. A selected list of published papers \nthat have been consulted is attached.\n\nThe organs of speech (Fig. 1) consist of the lungs (not shown in \nfigure), the trachea or windpipe, and the mouth, nose and throat \ncavities. The larynx is situated at the upper part of the windpipe \nand contains two muscular ledges known as vocal cords. They are \nso arranged as to form a straight slit through which the breath \npasses.\n\nIn the process of speaking, the lungs by their bellows-like action \nforce streams of air in and out through the vocal passages. As the \nstreams of air pass through the slit formed by the vocal cords, they \nare set in vibration so that the slit is alternately opened and closed. \nA train of sound waves is thus set up in the lower part of the throat. \nAs these waves pass out into the air, certain resonant and transient \ncharacteristics are impressed upon them by the vocal cavities and \nthe movements of the tongue, lips, etc. It is these variations which \nwe interpret as speech sounds.\n\nThese sounds are called unvoiced sounds and the vocal cords do \nnot enter into their production. They are produced by frictional \nvibration set up in the mouth itself.\n\nBoth groups may be divided into two classes\u2014those produced \nby a continuous flow of air which may be called the continuants \nand those produced by the sudden stoppage of the air which may . \nbe called the stops. In the former class are such sounds asa, v, f, \netc. In the latter class are such sounds as p, g, d, t, etc.\n\nAlthough the vocal cords lend quality to the voice they are \nnot highly essential for producing the distinguishing characteristics \nof the speech sounds. These are produced mainly by the mouth\n\nand nose cavities. This is evidenced by the fact that we can under- \nstand whispered speech, in which the vocal cords play no part. \nAs a matter of fact counterparts of the lungs and vocal cords can \nbe located quite outside the body and still produce intelligible speech\n\nA piece of apparatus known as the artificial larynx is a good \nillustration. An ordinary bellows serves as the lungs, a rubber tube \nas the windpipe, and a little metal box as the larynx. Inside the \nbox is a reed which vibrates with the passage of air, much like the \nold style automobile horn. The interrupted air stream is led into \nthe mouth by means of a short pipe which is attached to the metal\n\n. This device is used by people who have undergone a surgical \noperation known as tracheotomy. In this operation the larynx is \nremoved and the windpipe is terminated by a small hole in the \npatient\u2019s neck, through which he breathes. In this case, the bellows \nis dispensed with, and the rubber tube is fitted over the hole, so \nthat, when he breathes a train of sound waves is produced. In order \nthat he may speak these sound waves are directed into his mouth.\n\nThe mechanism of hearing (Fig. 2) is divided into three general \nparts, the outer ear, the middle ear and the inner ear. The outer \near consists of the external part and the ear canal. The middle ear \ncontains three small bones which are called the hammer, the anvil, \nand the stirrup. They connect the ear drum with the small window \nor diaphragm \u2018\u2018O\u201d\u2019 of the inner ear. In the inner ear is located the \norgan of hearing. It consists of a spiral cavity \u2018\u201c\u2018S\u201d surrounded by \nbone. The cavity is filled with liquid. As may be seen from the \nfigure a spiral ledge projects into the cavity. The liquid above this\n\n** The production of the simple speech sounds and also connected \nspeech was demonstrated with the above type of apparatus.\n\n+ Reproduced, with permission, from Textbook of Physiology by William \nH. Howell, M. D., W. B. Saunders Co., Philadelphia.\n\nledge is separated from the liquid below it by a flexible membrane. \nThe two windows \u201cO\u201d and \u2018\u2018E\u201d\u2019 retain the liquid at the base. At the \napex, the membrane is pierced by a small hole so that the liquid \nmay flow from the upper side of the ledge to the lower side. When \nthe window at \u2018\u201cO\u201d is pushed in, the window at \u2018EK\u2019 bulges out and \nvice versa. The nerves of hearing are distributed along the flexible \nmembrane.\n\nSound enters the ear as successions of minute condensations \nand rarefactions of the air, which are known as sound waves. They \ncause minute increases and decreases in the air pressure near the \near drum. These variations are called sound pressure and cause the \near drum to vibrate. It is supposed that the liquid in the inner ear \nsurges up and down the spiral in accordance with the frequency \nof vibration. If the frequency is low the pressure is relieved through \nthe flexible membrane near the apex. If the vibration frequency is \nhigh, the pressure is relieved near the base. In other words, tones \nof different frequency disturb the membrane in different positions \nalong its length. This separation enables us to identify the pitch of \nthe tone.\n\nSpeech waves are composed of a large number of tones of \ndifferent vibration frequency which change continually during utter- \nance. These tones cause disturbances which are spread out along \nthe membrane containing the nerves of hearing. The high tones \ndisturb one end of the membrane, the low tones disturb the other \nend of the membrane. If one could see this membrane it might \nresemble the keyboard of a player piano in operation. At one in- \nstant the keys at one part of the scale are disturbed; at a later \ninstant the keys at some other part of the scale are disturbed. In \nthe case of the ear, the pattern of the disturbance on the membrane \nis carried to the brain and interpreted as a speech sound.\n\nFig.3 shows the range of sound pressure and frequency which the \near can perceive. The vertical axis represents the sound pressuref fT \nand the horizontal axis represents frequency of vibration. In the \nsensation of tone, loudness is caused by the sound pressure and pitch \u00a9 \nis caused by the frequency of vibration. The lowest pitch that can \nbe sensed as a tone is of an order of 20 cycles, the highest is of an \norder of 20,000 cycles. The lower curve represents the sound\n\npressures at which the various pure tones are just audible. It will \nbe noticed that the ear is less sensitive to the low notes, that is, \na greater pressure is required for these tones to become audible. The \nupper curve represents the sound pressures at which the tones cause \na sensation of feeling or pain. The faintest tone that can be heard \nand the loudest tone that can be heard stand in a ratio of one million \nto one. The ear can detect any tone that falls in the area enclosed \nby the two curves.\n\nStudies on the wave forms of speech sounds have shown that \nthe pitch of man\u2019s voice is of an order of 128 cycles (vibrations per \nsecond), that of woman\u2019s voice of an order of 256 cycles. In both \ncases, overtones of the fundamental chord tone occur in the speech \nsounds. These studies have shown frequencies as high as 8000 or \n9000 cycles in various speech sounds. Studies on the interpretation \nof speech sounds have also indicated the presence of tones covering \na large part of the audible frequency range. This can best be seen \nby considering the parts of the auditory sensation area which are \nparticularly important to the interpretation of the various speech \nsounds, as shown in Fig. 4.\n\nAlthough in the above figure the speech sounds have been \ngrouped in sharply defined areas, it must be understood that \nactually the sounds overlap these areas somewhat, and that the \nindicated areas are those which are most important in the inter- \npretation of the sounds. The three voiced consonants which are \nsymbolized by the letters v, z, th (as in then), are exceptions and\n\nbelong in the unvoiced consonant area. The various speech sounds \ncover a frequency range up to 8000 cycles, and a sound pressure \nrange of 60 sensation units. The area above 8000 cycles does not \ncontribute appreciably to speech interpretation. It does, however, \ncarry a large part of the sounds of breathing which take place during \nspeech.\n\nIn general, woman\u2019s speech i is more difficult to ae se than \nman\u2019s. This may be due in part to the fact that woman\u2019s speech \nhas only one-half as many tones as man\u2019s, so that the membrane \nof hearing is not disturbed in as many places. We may suppose,\n\ntherefore, that the nerve fibres do not carry as much data to the \nbrain for interpretation. The greatest differences occur in the case \nof the more difficult consonant sounds. In woman\u2019s speech these \nsounds are not only fainter but require a higher frequency range \nfor interpretation. A range from 3000 to 6000 cycles for man\u2019s\n\nwoman\u2019s voice. Since the ear is less sensitive in the latter range \nand the sounds are initially fainter, their difficulty of interpretation \nis greater.\n\nThere is one other phenomenon of hearing which enters into \ninterpretation. When sounds containing a number of tones are \nincreased in loudness, the lower tones in the sound deafen the \nauditor to the higher tones. This deafening or masking effect \nbecomes very marked when the sound pressure of the lower tones \nis greater than twenty sensation units. In the case of speech, this\n\neffect impairs the interpretation of the higher pitched sounds. The \noptimum loudness for the interpretation of speech corresponds to \na cound pressure between 0 and 20 sensation units. If the sound \npressure is less than this, the fainter sounds are inaudible. If the \nsound pressure is greater, the masking effects impair the inter- \npretation of these sounds. 3\n\nAs in the case of speech, musical sounds consist of a funda- \nmental frequency and various overtones of the fundamental. The \ntones, however, are sustained for appreciable lengths of time, and \nwhen.they are changed, the changes usually take place in definite \nsteps, known as musical intervals. The pitch of the tone is deter- \nmined by the fundamental frequency. This frequency, however, \nneed not be present in the musical tone as the overtones which are \nmultiples of the fundamental may cause the correct pitch sensation.\n\nThe frequencies that are present in music depend upon the type \nof instrument and the character of the music. The composition \nof musical tones can be obtained by considering the construction \nof various types of instruments, and by wave form studies.\n\nMusical instruments may be divided into two general classes, \nstring instruments and wind instruments. With strings, the tones \nmay be produced by plucking, striking, or bowing, and are usually \nreenforced by resonating air cavities and sounding boards. With \nwind instruments, the tone may be produced with the aid of reeds \nas in the clarinet, or without reeds asin the flute. In the playing of \nhorns the lips act as reeds.\n\nEach class may be further subdivided into melody and harmony \ninstruments. In the former only one note is usually produced at \na time; in the latter several notes are usually produced simul- \ntaneously. In general, harmony instruments are capable of pro- \nducing notes over a much wider frequency range than melody \ninstruments. A given type of instrument of the latter class may, \ntherefore, include several instruments covering different frequency \nranges such as the bass, tenor and alto trombone.\n\nExperiments have indicated that notes of different frequency \nor pitch as produced by a musical instrument, appear about equally \nloud to the ear. This seems reasonable when we consider that the \near has played an important part in the design of such instruments. \nSince the sound pressures that are necessary for equality of loud- \nness of various tones are known, it is possible to approximately\n\ndetermine the range of pressures that are met with in covering the \nfrequency range of music.\n\nIn Fig. 5 contour lines of equal loudness are shown for fre- \nquencies from 32 to 4000 cycles. The frequency range has been \ndivided into three parts, bass, tenor and alto, and soprano registers, \ncorresponding to the notes produced by various instruments. The \ncontour lines indicate that the notes of the lower registers have \ngreater sound pressures than those of the higher. The range of \npressures for various instruments and for various musical notations \nof loudness are smaller for low notes. Direct measurements of the \npressures that are produced by various instruments when played \nby musicians show this tendency.\n\nThe contour lines for loud tones show a smaller change in \npressure in going from low to high notes than do the contour lines \nfor faint tones. If instruments are played so that the notes are of \napproximate equal loudness, an orchestra playing in a resonant \nroom where the sound reaching the players\u2019 ears is loud would \nproduce a smaller range of pressures than if it were playing in a \nvery well damped room.\n\nPercussion instruments such as drums and the various ac- \ncessory traps produce the greatest pressures that are used in music. \nAlthough the fundamental frequency of the notes which they emit \nis fairly. low, the notes are particularly rich in tones of higher fre- \nquency, which may extend as high as 10,000 cycles. Although these \nhigher tones die out rather rapidly, they are essential to good \ndefinition.\n\nThe organ, the piano, and the harp have the greatest compass \nand cover a frequency range from about 16 to 4000 cycles. All \nthree of these instruments are characterized by a rather prominent \nfirst overtone, so that their effective range extends as high as 8000 \nor 9000 cycles.\n\nOwing to their limited compass, the melody instruments are \namong the easiest to reproduce. In any given register these instru- \nments may be arranged in the following intensity order: wind \ninstruments, and string instruments, the violin producing the \nfaintest sounds. Asa class these instruments produce notes covering \nthe frequency range of 32 to 4000 cycles.\n\nFrom the auditory sensation area, we have seen that the ear \nis able to perceive a large number of tones which differ in sound \npressure and frequency. We have also seen that the voice and \nvarious musical instruments produce tones which cover a large \nportion of the auditory sensation area. In order to obtain informa- \ntion as to the relative importance of various parts of this area to \nthe sensory characteristics of speech and music, experiments have \nbeen performed in which the tones falling in various parts have \nbeen eliminated from the sounds by means of filters.\n\nWhen frequencies below 100 cycles, 200 cycles, 300 cycles and \non up to 1000 cycles are progressively eliminated from speech, the \ncharacter of the speech changes markedly. The terms \u201c\u2018timbre\u2019\u201d\u2019 \nor \u2018\u2018tone color\u201d best describe the characteristic which is lost. This \ncharacteristic appears to be associated with the fundamental and \nthe first few overtones of the voiced sounds and their presence is \nnecessary in order to convey the sensation of timbre. Frequencies \nbelow 300 cycles, however, do not appear essential to the correct \ninterpretation of the speech sounds.\n\nWhen frequencies above 8000 cycles, 7000 cycles, and on down \nto 3000 cycles are eliminated, the character of the speech again \nchanges markedly. The term \u201cgibilance\u2019\u201d\u2019 appears to best describe \nthe characteristic which is lost. It refers to the prominence of the \nhissing or frictional character of speech. If attention is directed to \nsuch sounds as s, f, th and z, the elimination of frequencies above \n6000 or 7000 cycles is readily detectable. It requires rather close \nattention to detect the elimination of frequencies above 8000 cycles. \nElimination of frequencies above 7000 cycles slightly impairs the \ninterpretation of the s and z sounds of woman\u2019s voice. Elimination \nof frequencies above 6000 cycles impairs the interpretation of the\n\nf and th sounds of man\u2019s voice, and the interpretation of the f, th, \nsand zsounds of woman\u2019s voice. The impairment due to eliminating \nhigher frequencies is usually greater in the case of female voices.\n\nAs in the case of speech the tone color or timbre of musical \ntones appears to be associated with the fundamental and the first \nfew overtones of the note produced. \u2018Timbre is probably more \nimportant in music than it is in speech as it is one of the things \nwhich distinguish the tones of various instruments. It appears \nthat, in general, the fundamental and the first three or four over- \ntones are necessary in order to distinguish the tones of various \ninstruments. When overtones higher than these are eliminated \nthe tones lose a characteristic which can best be described by the \nterms \u2018\u201c\u2018brilliance\u2019\u201d\u2019 or \u2018\u2018definition.\u201d\u201d The tones seem to lose life and \nbecome dull. The prominence of these characteristics varies with \nthe type of instrument, the composition of the music and the \npersonality of the musician.\n\nThe notes which are used most in music are contained in the \noctave below middle C and the octave above middle C, that is, \u00a9 \nfrom 128 to 512 cycles. The fourth overtone of 512 cycles has a \nvibration frequency of 8192 cycles, so that tones of this frequency \nand below occur frequently in music. A trained ear could no doubt \ndetect the elimination of frequencies above this range from the \nordinary run of music, but it is probable that the average individual \nwould have difficulty in detecting the elimination of frequencies \nabove even 6000 or 7000 cycles, unless he gave particularly close \nattention to the percussion instruments.\n\nOwing to the masking effect, the character of both speech and \nmusical sounds changes as the loudness of the sound is increased. \nThe higher notes are masked more and more which has the effect \nof accentuating the bass notes. The best simulation is obtained \ntherefore when the reproduced sounds are about as loud as the | \noriginal sounds. rs :\n\nThe effects discussed in the above paragraphs were illustrated by means \nof speech and music reproduced from phonograph records. In recording \nthe compositions, various frequency ranges were removed by means of \nelectrical filters, and different recording levels were used. Hence, the effects \ncould be shown by reproducing the records from an ordinary phonograph.\n\nThe records were prepared by Bell Telephone Laboratories and infor- \nmation concerning them can be secured from its Bureau of Publication, \n463 West Street, New York City.\n\nlin Inst., June (1922). \n2. The Frequency Sensitivity of Normal Ears by H. Fletcher and R. L. We-\n\nin the air was known to the ancient Greeks, and that objects \nare set in vibration by intense sounds must have been observed by \nprimitive man, but it was not until 1857, or less than a century ago, \nthat the first instrument was constructed for making a graphical \nrecord of sound waves. In that year L\u00e9on Scott patented in France \nan instrument which he called the phonautograph. In Fig. 1 is \nshown a picture of this instrument. A piece of smoked paper was\n\nattached to the cylindrical surface of a drum, which could be \nrotated by hand and moved forward by a screw. The center of a \ndiaphragm was attached to a stylus through a system of levers in \nsuch a manner that the stylus was moved laterally along the surface \nof the cylinder when the diaphragm vibrated. Over the diaphragm \nwas placed a barrel-shaped mouthpiece. When the drum was \nrotated, words spoken into the mouthpiece caused the stylus to \ntrace a wavy line upon the smoked paper. This wavy line was the \nfirst known record of sound vibrations.\n\nepoch-making invention. Edison constructed a machine very \nsimilar to the phonautograph but differing in two important details. \nThe smoked paper was replaced by a sheet of tinfoil and the stylus \nwas attached directly to the diaphragm, so that it traced an im- \npression of variable depth when the diaphragm vibrated, instead \nof a wavy line as in the case of the phonautograph. After such a \nrecord had been made the drum was set to the starting point and \nwith the stylus in place was again rotated at the same speed as \nbefore. The recorded sound was then intelligibly reproduced. Thus \nEdison gave us the first phonograph.\n\nIn subsequent models the tinfoil was replaced by a wax cylinder. \u2014 \nFor many years the wax record, either in cylinder or disc form, was \nused almost exclusively for the recording and reproducing of sound. \nAlthough many other methods of recording have been suggested, it \nis only in the last few years that records made photographically \nhave come into the commercial field as competitors. Both the wax \nand the photographic records are now being used in conjunction with \nmotion pictures.\n\nPhotographic records are now being made by many different \ntypes of apparatus. But they may be divided into two general \nclasses. In one of these classes the record is a trace of constant \nphotographic density but of variable width, while in the other it is \na trace of constant width but of variable density. An illustration \nof each of these is shown in Fig. 2. In one or two proposed methods \nthe record is a combination of both types.\n\nIn almost all systems experimented with today there is at \nleast one element which they have in common with the phonauto- \ngraph, viz.: a diaphragm which is set in vibration by the sound to \nbe recorded. As in the phonautograph the diaphragm may be \nmechanically connected to the engraving mechanism or recorder; \ner, again, it may be connected electrically as in most modern \nsystems. But in practically all of them the diaphragm forms an \nessential element.\n\nUnfortunately a diaphragm does not in general have the same \nresponse at all frequencies. A favorite experiment in lectures on \nelementary physics is to sound a tuning fork and with it, through \nthe air, set in vibration a second tuning fork. In this experiment it \nis important that the pitch, or the resonant frequency, of the two \nforks be very nearly the same, otherwise the motion set up in the \nsecond fork will be too small to be observable. Diaphragms, and in \nfact almost any other type of mechanical system, will have at \nleast one resonant frequency, which means that, under the action of \nsound waves, the response will in general be much greater in this \nregion than at other frequencies.\n\nIn the older methods of recording, resonance was purposely \nintroduced in order to obtain records of sufficient amplitude. The \nfrequencies lying in the resonance region were then much over- \nemphasized. The sound reproduced from such records had a blasting \nand metallic quality, and well deserved the title \u2018\u201c\u2018canned music.\u201d\n\nBecause of the complex nature of speech and music and of the \nereat amount of distortion introduced into the early recorders and \nreproducers it is not surprising that the quality of reproduction was \npoor, but it is really astonishing that the reproduced sounds were \nat all intelligible. In fact, it has been suggested that, had the \ncomplex nature of speech sounds been generally known at the time, \nthe invention of the telephone, which preceded the phonograph, \nmight have been delayed for many years since its inventor probably \nwould have dismissed his ideas as altogether impracticable.\n\nAlthough considerable distortion may permissibly be intro- \nduced by the recording and reproducing systems before the char- \nacter of the sounds is so changed that they can no longer be recog- \nnized, it is equally true that, if all classes of sounds are to be \nreproduced to a degree of fidelity where the ear cannot distinguish \nthem from the original, the amount of distortion must be kept\n\nextremely small. It is therefore necessary to diminish the distortion \nby the diaphragm to a negligible value.\n\nPrimarily that a diaphragm giving a uniform response may be \nused and that a record of sufficient amplitude may still be obtained, \nthe electrical method of recording has been developed which is \ntoday widely used in the production of commercial sound records. \nIn this method the pick-up diaphragm is made a component part \nof the recording microphone. Here we can content ourselves with \na small amplitude of motion and amplify the voltage generated to \nan amount sufficient for operating a rugged and distortionless \nrecorder. It may be of interest here to compare the amplitude of \nmotion of the diaphragm in the Edison recorder with that of the \nmicrophone used in the majority of present recording systems. In \nthe former the maximum amplitude required for the loudest sounds \nis about 0.001 inch, whereas in the latter under ordinary recording \nconditions it is only about one-tenth as great and the weight is \nonly one-twentieth as great. It can thus be seen how the problem \nof. design of a pick-up diaphragm is greatly simplified in the electrical \nmethod.\n\nIt is of course important that the rest of the recording system \nshall also be free from distortion. However, if a microphone of \nuniform response is available, the design of a distortionless recorder \nis made comparatively easy, for its sensitivity may to a large extent \nbe disregarded, inasmuch as the required power can in general be \nobtained by the use of vacuum tube amplifiers.\n\nIn the electrical method, extraordinary improvements have \nbeen made over the older systems in the elimination of distortion.\n\nThe problem of developing recording apparatus is in many \nrespects identical with that of developing a high quality radio \ntransmitter. In the former, however, there is the additional problem \nof distortion introduced by the record itself. If, for instance, a \nrecord is run at a speed of ten inches per second, and a tone having \na frequency of 5000 cycles per second is recorded, the length of one \ncycle on the record will cover a distance of only 0.002 inch. In the \ncase of wax records the needle must have a very fine point; and in \nthe photographic record the width of the light beam as measured \nalong the direction of motion of the film must be extremely small. \nAt whatever speed the record may be driven, there will always be \nsome frequency beyond which all tones will become more and more \nattenuated. While the loss of the higher frequencies does not in\n\ngeneral impair the tone quality to the same extent as does the \npresence of sharp resonance regions, yet it reduces the intelligibility \nof speech and the richness and brilliancy of musical sounds.\n\nThere is another type of distortion commonly present in \nreproduced sound, which is frequently designated as non-linear \ndistortion. This type of distortion is introduced when the excursion \nof any element of the system is not proportional to the stimulus. \nFor example, a pure tone, which is of sine wave form, such as is \nshown in Fig. 3, a, may be reproduced so as to have a wave form \nsimilar to that shown in b. Physically, distortion of the wave form \nin this manner is equivalent to the introduction of extraneous \nfrequencies. If the magnitude of these extraneous frequencies is \ntoo great, the tone quality will be very disagreeable. However, a \nsmall amount of distortion of this kind is not noticeable, for the\n\nreason that the primary will mask the extraneous tone. It is a well \nknown fact that a tone must be much more intense to be heard \nif another tone is sounded simultaneously.\n\nAnother type of distortion peculiar to recording is that intro- \nduced by a non-uniform speed of the medium on which the record \nis engraved. This may not always be serious, but, in certain cases \nof sustained tones, speed variations cause a disagreeable flutter \nand in some types of music a decided harshness of tone.\n\nOne of the most serious problems with which the radio engineer \nhas to contend is static interference. This also has its counterpart \nin sound reproduction from records. As the ether through which \nthe radio waves are sent is non-homogeneous because of extraneous \nelectrical disturbances, so the sound record is non-homogeneous \non account of the non-uniformity of the material on which it is \nengraved. The noise resulting from these irregularities is often \ndesignated as surface noise. In the case of the wax record, an \nappreciable part of this noise has its origin in the minute irregular-\n\nities of the material and, in the case of the photographic record, in \nthe finite size of the grains forming the photographic image. The \ndifficulties of eliminating this noise arise from the fact that the \nphysical intensity of audible sounds covers an exceedingly wide \nrange. Fig. 4 shows curves published by Wegel! on the sensitivity \nof the ear. The lower curve gives the threshold of audibility and \nthe upper curve the feeling level, i.e. the level of intensity of sound \nwhich becomes painfully loud. The ratio of pressures of the maxi- \nmum of these curves is about ten million. If a record of this extreme \nrange of volume were to be recorded the amplitude of the loudest\n\ntone would have to be ten million times as great as for the faintest \ntone. There is, in general, a maximum amplitude that a record can \naccommodate, which, for instance, in the case of the wax record is \nabout 0.002 inch. If a tone having an intensity near the feeling \nlevel is recorded at this amplitude, then the amplitude of a tone \njust audible would be only 0.000,000,000,2 inch. It is difficult to get \na material having a degree of homogeneity corresponding to this \nvalue. A similar condition obtains in the case of photographic \nrecords, where the pattern is formed by grains in the emulsion \nwhich have a magnitude somewhat less than 0.001 mm. depending \nupon the type of emulsion used. The range of volume considered \nhere is extreme. Practically it is not necessary to record a range of \nthis extent, but it serves to illustrate the extraordinary requirements\n\n1 The Physical Examination of Hearing and Binaural Aids for the Deaf, \nby R. L. Wegel. Proceedings of the National Academy of Sciences, Vol. 8, \nNo. 7, July (1922).\n\nplaced upon the recording medium. When the range of frequencies \nthat are tobe reproduced is increased the surface noise effect becomes \ngreater. As in the case of the different types of distortion discussed \nabove, the difficulties to be met are increased as the quality of \nreproduction is improved.\n\nSynopsis: This paper considers chiefly the frequency-response charac- \nteristics and limitations of the lateral cut \u2018\u2018wax\u2019\u2019 record. It shows that the \nfrequency range from 30 to 8,000 cycles can be recorded and reproduced \nfrom the record with practically negligible deviation from a flat frequency- \nresponse characteristic. The paper brings out the ease with which the \nrecord can immediately be replayed from the \u2018\u2018wax\u201d\u2019 as an aid in assisting the \nartist to obtain the best results. A brief description is given of commercial \nprocessing methods including both plating and pressing. \u2018These methods \ngive essentially a perfect copy of the original \u2018\u2018wax.\u2019\u2019 The time required \nfor this work has been considerably reduced of late so that a test pressing can \nbe obtained within three hours of the cutting of the original \u2018\u2018wax.\u201d\u2019\n\nhe the recording and reproducing of sound by the so-called \u201celectric\u201d\u2019 \nmethod with the \u2018\u2018wax\u2019\u2019 disc, the process may be considered as \nconsisting of eleven steps. In order, these are: (1) studio, with \nits acoustic conditions, (2) microphone, (3) amplifier, (4) electro- \nmechanical reeorder, (5) \u2018\u2018wax\u201d\u2019 record, (6) copying or reproducing \napparatus, (7) hard record or \u2018\u2018pressing,\u201d (8) electric pickup, (9) \namplifier, (10) loud speaker, (11) auditorium.\n\nWith this chain of apparatus the chief problem is that of making \nthe reproduced sound in the auditorium a perfect copy of that in the \nstudio. This isa matter of quality or fidelity of reproduction. There \nare other problems of cost, reliability, time required, etc., which are \nimportant but secondary to that of fidelity. While it may be necessary \nor convenient to introduce distortion in one of these links to com- \npensate for such unavoidable distortion as may occur in other links, \nexperience shows that it is desirable for the sake of simplicity, reliability \nand flexibility to reduce such corrective warping to a minimum and \nto make each step in the process as nearly perfect as possible. Per- \nfection of a complete system may be judged by the practical method \nof listening to the overall result. It is necessary, however, to analyze \neach element of the complete system. To do this, other more \nanalytical methods of test and standards of performance must be \nused. One of the most useful of these is the response-frequency curve. \nIn order that all frequencies be reproduced equally and that the \nordinary faults of resonance be avoided, this must be flat and free \nfrom sharp peaks. Good reproduction requires that frequencies from \n50 to 5,000 cycles be included without discrimination. If, however,\n\n1 Presented before Society of Motion Picture Engineers at Lake Placid, New \nYork, September 26, 1928.\n\nthe low frequency range be lowered to 25 or 30 cycles, a noticeable \nimprovement will be obtained with some classes of music, whereas \nif the upper limit be increased to 8,000 or even 10,000 cycles, the \nnaturalness and smoothness of practically all classes of reproduction \nwill be noticeably improved.\n\nA second important requirement in the judgment or analysis of \nany such system is that the ratio of output to input shall not vary \nover the range of currents or loudnesses (as well as frequencies) from \nthe minimum up to the maximum used. If this requirement is not \nmet, sounds or frequencies not present in the original reproduction \nwill be introduced. This type of distortion has probably been heard \nby all of us in listening to an overloaded vacuum tube amplifier and \nis often referred to as \u2018\u2018non-linear\u2019\u2019 distortion.\n\nA third requirement not entirely disassociated from the first two \nis that any shifts in the phase relations shall be proportional to \nfrequency.\n\nOur judgment of the degree of perfection needed in sound repro- \nduction systems is changing and growing more critical, so that what \nseemed excellent yesterday may be only fair today and tomorrow \nmay seem intolerable. It is therefore necessary that our consideration \nand analysis be continually more searching and fundamental.\n\nwax\u2019\u2019 recording and reproduction, only five are peculiar to the \n\u2018\u2018wax\u2019\u2019 method. These are the electromechanical recorder, \u2018\u2018wax\u2019\u2019 \nrecord, the copying apparatus, the \u201c\u2018pressing\u2019\u2019 and the pickup or \nreproducer. The extent to which the \u2018\u2018wax\u2019\u201d\u2019 method is capable of \nthe highest quality of reproduction will be disclosed by an examination \nof these five links. Any consideration of the practical advantages or \ndisadvantages of the method can logically follow this examination \ninto the quality possibilities.\n\nThe consideration which follows refers to the so-called \u201c\u2018lateral\u2019\u2019 \ncut record; that is, a record in which the groove is of constant depth \nand oscillates or undulates laterally about a smooth spiral. This is \nthe type used in the Western Electric Company disc record type of \nsynchronized motion picture system. Some, but not all of the con- \nsiderations and conclusions might apply to the \u201chill and dale\u201d type \nrecord. It is not the purpose of this paper to consider the relative \ncharacteristics of \u2018\u2018hill and dale\u201d and \u2018\u2018lateral\u2019\u2019 \u2018\u201c\u2018wax\u201d\u2019 records.\n\nIt is the task of the electromechanical recorder to take power from \nthe amplifier and drive a mechanical recording stylus. The present-\n\nday recorder is a highly developed apparatus based on extensive \nexperimental as well as theoretical studies. A diagrammatic view is \ngiven in Fig. 1.2. Recorders which have been supplied by the Western \nElectric Company have been designed to operate over a range of \nfrequencies from 30 to 5,500 cycles. A typical frequency characteristic\n\nis shown in Fig. 2. The device operates in linear fashion over the \nrange of amplitudes involved in speech and music. | As is seen, the \nresponse falls off below about 250 cycles. This falling characteristic \nis necessary in order that the maximum loudness be obtained from a \nrecord for a given spacing between grooves without cutting over\n\n2 High Quality Recording and Reproducing of Music and Speech,\u201d by J. P. \nMaxfield and H. C. Harrison, presented at 14th Midwinter Convention of the \nAmerican Institute of Electrical Engineers, New York, N. Y., Feb., 1926.\n\nfrom groove to groove. In order that a lateral oscillation in a groove \nmay represent constant intensity of sound or a constant energy over \na range of frequencies, not the amplitude of the oscillation but the \nvelocity, which is proportional to the product of the amplitude and \nthe frequency, must be maintained constant. With the characteristic \nshown with these recorders, constant velocity is obtained from about\n\n250 cycles to 5,500 cycles. Below 250 cycles an approximately \nconstant amplitude is obtained. If, therefore, sounds of constant \nabsolute intensity are to be recorded over this range of 30 to 250 \ncycles, there is equal tendency for sounds of the different frequencies \nin this range to over-cut the record groove. It may be corrected in \nreproduction by a suitable electric network. Such a network will \nincrease the subsequent amplification required but, as this additional\n\namplification occurs in the first stages, it is not expensive. Practically \nit has not been found necessary or desirable to introduce such a \ncorrective network since the correction has been largely cared for by \nthe characteristics of the pickups used.\n\nRecent development studies have established the possibility of \nflattening the response at the low frequency end and of raising the\n\nhigh frequency cut-off of the recorder. Fig. 3 shows a characteristic \nobtained with such a laboratory model. This shows uniform per- \nformance within +1 TU from 250 to 7,500 cycles and within \n+4 TU from 30 to 8,000 cycles. Although its immediate practical \nvalue might be limited by other portions of the system, this device \nis of great interest in that it establishes beyond question the fact \nthat an extremely broad range of frequencies can be successfully \nrecorded in the \u2018\u2018wax.\u201d\u2019\n\nThe broad, flat characteristic obtained with electric recorders has \nbeen made possible by so designing their elements that they constitute \ncorrectly designed transmission systems. In such a transmission \nsystem, whether it be an electrical recorder or a long telephone line, \na correct terminating impedance is required. \u2018The load imposed by \nthe \u2018\u2018wax\u2019\u2019 is somewhat variable but fortunately is rather small. \nIt has been found desirable to make the other impedances in the \nrecorder relatively large so as to dominate the system and thus \nminimize the effects of any changes in the impedance imposed on the \nstylus by the \u2018\u201c\u2018wax.\u201d\u201d The mechanical load used as a terminating \nimpedance and to control the device has consisted of a rod of gum \nrubber 25 cm. long. \u2018Torsional vibrations are transmitted along this \nrod. The rate of propagation is about 3,000 centimeters per second \nso that its length is equivalent to an ideal electrical line of about \n1,500 miles. The dissipation along this rubber rod is such that a \nvibration is substantially dissipated by the time it has travelled down \nthe line and back. It thus constitutes substantially a pure mechanical \nresistance, the magnitude of this resistance being approximately 2,500 \nmechanical abohms, referred to the stylus point as its point of \napplication.\n\nIn recording the usual procedure is to use a disc from 1 in. to 2 in. \nthick and from 13 in. to 17 in. in diameter, composed of a metallic \nsoap with small amounts of various addition agents to improve the \ntexture. This is shaved to a highly polished surface on a lathe. \nThis polished disc or so-called \u2018\u2018wax\u201d\u2019 is placed in a recording machine. \nIn Fig. 4 is shown what is essentially a high grade lathe arranged to \nrotate the \u2018\u2018wax\u201d\u2019 in a horizontal plane at a very uniform speed in a \ndefinite relation to the film with which it is synchronized. The \nrecorder with its cutting tool or stylus is moved relative to the surface \nof the disc, common phonograph procedure being to record from the \nouter edge of the disc towards the center, whereas in the Western\n\n3\u201c\u2018Some Technical Aspects of the Vitaphone,\u201d by P. M. Rainey, presented at \nthe meeting of the Society of Motion Picture Engineers at Norfolk, Va., April, \n1927.\n\nElectric Company theater system the direction of cutting is reversed. \nAfter a record has been cut into the \u201cwax,\u201d the \u201cwax\u2019\u2019 may be \nhandled and with proper precautions readily shipped from place to \nplace.\n\nThe shape of the groove varies somewhat in commercial practice. \nThe groove and stylus most commonly used with Western Electric \napparatus are shown in Fig. 5. The groove is approximately .006 in. \nwide and .0025 in. deep. The pitch of the groove is between .010 in. \nand .011 in. so that the space between grooves is about .004 in. Thus\n\nthe maximum safe amplitude is about .002 in. If this occurs at 250 \ncycles the corresponding amplitude at 5,000 cycles, assuming constant \nabsolute intensity of sound over this range, would be .0001 in.\n\nIt is important that a smooth groove be cut as any roughness in \nthe walls introduces extraneous noise in the reproduced sound. To \ninsure a truly smooth groove the surface of the \u2018\u2018wax\u2019\u2019 must be shaved \nto a high polish. The texture of the \u2018\u201c\u2018wax\u2019\u2019 must be fine and homo- \ngeneous which requires not only that the \u2018\u201c\u2018wax\u2019\u2019 composition be \ncorrect, but that it be operated at the proper temperature. \u2018\u2018Waxes\u201d\u2019 \nmay be obtained commercially which will operate satisfactorily over\n\nthe ordinary range of room temperature. The \u2018\u2018wax\u2019\u2019 must be \nlevelled in the recording machine with reasonable care. The stylus \nmust be sharp and so ground that the cut will be very clean. The \n\u201c\u201cwax\u2019\u2019 shaving is removed as cut by air suction. The operator is \naided in maintaining the correct depth of cut by the use of a so- \ncalled \u2018\u2018advance\u201d\u2019 ball which rides lightly on the \u2018\u2018wax\u2019\u2019 and serves to \nmaintain uniform depth of cut in spite of small inaccuracies of leveling \nof the \u2018\u2018wax\u201d\u2019 or deviations from planeness. The \u2018\u2018advance\u2019\u201d\u2019 ball is \nadjusted relative to the stylus by observing the cut with a calibrated \nmicroscope. A satisfactory operation of the recording machine re- \nquires an ordinarily skilled mechanic with reasonable experience.\n\nThe rate of rotation is dependent upon the diameter of the record \ngroove which is determined primarily by the length of time which it is \ndesired to have covered by a single disc. The controlling element is \nthe linear speed of the groove past the recorder or reproducer. In the \nWestern Electric system the speed varies from 70 ft. to 140 ft. per \nminute, in other words, of the same order of magnitude as with the \nfilm record. The wave-lengths also are about the same as for a \nsound record on a film. At the minimum linear speed the half wave- \nlength for a 5,000 cycle wave is .0014 in. If the minimum linear \nspeed is fixed at 70 ft. per minute and the groove spacing is fixed, \nthere is an optimum relation between the size of the record, the rate \nof rotation and the playing time. This is illustrated graphically in \nFig. 6.\n\nAfter a record has been cut, the sound may be reproduced directly \nfrom the \u2018\u2018wax\u201d\u2019 by using a suitable pickup or reproducer. Ordinary \nreproducers or pickups rest much too heavily on the records to be\n\nused on ordinary \u2018\u201c\u2018wax.\u2019\u2019 That this would be so is obvious from the \nfact that the vertical pressures between the point of the needle and \nthe record in an ordinary phonograph are of the order of 50,000 pounds \nper square inch. Obviously any such pressures would destroy a \ngroove cut in soft \u2018\u201c\u2018wax.\u2019\u2019 These high pressures have been necessary \nin order that the groove might properly drive the needle point of \nthe reproducer. Reduction of this pressure requires reduction of the \nimpedance offered by the needle point to transverse vibration.\n\n24Vo \nTm = MAX. PLAYING TIME IN MINUTES \nN = GROOVES PER INCH \nR = OUTSIDE GROOVE RADIUS-INCHES \nV, = MIN. LINEAR SPEED-FT. PER MINUTE \nr = INSIDE GROOVE RADIUS-INCHES\n\nFig. 6\u2014Relation between playing time and rate of rotation of disc for various \nvalues of R(R = 2r).\n\nThe design of a suitable \u2018\u201c\u2018wax\u201d\u2019 \u201cplayback\u201d\u2019 requires reduction of \nboth the mass and the stiffness of the reproducing system to a \nminimum. In the past such \u201c\u2018playbacks\u201d\u2019 have failed to reproduce \nthe higher and lower frequencies with much satisfaction. The device \nshown in Fig. 7 represents a large advance toward ideal reproduction.\n\nThe response of such a device when driven by a \u2018\u2018wax\u201d\u2019 record recorded \nat constant velocity over the frequency range is shown in Fig. 8. \nThis reproduction is not widely dissimilar from that obtained from \nfinished records with the best electric pickups now commercially \navailable and is sufficiently good to serve as a very valuable criterion \nin judging the quality of the record. The record may be played a \nnumber of times without great injury. The extent of the injury is \nindicated by the frequency characteristics obtained on successive\n\nplayings shown in Fig.9. They show little change in the low frequency \nresponse and a loss of about 2 TU per playing at high frequencies. \nThe practical value in studio work of being able to let an artist immedi- \nately hear and criticize the results of his own efforts can hardly be \noverestimated.\n\nFig. 9 Loss in response on successive playings of a wax playback driven by constant \nvelocity wax record.\n\nAfter a groove has been satisfactorily cut into the \u2018\u201c\u2018wax\u2019\u2019 record, \nthe usual procedure is to render the surface of the \u2018\u2018wax\u2019\u2019 conducting \nby brushing into it an extremely find conducting powder. It is then \nelectroplated. The technique in this step varies somewhat with the\n\ndifferent companies doing such work, although not in any fundamental \nmanner. \u2018The negative electroplate thus made may be used to hot- \npress a molding compound such as shellac containing a finely ground \nfiller. This first electroplate is called a \u2018\u2018master.\u2019\u2019 From it two \ntest pressings are usually made. If satisfactory the matter is then \nelectroplated with a positive, being first treated so that this positive \nplate may be easily removed. This positive is sometimes called an \n\u201coriginal.\u2019\u2019 From this in turn is plated a metal mold or \u201c\u2018stamper.\u201d\u2019 \nFrom these, duplicate \u201c\u2018originals\u2019\u2019 may be plated and from them, \nduplicate \u2018\u2018molds\u2019\u201d\u2019 or \u201cstampers.\u2019\u2019 These processes involve no \nmeasurable injury to the quality of the record and are comparatively \nsimple and extremely safe in practice. By this practice of making a \nnumber of duplicates it is possible to safeguard the \u2018\u201c\u2018master\u2019\u2019 and \ninsure against any accident which might destroy a valuable record. \nFrom a single \u2018\u2018stamper\u201d\u2019 it is not unusual to make a thousand finished \npressings. The time required for these operations is such that test \npressings are commonly obtained from the \u2018\u201c\u2018wax\u2019\u201d\u2019 in 12 hours. Recent \nrefinements in the art have reduced the time required so that finished \nrecords may, if necessary, be obtained in 3 hours after delivery of \nthe \u2018wax.\u201d\u2019\n\nVarious materials have been used in making the hard record or \n\u2018\u2018pressing.\u2019\u2019 In some cases the material has been made homogeneous \nand in others the surface is of a different material from that used in \nthe body of the record. Some have used a laminated structure. \nThere has not, however, been much latitude allowed the experimenter \nconcerned with materials for the hard record. \u2018The material has had \nto be quite hard and, in order to show a reasonable life, it has had to \ncontain sufficient abrasive to grind the needle quickly to a good fit. \nAt the beginning of the run of a new needle due to the small bearing \nsurface, the pressures are very high. They rapidly decrease so that \nwith an ordinary loud steel needle after one minute\u2019s wear in the \nordinary phonograph, the bearing area is increased to such an extent \nthat the pressure is only about 50,000 pounds per square inch. As \nthe needle continues to wear to a larger bearing surface, the pressure \nobviously continues to decrease. \u2018These high pressures and necessary \nabrasive characteristics of the record have introduced irregularities \nwhich are responsible for most of the extraneous noise commonly \nknown as \u201c\u2018surface\u201d\u2019 or \u2018needle scratch.\u201d\u2019 }\n\nThe \u201cpressing\u2019\u2019 copies the \u2018\u2018wax\u201d\u2019 record with a very high degree \nof accuracy so that if our attention be confined to frequency charac- \nteristics alone, the \u2018\u201c\u2018pressing\u2019\u201d\u2019 shows almost complete perfection.\n\nMoreover, it is cheap and durable, and reproduction of the sounds \nfrom this record calls for no fine adjustments or intricate apparatus \nas has been long evidenced by the broad use of the ordinary phono- \ngraph.\n\nThe major part of the extraneous or \u2018\u2018surface\u2019\u2019 noise found with \nthis method of reproduction comes from the material of the finished \nrecord. Recent progress has been made in reducing this noise. Asa \nresult of this, together with refinement in the plating processes, \nrecords used with Western Electric Company theater equipment \nduring the last two years have shown a reduction of 3 to 6 TU in \n\u201csurface\u2019\u2019 noise. \u2018This corresponds to eliminating 50 per cent to \n75 per cent of that previously present. It is not necessary to reduce \nthe level of \u2018\u2018surface\u2019\u2019 noise to the zero point but merely to the \nthreshold of audibility under the conditions of minimum auditorium \nnoise which are of interest. This noise masks the surface. More- \nover, it is not the absolute amplitude of the imperfection giving rise \nto \u2018\u201c\u2018surface\u2019\u2019 noise but the relative magnitude in comparison with \nthe useful sound amplitudes which counts. Thus, an effective re- \nduction in \u2018\u2018surface\u2019\u2019 could be made if we were willing to use larger \nrecords or if we were willing to reduce the playing time of the present \nrecords by increasing the spacing of the grooves and the amplitude at \nwhich the grooves are cut. Any large reduction in \u201csurface\u201d\u2019 noise \nmade by a reduction in the irregularities in the record material would \nopen the door to increasing the playing time of a record of given size. \nThere is no known absolute or fundamental reason why further \nimprovements in record materials may not be expected to reduce \nfurther the amount of \u2018\u201c\u2018surface\u2019\u2019 noise. Moreover, large advances \nin pickup design open distinctly new possibilities as to reductions in \nsuittacer\n\nIt has sometimes been thought that in order to reproduce high fre- \nquencies properly, the linear record speed would have to be increased \nor the size of the needle point reduced. At present the diameter of \nthe bearing portion of a representative needle is about .003 in. whereas, \nas mentioned before, the half wave-length for a 5,000 cycle wave is \n0014 in. The factor determining whether a needle will follow the \nundulation of the groove is not any consideration of the relative \ndiameter of the needle point and the undulation of the groove but \nrather the radius of curvature of the needle and the bend of the groove. \nAs indicated before, the amplitude at 5,000 cycles would be only \nabout .0001 in. if sounds of that frequency were as intense as those of \nlower frequencies (.002 in. at 250 cycles). Asa matter of fact, sounds \nof 5,000 cycles or more in speech or music are characterized by lower\n\nintensity than those of lower frequency. If, however, we assume an \namplitude of .0001 in. at 5,000 cycles and assume a linear record \nspeed of 70 ft. per minute, then the minimum radius of curvature \nof the undulation of the groove is .00193 in.t With the foregoing \nassumption, the radius of curvature of the undulation of the groove \nbecomes equal to that of the needle point at about 7,000 cycles. \nTaking into account the lower intensities of sounds encountered at \nthese high frequencies, it is obvious that present commercial needle \npoints are quite capable of following the high frequency undulations \nof the groove up to frequencies of at least 10,000 cycles. The limita- \ntions of high frequency reproduction commonly found in the past are \nassociated with limitations in the design of the pickup or reproducer \nand relate either to inability of the record groove to drive the needle \npoint, with resultant chatter, or inability of the pickup structure to \ntransmit high frequency motions from the needle point to the armature.\n\nLarge advances have been made within the last two or three years \nin designing electric reproducing structures. The mechanical im-\n\npedance at the needle point has been reduced so that the needle \npoint truthfully follows the undulations in the groove without necessi- \ntating excessive and somewhat destructive bearing pressures. At the\n\nsame time the transmitting structure has been so designed that a \nvery broad range of frequencies is properly transmitted from the \nneedle point to the armature. Moreover, proper mechanical loads \nhave been provided so that the motions after transmission are absorbed \nand hence not reflected back. This is another way of stating the fact\n\nthat resonance as ordinarily considered has been eliminated from \nthese structures.\n\nThe curves shown in Figs. 10 to 12 illustrate the steps which have \nbeen taken. The pickup shown in Fig. 11 is free from the resonances \nshown in that of Fig. 10. The resonances present in the earlier\n\nFig. 12\u2014Response of experimental pickup driven by constant velocity pressings.\n\npickup involved high needle point impedances in the region of these \nresonances. These high impedances involved large driving forces \ndestructive both to needle and record. Certain records were injured \nafter only a few playings with this reproducer. The later reproducer \nis characterized not only by considerably reduced average needle \npoint impedance, but, as shown by the curves, the resonance is\n\npractically eliminated and hence there is an even greater reduction \nfrom the maximum impedance which occurred in the earlier reproducer \nat resonance. Both needles and records have a relatively long life \nwith the later type pickup, which has been in commercial use for \nsome months. As is seen, the higher frequencies are reproduced in \nconsiderably better fashion. A third curve is given in Fig. 12. This \nwas obtained with a more recent experimental model in which a \nfurther large reduction in the needle point impedance had been \neffected and in which, in addition, a very much more rigid, though a \nlighter structure served to connect the needle point with the armature. \nThis model shows further reduction in wear and tear on the record \nand very greatly improved reproduction at the high frequency end \nof the scale.\n\nThe application of the processes of sound recording on \u2018\u2018wax\u2019\u2019 to \nthe synchronized film has involved meeting a number of conditions \nnot previously encountered in the phonograph field. One of the \nmost important of these relates to editing, cutting and rearranging \nof the picture. Various methods have long been used to copy or \n\u2018\u201c\u2018dub\u201d\u2019 a disc record. The prime requirement is that there be no \nsacrifice in quality. To attain this end records have sometimes \nbeen copied at very low speed. \u2018This method appears unnecessarily \nlaborious and slow and the results obtained are not altogether satis- \nfactory in the light of possibilities presented by pickups and recorders \nof the characteristics shown above. Rearrangement of material on \nrecords is entirely practicable, portions may be deleted or new portions \nadded either as a whole or the new sounds added to those already on a \nrecord\u2014in fact any changes of this type may be made which can be \nmade in the picture.\n\nThe detailed technique of \u2018\u2018dubbing\u201d\u2019 appears to offer no serious \ntechnical difficulties. The refinement reached and the extent of its \nfuture use may be expected to be governed by the demand in the \nsynchronized motion picture field.\n\nSynopsis: The light valve developed by Bell Telephone Laboratories is an \nelectromagnetic shutter consisting of afloop of duralumin tape formed into a \nslit at right angles to a magnetic field. Sound currents from the microphone \nand amplifier flow in this loop causing it to open and close in accordance \nwith the current variations.\n\nThe slit is focussed by a lens on the sound negative film. An incandescent\n\nribbon filament is focussed on the light valve, and the light passed by the \nundisturbed slit appears on the film as a line at right angles to the direction \nof the film travel. As the valve aperture is modulated by sound currents, \nthe film receives a varying exposure and a sound record of the variable \ndensity type is obtained. \n. For talking pictures such a sound film is made on a separate recording \nmachine synchronized with the camera and is printed alongside the picture \non the finished positive. The prints are displaced so that the sound is \nadvanced over the corresponding picture. This is in order that the sound \nmay be projected at a point of continuous film motion below the picture \ngate.\n\nIt is not difficult to specify the requirements of this type of sound \nfilm. So far as possible the exposure of the negative must be kept \nwithin the straight line portion of the Hurter and Driffield curve for \nthe emulsion chosen, and the print must be timed with the same \nrestriction. The development of the negative and of the print must \nresult in a positive where the transmission of each element of length \nis proportional to the exposure of the corresponding element of the \nnegative. The light modulator must be supplied with undistorted \npower from the recording microphone and amplifier. When the \npositive is projected, the striations of the sound track must be enabled \nto modulate the illumination of a photo-sensitive cell to retranslate \nthe photographic effect into electrical current which shall be a fair \ncopy of the microphone current generated by the original sound. \nFrom this point on the problem is the familiar one of sound re- \nenforcement, the film and cell having taken-the places of the sound \nsource and microphone.\n\nFig. 1 shows a photograph of the light valve, invented in 1922 by \nE. C. Wente of the Bell Telephone Laboratories. Essentially, it \nconsists of a loop of duralumin tape suspended in a plane at right\n\n1 Presented before Society of Motion Picture Engineers at Lake Placid, New \nYork, September 25, 1928. =F ;\n\nangles to a magnetic field. The tape, 6 mils wide and 0.3 mil thick, \nis secured to windlasses A and A\u2019 and stretched tight by the spring \nheld pulley B. At points C and C\u2019 insulated pincers confine the \ncentral portions of the tape between windlasses and pulley to form a \nslit 2 mils wide. Supporting this loop and adjusting devices is a \nslab of metal with central elevation D, which constitutes the armature \nof an electromagnet. The central portions of the loop are supported \non insulating bridges to lie 3 mils above the face of D; here the sides \nof the loop are centered over a tapered slot, 8 mils wide by 256 mils \nlong in this plane, opening to 204 mils by 256 mils at the outside face \nof the armature. Viewed against the light, the valve appears as a \nslit 2 mils by 256 mils.\n\nThe electromagnet core has a similar elevation opposing D across \nan air gap of 8 mils which closes to 7 mils when the magnet is energized \nfrom a 12 volt battery. A tapered slot in the magnet core begins \n8 mils wide by 256 mils long and opens with the same taper as the \nslot in the armature. When the assembly of magnet and armature is \ncomplete, the valve constitutes a slit 2 mils by 256 mils, its sides \nlying in a plane at right angles to the lines of force and approximately \ncentered in the air gap. The windlasses A and A\u2019, one of which is \ngrounded, are connected to the output terminals of the recording \namplifier. If the magnet is energized and the amplifier supplies a \nsine wave current from an oscillator, the duralumin loop opens and \ncloses in accordance with the current alternations.\n\nWhen one side of the wave opens the valve to 4 mils and the other \nside closes it completely, full modulation of the aperture is accom-\n\nplished. The natural frequency of the valve is set by adjusting the \ntension applied by the pulley B; for reasons which involve many \nconsiderations the valve is tuned to 7,000 cycles per second. Under \nthese circumstances about 10 milliwatts of A.C. power are required \nfor full modulation at a frequency remote from resonance; about one \none-hundredth of this power at the resonant frequency. The im- \npedance of the valve with protecting fuse is about 12 ohms.\n\nPLANE OF _ PLANE OF PLAN \nRIBBON OF VALVE IMAGE ON \nLIGHT RIB an FIL \nX 0.256\u201d ath p \n( x ) K or \u2018x a 128\")\n\nIf this appliance is interposed between a light source and a photo- . \ngraphic film we have a camera shutter of unconventional design. \nFig. 2 shows a diagram of the optical system for studio recording. \nAt the left is a light source, a ribbon filament 18 ampere projection \nlamp, which is focussed on the plane of the valve. The light passed \nby the valve is then focussed with a 2 to 1 reduction on the photo- \ngraphic film at the right. A simple achromat is used to form the \nimage of the filament at the valve plane, but a more complicated \nlens, designed to exacting specifications by Bausch and Lomb, is \nrequired for focussing the valve on the film. The undisturbed valve \nopening appears on the film as a line 1 mil by 128 mils, its length at \nright angles to the direction of film travel. The width of this line \nvaries with the sound currents supplied to the valve, so that the film \nreceives a varying exposure: light of fixed specific intensity through a \nvarying slit.\n\nFig. 3 shows a studio recording machine with the door of the exposure \nchamber open. In this machine the film travels at 90 feet per minute, \nand the sound track is made at the edge away from the observer. \nThe line of light, the image of the valve, overruns the perforations \nby 6 mils, extending toward the center of the film 122 mils inside the \nperforation line. The right-hand sprocket serves to draw film from \nthe feed magazine above and to feed it to the take-up magazine below; \nthis sprocket is driven from the motor shaft through a worm and \nworm-wheel. The left-hand sprocket engages 20 perforations and is \ndriven through a mechanical filter from a worm and worm-wheel\n\nsimilar to that driving the feed sprocket. The mechanical filter \nenforces uniform angular velocity of the left-hand sprocket which \ncarries the film past the line of exposure: the focussed image of the \nvalve: balancing of the flywheel which forms part of this mechanical \nfilter holds the angular velocity constant to one-tenth of one per cent, \ndespite the imperfections of the driving gears.\n\nSo far we have provided a means for driving the film and a means \nfor modulating the light thereon, but we have not chosen the average \nillumination about which the modulation is to take place. The \nmaximum exposure corresponds to the maximum opening of the \nvalve and is therefore double the average.\n\nbe developed and draw the Hurter and Driffield curve for this contrast \nfor the emulsion chosen for the negative sound record. The maximum \nexposure should correspond to the beginning of over-exposure, the \naverage should be half this. The Hurter and Driffield curve will give \nthe density of the over-exposure point for the chosen contrast and the \ndensity for half this exposure. Let the machine be run to expose \nfilm to light through the unmodulated valve for several values of \nthe lamp current. Develop the film and measure the densities due \nto the various values of lamp current. Select, by interpolation if \nnecessary, the lamp current which corresponds to half over-exposure. \nWith this current in the lamp the machine is ready to make a sound \nrecord, since the focussing of the valve has already been. done and \nmanufacturing specifications insure that the line of illumination shall \nlie, within 3 minutes of arc, at right angles to the direction of film \ntravel.\n\nConsider at this point the procedure in the recording studio. Adding \nsound to the picture introduces no complication of technique other \nthan to require sufficient rehearsing to make sure of satisfactory \npick-up of the sound: microphone placement must be established and\u2019 \namplifiers adjusted to feed the light valve currents which just drive \nit to the edge of overload in the fortissimo passages of music or the \nloudest utterances of speakers.\n\nIn Fig. 3 the photograph shows a photoelectric cell mounted \ninside the left-hand sprocket, which carries the film past the line of \nexposure. Fresh film transmits some 4 per cent of the light falling \non it, and modulation of this light during the record is appreciated by \nthe cell inside the sprocket. This cell is connected to a preliminary \namplifier mounted below the exposure chamber, and with suitable \nfurther amplification the operator may hear from the loud speaker the \nrecord as it is actually being shot on the film. Full modulation of \nthe valve implies complete closing of the slit by one side of the wave \nof current; this modulation should not be exceeded or photographic \noverload will abound.\n\nFig. 4 is a schematic diagram of the studio equipment for sound \nrecording. Provision is made for combining if desired the contri- \nbutions of several microphones on the set. This combination is\n\nunder the control of the mixer operator in the monitoring room, \nviewing the set through a double window in the studio wall. The \nmixer controls also the gain of the amplifiers for the recording machines.\n\nThe diagram shows relays which permit the mixer to connect the \nhorn circuit either directly to the recording amplifier or to one or the \nother of the monitoring photoelectric cells in the film recorders. \nThe direct connection is used in preparing the sound pick-up in the \nstudio: the program is rehearsed until satisfactory arrangement of \nmicrophones and of amplifier gain is effected. The electrical charac-\n\n== 500 w \nAT TENUATORS \n[J \nwa COILS \n= Sey: \nYP ' ORDI \nHey aes : i  (RECORDIN \n5 \u00a9 ] \nGAIN so Lely i Pee Pe \nR 4 CONTROL ! 2 ui \n(Zico E 2 -#E om re \\ p-86692\n\nteristic of this direct monitoring circuit is so designed that the sound \nquality heard in the horns shall be the same as the quality to be \nexpected in the reproduction of the positive print in the theater. \nAcoustic treatment of the walls of the monitoring room secures the \nreverberation characteristic of the theater, and the monitoring level \nis so adjusted that the mixer operator hears the same loudness that \nhe would wish to hear from the theater horns. It is capitally im- \nportant that the operator judge his pick-up on the basis of sound \nclosely identical in loudness and quality with that to be heard later in \ntheater reproduction.\n\ncircuit, the output of the recording amplifier is applied to the light \nvalves and the monitoring horns are connected to the photo cell \namplifiers on the recording machines. With no film in the machine \nand at a convenient lamp current a complete rehearsal is made to \nverify the operation of the valves at the proper level. Film is then \nloaded, cameras and sound recorders are interlocked and starting \nmarks made on all films by punches or light flashes.\n\nA light signal from the recording room warns the studio, which \nafter lighting up signals back its readiness to start. \"The machine \noperator starts the cameras and sound recorders, brings up the lamp \ncurrent to the proper value, and when the machines are up to speed \nsignals the studio to start. During the recording, the mixer operator \nmonitors the record through the light valves, thereby assuring himself \nthat no record is lost.\n\nIn the choice of emulsion for the sound negative, the usual designa- \ntion of speed may be disregarded, because it is desired to make the \nexposure of the unmodulated track many times the under-exposure \nof the emulsion used. The advantages of positive emulsion for the \nsound negative have come to be generally recognized; positive has \nbeen used by Bell Telephone Laboratories since 1924. The scale of \nEastman positive film is about 20 to 1; we adjust the recording lamp \ncurrent to give an illumination on the film for the unmodulated track \nof 10 times the under-exposure. After one lamp has been calibrated \nas described before it may be replaced when necessary by another \nin which the wattage in the ribbon filament is the same; the light \nemission is very closely correlated with the wattage. Where the \nunmodulated or average exposure is ten times the under-exposure \nminimum, 90 per cent modulation of the light can be permitted with- \nout running into under-exposure on the faint side of the wave. For \nsound currents reaching 100 per cent modulation of the light, 90 \nper cent of the wave is free from distortion; if the average light were \nhalved, still 80 per cent would be free from distortion. There is \ntherefore considerable latitude in the average exposure, and the \nnegative is satisfactory if the transmission of the unmodulated track \nlies between fairly wide limits.\n\nThe choice of the negative sound gamma is determined by the \npractice of the laboratory in regard to picture development. It is \nusual to see on the screen pictures whose overall gamma considerably \nexceeds unity. On the sound track the overall gamma should equal \nunity, and the development of sound negatives should be uniform, \nthough that of picture negatives is left to the judgment of the finisher.\n\nTheoretically, it should be immaterial what combination of reciprocal \nvalues is chosen for the negative and positive sound gammas. Prac-\n\ntically, we have to recognize the existence of ground noise in all \nrecords and take precaution to minimize it. No matter how ex- \ncellently we reproduce the fortissimo passages, our record is unsatis- \nfactory unless the ground noise is low enough for a wide volume range, \nthat is, a wide range in level between fortissimo and pianissimo. \nWhether our negative sound record is made on negative or positive \nemulsion, there is always the danger that in reproduction we shall \nencounter variations in transmission from point to point due to local \nvariations in the celluloid base, to local action of the developing agent, \nor to a developer excessively granular in action. The photoelectric \ncell is able to recognize variations of 1/10 of 1 per cent, whereas the \neye ignores contrasts under 2 per cent. These local variations in \ntransmission, continued to the positive print, constitute the ground \nnoise.\n\nThe remedy is, in part, to choose a developer as little granular in \nits effect as possible. In part, to insist on machine development of \nthe sound film with thoroughly agitated developer. Further, to \ncarry the sound development to a high gamma; this obviates to a \nlarge extent flow marks of the developer, and goes a long way to \nescape local variations in the base by developing the negative striations \nto be conspicuous in comparison.\n\nIn 1924 we concluded that the optimum choice was positive emulsion \ndeveloped to unit gamma for both sound negative and sound print. \nThis is feasible for sound records separate from pictures, but a compro- \nmise must be made for the combination of sound and picture in a \nsingle positive print. Here the positive development required for a \nsatisfactory picture is always to a gamma far above unity.\n\nIt is customary to develop picture negatives by inspection, having \nin mind the uniform positive development to be undergone by the \nprints from these negatives. The gamma of these positives need \nnever exceed 1.8; the sound negative then should be developed to \n0.55. In order not to disturb the practice of the film laboratory, we \nask that the positive development be standardized and its gamma \nascertained, the reciprocal of this gamma then arranged for in the \nstandardized negative development. A negative gamma above 0.5, \ntogether with the precautions of careful handling, permits the realiza- \ntion of an adequate volume range.\n\nIt is beyond the scope of this paper to discuss the details of manipu- \nlation and of choice of developer, but I wish to acknowledge the \ncooperation of Mr. J. W. Coffman in the solution of such problems. \nThe problem is the reduction of ground noise, and its seriousness is \nnot to be diminished by choosing a different recording method.\n\nIn printing the sound negative, a uniform density for the print of \nthe unmodulated track is desired. The volume of reproduced sound\n\nfor a given reproducing light source, varies directly with the average \ntransmission and the per cent modulation of this average. This \naverage density should be on the straight line portion of the positive \nHurter and Driffield curve, far enough to keep the denser negative \nportions from reaching the under-exposure region. For Eastman \npositive film a suitable transmission of the unmodulated portion of \nthe sound print is 35 per cent, referred to air, for the usual values of \npositive gamma: 1.4 to 1.8. At this average transmission only the \npeaks of the recorded sound will encroach on the region of under- \nexposure. For the reciprocally developed negative track the region \nof under-exposure will have been reached by occasional peaks on the \nother side of the wave, and such photographic distortion as exists will \nbe balanced between positive and negative.\n\nHere we appropriately consider the photographic distortion as it \noccurs in variable density records. If the entire negative exposure \nhas been confined to the under-exposure region of the emulsion chosen, \na huskiness will result in the reproduction which can not be corrected \nby any known technique. But if the unmodulated negative trans- \nmission, for a gamma of 0.55, is about 16 per cent referred to air, \n90 per cent of the wave will be clear of under-exposure, and experience \nshows that the ear detects no distortion. In telephonic terms, \neverything at a level 1 TU below full modulation will be free from \ndistortion, and the peaks will be substantially perfect. The same \nmay be said of the positive printed to an average transmission of \n35 per cent, provided the overall gamma approximates unity.\n\nIt has been calculated that if the overall gamma departs from \nunity by 0.2 in either direction, a harmonic of 5 per cent amplitude \nof the fundamental will be introduced. Experimentation has shown \nthat a 5 per cent harmonic is the least detectible. We state then the \ntolerance on the overall gamma for the sound track as 0.8 to 1.2. \nVariation of corresponding amount in the contrast of a picture print \nis intolerable; therefore greater latitude in contrast is permissible in \nthe sound record than could be tolerated in the accompanying picture.\n\nIn printing these sound negatives in combination with pictures for \nprojection in the theater, it is customary at the present time to print \none negative, masking the space needed for the other, then run the \npositive again through the printer with the other negative, masking \nnow the space already printed. In printing the picture negative, \nlight changes are made as usual; for the sound negative the light is \nregulated to result in 35 per cent transmission of the unmodulated \ntrack after positive development. Provision of suitable masks in the \ncamera has been made to show in the finder and expose on the film \nonly the portion which will be available for picture projection.\n\nIn the theater projector, the sound gate is located 14.5 inches \nbelow the picture gate, in order to project the sound record at a \npoint where the film is in continuous motion. Therefore in the \nprinting it is arranged to print the sound negative displaced along \nthe length of the positive enough to bring the sound 14.5 inches ahead\n\nNOTE: \nIN PRINTING, THE SOUND \nIS DISPLACED 14.5\" AHEAD \nOF THE PICTURE WITH WHICH \nIT IS SYNCHRONIZED.\n\nFig. 5\u2014Picture and sound track dimensions of synchronized sound film for standard \n35 mm. positive stock.\n\nof the corresponding frame. The printer apertures are chosen to \ngive a dark no man\u2019s land 17 mils wide between picture and sound \ntrack; the latter at the outside is separated 4 mils from the inner \nperforation edge. 7\n\nFig. 5 exhibits the present practice for the finished positive. \u00bb It will \nbe seen that the sound track covers 100 mils clear, and is illuminated\n\nin the projector by a line of light 80 mils long, 1 mil wide, centered on \nthe striations. This gives a margin of 10 mils at each end of the \nreproducing line, an allowance for lateral shifting of the film on the \nsprocket teeth.\n\nIn conclusion, let me estimate the quality of the sound record to \nbe expected. Assume that the recording lamp current has been set \nto within 5 per cent of the theoretical optimum, the overall gamma \nheld between 0.8 and 1.2, and the final average positive transmission \nis between 32 per cent and 38 per cent. \u2018Then the distortion of wave \nform due to photographic handling is so small that the ear can not \ndistinguish the record from a theoretically perfect one. The fre- \nquency-amplitude characteristic of the reproduced sound remains to \nbe stated.\n\nDue to the fact that the element of illumination, both in recording \nand in reproducing, is 1 mil wide instead of infinitely narrow, the \nfinal print will not reproduce the higher frequencies as efficiently as \nthe lower. For example, at the standard speed of 90 feet per minute, \nthe line of illumination covers on the film an entire cycle length of the \nfrequency of 18,000 cycles. This frequency is therefore extinguished \ncompletely. The drooping characteristic resulting from this effect, \ncalled the film transfer loss, may be largely offset by judicious choice \nof electrical characteristics and by taking advantage of the mechanical \ntuning of the light valve.\n\nIn Fig. 6 is shown in curve A the light modulation by the valve in \nrecording for constant sound pressure of various frequencies at the \ntransmitter; in curve B the overall characteristic of the reproduction \nin terms of electrical power delivered to the loud speaker for constant \nsound pressure at the transmitter in the studio. Experience shows \nthat curve B is close enough to flat; the success of the record, as of \nthe picture, depends on the director.\n\nSynopsis: The reproduction of the synchronized sound picture of today \npresents no serious problem of synchronization, for this factor has been \npractically eliminated by the perfection of electrical means for reproducing \nsound with equipment which may be coupled mechanically to the picture \nprojector.\n\nThe important problem of the present day, in connection with the \nreproduction of synchronized sound pictures, is the provision of suitable \nmeans for maintaining a constant speed of the sound reproducing mechanism \nin order that the pitch of the sound being reproduced may not suffer any \nsudden change which would be sensed by a good musical ear. Control \ncircuits using vacuum tubes with a frequency bridge as a speed standard \nwith provision for manual variable speed control are described and explained \nfor use with both A.C. and D.C. motors. Remote synchronization permitting \nthe recording of pictures and sound simultaneously on equipment located \nsome distance apart is obtained by a modification of the Michalke electric \ngear system.\n\nHEN Thomas A. Edison gave a demonstration of his talking \nmotion pictures nearly sixteen years ago one of his chief \nproblems was proper synchronization between his acoustic phonograph \nand the motion picture projector. It was then necessary to locate \nthe phonograph behind the screen in order to make the sound appear \nto come from the picture. A system of belts and pulleys running \nfrom one end of the theater to the other was used to secure synchro- \nnization with the projector in the booth.\n\nThe development of the electrical reproducer has made it possible \nto locate the turntable and reproducing mechanism in the projection \nbooth permitting a direct mechanical coupling between it and the \nprojector. The horns are located behind the screen and electrically \nconnected by wires with the electrical reproducer.\n\nHowever, such mechanical coupling between the projector and sound \nrecorder (either of the disc or film type) makes it necessary to provide \nvery close speed regulation on the projector motor, since variations in \nspeed produce proportional changes in the pitch of the sound.\n\n1 Presented before Society of Motion Picture Engineers at Lake Placid, New \nYork, September 24, 1928.\n\nA good musical ear while having a sense of absolute pitch of only \nabout 3 per cent is extremely sensitive to sudden changes in pitch. \nIt has been found that a sudden change in pitch as small as one half \nof 1 per cent may be noticed if made abruptly. In order to properly \ntake care of this requirement, therefore, the speed regulation or \nchange in speed of the motor drive over normal variations in line \nvoltage and load should be held within 2/10 of 1 per cent.\n\nThe absolute speed must also be held near these limits. since at \nthe end of a film it is necessary to switch from one projector to another \nwith minimum change in the pitch of the sound reproduction.\n\nA study of the voltage variations in power supply systems indicated \na range from 100 to 125 volts. At a particular location the normal \nvariation of voltage was found to be 5 per cent above or below the \nmean value with occasional momentary variations of as much as \n10 per cent above and below mean value.\n\nAn investigation of variations in frequency of the supply voltage \nshowed that in the large cities the frequency was held very accurately \nat 60 cycles. In New York City for example the frequency stays \nwithin one quarter of 1 cycle and does not change rapidly. However, \nin some small power systems the frequency varied as much as 5 cycles \nand in some cases was subject to rapid changes in frequency.\n\nThe load of the motor is due mainly to mechanical friction in the \nprojector and take-up mechanism. This load was found to be on \nthe average 1/10 of a horse power but subject to wide variations. \nIn the case of a new machine with a stiff adjustment of the take-up \nmechanism, the load was found to be as high as one-fifth of a horse\n\nA consideration of the variables just discussed imposes rather \nsevere requirements of speed control, the two extremes being (1) the \ncombination of low line voltage, low frequency and heavy load, \n(2) the combination of high line voltage, high frequency and light load. \nOrdinarily it might be possible to compromise and not provide for \nsuch an extremely unfavorable combination of requirements. How- \never, it must be borne in mind that in the case of a musical program \nthe failure of the speed regulating system for even as short a time as a \nfraction of a second would be a very serious matter causing the music \nto sound off pitch similar to a phonograph which has run down while \nin operation.\n\nAn examination of the standard commercial types of speed control \nindicated that there was nothing exactly suitable. The nearest \napproach to a suitable governor is the standard type of phonograph \ngovernor but this friction brake type of governor has serious objections \nif applied to a motor of considerable power output. In order, there- \nfore, to have a control system which would be free from maintenance, \nit was necessary to develop a special form of control system for the \npurpose. Fig. 1 shows a photograph of the A.C. motor and its control \ncabinet.\n\nFig. 2 shows the A.C. motor circuit which consists of a repulsion \ntype of motor coupled to a small auxiliary alternator providing a \nfrequency of 720 cycles which through a control circuit is made to \noperate a variable reactor across the armature terminals of the motor. \nIf the speed of the motor is too high, the control circuit produces a \nmaximum impedance in the reactor L,; thereby reducing the armature \ncurrent of the motor and causing it to slow down. While if the \nspeed is too low the control circuit causes the reactor Li to have a \nminimum of impedance increasing the armature current and causing \nthe motor to speed up.\n\nThis reactor ZL; is of the D.C. saturating type having two outer \nlegs with A.C. windings and the middle leg with a D.C. winding. \nThe A.C. flux circulates around the two outer legs. The D.C. flux \nflows from the middle leg and returns through the outer legs in \nparallel. When D.C. flux is sent through the middle leg it saturates\n\nreactance is old and was employed by Alexanderson as a magnetic \nmodulator in his early radio work.\n\nFig. 3 shows one element of the speed control circuit. This consists \nof a bridge circuit having one variable arm and three fixed arms. \nThe variable arm comprises a tuned circuit consisting of the in-\n\nductance LZ and the capacity C which are designed to tune at exactly \nthe frequency corresponding to the desired motor speed. When this \ncircuit is in tune it has a resistive impedance which is balanced by the \nfixed arm R of the bridge. The other two fixed arms on the left\n\nare windings of a transformer with a mid tap. If a voltage 2F, \nhaving a frequency of 720 cycles (which is the frequency corresponding \nto the desired speed of 1,200 R.P.M.) is supplied to the bridge it will \nbe apparent that the output voltage E\u00bb will be zero. If, however, \nthe speed is low the tuned circuit will have a condensive reactance \nwhile if the speed is high it will have an inductive reactance. The \noutput voltage HE\u00bb, will, therefore, change abruptly 180 electrical \ndegrees from a speed below 1,200 to a speed above 1,200. This \ncharacteristic is shown in Fig. 4.\n\nthe bridge. In this way an overall characteristic of the desired \nsharpness is secured using a comparatively small and inexpensive \ncoil and condenser.\n\nFig. 5 shows the complete control circuit. The output from the \nbridge circuit is supplied to the grid of tube V4 which is called the \ndetector tube. The plate voltage of this tube comes from the 720- \ncycle generator through the step-up transformer 74. The phase of \nthis voltage, therefore, remains constant. The phase angle of the grid \nvoltage, however, comes from the bridge output circuit through the \nstep-up transformer 73 and as previously explained suffers a sudden \nreversal of phase as the speed passes through 1,200 R.P.M. Fig. 6 \nshows the resulting current characteristic through tube V4. This \ncurrent flows through coupling resistance R; which drives the grids\n\nof tubes V; and V2 negative. This in turn reduces the plate current \nthrough tubes Vi and V2 and hence through the D.C. winding of \nthe inductance ZL; controlling the armature current of the motor.\n\nHOEGGLD \nchee. \nCi) 7 \npies \nCane \na aes \nROTOR - Z \nJOCCOIGSS ; \ni STATOR . \n| ees \ni - - \u2014C}\u2014\u2014 \n2 \nMOTOR FUSES \n110 V.A.C Seaiee \nEINE SWITCH \nGRD. A.C.CONTROL CABINET Q \nFig. 5\u2014A. C. control circuit diagram. \nLd \noz \n> WwW \npote \n= \nWe \noe \nOG \nLg \napn \nLu 0) \nO\n\nFig. 7 shows the performance characteristics of the motor. It will \nbe noted that the actual speed characteristic is practically flat. This\n\nflat characteristic is secured by a compensating network consisting \nof the resistances Re, R; and R, and the condenser C,. This com- \npensating network feeds back on the grid of tube Via portion of the \nvoltage drop across the D.C. winding of inductance L; thereby cor- \nrecting for the \u201cstatic fluctuation\u2019? of the control circuit. By a \nsuitable adjustment of this compensating resistance the control circuit \nmay be arranged to give flat regulation, under regulation or even \nover regulation if desired. Fig. 7 has been drawn with line voltage \nas the variable. A similar characteristic is also obtained with load \nas the variable instead of voltage.\n\nAn interesting point in connection with this compensation circuit \nis the necessity for avoiding hunting or surging of the speed. It is a \nwell-known property of all forms of governors that if they are adjusted \nto too great a sensitivity the speed instead of remaining constant will \nfluctuate up and down about a mean value. The simplest method of \npreventing such speed fluctuations is to decrease the sensitivity of \nthe governor allowing a bigger change in speed with load (or voltage) \nand then compensating for this change of speed or \u2018\u2018static fluctuation\u201d\u2019 \nby means of a delayed action compensator. This phenomenon is well- \nknown in the mechanical governor art and is described by Trinks in \nhis book \u2018Governors and the Governing of Prime Movers.\u2019 The \nelectrical equivalent of this mechanical system is obtained by intro-\n\nducing the condenser C2 in series with the high resistance R,. When a \nchange in current through the regulating reactance L; occurs the \ncorresponding change in voltage drop is not transmitted to the con- \ndenser C2 immediately, but C, changes its voltage after a certain \ntime lag (approximately 1 second), required to charge the condenser \nthrough the resistance Ry. The introduction of this time lag restores \nthe precision of the circuit to the flat characteristic desired without \nintroducing hunting.\n\nBy throwing the switch S; to the right the operator can disconnect \nthe tuned circuit control and substitute a potentiometer P; as a\n\nsource of grid voltage for tube Vs. By means of this potentiometer \nthe operator can adjust the speed of the motor at any speed from \n900 to 1,500 R.P.M. corresponding to 68 to 112 feet of film per minute. \nThis feature is employed for ordinary motion picture work where it is \nunnecessary to synchronize the picture with the sound. The regu- \nlation of the circuit under these conditions is sufficiently good for \nordinary motion pictures.\n\nAn interesting feature in this connection is that theaters in many \ncases have preferred to use the regulated speed position for ordinary \nmotion pictures as well as synchronized pictures. The reason for \nthis being that with the speed of the projector precisely controlled \nthe orchestra leader is better able to keep his orchestra in step with \nthe picture indicating apparently that closer speed regulation than is\n\nat present provided would be desirable for ordinary motion pictures \nas well as synchronized pictures.\n\nA.C, circuit. Since a strengthening of the field of the D.C. motor is \nrequired in order to reduce the speed it is necessary to reverse the \nphase relationship of the transformer 7,, so that the current in the \ndetector tube decreases at speeds above 1,200 instead of increasing \nas in the case of the A.C. circuit.\n\nThe operation of the circuit is as follows: When the line switch is \nfirst thrown the motor acts as an ordinary D.C. shunt motor and \naccelerates. At low speeds the output from the 720-cycle generator \nis low and consequently there is no plate voltage supplied to the tubes \nand no current through the auxiliary field winding. The field is, \ntherefore, weak and the motor speeds up. This condition is main- \ntained until the equilibrium speed of 1,200 R.P.M. is approached. \nThe phase angle of the voltage supplied to the grid of the tube V3 is \nthen in phase with the voltage supplied to the plate so that the grid \nof the tube goes positive at the same time that the plate is positive.\n\nThis causes a current to flow through the coupling resistances R; and \nR\u00bb. which drives the grids of tubes V; and V2 negative, thereby keeping \ndown the current through these tubes and hence maintaining a weak \nmotor field. The motor, therefore, continues to accelerate until a \nspeed of 1,200 R.P.M.isreached. At this point as previously explained \nunder the description of the bridge circuit, the phase of the output \nsuddenly reverses whereupon the grid of the detector tube goes \nnegative at the same time that the plate goes positive, thereby cutting \noff the current through the detector tube V3 and reducing the negative \nC voltage on the grids of tubes V; and V2. This increases the plate \ncurrent through the regulating field thereby stiffening the field of \nthe motor and checking its rise in speed. In practice the current \nthrough the detector tube is neither at one extreme nor the other but \nreaches an equilibrium at the speed of 1,200 R.P.M. A feedback \nnetwork having the delay feature for prevention of hunting is included \nin the same manner as previously described for the A.C. circuit. The \ncharacteristic curves for the D.C. motor are similar to those shown in \nFig. 7 for the A.C. motor.\n\nFor the operation of ordinary motion pictures the motor is changed \nto a simple shunt D.C. motor by the switch S,; and the speed varied \nby means of the field rheostat.\n\nIt might appear that the simplest method of securing synchronization \nin recording work would also be mechanical connection between the \nrecording machine and the camera. It has been found desirable, \nhowever, from a practical standpoint to have the camera movable \nwith respect to the recording machine as the recorder has to be \naccurately lined up and adjusted and is not essentially a portable \nmachine whereas the camera in ordinary motion picture work must be \na portable piece of equipment. It has been necessary, therefore, to \ndevelop a motor drive equipment which will satisfactorily interlock \nthe camera and the recording machine but leave the camera unit \nportable. It is essential that the interlock should hold not only \nduring normal conditions but during acceleration and deceleration. \nIn other words, the system must be the full equivalent of a mechan- \nically geared system. The principle employed is old being disclosed \nin a patent issued to Michalke in 1901. In Fig. 10, A and B are two \nunits which it is desired to interlock. Each unit has a three phase \nstator and a three phase rotor, the latter provided with slip rings. \nMagnetizing current for the system is supplied from an independent \nthree phase, 60-cycle source. If the rotors of A and B are in exactly\n\nthe same positions with respect to the stators it is evident that the \ne.m.f.\u2019s produced in them by transformer action will be identical as \nto voltage and phase. Consequently there will be no flow of current \nover the rotor leads and hence no torque developed. If, however, \nunit A is turned through a small angle then the phase of the e.m-.f.\u2019s \nproduced in the rotor circuits will differ from that in B and a current \nwill flow in the rotor circuits producing a torque which will tend to \nmake unit B assume the same position as A. If A is rotated con- \ntinuously B will follow it up to synchronous speed of the stator field \nat which point the torque will drop to zero since no e.m.f. is induced \nin the rotor of either machine.\n\nThe portion of the circuit shown in Fig. 10, is merely the equivalent \nof a mechanical gear, neither unit tending to rotate as a motor by \nitself. In order to produce such rotation, therefore, a distributor \nset is added as shown in Fig. 11, the distributor acting, to Use a \nmechanical analogy, as the driving gear of the system and each of the \nindividual units of the system as driven gears. The distributor is \nitself driven by a D.C. motor provided with the speed control circuit \npreviously described and shown in Fig. 9. The speed of the system \nis thus solely dependent on the D.C. driving motor and independent \nof the 60-cycle excitation frequency.\n\nIn practice the system is controlled by an operator at the distributor \nset and by means of switches any desired number of cameras, recording \nmachines, or projectors may be employed. The projecting machines \nare used in case it is desired to make up a sound record to accompany \nan ordinary motion picture film which has previously been recorded \nwithout sound accompaniment. It has been found that the system \noperates very satisfactorily and requires very little maintenance.\n\nWhen starting up for the first time it is necessary that the various \nunits should line up properly as to phase otherwise there will be a \nlocal flow of current in the rotor circuits, which will cause the motors \nto operate as induction motors and run away. Under running con- \nditions the system is very stable showing no tendency to hunt or surge \nbetween units, for the reason that being polyphase each phase as it \nbecomes inactive (when the induced e.m.f. passes through zero) acts \nas a damping winding for the other two active phases.\n\nIf the load on a particular unit of this system is varied there will \nbe a variation in the phase angle between this unit and other parts \nof the system in the same manner as in the case of a synchronous \nmotor of the ordinary type. The magnitude of this phase angle, \nhowever, does not vary more than 30 electrical degrees or 15 mechanical \ndegrees and is sufficiently small so that it produces an inappreciable \neffect on the synchronization.\n\nIn the design of this equipment, first consideration has been given ' \nto its precision and reliability in operation and the provision of \nadequate margins to care for all variations in service conditions. Asa \nresult it has been possible to maintain a high standard of quality in \nmusic and speech reproduction.\n\nSynopsis: The general problem involved in the design of a system \nsuitable to be used to record and reproduce sounds such as are required for \n\u2018\u2018talking\u2019\u2019 motion pictures is outlined. The general method of attack is \nindicated. There follows a description of the several pieces of apparatus \nwhich comprise the theatre equipment, including a discussion of some of \ntheir salient features and of the part each plays in the sound projector \nsystem.\n\nN order to reproduce in a theater the pictorial record of events \naccompanied by the sound associated with those events, it is, of\n\ncourse, necessary to add equipment to that installed to produce only \nthe silent motion picture. It is the purpose of this paper to outline \nand discuss briefly the major items of such equipment as developed \nby Bell System engineers.\n\nIn the design of sound equipment a primary requisite is that there \nshall be freedom from distortion. Distortion may be of the sort \nwhich is independent of load and is evident in that the intensity of \nsome portion or portions of the sound spectrum is increased or de- \ncreased in comparison with the rest; or there may be the distortion \nwhich is a function of the level at which the device is operated and is \ncharacterized by the reduction of a pure tone into fundamental and \none or more harmonics. \u2018This latter condition is most often the conse- \nquence of operating a vacuum tube amplifier above its proper energy \nhandling capacity.\n\nIt is the resonances of vibrating strings or reeds or air columns or \nvocal cords that give us the music we record but we are careful that a \nminimum of the resonances of the recording system itself shall go \ninto the record, and that any resonances of the reproducing system \nshall not appear in the output of the sound projectors. Aside from \nthe effects of overloading, the prevention of distortion is largely a \nmatter of getting away from resonance phenomena since it is the \ncharacteristic of the resonant system to respond with disproportionate \namplitude to stimuli in the region of its own natural period. The \nwhole story of the passage from sound energy through the various \nrecording and reproducing devices back to sound energy again is \none of contest with this fundamental physical phenomenon.\n\n1 Presented before Society of Motion Picture Engineers at Lake Placid, New \nYork, September, 1928.\n\nThere are in general two things one can do to avoid the harmful \neffects of resonance in vibration transmitting or transforming appa- \nratus: (1) the period of resonance of each piece of equipment can be \nmoved outside the range of frequencies one wishes to transmit, at \nthe same time providing damping means to minimize free vibrations; \n(2) the distortion produced by resonance in one piece of apparatus \ncan be compensated for or equalized by similar and opposite distortion\n\nin some associated apparatus. The first is not always easy of practical \naccomplishment in any particular device and generally results in an \ninstrument of very low inherent efficiency; the second usually involves \nloss of energy. In both cases increased amplification is required. \nThe sound record comes to the theater either as a wavy groove in a \ncomposition disc or as a striated track of varying density at one \nside of the picture film. It is the function of the apparatus being \nconsidered to derive from these records an electric current in which\n\nall the variations in pitch and loudness are accurately represented, to \nsuitably amplify this current, to effect its conversion into sounds \napproximating those from which the records were made, and to so \ndirect those sounds as to reasonably create the illusion that sound \nand picture are cognate.\n\nThe disc records do not differ essentially from those used in the \nordinary phonograph except that they are considerably larger and \nrun at a much slower speed so that a single record will play throughout \nan entire reel. The reproducer used is in some ways similar to that\n\nFig. 2\u2014Diagram of motion picture projector equipped for reproducing sound from \nfilm.\n\nused on the acoustic phonograph, the needle holder being connected \nto a clamped diaphragm. This diaphragm is of highly tempered \nspring steel and to it there is fastened an armature made of a special \nhigh permeability alloy and so arranged that as the diaphragm vibrates \nthe flux in the air-gap of a permanent magnet varies correspondingly, \nthereby inducing in appropriately placed coils currents which are the \nelectric representation of the wavy groove which the needle travels. \nThis reproducer is shown in Fig. 1. Although the energy delivered \nby this instrument is comparatively low it has a very uniform response \nover a wide frequency range. This result is largely brought about\n\nby moving all resonances out of the working range and by filling the \nmagnet chamber back of the diaphragm with a heavy damping oil. \nThe film used with the disc record, called a synchronized film, differs \nfrom ordinary film only in that one frame at the beginning is specially \nmarked to give the starting point.\n\nThe film sound record, as has been said, consists of a track of \nvarying density running along one side of the picture. This sound \ntrack is 1/10\u2019 wide. Differences or changes in intensity of sound \nare represented by differences in the density of the record, while pitch\n\nCONDENSER \nPHOTO- \nELECTRIC VACUUM \n2 10 TUBE \nMEGOHMS MEGOHMS \nT POLARIZING \n| BATTERY _\n\nis represented by the number of changes from dark to light and back \nagain in a given length of track. This sound record is converted \ninto a corresponding electric current by arranging that a narrow high \nintensity beam of light shall pass through it and fall upon a photo- \nelectric cell. The arrangement is shown in Fig. 2. The light from\n\nthe bright filament of the exciting lamp is focused as a very narrow \nline upon the film by passing through a system of lenses and an \naperture plate. The lamp filament is focused upon a slit of dimensions \n0015\u201d x 3/16\". The image of this slit is then brought to focus upon \nthe film as a .001\u201d line whose length has been reduced in passing \nthrough the aperture plate to .080\u2019.. This reduction in length allows \n010\u201d on either side for variations in position of the .100\u201d sound track. \nThe position and focus of the lens tube are fixed, but the carriage of\n\nthe exciting lamp is movable so that when replacing lamps the filament \nmay be properly brought on focus.\n\nA photoelectric cell of the type used is shown in Fig. 3. The \ncharacteristic of this device is that when it is polarized by a proper \nvoltage and is used within proper limits the current through it is pro- \nportional to the incident light. The circuit is shown in Fig. 4. It is \nto be noted that the polarizing voltage is supplied to the photo cell \nthrough a very high resistance and there is, therefore, obtained across\n\nthis resistance a voltage which is proportional to the light falling upon \nthe cell and accordingly bears a direct relation to the varying density \nof the sound track interposed between the exciting lamp and the \ncell.\n\nThe photo cell circuit is inherently one of high impedance. In \nsuch a circuit there are two matters which require attention. (1) \nLocal interference\u2014\u2018\u2018static,\u201d to use the radio expression, is most \nreadily picked up, and at this point where the energy level is low, \u00e9\n\nmay be appreciable in comparison with the sound currents themselves; \n(2) also the shunting effect of capacity between the electrical con- \nductors becomes noticeable, particularly at the higher frequencies.\n\nMACHINE #1 \nHORNS \nNS \nLa oR a SWITCHING PANEL \nBeOEre THEATRE \nTURNTABLE To \nOTHER \nUPPER \nTRANSVOX HORN i \nFADER als ae noun CONTROL ae HORN \nAMPLIFIER | JAMPLIFIER [] AMPLIFIER ape \nBOX\n\nHence a vacuum tube amplifier, which serves both to increase the \nenergy and to make that energy available across a low impedance \ncircuit, is closely associated with the cell upon the projector itself.\n\nThe cell and amplifier are enclosed in a heavy metal box or shield \nwhich is made fast to the frame of the projector and the projector \nitself is carefully grounded. This amplifier is shown in Fig. 5. It is \ndesigned to bring the level of the electric counterpart of the film sound\n\nrecord up substantially to the same energy value as that obtained from \nthe magnet coils of the disc reproducer. The filaments are heated \nfrom a 12-volt storage battery. Small dry batteries supply its plate\n\nFig. 9\u2014Amplifier Panel. \ncurrent and also the polarizing potential for the photo cell. These \nbatteries and the battery leads are shielded.\n\nVibration of a vacuum tube often produces sufficient motion of its \nelements with respect to each other to effect changes in the stream\n\nof electrons which appear when sufficiently amplified as noise from a \nloud speaker. In spite of all precautions there is a certain amount of \nvibration of the projector when in operation and it has therefore been \nnecessary to design a rather elaborate shock-proof mounting for the \nphoto cell amplifier.\n\nIt is evident from the relative location of apparatus as shown in \nFig. 2 that it is not feasible to print the film sound record directly \nbeside the picture to which it applies. Asa matter of fact, there is a \nspacing of 143\u2019 between picture and corresponding sound record and \na certain amount of slack is allowed between the sprocket which \ncarries the picture with an intermittent motion before the picture \nprojection lens and the sprocket which must carry the sound record\n\nwith a uniform motion in front of the photoelectric cell. In this \nconnection it is noteworthy that special precautions are necessary in \norder to prevent vibrations and speed fluctuations due to either \nvarying supply voltage or varying load from affecting the uniformity \nof rotation of this sound sprocket. This is taken care of by the very \neffective means of automatically controlling the speed of the driving \nmotor and by means of a mechanical device interposed between the \nsound sprocket and the rest of the moving equipment of the projector \nwhich effectively opposes the transmission of any abrupt change of \nspeed to this sprocket.\n\nThe control box which contains the apparatus for governing the \nspeed of the driving motor is arranged to hold the record speed the\n\nsame as that at which the records are made, i.e. 90 feet per minute \nin the case of synchronized sound and picture productions. By \nthrowing a switch the automatic feature may be cut out and the speed \nof the machine may then be manually controlled by the operator.\n\nThis completes the apparatus associated directly with the projector. \nThe general arrangement of the latest type of projection machine, \nequipped with a Simplex head, is shown on Fig. 6. Incidentally this \nprojector is also arranged to be fitted with the Powers or the \nMotiograph head. Fig. 7 shows a typical layout of a sound projector \nsystem as installed for use with talking motion pictures,\n\nAs in ordinary pictures, in order to run a continuous program, it is \nnecessary to use two projectors alternately. As the picture from one \nmachine is faded imperceptibly into that on the other so the sound \nrecord may be faded from one machine to the other without the \naudience being aware that a change has been made. At the end of \neach record or sound film the music overlaps the beginning of the next \nand a device called a fader is employed in making the transition. All \nthat is necessary is to turn the fader knob when the incoming machine \nis started. \u2018This fader is in fact a double potentiometer. In the upper \nor normal operating range the change in volume in moving from one \nstep to the next is hardly more than perceptible whereas in the lower \nrange used only in fading the steps are large and the volume decreases \nto zero on one machine and builds up on the other very rapidly. By \nchoosing the proper step in the upper range one can obtain any \nvolume of sound desired within reasonable limits and thereby equalize \nthe level obtained from different sound records. The fader is ordi- \nnarily installed with one or more auxiliary dials and handles inter- \nconnected so that it may be operated from any projector position. \nIn connection with the fader there is provided a switch for changing \nfrom the film to the disc input system and also a key for switching a \nspare projector in place of either of the regular machines. Fig. 8 \nshows a fader with one auxiliary position.\n\nFollowing the fader, we come to the main amplifier which raises \nthe energy of the feeble electric currents to a level adequate to supply \nthe loud speakers with sufficient volume to serve the particular \ntheater. Fig. 9 shows a typical amplifier panel. This combination is \ncapable of an energy amplification of about 100,000,000 times and is \nso designed that all frequencies in the range from 40 to 10,000 cycles \nare amplified practically equally. A potentiometer is provided on \nthe amplifier but while its handle is readily accessible it is ordinarily \nnot used after having once been set at the time of installation to give \nproper results in the particular theater. Necessary adjustments are\n\nmade on the fader. The amplifier shown consists of three units. \nThe first consists of three low power tubes in tandem, resistance \ncoupled, and requiring a 12-volt battery delivering 1/4 ampere to heat \ntheir filaments. The second consists of a single stage of two medium \npower tubes, connected in push-pull arrangement with filaments\n\nheated by low voltage alternating current. \u2018Two similar tubes in this \nunit operate as a full wave rectifier and supply rectified alternating \ncurrent for the plate circuits of the amplifier tubes of both the first \nand second units. The third unit has a single stage of high power \npush-pull amplifier tubes and push-pull rectifier tubes and also operates \nentirely on alternating current.\n\nThese three types are capable of arrangement into combinations to \nmeet the particular need. For small theaters only No. 1 and No. 2 \nare required. In the larger houses the high power unit No. 3 is added, \nwhile to meet exceptional conditions two or more of the high power \namplifiers may be operated in parallel from the output of No. 2.\n\nFollowing the amplifier there is an output control panel. This \nconsists of an auto-transformer having a large number of taps, the \ntaps being multipled to a number of dial switches, to which the sound \nprojectors or loud speakers are connected. By means of this panel, \nit is possible to match the impedance of the amplifier output to the \ndesired number of horns in order to obtain the most efficient use of \nthe power available and also to adjust the relative volume of the \nindividual horns.\n\nThe ordinary theater installation employs four horns, two mounted \nat the line of the stage and pointed upward toward the balconies and \ntwo mounted at the upper edge or above the screen and pointed \ndownward. This combination has been found to give good distri- \nbution throughout the house.\n\nThe loud speaker unit used with the horns in theater equipments \nis essentially that recently described by Messrs. Wente and Thuras.? \nAs brought out in this article, this unit shows extremely high efficiency; \nabout 30 per cent of the electrical power supplied is radiated in the \nform of sound. This is important since the higher the loud speaker \nefficiency, the smaller the power capacity of the amplifier needed in \nthe system. The frequency-response characteristic of a typical re- \nceiver and horn is given in Fig. 10. An individual horn may be \nequipped with two, four or nine loud speakers by using the throats \nshown in Fig. 11. The power capacity for continued safe operation \nof the horn with one, four and nine throats is approximately 5, 20 \nand 45 watts, respectively (electrical input). The number of horns \nused is dependent upon the particular installation and is related to \nthe directive characteristic of the horn. If it is necessary to disperse \nthe sound over a large angle, more horns are needed than when it is \ndesired to concentrate over a comparatively small angle. This \ndirective characteristic of the horn is important in talking motion \npictures as it is responsible for the illusion of the sound coming directly\n\nfrom the mouth of the horn; that is, from the screen. If the horn \nis replaced by a loud speaker of otherwise identical characteristics \nbut which radiates its sound over a very wide angle, there is a tendency \nfor.the sound to appear to come from a point some distance back of \nthe screen, thus tending to destroy the illusion.\n\nThe power supply equipment has been fairly completely covered in \ndiscussing various parts of the system. Under ordinary conditions \nthe requisite power is obtained from the electric mains in the theater \nexcept for the 12-volt battery required for some of the vacuum tube \nfilaments and for the electromagnets in the loud speakers and the \ndry cells used with the photo cell and photo cell amplifier. Where \n110-volt D.C. only is available there is a projector-driving equipment \nwhich operates on this voltage, but a D.C. motor driving a 60-cycle \ngenerator is required for supplying the amplifiers. Where 110-volt \nA.C. is available, it is only necessary to connect the projector motor \nand the amplifiers to this supply.\n\nLarge Power Capacity for Horn-type Loud Speakers,\u2019\u2019 by E. C. Wente and A. L. \nThuras.\n\nWESTERN ELECTRIC SOUND PROJECTING SYSTEMS \nFOR USE IN MOTION PICTURE THEATRES\n\nTo obtain a clear picture of the whole problem in all its com- \nplexities, it will perhaps be best to consider an installation from its \nvery inception and follow it through to completion. This treatment, \nwhile including mention of details which may possibly appear \nsuperfluous, will nevertheless present a fair idea of the scope of the \nwork involved.\n\nThe Electrical Research Products, Inc. supplies to theatres \nthe equipment of the Western Electric Sound Projector Systems, \nsupervises its installation, instructs the local personnel in its opera- \ntion, and provides inspection service during its subsequent use.\n\nOne of the most important factors in preparing a theatre for \nsound projection is the initial engineering survey, for upon this \nsurvey rests the decision of the type of system to be installed and \nthe determination of requirements for apparatus layout and of \npossible special treatment. Each theatre is an individual problem \nand must be considered as such.\n\nThe survey engineer gathers pertinent information and pre- \npares his recommendations. And while his recommendations are \ngiven due consideration, a special staff at the home office studies \nthe survey and renders final decision. Survey information must \ntherefore be complete. |\n\nIn the projection room the engineer must investigate the power \nsupply not only as to voltage and kind but to determine if there is \nample capacity to carry the load to be added, and if the regulation \nis suitable. The type and condition of the projection machines\n\n* Theatre Systems Engineer, Electrical Research Products, Inc., New \nYork, Nes\n\nmust be noted and the angle of projection learned. Is the space \nbetween machines ample for the attachment of the sound apparatus \nand for convenient operation? Is there an acceptable location \nin the room for the amplifying equipment and associated controls? \nIt is sometimes necessary, particularly in the older theatres, to \nrequire rearrangement of the projection equipment or possibly an \nenlargement of the projection room. Such changes are naturally \nwelcomed by the operating staff. Space, conveniently near but \nseparately enclosed, must also be found to house the storage \nbatteries.\n\nThe auditorium presents an item for careful study, for upon \nits characteristics is based the size of the system to be installed. \nThe types of systems available, varying in power output, meet \nconditions imposed by theatres from the largest to the smallest and \nare so arranged and selected that, while sufficient sound volume \nis at hand in each case, no theatre is overburdened with excess.\n\nAs blue prints of the auditorium are not always available it is \nfrequently necessary for the engineer to take the dimensions him- \nself, for the cubical volume is an important factor. Seating capacity \nand the distribution of seats also bear influence. The general \nacoustics of the house must be considered. What degree of rever- \nberation is present, bearing in mind the absorptive effect of the \naudience. Are there areas of echo or interference or of otherwise \nbad hearing? Are internal or external disturbing noises a factor? \nA final and specific study of acoustic properties can best be made at \nthe time of system testing upon completion of the installation, but \ngeneral effects can be and are observed during the survey process.\n\nThe stage must receive consideration as to size and to layout \nwith reference to location of the horns and of the special screen \nused with the system. If the theatre policy covers the showing of \nmoving pictures only, the horns may be located permanently on a \nsuitable structure. If vaudeville or stage presentations are in- \ncluded, however, they are generally placed in movable horn towers \nor else flown from battens.\n\nInformation sufficient to permit a knowledge of the conduit \nruns for the complete installation is also obtained at the time of \nsurvey.\n\nThe data collected, augmented by sketches prepared by \nthe engineer, in conjunction with blueprints of the building, if \navailable, thus provide the home office survey staff with a complete\n\npicture of the theatre. The type of sound projector system is ac- \ncordingly selected and the dates of shipment of equipment and \ninstallations are placed on schedule. If, perchance, architectural \nchanges in the theatre are required the theatre management is so \nnotified in the anticipation that alterations will be completed \nprior to the actual introduction of apparatus.\n\nAt the outset of an installation, it is important for the engineer \nto establish satisfactory personal contacts with the local people \nand organizations that will be related to the work. This is neces- \nsary that the work may proceed smoothly and without delay, as \nan opening date has by this time usually been fixed. While sufficient \ntime for the installation, testing and rehearsals is allowed, obviously \nno large safety factor can be tolerated. The engineer first meets \nthe. house manager, and subsequently the projectionists, the house \nelectrican and stage hands, and familiarizes himself with house \nconditions such as policy, house hours, labor conditions, best time \nfor work, etc. It is necessary for him to find a competent electrical \ncontractor, one, if possible, who already has knowledge of the \ntheatre. 4\n\nThe installation proceeds in accordance with a logical plan. \nIt is usual first to attach the driving equipment to the motion \npicture projector, for, while the projector mechanism itself remains \npractically intact, its driving motor is replaced by one with elec- \ntrically regulated speed control. A film speed of 90 feet per minute \nis assured for sound picture work, but variable speeds may be \nobtained, if desired, for silent film. As these motors require four \nor five seconds to pick up speed, it is advantageous to have the pro- \njectionists become accustomed to such features in order to be able \nto give full attention to the handling of film or disc record when \nthe time arrives.\n\nThe battery equipment is next installed with its switching \npanel and charging apparatus. This permits an investigation of \nthe condition of the batteries and ample time for charging if \nrequired. Furthermore, the projectionists may study this portion \nof the apparatus at some leisure and have explained to them proper \nschedules of charging. If a motor generator is required, it is set up \nat this time.\n\nThe amplifiers and associated equipment are placed upon the \nrelay racks and the fader with its auxiliary positions is mounted \nat the front of the projection room wall. The position of the monitor\n\nhorn is carefully selected so that the operating personnel will at \nall times be able to hear clearly, for it is by following the program \nthrough this horn that fader change-overs are made. .\n\nWhile the apparatus is being placed in position the conduit \ninstallation keeps apace. It should now be possible to connect the \nstorage batteries and A.C. supply to the amplifying equipment \nand check its functioning. The fader can be electrically joined to the \nfilm and disc reproducers associated with the production machines. \nA film-dise transfer panel is mounted above the master fader \nposition to permit either the film or disc reproducers to be used at \nwill, simply by momentarily depressing the proper push button. A \nsignal light gives visible evidence as to which circuit is set up. \nInterconnections between physically disassociated groups of equip- \nment are made through a junction box located in the projection \nroom. This method not only facilitates installation but provides a \nready means for circuit checking and trouble finding.\n\nWhile equipment and installation methods are constantly \nbeing simplified, it may be noted that a rather marked advance in \nthis direction is to be introduced within a few months. At the \npresent time the driving motor and disc turntable have their \nseparate mounting pedestals. The film reproducer and associated \namplifier are supplied as separate units to be assembled as an \nappendage. The film-disc transfer panel is an apparatus separately \nmounted on the wall. The new plan provides that these individual \npieces shall join into a single united whole. A pedestal is to be \nsupplied which will have space for placing the driving motor and \nmounting the dise turntable. The film reproducer and amplifier \nwill form a definite part of the pedestal. A film-dise transfer will \nbe associated with each unit which will permit the projectionist \nin making his machine ready for the succeeding selection to prepare \ncompletely for the change-over. As any equalizers which may be \nnecessary will also be included in the pedestal, special wiring or \nswitching of such units will be eliminated. A mechanical brake will \nprovide ready stopping of the projector. A model of this complete \ndevice has been installed in the booth in this auditorium and is \nbeing used in the projection of sound pictures.\n\nThe installation engineer next turns his attention to the stage. \nThe horns are placed in the mountings which have been constructed \nby now and are located in positions which experience has taught \nthe engineer may be roughly correct for this type of house. After\n\nthe loud speaking receivers are attached to the horns and con- \nnected in circuit, a cursory test follows to assure that no major \nerrors are present. This is the first opportunity to produce sound \nthrough the complete system. Any imperfections that are im- \nmediately obvious are corrected. Comparative tests are then made \nbetween the sound reproducers on the projectors equipped, so that \nvariations in volume can be compensated for. Each receiver and \nhorn is heard separately, and then in unison to check poling. \nReceivers incorrectly poled give a most unpleasant effect of pressure \nin the ears of the listener if he is at the junction of their sound \npaths.\n\nBefore proceeding further it is desirable to have in place \nthe screen used with the sound system. An obvious requisite of \ntalking pictures is that the sound shall seem to emanate directly \nfrom the action shown on the screen. If the sound source is placed \nimmedately behind the area of action in the picture, the illusion is \noptimum. While it is true that the human mind tends to adjust \nitself to disregard imperfections in this particular, extensive study \nhas shown that, where the sound originates from locations other \nthan at the visual source of action, an impression of unnaturalness \nis often present. The auditor is probably not aware of the cause of \nhis uneasiness, but he is unconsciously attempting to convince \nhimself that he is hearing aright. Imperfections of location become \nless annoying, of course, the farther one is away from the screen. \nTo present the best illusion a special semi-porous screen is used \nwhich permits sound from the horns located directly behind it to \npass through readily. With the present type in use, a small per- \ncentage of light is lost from the picture.\n\nWith the screen in place and the horns temporarily set, the \nacoustic tests can be made and the final check given to the system.\n\nAcoustic conditions of theatres are responsible for widely \ndifferent results obtained. Some houses give a mellow effect, \nothers make reproduced sounds appear metallic. The shape of \nthe interior has less effect in general than has the nature of the sur- \nface of the walls, ceiling and floor. If these surfaces are hard and \nsmooth like ordinary plaster, concrete, polished wood or glass, \nthey will reflect most of the sound striking them. Different fre-\u2014 \nquencies may be reflected in varying proportion. Where surfaces are \nsoft, thick materials such as heavy drapes, carpets, upholstery, or \nwhere the audience covers a relatively large space, a goodly portion\n\nof the sound is absorbed. Pilasters, moldings, cornices and such \nbroken surfaces disperse sound waves and tend to damp them out \nquickly.\n\nThe hard or reverberant type of theatre is especially annoying \nfor speech selections since the duration of audibility is so great \nthat a sound will remain to overlap the sound succeeding it. This \ngives an effect of poor articulation. For music reproduction, how- \never, a reasonable amount of reverberation is acceptable as it \ntends to give the effect of fullness and roundness. Another varia- \nble factor which further complicates the situation is the changing \namount of audience which in itself is an excellent absorbent. \nTheatres which may be entirely too reverberant empty will some- \ntimes give pleasing results when filled. Unfortunately, it is not \npossible to have a completely filled house at every performance. It \nwill therfore be seen that at best, a compromise must be tolerated.\n\nReverberation depends upon the rate of decay of sound and is \nmeasured by its inverse, the duration of audibility of a sound of \nknown initial intensity. The introduction of highly absorbent \nmaterial into a room will have greater absorptive effect on the higher \nfrequencies. If the thickness of the material is increased considera- \nbly the absorption of the lower tones will be increased while that \nof the higher frequencies will remain practically intact. Very thin \nfabrics such as cheese cloth or bunting absorb only extremely \nhigh frequencies.\n\nA highly damped auditorium will cause sound to appear to \nJack brilliance and definition and will have deadness which in \nextreme cases may be almost depressing. Greater electrical power \nwill be required to fill the house with comfortable sound volume, \nbut the effect of the audience will be relatively less than it is in a \nreverberant house.\n\nEchoes, generally speaking, are due to too great differences \nin the lengths of the sound paths of direct and reflected sound which \nreach an auditor in a given position in a room. An interval of about \n1/16 second is noticeable. Proper location of the source of sound, \nor placing draping material over the offending reflecting surface \nwill ordinarily correct echo.\n\nInterference is usually produced by the crossing of many \ntrains of sound waves reflected from various parts of the room.\n\nIt results in distortion and is overcome by proper placing of absor- \nbent.\n\nResonance occurs with a surface consisting of some thin, hard \nmaterial which is free to vibrate as a diaphragm at its own natural \nfrequency. Thus sound is reinforced. Distortion will probably \nresult, since only that portion of complex sound whose frequency \nis near the natural period of the resonator is affected.\n\nArmed with a knowledge of these general acoustic considera- \ntions and bearing in mind the three prime requisites to good \nhearing, i.e., that sound should be sufficiently but not unnaturally \nloud for all auditors, that successive sounds be clear and distinct, \nand that the components of complex sound should retain their \nrelative intensities, the engineer makes his acoustic adjustments. \nHe usually tests first for sound distribution and arranges his horns \nso that in so far as possible all portions of the house receive a like \namount. Good distribution can normally be obtained by proper \nflaring and tilting of the horns. Any acoustic peculiarities are \nnext investigated and where possible corrected. Slight variation \nin the location of the horns may result in quite marked differences \nin the results obtained, for which reason this test may involve \nmany hours of labor. It may be necessary to convince the manage- \nment that acoustic treatment is essential if good reproduction is to \nbe obtained.\n\nOne standard practice which is followed in all cases requires \nthat the horns and the back of the screen not occupied by the \nhorns be completely enveloped by absorbent drapes. If the sheet is \nset back from the proscenium arch, it is also usual to hang drapes \nin shadow box effect. This treatment eliminates back stage reflec- \ntions of sound.\n\nAfter the horn locations have been definitely fixed and the \nequipment given a final check, the engineer proceeds to calibrate \nthe theatre. He has available several test records embracing a \nrather wide variety of character of selection to cover various \ntypes of entertainment which will be shown in future in the theatre. \nEach record has been marked with a proper fader setting deter- \nmined in a standard theatre. By playing these records at the \nmarked value and adjusting the main volume control potentio- \nmeter on the amplifying equipment until the proper effects are \nobtained, it should be possible from then on to obtain good results \nby using normal fader volumes for any succeeding selections. In \nother words, this potentiometer setting which then remains fixed \ncompensates for the varying sizes of theatres and permits like effects\n\nof reproduction to be obtained universally. In addition, various \ntypes of selection require that varying degrees of volume be fed to \nthe different horns. From the test records the engineer is able \ndefinitely to select what these volumes should be and so establish \nthe horn setting values which are normally required. When these \npotentiometer values and horn settings are determined the theatre \nis considered calibrated. Such routine tends to standardize sound \nreproduction in so far as that is possible in theatres using sound \nprojector systems.\n\nWhile the installation has been progressing the operating \npersonnel have been instructed in regard to the functioning of the \nequipment and the proper method of handling it. At the outset \neach man has been given a copy of a complete operating instruction \nbulletin for his personal keeping for study and reference. By the \ntime of the completion of the work, and as a result of the running \nof the equipment during the acoustic tests the projectionist should \nnow be able to start up and adjust the amplifying equipment pre- \nparatory to running the show. He should be able to thread the pro- \njector with film through the sound gate compartment with surety \nand for film reproduction adjust his sound reproducer lamp. For \ndisc reproduction he should know how to place the starting mark \nof the film in its aperture and fix the needle of the reproducer on \nthe starting mark of the record.\n\nThe rehearsal of the opening program is an important event. \nIt not only enables the projectionists to demonstrate their ability \nin handling the equipment under real operating conditions, but \nalso gives them experience in actual change-overs of picture and \nfader. The engineer can acquaint the manager with the technique \nof running the show to obtain best results. While the physical \noperation of the apparatus is the responsibility of the projectionist, \nthe manager in the last analysis is responsible for the general effects \nin the theatre proper. Either he or someone assigned to the work \nshould be present in the audience during the showing of sound \npictures. Even though rehearsals of programs have indicated proper \noperating points of the various selections certain conditions arise \nwhich may require changes in volume. The observer is able to be \nin touch with the projection room for transmitting his wishes by \nmeans of a buzzer signal system and telephone circuit, which are \nprovided. It is, of course, essential that this observer not only be \nentirely familiar with the results which the system may be expected\n\nto give, but should also know how to originate requests for obtain- \ning the changes which he may want. Upon the ability and conscien- \ntious effort of this observer depends in a large measure the success \nof the sound picture programs.\n\nThe installation personnel remain at the theatre until it is \nevident that the equipment can be handled in a commendable \nmanner. At such time the installation is transferred to the service \norganization of the Electrical Research Products, Ine.\n\nThe function of the Service Department is evidenced bp its \nname. The service engineer in whose district the theatre is located \nis available for emergency calls at all times. In addition, he visits \nthe theatre periodically to check the equipment and recommend \nwhat minor adjustments may be necessary. He listens to the selec- \ntions on the program and makes any suggestion to the manager \nwhich may occur to him. In this manner all of the theatre systems \nare kept under frequent engineering supervision in an earnest \nendeavor to maintain the class of service which they may be \nexpected to give.", "title": "Bell Telephone Laboratories Incorporated: Sound Pictures", "trim_reasons": [], "year": 1929}
{"archive_ref": "sim_att-technical-journal_1926-01_5_1", "canonical_url": "https://archive.org/details/sim_att-technical-journal_1926-01_5_1", "char_count": 375898, "collection": "archive-org-bell-labs", "doc_id": 301, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc301", "record_count": 471, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/sim_att-technical-journal_1926-01_5_1", "split": "test", "text": "Correction of Data for Errors of Measurement\u2014W. A. = \nTheory of the Howling Telephone with Experimental \u2018a \nConfirmation\u2014Harvey Fletcher ...... 27 \nElectric Circuit Theory and the Operational Calculus S \nModel, Part K. Darrow . .... 96\n\nJ. J. Carty Bancroft Gherardi FP. B. Jewett \nE. B. Craft _ L. F. Morehouse O. B. Blackwell \nE. H. Colpitts H. P. Charlesworth H. D. Arnold \nR. W. King, Editor J. O. Perrine, Asst. Editor\n\nCompany, the Bell Telephone Laboratories, Inc., and the Associated \nCompanies of the Bell System\n\nAddress all correspondence to the Editor \nInformation Department \nAMERICAN TELEPHONE AND TELEGRAPH COMPANY \n195 BROADWAY, NEW YORK, N. Y. \n50c. Per Copy i oe Copyright, 1925 $1.50 Per Year\n\nPrevious to 1907 there were maintained in the Bell System \nthree laboratories and departments of development, research and \nexperiment,\u2014one by the American Telephone and Telegraph \nCompany at Boston, one by the Western Electric Co at \nChicago and one by the Western Electric Company at New York.\n\nIn 1907, in the interest of economy and efficiency, these were \nconsolidated so far as laboratory and experimental work were \nconcerned and the Bell System Laboratory at Bethune and West \nStreets, New York, was established. This was incorporated as \nthe Bell Telephone Laboratories, Inc., January first, 1925. The \nexpense of operation is divided between the American Telephone \nand Telegraph Company and the Western Electric Company ac- \ncording to the nature of the work done.\n\nIn the Bell System the American Telephone and Telegraph \nCompany undertakes, through constant association with the oper- \nating organizations, to formulate the requirements, present and \nfuture, of the Bell System. Out of these requirements come the\n\nroblems of the American a and Telegraph Company\u2019s \ntment of Development Research and the System lab- \noratory. After the problems have been eatisfactority ee \nthe Department of Development and Research adopts as s \nthe systems, equipment and apparatus thus produced which are \nthen specified for their proper uses in the Associated Companies \nthe Engineering Department of the American Telephone and \nelegraph Company. When different departments and companies \nare mentioned in this publication, so far as they are es of the \nBell System, they are parts of one working organization.\n\nN the brillant galaxy of investigators to whom we owe out \nknowledge of electrical science, Joseph Henry stands out as ot \nthe first magnitude; and for those who are associated with the Bell \nSystem, the present is a most appropriate time to review his researches \nwhich had an important guiding influence on the development ot \nelectrical communication. The present vear marks the fiftieth since \nthe invention of the telephone by Alexander Graham Bell, and among \nthe scientists with whom Bell conferred at that time, he gave a place \nof honor to Henry. Ina letter to his parents written in March, IS75, \nwhile he was busy in an effort to perfect the harmonic telegraph, \nand before he had turned his attention to the telephone, Bell wrote:\n\n\u201cNow to resume telegraphy. When | was in Washington, | had \na letter of introduction to Professor Henry, who is the Pyndall ot \nAmerica. [| had found on inquiry at the Institute of Technology, \nthat some of the points I had discovered in relation to the applica- \ntion of acoustics to telegraphy had been previously discovered by \nhim. I thought I would, therefore, explain all the experiments, and \nascertain what was new and what was old. He listened with an \nunmoved countenance, but with evident interest to all, but when | \nrelated an experiment that at first sight seems unimportant, [Twas \nstartled at the sudden interest manifested.\n\n\u201cT told him that on passing an intermittent current of electricity \nthrough an empty helix of insulated copper wire, a noise could be \nheard proceeding from the coil, similar to that heard from the tele \nphone. He started up, said, \u2018Is that so? Will you allow me, Mr. \nBell, to repeat your experiments, and publish them to the world \nthrough the Smithsonian Institute, of course, giving vou the credit \nof the discoveries?\u2019\n\n\u201cT said it would give me extreme pleasure, and added that T had \napparatus in Washington, and could show him the experiments mysell \nat any time.\n\nstrument working and he sat at a table for a long time with the empty \ncoil of wire against his ear listening to the sound. IT felt so much \nencouraged by his interest that I determined to ask his advice about \nthe apparatus I have designed for the transmission of the human \nvoice by telegraph. | explained the idea and said, \u2018What would \nvou advise me to do, publish it and let others work it out, or attempt \nto solve the problem myself?\u2019 He said he thought it was \u2018the germ \nof a great invention,\u2019 and advised me to work at it myself instead of \npublishing. IT said that I recognized the fact that there were mechan- \nical ditficulties in the way that rendered the plan impracticable at \nthe present time. IT added that IT felt that I had not the electrical \nknowledge necessary to overcome the difficulties. His laconic answer \nwas, \u201cGET IT.\u2019\n\n\u201cT cannot tell you how much these two words have encouraged \nme. Such a chimerical idea as telegraphing vocal sounds would \nindeed to most minds seem scarcely feasible enough to spend time in \nworking over. I believe, however, that it is feasible, and that I \nhave got the cue to the solution of the problem.\n\n\u201cProfessor Henry seemed to be much interested in what IT told \nhim, and cross-questioned me about my past life, and specially wanted\n\nJoseph Henry was born in Albany, New York, in 1799, and coming \nto full maturity of mind at the beginning of a century which will \nprobably never be surpassed for fruitful research in the field of elec- \ntricity, he demonstrated, at the very outset of his career, his right \nto stand for all time with the foremost investigators in this depart- \nment of natural science. Henry was, moreover, a many-sided man. \nHis distinguished career leads into many fields and before reviewing \nhis researches on clectro-magnetism we may note briefly the very \ndiversified and yet important character of his other work.\n\nDuring the latter half of his life, official duties as the director of \nthe Smithsonian Institution consumed an ever increasing portion of \nhis time, but he still found opportunity to prosecute many original \ninquiries,\u2014for example, into the application of acoustics to building, \ninto the best construction and arrangement of lecture rooms, and into \nthe strength of various building materials. As one of his first ad- \nministrative acts, he organized a widespread corps of observers for \nsimultaneous weather and meteorological reports by means of the \ntelegraph which was vet in its infancy. He was the first to have the \ndaily atmospheric conditions indicated upon a map of the country \nand to utilize this information in making weather forecasts.\n\nHe was an active and long-standing member of the Lighthouse \nBoard of this country and his diligent investigations into the efficiency \nof various illuminants and the best conditions for their uss\n\nimproved the beacons which dotted our coasts. During the dark \ndays of the Civil War, Henry clearly saw the tremendous advantage \nto be derived from a mobilization of the nation\u2019s scientific men for \ncooperative service. His vision, backed by his tremendous energy \nand ability, resulted in the formation of the National Academy of \nSciences, under a Congressional charter signed by Abraham Lincoln\n\nismi inspired by Henry, there re-occurred, in 1916, under presidential \nproclamation, a mobilization of the nation\u2019s scientific and technical \nmen.\n\nknown, his researches are of the most enduring character, and for all \ntime must enter intimately into the lives of all civilized mankind. \nHe was without peer among the American physicists of his time, and \nitis well attested by every record that he was a man of varied culture, \nof Jarge breadth and liberality of views, of generous impulses, of \ngreat gentleness and courtesy of manner, combined with equal firm- \nness of purpose and energy of action.\n\nLet us now turn to Henry's investigations of clectro-magnetism, \nwhich were among his earliest scientific undertakings. He began \nhis career in 1826 in New York State at the Albany Academy, where \nhe had only the apparatus he could construct with his own hands \nand, out of each vear, but a single month uninterrupted by other \nduties to devote to his researches. It was there-~independently of \nFaraday and on some fundamental points prior to him\u2014that Henry \ndiscovered the laws of current induction. At the same time he under- \ntook a study of the electromagnet which prepared the way for not \nonly the telephone and telegraph, but also for all types of dynamos \nand motors.\n\nThe electromagnet was discovered by Sturgeon in\u2019 England, but \nHenry's contributions to. our knowledge of it were so great that after \nhis work, a powerful instrument suitable for many uses replaced \nwhat had been a feeble toy. When he started his work on the electro- \nmagnet its design was not understood; when he had completed his \nwork he had developed a magnet, the design of which was understood \nand which could be adapted, according to the rules which he laid \ndown, to a multitude of purposes.\n\nWith reference to the making of electromagnets, Henry pointed \nout the improvements which resulted from insulating the conducting \nwire itself, instead of the rod to be magnetized, and by covering the \nwhole surface of the iron with a series of coils in close contact. This \nwas effected by insulating a long wire with silk thread, and winding \nthis around the rod of iron in close coils from one end to the other. \nThe same principle was extended by employing a still longer insulated \nwire, and winding several strata of this over the first, care being taken \nto insure the insulation between each stratum by a covering of silk \nribbon. By this arrangement the rod was surrounded by a compound \nhelix formed of a long wire of many turns instead of a single helix of a \nlew turns.\n\nThus Henry laid down the rules, which, in general, are followed \ntoday in the construction of commercial electromagnets; namely, \nthat the wire should be insulated, that it should be wound in layers, \nand that there should be several lavers, one above the other. He\n\nalso did another thing in his actual construction: he adopted what \nmay be called the spool construction, the placing of the windings on \nspools, and then the sliding of the spools on the core. Phat is a \nstandard method of building clectromagnets today.\n\nSoon after doing this work Henry built a magnet to be used at \nYale University, which was in its time a wonder and would even \ntoday be considered very powerful. He also built a series of magnets \nin which the emphasis was placed upon the lifting power in relation \nto the weight of the magnet and succeeded in designing one which, when \nenergized by a single small cell, could support 420 times its own weight.\n\nThe improvements which Henry made in magnets suggested to him \napplications of magnetic attraction to the production of mechanical \nmotion. He realized that electromagnets such as he built were easy \nto control, and believed that he could design a machine by which he \ncould get power from an electric current and this at a time when the \nonly source of current were primary batteries as the dynamo did not \nvet eXIst,\n\nHis electric motor was the first ever built to use electromagnets:! \nit Was extremely simple consisting of an electromagnet supported at \nits center by a pivot so that it could rock back and forth under the \nalternating pulls of two permanent magnets. To effect the reversal of \nmagnetization of the electromagnet and hence the alternation ot pulls, \nmercury cups were arranged so that wires would dip in them as the \nsuspended magnet rocked to and fro. These contacts were the proto- \ntype of the commutator which is found in every direct current motor \nand dynamo today. It is interesting to note the words in which \nHenry deseribed this invention. In Silliman\u2019s American Journal of \nScience for 1831 he wrote, \u201cT have lately succeeded in) producing \nmotion in a little machine by a power which [T believe has never before \nbeen applied mechanics -by magnetic attraction and repulsion. \nNot much importance, however, is attached to the invention since \nthe article in its present state can only be considered a philosophical \ntoy; although in the progress of discovery and invention it is not \nimpossible that the principle or some modification of it on a more \nextended scale may hereatter be applied to some useful purpose.\u201d\n\nossible applications of the principle there shown cannot fail to com- \nPI\n\n' Faraday has some years before shown that a wire carrying a current could be\n\ncaused to revolve continuously around the pole of a permanent magnet. Henry's \nadvance over this was considerable in that he materially increased the force causing \nmotion by employing the attraction between two magnets, one permanent and one \ngenerated by current. The motor using electromagnets throughout did not come\n\nmand our admiration. Of course, until the dynamo was invented \nat a later date, and a substantial electric current became available, \nthe motor could not be much more than he characterized it, \u201ca \nphilosophical toy.\u201d\n\nHenry also became interested in a determining whether an electro- \nmagnet could be operated from a distance so that the doing of some \nwork\u2014for example the ringing of a bell\u2014could be controlled from a \ndistant station. From his investigations directed to this end, Henry \nwas the first to appreciate that the effect of the resistance of long \nlengths of wire to the passage of electric current could be minimized \nby properly proportioning the battery and the magnet windings to the \nlength and resistance of the line wires.\n\nEfforts had been made by others prior to Henry's time to devise \nsuccessful electric telegraphs. They had failed, however, because \nthey did not know how to proportion their magnets and their batteries \nso as to operate over any substantial length of line. The literature of \nthat time contains a number of demonstrations of the impossibility \nof operating an electric telegraph, because scientists could arrange \ninstruments which would operate successfully when separated by \na few feet, or even one hundred feet, but they would not work at a \ndistance of thousands of feet because of the resistance of the long \nline wire.\n\nWhat Henry did was to determine the proportioning of the various \nparts of the system so as to secure operation. He found, when his \nmagnet was connected by a short wire to the battery, that the greatest \nmagnetizing effect was obtained by joining the cells of the battery in \nparallel, but that a series arrangement of the battery would give the \ngreatest pull if a long wire (a length of a mile or more was used in \nsome of his experiments) carried the current. He also obtained the \nbest operation over a short line when the magnet winding consisted \nof several distinct coils, all connected in multiple; and for operation \nover a long line he found it best either to connect these coils in series \nor to apply to the magnet a single long winding. Henry was there- \nfore the first to produce an electric telegraph, and more than that, \nthe transmission of electrical energy to a distance. That first tele- \ngraph paved the way for all the telegraph systems, all the ocean \ncable systems, and contained the principle of all telephone call bells.\n\nOne of Henry\u2019s greatest discoveries from the standpoint of electrical \nscience, but a discovery in which he must vield the first place to \nFaraday, is that of mutual induction\u2014the fact that a wire when \nmoving with respect to a magnetic field has an electromotive force \ngenerated in it. Although Henry made his discovery independently\n\nof Faraday, the latter was the first to make known his observations \nto the world, and it is no trifling index of Henry's character that he \nnever in any wavy intimated that he was entitled to share with Faraday \ncredit for the discovery.\n\nBecause Henry was anticipated in the publication of his observation \nof mutual induction, he does not appear to have left a verbal record \nof the steps of reasoning by which he was led to the discovery. How- \never, he does tell us what the arrangement of apparatus was and if \nwe bear in mind that he Was seeking a method ot generating an electri \ncurrent from a magnet--this magnet, in turn, being itself the product\n\nWriting of his original observations, Henry savs he \u2018succeeded in \nproducing electrical effects in the following manner, which differs\n\nfrom that emploved by Mr. Faraday and which appears to me to \ndevelop some new and Interesting Lacts;, A piece of copper wire, \nabout thirty feel long and covered with elastic varnish, Was closely \ncoiled around the middle of the soft iron armature of a galvanic \nmagnet... which, when excited will readily sustain between six \nand seven hundred pounds. The armature thus furnished with wire \nwas placed in its proper position across the ends of the magnet and \nfastened so that no motion could take place. The two projecting \nends of the helix were connected with a distant galvanometer by \nmeans of two copper wires each about forty feet long. This arrange- \nment being completed, TP stationed myself near the galvanometer and \ndirected an assistant at a given word to suddenly immerse the galvanic \nbattery attached to the magnet. At the instant of immersion. the \nnorth end of the needle was detlected 30\u00b0 to the west, indicating a \ncurrent of electricity from the helix surrounding the armature. The \neffect, however, appeared only as a single impulse, for the needle \nafter a few oscillations, resumed its former undisturbed position, \nalthough the action of the battery was still continued. [T was, how- \never, much surprised to see the needle suddenly detlected) from a \nstate of rest to about 20\u00b0 to the east, when the battery was suddenly \nwithdrawn from the acid, and again deflected to the west when it \nwas re-immersed. This operation was repeated many times ino suc- \ncession, and uniformly with the same result.\u201d\n\nIt was in this same paper that Henry announced his observation \nof the phenomenon of self-induction, a most important discovers \nand one for which he holds full credit for having first made it known \nto the world. He writes, \u201cI may, however, mention one fact which \n[ have not seen noticed in any work, and which appears to me. to \nbelong to the same class of phenomena as those before described; it \nis this: when a small battery is moderately excited by diluted acid. \nand its poles, which should be terminated by cups of mercury, are \nconnected by a copper wire not more than a foot in length, no spark \nis perceived when the connection is either formed or broken; but \nif a wire of thirty or forty feet long be used instead of the short wire, \nthough no spark will be perceptible when the connection is made, \nvet when it is broken by drawing one end of the wire from its cup of \nmercury, a vivid spark is produced... The effect appears \nsomewhat increased by coiling the wire into a helix.\u201d In a some- \nwhat later paper we find the following statement. \u201cA ribbon of \nsheet copper nearly an inch wide, and twenty-eight and a half feet \nlong, Was covered with silk, and rolled into a flat spiral similar to the \nform in which woolen binding is found in commerce. With this a\n\nvivid spark was produced, accompanied by a loud snap. \u2018The same \nribbon uncoiled gave a feeble spark.\u201d\n\nHenry tried many modifications of this experiment and in the end \ndrew the conclusion that the after-current he was observing was due \nto the inductive effect of the current in the wire upon itself, and that \nthis became particularly apparent when the wire was so coiled that \nits various turns lay close together. The discovery of mutual in- \nduction by Faraday and the discovery of self-induction by Henry \nconstitute two halves of a whole, and it is appropriate that to these \nmen should go equal recognition in the matter of having electrical units \nnames after them. Of the three units by which the properties of every \nelectric circuit are measured, the unit of capacity was named after \nFaraday, and unit of inductance after Henry; the third unit, that of \nresistance, recognizes the fundamental researches of Ohm.\n\nA few vears later, after having accepted the chair of physics at \nPrinceton University, Henry returned to the subject of induced cur- \nrents. In his earlier work he, like Faraday, had used the continuous \ncurrents which a voltaic battery generates. He now chose the cur- \nrents which flow when a Levden jar is discharged. To register the \ninductive effects of the fleeting currents of discharge Henry adopted \na device consisting of an unmagnetized needle placed in a small coil \nof wire. Through this coil the induced current had to flow. The \nuse of the needle as an indicator led Henry to an important observa- \ntion. He noticed that following a discharge, the direction of magnetiza- \ntion of the needle depended upon the distance across which the induc- \ntive effect had occurred. To account for this curious result, he ad- \nvanced the hypothesis later shown to be correct that the discharge \nis oscillatory.\n\nHere was the germ of a great discovery. The oscillatory character \nof the discharge is one of the fundamental and important properties \nof certain types of electric circuit. Henry did not have the facilities, \nhowever, for carrving his investigations in this field far enough to \nattract the attention of the scientific world. It was not until IN75. \nsome thirteen vears later, when Lord Kelvin was led independently \nby mathematical considerations to believe that the discharge 1s \noscillatory, that the significance of the phenomenon began to be \nunderstood.\n\nHenry's work contained the germ of vet another important dis- \ncovery. Some of his experiments on induction by Leyden jar dis- \ncharges involved the transmission of electric force without wires \nthrough distances as great as two hundred feet, and through the \nfloors and walls of buildings. And in similar experiments in which he\n\nobserved the effects of lightning flashes in place of sparks from a \nLevden jar, he found that he could get the lightning to magnetize \nneedles up to a distance as great as eight miles. This was about \nIS42. Here we have the earliest evidence of ether waves of the type \nthat the radio engineer emplovs. But again the significance of \nHenry's work was not recognized. This could only have come after \nmuch fuller investigation. However, it is instructive to reflect for a \nmoment on what might have been had Henry possessed the time and \nfacilities for carrying his work further. Needless to sav, there is a \nwide gulf between the wireless telegraph of today and its earliest \nprecursor with which Henry received an electromagnetic signal from \na lightning flash eight miles away, but it is wholly possible that, \nhad Henry not been called to other work, the world might have \npossessed a wireless telegraph capable of sending messages over sub- \nstantial distances many vears before it did.\n\nWriting of Henry, Simon Newcomb, the celebrated astronomer \nsaid? \u201cHis scientific work is marked by acuteness in cross-examining \nnature, a clear appreciation of the logic of science, and an enthusiasm \nfor truth without respect to its utilitarian results.\u201d A man of the \nhighest scientific ability, Henry spent the better part of his life as the \nhead of an institution dedicated to \u2018tthe increase and diffusion of \nknowledge among men.\u201d\u2019\n\n\u201cThe mantle of Franklin has fallen upon the shoulders of Henry,\u201d \nwrote Sir David Brewster,\u2019 the eminent English scientist, and it is \nreported that Abraham Lincoln declared, when he became acquainted \nwith Henry after assuming the Presidency, \u2018The Smithsonian In- \nstitution must be a grand school if it produces such thinkers as Henry.\u201d \nHe was, in every way and in the best that the word implies, a scientist, \nand the interest in scientific questions which dominated his life, \nremained with him to the very end,\u2014-almost the last words to pass his \nlips were whether the transit of the planet mercury had been suc- \ncessfully observed. If we use the word \u2018Dean\u2019\u2019\u2014so rich in academic \nassociation\u2014to stand at once for the greatest usefulness to one\u2019s \nfellowmen as well as for the highest achievements in the field of \nscholarship and research, for lifelong devotion to public service, for \nbreadth of view and tolerance regarding all questions, whether arising \nin science or directly out of human relations, and as epitomizing all \nthat is best and highest in man\u2019s intellectual life, we may well call \nJoseph Henry the Dean of American scientists.\n\n\u00b0 Biographical Memoir; prepared by Prof. Asa Gray in behalf of the Board of \nRegents of the Smithsonian Institution.\n\n4 accepted truth is the result of every-day experience. From \nthe simplest type of measurement, such as determining the length \nof a board with an ordinary tape measure, to the most refined type \nof measurement, such as determining the charge on an electron, \nerrors are bound to creep in.\n\nNow, a manufacturer must constantly make measurements of one \nkind or another in an effort to control his production processes and \nto measure the quality of his finished product in terms of certain of \nits characteristics, but, before he can safely determine the significance \nof observed differences in his production processes or in the quality \nof his product as given by these measurements, he must make allow- \nance for his errors of measurement; i.e., for the fact that the observed \ndifferences may be larger or smaller than the true differences. To \nmake such allowances for the errors of measurement of any character- \nistic, to find out what the true magnitude of the characteristic most \nprobably is, to find out, as it were, what a thing most probably is \nfrom what it appears to be, presents an endless chain of interesting \nproblems to be solved.\n\nThree important types of problems arising in engineering practice \nare discussed in this paper. They are:\n\n2. Error correction of data taken periodically to detect significant \nchanges in quality of product.\n\n3. Error correction of data taken to relate observed deviations in \nquality of product to some particular cause.\n\nThe solution of the first one is presented here for the first time. \nThe solution of the second has been generalized to include cases not \npreviously solvable. All three types of problems are illustrated.\n\nTyer 1 ERROR CORRECTION OF DATA TAKEN TO SHOW THE QUALITY \nOF A PARTICULAR Lor\n\nLet us take a specific problem first. Assume that we have a lot \nconsisting of 15,000 transmitters ' and a machine with which to measure \nthe efficiency of cach instrument. Suppose we make one observation \non cach transmitter-\u2014a total of 15,000 measurements. Suppose we \nfind, as in the distribution illustrated in Fig. 1, that one measure- \nment is in the efficiency range \u2014 1.75 to \u20141.50, 17 within the range\n\nKig. 1-\u2014Typical frequency distribution. Chart showing observed number of trans- \nmitters versus efficiency\n\n1.50 to \u20141.25 units, and so on. The vertical height of a point \nrepresents the number or frequency of occurrence of observations \nfalling within the corresponding interval laid off on the horizontal \naxis of the chart.\n\nSo far so good, but suppose a customer wants to buy these trans- \nmitters. We know that some transmitter which appeared to have an \nefficiency within the range of 1.25 to 1.50 units say, may actually have \nhad an etheiency within some other interval. We know too that, \nbecause of the errors of measurement, the transmitters appear to \ndiffer more among themselves than they really do. We therefore\n\n\u2018Of course, the efficiency of a transmitter does not remain constant during a\n\nseries of tests but these inherent variations in the transmitter may be considered, \nfor our purpose, as forming a component part of the resultant error of measurement\n\ndesire to find the most probable numbers of transmitters within tl \ndifferent intervals indicated in Fig. 1.\n\nthe interval of efficiency from Y to VY+dN is It is this \nfunction f7(XY) that we want to find. Similarly let us assume that \nthere is some function such that vives the observed \nnumber of transmitters appearing to have ethiciencies within\u201d the \ninterval Y to Y+dY where the measurements are made by a method \nwherein the probability of making an error within the interval x to a \n+dx is fi vide. [tis reasonable to expect that, if two of these func\n\ntions are known, the third can be easily determined. We shall pro \nceed to show that this is the case. Let us first tind the law of errot \nexperimentally.\n\ngiven magnitude in measuring the efficiency of any transmitter \nNaturally, the only way of doing this is to make a series of measure \nments on a single transmitter from which we can determine the \nobserved frequency of occurrence of measurements which differ from \nthe average by some fixed amount, and thus tind what percentage of \nthe total number of measurements may be expected to fall within \nanv given range on either side of the average. Common sense and \nintuition may tell us that we may expect to find a large percentage \nof the measurements within a narrow range on either side of the \naverage, that there will be just as many measurements greater than \nthe average by a certain amount as there are less than the average \nby the same amount, and that large deviations from the average may \nbe expected to occur with less frequency than small deviations \nSuppose we make 500 observations of the ethciency of a single trans \nmitter and find the distribution given in Fig. 2. Just as we might \nhave expected, the observed values of the efficiency of the transmitter \nare grouped symmetrically about the average of all the observed \nvalues. We see that the maximum deviation between observations \non a single transmitter is quite large (33\u00b0, ) compared with the actual \nmaximum differences observed between the efficiencies of the trans \nmitters.\n\nThe results reproduced in Fig. y sugvest that the deviations for the \ncase in hand are distributed in a manner closely approximating the\n\nbell-shaped distribution so familiar in the theory of errors. We often \nfind, as we do in this case, that the observed distribution can be closely \napproximated by a function fx(x) of the form\n\nwhere fr(x)dx is the probability that an error x will lie within the \ninterval x tox+dyx, \u00a2 is the root mean square or standard deviation, \nX is the arithmetic mean value and (Y \u2014Y) is the deviation x. The\n\nFig. 2\u2014Typical form of distribution of errors of measurement. Chart showing \nnumber of measurements on a single transmitter versus efficiency\n\nfunction f(x) is referred to in the literature as the normal law of \nerror. If we try to fit such a curve to the deviations\u00ae given in Fig. 2, \nwe obtain the results shown in Fig. 3. This figure is the same as \nFig. 2 except for the addition of the smooth normal curve of error \ncalculated for the observed data. Without further consideration, we \nshall assume the law of error to be normal and hence of the form\n\nWe have next to consider the choice of the function to represent \nthe true distribution f7-Y). Often we have reason to believe that this \n* If the average of the observed values of the 500 observations of efficiency given\n\nin Fig. 3 is assumed to be the true value of the efficiency of the transmitter, then the\n\ndeviation of an observed value from this mean is also the error of this observed value. \nWe shall use the terms \u201cerror\u201d and \u2018\u2018deviation\u201d interchangeably in this sense.\n\nis also approximately normal, and hence we shall consider first the \nmethod for finding the observed distribution f,(Y) for the special case \nwhen both the true distribution f7(.Y) and the law of error fr(X) are \nnormal; t.e., when they are both of the form given by Equation (1\n\nWe shall first obtain an experimental answer to this problem. \nSuppose we take, say, 1,000 instruments of some kind which are\n\nFig. 3\u2014Chart showing the observed distribution of errors fitted by a tvpical smooth \ncurve. Data of Fig. 2 fitted by normal law of error, Eq. 1\n\nknown to be distributed in normal fashion, in respect to some char- \nacteristic, with a standard deviation op. Let us measure each of \nthese instruments by a method subject to the normal law of error \nwhose standard deviation oy, is } or. The results of one such experi-\n\nment are given in Fig. 4. The observed frequencies of occurrence \nare represented by the circles. It was found that this observed \ndistribution could be closely approximated by a normal law f.(X)\n\nin a succeeding paragraph. The theorem is: When the true distri- \nbution f7(\u00a5) and the law of error fe(x) are both normal (hence ex- \npressible in form indicated by Equation (1)) with root mean square \nor standard deviations % and \u00a2,% respectively, the most probable ob- \nserved distribution will be normal in form with a standard deviation \nV 03.\n\nnormal as it should be if fr(X\u00a5) and fr(x) were both normal. We \nmust therefore, try some other function for f7CY).\n\nOf course, experiments might be performed for other types of true \nand error distributions, but in all such cases the results, as in the \nilustration just considered, would be subject to errors of sampling.\n\nFig. 4 -Experimental results shpwing effects of errors of measurement. Normal curve \nfitted to observed points, when the true distribution and the law of error are both \nnormal\n\nHence we shall proceed at once to the analytical treatment of the \nproblem.\n\nAssuming the law of error to be normal, we see that the fraction \nfe(xjdx of the number of objects having magnitudes between Y +x \nand Y+x+dsx will be measured with an error between \u2014x and \u2014x\u2014dx\n\nlor the particular case treated in a previous paragraph where both \nthe true distribution f7(Y) and the law of error fe(x) are normal, \nwe may write Equation (2) in the form\n\nthe true and error distributions respectively. Integration of Equation\n\nI-quations (4) and (5) are the analy tical eX pression for the rule stated \npreviously, for finding the observed distribution (..Y) when both \nthe true and error distributions are normal, because Equation (4\n\nshows it to be normal and Equation (5) expresses the standard devi- \nation o, of the observed values in terms of those of the true values \nand of the errors.\n\nIn practice, however, we often tind that the true distribution is \nnon-symmetrical or skew and can be more nearly approximated \nby the function\n\nwhere ky is a measure of the asymmetry or skewness, the modal o1 \nkey \nmost probable value of X being at a distance \u2014\u00b0 97 from the average \nSee Appendix 1 where another method of solution is given \n\u2018 This is often referred to in the literature of statistics as the secon \ntion. It is in fact the first two terms of the Gram-Charlier series\n\n18 BELL SYSTEM TECHNICAL JOURNAI \nvalue of YY. Substitution of this expression and a normal error\n\nfunction in Equation (2), vields upon integration \u00ae the following \ndistribution f.(.V) of the observed values\n\nWe see that the distribution f,CY), Equation (7), of the observed \nvalues is of the same form as that f7(.Y), Equation (6), of the true \nvalues. The standard deviation of the errors of measurement \u00a2p, \nas in the previous case, has equal weight with the standard deviation \n7, in influencing the standard deviation o, of the observed values. \nThe degree of asymmetry of the observed distribution as measured \nby the skewness k, is, however, less (Equation (8) ) than that of the \ntrue distribution as measured by the skewness ky of the true dis- \ntribution.\n\nNow we can correct the observed distribution, Fig. 1, for the errors \nof measurement, because we find that the observed frequencies, Fig. 1, \ncan be closely approximated by a function of the twpe defined by \nEquation (7). Knowing that the law of error, Mig. 3, is normal we \nconclude that the true distribution f7CY) must be a function of the \nsame type as f,(Y) was found to be except that the true standard \ndeviation 77 will be, from Equation (5), \u00a5/\u00a22+ 3, and the true skewness\n\nlated from the observed distribution, Fig 1, and 7%, can be determined \nby the data given in Fig. 3.\n\nThus finding the values of \u00a2; and kr and substituting them in \nEquation (6), we have the function f7(.Y) representing the true dis- \ntribution which we started out to find. Krom this knowledge of \nfr(X) we can now get the most probable frequencies of occurrence \nof the different efficiencies. Subtracting these frequencies from those \nobserved and shown in Fig. 1, we get the corrections plotted in Fig. 6, \nexpressed as percentages of the observed frequencies.\n\nWe are now in a position to summarize the practical routine to be \nfollowed in finding the most probable distribution f;(.Y) of quality \nwhen the observed distribution is given.\n\nTo tind f7CV), we must first Know the law of error fe(v). We \nmust show this to be normal and tind the standard deviation \u00a2%\n\nFig. 6\u2014Correction which must be applied to the observed distribution of transmitters \nFig. 1, because of the existence of errors of measurement\n\ncision we attain in finding f7(Y) depends upon the number of obser- \nvations made in finding 7p.\n\nHaving found \u00a2% to the required degree of precision, we must next \ndiscover whether or not the observed distribution f;(X) is either \nnormal or the second approximation. Standard statistical methods \ncan be used for this purpose.\n\nand, if f (XY) is second approximation, we know that f7(.X) is given \nby Equation (6), where op and ky can be found with the aid of Equa-\n\ntions (5) and (S) in terms of the observed values of op, o) and k.. \nIn other words we have\n\nCORRECTION OF DATA TAKEN PERIODICALLY TO DETECT SIGNIFICANT \nCHANGES IN QUALITY OF PRODUCT\n\nIrrespective of the care taken in defining and controlling the manu- \nfacturing processes, the units of a product will differ among them- \nselves in respect to any measurable characteristic. Random fluctua- \ntions in such factors as humidity, temperature, grade of raw material, \nand wear and tear on machinery may produce such differences be- \ntween units of a product. Such random variations in the factors \nunderlying the manufacturing process usually yield a product in \nwhich the units differ in random fashion according to some law of \nprobability.\n\nCustomarily, product is inspected periodically, and the data are \nanalyzed to determine if the observed difference in two samples is \ngreater than can be accounted for as a random variation. If it is, \nwe may assume that the manufacturing processes have changed \nsignificantly for some reason which further investigation should dis- \nclose. Now, the presence of errors of measurement effectively in- \ncreases the magnitude of the random differences to be expected from \none sample to another and hence makes it harder for us to detect \ntrends or fluctuations in product. Let us investigate this effect of\n\nSymbolically we may assume that the probability of production of \na unit of product having a characteristic Y within any range Y to \n+dN is fp( N)dX, where the characteristic Y is measured by a method \nsubject to a law of error fe(x), so that fe(x)dx represents the proba- \nbility of occurrence of an error x within the range x to x+dx. The \nproblem is to find the corresponding distribution f,CY) for the ob- \nserved magnitudes.\n\nObviously the observed magnitude Y, is the algebraic sum of the \ntrue value Y and the error x. Assuming that there is no correlation \nbetween these two quantities, the probability of a unit having a value \nof Y within the range Y to VY + dX being measured with an error\n\nwithin the range x to x + dy is fy(.N)dN fyp(xidx. Assuming that \nY,=N+.x we may write the probability\n\nbecause f,U.Y,) is obtained by taking into account that all possible \nvalues of x between + % and \u2014 * may be combined with a given Y,. \nThis integral is of the same form as that given in Equation (2). Inte-\n\nwhere as before \u00a2,=\\/o}+o;. This result is well known as the law \nof propagation of error.\n\nWhen f(x) is normal and fy XY) ts given by the first two terms of \nthe Gram-Charlier series, Equation (6), with skewness ky; and stand- \nand deviation o7, the observed distribution f,CY,) 1s of the same func- \ntional form as the true distribution f7(.Y) and has values of standard \ndeviation o, and skewness k&, given by Equations (5) and (8) in Part I. \nThis result appears to be new.\n\nNow for the case where the true distribution f7CY) and the law of \nerror fe(x) are both second approximation type, the integration ts \nsomewhat tedious, but we can approach a special case of this problem \neasily from a slightly different angle as indicated in Appendix 2. \nUnder certain special conditions therein set forth, the resultant dis \ntribution is also second approximation form with a skewness which\n\nExam ple of Applications to Determine Most Economical Way of \\easur- \ning Quality\n\nLet us next consider a very simple method of using the above \nresults to indicate the most economical method for determining the\n\nWhat is the most economical way of determining the quality of \nproduct within some predetermined range Y+AY with a known \nprobability P, where XY is the average quality? Let us assume that:\n\ndy =cost of making each measurement, \n=number of units selected, \nts =number of measurements made on each unit, \no, =standard deviation of the errors of observation. \no,=o0,=standard deviation of the true distribution\n\ncent. of the observations, and hence AY =3cy. \nThe average of 7, measurements made on one unit is the observed\n\nOver 99 per cent. of the averages of samples of size .V drawn from \na product whose law of distribution is f7(.X) where fy(Y) is either \nnormal or second approximation may be expected to lie within the\n\nfalls outside these limits, this fact is taken as probably indicating the \nexistence of a trend or cyclic fluctuation in product, the cause of \nwhich should be sought. The presence of errors of measurement \nincreases the separation of these limits to 6o, trom 6e;. Our pre- \ncision of detecting trend or cyclic fluctuation is thereby decreased.\n\n25 per cent. greater than oy. In some instances \u00a2@ has been found \nto be nearly 50 per cent. greater than o;.\n\nERROR CORRECTION OF DATA TAKEN TO RELATE OBSERVED \nDEVIATIONS IN QUALITY OF PRODUCT TO \nSoME PARTICULAR CAUSE\n\nIn many practical cases it is not possible to write down an equation \nto show how the quality of a finished product depends upon the \nfactors controlled by different manufacturing steps. To cite one such \ncase, we may know that the quality of the finished article depends \nupon the control of the temperature to which some of the piece parts \nare heated in the process of manufacture. Thus the microphonic \nproperties of carbon depend upon the temperature to which the \ncarbon is heated. In cases where the relationship between quality \nand some factor (such as temperature in the above illustration) can \nonly be determined through a study of the correlation existing between \nthe quality and the particular factor, use must be made of the correl- \nation coefficient r which is defined as\n\nwhere x and y represent respectively deviations from the average \nquality Y and the average magnitude Y of some factor which is\n\nto be controlled by the manufacturing process, and .V is the number of \nobservations. Now, if errors of observation are made in determining \nx and y, the observed correlation coefficient ry,y,, is known to be given \nby the expression\n\ngx, and oy, being the root mean square errors of observation of x\u00bb and \ny respectively.\n\nAttention is directed to Equation (10) which shows that the observed \ncorrelation coefficient 7, is always less than the true correlation co- \nefficient: 7, irrespective of the number of observations made. Ob- \nviously, this point is of considerable commercial importance as we \nshall now see.\n\nIf the observed correlation is small, we customarily assume that \nthere is little need of trying to control the quality Y by controlling the \nmanufacturing factor Y, whereas this conclusion cannot be justified \nunless it can be shown that the true correlation has not been masked \nby the errors of measurement.\n\nThis point has had to be taken into account in the development \nof machine methods for testing transmitters and receivers, because \nthe calibration curves of the machines in terms of ear-voice tests \ndepend upon the correlation coefficient.\n\nIt may be of some interest to certain readers to note that the results \ngiven in Equations (4) and (7) can also be obtained in the following \nway by the method of moments so often used in statistical investi- \nvations.\n\nIf we substitute a normal form for fr(Y) in Equation (11) and \nsolve for the moments of {,CY), we find that the odd moments are zero\n\nand the ratio of the 4th moment to the square of the 2nd is numerically \n3 which indicates that f,CY) is normal in form.\n\nA similar substitution of the 2nd approximation form for fy(Y) in \nEquation (11) yields a distribution f,.Y) from whose moments we\n\ndeduce Equation (7). Use is made in this proof of the easily demon- \nstrated theorem that \n/ =O \nx \nif i where is the jth derivative of the normal law function,\n\nIt is well known that the normal law of distribution may result from \na system of \u00bb (nv being large) causes each of which produces an incre- \nment AY measured from some fixed origin with a probability p=! and \nno increment with a probability g= 3. Furthermore the second ap- \nproximation may result from a similar system in which p+g and \u00bb \nis large. Under such systems of causes, the probabilities of the oc- \ncurrences of m, u\u20141,...3, 2, 1, 0 increments are given by the suc- \ncessive terms of the point binomial (p+4q)\u201d.\n\nLet us assume that the symbols pr, gp, v7, AN and pp, de, Ne, AX \nrefer to the systems of causes controlling the product and_ errors \nrespectively. The probabilities of observed combinations my; + \nneAx, (np\u2014WAN + (ne\u20141)Ax,... are given by the successive terms \nof the expansion (pr+qr)\"\u2019 (petgqr)\"\u00ae. Now for the special case \npr=pr=p and AX =Ax we have the resultant probability distribu- \ntion with skewness \nq\u2014P\n\nNow if p=q, the skewness kj is zero and the observed distribution \nis more nearly normal than either component, and its standard devi- \nation \u00a2o is the square root of the sum of the squares of %p and 7p. \nThis result is similar to that given by Equation (4) of this paper.\n\nWe may also consider by this method a case not treated in this \npaper. When the skewness ky of the true values is equal to that\n\nkr of the law of ertor, or, more particularly, when mp=ne_=n, pr= \nPe=p, dr=qi=q, p= q, we see that the observed distribution is given\n\nor \ntimes the standard deviation of either of the true or error distributions.\n\nThe Theory of the Operation of the Howling \nTelephone with Experimental Confirmation\n\nsufficient numerical data are given to enable one to calculate the intensity \nand frequency of howling for various types of systems. Detailed con- \nsideration is given to the following three systems, namelv, one where the \ntransmitter and receiver disphragms are coupled together mechanically \nby a lever system, one where they are coupled by a small box of air, and \none where they are coupled by a long tube of air. The ty pe ot electrical \ncircuit to use with each of these systems depends upon the type of perform-\n\nfront of the mouthpiece of the transmitter, a shrill note is \nemitted. A sustained oscillation is set up in the electro-mechanical \nsystem which is frequently called \u201chowling\u201d or \u201csinging\u201d or \u201chum- \nming.\u201d\n\nThis phenomenon was first observed by A. S. Hibbard ot \nthe United States in IS90.) Frank Gill was the first to publish an a \ncount of the phenomenon. He first noted that the pitch of the howling \nnote was changed by reversing the telephone receiver connection. \u2014 In \nsummarizing further his experimental results, he states \u201cthat the pitch \nof the note appears to be determined by the length of the column of air \nbetween the two diaphragms and the conditions of the circuit. As the \nperiodic time of the circuit is increased, the time of the note rises. To \nsome extent, the pitch is governed by the rate of the diaphragm, but I \ndo not think this is so important a factor as the others. The main \nfactors appear to be the angle of lag and the length of the column of \nair between. the diaphragms. Although the vibration is a forced \none, we could almost see that its rate is largely dependent on the \nfree period of the circuit.\u201d !\n\nIn 1908 Kennelly and Upson extended Gill's work and made ex \ntensive experimental investigations of the case in which the trans \nmitter and receiver are coupled together acoustically by means of a\n\n'Taken from a paper on \u2018Notes on the Humming Telephone\u201d by F. Gill, read \nat a meeting of the Dublin Loeal Section of the Society of Telephone Engineers\n\nand published in the Journal of the Institution of Electrical Engineers, Vol. NXNXI, \n1901,\n\nhollow circular tube of varying lengths and electrically by means of \nan induction coil, The summary of the conclusions is as follows: *\n\n(1) The mean frequency of the humming-telephone note is de- \ntermined solely by the receiver diaphragm, and its natural free rate \nof vibration. (2) The ascending intersections of the frequency \nzig-zag with the mean frequency line will be formed approximately \nat tube lengths of (8. 4+) v mo cm. for one connection, and of (144 \nm) von, em. for the other connection, of the receiver; where v is the \nvelocity of sound in air, 7 is the mean frequency in cycles per second, \nand m is any positive integer, within the working range of the tube. \nThe constants 3,4 and 1/4 may be modified by the presence of con- \ndensers, and other circumstances. (3) The range of pitch variation. \nand the breaking positions, are determined by the transmitter, and \nby the reinforcing capability of the system. For systems that are \nweak, either electrically or acoustically, the range of pitch, above or \nbelow the mean, will be small. (4) The primary current, as measured \nby a DC instrument, is ordinarily a minimum at the mean frequency, \nand a maximum at a break. (5) Transmitters may be tested for \neffectiveness, by measuring their hum-extinguishing resistances in \nthe primary or secondary circuit. The tube length should be such \nas to produce mean frequency if one connection of receiver only is \nused, but should favor both connections equally, if both connections \nof receiver are used.\u201d\n\nThey also give a first approximation theory to account for the \nchanges in frequency as the length of the coupling tube is changed.\n\nIn 1917, H. W. Nichols gave the general equations for the \nspecial case where the two diaphragms act as pistons closing the \nends of a tube of air. This case was given as an illustrative example \nof the \u201cTheory of Variable Dynamical Electrical Systems.\u201d\n\nThis paper gives a theoretical treatment of the behavior of a system \ncontaining a transmitter and a receiver coupled together acoustically \nand electrically, and with a source of electrical energy feeding the \ntransmitter. Formulae are deduced which give the frequency and \nintensity of howling in terms of the physical constants of the system. \nNumerical calculations are given sufficiently detailed) solution \nof some special cases are given to enable one, who is interested in \nusing the howling telephone as a source of alternating current. or \nfor other experimental work, to design the set for his particular \npurpose.\n\n?\u201cHumming Telephone\u201d by 4. E. Kennelly and Walter L. Upson, American \nPhilosophical Society, July 20, 1908.\n\nThe elements of a telephone system which is howling are the trans- \nmitter, the receiver, the mechanical coupler and the electrical coupler \nas indicated in Fig. 1. Hf there is a source of electrical power in the \nelectrical coupler, which is released by movements of the transmitter \ndiaphragm in the form of clee trical Vibrations, and also, if there is a\n\nhowling will result. In other words, if the gain in the transmitter \ndue to its amplifving action is just equal to the losses in the electrical \nand mechanical circuits, then a steady oscillatory state will be \nmaintained. The problem is to determine the aature of these \nrelationships.\n\nAssume that the conditions are such that a steady oscillatory \nstate has been set up. Under such conditions let 7) be the electrical \nimpedance of the transmitter, R the impedance looking away from \nthe transmitter terminals into the electrical coupler, and Zr the \nimpedance of the receiver. [tis well known that the impedance Zp \nis dependent upon the velocity of motion of the receiver diaphragm. \nAlso, T is dependent upon the amplitude of motion of the transmitter \ndiaphragm as well as upon the direct current supplied to it. Conse- \nquently, the impedances detined above are not only dependent upon \nfrequency but also upon the mechanical coupling and magnitude of \nthe current supplied to the transmitter.\n\ncurrent flowing through it Loth expressed in root mean s;uare values\u2018, \nthen \ne=(I+R)i (1) \nIt is cousenent to define a quantity JZ which shall call the uni- \nlateral mutual impedance by the qeuation\n\nwhere \u00a2; is the electromotive force created in the transmitter when \na current 7; flows in the recetver circui.. [It is a quantity which is \nclosely related to the etfectiveness of the mechanical coupling and \nthe efficiencies of the transn: iter and receiver.\n\nIf the electrical coupler be considered part of the receiver, and \nthe transmitter and rec iver circuits are connected together as in \nVig. J, then e=e,, and f=7,. Consequently\n\nthe eoadition .or sustained oscillation. This condition is in effect \na pair of condtiiors, as the two sides of the equation must be equal \nboth in amplitude and in phase. These two conditions are sufficient \nto determine the irequency and intensity of howling.\n\nIn order to express J\u00e9 and Rin more fundamental physical constants, \nit is necessary to examine more closely the mechanical and electrical \nconnections. Before doing this for some important special cases, \nit will be necessary to discuss some of the electro-dynamical properties\n\nFor the sake of clarity the discussion will be confined to permanent \nmagnet receivers and carbon transmitters. The modifications neces- \nsary for other types of instruments will, | think, be evident from the \ndiscussion. Representing by Fr and Fr the forces acting on the \ndiaphragms of the receiver and transmitter respectively, and by y\u00a5 \nand s their displacements, we have the following equations defining \nthe \u2018stiffness factors\u2019\u2019 Spr and\n\n\u2018In what follows all quantities involving periodic variations will be expressed \nas root mean square values unless otherwise specified, and the vector notation \nwill be used for denoting phases.\n\nThese factors are usually complicated functions of the frequency \nwhile Sp likewise depends on the kind and amount of agitation. In \nthe case of a system of a single degree of freedom which mav_ be \nregarded as a first approximation to this case\n\nwhere w is 27 times the frequency. When referring to the movements \nof a diaphragm, the quantity m represents the mass, r the mechanical \nresistance, and s the elastic constant. The stiffness factor S divided \nby jw is usually called the mechanical impedance.\n\n<x \n> \nVALUES OF THE FORCE FACTOR \nOF THE RECEIVER \nz . FORCE FACTOR Z \n< } \n| \n30 \n0 200 400 600 800 1000 (200 1400 \nFREQUENCY \nFig. 2\n\nAn important constant which enters into the determination of the \nunilateral mutual impedance \\/ is the force factor of the receiver \nwhich will be designated by Z. It is defined as the force in dynes \nacting upon the diaphragm per unit of current. For the receivers \nused in this investigation, its values in magnitude and phase are shown \nfor various frequencies in Fig. 2. These were determined by the \nmethod outlined by Wegel2 In the region of the resonant frequency \nits value in absolute units can be approximately represented by\n\nThe impedance Zp of the receiver varies with frequency and de- \npends upon the load on the diaphragm. If S is the loaded stiffness \nof the diaphragm, that is, its resistance to force under actual working \nconditions, and Z; is the impedance of the receiver when the diaphragm \nis prevented from moving, then it is well-known that\n\nIt was found that Zy expressed in ohms could be represented in the \nfrequency region near resonance by the formula\n\nThe electromotive force e created in the transmitter, the direct \ncurrent J flowing through it, and the displacement of the diaphragm \nare related in a rather complicated way. For describing this rela- \ntionship it is convenient to define a modulation factor h# by the \nequation\n\nwhich shows that the modulation factor is also an important one in \ndetermining the unilateral mutual impedance. For sustained \noscilation the factor Ji does not enter into the periodic variation \nand may be thought of as an electro-mechanical impedance between \nthe electromotive force created in the button and the displacement \nof the diaphragm of the transmitter. However, for a different condi- \ntion of sustained oscillation which results in giving 2 a different mag- \nnitude the value of 2 changes. In other words / is dependent upon \nthe agitation of the carbon as represented by s, and also upon the \ndirect current supplied to the transmitter. [It is mainly this variable \ncharacter of A that makes it possible to fulfill the conditions for sus- \ntained howling.\n\nSimultaneous measurements of e, J and s were made upon several \ntransmitters of the type used in this investigation. From the results \nobtained and from the defining equation (10) for /, it was found that\n\nthe following empirical equation would represent approximately the \nrelation between /, J and s, namely\n\nwhere 2 is expressed in) microns and J in amperes \u00a2 in volts and h \nin ohms per micron. To facilitate solving for z when k and TJ are \ngiven, a set of curves showing this relation is given in Fig. 3.) It \nis this modulation factor # which measures the efficiency of the trans- \nmitter button.\n\n- Relation Between the Modulation \nFactor and the Displacement \n\u201c2\u201d of the Transmitter Diaphragm\n\nIt is also necessary to know the dependence of 7 upon 2 and J. \nTo obtain this relation corresponding values of e and Vy the DC \ndrop across the transmitter as measured by direct current measuring \ninstruments, were obtained for various degrees of agitation and \namounts of direct current. Four transmitters were used in establish- \ning the relation, the results being shown in Fig. 4. Then, for any \nvalue of the supply current J a value of 7 can be obtained from V. \nFrom the corresponding e a value of and s can be obtained from\n\nand 6 were obtained. It is thus seen that for a given type of trans- \nmitter if the direct current and any one of the four quantities e, /, \u00a2, \nor T are known, the others are determined and may be obtained from \nsuitable curves.\n\nTHE RELATION BETWEEN THE MODULATION FACTOR fh \nAND THE TRANSMITTER RESISTANCE \u201cT\"\n\nCommercial receivers and transmitters have constants which vary \nlargely from those given above. These values represent the general \nbehavior of such instruments and are useful in understanding their \noperation in a howling circuit. Inasmuch as the performance of such \ninstruments particularly the transmitter depends very largely upon \nthe condition of operation the constants given cannot be applied \nwith confidence to conditions greatly different from those mentioned \nin the paper. With these facts concerning telephone instruments \nin mind we are now in a position to treat some special cases.\n\nCASE 1\u2014DIAPHRAGMS CONNECTED MECHANICALLY BY A RIGID \nAND WEIGHTLESS LEVER\n\nTo illustrate the method of solution this special case will be solved \nin some detail. A diagrammatic sketch illustrating the connections \nis shown in Fig. 7. Neglecting the reaction of the air, the vibration of \nthe receiver diaphragm is controlled by the force Zi exerted by the\n\nreceiver winding and the opposing force Y exerted by the connecting \nrod.\n\n13) \n= ( \nSr \nIf the lever is rigid and weightless and has an arm ratio c, then \nFr=cFr (14) \nand due to the restraint \ncx \nY= cs = (10) \n5; \nUsing these equations together with equation (11) it is seen that \nIhZ \n(16) \nCSR\n\nwhere Rpc is the direct current resistance of the receiver winding \nand k is the line resistance. The condition (3) for howling then \nbecomes\n\nPhis is equivalent to two scalar equations and taken together with \n19) and the curves of Fig. 6 gives the necessary four equations to\n\nThe solution, however, is not) straightforward since the rela \ntion between 7. and J is only given empirically by a set of \ncurves. By \u201ccut and trv\u201d methods the solution for anv numerical \ncase can be obtained. The last term of (20) is usually nevgligibl \nor at least it is of second order of magnitude.  \u00a2 onsequently, th \nsum ot the phase angles of the other factors must be APpronimiate hy \nequal to the phase of Z. This completes the formal solution for\n\nThe solution of a numerical case throws considerable light) upon \nthe physical phenomenon taking place, and also upon the method of \ncalculation. Let the arm ratio be unity, a case corresponding to that \nwhen the diaphragms are connected directly together, and assum \nthat the supphy current is furnished by a battery of 24 volts throug! \na line having a resistance of 300 ohms. Using the constants for the \nreceivers and transmitters given above and expressing f in kiloewcles \nTin ohms, J in amperes and # in ohms per micron, equations (14\n\nIf J is positive there is no solution for f/, since the angle of the first \nfactor is in the first quadrant, and that of the second factor either \nin the first or second; consequently, the phases cannot match at any \nfrequency. If the supply current is reversed, then J is negative or \n180\u00b0 is added to the phase of the left hand member making it a positive\n\nIt a value of \u00a2 equal to 2.7, which is approximately equal to the \nsquare root of the ratio of mechanical impedances of the two dia- \nphragms, then the solution for reversed DC supply becomes\n\nIt is thus seen that changing the ratio arm has increased the howling \nintensity, but the increase for the various clements is greatly different. \nThe frequency is slightly lowered, the values of # and J have been\n\nreduced by 26\u00b0, and 14\u00b0 respectively, while the values of y, 2, 7, \nt and have been increased 40000, 105\u00b0C, 5706, 2206 and respect- \nively.\n\nIf the circuit of Fig. 7 is modified as shown in Fig. 8, the induct- \nance L being very large, then the condition for howling becomes\n\nThe solution for values of A= 1 mi, K=1/2 mf, and K= 1/5 nit \nare given in Table T. When A= 1 mf and the supply current is direct \nthe solution which satisfies the phase equality is f= 506. This corre \nsponds to h= 220 which is an impossible value. Therefore, no howling \nwill be sustained for this condition. For A= 1 2 mf the svstem will \nhowl for both direct and reversed supply current, the frequency \nchanging suddenly from 839 to L119 eveles as the current is reversed \nwhile the other variables change only slightly.\n\nIt is interesting to note the change in the howling frequency as \nthe value of AK increases. When the supply current is negative, and \nfor values larger than 1 mf, the frequency of howling is always close \nto 1000, as K goes from 1 to 1,2 the frequency increases to above \n1100. For smaller values of A the frequency continues to slowly \nincrease until, for values smaller than 1 3, the system ceases to \nsustain oscillations. \u2014 For positive values of supply current no howling \nwill result until A becomes smaller than 2.3 where the frequency is \naround 800.) The frequency then increases reaching a howling fre- \nquency around 1000 for A= 17.) For smaller values of K no howling \nwill be sustained.\n\nIt will be assumed that the air chamber is so small that the phase \nof the pressure variation is the same on both diaphragms. Let V \nbe the volume of air between the diaphragms. Then\n\nwhere Vis the volume of air in the undisturbed state and Or and Q; \nare the effective areas of the receiver and transmitter diaphragms \nrespectively.\n\nThe pressure variation in the chamber (changes considered adiabatic \nis given by\n\nWhen the steady state is set up this may be considered a vector \nequation and the variables expressed in rms values.\n\nIn this case the ratio between zs and y is not fixed, but depends upon \nSr which is a function of the frequency.\n\nFor the transmitter and receiver used \nQr = 6.5, \n10.5. \nLet the volume of entrapped air be taken as 10 cc., then\n\nwhere J is expressed in amperes, 7 in ohms, f in kiloeveles and # in \nohms per micron.\n\nComparing this to the case where the diaphragms are coupled by a \nlever having an arm ratio 2.7 it is seen that the air coupling produces \na greater e.m.f. in the transmitter and only a slightly increased AC \ncurrent. The receiver diaphragm in this case, however, has a smaller \namplitude than the transmitter diaphragm. At this\u201d particular \nhowling frequency the transmitter diaphragm stiffness is only about \n1/4 that of the receiver diaphragm stiffness which explains this \nanomalous result. Also, it will be seen that the diaphragms vibrate \nalmost oppositely in phase.\n\nThese cases are sutticient to illustrate the method of calculation, but \nthere is one other important case for which I desire to give the results \nas this is the case handled experimentally by Kennelly and Upson.\n\nCase III\u2014DIAPHRAGMS CONNECTED ACOUSTICALLY BY A TUBE OF \nAIR OF UNIFORM CROSS-SECTION WITH AN \nAIR CHAMBER AT ENpDs\n\nIn this case the two diaphragms are connected acoustically by the \nair, but since the tube has considerable length phase differences exist\n\nat different points along it. The connections are shown schematically \nin Fig. 9. \nThe equation of motion for the receiver diaphragm is\n\nwhere dpr and dpr are the pressure variations in the air chambers \nat the receiver and transmitter ends of the tube respectively.\n\nThe equations of motion for a gas in which the movements are \nsmall and in only one direction and in which the fluid friction is \nneglected are as follows: \u00b0\n\nwhere \u00a2 is the velocity potential, \u00a2 the time, a the velocity of sound in \nthe air, x the distance along the tube, p the pressure and p the density \nof the air.\n\nFor the case in which we are interested, a sinusoidal oscillation is \nsustained, so that the special solution\n\nis suitable for our problem. Quantities A and B are arbitrary con- \nstants which are determined by the end conditions. Substituting \nthis value of \u00a2 in equation (35), there results\n\ndp = \u2014pjwe\u2019 (A cos\u00ae* +B sin\u201c) (38) \na a \nIt remains then to determine the arbitrary constants A and B.\n\nAt the receiver end of the tube, the displacement, \u00a2x of the air \ndiaphragm across the end of the tube is related to the displacement \ny of the receiver diaphragm. This relationship is established by the \nfollowing consideration. If g is the cross-section of the tube, the \nincrease in volume in the air chamber is given by\n\nAssuming that the air chamber is so small that the pressure change \nat any instant is the same throughout, and that it takes place adia- \nbatically, we have:\n\ndpr=\u2014} dVr (40) \nVr \nCombining equations (35), (39), and (40), we obtain: \nQSrtr = OrZi\u2014 ( Sr + Ox\u00ae )d pe (41) \np \nSimilarly, \ngSr\u00e9r = (= S7 +O *\\dp (42 \nYP\n\nThen the following conditions must be fulfilled at the two ends of a \ntube of length 7.\n\nyp/ Or yp Or \nOr ~ On , \u00ab o \n+S) + Sr Spr 32) \nOr V1 vp \nThe unilateral mutual impedance J is given by \nThZ \nM= \nN sin\u2014 +P cos \nad ad \nThe condition for sustained howling becomes \nIZ wl wl \nh=Nsi +P cos\u2014. 54)\n\nIf the two diaphragms work directly into the connecting tube as \npistons, then Qr=Qr=q=Q and Vr=Vr=O0 and the expressions \nfor MW and S become\n\nThe method of solution is the same as that given for the sim ler \ncases, although it is evident that the actual work of calculation is \nmore involved.\n\nIt is seen that in such a system the intensity and frequency depend \nupon a large number of quantities, namely: S; and Spr, the diaphragm \nstiffness factors; Og and (Ur the effective areas of the two diaphragms; \nVr and Vy the volumes of air entrapped between the diaphragm \nand the opening into connection tube; the length / and the cross \nsection q of the connecting tube; the pressure a, the density s, and the \nvelocity of sound_a for the*gas in the connecting tube; the resistance \nT, direct current_J and modulation factor / of the transmitter; and\n\n7 These two equations were given by H.W. Nichols in essentially this form in \nthe Physical Review, Vol. 10, p. 171; 1917.\n\nthe force factor and impedance of the receiving circuit. Modification \nof any of these may produce marked changes in the resulting howling.\n\nThe way the length / enters the formula (54) for sustained howling \nindicates that the curves representing the possible frequencies of \nhowling, that is, frequencies which produce equality of phase on both \nsides of the equation, vary periodically with the length.\n\nThe intersection of the branches of these curves on any given \nfrequency line will be separated by distances corresponding to rs \nthat is, corresponding to a wave length at the pitch corresponding \nto f. Also, if the supply current is reversed, that is, the sign of J\n\nchanged, and the length of the tube varied until the frequency of \nhowling is brought back to the original value, the change in length \na \nmust be equal to \u2014- \n2f\n\ncosine factors. Adding a half wave length is equivalent to adding 7 \nto the angle which makes the left hand member the negative of its \nfirst value, and consequently, restores the phase equality.\n\nUsing the circuit shown in Fig. 10 for the electrical coupling, the \nfrequency of howling was computed for various tube lengths, the \nresults being given in Fig. 11.\n\nThe instrument constants were those used before, the other values \nbeing Vr=1.6ce., Vr=6.4 ce., and g=.97 cm\", a=3.43 X 104 cm/sec. \no =.001203 gm/cm*. Using these values the formulae for N and P \nbecome \nN= (\u20141.31f\u00b0+7.5f3 \u2014 9.68f 13.26 )X 105+ j(.141f' \u2014 .63/\u00b0+.36) X 108,\n\nThe points on the calculated curves of Fig. 11 were obtained by \ndirect experimental observation with the circuit shown, and with \nvarious lengths of brass tube coupling the transmitter and receiver \ntogether. The agreement between the calculated and observed\n\nvalues is well within the experimental error involved in determining \nthe constants used in the calculation.\n\nIn Fig. 12 are shown similar calculated curves for a transmitter \ncalled \u201chollow,\u201d that is, for one having a lower natural period of \nvibration. It is coupled to the same receiver as used before. The \ndotted curves in each case represent the behavior for reversed current.\n\nIn Figs. 13 and 14 are shown the probable frequencies of howling \nfor these two transmitters as the tube length of the coupler Is In- \ncreased The shaded areas are the so-called breaking points where \nthe howling may be at either of the frequencies shown.\n\nWith these facts in mind let us review the conclusions reached by \nKennelly and Upson given in the beginning of this paper. It is seen \nthat conclusion\n\ns not warranted. The transmitter and. circuit \nconditions as well as the receiver diaphragm influence the mean\n\nof the curves representing the relation between frequency and tube \nlength is correct and the explanation has just been given. This \nperiodic relation is not only true of the mean frequency line but for \nevery constant frequency line.\n\nfactors including the circuit and end conditions. Conclusion (3) is \npartially correct, the range of the howling frequencies depending \nupon the efficiencies of the transmitter, receiver, and circuit is evident \nfrom equation (54). Calculations show that conclusion (4) is generally\n\nWhen the transmitter and receiver are coupled by the air in an \nopen room the behavior is somewhat similar to the case yust solved \nThe size and shape of the room as well as the disposition of articles \nof furniture will all influence the intensity and frequency of howling \nIn general when the two instruments are moved apart the trequency\n\nR Impedance looking away from Transmitter Ti \nZr Impedance e of Receiver.\n\nV, Volume of Air in Front ot Receiver Diaphrag \nV, Volume of Air in Front of Fransmitter Diaphrag\n\nOn Effective Area of Receiver Diaphragm \nOr Effective Area of Transmitter Diaphrag: \npb Air Pressure.\n\nce Displacement of Au Particle at Receiver End of 1 \nDisplacement of Air Particle at Trans nitter End of 1\n\nelectrotechnics are to the theory of the propagation of current \nand voltage along transmission systems. Of such transmission sys- \ntems the simplest is the non-inductive cable. The theory of the \nnon-inductive cable is not only of great historic interest, relating as it \ndoes to Kelvin\u2019s early work on the possibility of transatlantic teleg- \nraphy, but is also of very considerable practical importance today, \nand serves as a basis for the theory of submarine telegraphy over long \ndistances. We shall therefore consider the propagation phenomena \nin the non-inductive cable in some detail.\n\nThe propagation phenomena in any type of transmission system \nare isolated and exhibited in the clearest possible manner when we \nconfine attention to the infinitely long line, with voltage applied \ndirectly to the line terminals. Furthermore, as we shall see later, \nthe solution for the infinitely long line is fundamental and can be \nextended to the more practical case of the finite line with terminal \nimpedances. We therefore, in this chapter, shall confine our atten- \ntion to the case of the infinitely long cable with voltage applied directly \nto the cable terminals.\n\nConsider a cable of distributed resistance R and capacity C per \nunit length, extending from x=0 along the positive x axis. From a \nprevious chapter (see equations (64) and (65) ), we are in possession \nof the operational equations of voltage and current; they are, for the \ninfinitely long line,\n\n(162) \nap \u00a2 var Vo= \\ We \nwhere a=x\u00b0RC, and V. is the terminal cable voltage at x=0. Let \n\u201cunit then \nV=e7 Var, (164)\n\nThe solution of (164) for V was considered in some detail in the \npreceding chapter; it is, by (129)\n\nwhere t=4t/a=4t/x\u00b0RC. Series expansions of this solution were \nalso given. Another equivalent form is, by (131)\n\nThis last form, recognizable also from inspection of the series expan- \nsion (132), is useful because the integral term is what is called the error \nfunction and has been completely computed and tabulated.\n\nBefore discussing these formulas and the light they throw on propa- \ngation phenomena in the non-inductive cable, we shall derive the \nsolution for the current. A very simple way of doing this is to make \nuse of the differential equation (57)\n\nIt is worthwhile verifying the formula by direct solution from the \noperational equation (165). From formula (g) of the table of in- \ntegrals, we have\n\nComparison with the operational equation shows that they are iden- \ntical, within a constant factor provided we put =a 4. Conse- \nquently the solution of (165) is \nCc ! \n[= e~ = \nN N cRt\n\nexample of the utility of the table of integrals in solving operational \nequations.\n\nThis formula is easily calculated for large values of \u00a2 by expanding \nthe exponential function; it ts\n\nThe propagation phenomena of the non-inductive cable are there- \nfore determined by the pair of equations\n\nspt; that is, of 4\u00a2 divided by the total \nresistance and capacity of the cable from x=0 to x=x. The same \nstatement holds for the form of the current wave: its magnitude,\n\nhowever, is inversely proportional to xR, or the total resistance of \nthe cable up to point x. Consequently a single curve, with proper \ntime scale serves to give the voltage wave at any point on the cable. \nSimilarly a single curve, with proper time and amplitude scales, \nserves to depict the current wave at any distance from the cable \nterminals. These curves are given in Figs. 3 and 4.\n\nReferring to the curve depicting the current wave, we observe that \nit is finite for all values of \u00a3>0; consequently, in the ideal cable, the \nvelocity of propagation is infinite. This is a consequence, of course, \nof the fact that the distributed inductance of the cable is neglected. \nActually, of course, the velocity of propagation cannot exceed the\n\npoint will be discussed and explained more fully in connection with the\n\nIts subsidence to its final zero value is very slow; for example, when \n7 = 100 its value is still \n(0.10). \nTurning to the voltage curve, Fig. 4, we see that it is negligibly \nsmall until 7 reaches the value 0.25, at which point it begins to build \nup. Its maximum rate of building up occurs when 7+=2/3, after\n\nwhich it builds up more and more slowly. Its approach to its final \nsteady value is in accordance with the formula\n\nEven, therefore, when 7 is as great as 100, V differs sensibly from its \nultimate value, unity, its value being 0.8876.\n\nThe power curve VI is given in Fig. 5. V.I is the rate at which \nenergy is being transmitted past the point x of the cable.\n\nThe fact that the form of the current and voltage waves depends \nonly on 4t/x\u00b0RC is at the basis of Kelvin\u2019s famous \u201cKR\u201d law, long \napplied to cable telegraphy and sometimes incorrectly applied to \ntelephony. When the first transatlantic telegraph cable was under \nconsideration, Kelvin attacked the problem of propagation along the \nnon-inductive cable and arrived at formulas equivalent to (169) and\n\n(170). From these formulas he announced the law that the \u2018\u2018speed\u201d\u2019 \nof the cable, i.e., the number of signals transmissible per unit time, \nis inversely proportional to the product of the total capacity and total \nresistance of the cable (KR in the English notation). To see just \nwhat this means requires a little digression into the elementary \ntheory of telegraph transmission.\n\nTelegraph signals are transmitted in code by means of \u201cdots\u201d \nand \u2018\u2018dashes.\u201d\u201d. The \u201cdot\u201d is the signal which results when a battery \nis impressed on the cable for a definite interval of time, after which \nthe cable is short circuited. A \u201cdash\u201d is the same except that the \ntime interval during which the battery is connected to the cable is \nincreased. The \u201cdots\u201d and \u2018dashes\u2019 are separated by intervals, \ncalled \u201cspaces\u201d, during which the cable is short circuited. Now when \nthe cable is short-circuited we may imagine a negative battery im- \npressed on the cable in series with the original battery. Conse- \nquently the current in the cable, corresponding to a signal composed \nof a series of dots, dashes and spaces, will be represented by a series \nof the form\n\nwhere, in the cable under consideration, J(\u2018) is given by (168). fy is \nthe duration of the first impulse, fg\u2014f; of the first space, fs\u2014 tz of the \nsecond impulse, etc.\n\nxCR \nThen the signal can be written as \n2 \no(7) \u2014 +O(7\u2014T2)\u2014 (172) \nxRV x ' \nNow if the relative time intervals 7;, 72... are kept constant (as the \nlength of the cable is varied), the actual time intervals f;, fo... are\n\nproportional to x\u00b0CR or to KR, and the wave form of the total signal \nis independent of KR, when referred to the relative time scale r.\n\nHence, if 7 is the total time of the signal, 7 is proportional to x\u00b07CR \nor to KR). Phat is to say, if the duration of the component dots, \ndashes, spaces of the signal are proportional to the \u201cKR\u201d of the \ncable, the wave form of the received signal, referred to the 7 time \nscale, is invariable, and the total time required to transmit the signal \nis proportional to the \u201cKR\u201d of the cable. Now the maximum theo- \nretical speed of transmission on the cable is limited by the require- \nment that the received signal shall bear a recognizable likeness to \nthe original svstem of dots and dashes: in other words there is a\n\nmaximum allowable departure in wave form between received and \ntransmitted signals. If, therefore, the actual speeds of two cables \nare inversely proportional to their \u201cKRs,\u201d the wave form will be the \nsame. This establishes Kelvin\u2019s \u201cKR\u201d law. As a corollary, if the \nlength of the cable is doubled the speed of signaling is reduced to one- \nquarter, assuming the same definition of signals.\n\nThe foregoing will be somewhat clearer, perhaps, if we refer to \ncurves, 4, 5, 6, 7 of Fig. 6 which illustrate the distortion suffered by \nelementary dot signals in cable transmission. Curve 4 shows the \ndot signal produced by a unit battery applied to the cable terminals \nfor a time interval /=2 a while curves 5, 6 and 7 are the cor- \nresponding dot signals when the battery is applied for the time in-\n\naffect the shape of the transmitted dot, whi \nspeed has reached its theoretical maximum. These curves, it should \nbe observed, can be interpreted in two wavs. First, we can regard the \nlength x of the cable as fixed and the duration of the impressed dot \nas varied. On the other hand, we can regard the actual duration of \nthe impressed dot as constant and the length of the cable as varied \nFrom the latter standpoint the curves illustrate the progressive dis \ntortion of the signal as it is transmitted alone the cable\n\nThe dot signal of relative duration 7 can be written as \nD=TI(r), \n=I(r)\u2014I(7-7), \nand the second expression can be expanded in a Tavlor\u2019s series, giving\n\nHence when the dot signal is of sufficiently short relative duration \nI\u2019, the wave shape of the received signal is constant, 7\u2019(7), and its \namplitude is proportional to the relative duration of the dot.\n\nThis can be generalized for any type of transmission svstem: \nLet the dot signal be produced by an e.m.f. f(t) of actual duration 7\u2019. \nThen the received dot signal, by formula (31), is\n\nHence for a sufficiently short duration of the impressed e.m.f. the \nreceived dot signal is of constant wave form, independent of the shape \nof the impressed e.m.f., and its amplitude is proportional to the time\n\nintegral of the impressed e.m.f. These principles are of considerable \npractical importance in telegraphy.\n\nThe leaky cable, that is, a cable with distributed leakage conductance \nG in addition to resistance R and capacity C, is of some interest. The \ndifferential equations of the problem are given in equations (70); the \noperational formulas for the case of voltage directly impressed on the \nterminals of the intinitely long line are\n\nWriting CRx? =a and RGx? =8, G/C=X, and assuming a \u201cunit e.m.f.\u201d \nimpressed on the cable, this becomes\n\nThese equations are readily solved by means of the table and formulas \ngiven in a preceding chapter.\n\nBut first let us attempt to solve the operational equation (175) for \nthe voltage by Heaviside methods, guided by the solution of the \noperational equation\n\nof the preceding chapter. Expand the exponential function in (175) \nin the usual power series; it is\n\nNow identify Vp with 1 \\/ xt in accordance with the Heaviside rule, \nand 1/pwith jdt. We get\n\nNow in the terms of the expansion (178) identify p\u201d with d\"/dt\" \nand substitute (180); we get\n\nThis series is hopelessly complicated to either interpret or compute. \nIt is, in fact, an excellent illustration of the grave disadvantages under \nwhich many of Heaviside\u2019s series solutions labor. We shall there- \nfore attack the solution by aid of the theorems and formulas of a \npreceding section. The simplicity of the solutions which result is \nremarkable.\n\nNow the operational formula for the voltage in the non-leaky cable \nis (see equation (164) )\n\nIn order to distinguish between the two cases, let us denote the voltage \nin the latter case by V\u2019; thus\n\nthe superiority of the definite integral to the series expansion- -compare\n\nwith the series expansions \u2014and secondly as exhibiting \nclearly the etfect of leakage on the propagated waves of current and \nvoltage. We see that in both the current and voltage the effect. of \nleakage is two-fold: first it attenuates the wave by the factor e \u2122 \n(X=G C), and secondly it adds a component consisting of the pro- \ngressive integral of the attenuated wave. This, it may be remarked, \nis the general effect of leakage in all types of transmission svstems. \nIts effect is, therefore, easily computed and interpreted.\n\nFormulas (185) and (186) are very easy to compute with the aid \nof a planimeter or integraph; or, failing these devices, by numerical \nintegration. However, for large values of f, the character of the waves \nis more clearly exhibited if we make use of the identity\n\nThe first two terms of these formulas are clearly the ultimate stead, \nstate values of the voltage and current waves, and can be deter \nmore\n\nmined by evaluating the infinite integrals. A far simpler at \ndirect wav, however, is to make use of the fact that the ultimate \nsteady values of Vand J are gotten from the operational formulas \nby setting p=0. That this statement is true is easily seen if we \nreflect that the steady voltage and current are gotten from. ti \noriginal differential equations of the problem by assuming a steady \nstate and setting d dt=0. \nFrom the operational formulas we get, therefore\n\nUsing the values of WV\u2019 and J\u2019, as given by (169) and (170), it is ex \ntremely easy to compute V and I, for large values of \u00a2, from (193 \nand (194),\n\nSo far we have considered the current and voltage waves in re- \nsponse to a \u201cunit e.m.f.,\u201d\u201d impressed on the cable at \u00ab=O. It is of \ninterest and importance to examine the waves due to sinusoidal \ne.m.fs., suddenly impressed on the cable, particularly in view of\n\n(194-a) \n\u2014 cos ut | sin wt.h'(t)dt. \nJo \nSimilarly, if the impressed e.m.f. is cos at, \nx-(t) =cos wt | cos wt.h\u2019(t)dt \neJ0 \n(194-b)\n\nThe investigation of the building-up of alternating currents and \nvoltages, therefore, depends on the progressive integrals\n\nFor the case of the voltage waves on the non-inductive, non-leaky \ncable these integrals, by aid of equations (169), become, if we write\n\na ELECTRIC Cll \nFor small values of 7 and \u2019 these integrals can be numerically \nevaluated without great labor. Mechanical devices, such as the\n\nthe Coradi Analyzer gives these progressive integrals automaticalls\n\nIt may be said, therefore, that a complete mathematical investiga \ntion of the building-up of alternating current: and voltage waves \non the non-inductive cable presents no serious ditticulties, although \nthe labor of computation is necessarily considerable. One tact makes\n\nposed. This is, if the foregoing integrals are calculated for a given \nvalue of w\u2019, the results apply to all lengths of cable and all actual \nfrequencies w\n\nsuch that aw is a constant. Then if we double the \nlength of the cable and quarter the frequency, the integrals are un- \nattected.\n\nThe solid curve of Fig. 7 shows the building-up of the cable voltage \nin response to an e.m.f. cos wf, impressed at time f=). The fre-\n\nculated from equations (194-b) and (194-e). The dotted curve shows \nthe corresponding steady-state voltage on the cable; that is, the voltage \nwhich would exist if the e.m.f. cos wt had been applied at a long time \npreceding \u2018= \u00b0. We observe that, for this frequency, the building-up \nis effectually accomplished in about one cvcle, and that the transient\n\nThe case is very much different when a higher frequency is ap- \nplied. Fig. 8 shows the building-up of the alternating current in the \ncable when an e.m.f. sin wt is applied at time \u2018=0. The frequency \nis so chosen that w\u2019=aw/4=107. The outstanding features of this \ncurve are that the initial current surge is very large compared with \nthe final steady-state, and that the transient distortion is relatively \nvery large. It is evident that the frequency here shown could not be\n\nemployed for signaling purposes. This curve has been computed \nfrom the steady-state formulas, and equations (160) and (161) for \nthe transient distortion.\n\nIf the applied frequency w, 27 is very high, the steady-state becomes \nnegligibly small, and the complete current is obtained to a good \napproximation by taking the leading terms of (160) and (161). Thus \nif the applied e.m.f. is sin wf, and w is sufficiently large, the cable \ncurrent is\n\nWe now take up the more important and difficult problem of \ninvestigating the propagation phenomena in the transmission line. \nThe transmission line has distributed series resistance R and _ in- \nductance L, and distributed shunt capacity C and leakage conductance \nG. It is the addition of the series inductance L which makes our \nproblem more difhcult and at the same time introduces the phenomena \nof true propagation with finite velocity, as distinguished from the \ndiffusion phenomena of the cable problem. The cable theory serves \nvery well for the problems of trans-oceanic telegraphy * but is quite \ninadequate in the problems of telephonic transmission.\n\nIf J denotes the current and V the voltage at point x on the line, the \nwell known differential equations of the problem are:\u2014\n\nd } \nL +R) I=\u2014y, \n( dt x0 \n(195) \n(c +G)V=- \nd x \nReplacing d/dt by p, these become \n(Lp+R) I= \u2014 \u2014V, \n(196) \n(Cp+G)V= \u2014 \u2014I. \nx \nFrom the second of these equations \noV \n\u2014 = \nox \nand substitution in the first gives \n: (Lp+ R)(Cp+G)I=\u2014.I. (197) \nSimilarly if we eliminate J, we get \n(Lp+R)(Cp+G) V=\u2014\\V. (198) \n3 \u00a7 With the installation of the new submarine cable, continuously loaded with per \n; malloy, this statement must be modified. In this cable, the inductance plavs \n3 important part, and is responsible for the greatly increased speed of signaling\n\nBe \nwhere \u00ablf and B are arbitrary constants, substitution shows that the \nsolution satisties the differential equation for V7 provided\n\nNow restricting attention to the infinitely long line extending along \nthe positive x axis, with voltage |, impressed directly on the line at \nv=0, the retlected wave vanishes and we get\n\n= ,[(b+p)?\u2014 (202) \nwhere \nv=1/V LC, \nRG \n& \nThen setting V,=1, the operational equations of the problem become \n(203) \nG e Vip+p 0 \nI=0( C+ \\p (204) \np p)* =o\"\n\noperational equation (205). This equation can be solved by aid of \nthe operational rules and formulas already given. The process is \nrather complicated, and there is less chance of error if we deal instead\n\nNow let us search through our table of definite integrals. We do \nnot find this integral equation as it stands, but we do observe that \nformula (m) resembles it, and this resemblance suggests that formula \n(m) can be suitably transformed to give the solution of (208). We \ntherefore start with the formula\n\n7) for being the Bessel \nfunction of order zero. We now transform (m) as follows:\n\n| eH J, )at= = (m.3 \nwhere \\;=-\u2014-A and w,=\u2014- wu. (They are, of course, as yet, arbitrary \nparameters, except that they are restricted to positive values). \n(4) Now if we compare (m.3) with the integral equation (208 \nfor F, we see that they are identical provided we get \nMi>Ppy \n\u20141, \nwhich is possible, since p>o. \nIntroducing these relations, we have\n\nHere J, denotes the Bessel function of imaginary argument; thus \nIt follows from (m.4) and the integral equation (208) that\n\nHaving now solved for F= F(t), the current and voltage are gotten \nfrom equations (206) and (207). Thus\n\nHere I)(\u00a2 VY 72?\u2014x\u00b0/v\") is the Bessel function of order 1: thus \u2014iJ,(iz) = \nI,(z). The function is entirely real. The derivation of formula \n(211) is a little troublesome, owing to the discontinuous character of \nthe function F: the detailed steps are given in an appendix.\n\nThe preceding solution depends for its outstanding directness \nand simplicity on the recognition of the infinite integral identity (m \ninto which the integral equation of the problem can be transformed. \nWhen such identities are known their value in connection with the \nsolution of operational equations requires no emphasis. On the \nother hand, we cannot always expect to find such an identity in the \ncase of every operational equation; and, particularly in the case of \nsuch an important case as the transmission equation it would be \nunfortunate to have no alternative mode of solution. Fortunately \na quite direct series expansion solution is obtainable from the oper- \national equation, and this will now be derived. As a matter of con- \nvenience we shall restrict the derivation to the voltage formula\n\nAs a further matter of mere convenience we shall assume that G=4, \nso that \u00a2o=p and (203) becomes \nV =e (203-a) \nwhere \nThe method holds equally well for the current equation (204) and \nfor the general case op. \nWrite (203-a) as\n\nIf the coethcients 3), bs... are evaluated, a simple matter of elemen- \ntary algebra, the foregoing expansion in the retarded time t\u20147 will \nhe found to agree with the solution (211) when @ is put equal to p.\n\nWe shall now discuss the outstanding features of the propagation \nphenomena in the light of equations (210) and (211) for the current \nand voltage. We observe, first, that we have a true finite velocity of \npropagation 7=1 y LC. No matter what the form of impressed \ne.m.f. at the beginning of the line (x=), its effect does not reach the \npoint x of the line until a time f=x v has elapsed. Consequently \nv=x fis the velocity with which the wave is propagated. This is a \nstrict consequence of the distributed inductance and capacity of the \nline and depends only on them, since v=1 LC. It will be recalled \nthat in the case of the cable, where the inductance is ignored, no \nfinite velocity of propagation exists.\n\nThe question of velocity of propagation of the wave has been the \nsubject of considerable confusion and misinterpretation when dealing \nwith the steady-state phenomena. It seems worth while to briefly \ntouch on this in passing.\n\nAs has been pointed out in preceding chapters, the symbolic or com- \nplex steady-state formula is gotten from the operational equation by \nreplacing the symbol p by where 7= \u2014 andw, 27 is the frequency. \nIf this is done in the operational equation (203) for the voltage, the \nsymbolic formula is\n\nV=e \u201cVN \nIf the expression Is separated into its real and im- \naginary parts we get an expression of the form\n\nalong the line with velocity dx dt=v, 8, the phase of the wave will \nremain constant. This is interpreted often as meaning that the\n\nvelocity of propagation of the wave is @ 3. Now since 3 is greater \nthan unity and only approaches unity as the frequency becomes \nindefinitely great, the inference is frequently made that the velocity \nof propagation depends upon and increases to a limiting value \nwith the frequency. This velocity, however, is not the true velocity \nof pr ypagation, which is always but is the veloc ity of phrase propagation \nin the steady-state. This distinction is quite important and failure \nto bear it in mind has led to serious mistakes.\n\nReturning to equation (211) and (210) we see that after a time \ninterval /=x v has elapsed since the unit e.m.f. was impressed on the \ncable, the voltage at point x suddenly jumps from = zero to the \nvalue e \u2019*\u00b0 while the current correspondingly jumps to the value\n\nwhich will be recognized as the steady-state attenuation factor for high \nfrequencies. Similarly y CL is the steady-state admittance of the \nline for high frequencies. The sudden jumps in the current and \nvoltage at time f=y v are called the heads of the current and voltage \nwaves. If, instead of a unit e.m-f., a voltage f(\u00a2) is impressed on the \nline at time \u00a2=0, the corresponding heads of the waves are fioje * \nand VC. L floje for voltage and current respectively. These \nexpressions follow at once from the integral formula\n\nx(t) = f(t\u2014r)h(r)dr \ndt, 0 \n=f(o)h(t)+ | f'(t\u2014r)h(r)dr. \nThe tails of the waves, that is, the parts of the waves subsequent \nto the time \u2018=x v, are more complicated and will depend on the \ndistance x along the line and on the line parameters p and o. The two \nsimplest cases are the non-dissipative line, and the distortionless line \nThe ideal non-dissipative line, quite unrealizable in practice, is one \nin which both R and G are zero. \u2014 In this case p=o=0, and formulas \n(210) and (211) become \nt<x/2,\n\nBoth current and voltage jump, at time \u00a2=.x/v, to their steady values. \nIf an e.m.f. f(t) is impressed on the line at time \u00a2=0, the corresponding \ncurrent and voltage waves are\n\nConsequently the ideal non-dissipative line transmits the waves \nwith finite velocity v, without attenuation or distortion. Such a line \nis, of course, the ideal transmission system.\n\nThe non-dissipative line is, of course, purely theoretical and un- \nrealizable in practice; the disfortionless line is, however, approxi- \nmately realizable, and as the name implies, transmits without distortion \nof wave form. The distortionless line is one in which the line con- \nstants are so related that\n\n=e for x/v. \nFurthermore, if the impressed e.m.f. is f (4), the corresponding current \nand voltage waves are :\u2014 \nI=0 for t<x/v, \naxf(t\u2014x/v) for \nV=0 for i<x/y, \n=e *f(t\u2014x/v) for t2=x/v. \nThe distortionless line, therefore, transmits the waves without dis- \ntortion of wave form, but attenuates the waves by the factor e~**. \nSuch a line is an ideal transmission system as regards preservation \nof wave form, but introduces serious attenuation losses. For example, \nif a line has normally negligible leakage, and leakage is introduced\n\nto secure the condition R/L=G/C, the line is thereby rendered dis- \ntortionless but the attenuation is doubled.\n\nOne of Heaviside\u2019s most important contributions to wire trans- \nmission theory was to point out the properties of the distortionless \nline, its approximately realizable character, and to base on it a correct \ntheory of telephonic transmission.\n\nThe character of the wave propagation when the parameters \np and o are not restricted to special values, can only be roughly in- \nferred from inspection of the formulas, and then only when the prop- \nerties of the Bessel function 7, and J; have been studied. Fortunately \nthese functions have been computed and tabulated for small values \nof the argument, and have simple asymptotic expansions for large \nvalues. It is therefore a simple matter to compute and graph a \nrepresentative set of curves which show the current and voltage \nwaves for various values of p, \u00a2 and x. For this purpose it is con- \nvenient to introduce a change of variables and write:\n\nFigs. (9) to (18) give a representative set of curves illustrating \nthe form of the propagated current and voltage waves for different \nlengths of line, and different values of the line parameters @ and }, \nor p and oa.\n\nThe curves of Figs. (9) and (10) show the current entering the \nline in response to a unit e.m.f. applied at time \u00a2=0. The line is \nassumed to be non-leaky (b)=0) and is computed for two different \nvalues of the parameter a. We see that the current instantly jumps to \nthe value \\/ C/L and then begins to die away, the rate at which it dies\n\nbegins to die away provided x\u00bb and a are such that ax <2. Hf, however, \nwe are considering a point at which av >2, the current begins to rise \ninstead of fall after the initial jump, and may attain a maximum \nvalue very large compared with the head before it starts to die away. \nThis is shown in the curves of Figs. (11), (12) and (13), also computed\n\nfor the non-leaky line (b=0). From these curves we see that, as \nthe length of the line and the parameter a increase, the relative mag- \nnitude of the tail, as compared with the head of the wave, increases. \nFinally when the line becomes very long, the head of the wave be-\n\ncomes negligibly small, and the wave, except in the neighborhood of \nits head, becomes very close to that of the corresponding non-inductive\n\nAn interesting feature of both current and voltage waves is that \nwhen a sutficient time has elapsed after the arrival of the head of the \nwave, the waves become closer and closer to the wave of the cor- \nresponding non-inductive cable; that is, to the cable having the same \nR,C and G. Consequently the inductance plays no part in the \nsubsidence of the waves to their final values.\n\nELECTRIC CIRCUIT THEORY \nCurves (16), (17) and (18) illustrate the voltage wave for several \nconditions. After the arrival of the head, the wave slowly builds \nup to its final value. Curve (1S) represents the case where the line \nis very nearly distortionless, showing how completely the distorting \ntail of the wave is eliminated. \n\u20186 \n12 \n260260 280 300 20 340 360 380 \nFig. 15\u2014Propagated current in line; x =200 \n\u00bb \n2 L 2 ( \nb= VG \nMultiply ordinates by vy C/L.e \nas = \na3 \n| | \n0 | Values of | \n0 2 4 6 8 10 2 \u20184 \u20ac 18 2 \nFig. 16\u2014P voltage fine; WS G \nig. 16\u2014Propagated voltage in line; \u2014-\\ G=0. \n\u2018 So far we have confined attention to the current and voltage waves \nin response to a unit e.m.f. applied at time \u00a2=0 to the line terminals. \nH Of much greater technical importance is the question of the waves \n: In response to a sinusoidal e.m.f. suddenly applied to the line termi-\n\nnals. In order to investigate this important problem it is convenient \nto divide the expressions for the current and voltage waves as given \nby equations (210-a) and (211-a) into two components. We write \nfor\n\nwhere, by definition, J(f) and W(t) are the differences between the \ntotal waves and their heads. The advantage of analyzing the waves \ninto these components is that the distortion of the waves is due to\n\nJ(t) and W(t) respectively, while the first component of (210-b> and \nen \n(211-b) introduce merely a delay. Thus, if the e.m.f IMpressed \nat time 1s the corresponding waves for or r2 x. are \nC \nhe : \nes |= \u00e9 \\ \nto 2] \n| \nd d \nwhere J\u2019(t) =\u2014J(t) and W(t W(t \ndt dt\n\nThe integrals of (212) and (213) can be computed and analyzed \nin precisely the same way as discussed in connection with the non \ninductive cable problem, and are of very much the same character \nas the alternating current waves of the cable. In the total waves, \nhowever, as given by (212) and (213), a very essential difference is \nintroduced by the absence of the first terms, which represent undis \ntorted waves propagated with velocity v7. Thus, if the impressed\n\nNow the first terms of (214) and (215) are simply the usual steady \nstate expressions for the current and voltage waves when the fre- \nquency is sufficiently high to make the steady-state attenuation \nconstant equal to a and the phase velocity equal to e. Furthermor \nthe integral terms become smaller and smaller as the applied fre \nquency w/27m is increased. It follows, therefore, that for high fre \nquencies the waves assume substantially their tinal steady value \nat time \u00a2=x /v, and that the tails of the waves, or the transient \ndistortion, becomes negligible. \u201cThis is a consequence entirely of the\n\npresence of inductance in the line, and shows its extreme importanc\u00ab \nin the propagation of alternating waves and the reduction of transien \ndistortion,\n\nIt should be pointed out, however, that if the line is very long and \nthe attenuation is very high, the integral terms of (214) and (215) \nare not negligible unless the applied frequency is correspondingly very \nhigh. For example, on a long submarine cable, the a.c. attenuation \nis so large that the first terms of (214) and (215) are very small, and \nJ(t) is very large compared with e \u201c. Consequently here \nthere is very serious transient distortion and alternating currents \nare therefore not adapted for submarine telegraph signalling.\n\nThis discussion may possibly be made a little clearer, without \ndetailed analysis, if we recall the discussion of alternating current \npropagation in the non-inductive cable of the preceding chapter. \nKrom that analysis it follows that, when the applied frequency w/27 \nis suthciently high, the integral term of (214) becomes approximately\n\nJ\"(t) \nand the complete current wave is \nsin w(t\u2014x/v)+-\u2014J\u2019(t) (216) \nL w \nand similarly the voltage wave is \ne-**sin w(t\u2014x/v) + W(t). (217) \nNow if the total attenuation ax is large the last terms of (216) and \n(217), before they ultimately die away, may become very large com- \npared with the first terms, which represent the ultimate steady-state. \nAppendix to Chapter V/I, Derivation of Formula (211)\n\nThe only troublesome question involved in deriving (211) from \n(207) and (209) is that we have to differentiate with respect to x, in \naccordance with (207), the discontinuous function F(t). To accom- \nplish this we write (209) in the form\n\nwhere $(f) is defined as a function which is zero for t<x/v and unity \nfor t2x/v. Clearly this is equivalent to (209) and permits us to deal\n\nwith F(t) as a continuous function. Now, in accordance with (207 \nperform the operation of differentiation upon (209-a): we get\n\nand that the whole contribution to the integral occurs at t=x \nWith these points clearly in mind, the expression\n\nCHAPTER VIII \nPROPAGATION OF CURRENT AND VOLTAGE IN: ARTIFICIAL \nLINES AND WAVE FILTERS \nThe artificial line here considered is a periodic structure, com- \nposed of a series of sections connected in tandem, each section con- \nsisting of a lumped impedance 3; in series with the line, and a 'umped\n\nimpedance z, in shunt across the line. In the artificial line which \nwe shall consider it will be assumed that the voltage is applied at \nthe middle of the initial or zeroth section, as shown in Fig. 19. This \ntermination is chosen because of its practical importance, and be-\n\ncause also of the facet that the mathematical analysis is simplified \nthereby. Furthermore any other termination can be regarded and \ndealt with as an additional terminal impedance, so there is no essen \ntial loss of generality involved.\n\n1. The artificial line is often used as a model of an actual trans \nmission line and it is therefore of importance to determine theoretical, \nthe degree of correspondence between the two.\n\n2. The solution for the corresponding transmission line with con \ntinuously distributed constants is derivable from the solution for the \nartificial line by keeping the total inductance, resistance, capacity \nand leakage constant or finite, and letting the number of sections \napproach infinity.\n\n3. The artificial line is very closely related, in its properties and \nperformance, to the periodically loaded line, and its solution is, to \na first approximation, a working solution for the loaded line.\n\n1. The structure is of great importance in its own right, and when \nthe impedance elements are properly chosen, constitutes a \u201cwave \nfilter.\u201d\n\nWe shall now derive the operational and symbolic equations which \nformulate the propagation phenomena in the artificial line. Let /, \ndenote the mesh current in the ath section of the line; 7,\u2014, the mesh \ncurrent in the (z7\u20141)\"' section, ete. Now write down the expression \nfor the voltage drop in the #'\" section; in accordance with Kirch- \nhott's law we get:\n\nential equation, but the method of solution is essentially the same. \nWe assume a solution of the form\n\nI, =Ae-\"l + Berl (219) \nwhere A, B and VT are independent of , and substitute in (218). \nAfter some simple rearrangements we get\n\nEquation (218) is clearly satistied by the assumed form of solution, \nand furthermore leaves the constants 4A and B arbitrary and at our\n\nwhere 42:. \nNow by reference to equation (219) it is easily seen that Pois the \npropagation constant of the artificial line, precisely analogous to the\n\npropagation constant y of the smooth line. In terms of 1 \npedances 2; and s., the propagation constant of the artificial line ts \ndetermined by (221). This equation may either be regarded as an\n\noperational equation or a symbole equation, depending on whether \nthe impedances are expressed in terms of the operator Pp or in terms \nof tw, where w is 27 times the frequency.\n\nNow return to equation (219) and Jet us assume that the line ts \neither infinitely long, or, what amounts to the same thing, that it ts\n\nclosed by an impedance which suppresses the retlected wave. We \nassume also that a voltage V. is impressed at mid-series position of the \nzero\u201d section (w=). Equation (219) becomes\n\nwhere K is the characteristic impedance of the artificial line (at mid \nseries position). Hence by (223) and (222)\n\nBefore proceeding with the operational equation, and the investi- \ngation of transient phenomena in artificial lines, it will be of interest \nto deduce from the foregoing the unique and remarkable properties \nof wave filters in the steady state. For this purpose we return to \nequation (221)\n\nNow suppose that the series impedance 2; is an inductance L and \nthe shunt impedance zz a capacity C, so that, symbolically,\n\ncos 0=1\u20143 wLC (227) \n0. Consequently if \n4 is a real quantity the ratio of the absolute values of the currents in ad-\n\njacent sections ts unity, and the current is propagated without attenuation.\n\nInspection of equation (227) shows that @ is real provided the right \nhand side lines between +1 and \u20141: or that w lies between 0 and \n2/-\\y LC. Consequently this type of artificial line transmits, in the \nsteady state, sinusoidal currents of all frequencies from = zero to \n1 ry LC without attenuation. It is known as the low-pass filter.\n\nIf we invert the structure, that is, make the series impedance 2, \na capacity C and the shunt impedance 22 an inductance L, so that\n\nThis type of artificial line transmits without attenuation currents \nof all frequencies for which the right hand side of (22S8-a) lies between \n+1 and \u20141; that is, all frequencies from infinity to a lower limiting \nfrequency 1/47 LC, while it attenuates all frequencies below this \nrange. It is known, on this account, as the high-pass filter.\n\nIt is possible by using more complicated impedances to design \nfilters which transmit a series of bands of frequencies. We cannot, \nhowever, go into the complicated theory of wave filters here, which \nhas been covered in a series of important papers. One point should \nbe noted, however: transmission without attenuation implies that \nthe impedance elements are non-dissipative. Actually, of course, \nall the elements introduce some loss, so that in practice the filter \nattenuates all frequencies. Careful design, however, keeps the \nattenuation very low in the transmission bands.\n\nWe shall now derive the indicial admittance formulas for some \nrepresentative types of artificial lines and wave filters from the oper- \national formula\n\nWe start with the so-called low-pass filter on account of its sim- \nplicity and also its great importance in technical applications. This \ntype of filter consists of series inductance L and shunt capacity C \nThe general case which includes series resistance R and shunt leakage \nG has been worked out (see Transient Oscillations, Trans. A. 1. ke. k., \n1919). The solution is, however, extremely complicated and will not \nbe dealt with here. We shall, instead, consider the important and \nilluminating case where the series and shunt losses are so related\n\ns) that the preblem is reduced to the solution of the operational \nequation for <1 Writing w,=2/+/LC, we have\n\nNZ V/1+(p/w.)? V1+(p p \n\\ L +. 2 \nNow refer to formula (72) of the table of integrals; writing y L C=k, \nwe see by Theorem V that\n\n/ / ( \nike, to a unit e.m.f. applied te the initial section (2 =0 From (232 \nand (234) it follows at once that \n+. | dre ] ri \nR So Jo \nIntegrating the second member by parts and noting that .1.(0) =0 \nthis reduces to \n231) 1 \\ \nAy | e Ge\u2019 \nk, \nwhich is the indicial admittance formula for the quasi-distortionless \nlow-pass filter, or artificial line.\n\nBefore discussing these formulas, it is of interest to derive the \nformula for A) by power series expansion. Formula (233) can be \nwritten\n\n) l l \nk Pp \\ Il+(w p +4 p \nThis can be expanded in a series in inverse powers of Pp; thus \n2) 2n+2 /w 2 \nAe ' ( ) \n32) 271! \\p, \n(2n+3) (2n+4 \nnal \\ p \nReplacing 1p\" by fa! in accordance with the Heaviside Rule we get \nLIN 2 L'(2n+3 2 \n2!(2n-+5)! 2 \nk, ; Phis can be recognized as the power series expansion of (254 \nThe artificial cable is also of interest and practical importance \nIn this structure the series impedance is a resistance R and the shunt\n\nFor large values of \u00bb and \u00a2 this series is difficult to compute or in- \nterpret. It can, however, be recognized as the series expansion of the \nfunction\n\nwhere J,,(2t/RC) is the Bessel function J, of order \u00bb and argument \n(2\u00a2/RC). This solution, it may be remarked, can be derived directly \nby a modification of the integral formula (7).\n\nIt is beyond the scope of this paper to consider other types of \nartificial lines and wave filters; for a fairly extensive discussion the \nreader is referred to \u2018Transient Oscillations in Electric Wave-Filters,\u201d\u2019 \nB.S. T. J., July, 1923. The low-pass wave filter, however, both in \nits own right and on account of its close relation to the periodically \nloaded line, deserves further discussion.\n\n(\u2018wt \n{\u00b0 Jon(r) dr (234) \nk, 0 \nwhile for the quasi-distortionless low-pass wave filter \nA | e AT Jo,(r)dr (235)\n\nComputation and analysis of these formulas involve an elementary \nknowledge of Bessel functions. The properties necessary for our \npurposes are briefly discussed in an appendix to this chapter.\n\nFig. 21\u2014Low pass wave filter. Indicial admittance of third section (1 =2 \nMultiply ordinates by vC/L\n\nValues of wt \nFig. 22\u2014Low pass wave filter. Indicial admittance of fifth section (m=4). \ng I\n\nthat is, the current in response to a steady unit e.m.f. applied at \ntime \u00a2=0, are shown in the curves of Figs. 20, 21 and 22, for the \ninitial or zeroth, the 3rd and the Sth sections, respectively. These\n\nabove are sufficient to give a reasonably comprehensive idea of the \ngeneral character of these oscillations and their dependence on the \nnumber of sections and the constants of the tilter.\n\nIt will be observed that the current is small until a time approxi- \nmately equal to 2a w =nyLiC, has elapsed after the voltage is \napplied. Consequently the low-pass filter behaves as though cur- \nrents were transmitted with a finite velocity. of propagation w, 2 \n1 y LyCy sections per second. This velocity is, however, only ap- \nparent or virtual since in every section the currents are actually \nfinite for all values of time >0.\n\nAfter time (=n 1,02 has elapsed the current oscillates about the \nvalue Tok with increasing frequency and diminishing amplitude. \nPhe amplitude of these oscillations is approximately\n\nand their instantaneous frequency (measured by intervals between \nzeros)\n\nThe oscillations are therefore ultimately of cut-otf or critical frequency \nw 2r in all sections, but this frequency is approached more and more \nslowly as the number of filter sections is increased.\n\nFigs. 238, 24. 25, give the indicial admittance in the 100th, 500th \nand 1000th section of the low-pass wave filter. The filter itself seldom\n\nhe embodies more than 5 sections Phe case of a large number of se \nhe tions is of interest, however, because 11 represents a first: approxi\n\n12 +\u2014\u2014\u2014+\u2014 \nip- \nle. \na6 + + + \u2014 \nen \n' | | \n+ \n960 980 1000 1020 5 00 0 40 160 \nFig. 24\u2014-Low pass wave filter Indicial admittance f 500th se $09 \nMultiply ordinates by L \nCy \n\u2014 \nto-\u2014 \nth \nmM \na8 t + \nnear H \n| \n| \n| | \n02 \n| | \n| \n} | | \n960 1980 2000 2020 2040 2060 2080 2120 2140 2160 \nFig. 25\u2014-Low pass wave filter. Indicial admittance of 1000th section (2 = 999 \nMultiply ordinates by vc L \nline is ideal and unrealizable, its study is of practical importance \nbecause in this type of line the effect of the discontinuous character \n: of the loading of the periodically loaded line is isolated and exhibited\n\nThe dotted curves represent the current in the corresponding \nsmooth line. For the smooth line, the current, as we have seen, is \ndiscontinuous, being identically zero for a time v=\u201d and having an \ninstantaneous jump to its final value \u00a5 \u2018C/L at vt=n. The current \nin the artificial or periodically loaded line differs from that in the cor- \nresponding smooth line in three important respects: (1) the absence \nof the abrupt discontinuous wave front, (2) the presence of super- \nposed oscillations, and (3) the absence of a true finite velocity of propaga- \ntion. It will be observed, however, that the current in any section \nis negligibly small or even sensibly zero until v=, so that the current \nis propagated with a virtual velocity 1/+/LC per section. The pres- \nence of a well marked wave front is also evident although this is not \nabrupt, as in the smooth line. The effective slope of the wave front \nbecomes smaller as the current wave travels out on the line, decreasing \nnoticeably as the number of sections is increased. When the number of \nsections becomes large, however, the decrease in the slope is not rapid, \nbeing in the 500*\" section about 60 per cent. of that in the 100\u00b0\" section.\n\nThe superposed oscillations are of interest. These are initially \nof a frequency depending upon and decreasing with the number of \nsections, 2, but in all sections ultimately attaining the frequency\n\nwhich is the critical or cut-off frequency of the line, above which \nsteady-state currents are attenuated during transmission and below \nwhich they are unattenuated. When vt is large compared with \nthe amplitude of these oscillations becomes ./1/zvt so that they \nultimately die away and the current approaches the value 1/C/L \nfor all sections. The current in the loaded line is thus asymptotic \nto the current in the corresponding smooth line and oscillates about \nit with diminishing amplitude and increasing frequency.\n\nSince the abscissas of these curves represent values of 2v\u00a2=2t/1/LC, \nand the ordinates are to be multiplied by C/L to translate into \nactual values, the curves are of universal application for all values \nof the constants L and C.\n\nThe investigation of the building-up of alternating currents in wave \nfilters and loaded lines is very important. It depends for the non- \ndissipative case on the properties of the definite integrals\n\nELECTRIC CIRCUIT THEOR) 93 \nwhere w=w/w, and w=2z times the applied frequency. The mathe-\n\nmatical discussion is, however, quite complicated and will not be \nentered into here. The reader, who wishes to follow this further, \nis referred to Transient Oscillations, Trans. A. I. E. E.. 1914 and \nTransient Oscillations in Electric Wave Filters, B.S. T. J., July, 1923.\n\nThe Bessel Functions of the first kind, J..(x) and J,,(x). are detined, \nwhen # is zero or a positive integer, by the absolutely convergent \nseries\n\n+546 (In + \n2.4.6.(2n+ 2) (2n+4)(2n+6) \nIn the following discussion of the properties of these functions it will \nbe assumed that the argument x is a pure real quantity. \nFor large values of the argument (x large compared with 7), the \nbehavior of the functions is shown by the asymptotic expansions: \u2014\n\nWhen the argument ts less than the order (0Sx<n) the function \nis very small and positive, and is initially zero (except when a=0). \nIn the neighborhood of x=\u201d, the function begins to build up and \nreaches a maximum a little beyond the point x=. Thereafter the \nfunction oscillates with increasing frequency and diminishing ampli- \ntude, and ultimately behaves as\n\nThis approximate formula is valid only where x>n, its accuracy \nincreasing with x and with w. For all orders of m it is quite accurate\n\nBUNDANT evidence of many kinds exists to show that each and \nevery distinct sort of atom is especially adapted to possess \nenergy, notin any random quantity whatsoever, but in certain peculiar, \ndetinite, characteristic amounts. An atom having energy in one of \nthese particular amounts apparently cannot add arbitrary quantities \nto its store, nor yield up arbitrary quantities from it; whenever the \natom receives or whenever it gives energy, it receives or gives only \njust so much as is exactly sufficient to raise or reduce its) supply \nto some one among the others of these distinctive quotas. For \neach of the chemical elements there is a great system of these dis- \ntinctive energy-values. They are determined chiefly by analyzing \nspectra, and for most of the elements\u2014the exceptions being those \nof which the spectra are excessively complicated\u2014\u2014many of them \nhave been evaluated very accurately and set down in tables. \nThe system of distinctive energy-values for any element is a very \nimportant feature of that element; perhaps, indeed, the most im- \nportant feature of all.\n\nIt is customary to say that when an atom acquires or surrenders \nenergy, it passes from one into another state; the various states cor- \nresponding to its various distinctive energy-values are called its \n\u201cStationary States.\u201d\u2019 This is a name which suggests, and is doubtless \nmeant to suggest, that the energy-value of the atom is but one among \nmany of its features, all of which change when the energy-value \nchanges. This is a legitimate idea; theorizing about the atom consists \nin speculating about just such features. But the reader will go far \nand grievously astray if he lets the name signify to him that many \nof them are directly and definitely known. In some few cases there is \ngood reason to believe that we know the magnetic moment of an \natom in its normal state. Beyond these the energy-values are all \nthat are known. If the reader chooses everywhere to replace \u201c\u2018Sta- \ntionary State\u2019 by \u201cenergy-value\u2019\u201d he will be holding fast to the \nphysical reality, to the one thing not liable to be compromised by the \nfuture trends of thought.\n\nAn atom may pass from one Stationary State to another because \nof colliding with an electron or another atom of the same or a different\n\nkind; or by absorbing radiation it may pass from one Stationary \nState to another of higher energy; or it may pass spontaneously\n\nthe \ndifference \\U between the energy-values of the initial Stationary State \nand the final one by the equation\n\nThe same equation governs the last case but one. in which it con- \nnects the frequency of the absorbed radiation with the enerev-ditfer- \nence between the two Stationary States from and into which the atom \npasses. On this equation is founded the method of analyzing spectra \nwhich is the most accurate and most widely applicable method. of \ndetermining the energy-values of Stationarv States. The other \nways in which atoms are caused to pass from one State to another \nlead to methods of determining these states, which are almost useless \nfor accurate measurements, but invaluable as controls.\n\nThe energy-values of the various Stationary States of an atom \nare interrelated, and sometimes it is possible to express a long sequence \nof them by means of a simple or a not very complicated formula. \nThere are also interrelations between the distinctive enerev-values \nfor different elements; and this statement is meant to apply also to \natoms from which electrons, one or more, have been detached, which \nshould be considered as distinct though not as stable elements. There \nare unmistakable numerical relations among the Stationarv States \nwhich come into being when atoms are subjected to electric or to \nmagnetic fields. Finally there is the important principle that the \nspontaneous transitions between various pairs of states, which re- \nsult in spectrum lines, do not occur equally often; and vet the rela- \ntive oftenness or seldomness of their occurrence is itself regulated \nby laws. One finds instances in which transitions from a state v4 \nto a state B, are just twice as common as transitions from A to a state \nBy close to By. One finds instances in which transitions from a state \nA toa state B do not occur at all under usual conditions and an atom \nin state A cannot get into state B without touching at some state C \nfrom which A and B are both accessible. It is possible so to arrange \nthe Stationary States of an atom that by looking at the situations \nof any two states in the arrangement, one can tell immediately whether \ndirect transitions between them do occur or do not: and this arrange- \nment is found to be suited to, even to be demanded by the numerical \ninterrelations to which I alluded above. Upon these facts the classi-\n\nfications of the Stationary States are founded, and the notations | \nwhich they are named.\n\nThe atom-model to which this article is devoted, the atom-medel \u00ab | \nRutherford and Bohr, is designed to interpret these facts of t]\n\nterpret. certain experiments \u2014chiefly, though not altogether, exper \nments on the deflections suffered by minute flying charged particle \nwhen they pass through matter \u2014 which indicate that an atom consist- \nof a positively-charged nucleus with a congeries of electrons around it \nSpecifically, the results of thes\u00e9 experiments agree with the notic: \nthat the Vth element of the Periodic Table consists of a nucleus wit! \npositive charge .Ve and .V electrons surrounding it; and this is the \nsimplest and most satisfying notion with which they do agree. Yer \nthere is something paradoxical about this atom-medel; for electrons \ncould neither stand still nor vet revolve permanently in orbits around \na nucleus, if they conformed to the laws of electrostatics. Also \nthere must needs be something paradoxical about any attempt to \ninterpret the Stationary States by this model, for there is nothing \ninherent in it to make any energy-value preferable to any other \nUnder these circumstances Bohr's procedure was, resolutely to accept \nboth paradoxes at once, and to say that the electrons can revolve \npermanently in those and just those particular orbits, whereby the \nenergy of the atom assumes the particular values which are those of \nthe observed Stationary States. This is easy to sav; but it is not \nimportant, unless one succeeds in showing that those and only those \nparticular orbits are set apart from all others by some peculiar feature, \nare distinguished by conforming to some particular principle, which can \nbe exalted into a \u201cLaw of Nature\u2019 to complement or supersede the \nlaws of electrostatics. Otherwise the atom-medel would be of no value \nThus in order to make the test of the atom-model, it is necessary \nto trace these orbits. One is confronted with this problem of orbit- \ntracing: Given: the observed energy-values of the Stationary States; \nrequired: to trace the orbits such that, when the electrons travel in \nthem, the energy of the atom has these observed values. If this \nproblem cannot be solved, it is impossible to take the next and es- \nsential step of ascertaining whether these particular orbits are dis- \ntinguished in any particular way from all the other conceivable ones. \nIn the case of a single electron revolving about a nucleus, this \nproblem is sometimes soluble. Tf the mass of the electron is regarded \nas invariable, and no outside influences are supposed to act upon the \natom, then the solution is comparatively easy to attain. It was \nperformed in the Second Part of this article. If an external magnetic\n\nfield is superposed, the problem is scarcely more ditheult: if an external \nelectric field is superp sed, itis ditheult but soluble provided ilwavs \nthat the mass of the electron be supposed invariable It the mass of \nthe electron Is supposed to vary with its =peed as the theory of rela \ntivity. requires, and as certain experiments suggest, the problem \nremains soluble provided that no outside influences act For all \nthese cases the orbits which vield the observed energy-values have \nbeen traced; and certain features have been shown to be common \nto all of these \u201cpermitted\u201d orbits, and to no others, so that by these \nfeatures the permitted orbits are sel apart from all the rest Inve rsely, \nanvone who is told what these features are, and who is sutticientls \nadept in dynamics, can trace all the orbits which display them and \ncalculate the energy-values for these orbits and so predict the energy \nvalues of all of the Stationary States of an atom consisting of a nucleus\n\nand a single electron. Such orbits are known as quantized orbits \nThe rules whereby they are set apart from all the multitude of orbits \nnot permitted are the Quantum Condtitons; which some one, it is to \nhe hoped, will some day succeed in deriving from a general Principle \nof Quantization.\n\nThe most general way of phrasing these conditions is ditheult to \ngrasp, and the more intelligible ways are not the most general. The \nmost general conditions yer formulated are not adequate for all \ncases; the completely adequate principle is vet to be discovered \nkor the purposes of this summary and of most of what follows, a \nvery limited expression of the Quantum Conditions will be sufficient. \nIn the Second Part of this article it was proved that the permitted \norbits of an electron of invariable mass revolving in an inverse-square \nfield such as is supposed to surround a bare nucleus, are certain \nellipses. Tt was further stated, without proof, that if the electron ot \ninvariable mass revolves in a central tield which deviates slightly \nfrom an inverse-square field, then the permitted orbits are certain \n\u201crosettes\u201d or precessing ellipses each orbit may be traced by im \nagining an ellipse revolving steadily in its own plane around the source \ncf the central field (IT will sav the nucleus) at one of its foci. All \nthe orbits are resettes; the permitted orbits are certain rosettes which \nare distinguished from the others by a distinctive feature. One ways \nof expressing this feature involves the angular momentum pp, and \nthe radial momentum p, of the electron. In terms of the mass \nof the electron, its distance \u00a2 from the nucleus, and its angular velocity\n\n4 \nSOME CONTEMPORA \nSOME CONTEMPORARY DPA \n; \nicles \nTit \nti \nvit] \n1} \n\u201cOns \nund \nto \nher \n\u2018ept \nthe \nnot \n1OSse \nthe \nlue \nx \nthis \nNes. \nthis \nded \n4 \nWas \na\n\nThe \u201cprinciple of quantization\u201d is, that the permitted orbits ar \nmarked out from all others in that they fulfil these conditions, whic \nare the Quantum Conditions:\n\nIn these equations each integral is taken around one complete cycl' \nof the corresponding variable; # stands for Planck's constant, and \nn and k take the values of all positive integers, k never surpassing 1\n\nThere is an alternative way of phrasing these quantum-conditions, \nwhich is much easier to visualize; but it emphasizes what are probably \naccidental features of the permitted rosettes, rather than fundamental \nones. The rosettes are, as I have said, precessing ellipses; the major \naxes 2a and the minor axes 2) of these ellipses are, for the permitted \nrosettes\n\nin which # and k& take as before the values of all positive integers, \nk not surpassing 7.\n\nExactly the same principle governs the permitted orbits of | an \nelectron revolving in a perfect inverse-square central field, but varying \nin mass when its speed varies, as the theory of relativity requires \nIn this case also the orbits are rosettes, and the permitted orbits \nare particular rosettes set apart from all the others in that they fulfil \n(2) and (3), therefore automatically (4) and (5). The energy-values \nof these permitted orbits agree closely with those of the observed \nStationary States of hydrogen and of ionized helium, the atoms of \nwhich are the only atoms believed to consist of a nucleus and one \nelectron. Inversely, the orbits required to interpret the observed \nStationary States are set apart from all the other conceivable orbits \nby the features expressed by (2) and (8), and by (4) and (5). On \nthese close numerical agreements for hydrogen and ionized helium, \nand on other numerical agreements for the same atoms arising when \nexternal fields are applied, the prestige of Bohr\u2019s atom-model is \nfounded.\n\nThe integers 7 and k, the total quantum number and the azimuthal \nquantum number, are used as indices to symbolize the various Station- \nary States of hydrogen and ionized helium to which they correspond. \nThus the symbol \u20188,\u201d stands for the Stationary State of either atom,\n\nrosette, or precessing ellipse, for which m=3 and k=2. Orbits for \nwhich k=\u201d are circles; orbits for which <7 are \nand the farther k falls below \u00bb the more eccentric (narrower) is the \nellipse, although its major axis is independent of k. [I have repeated \nall of these statements about precessing ellipses and their quantum- \nnumbers, because a great part of the speculation about atoms pos- \nsessed of more than one electron consists of persevering and obstinate \nattempts to interpret their behavior by as nearly as possible the same \nideas.\n\nIt is essential to remember also that all the enerey-values of the \nStationary States are reckoned from the \u201cState of the Lonized Atom,\u201d \nin which State the energy\u2014i.e., the energy of a system composed of \none atom deprived of an electron, and one electron far away-\u2014is \nequated to zero.\n\nN. INTRODUCTION TO THE SPECULATIONS ABOUT ATOMS \nWITH MORE THAN ONE ELECTRON\n\nAll atoms, except those of hydrogen and ionized helium, possess \nmore than one electron. There is much evidence of various kinds \nfor this assertion; and certainly the spectra of these other atoms \ncannot be interpreted as those of the first two have been. Thus we \nare confronted with the problem of a system composed of a nucleus \nand more than one electron. The similarities between the spectra of \nhydrogen and ionized helium, and those of other elements, are im- \nportant enough to make it desirable to use the same sort of explana- \ntion. We imagine the various electrons, when there are two or more, \neach to describe certain permitted orbits, set apart from the multi- \ntude of other conceivable orbits by peculiar features expressible by a \nPrinciple of Quantization.\n\nHere at the outset we meet with the great hindrance to success \nin this problem. It is not possible to determine what features are \ncommon to permitted orbits, for it is not possible even to trace the \npermitted orbits. The general problem of tracing the paths of three \nor more bodies, attracting or repelling one another according to the \ninverse-square law, remains unsolved. Considering that for cen- \nturies the related but simpler problem of celestial mechanics has been \nunder continual and powerful attack, the general problem may fairly \nconfidently be regarded as insoluble. There is very little hope of ever \ndominating it to such an extent, that the spectra of atoms with two \nor more electrons can be interpreted exactly by Bohr\u2019s atom-model,\n\nor can be used as strong support to that theory. Tf those two spectra \nof hydrogen and ionized helium were unknown, it is unlikely that \nthe atom-medel would ever have been suggested; it is more than \nunlikely that the atom-model could ever have been regarded as \nsatisfactory. To this day the prestige of the atom-model results \nalmost entirely from its achievement with those two spectra.\n\nWhy then trouble with applying it to the interpretation of other \nspectra? Several good reasons can be given. For instance, it may \nbe that a svstem of several electrons about a nucleus acts in some \nrespects as a unit that its motion can be considered in some ways \nas the motion of a rigid body, that principles of quantization can be \nfound for the system as a whole, similar to the principles used for \nquantizing the smaller and vet perhaps not more consolidated system \nwhich a single electron is. Here and there in the discussion we shall \nfind indications that this way of thinking is suitable.\n\nAgain, one is justified in arguing that if in simple cases a certain \nlaw is proved, and if in complex cases neither that nor any other law \ncan be proved nor disproved, than we should assume that the law \nproved for the simple cases extends over the complex ones. Few \nevents in this world take place under such conditions that conserva- \ntion of energy can be proved to prevail during them; vet, from the \nfact that conservation of energy has been verified in whatever events \nit has been tested, we do not hesitate to infer that it prevails in all. \nBohr\u2019s model having been so strongly fortified by the data for the \nonly two atoms for which it can be completely tested, why not assume \nit for the others?\n\nAnd tinally, there is the point that many of the data of experiment \nare almost universally expressed in terms of the model, so that the \nphysical literature of today is almost) incomprehensible without \nsome knowledge of it. Unfortunate as this is, it shows that the model \nis a Valuable aid for visualizing the facts. This justifies any model; \nbut must not be construed as evidence for it.\n\nIt will be expedient to divide the subject substantially under these \nfollowing headings.\n\n(a) Lhe IHelium Atom. This, as the case of an atom composed \npresumably of a nucleus and two electrons, comes nearest to being \namenable to calculation. Certain) mechanically possible orbits of \nthe two electrons, possessing the peculiar features of the \u201cpermitted\u201d \norbits of a single electron revolving around a nucleus, have been \ntraced and their energy-values calculated. Not one of them has \ngiven the observed energy-value of a Stationary State of the helium \natom. It is the consensus of opinion that whatever the features\n\nwhich distinguish the permitted orbits may be. they are not those \nwhich prevail in the hydrogen atom.\n\n(b) Alkali- Metal Atoms. Vor these there is reason to believe that \none electron is normally located far bevond all the others, and may \nbe supposed to revolve around a \u201cresidue\u201d consisting of all the others \nand the nucleus. At a ereatl distance, the held due to this residue\n\nof charge +e,\u2014a hydrogen nucleus; for. from a great distance, the \nnucleus and the electrons of the residue will seem almost to coincide \nin place. Nearer in, the forces due to the electrons of the residue \nmay be suppesed to compound with that due to the nucleus in suc! \na way that a central field, not varving as the inverse square, results \nThus rosette orbits may be expected (for this reason | quoted the \nprinciple of quantization for such orbits in Section M). An enormous \namount of effort has been spent in constructing central tields, such \nthat the rosette orbits obeving the quantum-conditions (2) and (3 \nhave nearly the energy-values which the Stationary States of thes \natoms are known to possess. Always, the emission of a spectrum \nline is supposed to result from a transition of the outermost or valence \nelectron from one orbit to another, the electrons of the residue being \nscarcely or not at all affected. Such is the general explanation for the \nfar-reaching and yet imperfect resemblance of the spectra of these \nmetals to that of hydrogen.\n\n(c) Other Elements. As one passes across the Periodic Table \nfrom left to right along any row, the spectra rapidly lose resemblance \nto the hydrogen spectrum. This is taken to mean that the assump \ntion used for alkali metals -the assumption that one electron lies far \nbeyond the others, and executes transitions while the others remain \nunaffected \u2014 departs progressively further from the truth. Evidence \nexists that simultaneous transitions of two electrons occur, and vers \nlikely yet more drastic rearrangements taking place ex blo:\n\n(d) Building of Atoms by Consecutive \u2018Binding\u2019 of Electrons. An \natom composed, when complete, of V3 electrons arranged about t \nnucleus bearing the charge 4+-Ze, may have been formed originally in \nZ stages by the consecutive advent of Z electrons, the first annexing \nitself to the bare nucleus, the second joining itself to the svstem com \nposed of the nucleus and the first, and so on until as many have \narrived as the nucleus is able to hold. Each of these stages should \nbe accompanied by the emission of lines belonging to a particular \nspectrum; the ordinary hydrogen spectrum accompanies the forma \ntion of a hydrogen atom by the step-by-step binding of an electron \nto a nucleus of charge e, the ionized-helium spectrum accompanies\n\nthe joining of the first electron to a nucleus of charge +2e, the neutral \nhelium spectrum the adhesion of the second electron to such a nucleus \nSpectra corresponding to the latest four or five stages, in the forma \ntion of atoms having many electrons when completed, have been \nobserved. To a certain extent, but not entirely, an atom with Z \nelectrons and a nuclear charge Z resembles an atom with Z electrons \nand a nuclear charge Z+1. Toa certain extent, therefore, each atom \nin the Periodic System may be regarded as resembling the last stage \nbut one in the formation of the next following atom. This fact is \nimportant in the interpretations of the Periodic Table.\n\n(e) We next take account of the fact that the sequences \nof Stationary States, mentioned in the elementary theory and descrip- \ntion of spectra, are actually sequences of groups of Stationary States; \nand inquire what may be supposed to differentiate the several states \nof a group from one another. An elaborate formal theory is based \non the assumption that all of the electrons of what I have called the \n\u201cresidue\u201d of the atom revolve, if not literally as a rigid block, at least \nwith a resultant angular momentum which itself is quantized; and \nthat the outermost electron revolves in its own orbit around. this \nresidue, the different Stationary States of the group differing from \none another in respect of the inclination of the orbit to the axis of \nrotation of the residue. The theory is not quite coherent with what \nhas gone before; and for that reason the reader should try to separate \nits essential qualities from its accidental ones.\n\n(t) dagnetic Properties of Atoms. A magnetic field should treat \na system of electrons revolving around a nucleus in the same way as \nit treats one electron, as was said in the Second Part of this article. \nOne would expect that in this case, if in any, the behavior of complex \natoms would resemble that of the hydrogen atom; yet there is a \nstriking and inexplicable contrast. This, like the spectrum of the \nhelium atom, shows that either the quantum conditions governing the \nhydrogen atom are not universal, or the expressions hitherto found \nfor the quantum conditions are too limited. From the responses of \natoms to magnetic fields something is learned about the magnetic \nproperties of atoms and their residues, some part of which can be \ntested by direct experiment; and these experiments include what are \nprobably, all things considered, the most perplexing and fascinating \nones of recent years.\n\n(g) Interpretation of X-ray Spectra. X-ray spectra are analyzed \nas other spectra are, and each absorption and each emission of an \nX-ray by an atom is associated with a transition between two Station- \nary States; these \u2018X-ray Stationary States\u2019\u2019 however are distin-\n\nguished from the others, by the circumstance that every one of them \ninvolves the absence of an electron from the atom; consequently they \nmay be described as Stationary States of an atom-residue. There is \nreason to believe that each distinct State involves the absence of a \nparticular one, or of one out of a particular group, of the electrons \nbound to the nucleus during the earlier stages of the imagined building \nof the atom by successive \u201cbinding\u201d of electrons. The speculations \nabout X-ray spectra consist largely in attempts to correlate the \nindividual States with absences of particular electrons.\n\nhelium atom\u201d as van Vleck calls it, is one of the most tantalizing in \ncontemporary physics. One feels confident a priori that the same \nquantum conditions as suffice so beautifully to constrain the one; \nelectron atom to vield the hydrogen spectrum should also. suffice, \nwhen applied to the orbits of two electrons, to vield the spectrum of \nneutral helium. Yet the various pairs of orbits conforming to these \nquantum conditions, which have already been traced, have been \nshown (with vast expenditure of intellectual labor by some of the \nablest mathematical physicists of our time) to entail energy-values \nfor the Stationary States which are hopelessly incorrect.\n\nFor instance, one might assume that when the helium atom is in \nits Normal State, the two electrons are revolving in a common circular \norbit about the nucleus, being at each instant located at opposite \nends of a diameter; and that this permitted orbit is determined by \nthe condition that the angular momentum of each electron, or per- \nhaps that of both together, is h 27. This seems an obvious general- \nization of the Quantum Conditions for hydrogen; but it vields a \nfalse energy-value for the Normal State; and there is nothing more \nto be said. Kemble and van Vleck demonstrated that no arrange- \nment in which the two electrons are symmetrically placed relatively \nto a line through the nucleus entails the proper energy-value for \nthe normal state. This still leaves open the possibility that the two \nelectrons are unsymmetrically placed-\u2014\u2014a possibility which to some \npeople seems repellent enough to be excluded. Born and Heisen- \nberg calculated the energy-values corresponding to pairs of orbits \none of which lies far beyond the other at all points, and both of which \nare concordant with the Quantum Conditions. These ought to \nhave agreed with the energy-values of the Stationary States which \nare remote from the normal state and near the state of the ionized\n\natom; but they did not. This result is commonly regarded as the \nstrongest evidence for the belief that the Quantum Conditions valid \nfor an atom with one electron are not valid for an atom with two.\n\nThe atom-medel favored by Kramers, and hence presumptively by \nBohr, to represent a helium atom in its normal state, involves two \nelectrons moving in orbits which are not coplanar nor even plane. \nPlanes tangent to the two surfaces upon which the orbits are traced \nintersect each other at 120\u00b0 along a line passing through the nucleus, \nand the electrons pass simultaneously across this line at opposite \ncrossing-points. \u201cThese orbits conform to the Quantum Conditions; \nand the resultant of the angular momenta of the two electrons, which \nis the angular momentum of the entire atom, is equal to # 27. This \natom-model likewise fails to have the right energy-value for the \nnormal state.\n\nThe alkali metals (lithium, sodium, potassium, rubidium and \ncaesium) are elements of which the atoms are easily deprived of a \nsingle electron apiece; one electron of each atom is, as the phrase \ngoes, exceptionally \u201cloosely bound.\u201d Many facts combine to indi- \ncate this; for instance, each of these elements enters with violence \ninto chemical combinations, and the compounds which each forms are \nsuch as to suggest that its atom yields up one electron to the atom or \natoms which join with it. Again, when a salt of one of these metals \nis dissolved, the molecules split up and the atoms of the metal are left \nwandering around in the solvent minus one electron apiece, while \nthe atoms of the other element each hold on to one captured electron. \nMore definite vet is the direct evidence that the ionizing potentials \nof the alkali metals are lower than those of any other elements in the \nsame rows of the Periodic Table, those of rubidium and caesium being \naltogether the smallest known. These alkali metals follow, in the \nPeriodic Table, immediately after the five noble gases helium, neon, \nargon, xenon and krypton respectively. These gases are chemically \nall but absolutely inert, almost never entering into combinations. \nTheir ionizing-potentials are higher than those of any other elements \nin their respective rows of the periodic table, and those of helium \nand neon are the greatest known. The atoms of each of the alkali \nmetals are much larger than those of the preceding inert gas.\n\nFrom all these facts the inference is drawn, that the atom of each \ninert gas consists of a nucleus and electrons, at least the outermost\n\nones of which are arranged in a peculiarly stable and symmetrical \nfashion (as for instance, in a group of eight at the corners of a cube, \nthough this is by no means sure); while the atom of the next following \nalkali metal consists of just this sort of arrangement or \u201c\u2018inert-gas \nshell.\u2019 now to be known as the \u201cresidue\u201d or \u201ckernel,\u201d and of one \nadditional electron now to be known as the \u201cvalence-electron,\u201d\u201d usually \nmuch farther away from the nucleus\n\nIf to such an atom-model we apply the doctrine of Stationary \nStates, we mav infer that for each and every arrangement of the \nelectrons in the residue, or (to use more general terms) for each condi- \ntion of the residue, there is a whole svstem of Stationary States ditter- \ning from one another only in that the valence-electron travels in \ndifferent ones among a svstem of quantized orbits. These orbits we \nmay suppose to conform to the quantum-conditions (2) and (3), at \nleast until convincing evidence is brought to the contrary. Such in \nfact is the interpretation of the svstem of Stationary States, transi \ntions between pairs of which are responsible tor the \u201coptical spectrum\u201d \nof each alkali metal.\n\nAn electron at a very great distance from the kernel of such an \natom will experience an attraction towards it, practically indis- \ntinguishable from the attraction which would be exerted by a single \n(hydrogen) nucleus of charge +e. One might sav that the (Z\u20141 \nelectrons surrounding the nucleus of charge + Ze effectively cancel \na portion +(Z\u20141)e of the nuclear charge; or to use a more common \nword, that they \u201cscreen\u201d it. As the imagined distant electron moves \ninward towards the kernel, the screening will cease to be pertect \nAn effect should occur analogous to the \u201cstray tield\u201d which penetrates \nthe meshes of a grid; since the electrons of the kernel do not form a \ncontinuous shell of electricity enclosing the nucleus, the latter should \nmake itself felt through the interstices, although this effeet may be \ndiminished by the swift motion of the electrons. All this ts specula- \ntion of the wildest kind. The only deduction reasonably sate is this, \nthat very far outside the kernel the tield will be very nearly the inverse- \nsquare field due to a hydrogen nucleus of charge +e; very near to the \nkernel the field will be quite incalculable '; while in between the very \nfar-out and the very-near-in region, there will be an intermediate \nregion, in which there may be some chance of finding an adequately \napproximate expression for the field. On the existence of such a\n\nUnless it is violently simplified by some agency or restriction of which at present \nwe know nothing.\n\nrests all the present hope of achieving numerically valid theories in \nthis division of atomic physics.\n\nOne agreement between this theory and certain data may be demon- \nstrated without making any specific approximation. The farther \naway the valence-electron remains from the kernel, the more nearly \nidentical with the field of a hydrogen nucleus is the field in which it \nrevolves, the more nearly should it behave like the electron of a hydro- \ngen atom. Consider for instance, in a hydrogen atom, the orbit which \nvields the Stationary State for which the total quantum-number and \nk the azimuthal quantum-number are both equal to 5. This orbit \nis a circle of which the radius is 10-7 cm.; far larger than the radius of \nany inert-gas atom, presumably a fortiort far larger than the kernel \nof any alkali-metal atom. Were the valence-electron of such an \natom to describe this circle, it would pass everywhere in a field very \nnearly like that of a hydrogen nucleus, and should very nearly con- \nform to the quantum conditions for this field. It follows that an \norbit drawn in the actual field, obeying the quantum-conditions \nn=5 and k=5, would be very nearly such a circle with very nearly \nthe same energy-value. The inference is drawn that for high values \nof n and k, the Stationary States of an alkali-metal atom should be \nvery nearly identical with those of hydrogen. These orbits which lie \nfar out from the kernel of the alkali metal atom, or from the nucleus \nof the hydrogen atom, have small energy-values. It may therefore \nbe said that if we tabulate the Stationary States of the two atoms in \norder of decreasing energy-value, then the farther along the two \ntabulations we go, the more nearly should the two systems of Station- \nary States coincide.\n\nThis is found to be true, under a limitation. The limitation is an \nimportant aid in interpreting the arrangement of the Stationary States. \nIt will be recalled from the First Part of this article that the Station- \nary States of the sodium atom are arranged in several sequences (there \nillustrated as columns in Fig. 7) known as the s-sequence and the p- \nsequence and the d-sequence and the f-sequence and others; and to \nthese sequences successive values 1,2,3,4.... of a symbol k were ap- \npended. One basis for this classification is that when it is made, the \noccurrence or non-occurrence of transitions between any pair of \nStationary States, under normal conditions, can be determined by \napplying the \u201cselection-rule\u201d that only such transitions occur as \ninvolve a change of one unit in k. Now there are two reasons for \nsupposing that the only transitions which can occur are those in \nwhich the Azimuthal Quantum-number of the valence-electron changes \nby one unit. Unfortunately it is not possible to introduce these two\n\nreasons with all the necessary background without too long a stoppage \nof the main current of this argument.2 I must therefore set it down \nas an assertion, that the selection-rule is deducible from the assump \ntion that the value of & is the Azimuthal Quantum-number of the \nvalence-electron; which thus is 1 for all the Stationary States of the \ns-sequence, 2 for each State belonging to the p-sequence, 3 for the \nd-sequence, and 4 for the f-sequence. The feature common to the \nvarious Stationary States of a sequence is, therefore, the Azimuthal \nQuantum-number of the valence-electron\u2014if this atom-model is valid\n\nThis being assured, the conclusion is drawn that, since & is higher \nfor the f-terms than for the d-terms, higher for the d-terms than \nfor the p-terms and higher for the p-terms than for the s-terms; since, \ntherefore, the f-orbits are ceteris paribus more nearly circular than \nthe d-orbits and less inclined to stretch down into the neighbor- \nhood of the kernel, the d-orbits more nearly circular than the p-orbits \nand the p-orbits more nearly circular than the s-orbits\u2014therefore \nthe approximation of the sodium terms to the hydrogen terms will \nbe most nearly perfect for the f (and higher) sequences, less so for \nthe d, less for the p and worst for the s-terms. This also is veritied. \nIt reinforces the opinion that the k-values assigned to the various \nsequences are actually their azimuthal quantum-numbers.\n\nAs the different States of a single sequence share a common Azimu- \nthal Quantum-number, they must differ\u2014supposing always that this \natom-model is valid\u2014in their Total Quantum-number. Consecutive \nStates of a sequence presumably have consecutive values of the \nTotal Quantum-number (although sometimes one meets with a break \nor a jolt in the continuity of a sequence, suggesting a departure from \nthis rule). The meanings of the Total Quantum-number 2 and of the \nAzimuthal Quantum-number & for elliptical orbits are such, that \u00bb \ncan never be less than k. Hence the value of \u00bb for the first Station- \nary State of the s-sequence may be unity, or greater; but the values \nof for the first terms of the p-sequence, the d-sequence and the f- \nsequence may not be less than 2, 3, and 4, respectively.\n\nStrange as it may seem, there is no perfectly satisfactory way of \ndetermining the value of m for all Stationary States. Generally \nit happens that the various States of an f-sequence, that of sodium \nfor example, agree so closely with those States of hydrogen which \nform an my sequence, that there is little hesitation in attaching to \neach of the f-States the same value of \u201d as is borne by that State \nof the hydrogen atom which coincides with it so nearly. For instance,\n\n? These being (verbum sapienti) the argument associated with the name of Rubin \nwicz, and the argument deduced from the Principle of Correspondence.\n\nthe first f-State of sodium has very nearly the same energy-value \nas the 4, State of hydrogen; the second f-State of sodium nearly \ncoincides with the 4, State of hydrogen, and so forth along the se- \nquence. Hence to the successive States of the f-sequence of the \nsodium atom one attaches with confidence the symbols 44, 54, 6;, and \nso onward. In some cases this is practicable for the terms of the \nd-sequence also; but never for those of the s-sequence. The Stationary \nStates of the s-sequence depart so far from those of hydrogen, that \none cannot with any security conclude what values of the Total \nQuantum Number should be assigned to them. [It used to be assumed \nthat w#=1 for the first term of the s-sequence and n=2 for the first \nterm of the p-sequence, and the usual notation for the Stationary \nStates reflects this supposition; which however is neither necessary \nnor probable.\n\nAll of the foregoing interpretations are based upon a theory of the \nalkali-metal atoms which may be summarized in this way: as the \nhydrogen atom is supposed to consist of a nucleus surrounded by an \ninverse-square field through which an electron travels always in one \nor another of certain orbits determined by quantum-conditions, so \nalso the alkali-metal atom is supposed to consist of a kernel sur- \nrounded by a not-inverse-square field) through which electron \ntravels always in one or another of certain orbits determined by \nidentical quantum-conditions. As the Stationary States of the \nhydrogen atom correspond each to a certain orbit and are designated \neach by certain values of two quantities 7 and k, or for short by a \nsvmbol a, indicating the features of that orbit, so also the Stationary \nStates of the alkali-metal atom correspond each to a certain orbit and \nare designated each by a symbol ay. For the hydrogen atom we recog- \nnize the proper a, for each Stationary State because of the wonderful \nnumerical agreement between Bohr\u2019s theory and the experimental \nvalues for the energy of each State. For the alkali-metal atom we \ncan only guess the proper my for each Stationary States from = indi- \ncations of much lesser evidential value. We suppose, however, that \nk=1,2.34 for the various States of the s, p, d and f sequences, respec- \ntively; so that the s-sequence is like the 7, sequence of hydrogen, the \np-sequence like the #2 sequence, and so on. Of the values of m we \nare moderately sure for the f and d sequences, quite uncertain for \nthe terms of the s and p sequences.\n\nOne may now wonder whether it is possible to invent a central \nfield, such that the orbits traced in it according to the quantum- \nconditions (2) and (3) would vield a series of energy-values agreeing \nwith the observed energy-values of the Stationary States of (let me\n\nfor the sodium atom involves ten electrons around the nucleus in \naddition to the one \u2018\u2018valence\u201d electron for the benetit of which the \nfield is being devised; and one might expect these ten electrons to bx \nrushing around the nucleus in uncoordinated and non-recurring paths \nnever at any two instants similarly placed and similarly moving, \nnever at any two instants exerting the same influence upon the valence\n\nelectron. Yet the Stationary States of the sodium atom are as \nsharply defined as those of the hydrogen atom; and this may \nthought to mean that the ten electrons of the kernel are constrained \nto a unity and a fixed relationship, like that of the members of a \nmachine if not like that of the parts of a rigid body, which translates \nitself into an influence upon the valence-electron not unlike that of a \ncentral field.\n\nAt all events, several physicists working independently in) various \nnations have taken the not inconsiderable trouble of devising central \nfields to fulfil the condition required; and they appear to have ac \nchieved a respectable success. It is not easy to decide what this \nsuccess requires the rest of us to believe; perhaps it is formally possibl \nto devise a central field to account for amy set of Stationary States: \n1am not sure whether this question has been adequately examined \nSome have felt confident enough to sav that the results show which of \nthe Stationary States correspond to orbits of the valence-electron \nwhich \u201cpenetrate into the kernel\u201d and which to orbits that remain \nin all their circuit quite outside of the kernel. It is to be hoped that \nthis problem will become clearer in the next few vears. At this point \nI will add only, that the orbits traced for the valence-electron are \nrosette orbits in which the precession ts very rapid, so that consecutive \nloops of a rosette are inclined at a considerable angle to one another \nIn the model for the hydrogen atom, the consecutive loops of a rosette \norbit lie so close together as to be indistinguishable when drawn to \nscale on an ordinary sheet of paper (the separation between them \nwas much exaggerated in Fig. 3 of the Second Part of this artick \nIn these atom-models, the orbit looks rather as if it were drawn along \nthe edges of the blades of an electric fan.\n\n\u00a9. INTERPRETATION OF THE OPTICAL SPECTRA \nOF OTHER ELEMENTS \nAs soon as we step from the first column of the Periodic Table into \nthe second, the obstacles to such a theory as we have hitherto tried\n\nwhich seems to bear upon the arrangement of the electrons in |}, \natoms; but some of it leads to conclusions opposite to those which ||, \nremainder suggests.\n\nOn the one hand, line-series are discernible in the spectra of cle. \nments in the second and the third columns of the Table, and even \nin those of some others; and from these line-series, svstems of Station- \nary States are deduced which resemble those ascertained for the alkali- \nmetal atoms; and it is natural to extend the same explanation from \nthat case to these, supposing again that each atom consists of a \nnucleus and a certain number of electrons, all but one of which are \ntightly bound into a residue, around which the one remaining electron \ncirculates in one or another of various quantized orbits.\n\nOn the other hand, the chemical behavior of these elements does \nnot confirm this easy classification of the NV electrons of an atom into \n(V\u20141) very-tightly-bound electrons and one which is very loosely \nbound. Thus, the atoms of elements of the second and third columns \nof the Periodic Table\u2014\u2018\u2018alkaline-earth metals\u201d and \u2018earth metals,\u201d \nas they are called\u2014when floating in water as the fragments of mole- \ncules of dissolved salts of these elements, are found to be deprived \nof two and of three electrons, respectively; and the composition \nof these salts is such as to suggest that the atoms of the other element \nor elements involved in them have annexed two or three electrons, \nrespectively, from the alkaline-earth atom or from the earth-metal \natom. These facts suggest rather that the .V electrons of an alkaline- \nearth atom, or of an earth-metal atom, should be classified into (V\u20142 \nor (V\u20148) very-tightly-bound electrons and two or three which are \nloosely-bound, respectively. The very tightly bound electrons will be \nequal in number to, and presumably arranged like, the electrons of the \natom of the next preceding inert gas. Henceforth I will reserve the \nword \u2018\u2018kernel\u2019\u2019 for such a system, and the word \u201cresidue\u201d for what \nis left behind when one electron is separated in fact or in imagination \nfrom the atom. Thus these two words will not mean the same thing \nexcept in special cases, such as those of the alkali-metal atoms.\n\nSpecifically, let us consider the four consecutive elements argon \n(inert gas, 18th element of the Periodic Table), potassium (alkali \nmetal, 19th element), calcium (alkaline-earth metal, 20th element \nand scandium (earth-metal, 21st element).\n\nThe evidence from chemistry and from electrolysis impels us to \nthink that the argon atom consists of a nucleus surrounded by (eigh- \nteen) electrons tightly bound, in a stable and imperturbable arrange- \nment; that the potassium atom consists of a kernel much like the \nargon atom, with one additional electron loosely bound and hence\n\nof kernel and two loosely-bound electrons, t \nsame sort of kernel and three outer electrons.\n\nThe Stationary States of the calcium atom resemble, in their \narrangement, those of the potassium atom sufficiently to make the \nsame general sort of an explanation desirable,\u2014to make it desirable \nto suppose that one electron is loosely-bound and remote from the \nnucleus, the other nineteen tightly-bound and near the nucleus; one \nloosely-held electron versus nineteen tightly-held ones. But the \nevidence from chemistry and electrolysis demands two loosely-held \nelectrons versus eighteen tightly-held ones.\n\nOne might try to evade the dilemma by supposing that the calcium \natom is a sort of three-stage construction, with eighteen electrons \ncongregated in a kernel around the nucleus, a nineteenth far out bv \ncomparison with the nucleus, a twentieth far out by comparison with \nthe nineteenth. For interpreting spectra, the residue of the atom \nwould be the kernel or \u201cinert-gas shell\u201d and the nineteenth electron, \nthe valence-electron would be the twentieth. For interpreting chemi- \ncal data, the residue of the atom would be the inert-gas shell. This \nconception would rescue the interpretation of the calcium spectrum \nmade after the fashion of the one just expounded for alkali-metal atoms. \nIt would probably demand a larger atom, or a more shrunken kernel, \nthan other data will allow.\n\nOr one might suppose that the nineteenth and the twentieth elec- \ntron are on the whole about equally remote from the nucleus, and vet \nit is possible for one of them to change over between any two of a \nvast system of quantized orbits without greatly affecting the other. \nThere is certain evidence for this conception which I shall presently \nnarrate.\n\nOr one might suppose that the nineteenth and the twentieth electron \nare a system by themselves, and that each Stationary State corresponds \nto a particular configuration of this svstem, so that each line of the \nspectrum is attributed to a leap not of either electron separately but \nof both together. This idea seems to be gaining ground rapidly in \ndealing with atoms composed of a kernel and several outer electrons,\n\nthree or four or five or six or seven. The preceding notion might ty \nbrought under it as an especial case. If it is accepted the theory o \natoms other than the alkali-metal atoms will inevitably be mor \ncomplex than the theory mentioned for these in section P.\n\nAn interesting feature of some of these spectra discloses that the \nresidue of the atom may exist in either of two distinct states. It will \nbe recalled that the energy-values of the Stationary States have been \nmeasured from the state of the ionized atom, to which the energy \nvalue zero is assigned. In this fundamental state, one electron and \nthe residue of the atom are completely sundered; and the energy- \nvalue of any other Stationary State is the energy required to tear the \nelectron completely out of the atom when the latter is initially in that \nStationary State. This definition implies that the state attained when \nthe electron is completely separated from the rest of the atom is de- \nterminate and unique. Such must be the case if the atom consists of \nan invariable nucleus and one electron, as in hydrogen; but if the \natom contains several electrons, there is no a priori reason for ex- \ncluding the possibility that there may be several \u2018states of the ionized \natom\u201d; in each of these states one electron will be far away, but the \nresidue will have as many different arrangements as there are different \nstates. Extending this idea, one infers that there may be two or \nmore distinct sets of Stationary States for certain elements, each set \nculminating in a different final configuration of the residue,\u2014that \nis to say, of the ionized atom.\n\nSeveral instances of atoms possessing two such distinet families of \nStationary States are known; the most noted is probably that of neon, \nbut I will describe the case of calcium, lately interpreted by Russell \nand Saunders and independently by Wentzel. Two families of terms \n\u201cprimed\u201d and \u201cunprimed,\u201d had been identified in the spectrum of this \nelement, and important sequences of each could be followed \u2014 sufti- \nciently far to make the extrapolation to the limit not too daring. The \nhmits were different, showing that the amount of energy required to \nseparate an electron from an atom initially in its normal state had \ntwo values differing from one another by 1.72 equivalent volts. Con- \nsequently the residue may remain (it is not necessary to assume that \nit can long remain) in either of two States differing from one another \n(when the extra electron is far away) by this amount.\n\nAt this point a very significant numerical agreement enters upon \nthe scene. The residue of the calcium atom, the \u2018onized-calcitum atom, \nhas itself a spectrum which is known, and from which its system of \nStationary States has been learned and mapped. Like the systems of\n\npd and other sequences. The Normal State of the ionized-calcium \natom belongs to the s-sequence; following the usual custom it) may \nhe called the (1, s) State. The State of next lowest enerev-value, \nthe \u201cnext-to-normal\u201d\u2019 State (so to speak belongs to the d sequence, \nand may be called the (3, d) State. The energy-ditference between \nthe (1, s) State and the (3, d) State is 1.69 volts. This agrees within \nthe error of the experiments with that value 1.72 equivalent volts, \nwhich was found for the energy-ditterence between the two conditions, \nin either of which the residue of the calcium atom might be left after \nthe twentieth electron is abstracted. This agreement shows that the \nextraction of the 20th electron from a calcium atom mav leave the \nresidue either in the (1, s) State or in the (3, d) State.\n\nIf now we remember that the ionized-calcium atom is comparable \nwith the potassium atom (and with alkali-metal atoms generally \nhaving as it does eighteen electrons very tightly bound as a kernel \naround the nucleus and one electron loosely held then it is reason- \nable to use the same interpretation of its Stationary States as was \nexpounded in Section P; and to suppose that when the ionized-cal- \ncium atom is in the (1, s) State that loosely-held electron is revolving \nin a certain 2, orbit, and when the atom is in the (3, d) State the \nelectron is revolving in a certain m3 orbit. Thus the extraction of \nthe 20th electron of the calcium atom may be supposed to leave the \n19th electron sometimes in the one, sometimes in the other of these \ntwo orbits.\n\nWe may now inquire whether the 19th electron will always remain \nin its #2, orbit, or in its a, orbit as the case may be, when the 20th \nelectron reenters the atom, descending from one orbit to another. \nHere it is necessary to watch one\u2019s mental steps very closely; for one \nis liable to slip into the naive notion of a particular orbit, say for \ninstance a 33 orbit, as a fixed and permanent railway-track around \nwhich the electron continually runs until something violent derails \nit. This could not be true unless (to take this special case) the 2 dth \nelectron had no influence whatever upon the I@th. Were it so, every \nStationary State of the one family would differ by the same amount. \n1.69 equivalent volts, from the corresponding State of the other \nfamily. In fact, the energy-difference between corresponding States \nvaries from one pair to another. This may well be simply because \nthe approach of the 20th electron so alters the forces acting upon \nthe 19th, that its orbit is changed both in geometry and in\u2019 energy- \nvalue, while remaining still identified with the same values of its \nquantum-numbers. The experiments neither prove nor disprove \nthis; it is commonly accepted as true.\n\nIt is a very important fact that the atom may pass from a State o \none family to a State of the other,\u2014in terms of the model, that thi \n19th electron passes from its m3 orbit to its m orbit, and simultane \nously the 20th electron makes some transition or other of its own \nThe emitted radiation contains the energy resulting from both change- \nsimultaneously, fused together without any discrimination.\n\nI next point out that the processes whereby the lines of an optical \nspectrum are emitted may be regarded, if this theory of the atom is \nvalid, as stages in the gradual formation of an atom. Consider the \nhydrogen spectrum to begin with; each line is emitted as the atom \npasses from one Stationary State to another of lower energy-value, \nthe state of least energy being the Normal State of the perfected atom \nand the state of greatest energy being the condition in which the \natom-residue and its electron are torn apart. The various lines of the \nspectrum correspond to various partial steps along the path from the \nlatter of these states to the former, to various stages of the formation \nof a hydrogen atom from two separated parts. The specific conception \nof each Stationary State as a definite orbit of the electron about a \nnucleus merely reinforces this way of envisaging the process. In the \nspectra of ionized helium and of neutral helium, we read the testimony \nof the gradual formation of a helium atom out of a nucleus and two \nelectrons initially quite dissevered. The various lines of the ionized- \nhelium spectrum correspond to different stages in the advance of an \nelectron from the state of freedom to the state of most stable asso- \nciation with a nucleus of charge 2e, or in Bohr\u2019s language, to different \nstages in the \u201cbinding\u201d of an electron by a nucleus of charge 2e. The \nvarious lines of the neutral-helium spectrum correspond to stages in \nthe \u201cbinding\u201d of a second electron by a system composed of a nucleus \nof charge + 2e and an electron already bound to it. Thus the two \nspectra of helium testify to two consecutive processes in the upbuilding \nof a helium atom out of its constituent parts.\n\nThe process of building up an atom, by successive adhesions of \nelectrons to an incomplete electron-system surrounding a nucleus \nthat is to say, the process of building a system of Z electrons around a \nnucleus bearing the charge Ze, out of a system of (Z-b) electrons \nsurrounding the nucleus, by consecutively adding 6 electrons one \nafter the other\u2014evidently occurs very profusely in intense high- \ncurrent high-voltage discharges in vapours, such as the condensed\n\npark and above all Millikan\u2019s \u201cVacuum Spark.\u201d To take instances \nfrom the work of Millikan and Bowen, Paschen, and Fowler: in the \n.pectra of such discharges lines have been identitied which belong to \natoms for which Z=14 and b} has the several values 1.2.3.4 (four \nstages in the building of a silicon atom); and to atoms for which \nZ=10+6 and b has the several values 1,2,3,4,5,6. Many of these \nspectra of multiply-ionized atoms have not yet been analyzed, but the \nwork is proceeding rapidly. There is reason to hope that within a \nfew years we shall be in possession of interpreted spectra not only of \nmany systems of Z electrons about a nucleus of charge Ze, but also of \nmany systems of fewer than Z electrons about nuclei of charge + Ze. \nThis may be highly important, as I will try to show by an illustra- \ntion. We will consider two consecutive elements of the periodic table; \nsodium (Z=11) and magnesium (Z = 12).\n\nA Mg atom is imagined as 12 electrons around a nucleus of charge \n+12 e. It is formed when one electron joins itself to a Mg+ion, \nwhich is composed of Ll electrons about a nucleus of charge +12 e \nFor this process a spectrum is emitted, the so-called are spectrum of \nMg or \u201cMgl\u201d spectrum, which is known and analyzed. [t shows that \nthe normal state of the Mg atom is an s-state (probably of total \nquantum-number 3). It is likewise a singlet-and-triplet-spectrum. \nThe first of these facts is taken to mean that the valence-clectron, or \ntwelfth electron (the reader will see the reason for this usage, the electron \nin question being the last annexed out of the twelve) of the Mg atom \nmoves in a 3, orbit. The second is taken to mean something or other \nabout the residue of the atom, as will be shown in section S.\n\nThis residue of the atom is itself formed when one electron joins \nitself toa Mg+ +ion, which is a group of 10 electrons about a nucleus \nof charge +12 e. In this process the so-called spark-spectrum of Mg, \nor spectrum, is emitted. Tt is known and analyzed. [t shows \nthat the normal state of the Mg+ion is an s-state (probably of total \nquantum-number 3). [It is a doublet spectrum. The first of these \nfacts is taken to mean that the valence-electron or eleventh electron of \nthe Mg+ion, moves in a 3, orbit. The second is taken to mean some- \nthing or other about the residue of the Mg++ion.\n\nA very interesting question now arises: is the Mg+ion actually the \nsame as the residue of the Mg atom?) In other words: when a 12th \nelectron is added to the group of Ll electrons about a nucleus of \ncharge+12 e, is the group of eleven left unchanged? Lf so, we have \nknowledge about this group from two sources. The character of the \nMgl spectrum (the fact that it is a singlet-and-triplet) spectrum) \nteaches something about the group, though what it is is far from\n\nThis suggests that it would be a most desirable achievement t \nproduce the spectra due to groups of (Z\u20146) electrons congregated \nabout a nucleus of charge +Ze, for some value of Z (the higher the \nbetter) and all values of 6 from O to (Z~\u20141). Were this done we \ncould almost lay claim to having witnessed the creation of an atom \nfrom fundamental particles common to all matter. We could not \nquite make this claim, since the nucleus of charge +Ze would still \nremain characteristic of that one kind of atom alone; but we should \nhave made a substantial approach to it. However, there is no im\n\nthe spectrum expected for Li++ (ie. for Z=3 and b=2) acts as a \nbarrier against utterly tearing down the electron-structures of higher \natoms so that they can rebuild themselves before our eves from the \nfoundations.\n\nThe next important question may be introduced in this fashion, \nSuppese that nothing were known about the spectrum called Mell, \ntherefore nothing about the process of adding an eleventh electron \ntoa group of ten around a nucleus of charge 12 e. Knowledge would \nstill be available about the process of adding an eleventh electron to \na group of ten about a nucleus of charge 11 e; for this is precisely \nthe process which creates the neutral sodium atom out of the Na-+ion, \nand results in the emission of the Nal spectrum or are spectrum of \nsodium. \u2018This spectrum is a doublet spectrum, and it shows that the \nnormal state of the sodium atom is an s-state, probably of total \nquantum number 3. This last fact is taken to mean that the eleventh \nelectron in a group of eleven electrons about a nucleus of charge \n+11 e, is revolving in a 3; orbit. Could we have assumed that there- \nfore the eleventh electron, in a group of eleven electrons about a \nnucleus of charge +12 e, is revolving in a 3; orbit? There is no a \npriort certainty of this: but the observations on the MgII spectrum, \nas we have seen, confirm it (and also that the residue of the Mg+ion \nis like the residue of the Na atom, in causing the next added electron\n\nWere this generally true we could say that each atom in the period \ntable is like the residue of the next atom following it; and that the \nmth electron in the mth atom is revolving in the same sort of orbit \nas the outermost electron of the mth atom, for every value of 7 an \nfor every value of m less than that value of 1.\n\nHowever, it is not always true. To take another specitic instance, \nconsider the two elements potassium (Z=19) and calcium (Z = 20 \nThe spectrum KI, which is due to a nineteenth electron joining a \ngroup of 18 about a nucleus of charge +19 e, and the spectrum Call, \nwhich is due to a nineteenth electron joining a group of IS about a \nnucleus of charge +20 are dissimilar. The dissimilarity is not \nquite so great as to affect the normal states of the two systems, Ko and \nCa+, composed of nuclei of charge 19 e and 20 e each surrounded by \n19 electrons; both have as normal state an s-state, apparently of \ntotal quantum-number 4; it is inferred that in each, the d?th electron \nrevolves in a7, orbit. If we consider, however, the first of the d-states \n(to which the total quantum-number 3 is commonly assigned), we \nsee that in the KI spectrum it has a much larger energy-value than \nthe Normal State, while in the Call spectrum it has nearly the same \nenergy-value. A short leap of the imagination leads to the conclusion \nthat if we could examine the spectrum produced by a 19th electron \njoining a group of 18 about a nucleus of charge +21 e, the d-state in \nquestion would have a smaller energy-value than any s-state. In \nthis case it would be the Normal State itself,\u2019 and we should say that \nthe 19th electron, in a group of 19 surrounding a nucleus of charge \nZe, revolves in a m, orbit if Z=19 or 20, but in a ny orbit if Z= 21.\n\nThis system of 19 electrons around a nucleus of charge 21 e@ is a \ndoubly-ionized scandium atom, Se++-. Its spectrum has not been \nproduced, so that the foregoing sentences are still somewhat specu \nlative. What gives them value is the inference that scandium marks \na sort of a breach in the regularity of the Periodic System. For \nmost of the elements in the Periodic System, it can be said that the \natom consists of a residue which is like the atom of the preceding \nelement, and an additional electron; and that in its turn this atom \nresembles the residue of the atom of the element next following. To \nthis the regular periodicity of the properties of the elements ts ascribed \nBut when we reach an element of which the atom has a residue dis \ntinctly different from the atom of the foregoing element, then the \nregular variation of the physical and chemical properties is inte \nrupted. Scandium, as a matter of fact, is the first of a group of\n\nIn the First Part of this article the impression may have been left that the Normal \nState of every atom is an s-state. This is not true; in some known cases the Norma! \nState is a p-state, in others an f-state.\n\nelements, the intrusion of which into the Periodic Table brings about \na disruption of the simplicity of its first three rows. There are other \nsuch intrusive groups of elements, notably the celebrated groups of the \nrare earths. [It is supposed that wherever such a group commences, \nthere the residue begins to vary from one atom to the next. The spec- \ntroscopic evidence is lacking; it is awaited with extreme interest,\n\nThe reader will very probably have seen one or more tables of the \ndistribution of electrons in atoms; tables in which it is stated, for \ninstance, that the atom of sodium contains two electrons moving in \n1, orbits, four in 2; orbits, four in 2s orbits, and one in a 3, orbit; \nor more succinctly that it contains \u201ctwo 1,, four 2;, four 2. and one 3; \nelectron.\u201d Such tables are built by piecing together bits of evidence, \nsome of which are such as I have described in this section, while \nothers are inferences from X-ray spectra, magnetic properties, or \nobservations of still other kinds. That they are still highly speculative \nis confirmed by the fact that they are continually being remodeled. \nIf we could produce the spectra corresponding to all the stages of \nformation of an atom, we should be able to set up a tabulation more \nreliable than any vet put together. Even then, however, we should \nbe contronted with the question whether the addition of a new electron \ntoa kernel fundamentally alters the distribution of those already there.\n\nHaving considered the facts at such length in this section, we are \nentitled to consider the theory. In the coupled cases of hydrogen and \nionized helium it was shown by experiment, and rendered plausible \nby theory, that the Stationary States of the element with one electron \nand a double charge on its nucleus correspond exactly to those of the \nclement with one electron and a single charge on its nucleus, and are \nendowed with fourfold the energy of these latter. This conclusion \ncan be extended to cover the case of a valence-electron circulating in \nan orbit ata great distance from a kernel composed of (Z\u20146) electrons \nand a nucleus bearing the charge +Ze. The field due to the kernel \nwill at great distances approximate the field due to a solitary nucleus \nbearing the charge be. We have seen already that when b=1 (so \nthat the total charge on the nucleus balances the total charge of the \nelectrons, valence-electron included) the Stationary States corres- \nponding to orbits for which \u00bb and & are large coincide with Stationary \nStates of hydrogen. [t follows equally that when )=2, the Stationary \nStates for which \u00bb and & are large have approximately fourfold the \nenergy of stationary states of hydrogen, and coincide approximately \nwith Stationary States of ionized helium. This is verified by experi- \nment, and so are the corresponding conclusions for the cases 6=3 and \nb= 4,\n\nHeretofore in the Third Part of this article TE have repeated the \nprocedure adopted in the First Part, simplifving the actual facts by \nwriting as though the Stationary States of each atom were arranged \nin sequences of individual terms, each sequence being distinguished \nby a particular value of the Azimuthal Quantum Number Here as \nthere, it finally becomes necessary to concede the complexity of the \nfacts, and recognize that the sequences in question are sequences not \nof individual terms, but of groups of terms. Thus for instance the \nsodium atom possesses a p-sequence, not of individual terms but of \npairs of terms\u2014a pair 2p; and 2p2, then a pair 3p; and 3p., then a \nmultitude of other pairs. For another instance, the mercury atom \nexhibits a p-sequence not of individual terms but of triads of terms \na triad 2p; and 2p2 and 23, then a triad 3p; and 3p. and 3p,, and \nthen a procession of other triads. These sequences are collected \ninto systems: an s-sequence and a p-sequence and a d-sequence and \nseveral more constitute a system. There are singlet: svstems and \ndoublet systems and triplet systems and systems of still higher \u00bbru/si- \nplicity; and each kind of system is distinguished by a certain manner \nof grouping of the terms which form its various sequences. Note \nworthy and peculiar laws govern these groupings; in a doublet svstem, \nfor instance, the s-sequence consists of individual terms, but all the \nothers consist of pairs of terms; in a quartet system, the s-sequence \nis made up of single terms, the p-sequence of triads of terms, the \nremaining sequences of groups of four terms each. From the First \nPart of this article reprint a Table showing how the terms are grouped \nin systems of all multiplicities from the singlet to the octet. The \nnumbers opposite the name of each system and under the letters of \nthe various sequences show how many terms belong to each group \nin the various sequences of that system.\n\nleach atom possesses one or more such systems of Stationary States; \u2018 \nand the particular types which an element displays depend in a ia \ndetinite and fairly clear manner upon the position of the element in 4 \nthe Periodic Table, being in fact one of the most distinctive of the q \nperiodically-varving qualities. Each atom with an even number of ? \nelectrons exhibits svstems which are all of odd multiplicity, and each j \natom with an odd number of electrons exhibits systems which are all 4 \nof even multiplicity; thus magnesium, with twelve electrons, has a | \nsinglet system and a triplet system, while sodium and once-ionized 4 \nmagnesium, each with eleven electrons, have each a doublet system, 4 \nand neon with ten has a singlet, a quintet and two triplet systems.\u2018 : \nRemembering what was said about the consecutive binding of elec- : \ntrons, it will be noticed that these facts show a regular difference be- : \ntween the binding of the Vth electron when WV is odd and the binding \u2018 \nof the Vth electron when NV is even. Otherwise expressed, they show 4 \nthat a kernel of .V electrons treats an oncoming member in one or \u2018 \nanother of two distinctive ways, according as .V is even or odd. The q \ninfluence of magnetic fields on spectra likewise shows that this com- : \nplexity of the Stationary States is a quality not negligible, but primary. 4 \nThe features of the atom-model hitherto described must be supple- ; \nmented with some new one if it is to cope with such facts as these. : \nWe have represented (for example) the sodium atom in its 2p state : \nby a \u201c\u2018valence-electron\u201d cruising with angular momentum 2(h/27) : \nin.an orbit around a \u201ckernel\u201d composed of ten electrons and a nucleus. : \nBut there are two such states instead of one; if the angular momentum ; \nof the valence-electron is to be equal to 2(4 27) for each of these, \u2018 \nsome other not yet mentioned feature of the atom must discriminate i & \nthe two. One might, of course, again proceed as we did in discussing : \nthe \u201cprimed terms,\u201d by assuming that the kernel of the atom is in ; \none condition when the atom is in the 2p; state, and in another slightly i \ndifferent condition when the atom is in the 2p, state. This would : \nprobably entail as many different conditions of the kernel as there ; \nare pairs of terms in the sodium spectrum \u2014a great number, and yet \n| small in comparison with the multitude which would be required by : \nother atoms; vet such may be the eventual theory. However, it is : \npossible to construct for these facts an atom-model out of two re- i\n\nvolving parts, whereby different Stationary States of a group are \nrepresented not by varying the condition of either part separately,\n\nHydrogen and ionized helium are not included under this rule. Helium shows \na singlet and a doublet system together, a combination which violates the rule as 4 \nstated, unless the doublet system is really a triplet system in which two states of \" \neach triad are too close together to be distinguished.\n\nbut by varying the relative orientation of the two. Although this \ntheory has not been harmonized with those which [| have hitherto \nrecited, it is competent in its own field; and for that reason I present it \nWe will imagine that the atom is represented by a combination of \ntwo flywheels, two whirling objects, endowed each with angular \nmomentum. These angular momenta are vectors, pointing along \nthe directions of the axes of rotation of the respective flywheels, and \nhaving certain magnitudes. 1 will designate them temporarily as Py \nand Pr, each symbol standing for a vector generally and also (when \nin an equation) for its magnitude. The angular momentum of the \nentire atom, which is necessarily constant in magnitude and in direec- \ntion so long as the atom its left to itself, is the resultant of Py and \nPria vector, pointing along the direction of the so-called \u201cinvariable \naxis\u2019 of the atom. I designate it by Pa. The following equation \nshows the relation between the magnitudes of these three angular \nmomenta and the angle 0 between the two ftirst-named, the angle \nwhich describes the relative orientation of the axes of rotation of the\n\nRemembering the successes which in dealing with the spectrum of \nhvdrogen have resulted from assuming that the angular momentum \nof the entire atom is constrained to take only such values as are \ninteger multiples Jh/2z of the quantity / 27, we make the same as- \nsumption here. We further make the same assumption for each of \nthe flywheels separately; the magnitudes of the angular momenta P, \nand Pr are supposed to take only such values Vh 27 and Rh 27 as \nare integer multiples of the same quantity # 27.2 These particular \nassumptions, frankly, are foredoomed to failure: but the failure will \nbe instructive.\n\nMaking all these assumptions together, we see that in effect we \nhave laid constraints upon the angle 0 which measures the relative \norientation of the two flywheels. For if Py is an integer multiple of \nh/2n, and Pe is an integer multiple of # 27, then Py which is fully \ndetermined by equation (6) cannot be an integer multiple of & 27 \nunless 0 is very specially adjusted. To illustrate this by an instance \n(which will be clearer if the reader will work it out with arrows on a \nsheet of paper): if Py and Pr are each equal to the fundamental \nquantity A/2z7, and if P4 must itself be an integer multiple of 4 27: \nthen cos 0 must take only the values, +1, \u20143, \u20141, which vield the\n\n* All that is actually being assumed is, that Py and Py, and Py are all integer\n\nmultiples of a common unit; nothing in this section will indicate either #2 or any \nother value as the precise amount of that common unit.\n\nvalues O, 27 3, 7 for 0, which vield the values 24/27, h/27, 0 for P, \nAny other integer values for Py (h 27) are unattainable by any \nvalue of Q whatsoever; anv value of 80 not among these three would \nvield a value for Py not an integer multiple of 4/22, which is con- \ntrary to the assumptions. Thus, the assumptions that the atom \nis a conjunction of two whirling parts, and that the atom altogether \nand each of its two parts separately whirl with angular momenta \nwhich are constrained to be integer multiples of a common factor \nthese assumptions lead to the conclusion that the relative inclination \nof the two revolving parts is constrained to take one or another of a \nstrictly limited set of values.\n\nThis essentially is the model devised by Land\u00e9 to account for the \ncomplexity of the Stationary States. The several Stationary States \nwhich form a group belonging to a sequence\u2014-in other words, which \nshare a common value of 2 and a common value of k, like the 2p; and \nstates of sodium or the 8d), 38d, states of mercury\u2014are supposed \nto resemble one another in this, that each of the whirling parts sep- \narately has the same angular momentum in every case; and to differ \nfrom one another in this, that in the several cases the two whirling \nparts are differently inclined to one another, so that the angular \nmomentum of the entire atom differs from one state to the next. \nThe ditterent Stationary States which share common values of 1 and \nk are supposed to correspond to different orientations of the two \nparts of the atom and to different values of its angular momentum.\n\nI will now no longer disguise the fact that these whirling parts are, \nor at any rate have been, supposed to be precisely the valence-electron \nand the residue. To the former we should therefore assign these values \nfor the angular momentum Py: the value #27 for every state be- \nlonging to an s-sequence, the value 2h 27 for every p-state, 3h, 27 \nfor every d-state, and so on. Then to the angular momentum Px \nof the residue we should assign a suitable constant value; a \u2018\u2018suitable\u201d \nvalue in this case being such a one, as would yield the proper grouping \nof terms in the various sequences of the system which the atom under \nconsideration is known to have. Thus, for an atom-model to represent \nsodium with its doublet system we require a value for the angular \nmomentum of the residue, such as will vield one permitted orientation \nwhen the atom is in an s-state (Pj; =f, 27), and two when it is in any \nstate for which Py=kh/27 and k is any integer greater than unity.\n\nfor, as was shown in the illustrative instance a couple of pages back, \nit vields three permitted orientations when Py=h/27, and (as can \ncasily be sl ) three for each and every other value of Py whict \neasily be shown) three for each and every other value of Py which\n\nis an integer multiple of 4 27. Thus it would form an idequate \nmodel for a system of Stationary States in which every group of \nterms in every sequence was a triad; but this is not a doublet svstem. \nnor even a triplet system, nor any other observed svstem whatever \nTo make this long story short; it is impossible to simulate anv of the \neight groupings of terms set forth in the eight lines of Table I by\n\nassuming that Py, Pr and P4 are all integer multiples of A 27 (or \nof any other common factor \nIt is in fact necessary to put Py equal, not to # 2r and to 2h 2x\n\nsand p and d states, respectively. This use of \u201chalf quantum num- \nbers\u201d makes it possible to produce an adequate medel for an atom \npossessed of a doublet svstem, by assuming that the angular momen \ntum Px of its residue is always / 27, and that its two whirling parts \nmust always be so inclined to one another that the angular momentum \nof the entire atom is an integer multiple of 4 27\n\nkor (to work out one example. and one only) when we make \nPr=h 2r and Py=}(h 27), then the greatest possible resultant \nthat can be obtained by combining these vectorially is * (A 2x) and \nthe least possible one is } (h 27); these two extreme values being at \ntained when the two component vectors are parallel and when they are \nanti-parallel,\u201d respectively. If we permit for the resultant only such \nvalues as are integer multiples of # 27, then there is only one that is \npermitted: the value hk, 2x7\u2014 for this is the only such value Iving within \nthe possible range. Next, put Pr=h/ 227 and Py =3 (h 27). All pos\n\nA more drastic use of \u201chalf quantum numbers\u201d is required to obtain \nan adequate model for atoms showing singlet and triplet and other\n\n7 The diagrams with arrows, offered by Sommerfeld in the fourth edition of hi \nclassic book, are very helpful in studying these models. Incidentally Sommerfeld\u2019s \nalternative way of arriving at the groupings of multiplet terms by compounding \nvectors is instructive.\n\nsystems of odd multiplicity. Thus to produce a singlet system it is \nnecessary to put Pr=} (h 27) always; to set Py equal to } (h/ 27) for \nall S-states, to} (h 27) for all P-states, and so forth; and to suppose \nthat the two whirling parts of the atom are constrained to take only \nsuch relative orientations as vield values for P4, the angular momen- \ntum of the entire atom, which are odd integer multiples of | (/ /27). \nIt is easy to see that there is but one such orientation for an s-state, \none for a p-state, and one for any other kind of state. To produce a \ntriplet, or a quintet, or a septet system, it is necessary to put Pr \nor 3 (hk or (h/ 27), respectively; and to retain the \njust-stated assumptions about Py and P4.\n\nThe question whether these models have any intrinsic truthfulness \nhas now become acute. If there is any doctrine in contemporary \natomic theory which appears to be multiply tested and approved, it is \nsurely the doctrine that the angular momentum of the valence-electron \nis always an integer multiple of 4/27. Yet in this passage I have \nspoken as if this principle had been indifferently and casually dis- \ncarded, and replaced by a new principle to the effect that the angular \nmomentum of the valence-electron is alwavs an odd-integer multiple \nof } h 2x. It is hard to evade or mitigate this arrant contradiction.\n\nA way out may possibly be found by suggesting that the partition \nof an atom into \u201cresidue\u201d and \u2018\u201c\u2018valence-electron,\u201d\u201d while appropriate \nwhen calculating energy-values by the method mentioned in Section \nP, is not appropriate in this instance; that the two whirling parts of \nthe atom are respectively a system composed of a part of the residue, \nand a system composed of the rest of the residue and the valence- \nelectron. This seems most admissible for such an atom as magnesium, \nconsisting as it is supposed of what I have called a \u201ckernel,\u201d and two \nadditional electrons outside. The two whirling parts may be the kernel \nrotating as a unit, and the pair of outer electrons also rotating as a \nunit. It may be profitable to push the analysis even further, and to \nconsider the two outer electrons each as an entity possessed of angular \nmomentum, their two angular momenta combining with one another \nin such a fashion as | have lately described for the two parts of the \natom; this resultant angular momentum of the two may then figure \nas the Py emploved in constructing the atom-model. There are \ndecided possibilities in this way of thinking; but it is doubtful whether \nthe dithculty about half-quantum-numbers can ever quite be removed.\u00ae\n\n* An unfortunate feature of Land\u00e9\u2019s model in its original form is that it requires \nus to believe that the residue of each atom is different from the completed pre- \nceding atom. For instance since Mg has a singlet and a triplet system, its residue \nmust have sometimes and sometimes P,=%h\n\nIt may be recalled from the First Part of this article that the ditterent \nStationary States of a group, sharing a common value of 7 and a com \nmon value of k, are distinguished from one an: ther bv having ditferent\n\ntransitions between two states, in each of which 7=0, are likewise \nmissing. This numeral is correlated with the angular momentum P 4 \nof the entire atom, in the theory here outlined. For systems of even \nmultiplicity, Pa is equal to jh 2x; for systems of odd multiplicity,\n\nThe various Stationary States of a group differ slight \notherwise, of course, they would never have been discerned. The \nenergy-value of an atom must be conceived therefore as depending \nnot merely upon 2 and k, not merely on the rates at which the two \nwhirling parts are separately spinning, but likewise upon then mutual \norientation and hence upon j.' [In this theory, the dependence of energy \nupon orientation must be postulated outright. We shall presently \nmeet with a case in which the dependence of energy upon orientation \ncan be foreseen, even in detail.\n\nIt appears from all these speculations, that a transition between \ntwo Stationary States is no longer to be cone ceived merely as a simple \nleap of an electron from one geometrically-detinite orbit into another. \nA leap is indeed supposed to occur, but it is accompanied by a turning- \ninward or a turning-outward of the axes of rotation of the two spinning \nparts of the atom. The radiation which comes forth isa joint product \nof these two processes, in which however no features of either sepat\n\nately appear; only the net change in the energy \u00ab \nalgebraic sum of the energy-changes due to each process separately, \nis radiated as a single fused unit. Nature does not make the separ-\n\nHaving used an orientation-theory to interpret the complexity of \nthe Stationary States, we will now consider an orientation-theory \ndeveloped to account for the effect of a magnetic field upon the Station- \nary States. There, it was supposed that the various States belonging \nto a single group are distinguished by various orientations of two \nspinning portions of an atom, relatively to one another. Here, it will \nbe supposed that the various States which replace each individual \nState, when a magnetic field acts upon the atom, are distinguished\n\nby various orientations of the spinning atom relatively to the fie! \nIt will presently be seen that the evidence for the orientation-theo: \nis much more abundant and more nearly direct, in this case of ma. \nnetically-excited Stationary States, than in that former case of mi \ntiplets. This case in fact was the earliest to which an orientation- \ntheory was applied; but for it, some quite different form of theory \nmight have been developed for multiplets. Even here the data ani \nthe theory are not entirely concordant; but the concordance is sv \nextensive, that the discord is sharply localized and identifiable.\n\nIrom the Second Part of this article (Section L) I quote the prin \nciple that an electron (of mass gw) revolving in an orbit with angulai \nmomentum P is equivalent to a magnet of which the magnetic momen! \nM is proportional to P, being\n\nBoth P and M are vectors normal to the plane of the orbit and henc \nparallel to each other. If several electrons are revolving in divers \nplane orbits about the same nucleus, their separate angular momenta \nmay be summed vectorially into a vector which is the angular mo \nmentum of the entire system, and their separate magnetic moments \nmay likewise be summed vectorially into a vector which is the mag. \nnetic moment of the entire svstem; and these two summation-vectors \nwill be parallel to one another, and related by the foregoing equation. \nHence a rigidly-connected revolving framework of electrons\u2014if such \na thing there be\u2014may be treated like a single electron, insofar as the \nratio of magnetic moment to angular momentum is concerned. Wher- \never in the course of this article we have envisaged electrons, kernels, \nor atoms revolving with angular momenta prescribed as integer \nmultiples of //27 or of $h/27, we might have imagined these as mag- \nnets with magnetic moments prescribed as integer multiples of eh /4ruc \nor of }eh/4ruc.2 This is not necessary; though the relation between \nangular momentum and magnetic moment is derived directly from an \nequation valid for perceptible electric currents, it might not be true \nfor individual electrons. Nevertheless we shall arrive at striking \nresults, by supposing that it is.\n\nWhen a magnetic field is applied to a multitude of radiating atoms, \nmost of the lines of their spectrum are replaced by groups of several \nlines each, or \u201csplit up\u201d? into several components, as the phrase ts. \nThis signifies that each of the Stationary States of each atom is ap- \nparently replaced by several. One may infer that when an atom is\n\n* The quantity e/4ruc, the presumptive magnetic moment of an electron circu- \nlating in an orbit of angular momentum h/2z, is known as the Bohr magneton.\n\nintroduced into a magnetic field, each of its Stationary States is modi- \nfied into one or another of several new States, differentiated from one \nanother and from the original State to a small but appreciable extent. \nThis might arise from some distortion or internal alteration of the \natom by the field; and it will probably be necessary to adopt this view \nin some cases. But there is also a simpler effect which the magnetic \nfield may have upon the apparent energy-values of the Stationary \nStates, an effect not involving any deformation of the atom by the \nfield\u2014to wit, an orientation-effect similar to that which was assumed \nto account for multiplets. This we proceed to examine\n\nIf an atom which is a magnet is floating in a magnetic field, it \nexperiences a torque which tends to orient it parallel with the tield \nBy saving that an atom is parallel or oblique to the field, | mean \nthat the magnetic moment of the atom and therefore also its angular \nmomentum, are directed parallel or obliquely to the field; and this \nusage will be maintained. Owing to this torque it is endowed with \nenergy due to the field, in addition to its own intrinsic energy; this \nadditional energy, which depends upon the inclination of the atom to \nthe field, I shall call its extra magnetic enerey. If the atom turns in \nthe field, the amount of its extra magnetic energy changes; and if its \nmagnetic moment suddenly changes, its extra magnetic energy also \nchanges unless it simultaneously turns by just the right amount to \ncompensate the change. If at the moment of passing over from one \nof its Stationary States to another, its inclination or its magnetic \nmoment or both are changed; the amount of magnetic energy which it \ngains or loses will be added (or subtracted, as the case may be) to the \namount of energy which it gains or loses because of the transition. \nThe frequency of the radiation sent out or taken in by the atom will \nbe equal to 1/h times the sum of two energy-changes of distinct \nkinds\u2014not, as in the absence of magnetic field, to 1\u2019 times the \nenergy-difference between the two Stationary States alone. Thus \nthe effect of a magnetic field upon spectrum lines might be ascribed, \nnot to any deformation of the atom by the field, but to changes in the \norientations or in the magnetic moments of > the atoms occurring \nat the instants when they make their transitions. The question for \nus now is, whether the actual details of the observed etfect can be \ninterpreted in this manner.\n\nExpressing the foregoing statements in formulae, in which MJ \ndenotes the magnetic moment of an atom, //J the magnetic field, and \na the inclination of the atom to the field, we have for the torque which \nthe field exerts upon the atom\n\nIn this last expression it is tacitly assumed that the extra magnetic \nenergy is zero when the atom is oriented crosswise (at right angles) to \nthe field. This is not an arbitrary, but a quite essential convention, \njustified from the atom-model.'\"\") Suppose now that the atom passes \nbetween two stationary states S\u2019 and S\u201d, in which its internal energy, \nits magnetic moment and its inclination are denoted by U\u2019, A\u2019, a\u2019 \nand U\u201d, JM\u201d and a\u201d, respectively. Were there no magnetic field, \nthe frequency radiated would be\n\nthe term Av representing the displacement of the line by the field. \nThe question is, whether this term can be equated to the observed \ndisplacements.\n\nConsider the most tractable cases, those in which the so-called \n\u201cnormal Zeeman effect\u2019 is observed. In these cases a line of fre-\n\nquency vy, is replaced by three, of which the frequencies are \nVotwll, vo, ve\u2014wll (12)\n\ncorresponding to three values for the displacement Av, which are \nexpressed by\n\nThe quantity \u00bb occurring in these expressions is a specific numerical \nconstant. Comparing these with the expressions for Av in (11), we \nsee that if our model is to be used to interpret the observations, then \nfor the first of the three observed lines J/\u2019 cos a\u2019 must be greater \nthan MW\u201d cos a\u201d by the amount / @; for the second, AZ\u2019 cos a\u2019 must \nbe equal to J/\u201d cos ae\u201d; for the third, AZ\u2019 cos ae must be less than \nM\u201d cos ae\u201d by the amount wh.\n\nAnother way of putting these statements is, that in order to interpret \nthe normal Zeeman effect in this manner it must be supposed that \nwhenever a transition occurs, the projection of the magnetic moment\n\n\u201d The action of the magnetic field upon the revolving electron imparts to it an \nextra angular velocity about the direction of the field (the Larmor precession) and \nhence an extra kinetic energy which (to first order of approximation) is proportional\n\nto \u2014cos a and is zero when a=7/2. This extra kinetic energy is the extra magnetic \nenergy AU. It is profitable to_derive the entire theory in this manner.\n\nupon the direction of the field\u2014for this is precisely what / cos @ is \neither does not change at all or else changes by +wh. Sometimes \nit acts in the first of these wavs, sometimes in the second, sometimes \nin the third; but never in any other.\n\nThis would result, if the behavior of the atom floating in the mag- \nnetic field were governed by two rules; first, that it may orient itself \nonly in certain \u201c\u2018permitted\u201d directions such that J cos a, the pro- \njection of its magnetic moment upon the tield-direction, assumes \n\u201cpermitted\u201d values which are integer multiples of wh; second, that \nwhenever a transition occurs J/ cos @ either retains the value which \nit had initially, or else passes to one or the other of the two adjacent \npermitted values.\n\nThe first of these rules is stated more rigorously than is quite neces- \nsary; all that is required is to sav that J cos @ is permitted to take \nonly such values as belong to an equally-spaced series with intervals \nequal to wh. The second rule is necessary.\n\nThe theory of the normal Zeeman effect is simply, that the atom \ndoes behave according to these rules. Radiation of th\u00e9 frequency v \noccurs, either when the magnetic moment of the atom does not change \nand the atom does not turn, or when the magnetic moment changes \nand simultaneously the atom turns just so as to keep the projection \nof the magnetic moment on the field-direction constant. We shall \nlater see that the latter of these two alternatives is the accepted one. \nIt must be supposed that the atom, so to speak, capsizes when it \nemits the frequency v, while floating in a magnetic field; it flops over \nat the same moment as it passes from one stationary state to another. \nRadiation of the frequency vo+Av or of the frequency v,\u2014Av occurs, \nas we shall see, when the magnetic moment of the atom changes; in \nsome cases the atom capsizes during the process, in others it does not.\n\nI now translate the foregoing rules from the language of magnetic \nmoments to the language of angular momenta. The first rule is, \nthat the atom may orient itself only in certain permitted directions \nsuch that P cos a, the projection of the angular momentum upon the \ndirection of the magnetic field, assumes permitted values which are \nconsecutively spaced at intervals of (2uc e)wi.\n\nNow it is a fact of experience, that in the cases of the normal Zeeman \neffect,\n\nThe rule therefore reads, that the projection of the angular momentum \nof the atom upon the direction of the magnetic field is constrained to take \ncertain permitted values, spaced at intervals of h/2r.\n\nWe have supposed, in dealing with multiplets, sometimes that the \nangular momentum of the entire atom is constrained to take such \nvalues as are integer values of 4/27, and sometimes that it is con- \nstrained to take such values as are odd-integer multiples of } (h/2z7).\" \nIn either case the permitted values of the angular momentum are \nspaced at equal intervals; and as the rule for the component of the \nangular momentum along the direction of the field bears the form \nwhich it does, we may well suppose that something in the order of \nnature constrains both the angular momentum and its projection to \naccept only values which form a sequence spaced always at that \ncurious interval h, 27.\n\nThe total number of permitted orientations will obviously be \nlimited by the actual magnitude P of the angular momentum. This \nbeing supposed always to be an integer multiple of 3 2/27, let it be \nwritten P=2/(4 h/2x). The permitted orientations are those which \nyield a series of values for the projection P cos @ spaced at intervals \nh/27; let these be written\n\nof which there are (2/+1) in all. On the other hand, it may be that \nthe atom is prevented from orienting itself parallel to the field; that \nthe least permitted angle between the axis of the atom and the direc- \ntion of the field is some angle yielding a projection A\u00bb intermediate \nbetween P and (P\u2014h,/27). Then there are 2J permitted orientations \naltogether.\n\nSummarizing the results of this last paragraph: if the angular \nmomentum of the atom is an integer multiple 2/(3 h/27) of the fun-\n\n\" It was remarked at the beginning of Section S that the evidence to be presented \nin that Section would support neither 4/27 nor any other particular numerical value \nfor the fundamental unit of angular momentum; here, however, we have evidence \nfor that value.\n\nthe atom is permitted to take either (2J+1) or 2/ distinct orienta- \ntions in the field; the former number if it is, the latter if it is not \npermitted to set itself quite parallel to the field.\n\nIt will now be shown that these are by no means idle speculations; \nthey bear directly upon certain facts accessible to observation. Before \nbringing up these facts it is necessary to abandon the policy of speak- \ning exclusively about the \u201cnormal\u201d Zeeman effect. This \u2018normal\u2019 \neffect received its adjective because it agrees so excellently with the \noriginal theory devised years before quanta were dreamt of to explain \nthe effect of magnetic field upon spectra. It is essentially because \nof this agreement that it is possible to develop the contemporary theory \nof the \u201cnormal\u201d effect in a perfectly deductive fashion, using no \nnew assumptions beyond those general ones of the orientation-theory. \nMost spectrum lines, however, are affected by a magnetic field in ways \nnot compatible with the original theory; which is a consequence of \nthe fact that the set of new Stationary States, whereby a magnetic \nfield supplants each original Stationary State, in most cases does not \nconform to the laws previously set forth.\n\nThe laws to which it generally does conform were read from the \nspectra by Land\u00e9. The one feature in which the foregoing theory \nquite agrees with these laws is its prediction of the total number of \nStationary States. A Stationary State for which the angular mo- \nmentum of the atom is determined, by virtue of the theory of multi- \nplets which filled the preceding section of this article, as being \n2J (jh/2mr), is actually found to be supplanted, when a magnetic \nfield is impressed upon the atom, by 2J new Stationary States. This \nis in agreement with one of the two alternative predictions made a \nfew paragraphs supra; to wit, with the prediction derived from the \nassumption that the atom cannot set itself quite parallel to the \nfield. This agreement between the orientation-theory of multiplets \nand the orientation-theory of Zeeman effect considerably strengthens \nboth.\n\nThe several Stationary States replacing a given original State \nare always equally spaced; but the spacing differs in amount \nfrom the value wl/h or ellh/4zuc exhibited when the normal \nZeeman effect occurs, and which we found it possible to deduce \nfrom the simple orientation-theory. The difference is this, that \nthe actual spacing is a multiple of the value w//h by a factor g \n(generally lying between } and 2) which depends upon the original \nState:\n\nThe only ways hitherto used to accommodate the atom-model \nthis surprising and inconvenient factor g are tantamount to assumi!\n\nthat it enters into the relation between angular momentum J/ and \nmagnetic moment P? which was derived in Section L and written down \nhere as equation (7); which relation is accordingly modified without\n\na very unsatisfying procedure. Lande found it possible to mitigat \nthis process somewhat and at the same time produce a partial ex- \nplanation of the formula quoted in the First Part of this article, \nwhereby g is related to the factors A, R and J which, in the atom- \nmedel of the two whirling parts, measure the angular momenta of \nvalence-electron and residue and entire atom respectively in terms of \nthe common unit 4 27. This explanation involved the postulate \nthat g=1 for the valence-electron and g=2 for the residue. It would \ntherefore be necessary to justify, or to postulate without justification, \nnot a multitude of such relations as (16) with a multitude of unforeseen \nvalues of g, but only a single such relation with a single unforeseen \nvalue of g. This is bad enough, but not so bad as if it were inevitable \nto assume that 1/7, P may have a dozen different values in different \ncases.\n\nIt is now the occasion to recur to the extraordinary experiments \nof Gerlach which disclose the magnetic moments of individual atoms \nand verify the supposition that certain orientations are permitted \nand others inhibited. These experiments having already twice been \nmentioned in this series of articles, I shall spend no more space upon \nthe method than is necessary to say that atoms in a narrow stream \nare sent flying across an intense magnetic field with a strong field \ngradient, by which they are drawn aside. Were the atoms tiny \nmagnets oriented randomwise in all directions, the beam would be \nbroadened into a fan; one edge of the fan would be the path of atoms \noriented parallel to the field, the other edge would be the trajectory \nof atoms oriented anti-parallel to the field, while the space between \nthe edges would be filled by the orbits of atoms pointed obliquely \nto the field. Actually Gerlach observed not the whole fan, but two \nor several separate diverging pencils of atoms, and between them \nvacant regions traversed by none. Certain orientations of atoms to \nfield were unrepresented in the beam. Here for the first time there is \ndirect evidence of discrete Stationary States, of quantum permissions \nand quantum inhibitions, not deduced from observations upon transi-\n\ntions but drawn forthright from viewing atoms in equilibrium in their \nNormal States.\n\nWhen from the diverging pencils one proceeds to determine the \norientations of the atoms and their magnetic moments, one is con- \nfused by a possibility made clear in the foregoing pages, but unsus \npected at the time when the first of these experiments were performed. \nI illustrate with the case of silver, the atoms of which flock into two \ndiverging pencils with a quite vacant space between. At first it was \nnaturally supposed that one pencil consists of atoms oriented parallel, \nthe other of atoms oriented anti-parallel to the tield. The detlections \nof the two pencils are such, that if this assumption is true then the \nnumerical value of the magnetic moment of the silver atom agrees \nwithin the error of experiment with the value of eh 4druc \u2014agrees, \ntherefore, with the notions that the angular momentum of the silver \natom in its normal state is #27 and that the magnetic moment stands \nin the right and proper ratio e 2uc to the angular momentum. The \ndata were supposed to prove these notions. They also agree, how- \never, with the suppositions that one pencil consists of atoms inclined \nat 60\u00b0 to the field and the other of atoms inclined at 120\u00b0; in which \ncase the magnetic moment of the silver atom would be eh 4ruc, \nsuggesting that the ratio of magnetic moment to angular momentum \nhas twice the right and proper value. This inextricable tangling of \nthe effect of orientation with the effect of magnetic moment makes it \nimpracticable to deduce quite so much from the data as was at first \nthought possible; but plenty still remains. It is found that copper and \ngold behave like silver, as was to be expected from their positions \nin the Periodic Table. It is found that lead atoms, and (most sur- \nprising of all!) cron atoms are not deflected at all; so that either the \nmagnetic moments of their parts balance one another completely, or \nelse they all orient themselves crosswise to the field. Nickel, on the \nother hand, behaves as though its atoms had each a magnetic moment \nsurpassing 2eh,4auc, while thallium responds as though that of its \natoms were much less than eh 4ruc. Finally \u2014lest the results seem \ntoo gratifving\u2014it is found that bismuth atoms are deflected in a \nmanner quite unforeseeable.\n\nThere is not time nor space to speak of the other method for de- \ntermining the magnetic moments of atoms, by measuring the sus- \nceptibilities of great quantities of them in gases or solutions; but the \nmeasurements so made are also very helpful in determining the \nmagnetic moments of various atoms and ions various groupings, \nthat is to say, of electrons around nuclei.'\"? All such data are of im-\n\n>For the status of such measurements in 1923, the first of this series of articles \nmay be consulted. (This Journal, September, 1923.)\n\nmense value, and no theory of the atom can be spared from the demai \nthat it confront them and account for them.\n\nBy the term \u201cX-ray\u201d the reader may understand any radiatior \nof which the frequency v is so high, that the energy Av of a singk \nquantum is several times as great as the energy required to remov \nthe most-easily-detached electron from an atom; greater, for instance \nthan 100 equivalent volts, so that the wavelength of the radiation is \nless than some 125 Angstrom units. The emission or the absorption \nof such radiation by an atom involves too great an energy-change \nto be attributed merely to a displacement of the valence-electron or \neven to combined displacements of the valence-electron and one or \ntwo others. This definition leaves a sort of \u201ctwilight zone\u201d of radia- \ntions having frequencies somewhat but not much greater than 1 4 \ntimes the ionizing-potential of an atom. Little is known about such \nradiations, and in this place they will not be considered.\n\nThe absorption of an X-ray quantum by an atom results in the \nextrusion of an electron from the atom. The emission of an X-ray \nquantum results from the passage of an electron within the residue of \nthe atom from some original situation to the situation vacated by \nthe extruded electron, or else into a situation vacated by an electron \nwhich itself has moved elsewhere within the atom. These state- \nments contain the theory of the vast amount of data piled up by \nobservations upon the emission and absorption of X-rays by matter.\n\nTo express the same statements rather differently: X-ray absorp- \ntion-spectra and X-ray emission-spectra reveal, when analyzed for \nStationary States in the manner used in analyzing optical spectra, \nthat each atom with several or many electrons has a considerable \nnumber of Stationary States, distinguished from those we have here- \ntofore discovered in that each of them involves the absence of one electron \nfrom the atom. Each of them is therefore strictly an \u201c\u2018ionized-atom \nstate,\u2019 and yet there are several of them with extremely different \nenergy-values. This signifies that the extraction of an electron from \nan atom rich in electrons may leave the residue in any one of several \ndistinct conditions. These distinct conditions are the distinct Sta- \ntionary States, transitions between which are responsible for X-ray \nspectra. Owing to this striking difference between the Stationary \nStates hitherto described and these latter, I shall refer to these as the \n\u201cX-ray Stationary States\u2019\u2019\u2014not that this name is a_ particularly \ngood one.\n\nAbsorption of an X-ray quantum by an atom, then, results in a \ntransition of the atom from its normal state to one of the \u2018X-ray \nStationary States.\u201d Emission of an X-ray quantum by an atom \nresults from a transition of the atom from one into another of its \n\u201cX-ray Stationary States\u2019\u2019\u2014a transition which begins in a condition \nin which the atom lacks one electron, and ends in another condition \nin which the atom lacks one electron. To take instances: radiation \nof an adequate frequency falling upon an atom in its normal state \nmay put it over into an X-ray Stationary State known as the L;; state, \nan electron being extruded. Radiation of an adequate frequency \n(a higher frequency will be required) falling upon a similar atom in \nits normal state may put it over into another X-ray Stationary State \nof higher energy, known as the A state, an electron being extruded. \nThe atom in the A state may then spontaneously pass over into the Ly; \nstate, emitting a radiation belonging to the X-ray emission-spectrum, \nits frequency being 1/A times the energy-difference between the K \nstate and the Ly, state. Later the atom may pass into still another \nstate, such as the J; state, by emitting radiation of some frequency \nv\u2019; the energy of the J/; state is therefore less than that of the Ly, \nstate by the amount fv\u2019; calculating it thus, and then applying to nor- \nmal atoms a stream of electrons or of quanta having energy just \nadequate to put them over into this J/; state, we find that this \neffect is duly produced.\n\nThus there is a thoroughgoing analogy between the genesis of \noptical spectra by transitions between the optical Stationary States, \nand the genesis of X-ray spectra by transitions between the X-ray \nStationary States. The differences between the two kinds of spectra \nseem all to derive from the one fundamental difference between the \ntwo kinds of Stationary States; the former do not involve the absence \nof an electron from an atom, the latter do. In the optical region, \nfor instance, we find that an atom in the normal state cannot be put \ninto a particular excited state by any radiation except one of just \nthe right frequency v, for which /iv, is equal to the energy-difference \nbetween the normal state and the excited state in question. In \nthe X-ray region, we find that an atom can be put into the A-state \n(for instance) by any radiation of frequency equal to or exceeding \nthat critical frequency \u00bb, for which hv, is equal to the energy-difference \nbetween the normal state and the K state. This difference in be- \nhavior occurs because in the former case a quantum of radiation \nhaving frequency v exceeding \u00bb, would have no place to put the left- \nover energy h(v\u2014yv.), whereas in the latter case this extra energy \ncan be and is delivered over to the extracted electron as kinetic\n\nenergy, with which it flies away. This is known positively; for \nextracted electrons can be observed, and their energy measured.\n\nSpontaneous transitions from each of the X-ray Stationary Stat: \noceur to some, but not to all, of the States of lesser energy. Sor \nare evidently inhibited; and it is possible to lay down rules of sel \ntion, distinguishing those which are permitted. The complicated \nsystem of rules originally proposed has yielded place to a much simple: \none, exactly similar to the one prevailing in the optical spectra. That \nis to say: it appears to be possible to assign to each of the X-ray \nStationary States a certain value of a numeral k and a certain value \nof a numeral j, such that the only transitions which actually occur \nare those in which & changes by one unit and / either changes by on \nunit or does not change at all; while transitions between states in \nboth of which 7=0 are specially excluded. Furthermore, the various \nvalues of k and j thus assigned to the several X-ray Stationary States \nare identical with those assigned to the several States constituting a \ndoublet svstem, such as we have met already in Section S, such as \nthe sodium atom possesses; so that there is a complete correspondence \nbetween the system of X-ray Stationary States which every atom \nrich in electrons possesses, and the doublet system of optical Station- \nary States which only certain atoms possess. A part of this corre- \nspondence is expressed in the following Table:\n\nTABLE Il \nValues of k: 1 1 2 2 1 2 2 3 3 \nValues of 7: 1 1 1 2 1 1 2 2 3 \nStationary States of \nDoublet system: Is 2s 2pi 2pr 3s 3p2 \nX-ray system: iN L, Li Lin M, My Mir Mi My\n\nNo doubt the implications of this close correspondence are deep; \nbut just what they are is not yet obvious.\n\nThe fact that the residue, left behind when an electron is extracted \nfrom an atom, may exist in any one of several distinct States, is quite \nnaturally interpreted as meaning that the various electrons of the \ncomplete atom are variously situated, or revolving in various distinct \norbits; so that the several X-ray Stationary States differ essentially \nin this, that ditterently-located electrons have been removed, leaving \ndifferent places untenanted. This notion is easily combined with the \nidea that an atom is formed, or at all events behaves as though it \nhad been formed, by successive self-annexations of electrons to a\n\nsecutive adhesions of electrons, each of which settles down into a \npeculiar orbit and remains there more or less unperturbed as the later \n~omers immigrate one after the other into the svstem. Mav not \nthen the process of X-ray absorption consist in a powerful intruding \nentity, electron or quantum, violently invading the interior regions of \nthe atom and casting out one or another of the deeper-lying earlier- \nadded electrons, while the later-added ones nearer or upon the frontier \nremain attached?) May not X-ray emission consist in the passage of \none of these latter electrons into the orbit formerly held by its prede- \ncessor, now unexpectedly reft away and its place left empty?\n\nAlthough an affirmative answer to these questions involves a very \nliteral and concrete conception of electron-orbits, most physicists \nmake it, and would like to prove it. The chief difficulty lies in the \nfact that all information about X-ray Stationary States is primarily \ninformation, not about the prior condition of the electron which is \ngone, but about the final condition of the residue which is left behind.\n\nThe data show at all events that there are not nearly so mans \nconditions of the residue, as there are electrons of the completed atom; \nfrom which it is fairly safe to conclude that the electrons of the atom \nare so arranged, that any one of several different electrons may be \nremoved and the residue be left always in the same condition \ntherefore, that the electrons are arranged in groups, each electron \nbeing situated essentially like every other of its group. In discussing \nthe formation of an atom by successive binding of electrons, it was \nremarked that several electrons may be bound in orbits each char- \nacterized by a common value of \u00bb and a common value of &. These \ntwo ideas may coincide; and great efforts are being made to bring \nthem into entire coincidence. The evidence indicates, for example, \nthat the first ten electrons bound to a nucleus are divided into four \ngroups. Absence of an electron from one of these groups entails \nthat the atom is in the A state; absence of an electron from the second, \nthird, or fourth group brings it about that the atom is in the L,, or \nLy, or Ly, state, respectively. So much the X-rav data do show \nrather definitely; although the actual number of electrons in each \nof the four groups cannot yet be deduced. If one could prove a priori \nthat the first ten electrons annexed by a nucleus settle down into \norbits of four distinct kinds, the achievement would be a great one. \nIntimations that something of this sort has been achieved are made \nevery now and then; but it is diftheult to tell whether the assertions \nwhich are made have been derived cogently from a principle or are \ninspired guesswork.\n\nof which was until a couple of years ago regarded as perfectly d \ntinct: but at this moment it is beclouded by one of the curious c \ntradictions so abundant in the Theory of Atomic Structure. Brief \nthe typical phenomenon is this: the differences between the energy- \nvalues of the K, Ly, and Ly, states agree notably well with wha: \nwould be expected if the complete atom contains a few electrons \nmoving in 1; orbits, a few in 2; orbits and a few in 22 orbits about the \nnucleus; and if the K state corresponds to absence of an electron of the \nfirst group, the Ly, state to absence of one out of the second and the \nLy; state to absence of one out of the third. (The reason why calcu- \nlations can be made for so indefinitely-phrased a model is this, that \nthe field due to the highly charged nucleus of a massive atom should \ndominate over those of the individual electrons so that it does not \nmake very much difference how many are supposed to be in each \ngroup.) The natural inference is, that the rest of the atom remains \nunchanged or little changed when any one of these electrons is ex \ntracted. In this case the Azimuthal Quantum-number of the residue \nshould differ by one unit for the two states Ly, and Ly. Consulting \nTable II, one finds that the quantity called k, which obeys the char- \nacteristic selection rule of the Azimuthal Quantum-number, is the \nsame for Ly, as for Ly. This is an illustration of the collisions be- \ntween two sets of inferences which unsettle the supposedly firmest \nachievements of this theory.\n\nOf the theory of molecules, a subject large enough for an article \nby itself, | can say here nothing more than that it attains some re- \nmarkable successes, achieved by and therefore fortifying some of the \nassumptions made in these pages; notably the assumption that \nAngular Momentum is a thing required in Nature to assume discrete \nvalues spaced at intervals of h, 27.\n\nThe tinal part of this long article has been very unlike the Second \nPart, in which an atom-model for the atoms of hydrogen and ionized- \nhelium was constructed and endowed with certain fundamental qual- \nities, so that it reproduced almost all of the relations of these atoms \nto radiation with a truly striking fidelity. This Third Part by con- \ntrast has been a thing of shreds and patches. Models for many atoms \nhave been brought forth, but they have not been thoroughly adequate \nand they have not been concordant with one another. Some were \ndesigned with the same fundamental qualities as those given to the \nmodel for hydrogen; and scarcely more can be said for any of them, \nthan that it does not positively clash with the properties of the element \nfor which it is devised. Others were made competent to deal with a\n\ncertain limited set of facts (as the grouping of terms in multiplets \nby giving them qualities gravely in discord with those attributed \nto the atom-model for hydrogen; and then they proved themselves \nsurprisingly well able to account for isolated facts of quite a different \nsort (as the effect of magnetic fields upon atoms). The presentation \nin these pages is naturally very far from complete; had it been com- \nplete, it would have filled a book and not an article. But if it had \nbeen complete, the eventual impression would have been the same; \nan impression of confusion, vet of a confusion full of hope.\n\nFor the \u201cTheory of Atomic Structure\u201d is distinguished especially \nby this, that it is not one theory but a multitude of partial theories, \neach designed and competent to cover a limited family of the abound- \ning data, each struggling to overlap and to absorb the others. It \nmay be compared with a cross word-puzzle or a map-puzzle, in which \nthe beginnings of a solution have been made in half-a-dozen corners \nand patches, while wide blank areas adjoin and separate them, and \nsome of the partial solutions already entered upon the tield may finally \nyield to others which can be unified into the perfected pattern. Or \nit may be compared with those maps of polar regions, in which here \nand there a properly-surveved island or little strip of coastline emerges \nfrom the blankness of the unexplored realms, and some of them are \ncertainly misplaced relatively to the others and will be shifted on the \nmap when all the geography is at length made known. Or it may be \ncompared with the state of a congealing metal, in which a multitude \nof little crystals have formed themselves about casual nuclei of crys- \ntallization; each is oriented in a different way, and when two of them \ngrow into contact with one another thev clash and cannot merge, \nthey stand blocking and thwarting one another. It may be neces- \nsary to reliquefy them all and make a new attempt to change the \nformless mass into a single crystal.\n\nMeanwhile the work is driven forward with the fervor of discovery \nand exploration, in this period which Russell finely called the \u201cHeroic \nAge of Spectroscopy,\u201d\u2019 and not of spectroscopy alone. Many, though \nnot so many as are needed, are busy with determining the Station- \nary States by deciphering the rich and cryptic spectra of some among \nthe numerous unstudied elements\u2014enormously numerous, taking into \naccount how many kinds of ionized atoms there are; and others with \nthe assembling of new photographs of spectra made under the most \nvaried sorts of excitation, with other aids to discriminating the lines; \nand others with the impressing of electric and magnetic tields upon \nradiating atoms; and others are engaged in measuring the intensities \nof lines. Yet other experimenters are determining the magnetic\n\nmoments of various atoms in all the possible ways. Some are seeking \nnew phenomena which may result from Stationary States and fron \ntransitions, and occasionally they are rewarded with brilliant example: \nsuch as that vivid demonstration of the atom-magnets which Gerlach \neffected, or such as the passage of an atom from one State to anothe: \nwhile it transfers the liberated energy to another particle directly \nand produces a chemical change. Others are finding the processes \nresulting from the Stationary States manifest on an unearthly scale \nwithin the stars.\n\nThe theorists likewise are at work with furious industry. Now and \nthen a set of data hitherto rebellious is suddenly systematized, usually \nin a manner not quite concordant with the other theories holding \nother parts of the field. Attempts are made to unify one partial \ntheory with another, usually unsuccessful. Sometimes an authorita- \ntive thinker, despondent over the continuing contradictions, tries to \ncut all the knots by declaring that one or another of the conflicting \nmodels is entirely fallacious, and that the numerical agreements on \nwhich it is founded are a delusion and a snare. Another is driven \nto concede that the conflicts are destined to endure forever, and \naccepts all of the partial theories as equally valid, or else paraphrases \nthem in ingenious words which veil the contradictions, yet leaving \nthese essentially unabated. Others, abandoning the general problem, \nhave returned to the question of the hydrogen atom, and for this they \nare trying to rephrase or reshape the Quantum Conditions in a manner \nmore satisfactory to themselves; sometimes with the aid of new and \nunfamiliar forms of mathematics, apparently expecting that when \nthese become habitual to the human mind, the mystery of the Quan- \ntum Conditions will seem simple and clear. That, of course, always \nremains a possibility\u2014that the human intellect will accustom itself \nso thoroughly to the new systems of ideas that they will cease to \nseem incoherent, as the human ear has so accustomed itself to the \nharmonic innovations of successive generations of musicians that the \ntones which seemed outrageous discords to the audiences of Beethoven \nnow are to us monotonously sweet. To our minds the various divi- \nsions of the Atomic Theory are still discordant. It would not be \nfair to leave any other impression of this strange and fascinating \ntheory; inchoate but full of promise, immature but gathering force, \na fantastic assemblage of failures and successes; irreconcilable with\n\nall other theories, irreconcilable even with itself, and yet perhaps \npredestined to refashion all the science of physics in its own image.\n\nSynopsis: The paper is based on radio transmission tests from station \n2XB in New York City to two outlying field stations. It is a detailed study \nof fading and distortion of radio signals under night time conditions in a par- \nticular region which may or may not be typical.\n\nNight time fading tests using constant single frequencies and bands of \nfrequencies in whic h the rece iving observe itions were recorded by os illograph \nshow that the fading is selective. By selective fading it is meant that \nent frequencies do not fade toge \u2018ther. From the regularity of the frequency \nrelation between the frequencies which fade tog ether it is cone luded th it \nthe selective fading is caused by wave interference. The signals appear to \nreach the receiving point by at least two paths of different lengths. The\n\npaths change slowly with reference to each other so that at different time s \nthe component waves add or neutralize, going through these conditions \nprogressively, The two major paths by which the interfering waves te ivel\n\nare calculated to have a difference in length of the order of 135 kilometers for \nthe conditions of the tests. Since this difference is greater than the distance \ndirectly from transmitter to receiver it is assumed that oe path at \nmust follow a circuitous route, probably reaching upward th rough higher\n\natmospheric regions. Various theories to explain this are briefly reviewed.\n\nThe territory about one of the receiving test sti itions in Connecticut is \nfound under day time conditions to be the seat of a gigantic fixed wave inter \nference or diffraction pattern caused in part by the shadowing of a group of \nhigh buildings in New York City. The influence of this pattern on night \ntime fading is discussed. It is considered a contributing but not the control} \ning effect.\n\nTests using transmission from an ordinary type of broadcasting trans- \nmitter show that such transmitters have a dynamic frequency instability or \nfrequency modulation combined with the amplitude modulation. At night \nthe wave interference effects which produce selective fading result in  dis- \ntortion of the signals when frequency modulation is present. It is shown \nthat stabilizing the transmitter frequency eliminates this distortion. A \ntheory explaining the action is given. The distortions predicted by the theory \ncheck with the actual distortions observed.\n\nA discussion of ordinary modulated carrier transmission, carrier sup- \npression, and single side band transmission is given in relation to selective \nfading It is shown that the use of a carrier suppression system should \nreduce fading.\n\ntechnique of radio telephone broadcasting is to consolidate and \ncontinue its remarkable progress, is the mechanism of the transmission \nof radio signals through space. In many receiving situations \nlargest apparent defects present in the reproduced signal are those \nsuffered not in the terminal apnaratus but in transit through space, \nand in these cases better methods of utilizing the transmitting medium \nIn the present \npaper we are reporting some investigations in this field of radio trans- \nmission which have uncovered a number of interesting facts and have \nled to at least one conclusion which is of practical utility.\n\nNight time transmission, which is the usual case in broadcasting, \nin many places commonly marred by fading and sometimes by actu \ndistortion of signals. Often these occur in certain areas not mor \ndistant from the transmitting station than other areas which enjo: \nfreedom from such annoyance. Selecting a particular instance of thes \ndifficulties in an area near New York City which, in so far as can ly \njudged at present, is probably a typical instance, we have subjected i \nto an intensive experimental study to determine what is the inheren: \nnature of the troubles and if possible how they may be alleviated \nIn doing this it has been necessary to employ novel forms of tests \nespecially fitted to bring out in a concrete way the phenomena being \ninvestigated.\n\nAs the radio art has progressed from spark telegraphy into continu- \nous wave telegraphy and into high quality radio telephone broadcast- \ning, increasing demands have been made on the transmission medium \nto deliver at the receiving point a true sample of what was put into it \nat the transmitting station. The requirements have grown in rigor \nbecause in telegraphy the end has been to develop increased reliability \nof communication at longer ranges and in telephony the medium is \ncalled upon to transmit a highly complex form of intelligence.\n\nOf the requirements placed on the transmission medium by modern \nuses, those imposed by telephony are far more exacting than those for \ntelegraphy. In telegraphy a single frequency, or at most a narrow \nband of frequencies sent out intermittently in accordance with a dot \nand dash code must reach the receiving station in such shape that it \nmay be converted into audible sound for aural interpretation or into \ncurrent pulses for the operation of relays or recording instruments. \nLeaving aside noise, the principal requirement is a sufficient freedom \nfrom fading so that signals can be interpreted or recorded without \ninterruption. In radio telephony, as at present practiced in broadcast- \ning, there is transmitted a modulated high-frequency wave comprising \na relatively wide band of frequencies, usually at least 10 kilocycles.\n\nSuch a modulated high-frequency wave drawn out in the familiar \ngraphical representation is a comparatively simple-looking thing, but \nanalyzed into its elements and studied in detail it is revealed as being \nan intricate fabric of elemental waves so interwoven with each other \nthat no one of them can be disturbed without changing in some \ndegree the complexion of the whole. For perfect results the whole \nband must arrive at the receiver with an amplitude continuously \nproportional to that leaving the transmitter, or the inflections or expres- \nsion of the speech or music will not be correctly reproduced. All the \ncomponent frequencies within the band must be unchanged in their \nrelative amplitudes lest the character of the sounds be altered. Even \nthe relative phase relations of the various frequencies must be preserved \nor, as will be shown later, the interaction of the two side bands in the \nreceiving detector will result in the partial loss of some of the frequency \ncomponents,\n\nIt is not long since the time when radio was supposed to be the \nperfect medium for voice transmission it being presumed that since the \nether of space (if there be such a thing) was substantially perfect in \nits electrical characteristics it must transmit frequency bands carrying \ntelephone channels without distortion of any kind. This may be true \ntheoretically of a pure ether but in fact, the ether used for radio com- \nmunication is filled with a number of things ranging from gaseous \nions down to the solid bed rock of the earth. It is rather to be ex- \npected that these will affect the progress of electromagnetic waves \nand we know from experience that they do. Diurnal variations of \nattenuation, fading, directional changes, dead spots and the like \nare already well known phenomena resulting from the complexity \nof our transmission media, although no entirely adequate explana- \ntions of their causes have been certainly established. One of the \nmost recent manifestations of the effects of irregularities in\u2019 trans- \nmission through space is in the distortion of the quality of telephone \nsignals. \u2018This was perhaps first noticed in the use of short waves for \nbroadcasting it being found that frequently the transmission was so \ndistorted that after detection the signals such as speech and music \nwere in severe cases almost unrecognizable.\n\nFor some time after quality distortion was recognized as a character- \nistic of existing short wave transmissions, it was thought that for the \nlower broadcasting frequencies at least, it was present only at night \nand at relatively very great distances from the transmitter. However,\n\ncareful observations demonstrated that there were points relativel \nnear New York City where quality distortion from several broad \ncasting stations in the city was marked at night and in at least on \ncase was detectable even in daytime. When station 2XB the Bel! \nTelephone Laboratories\u2019 experimental station at 463 West Street, \nNew York City, was used to transmit test signals, it was found that \nquality distortion could be observed in northern Westchester county \nand in southern Connecticut at distances of about 30 to 50 miles from \nthe transmitter. Fading was also pronounced and it was noted as a \nsignificant fact that distortion was always accompanied by some fading\n\ntrials it was noticed that at a particular point near New Canaan, \nConnecticut, signals from 2XB were much weaker and more dis- \ntorted than signals from 2XY, the experimental station of the Amer- \nican Telephone and Telegraph Company at 24 Walker Street, New \nYork, even though the transmitter at 2XNB was about ten times more \npowerful. Daylight field strength measurements at this point showed \nthat the field strength of 2NB was only one-third that of 2XY. This \nled to the rather startling conclusion that there is a ratio of 100 to 1 \nin the power efficiency of transmission to that particular receiving \npoint from these two transmitting stations in New York which are \nonly about one mile apart.\n\nIn order to throw some light on this state of affairs a field strength \nsurvey was made by G. D. Gillett which resulted in the field strength \ncontour map! here reproduced in Fig. 1. The contours on this map \nshow that there is a series of long nearly parallel hills and valleys of \nfield strength which, extrapolated, would converge in lower Man- \nhattan and which extend out to the northeast as far as it was thought \nworth while to follow them. There has occurred to us no better\n\nexplanation of this hitherto uncharted form of field strength distribu-_\n\ntion thap that it is a gigantic wave interference pattern. A detailed \ndiscussion of this theory is given in another section of this paper. \nThe fixed pattern shown by Fig. 1 is definitely present only in the \ndaytime but that it is fixed is attested by the fact that a second survey \nmade about a year later checks with the original one quite closely. \nAt night fading is pronounced in the area covered by the pattern and it \nis apparent that some other factors must enter. As a result of an \nendeavor to check up the pattern at night it was discovered that \n\u2018This map was prepared by Mr, Gillett using the methods discussed in a paper \nDistribution of Radio Waves from Broadcasting Stations Over City Districts,\u201d\u2019\n\nquality distortion was, in general, most evident at places which were, \nby day, in the valleys of the field strength diagram and a point in ons \nof these valleys near Stamford, Connecticut, was selected for the estab \nlishment of a temporary field test station. The interior of this sta\n\nphotograph, Fig. 2. At this place apparatus was set up to enable a \nstudy of the nature of the distortion in signals from 2XB. Manv of \nthe records discussed in\u2019 succeeding paragraphs were taken at this \nStamford field station. Others were taken near Riverhead. Long \nIsland, which was also found to be well located for such work. Fig. 3\n\nis an outline map showing the relative positions of these field receivin, \nstations and the transmitting station.\n\nThe reason for settling down at a fixed point in this way was t \nattack the problem from a new angle. The field strength survey \nand aural observations had yielded much interesting information \nbut did not appear at that time to shed a great deal of light on the \nquality distortion so it was decided to attempt, by an oscillographic\n\nstudy of received signals sent out under rigorously controlled con- \nditions, to determine just what alterations these signals suffered in \ntheir journey through space.\n\nIn finding such distortions the ear is, of course, the primary testing \ninstrument or indicator of trouble, for, if the trained ear is unable to \ndetect anything wrong with a received signal in comparison with its \noriginal counterpart it is safe to say that nothing detrimental of import- \nance has happened to it. But the ear is a poor quantitative indicator \nand furnishes no permanent or easily analyzed record of its observa- \ntions. It is evident that if we are to study quantitatively the char- \nacteristics of radio transmission which give rise to quality distortion,\n\nwe must devise tests which will disclose changes, of whatever kind. \nn the relations between the various component frequencies of the \ntransmitted band and furnish interpretable permanent records. In\n\nFig. 3\u2014Outline map showing locations of transmitting station and receiving test \nstations\n\nThe variable factors in radio transmission which may be directly \ncontrolled are located at the transmitter and receiver. We have as \nyet no tangible means of controlling the transmitting medium, but it \ncan be studied indirectly through the characteristics of the received \nsignals. Obviously, it is desirable in the interest of simplicity to \nstabilize the apparatus variables to the extent that they may be \nidealized in considering observed results. Furthermore, at both the \ntransmitter and receiver, it is desirable to make the antenna arrange- \nments of the simplest form. For our work the normal antenna arrange- \nment at station 2XB was used perforce since any important changes \nwould have constituted a major operation. It is far from a simple \narrangement, as shown in Fig. 4 which is an outline elevation and plan \nof the antenna and building at 463 West Street, New York City. For- \ntunately there are no buildings considerably higher than the antenna \nwithin a distance of several wave lengths.\n\nAt the receiving test stations both loop and vertical antenna were \nused; but in most of the experiments a simple vertical antenna was \nemployed. It was constructed of brass tubing, 30 feet long, and \nguyed in a vertical position. A galvanized iron pipe 12 feet long \nwas driven in the earth for a ground connection. The vertical re- \nceiving antenna projected through the roof of the test station building\n\nat Riverhead, L. [., as shown in Fig. 5. The receiving antenna wa \nnot tuned but was connected to the radio receiver through fixed \ninductive coupling.\n\nThe carrier power in the transmitting antenna normally remains \nfairly constant, except for minor variations in voltage of the suppl) \nmains, and with a little care on the part of operating personnel, th\u00ab\n\nantenna current can be kept within the limits of a | per cent. varia- \ntion, which is small compared with the signal fading usually ex- \nperienced.\n\nThe stabilization of the frequency was of the greatest importance \nsince in some of the tests it was desired to beat or heterodyne the \nsignals down to audio frequencies and pass them through narrow \nband filters. To provide this stability engineers of the Bell Telephone \nLaboratories arranged the 5-kw. transmitter at station 2XB to obtain\n\nits carrier frequency by amplification of the output of a 6LO-ke. piezo- \nelectric crystal oscillator.\n\nWhen desired some of the antenna current from the output of the \ntransmitter was rectified and the resulting current was sent over a \ntelephone line to the receiving station so that the frequency and wave\n\nFig. 5\u2014Receiving test station near Riverhead, L. I. showing vertical antenna pro \njecting through roof of building\n\nform of the modulating signal could be seen and photographed at that \npoint, thus guarding against any possible distortion in the trans- \nmitter and enabling a direct \u201cbefore and after\u2019 comparison to be \nmade. The telephone circuit was also used for communication \nbetween engineers at the two terminal stations.\n\nAt the receiving station double detection receivers and audio fre- \nquency amplifiers were employed. These did not have entirely \n\u201cflat\u201d transmission characteristics over the audio frequency band, \nbut in most of the tests this was of no importance. In cases where \nit affected the results the making of necessary corrections was a simple \nmatter. In tests involving beating the received signals down to audio \nfrequencies through the agency of a local heterodyning frequency,\n\nthis was supplied from a shielded vacuum tube oscillator which on \ncomparison with a standardized piezo-electric oscillator was found \nto possess the required stability. The double detection type receivers \nwere used for no other reasons than their availability and their con- \nvenience for quantitative work. The beating down oscillator within \nthe sets and the intermediate frequency step passed through in the \nsets by received signals do not figure in the following discussion of \ntest methods but, of course, in each case the necessary set tuning\n\nVW \n| |Frequenc rR \n| quency Paper Re \nMaster \nIscillator| apn \n| for Wave Shape \nrary\n\nIn this work the moving coil type oscillograph was used throughout \nfor the purpose of making photographic signal records. As  indi- \ncated in Fig. 6 two oscillographs with elements connected in series \nwere employed; one for the purpose of making a continuous record \nof the variation in the amplitude of the signal using a slow moving \nphotographic paper tape and the other to obtain the wave shape of \nthe signal by means of the usual high speed photographic film drum. \nAn element of one oscillograph was also used at times to record on \nthe film drum the wave shape of signals rectifed at the transmitting \nantenna and sent over the telephone lines.\n\nIn considering these various records perhaps we had best look \nfirst at the simpler ones and then proceed in a more or less orderly\n\nfashion to the more involved ones. The simplest records are fading \nrecords of the unmodulated carrier frequency of 610 ke. At the \nreceiver the carrier was heterodyned with a local oscillator to pro- \nduce a beat tone of about 250 cycles which was fed through amplitiers \nto the oscillograph elements.\n\nFig. 7\u2014Interior view of Riverhead testing station showing recording apparatus\n\nthe received carrier signal with time, is given in Fig. 8. [t shows a \ntypical fading record made at Stamford, Conn., May 16, 1925. The \ntiming interval on strip 6 is 2.6 seconds.\n\nThe feed of the photographic paper tape through the oscillograph \nwas varied somewhat during the course of the experiments but was \ngenerally in the range of 6 to 12 inches a minute. At this rate the\n\nwidth corresponding to twice the amplitude of the signal, as both t] \npositive and negative half-eveles are recorded. It will be observe \nthat the outer limits of the band corresponding to the peaks of th \nsine wave are darker than the center portion of the record. This i \ndue to the fact that the rate of change of the movement of the ligh\n\nFig. 9--Wave form of beat note signal for single-frequency test. Center trace signal \nfrom vertical antenna, upper and lower traces signals from loop antenna receivers\n\nspot on the record is a minimum at the peak of the signal; hence, a \ngreater quantity of light affects these portions of the record. This \nshading effect was very useful in the way it brought out changes in \nthe distortion of the received signal. This is discussed fully in another \nsection of the paper. The fuzzy irregular outline on portions of the \nrecords is caused by static and radio noise. The timing marks on the \nrecord allow a measurement of the time interval between points of \nminimum signal. Fig. 9 is a sample oscillogram of the wave shape \nof a beat note signal recorded by the method described above. \nMarked changes in the fading cycle or time interval between points \nof minimum signal may occur within a period of a few minutes, and\n\ntrom day to day there is often evidenced a modification of the general \nharacter and the recurrence of these changes. An example of this \nchange in a short period of time is well illustrated by the oscillograms \nin Fig. 10. Strips 1, 2 and 3 form a continuous record starting at \n1:52 a.m.; strips 4, 5 and 6 start at 2:16 a.m.; and strips 7, 8 and 9 \nstart at 2:37 a.m. These are three sections of a continuous record\n\nFig. 10\u2014Single-frequency fading record, showing variation in rapidity of fading,\n\nselected for the purpose of showing the decrease in the fading period, \nin a 45-minute interval. The timing interval on strip 10 which ap- \nplies to these records is 5 seconds. In this particular record only \nhalf of the audio signal was recorded, the edge of the strip being the \nzero line.\n\nThese single frequency fading records do not offer very much to \nwork on. There is, however, just enough suggestion of regularity \nabout them to annoy one with the thought that perhaps they may \nfollow some definite combination of periodicities and with this in \nmind we have taken sections of two different records and subjected \nthem to a harmonic analysis.\n\nSo far we have been able to draw no more useful conclusions from \nsuch harmonic analyses than that the heterogeneous scattering of \nharmonic values is about what one would expect from the looks of \nthe curves.\n\nOne significant thing about these oscillographic single frequency \nfading records is that they show no high speed fading of important\n\nmagnitudes. Occasionally one cycle of the beat tone will be some- \nwhat upset by a sudden change in the amplitude but in general n) \nchanges which consistently distort the wave form were observed.\n\nThe slow fading may be considered as a modulation and on this \nbasis the received signal is seen to be composed of the original con- \nstant carrier frequency accompanied by very narrow side bands \noccupying at best perhaps a fraction of a cycle.\n\nThe next progressive step in the radio transmission studies is \nnaturally from a single frequency to two or more frequencies trans- \nmitted simultaneously. By the use of two crystal oscillators at the \ntransmitter two separate and distinct radio frequency signals were \ntransmitted simultaneously. These crystals were ground by the \nBell Telephone Laboratories to oscillate at 610,000 cycles and 609,750 \ncycles. The amplitudes of these signals at the transmitter were \ncontrollable so that it was possible to make them equal, or one larger \nthan the other, equivalent to the relative magnitudes usually found \nfor the carrier and single side-band transmission case. Records \nwere obtained of the variation of these radio signals, but none is \nreproduced here since the information shown by them can be just as \neasily abtained from the triple frequency records shown below.\n\nRadio transmission on three frequencies is readily obtained by \nmodulating the carrier with an audio frequency tone, and observing \nthe three frequencies separately at the receiver.\n\nwhere @ is a constant proportional to the percentage modulation. \nThese three frequencies are not merely a mathematical fiction \nbut are physically existent as three separate waves bound together \nonly at their point of origin. \nTo adequately record them separately by means of the oscillograph \nadvantage was taken of the fact that a group of frequencies beaten\n\nwith a single frequency differing from them by a small amount and \ndetected may thereby be reduced to audible frequencies without \nhaving their interrelations of phase, amplitude or difference frequency \ncomposition, changed in any respect. For instance if the frequencies \nexpressed above are beaten with a local constant frequency, \nB COS qt T y \nthe resultant lower or difference frequencies will be \nkBAa \ncos \n+kBA_ sin [(p\u2014gq)t\u2014y\u00a5] \nkBAa \neos\n\nEach one of the three components has been changed in amplitude \nby the same factor kB representing the efficiency of detection. Each\n\nand each has had its instantaneous phase shifted by an angle \u2014y. \nRelative to each other they remain unchanged.\n\nwas 608,375 ke. so that the resulting three audio frequencies were \n1,875 cycles, 1,625 cycles and 1,375 cycles.\n\nAs indicated in Fig. 11 in order to make a record of these signals \nthey are separated at the receiver by means of band filters. These \nfilters and others similar in type for other modulating frequencies\n\nwere designed and made by the Bell Telephone Laboratories es- \npecially for this work. The band filters used for the purpose of \nselecting the carrier and side-band frequencies had a cutoff of 40 \nTransmission Units 250 cycles from the mid-band frequency.\n\nThese cutoffs together with the position in the frequency range \nof the pass bands of the filters preclude any troubles from cross modu- \nlation of the radio carrier and side bands during the beating down \nprocess. The products of such cross modulation would be fre- \nquencies which are multiples of 250 cycles and these cannot pass the \nfilters. On the other hand the beaten down frequencies will pass \npractically intact, since as has been shown by the previously de- \nscribed single frequency tests, each of the three frequencies received \nalthough subjected to amplitude modulation by fading, represents \nonly a very narrow band of frequencies for which the filter pass bands \nwere of adequate width.\n\nAs the modulating tone was carefully calibrated to 250 cycles and \nthe filters adjusted to transmit the frequencies specified, it was only \nnecessary to transmit the carrier while adjusting the receiving beating \noscillator. The following procedure for this adjustment was found \nto be very successful. A local audio frequency oscillator was set to \nthe reduced carrier frequency of 1,625 cycles, and its output con- \nnected to a telephone receiver. The audio beat note from the radio \nsignal and local beating oscillator was reproduced by a loud speaker \nand its frequency adjusted to zero beat the 1,625-cycle tone from \nthe telephone receiver.\n\nWhen this adjustment had been completed the carrier was modu- \nlated with the 250-cycle tone, and the side-band signals automatically \npass through their respective filters.\n\nThe signals from the outputs of the filters were amplified, and \nrecorded separately by the three oscillograph elements. The sample \nrecords shown in Fig. 12 are representative.\n\nStrips 1, 2 and 3 are taken from a long record obtained May 7, \n1925, 3:22 a.m. The upper trace is a record of the upper side-band \nsignal, the center trace the carrier, and the lower trace the lower \nside-band. Strips 4, 5 and 6 are from a section of a similar type of \nrecord made May 28, 1925, 1:06 a.m., where the carrier was modu- \nlated with a 500-cycle tone and different filters were used. In this \nrecord the upper trace is the lower side-band and the lower trace the \nupper side-band.\n\nIt will be noticed that the timing interruption appears only in the \nside-band signals, as the tone was interrupted before modulation \ntook place, and that the amplitude of the carrier signal is not affected\n\nby the interruption of the modulating tone. This makes it very \neasy to identify the side-band signals. These records give an ex \ncellent graphic picture of ordinary radio telephone transmission, \nbringing out the fact that three truly individual frequencies are \ntransmitted to repre \u2014 one.\n\nIn Fig. 12, strips 1, 2, and 3, the relative amplitudes of the three \nsignals are very stat in snraiatiti to the relative amplitudes of\n\nFig. 12\u2014Fading record showing individually the fading of carrier and side-ban\n\nfrequencies. Made at Riverhead, L. 1. Timing interruptions in side-band signals, 5\n\nthe signals as they existed in the ether at the receiving point. Before \nthis record was made a transmission characteristic of the complete \nreceiving circuit, including the osciliograph elements, was obtained, \nusing a local transmitter with modulated carrier for the purpose of \nmaking the measurement. The gain of the audio amplifiers at the \noutputs of the filters was adjusted to give substantially uniform \ntransmission on each of the three frequencies corresponding to the \ncarrier and side bands of the radio frequency signal.\n\nAs shown in Fig. 11, a telegraph key is placed in the circuit of the \ncenter oscillograph element, for the purpose of placing identifying \nsignals on the records. An example of these identifying signals is\n\nshown in Fig. 12, strip 4, which gives the date and time the record \nwas started, July 23, 1925, 2:06 a.m. (Eastern daylight saving time).\n\nThe record in Fig. 13 is of the carrier and side-band signals with \n500-cycle modulation made at Riverhead, L. I., May 25, 1925, 1:25 a.m. \nMore gain was used in the side-band amplifiers for this record in \norder that the effects of fading could be brought out more prominently, \nIn this record only half of the side-band signals were recorded, the\n\nFig. 13-\u2014Fading record of carrier and side-band signals, made at Riverhead, L. I. \nliming interruptions in side-band signals, 5 seconds apart\n\nzero reference line being at the edge of the strip. The upper trace \nis the upper side band, the center the carrier and lower trace the \nlower side band. Where the traces of the signals overlap a darker \nrecord is obtained. This record may be confusing at first but if \nstrip 5 is examined where the amplitudes of the signals are not so \nlarge a better picture of the form of the record will be obtained.\n\nIt is obvious from these records that the carrier and side-band \nsignals do not fade together as a unit. The carrier may pass through \na zero value with still considerable amplitude in the side-band signals \nas in strips | and 38. In the first case, strip 1, the three frequencies \nsuccessively fade through points of minimum signal in the order lower \nside-band, carrier and upper side-band; and in the second case, \nstrip 3, the three frequencies fade through points of minimum signal \nin the reverse order. This is a definite indication of selective fading; \nthat is, fading is a function of frequency as well as time.\n\nAn endeavor to form an explanation of the cause of this selective \naction in fading must be largely in the nature of speculation. Fur- \nthermore, since our data consist in the results of things which have \nhappened rather than in any first hand information on the processes \nof the happening, the building of an explanation is a synthetic pro- \ncess. In general for any given set of facts it is possible to synthesize \na number of explanations. Bearing this philosophy in mind we have \nconsidered various theories in connection with our observations and \nhave concluded that simple wave interference as a major cause of \nthe signal variations is at present the most likely ial enatne While \nwave interference may be called a major cause it should perhaps also \nbe called a secondary cause since the assumption of wave interference \npresupposes for its origin, primary causation by some physical state or \nconfiguration of the transmission medium. Speculation as to the \nnature of this primary cause is one stage further removed from the \ndata contained in our oscillographic records than is the assumption \nof wave interference.\n\nSince it is desirable in the remainder of this discussion to point \nout the evidences of wave interference, let us consider brietly the \nnature of this phenomenon.\n\nTo avoid any possible confusion of terms let it be said that by \n\u201cwave interference\u2019? we mean a particular physical phenomenon in \nwave transmission and have no reference whatever to static, signals \nfrom other stations, or any other of the forms of radio noise which are \ncommonly designated by the word \u201cinterference\u201d when they hinder \nthe reception of desired signals.\n\nWhen two single frequency plane polarized wave trains start out \nat the same time from a common source and travel by different \nroutes to meet again at a distant point the naturg of disturbance at \nthat point is determined by the relative space phases of the planes of \npolarization and time phases of the amplitude of the two arriving \nwaves.\n\nIf we let FE represent the vertical resultant of the electric field, \nwhich would be the only part affecting a simple vertical antenna, \nsuch as we have used in most of our tests, then\n\nE =e, sin 27(Ft+d,) +e. sin 27( (1) \nwhere F is the frequency and d; and dz are the distances along the \nrespective paths measured in wave lengths and e; and e2 are the \nvertical components of the two waves. These two sine terms may \nbe thought of as two vectors differing in phase.\n\nthat is, the difference in length of the two paths must be an exact \nwhole number of wave lengths. The condition that the two waves \ncancel each other giving a field\n\nthat is, the difference in length of path must be an exact odd number \nof half wave lengths.\n\nThus if the two components e; and \u00e92 are equal, the resultant vertical \nfield E will go through values ranging from (e;+e:) down to zero \nas the path lengths change relative to each other. If the two waves \ndo not have exactly the same amplitude, the minimum value of / \nwill be something more than zero.\n\nDifferences in attenuation of the two waves and differences in \ntheir direction of arrival will modify the relative amplitudes of e; and \n\u00e9g but will not modify the time relations required for minima of the \nresultant field unless we assume that at the time of a minimum \nneither wave has an appreciable vertical component. Since the \nconsequences of such an assumption do not accord with our experi- \nmental data we have considered that it may be left out of account \nin the present discussion.\n\nThis is obviously a picture which fits in very well with the simple \nsingle frequency fading records. The major maxima and minima \noccur when the conditions of equations (2) and (3) are met and e; \nand e, are nearly equal. On the other hand it seems doubtful that \nthe picture can be so simple. If we suppose two wave paths why \nnot three or more? Additional paths would add irregularities to \nthe fading and it would not be necessary to assume as great a degree \nof irregularity in the changes in any one path. But with an increasing \nnumber of paths the various arriving waves would tend to average \nto a more or less constant mean value and large departures from \nthis mean would become rare. The fact that the fading signal con- \ntinually covers a large range of amplitude, with the maximum many \ntimes the minimum, definitely points toward there being but a very \nsmall number of major paths, probably not more than two.\n\nConsidering now the question of selective fading in relation to \nwave interference we refer back to equation (2).\n\nIf we assume the distances to be measured in any desired units \nand call them d,\u2019 and d.\u2019 our equation will still hold provided we divide \neach distance by the wave length measured in the same units, thus\n\nmay call the frequency spacing interval. That is, with changing \nfrequency EF will go through maximum values with frequency at a \nseries of frequencies beginning theoretically with zero and extending \nupward in regular spacing to infinity.\n\nThe spacing interval is obviously that number of cycles which cor- \nresponds to the lowest finite frequency in the series, namely, the \nfrequency for which the distance (d;\u2019\u2014d,\u2019) is just one wave length \nsince when x =unity equation (4) becomes\n\nBy using the same process on equation (3) we find that E has \nminimum or zero values at another series of frequencies having the \nsame spacing interval but lying midway between the frequencies at \nwhich maxima occur.\n\nThus it is apparent that with fixed path length difference the \namplitude of the field F will be different for different frequencies, \nranging from maxima of (e;+e2) down to minima of zero if the polar- \nization planes and amplitudes of the two vertical components are \nequal.\n\nFurthermore, still thinking of equation (1) as representing two \nvectors, it is evident that the phase of the resultant field is different \nfor different frequencies even though these different frequencies \nhad exactly the same starting phase at the source.\n\nIf the paths are changing with time, the field at a given point, as \nhas already been pointed out, will go through time fluctuations. \nAnother way to look at this is that there is a space pattern of maxima\n\nand minima and as the paths change the plane section of the pattern \ntaken by the surface of the earth wanders so that at any one point \nthe field is continually fading in and out as the maxima and minima \nglide by it. Each frequency has its own pattern differing from those \nof its neighboring frequencies in such a way that at any given point \nthe relation between amplitude and frequency is that just discussed\n\nFig. 14\u2014\u2014Plotted curves of signal amplitudes condensing a long fading record, part of \nwhich is shown in Fig. 13. Numbers along time axis correspond to successive 25 \nsecond timing interruptions\n\nabove. Thus as the paths change and the patterns shift the different \nfrequencies fade not simultaneously but progressively.\n\nIn the above analysis of wave interference it has been assumed \nthat all frequencies traveled from transmitter to receiver over a \ngiven path in the same elapsed time. This does not mean that they \nnecessarily follow exactly the same route on this path since they \nmight follow somewhat different routes of equal length or if their \ntransmission velocities were different they might follow different \nroutes of unequal length and still come within the definition of a \n\u201cpath.\u201d It seems reasonable to assume that over the width of an\n\nSTUDIES IN RADIO BROADCAST TRANSMISSION 165 \nordinary transmitted band the various frequencies are treated alike \nby the medium and the simple assumption that they follow the same \nroute with the same velocity is justified. If none of these assump- \ntions is correct but the departure is not large the effect will be merely \nto introduce slight irregularities into the spacing interval and the \ngeneral nature of the result will not be changed.\n\nLet us now examine more closely the record, a part of which is \nshown in Fig. 13.) A portion of this has been condensed into the \ncurves of Fig. 14. One unit along the time axes of these curves \nrepresents a 25-second interval.\n\nTo obtain these curves the amplitude of the signal has been scaled \noff and plotted, ignoring all the minor irregularities. From this \nrecord the relative fading characteristics of these single frequency \nsignals 500 cycles apart are more easily seen, and it is possible to \ncontrast the time of occurrence of points of minimum signal for any \npair of them.\n\nFor the frequency difference of 500 cycles (610.5-610 and 610\u2014 \n609.5) these times are obviously quite different but there is no clearly\n\ndiscernible relation between them. The curves for 1000-cycle dif- \nference (609.5-610.5), however, show a striking relation in that the \nmaxima and minima of the two are opposed fairly regularly over the \nentire 33-minute interval covered by the plot. This means that \nwhen one frequency has a minimum amplitude the other has a mavxi- \nmum and vice versa. Certainly this suggests a wave interference \ninvolving only two major paths whose difference in length is such \nthat the spacing interval is 2,000 cycles. The path difference appears \nto be changing somewhat irregularly but at an average rate of the \norder of one wave length (or approximately 500 meters) per minute.\n\nBefore speculating further on the numerical values which may be \nderived from this part of the data we had perhaps best consider some \nother records of a somewhat different kind which are better adapted \nto provide such values. But first let us reiterate that these are \nnight-time etfects.\n\nDuring the day signals substantially uniform in amplitude are \nreceived. An example of the type of transmission obtained in the \ndaytime is given in Fig. 15, which is a record of the carrier and side- \nhand signals received with substantially the same terminal condi- \ntions with the exception of the time as that existed when the records \nshown in Fig. 12 were made.\n\nThe abrupt change in the amplitude of the side-band signals was \ndue to an intentional change at the transmitter in the input level \nof the tone modulating the carrier, and accordingly the amplitude \nof the carrier did not change. The timing interval is 5 seconds.\n\nThe familiar fading record is limited to two axes, amplitude and \ntime. So far we have extended this cramped perspective somewhat \nby observing as many as three separate fading records spaced at \naudio-frequency intervals along the frequency axis. Even these \nthree narrow lookouts upon the wide range of ether transmission \nhave indicated amplitude relations along the frequency axis which\n\nFig. 16\u2014Diagram of system used to obtain records of selective fading or \u2018band \nfading\u2019\u2019 records\n\npromise to open a new line of attack upon the problem of night-time \nfading. But the desirability of knowing what takes place in the \ninterval unrevealed by these cracks in the fence becomes obvious. \nWe should like to know the relative amplitude of frequencies over a \nwide band, and the change in this relation with time.\n\nSince it is not a simple matter to record simultaneously the ampli- \ntude of a large number of waves of frequencies separated by say one \nhundred cycles in the radio-frequency range a single frequency in \ncombination with a frequency stepping device at the transmitter \nhas been adopted. The circuit arrangement is shown diagrammatic- \nally in Fig. 16. The rotary contactor bringing into the circuit suc-\n\ncessively a total of fifteen small condensers across the main condenser \nof the transmitter oscillator shifts the frequency in steps over an \nadjustable range. The contactor is rotated at the rate of nine revo \nlutions a minute, which is sufficiently slow to show definite ste ps in \nthe oscillograph record. At the receiving end a local oscillator sup- \nplies a radio-frequency wave for beating the incoming frequencies\n\nA long oscillograph record of this stepped frequency gives a sort \nof moving picture of the fading for the entire band covered. A \nsample of such a record is shown in Fig. 17 with alternate pictures\n\nthe two-way traversal of the frequency band successive pictures \nare reversed. If a series of such built-up pictures as these could be \ntaken rapidly on moving picture film, and projected successively \nupon a screen we should have before us an animated view of band \nfading. And according to the results of experimental investigation \nthe subject offers a lively theme for such a presentation. The peaks \nmid and depressions glide nervously back and forth across the setting. \nThe successive pictures of Fig. 17 (which, by the way, were selected \nfor their half-tone reproduction possibilities rather than as first \nclass examples of the records taken) illustrate a rather leisurely \nmovement of this sort. These ten built-up photographs cover a \nperiod of slightly more than one minute. In the first seven pictures \na depression appears at the left, while in the last three this depression \nseems to have made an exit followed by the simultaneous entrance of \nanother from the opposite wing of the stage. Evidence of such\n\norganized spacing of the minima is present in all of these night-time \nband fading records. As has already been suggested such evidence \nhas an important significance, but before going into this phase ot \nthe subject again let us examine a little more in detail the structure \nof these band fading records.\n\nThe steps in any one picture of Fig. 17 are, as we have said, snap- \nshots of the wave amplitude for successively different radio frequen- \ncies taken about a quarter of a second apart. The fact that the \nfifteen snap-shots used to build up a single picture are not taken \nsimultaneously causes a skewing of the outlines when movement \nof the depressions as shown in Fig. 17 occurs. If, for example, we\n\nFig. 18\u2014Three dimensional diagram, showing the method of interpreting band fading \nrecords\n\nwere to take fifteen separate and successive snap-shots of a mountain \nthrough fifteen long vertical slits side by side it would be possible to \ncombine the narrow sections so as to form a true picture of the peak. \nNow, if by some prodigious act of nature the mountain were shifted \nsuddenly to one side and back again during the time we were taking \nthe fifteen successive snap-shots through the vertical slits, the com- \nbination of them would form a profile quite different from that ob- \ntained when it was stationary. Or if it were simply moved steadily \nacross the field of vision during the time the snap-shots were being \ntaken one slope would be made to appear precipitous while the other \nwould be leveled to a gentle grade in the finally built-up picture.\n\nThe character of this skewing, then, and its magnitude depend upon \nthe rate at which the object being photographed in vertical sections \nmoves, and the direction of the movement.\n\nIn Fig. 18 is shown an imaginary night-time band fading record in \nthe \u201cassembled\u201d form. Since such a record contains frequency as a \nthird dimension, in addition to amplitude and time as shown in the \nordinary fading record, our simple fading curve has assumed the \nbroader aspect of a surface, the selective fading making more or less \nparallel \u2018valleys\u2019 running across it. The step-frequency svstem of \nrecording the points amounts to photographing sections of this solid. \nThe important point to be kept in mind is that these sections are \nnot perpendicular to the time axis. If they were, the skewing previ- \nously described would not be present. By setting these sections up in \ntheir true relation to the time axis, however, and filling in to produce \na continuous surface such as is shown in Fig. 18 the result is correctly \nrepresented. In order to make a detailed and accurate study of the \nband fading records, therefore, it is desirable to construct from the \noscillograph sections the complete surface by the method suggested.\n\nIn Fig. 18 the trace of minima crossing the band is shown by J, \nM' and M\u201d. Picture sections obtained as our recording apparatus \nliterally moves back and forth across this frequency band are shown \nas (a-b-c-d), (b-c-a\u2019), (a\u2019\u2014b\u2019-c\u2019), etc. It will be evident that the \nsection P), for example, will, in case a minimum is crossing rapidly, \nappear entirely unrelated to section P:. When the minima run \nnearly parallel to the time axis (slow changes in transmission condi- \ntions) the successive pictures P;, P2, Ps, etc., will reveal their rela- \ntion by direct comparison.\n\nActually to obtain frequency-amplitude sections perpendicular \nto the time axis in Fig. 18 would require the simultaneous trans- \nmission and reception of a large number of frequencies spaced at \nshort intervals along the frequency axis. A more practical thought \nis to speed up the process and though this seems very simple at first \nconsideration, it will be shown later to involve a particular kind of \ndistortion which cannot be separated out as easily as the skewing \nencountered by the more deliberated method.\n\nNow that we are familiar with the data, Fig. 19 showing, partially \nsuperimposed in vertical strips, the outlines of successive built-up \npictures of the frequency traverse will be of greater significance. \nDuring the steady periods there appears within the 2,280-cvcle \nband covered by these data approximately one complete cycle of \nselective fading. The lack of flatness in the audio-frequency-trans- \nmission characteristic of terminal apparatus has caused the suppression\n\nof amplitudes toward the right side of these sections. Keeping in \nmind also the skewing inherent to this system of presentation during\n\nrecord. The relative position of these minima gives us an interesting \ninsight into the nature of the night-time transmission path.\n\nbeen plotted against time as in Fig. 20.) The widths of the frequency \nbands covered in this case are indicated. This picture is essentiall \na bird\u2019s-eye view of band fading records such as are illustrated in \nidealized form by Fig. 18, the amplitude axis being perpendicular \nto the page. It reveals the presence of minima spaced at more o1\n\nFig. 20\u2014\u2014Plotted curves which condense a long band fading record so as to bring \u00ab \nthe frequency spac ing interval of the selective fading\n\ndepressions in regular spacing beyond the scope of our pictures, for \nwhen one minimum slides out of sight another appears to take its \nplace from the opposite side of the band. The minima traces shown \nin broken line were outside the record but were located by extra- \npolating the sections.\n\nOther depressions of small amplitude appear to be superimposed \nupon the major changes but the present data appear inadequate to \ngive reliable information concerning them. These minor depressions \nseem most evident during periods of rapid change.\n\nThe presence of these major minima in regular array bears a marked \nsimilarity to the familiar wave interference case in light and tits in \nvery nicely with the theory detailed in previous paragraphs. Assume \nfor a moment the simple case of two transmission paths producing \nsuch an effect and account for the difference in their lengths by pre \nsuming that one path follows more or less closely along the surface \nof the earth while the other seeks higher altitudes and in some fashion \ngets back to earth at the receiving station.\n\nThe mean frequency difference or spacing interval between suc- \ncessive minima for the records given in Fig. 20 is approximately \n2,200 cycles. Therefore, the mean wave length difference in length of \npath from equation (5) is 277 wave lengths, or 136.5 kilometers.\n\nIt is evident that the errant waves following the second path must \nhave been led a devious route. While this is about all the informa- \ntion which can be deduced directly from these data it is interesting \nto speculate further with the information along the lines of some of \nthe theories which have been proposed to account for such wave \ndeflections. For instance there is the Heaviside layer theory in which \nthere is supposed to be a more or less well detined reflecting layer in \nthe upper atmosphere. For this we would visualize our high alti- \ntude waves as proceeding in a straight line up to the layer, being \nreflected, and striking back to earth at the receiving station.\n\nSince the distance from transmitter to receiver was 110 kilometers \nthe length of the secondary path was 1104 136.5 or 246.5 kilometers. \nBy triangulation the height of the assumed reflecting layer may be \ndetermined as very nearly 110 kilometers or equal to the distance \nfrom transmitter to receiver, and the angle of incidence is 26.5 degrees.\n\nAs vet no positive information has been acquired concerning the \nvariation of difference in length of two major night-time transmission \npaths with direct distance from the transmitter. If the path differ- \nence is due to reflection from an overhead layer, the expected rela-\n\nWhen Ad is the difference in length of path, y is the direct distance \nand / is the vertical height of the laver.\n\nAn investigation of this relation would probably do much to prove \nor disprove the reflection theory.\n\nAt this point it is well to recall the results of earlier tests in which \nit was observed that single frequency waves separated by 1,000 cycles \nfaded in approximately an inverse relation also indicating a spacing \ninterval of about 2,000 cycles. The agreement of these earlier records \nis particularly noteworthy since about three weeks elapsed before \nthe more detailed band fading records were made.\n\nFig. 20 shows a time variation in the frequency position of the \nminima which is explained as due to a variation in the difference of \npath length. If we indulge in further speculation along the line of \nlaver phenomena we conclude that the reflecting layer is rising and \nfalling. It is improbable that the whole layer would rise and fall\n\ntogether so we conclude that undulations occur along its surface. \nThese undulations in themselves would cause the length of path of \nthe wave reflected toward the receiver to undergo a continual change. \nThey would also introduce minor retlections from surfaces more \ndistant than that responsible for the major effect which may be \nresponsible for the more rapid, low amplitude fading which is usually \nsuperimposed upon the slow changes. Obviously, the character of the \nfading would in the event that it is caused by undulations along the \nreflecting layer, be determined by the amplitude and direction of \nmovement over the surface.\n\nIf, on the other hand, we examine the possibilities of theories such \nas those proposed by Nichols and Schelleng, Larmor and others in \nwhich the action of free electrons in the atmosphere is invoked we \nmight visualize the waves on the second path as following a curved \ntrajectory. Or we might have the two sets of waves start off to- \ngether, become split by double refraction and eventually come. to- \ngether again. Perhaps their planes of polarization will have been \nrotated. In fact it is possible to build up what appears, we must \nconfess, to be a highly imaginary explanation in which the wave \ninterference is accounted for not on the basis of any great difference \nin path length but by the assumption that the amount of rotation is \nsuch a function of frequency that a change of about 2,000 evcles adds \nor subtracts a complete rotation, and the further assumption that one \nset of waves has had its plane of polarization rotated through several \nmore complete rotations than has the other. The synthetic possi- \nbilities are almost endless and we must wait upon further data more \nvaried in character before the facts can be established. In the present \ninvestigation we have not attempted to determine the mechanism of \nthe transmission medium except insofar as it could be inferred from \nthe results of our tests which were aimed at finding out just how radio \nsignals look after they have been subjected to a trip through this \nmechanism.\n\nReturning to the solid band fading record illustrated in Fig. IS, \nlet us form some conception of the appearance of this figure were it \nextended toward the much higher and lower frequencies using as a \nbasis of this conception the supposition that the existing record is \nsystematically distorted by wave interference. For a given rate of \nchange in the physical difference in length of path, such as would be \nencountered in the simple reflection case, the rate of movement of \nthe minima across the band fading pictures would vary directly with \nthe frequency. Therefore, we can extend the narrow section shown \nin Fig. 18 to form a wide band fading record such as is shown in\n\nFig. 21, wherein we are looking down upon the distorted surface, th\u00ab \nminima being traced by the light lines. Toward the short wave end \nof the band it is evident that a fading record for a single frequency \nrepresented, for example, by a section parallel to the time axis and \nperpendicular to the page, a\u2014a\u2019, would show rapid fading, while a \nsimilar record at the long wave end of the range as b\u20140}\u2019 would give \nslow amplitude changes. Such sections representing theoretical\n\nFig. 21\u2014 Theoretical diagram obtained by extrapolating band fading records to show \nhow the rapidity of fading might be expected to change with the wave-length\n\nsingle frequency fading records are shown at the bottom of Fig. 21. \nThe relative fading rates for long and short wave lengths as indicated \nby these idealized characteristics, are in accord with general experience\n\nIn describing the stepped-frequency method of obtaining band \nfading records allusion was made to distortion which might result from \nspeeding up the process. Suppose that we were to use a very small \nrotating condenser in parallel with the main condenser of the trans- \nmitter oscillator for changing the frequency, and that this condenser \nwere capable of changing the frequency sinusoidally about a mean \nvalue. Then we could represent the variation in frequency with time \nas is shown by the curve C; in (a) of Fig. 22. Now if the energy\n\ni i transfer from transmitter to receiver takes place over two paths of \nd different lengths one wave will constantly lag behind the other. \nA\" This lag may be measured as a time interval. In Fig. 28 are shown \na two waves, (a) and (b) of constant amplitude but with frequency \na \nssoida\n\nFig. 22\u2014Curves showing the relative effect of transmission time lag in sinusoidal and \nstep-by-step methods of frequency variations\n\nn Fig. 23\u2014Diagram showing the effect of frequency modulation \n\" modulation. The wave (b) representing the indirect wave, it will \nr be noticed, lags behind the direct wave represented by (a). The \n\" amount of this lag is determined by the difference in length of path \np and the transmission velocity. If we were to receive only one wave,\n\nconstant amplitude field (providing the high-frequency character- \nistic of the receiver is flat over the range of frequency variation). \nBut when two or more distinct paths exist, the combination at the \nreceiver becomes complex. This is evident in curve (c) shown in \nFig. 23 which is a direct summation of (a) and (b), and in (d) which \nis the envelope of (c). The amplitude is subjected to variations \nwhich did not exist at all in the original wave.\n\nWe might set up an equivalent effect right at the receiver by con- \nstructing two small local oscillators having the same characteristics \nas the transmitter oscillator. The two small rotating condensers \nwould be driven by the same motor but the rotor of one would be \nshifted backward in phase relation to the other so as to simulate the \ncase of transmission lag over the longer path. The relative fre- \nquency characteristics of the two may then be represented by curves C, \nand C;)\u2019 in (a) of Fig. 22.\n\nThe frequency of the signals arriving over devious paths at the \nreceiver may be put in the form of an equation as,\n\nFor a difference in length of path equal to 300 wave lengths at a \nfrequency of 600,000 cycles per second, for example, the time lag \nof one wave behind the other will be equal to 300 600,000 second or \n1 2000 second. The lag of one of the condensers behind the other in \nthe \u201cequivalent\u201d case described above would be then for 30 cycles \nper second rotation of the condensers, 30/2000 times 360 degrees \nor 5.4 degrees. The lag of 5.4 degrees represents the lag of the con- \ndenser rotor so that the frequency lag will depend entirely upon the \nrate of change of frequency by the rotating condensers at any \ngiven instant.\n\nNow to determine the resultant wave at the receiver we must know \nboth amplitude and relative phase of the components arriving over \nthe different paths. The amplitude will be constant, and we shall \nassume known, although it may actually follow slow changes with\n\nattenuation or variations in length of path. The relative phase must \nbe determined from equations (6a) and (6b). Knowing the fre- \nquency variation with time we may by integrating the following \nequation determine the phase relation at any time (1\n\nSubstituting the general relation for F, and F, from equations \n(6a) and (6b) we have,\n\nEvidently the relative phase (40) will be the difference between \nthese two giving,\n\nThe equation is not in itself very illuminating, but what it tells us \ngenerally is that if we represent two frequency modulated waves \ntravelling over paths of different lengths to a distant receiver by \nrotating vectors, these vec:ors are constantly shifting their relative \nposition. The magnitude of the shift at any instant is given by the \nvarying angle A\u00ae. Due to a change in the angle included by the two \nvectors their resultant will undergo an amplitude change, the serious- \nness of which we will consider later.\n\nThus far in the discussion of frequency modulation by means of a \nrotating condenser we have assumed sinusoidal changes in frequency. \nThe ordinary condenser departs considerably from such a perform- \nance. By considering the application of the integral equation for \nA@ to such a case it will be recognized that the relative space posi-\n\ntions of the vectors representing the direct and indirect waves will \nbe subjected to changes at every point where the slope of the fre- \nquency-time curve departs from a simple sine relation. The degree \nof distortion due to the presence of such irregularities may be con- \nsiderable.\n\nIn Fig. 24 are shown some samples of \u2018\u201c\u2018wobbled\u201d\u2019 carrier frequency \nrecords obtained at Stamford, Connecticut. For these records the\n\nFig. 24\u2014Sample fast records showing distortion produced by intentional frequency \nmodulation. a@ day record, 6 and \u00a2 night records\n\ncarrier was wobbled at the rate of about 10 cycles per second. There \nis some uncertainty as to the range of frequency variation for these \nrecords although it was probably in the order of a few thousand \ncycles. By means of a constant frequency local oscillator the radio- \nfrequency wave was stepped down in frequency to audio values which \ncould be amplified and recorded.\n\nThe record (a) of Fig. 24 represents stable day-time reception. \nThe record shows amplitude modulation due to the receiver char- \nacteristic alone. If the receiver were, as is desirable, capable of \namplifying all the frequencies present in the received wave in the \nsame ratio this record would be of constant width. In the sub- \nsequent examination of night records we must keep in mind the fact \nthat the terminal apparatus is responsible for a certain part of the \namplitude modulation. Its influence is readily recognizable.\n\nThe night-time records shown in (b) and (c) reveal a distinct dis- \ntortion of the envelope aside from that present in the daytime record. \nPeaks appear and disappear within time intervals sometimes as short \nas a fraction of a second.\n\nThe record in Fig. 25 represents a slow picture of the changes shown \nin (b) and (c) of Fig. 24. If these wobbled frequency waves are \nstudied carefully it will be noted that where a single peak stands at \none moment there gradually comes in view another as if it were sliding \nfrom behind the first. The cycle length being about 1/10-second \nwe may get some idea from this series of the rate at which the changes\n\nFig. 25\u2014Sample slow record showing distortion produced_by intentional frequency \nmodulation. Night record\n\ntake place. The presence of so many peaks in these records is at- \ntributed in part to the fact that the rotating condenser used gave a \nfrequency change which was far from a simple sinusoidal relation.\n\nLet us now return to the stepped-frequency method of obtaining \nthe band fading pictures and ascertain why it has certain advantages. \nIn (b) of Fig. 22 is shown the \u201cequivalent\u201d characteristic for the \nstepped condenser. During 1, 2000 of a second (for the conditions so \nfar assumed) in each step distortion may occur due to transient \nconditions, but during the remainder of the quarter second assigned \nto each step (for the records so far taken) a steady state is reached. \nThus, theoretically, distortion occurs only during about 1 500 of the \nstep interval. In (b) of Fig. 22 the lag is greatly exaggerated for pur- \nposes of illustration. This means simply that we have maintained \nconstant frequency for a sufficient length of time to establish, before \ntaking our picture, a fixed interference condition over the region \nincluding transmitter and receiver at least.\n\nThus far we have been dealing with the unstable phenomena of \nnight-time transmission. Our interest has been directed almost\n\nentirely toward variations with time. While the presence of wav \ninterference has been detected, and the movement of this inter \nference effect across the frequency band has been recorded, littk \neffort has been made to form a picture of such interference in its\n\nspace relation. A discussion of similar stable, daytime phenomena is i \ntherefore not out of place, and particularly so in view of an evident \nrelation of the fickle nocturnal interference phenomena to the steady i \nstates which follow the appearance of daylight. \u2018\n\nIn a previously published map of field strength distribution in : \nNew York City,* it was indicated that the congestion of high buildings .\n\nFig. 26\u2014Map showing location of radio obstruction on Manhattan Island as de \ntermined by the intersection of lines between various transmitting points and their \ncorresponding shadows\n\n; just below Central Park cast a heavy shadow. More recently it has \nbeen determined from observations on a portable transmitter, set \nup at various points, that this building center is a consistent per- \nformer. The position of this obstruction is determined in Fig. 26 \nwherein only partial contours from maps for the indicated sites are \ngiven to prevent confusion. The intersection of these lines from \ntransmitter to shadow, falls at approximately 38th Street in the vicinity \nof Sixth Avenue.\n\nThe dissipation of wave energy at such a point is probably the \ncomposite effect of many adjacent structures. Fig. 27 gives an \nelementary idea of how this can occur. The structures filling in\n\nBROADCAST TRANSMISS| \\ 1S] \nmee each block are, of course, very well connected electrically by means \nter of pipes, cables, ete., with those of adjacent blocks. Between each\n\nttle oscillating circuit (which is pictured as consisting of two buildings \nsi with earth connections) there exists a coupling which binds the whole \na is j system together more or less flexibly. Thus the obstacle offered by \nent \nrereere \u2014, \nngs r ror \n} \n: \n3 a \ni \nFig. 27\u2014Idealized picture of equivalent electrical circuit characteristics of hig! \nbuildings \na group of buildings might be of a selective nature, and evidently its \nfrequency characteristic may vary with direction.\n\nSuch an aggregate would, in addition to absorbing wave energy, \n\u2018 produce a change in velocity or a refraction of the wave front. Some \neir indication of such an effect will be discussed later. Before leaving the\n\nsignificance. \n\u2018a From the transmitter a wave front expanding outward and upward \net encounters an obstruction which we shall assume is near the earth \n\u2018i plane. The net result of this encounter is a weakening of the wave \n6 over an area nea: this plane, and probably a distortion of the energy- \nba bearing fields. We might then imagine this shadow to be a tunnel- \n\u201d like region extending along the earth beyond the obstruction, and as \ny having definite vertical as well as horizontal limits.\n\nFig. 28\u2014Aerial photograph of Manhattan Island showing locations of transmitting \nstation and obstructing high building area\n\nSTUDIES IN RADIO BROADCAST TRANSMISSION 183 \nSuch barriers to wave travel, situated within a short distance from \nthe source, seem, as we might expect, to have a more extensive and \nserious influence upon effective broadcast distribution than similar \nobstructions at greater distances. \nIt will be noticed that the obstruction falls very nearly upon the \ndirect line from the transmitter to the Stamford testing station. \nThis will also be evident later after an understanding of Fig. 29.\n\nwherein the position of the \u201cBand Near Sound\u201d represents also the \nbearing of the Stamford station. The Riverhead station is not \ndirectly in line with the major obstruction.\n\nIn certain sectors of the field strength contour map for station 2XB \nthere appears to exist a kind of wavy displacement of the contour \nlines forming a partial pattern of peaks and depressions side by side. \nIn general, this pattern must be differentiated from an ordinary \nshadow area. A remarkable example of this sort of field distribu- \ntion is shown in Fig. | which is one section of a field strength survey \nmade for station 2XB. These contours are based entirely upon \ndaytime measurements, and represent a condition which is stable \nthroughout the daylight period. Considerable difference in\u2019 signal \nlevel is apparent within short distances across the direction of wave \npropagation. Two pronounced low. signal channels extend ap-\n\nproximately north-east across this region. These shift with change \nin frequency of the transmitted wave. Fig. 30 illustrates the space \nrelations for such a movement. The full line curve shows a partial cross \nsection of the contour map of Fig. 1 taken along a line approximately \nperpendicular to the direction of transmission 110 wave lengths from \nthe transmitter. This represents relative field strength values for\n\nPig 30> -Cross-section of wave interference pattern showing change with frequency \n610.0-kilocyele radiation. When the frequency is raised to 635.0 \nkiloeyecles, there occurs a movement of the peaks and depressions as\n\nis shown by the broken line of Fig. 30.) Apparently the increased \nfrequency causes these channels to be crowded together.\n\nIf we take sections of the field strength contour pattern in Fig. | \nand examine carefully the relative amplitude of peaks and depressions \nrepresented by these wavy lines we shall find that the ratio of field \nstrength of the peaks to that in the depressions increases with dis- \ntance from the transmitter. That is, the channels become more \nsharply defined as we move away from the transmitter. This ratio \nis shown approximately by the curves of Fig. 31. If these peaks \nor depressions were simple shadows they would maintain their relative \nvalues at a distance from the source or even tend to \u2018heal\u2019 causing \nthe ratio to fall rather than rise as is actually the case.\n\nWithin 14.4 wave-lengths (7.1 km.) of the transmitter the pattern, \nso apparent beyond 30 wave-lengths, merges into one deep shadow \na cross-section of which is shown in Fig. 29. The abscissa of this \ncurve is in degrees measured from the transmitter so that the center \nof the two most distinct low field strength channels extending north-\n\neast may be inserted with their true radial relation. The two most \nevident in Fig. 1 are shown to be west of the line extending from \ntransmitter through the center of the obstruction located in Fig. 26. \nThe presence of Long Island Sound east of the geometrical center \nof the shadow has made an extensive survey of this section imprac-\n\nFig. 31\u2014Plot showing intensity of definition of wave interference pattern\n\ntical. However, a single section taken across the Sound at about \n90 wave-lengths from the station shows quite unquestionably the \npresence of a low channel about as indicated to the right of the ob- \nstruction designated in Fig. 29.\n\nWe have, therefore, a deep shadow with a more or less orderly \narray of maxima and minima within its limits. These maxima \nand minima grow more distinct at a distance from the transmitter, \ncontrary to what we might expect for ordinary shadows. Further- \nmore, we find that they move as the frequency is changed. These \nfacts lead to the belief that the phenomena in question are due to \nwave interference such as has already been described in connection \nwith night-time fading, but characterized by very much smaller path \ndifferences. This daytime interference condition is fixed while we \nhave seen that the nocturnal patterns appear to wander continually. \nTo explain this more in detail let us return to the shadow and con- \nsider the phenomena which might accompany it in a little more \ndetail.\n\nThe study of light has made available much information concern- \ning the subject of wave interference. It is known, for instance, that \nthe edges of shadows are not sharply discontinuous changes from \nlight to darkness, but that a series of dark and light bands, called\n\ndiffraction fringes, are interposed between the full light and full dark \nareas. In our radio case the distance from the source to the obstruc- \ntion and the dimensions of the obstruction are both very much smaller, \nin comparison with the wave length of the radiation, than for any \nordinary case in light, but apparently the phenomenon is of the \nsame general nature. By applying the ingenious principle of sec- \nondary sources used by Huyghens we might theoretically determine \nthe distribution of the field beyond an obstruction placed in the \npath of the advancing radio waves. The basis of this principle is\n\nFig. 32-\u2014Theoretical cross-section of radio shadow and associated wave interference \npattern\n\nthe assumption that each elementary part of the advancing wave \nmay be considered as a tiny transmitter. The effect at any point \nbehind an obstruction, therefore, becomes the resultant effect, con- \nsidering phase as well as amplitude, of the waves from all these minia- \nture sources.\n\nIn Fig. 32 the region between vertical lines (a) and (b) represents \nthe geometrical limits of the cross-section of a well defined shadow \ntaken some distance behind the obstruction. An analysis of the \nresultant field using Huyghens\u2019 construction would show variations \nin intensity somewhat as represented by the full line. In other \nwords the shadow will not be distinct but will have alternate maxima \nand minima within its geometrical limits and similar variations \nbeyond the edges.\n\nIt is very likely, of course, that even in case the foregoing specu- \nlative analysis of the contour pattern extending north-east of 2XB \nis fundamentally correct, a great many other influences than that\n\nof obstruction enter into the final field distribution. Relative atten- \nuation of water and land appear to influence the distribution con- \nsiderably though not as definitely as do steel structures close to the \ntransmitter. Distinct minima appear both on the Hudson and on \nthe Sound along radial lines extending from the transmitter.\n\nProbably refraction of the wave front in passing across shore lines \nalso enters into the shaping of this pattern.\n\nPerhaps as good an elementary picture as any of the phenomena \ncausing these patterns is that of a \u2018\u201c\u2018dent\u2019\u2019 produced in the wave \nfront by an encounter with a portion of New York City\u2019s impressive \nskyline. Since radio waves travel in a direction perpendicular to \nthe plane containing the electric and magnetic fields, opposite sides \nof this \u2018\u201c\u2018dent\u2019\u2019 would cross over one another with the result that an \ninterference pattern would appear beyond the obstruction. An \nanalogous situation exists when a water ripple passes a cluster of \nmarsh grass which, damping its motion and retarding its progress \ncauses part of the advancing front to converge and cross beyond the \nobstruction.\n\nThere is evidently a relation between day patterns such as have \nbeen discussed and night-time conditions. Just what this relation \nis offers some further opportunity for conjecture. In the first place \nquality distortion in transmission at night was, as previously ex- \nplained, observed over parts of the region covered by the pattern \nshown in Fig. 1. The worst distortion seemed to be somewhat \nassociated with the low field strength regions in this daylight survey. \nThe distortion seemed also to be worse along the low channel ex- \ntending in the direction of New Canaan, Conn., and beyond the \n100-wave-length circle. It was particularly bad at a distance of \nsome 140 wave-lengths from the station along this low channel where \nthe field strength became so low in the daytime as to be unmeasurable \nwith the set employed for the work. Accompanying the poor quality \nwere fading and marked directional shifts.\n\nQuality distortion though not so consistently severe at the River- \nhead station as in the vicinity of Stamford was at times easily detect- \nable by audible tests. Due to rapid attenuation of the radio waves \ntraveling from the site of 2XB across Manhattan and the length of \nLong Island the field strength around Riverhead is generally low \nwith higher levels north and south on the open waters of the Sound \nand Ocean respectively. Night-time fading at this point was rep- \nresentative of the variety which is usually found at distances of \napproximately one hundred miles from a broadcast transmitter. \nThe situation at Riverhead appears to be somewhat the same as\n\nthat which may exist over a large part of the broadcast area at a \ndistance from the transmitter, while in the Westchester region we \nhave an extreme and rather special circumstance. Field strength \nsurveys have shown that there are indications of a daytime inter- \nference pattern over the Riverhead area but this pattern, such as \nit is, appears to be irregular and to lack the definition which makes \nthe Westchester pattern so remarkable.\n\nOn the basis of the Westchester data alone we might build up a \ntheory to the effect that night-time shifts of the stable daylight \npattern were in some way responsible for quality distortion fol- \nlowing the departure of daylight. Such a thought applied to the \nRiverhead case does not seem so reasonable since here the pattern is \nabout one-quarter as distinct in terms of the ratio of maxima to \nminima values as the Westchester pattern. If, however, we presume \nthat quality distortion may be expected in areas where daytime \nsignals arrive considerably attenuated or so interfering as to simulate \nsuch an attenuated condition both situations are satisfied. After a \nconsideration of the evidence at present available, such a conclusion \nseems attractive; that is, a daytime wave interference pattern alone \nis only an agency in night-time quality distortion in so far as its \nminima in combination with the general shadow effect are responsible \nfor a low signal directly transmitted. Perhaps, in other words, the \ndaytime field strength is a measure of direct night-time transmission, \nthere existing in combination with this direct path at night a second, \nvariable route of greater effective length. Probably close to the \n\u201cindirect\u201d \nbut shadows or interference may materially modify the ratio.\n\nBy receiving simultaneously at several points the signal coming \nfrom a distant transmitter, it ought to be possible to detect the \nmovement in space of these interference bands we have been dis- \ncussing. The question immediately arises as to how far apart these \ndistributed receivers can be placed without giving us an entirely dis- \ncontinuous and misleading picture. For the first step toward record- \ning space variations, in the vicinity of the Riverhead testing station, \nthe receivers were spaced 1,16 wave length (30.5 meters), as illus- \ntrated in Fig. 33. It is necessary in making such determinations to \ntransmit a single radio frequency, since we have already found that \nthe interference bands for one component of a modulated wave are \nlikely to be in a different position than those for another.\n\nIn order to receive and record the radio frequency wave it is, as \nhas already been shown, convenient to use a local oscillator to beat \nit down to audible values. Since several oscillators for the separate\n\nemployed. This beating oscillator was situated at the testing station \nand the receiving antenna at this point was used as a radiator. In \norder to prevent overloading, the local receiver, the coupling to the \nreceiver input coil was balanced to give a minimum of the local signal.\n\nFig. 34 is a sample of the record obtained. The continuous shadow \nband at the top represents the local receiver output. One oscillator\n\nelement was used for the other four receivers, their signals being \nrecorded successively by a commutating device. Incidentally the \ninteraction between these receivers was checked by observing the \noutput of any one, while changes were made in the tuning of the \nothers. The antenna was, however, so nearly aperiodic that no \nrecognizable distortion or reradiation phenomena could be detected.\n\nFig. 35 illustrates compactly variations recorded by the oscillo- \ngraph records (of which Fig. 34 is a sample), for a representative \nperiod of about five minutes. Even within the dimensions of 1/16\n\nFig. 35\u2014Curves showing single-frequency fading on spaced receivers, condensed from \nlong record\n\nwave length there appears to exist transient field strength gradients \nin the direction of transmission. This is shown by a change in rela- \ntive values, in the upper set of curves which represents field strength \nat points 1/16-wave length apart in the direction of transmission. \nThe deviation is particularly noticeable in the relation between \nvalues for the local receiver and the \u2018\u2018West receiver\u2019 which is in the \ndirection of the transmitting station.\n\nThe lower set of curves, representing similar values across the line \nof transmission are much more nearly parallel. From the data so far \nobtained for the Riverhead testing site, it seems that transient night- \ntime field strength gradients are more generally evident in the direc-\n\ntion of transmission than perpendicular to this direction. Upon \nthese limited data one might be tempted to predict the presence of \ninterference bands across the line of transmission.\n\nThe above discussion concerning space relation of field strengths \nhas been included merely by way of contributing an additional bit \nof evidence to the theory that the erratic type of fading ordinarily\n\nFig. 36-\u2014Single-frequency fading record from vertical antenna and two-loop antenna \ncrossed at right angles\n\nexperienced at night time is due to wave interference. The picture \nis very small in terms of wave lengths but considering its content, its \nvery limits seem to imply wave interference rather than attenuation \nalone.\n\nIn connection with the wave interference theory thus far suggested \nas responsible for a major part of fading Fig. 36 is introduced as \nadded evidence. The middle record of this group represents am- \nplitude changes in the night-time reception of a carrier wave upon \na vertical antenna. The upper and lower records represent the \nsame for two loops turned at right angles to one another in the hori- \nzontal plane. By daytime tests the interaction of this combina- \ntion was found to be negligible. Night-time fading recorded simul- \ntaneously for these three separate receivers occupying as nearly the \nsame point in space as was possible, show that a high amplitude\n\nsignal may be coming in on both loops while the vertical antenna \npick-up approaches zero. Several points of this kind are marked by \narrows below the middle trace in Fig. 36.\n\nThere are at least two simple possibilities which might account \nfor these relations. In case the wave approaches the receiving \npoint from directly overhead, the vertical antenna would receive a \n\u201czero\u201d signal while the loops would pick up an amount depending \nupon the state of polarization. If this be true, the records indicate a \nvery rapid shift from the vertical direction of reception since the \nantenna minima are short lived most of them lasting at best a small \nfraction of a second.\n\nOn the basis of wave interference it is apparent that two waves \napproaching the receiving point in a 90-degree space phase relation \nand 180 degrees out-of-time phase could give a maximum signal \non the two loops while that received on the vertical antenna was a \nminimum.\n\nA compromise between these two viewpoints is probably a better \nguess than either one of them taken alone. That is, the existence \nof minima on the vertical antenna at the same moment that a strong \nsignal is coming in on the loops is perhaps due to the interfering \ncombination of waves having components in both the vertical and \nhorizontal planes.\n\nSo far the data shown have been limited to the results of observa- \ntions taken on special forms of transmission which are simplified \nfor the purpose of clearly exposing the basic facts. We wish now \nto consider some of the more practical aspects of signal distortion. \nThe first test which we made at our field test station was to record on \nslowly moving photographic paper tape and on the high speed film, \nthe detected audio signal which resulted when the transmitter was \nmodulated by a pure 26-4-cycle tone.\n\nFig. 37 is a sample of the general type of audio signal record ob- \ntained and Fig. 38 shows copies of the wave shape of the received signal, \nat particular times corresponding to the numbers of the oscillograms \non the records in Fig. 37. The abrupt displacement of the timing \ntrace indicates the point on the long record at which the snap-shot \noscillogram was made. A peculiar characteristic of these records is \nthe dark shadowy lines weaving back and forth through the band \nrecording the complete signal. These dark lines correspond to the \nkinks in the wave shape shown in Fig. 38. As explained before, the \ndarkening of the record is caused by the greater quantity of light\n\ntrace signal from vertical antenna receiver and lower trace signal from loop antenna \nreceiver, timing marks 2.6 seconds apart\n\nFig. 38\u2014Wave form of signals corresponding to numbered positions indicated \u00a21\n\naffecting the record at these peak points. At the same time these \nobservations were made, the wave shape of the signal rectified from \nthe antenna current at the transmitter was recorded by an oscillo- \ngraph. These oscillograms showed the signal to be free from dis- \ntortion at the transmitter.\n\nThe weaving of these shadowy traces together with their width \ngives a record of the change in phase and amplitude of the irregu- \nlarities in the wave shape of the signal. Although the wave shape \nof the signal is continually changing, it persists in substantially the \nsame form for a great many cycles. Thus the record shows that, \nin the transmission of this simple tone modulated signal from the \ntransmitting to the receiving antenna, it has been so modified that \nentirely new frequencies appear at the receiver. This receiver was \nshown by local tests to be free of any appreciable distortion within \nitself. While these new frequencies look like harmonics of the mod- \nulating tone in the snap-shot record it is obvious from the slow record \nthat they are not true harmonics but that they differ from the har- \nmonics by a very small amount and are incommensurable with the \nmodulating tone since they undergo progressive but irregular phase \nchanges with reference to it.\n\nThese records represent in a nutshell the signal distortion problem \nas it first presented itself to us. Our work then consisted in raveling \nout the complicated relations so that their nature could be ascer- \ntained and a theory of the causes established. In this paper, in the \ninterest of clarity of presentation we have departed considerably \nfrom the actual order of the experimental work but at this point \nperhaps the actual order is best to follow for a moment.\n\nWith such a weird-looking distortion to analyze, and if possible \neliminate, our first thought was as to whether the terminal apparatus \nmight not involve unrecognized peculiarities which would be a con- \ntributing cause. Local tests and daytime tests of the receiving \nsystem absolved it from doubt and attention was focussed on the \ntransmitting apparatus.\n\nIt was suspected that present day radio telephone transmitters leave \nsomething to be desired in regard to what we may call, for lack of a \nbetter term, their dynamic frequency stability. A very large per- \ncentage of the transmitters in use throughout the world today pro- \nduce amplitude modulation of the carrier by the action of modu- \nlating tubes directly upon an oscillating tube circuit. It is to be ex- \npected that the cyclic changes in circuit conditions occurring at the \nmodulating frequency will have some cyclic effect on the absolute \nfrequency of the carrier and that this effect will be in the nature of a\n\nwobbling or rapid shifting back and forth in frequency of the ampli \ntude modulated carrier. In other words the carrier and side-bands, \nwithout change in their relative frequencies, would be subjected to \n\u201cfrequency modulation.\u201d\n\nThis sort of thing should be clearly distingushed from the slow \nwandering of frequency which, for instance, causes beat notes between \ncarriers of different stations to drift gradually in pitch. What we \nhave called \u201cdynamic instability\u201d is so rapid (being governed by the \nevclic variations of the modulator) that it is difficult to observe by \nanv aural method. Since the transmitter being used for our tests\n\nwas a member of this almost universal class which employs modulating \nelements directly associated with the oscillator elements we deter- \nmined to study this aspect of the transmission.\n\nThe following test was made to find out the extent of the fre- \nquency variation during the period of the modulating cvcle. A \nschematic of the testing circuit arrangement is shown in Fig. 39. \nThe plan was to modulate the carrier with 33 evcles, a tone so low \nin frequency that it would not be ethciently transmitted through the \naudio frequency amplifier connected to the output of the radio re- \nceiver. Then upon beating the received modulated carrier signal \ndown to a frequency of about 1,000 cycles, an oscillogram of this \nsignal would show a 1,000-cycle signal with a 33-cyvcle modulation in \namplitude. Frequency modulation, if present, should then be easily \ndiscernible from the record. This experiment was made for day- \ntime transmission and oscillograms (A) and (B) shown in Fig. 40 \nwere obtained, one with the frequency of the beating oscillator greater \nthan the carrier frequency, and the other with the beating oscillator \nfrequency less than the carrier frequency. Both of these oscillo- \ngrams show by the change in the frequency of the beat note signal\n\nthat frequency modulation occurs in the transmitter circuit. The \n. \nfrequency change is very apparent on the oscillograms when the\n\nlengths of one evele at maximum and minimum amplitudes are com- \npared. The reality of the effect is demonstrated in the two records, |\n\nbig. 40 \u2014-Oscillograms showing frequency modulation accompanying amplitude \nmodulation\n\nthe beating frequency is moved in frequency from one side of the \ncarrier to the other.\n\nThe next step was to determine to what extent a stabilization of \nthe carrier frequency to stop frequency modulation would affect the \ndistortion of signals. True, master oscillator transmitters capable \nof giving the desired stability are not a new thing in the art. Several \nsuch transmitters were built by the Western Electric Company some \nvears ago and used successfully ship-to-shore radio telephone \nexperiments * in which frequency stability was of considerable im- \nportance. To modify the ordinary broadcasting transmitter to in-\n\n2 See Fig. 1 and accompanying discussion in: Radio Extension of the Telephone\n\nSystem to Ships at Sea by Hl. W. Nichols and Lloyd Espenschied Proc, I. R. E., Vol. \nIT No.3.\n\nfel ) FEIN EC / JOIN AE \n\u201cce ae reased frequency points with reference to the modulation cycle when \n| \nTak\n\nclude this feature involves major mechanical changes and in order to \nprovide a suitable arrangement for these tests the Bell Telephone \nLaboratories engineers merely added to the existing transmitter at \nstation 2XNB a temporary separate oscillator and high-frequency \namplifier which could be connected to drive the oscillator tubes of \nthe set as amplifiers. That this was free from frequency modulation \nis seen by comparing (C) of Fig. 40 with (A) and (B\n\nFig. 41-\u2014-Slow record of signal detected from tone modulated transmission with stabil- \nized carrier showing reduction in distortion. Made at Stamford, Conn., Oct. 10, 1924,\n\nwith Fig. 37. Fig. 41 like Fig. 37 is the detected result of a signal \nwhich started from the transmitter as a pure tone modulated signal, \nbut it shows that much of the wave form distortion has disappeared, \nthere remaining only a residuum which characteristically appears \nat the lower amplitudes of the signal. The probable cause of this \nresidual effect will be discussed later. Tests of speech and music \nwere concurrent with these findings. Using the normal transmitter, \nnight-time transmission as received at the test stations was seriously \ndistorted. When the stabilizing arrangement was emploved this \ndistortion was apparently eliminated except at the minima of fading.\n\nHaving arrived then at this practical result we wished to make \nfurther confirming tests, and tests to determine the whys and where- \nfores of the result. We have already detailed the more basic of \nthese tests in previous sections of this paper and are now ready to \nconsider the practical distortion records more carefully and to build \nup a theory to explain them.\n\nThe records shown in Fig. 42 are similar to the records in Fig. 37. \nThey are shown here to illustrate the difference in the characteristics\n\nof the wave form distortion variation that occurs from day to day. \nAll these records were made at Stamford, Conn.\n\nStrips 7 and S\u2014Jan. 24, 1925 \u2014-8:00 a.m. \nPhere is a marked ditterence in the records obtained on January 23 and \n24, which were made at the time an effort was being made to determine\n\nbig. 42--Slow record of signal detected from tone modulated transmission taken on \ndifferent days showing the changes in the character of the distortion\n\ntwisted appearance of the record obtained on January 24 is not very \ncommon in the records obtained. Most of the records have character- \nistics similar to those shown in Fig. 387. In the January 24 records \nthere is a marked change in the characteristic configuration of the \nVariation,\n\nIn order to obtain a record of the amount of wave form distortion \nresulting from frequency modulation present in the detected audio \nsignal the circuit arrangement shown in Fig. 43 was used. This circuit \nwas designed to analyze the wave form distortion when a 250-cycle\n\nsignal was used to modulate the carrier. Special precautions were taken \nto obtain a pure 250-cycle modulating tone. The wave shape of the \nsignal detected from the carrier at the transmitter was frequently \nchecked by observations with an oscillograph. The signals detected \nfrom the antenna current at the transmitter, both for the normal\n\nJ \nFig. 43-\u2014Diagram of system used to obtain \u2018Harmonic\u2019 analysis distortion records\n\ntransmitter with frequency modulation and for the stabilized carrie: \ntransmitter, were practically simple sine waves. The output circuit \nof the radio receiver was connected to a group of filters designed to \ntransmit narrow bands of frequencies straddling the harmonics of \n250-cycles.\n\nWhile below we have referred to the frequencies passing these \nfilters as \u201charmonics\u201d it should be borne in mind that they are not \nnecessarily (rue harmonics since they deviate verv slightly from the \ntrue harmonic relation. The purpose of the test was to procure a record \nwhich would show at a glance the presence or absence of wave form \ndistortion.\n\nThe input circuits of the filters were connected in parallel and the \noutput circuits separately connected to the audio amplitiers arranged \nto operate the oscillograph elements. The input of one amplitier was \narranged so that it could be switched either to the output of the tilter \npassing 250-cvcles or the output of the radio receiver. In this way a \nrecord could be obtained of either the whole tone from the receiver or \nonly the 250-cycle component.\n\nIn Fig. 44, Strip 1 is a harmonic analysis record of the audio tone \ndetected from the carrier and both side bands, transmitted with a \nstable carrier frequency. Strip 2 is a section of a record made a few \nminutes later when an unstabilized carrier was being used. On this \nrecord the lower trace is the 250-cvcle component, the center trace the\n\n200-cvcele component, and the upper trace the 750-cycle component. \nPhe upper and lower traces have their zero lines at the edges of the \nstrip. This record was made at Riverhead, L. I., April 30, 1925, at \n3:35 am. Strip 2 isa section of a record made a few minutes later when \nin unstabilized carrier was being used.\n\nThe gain in the audio amplifiers connected to the outputs of the \nhilters was adjusted to give nearly uniform transmission through the\n\nbig. 44--Slow record made with system diagrammed in Fig. 43. Contrasting the \nlistortion of detected tone transmitted by stabilized and unstabilized carrier frequency\n\nreceiving and recording apparatus for the frequencies recorded. Hence \nin these records the relative amplitudes of the fundamental and \nharmonies of the signal are directly comparable.\n\nStrips 3, 4 and 5 in Fig. 44 are taken from a record made for the \npurpose of obtaining a comparison of the wave form distortion sus- \ntained by the detected audio signal transmitted by the normal trans- \nmitter with frequency modulation present and by a stable frequency \ntransmitter. In each strip the lower trace is the whole tone from the \noutput of the radio receiver, the middle trace the second harmonic \n500 eycles) and the upper trace the third harmonic (750 cycles). \nStrip 8 and half of Strip 4 give the record obtained when the normal! \ntransmitter was used, and the remainder is the record obtained when \nthe moditied transmitter was used. There was a few minutes\u2019 difference \nin time between the ending of one transmitting condition to the begin- \nning of the next during which the master oscillator control was switched\n\nthe making of this record, so that the results obtained from the two \ntransmitters are directly comparable.\n\nThe record of the signal from the normal transmitter shows an \nabundance of second and third harmonics, at times equal in amplitude \nto that of the whole tone signal. The latter, of course, includes these \nharmonics. It will be noted also that dark line shadows run through \nthe trace of the whole tone, indicating the presence of the wave form \ndistortion. The signal from the stable frequency transmitter as shown \nby the record is practically free from wave form distortion. The\n\nindicate wave form distortion. This record is substantial evidence \nthat a great deal of the wave form distortion may be eliminated when \nthe carrier is stabilized. However, the selective fading still remains.\n\nThe selective fading we have already explained more or less satis- \nfactorily and we tind that it does not materially atfect the wave form \nof audible trequencies transmitted by a modulated stabilized carrier \nunless its changes are more rapid than any we have recorded. The \ncrippled state of originally perfect tone waves after they have been \ntransmitted by an unstabilized carrier, we have just observed. Now \nlet us consider the possible causes of this ditference. The carrier stabil- \nization referred to here, may we repeat, is not stabilization against \nslow variations in frequency from second to second or from hour to \nhour but rather against rapid variations within the evcle of the modu- \nlating frequency.\n\nThe reason for such changes over the modulating evcle is that \nthe variation of the impedance of a vacuum tube across the \noscillating circuit necessarily causes a variation in the nature period \nof the oscillation. As a simple case, the circuit in Fig. 45 is given,\n\nwhen rp is the plate resistance and the remaining constants are given \nin the illustration.\n\nDirect modulation by the usual method involves a cyclic change in \nthe value of plate resistance. Hence, according to the above equa- \ntion, there results a cyclic change in frequency which, though rela- \ntively small, becomes of the utmost importance when subjected to \nthe peculiar phenomena of night-time transmission.\n\nBy making certain assumptions concerning the nature of frequency \nVariation as amplitude modulation takes place, it is possible to work \nout distorted waves corresponding to various assumed wave inter \nference conditions at the receiver. Perhaps the most simple and \ninstructive means for producing these distorted waves is by a graph- \nical method.\n\nThe equation for modulation of a high-frequency wave by a single \ntone may be written\n\nWhen 1 represents the unmodulated amplitude of the wave, & is a \nfactor determined by degree of modulation, v is an angular velocity \nof the tone wave and p is the angular velocity of the high-frequency \nwave. The amplitude factor in this equation may be considered as a \nvector which is undergoing a change in length in accordance with \nthe term included in the brackets. For the purpose of our analysis \nwe shall include the angular velocity imparted to this vector by the \nlast term in the above equation, since we are interested in the en- \nvelope of the resultant high-frequency wave at the receiver and the \nrelative phase relations for two waves directly and indirectly trans- \nmitted combining to form this resultant. Since both carrier waves \nare of the same mean frequency only the relative position need be \nconsidered.\n\nNow in our graphical determinations for the case of two trans- \nmission paths different in length, we represent the two effective fields \nby vectors varving in length in accordance with the amplitude factor \nof equation (15). However, due to the difference in length of path,\n\nIn addition to the lag in amplitude there will be a lag in frequency \nchange over the frequency modulation cvele. This lag which has\n\nFig. 46\u2014Graphical method of synthesizing distorted wave forms caused by frequency \nmodulation\n\nalready been shown in connection with the analysis of distortion in \ncertain types of band fading records (see Fig. 22), becomes a change \nin the relative phase angle of the vectors under consideration. Thus \nour picture finally becomes one of two vectors changing in length, \nthe changes in one continually lagging the changes in the other, the \ntwo vectors at the same time undergoing what we might term a \nrelative angular wobble. \u201c\n\nIn Fig. 46 these relations are produced graphically. For our pur- \nposes we might assume that the vector representing one field is tixed\n\ninstant, for example, the directly transmitted field may be repre \nsented by a; in this figure. Assuming a difference in length of pat! \nwe may compute on the basis of the integral equation (13), the rela \ntive phase position of the vector representing the indirectly trans \nmitted field \u00ae). The relative amplitude of this vector may also |x \ndetermined by substituting Ag in equation (14).\n\nAfter establishing a sutficient number of vectors to represent. thi \nevelic variation we may combine the respective components to obtair\n\ng. 47 Synthetic wave forms showing distortion due to frequency modulation\n\nshown as Ry, Rs, Rs, ete., a broken line being drawn through their \nextremities to identify their positions. Now, if we plot these re- \nsultants as vertical ordinates in their successive time relation as \nshown on the lower right of Fig. 46, we have the envelope of the \nresultant wave at the receiver.\n\nWhen the mean position of the two vectors (a) and (b) in Fig. 27 is \nISO degrees separation, the signal is experiencing a fading minimum. \nWhen they are on the average in phase the amplitude is at a maximum. \nWe can, therefore, trace a relation between quality distortion and \nfading by such an analysis, assuming a constant percentage modula- \ntion. Fig. 47 shows a series of high-frequency wave envelopes ob- \ntained by this method of graphic analysis. The mean vector rela- \ntion is represented by @, and for @, = 180\u201cdegrees the fading may be \nconsidered at a minimum. The waves shown in Fig. 47 being en- \nvelopes of the high frequency will undergo certain changes in the \nprocess of detection. These, however, would only slightly modify \nthe wave,\n\nSTUDIES IN| RADI \nFor purposes of comparison, a set of oscillograph pictures of rep- \nresentative received wave shapes is shown in Fig. 4S These repre-\n\nsent the actual effect. of night-time transmission with frequency \nmodulation between 463 West Street, New York City and Stamford,\n\nFig. 48 \u2014Oscillograms showing actual wave forms with distortion re y \nquency modulation \nto correspond with the order shown in Fig. 47. There exists a striking\n\nsimilarity. Occasionally, however, the shapes predi ted may depart \nconsiderably from those obtained experimentally. As an example of \nsuch a departure, the record (h) in Fig. 48 has been included. Such \nunusual samples may be due to a combination of waves arriving over \nmore than two paths or it may be that the time variation of the \nfrequency is far from the simple sinusoid which we have assumed. \nAs a matter of fact, a critical mathematical treatment of this case \nshows that only an approximation of such a sinusoidal condition\n\npossible since as has been shown by Carson,' a frequency modulated \nwave of this character consists of an infinite series of fixed frequencies \nspaced at regular intervals either side of a \u201cfundamental\u201d carrie: \nwave. Obviously only a small part of such a series could get out of \nthe transmitter or into the receiver due to circuit selectivity. For \nthe lower modulating frequencies, however, the approximation \u00a7 in- \nvolved in the assumption of a simple sinusoidal variation is not far \nwrong since the amplitudes of these side frequency components fall \noff rapidly as their order in the series increases. While 150 wave \nlengths difference in path length has been assumed for the synthesis \nof the wave shapes in Fig. 47, this difference may according to the \ndata obtained amount to much more than this.\n\nIt may well be asked why this frequency modulation, since it pro- \nduces such marked distortion at night in certain places, does not also \ngive rise to distortion by day or in locations where transmission is \nsteady. A full answer to this question would be far from simple. But \nin brief it is because the carrier and side-bands shift in absolute fre- \nquency together as a unit so that their relative or difference frequencies \nwhich determine the audio signal remain unchanged. Another way to \nput it is that the detector operates on the envelope of the high-fre- \nquency signals and is blind to the frequencies contained within the \nenvelope except insofar as they affect the latter. However, since \nfrequency modulation appreciably widens the frequency band occu- \npied by the radio signals it is to be expected that the tuned circuits \nin the receiver would have some reaction on those louder portions of \nthe signal for which the amplitude modulation and therefore the \nfrequency modulation is large. The perfection with which broadcast \nsignals may be received under suitable conditions leads one to believe \nthat this effect must be small.\n\nIt has been shown that serious wave form distortion of the repro- \nduced signal may result if frequency modulation occurs with the am- \nplitude modulation and the transmission is subjected to night-time \nconditions. This distortion from frequency modulation can be elimi- \nnated by stabilizing the carrier frequency. There remain some wave \nform distortion and the annoying amplitude changes caused by selec- \ntive fading which is one of the most serious present day problems in \nradio transmission. Let us now consider the nature and cause of this\n\n* See \u2018\u201c Notes on the Theory of Modulation,\"\u2019 by John R. Carson, Proc. Institute \nof Radio Engineers, February, 1922.\n\nresidual wave form distortion and some further consequences of selec- \ntive fading under the assumption that there is no frequency modula- \ntion involved.\n\nThe process of detecting audio signals from radio frequency signals \nis, at least in its simpler aspects, well understood, but it may not be \ngenerally appreciated that the action is such that the detected signals \nmay be greatly modified by changes in the relative amplitudes and \nphases of the carrier and side-band components such as may \nresult from their transmission through the medium. That the ampli- \ntudes and phases of the carrier and side-band signals are not neces \nsarily received in the same relation that existed as they left the trans- \nmitter has been pointed out earlier, in the discussion on selective\n\nfading. \nThe usual expression for a high-frequency carrier wave of frequency \np/2z modulated by a low-frequency wave of frequency v 27 is \ne=A[l+a sin sin pt\n\nwhere Al is the carrier amplitude, a, the percentage modulation and \n@ the starting phase of the modulating tone with reference to the \ncarrier. Expanded into its components this becomes\n\ncos (pt+it+@ the upper side band \nCOS (the lower side band \n+A, sin pl the carrier) \nwhere \u00a2,;=\u00a22=6 and 4A,=A.,=A;=A as the waves leave the trans-\n\nIn the receiving set this funct\u2018on is squared by the action of the \ndetector and, neglecting direct currents and frequencies above the\n\noriginal modulating tone and the second term the second harmonic. \nFrom this expression several conclusions can be immediately\n\nfactor but the second term does not. Thus, if selective fading erases \nthe carrier at any time, reducing its amplitude to zero or a small value \nthe signal, represented by the fundamental tone, practically dis- \nappears, even though the side-bands have not faded out, and there remains \nonly the harmonic. This is the residual distortion shown in Fig. 41 \nand which can often be heard during a fading out period. It is \ncaused by the two side-bands beating together in the detector. We \nhave here exposed a fundamental defect in the usual form of modu- \nlated signal transmission. The amplitude of the received signal is \nsubject to all the whims of the carrier and to paraphrase freely an \nold saving we might remark that a signal is no stronger than its \ncarrier. We may at once conclude that one way to reduce fading \nis to suppress the carrier and resupply a constant amplitude carrier \nat the receiving station.\n\nAnalyzing further the first term of the expression representing the \ndetected signal, the first part of the bracketed portion results from \nbeating together in the detector of the carrier and upper side-band \nand the second part from the carrier and lower side-band. \u2014 It is clear \nthat one of the side-bands may fade out completely and the other \nwill still bring in the signal, provided the carrier is not also lost, with \na phase shift to be sure but nevertheless not seriously reduced in \namplitude. In telephony this kind of phase shift is relatively un- \nimportant. Here we have an evident advantage in transmitting both \nside-bands since they support each other's frailties. But if the twe \nside-bands suffer phase shifts in transmission, as we have earlier \nshown may be produced by wave interference, such that @; and @\u00bb \ndiffer by w radians or LSO degrees, the two components will cance! \neach other provided their amplitudes 4; and A\u00bb remain equal. In \nother words all three components\u2014carrier and both side-bands\u2014\u2014may \narrive at the receiver with full amplitude and vet no signal will be \ndetected from them except a second harmonic component. This is \nobviously a disadvantage of transmitting both side-bands since, at \nsuch an instant, if one of them were eliminated the signal would \nreappear.\n\nWe conclude that there is, on the basis of such a brief analysis, \nnot much to choose between single side-band and double side-band \ntransmission when the carrier is transmitted also.\n\nBut if we wish to realize the advantages of carrier suppression a \nchoice is not difficult. A carrier suppression system in which both \nside-bands are transmitted requires that the replacement of the \ncarrier at the receiving station be done with almost absolute accuracy \nas to frequency and phase, a thing which involves very serious prac-\n\n; \n\\ \nSes tical problems. On the other hand if but a single side-band is trans \nue 4 mitted the difficulty is reduced to placing the carrier within a vers \nfew cycles of its correet: position. The allowable departure will \nIns i depend on a number of things but there is reason to believe that for \n1] 3 high quality transmission it must be very small, perhaps no greater \nis than two or three cycles. \nVe With the single side-band carrier suppression method, invented by \nlu John R. Carson, the radiation is stripped down to the minimum which \nIs will fully transmit the telephonic signals and this reduces to a minimum \nan the exposure of the signals to the ravages of selective fading. If the \nits : spacing interval of the fading is relatively narrow as in the cases we \nng have examined hereinbefore, this form of transmission would not \ner i fade seriously in average volume but would be subjected to a continual \nchanging of its frequency-amplitude characteristic, that is to say \nhe individual frequency components would fade progressively as the \nm minima of the selective fading wandered back and forth across the \n1d d frequency range encompassed by the single side-band. If the spacing \nar interval of the fading were very large so that the minima were very \ner ; broad of if some other, at present unexplored form of fading which \nth covers a wide band at one time were acting, the signal would fade in \n1) average volume but the range of its variation would be only the square \nn- ; root of that of a carrier transmitted signal, since only the side-band \nh i would fade and the locally supplied carrier would remain unchanged. \n( : The extent to which these theoretically drawn conclusions may be \n\" realized in practical application Is vet to be determined but we have a \nD\u00bb few records bearing upon the matter which at least do not run con- \ntrary to them. \nn All of the transmission tests where the radio signal was beat with a \n\\ local oscillator and the detected beat note observed, were equivalent \ne to single frequency single side-band transmission with carrier sup- \nIS pression, the local oscillator functioning as the carrier suppressed at \nit the transmitter. In this case, for which a number of records have \n(| already been shown, the detected signal is in proportion to the product \nof the amplitudes of beating oscillator and received radio signal. The \nf. phase of either does not affect the amplitude of the audio signal. Hence, \n\u2018| the only important modification of the original signal is the variation \nin the amplitude resulting from selective fading. \n7 Unfortunately we have no records in which a direct comparison is \n1 made between single side-band transmission with and without carrier \n= suppression but the case can be visualized from the record shown in \nFig. 12 or 13. Here each one of the frequencies recorded may be looked \nupon as a single side-band frequency which has been detected through\n\nthe agency of the resupplied carrier of the beating oscillator used to \nbring them down to audio-frequency. If now we were to take two of \nthese frequencies shown on the record and multiply their amplitudes \ntogether at each point we would obtain the amplitude of the signal \nwhich would result if one of them were a single side-band and the \nother its accompanying carrier. It is obvious that the fading varia- \ntions would thereby be increased in amplitude and rapidity.\n\nIn order to obtain a comprehensive picture of the relative advantages \nof radio transmission using a Carrier and one side-band as compared\n\nFig. 49 \u2014Diagram of system used to obtain records of transmission with carrier and \none side band and carrier and both side-bands\n\nwith the common practice of transmitting both side-bands, the follow- \ning tests were made. The schematic diagram of the circuit arrange- \nment is shown in Fig. 49. At the transmitter the carrier and both \nside-bands are transmitted and at the receiver they were selected out \nby means of filters in the manner previously explained. The signals \nfrom the filters corresponding to the carrier and lower side-band were \napplied to the input of a detector circuit and from its output the de- \ntected difference signal was selected by a low-pass filter. This signal \nwas equivalent to that which would be received if only the carrier and \none side-band were transmitted. From the output of the radio receiver \na branch circuit goes to a low-pass filter which transmits only the \nsignal detected from the carrier and both side bands, suppressing from \nthis circuit the higher frequency signals corresponding to carrier and \nside-bands produced by the beating oscillator and received signals. \nBy making simultaneously a record of these two signals a direct \ncomparison is obtained of the effect of selective fading on their am- \nplitudes. Fig. 50 shows samples of several such records made at \nRiverhead, L. I. The modulating frequency for strips 1, 2 and 3 is \n250-cycles, and for strips 4 and 5, 500-cycles. The record on strip : \nis shown on account of the peculiar characteristic of the signal fading, \nfor considerable periods of time remaining at relatively low amplitude.\n\nIn these oscillograms the upper trace is the record of the signal from \nthe carrier and both side-bands, and the lower trace the signal from \nthe carrier and lower side-band.\n\nFig. 50-\u2014\u2014Slow record comparing the signal detected from carrier and one side-band\n\nwith signal detected from carrier and both side-bands. Made at Riverhead, L. 1.\n\nUpper trace carrier + both side-bands, lower trace carrier + one side-band. Strips 1\n\nsignal is detected from both side-bands. \u2018The amplitude of the signal \nfrom both side-bands in some instances is very small but appreciable \namplitude is still indicated at the same instant for the signal from one \nside-band. This is explained as meaning that the side-band phases \nwere such as to make the component signals 180 degrees out of phase \nafter detection and that the amplitudes of the components were \npractically equal. The reverse situation is also observed where the \namplitude of the signal detected from the lower side-band is zero and \nappreciable signal is recorded for the case where both side-bands are \nused. This is interpreted to mean that the side-band signal was \neliminated by selective fading. In this event it was, of course, not \ncontributing to the signal which was detected from both side-band \nsignals. The recorded signal comes from the other side-band which \nevidently was not eliminated at that instant by selective fading.\n\nVisual observations made with the cathode ray oscillograph, which \nunfortunately furnishes no permanent record of transient effects, con- \nfirmed the strip records in regard to the reality of there being side-band \nphase variations. From equation (17), it is seen that if these varia- \ntions occur the fundamental\u2019of the detected tone signal at the receiver \nwill not bear a fixed phase relation to that detected from the trans- \nmitting antenna current while if there are no such changes the phase \nbetween these two tones would remain constant. The locally detected \ntone and the tone detected from the transmitting antenna current and \nbrought to the receiving station over telephone Wires, were applied \nto the two pairs of deflecting plates in the cathode ray oscillograph \nSince the detlections caused by these two pairs of plates are at right \nangles to each other the resulting Lissajous figure from two sine waves \nof the same frequency will be a slanting line, an ellipse or a circle \ndepending on their phase and amplitude relation. The actual figures \nwere observed to change progressively through this range of shapes, \nthe changes following roughly the magnitude and rapidity of the \nfading. The efiect of amplitude changes on such figures is quite \ndistinct from the effect of phase changes and there was no difficulty \nin separating out the evidence of large phase changes.\n\nConsidering only the above theories and facts there appears to be a \nreasonable basis for a conclusion that the best form of radio transmis- \nsion for use in broadcasting is single side-band with carrier suppression. \nBut on practical grounds we do not believe such a conclusion is justified, \nThe fading and distortions which we have made much of in the pre- \nceding pages are not experienced by the majority of broadcast listeners \nwhen they listen to local stations. To require these listeners to prov ide \nthemselves with more complicated and expensi\\ e receivers, simply to \nallow more distant or less favorably situated listeners to obtain better \nreception, seems neither reasonable nor desirable. The art offers \nseveral other possible avenues toward improvement much less difficult \nof application and it must be remembered that radio broadcasting is \nalready reaching a degree of standardization and a volume of existing \nreceiving equipment which rules that changes must come slowly and \nwithout serious prejudice to the existing order.\n\nSubject to the limitations imposed by the scope of our investigations \nthe following conclusions may be drawn:\n\nFading can be quite sharply selective as to frequency and the evi- \ndence points toward wave interference as the cause.\n\nThe evidence for wave interference indicates that some of the energy \nof received signals reaches its destinations by a circuitous route and \nsuggests that this route is by way of upper atmospheric regions.\n\nQuality distortion may result from dynamic instability of the trans- \nmitter.\n\nAbstracts of Bell System Technical Papers \nNot Appearing in this Journal\n\nNew Methods and Apparatus for Testing the Acuity of Tearing. \nHarvey Furercuer. This paper presented before the American \nOtological Se CIEL, classifies hearing tests in four groups according to\n\n1. Industrial or those made for determining the fitness of a candi- \ndate for employment. In certain types of work it is particularly \nimportant that a prospective employee meet a definite requirement \nfor acuity of hearing. Tests made in the army and navy for various \nbranches of service are conspicuous examples of this kind of test.\n\n2. Educational or those made for determining the degree of hearing \nof school children both in the public schools and in the schools for the \ndeat for the spectal purpose of determining the proper methods to be \nused in their education.\n\nIt is highly desirable that a single scale be used for representing \nthe degree of hearing which is independent of the method used and \nwhich has a general application to the four purposes enumerated. \nSuch a seale is proposed and it is shown how the commonly made \nvoice test, watch tick, acoumeter, coin click and tuning fork tests \ncan be expressed in terms of hearing loss units on this seale.\n\nThe paper is concluded by summarizing the different methods for \ntesting the acuity of hearing which are as follows: (1) voice tests, \n2) phonograph audiometer, (38) hearing loss for speech calculated \nfrom audiogram, which audiogram may be obtained in three ways, \n(a) tuning forks (constant initial amplitude), (b) tuning forks (com-\n\nThe Relation Between the Loudness of a Sound and Its Physical \nStimulus. J. C. STEINBERG? Experiments with many types of \nsounds have shown that the loudness of a sound is a function of its\n\nsounds whose calculated values L are equal will appear equally loud \nto the average normal ear. P; is the r.m.s. pressure of the ith com \nponent of the sound wave. The weight and root factors HW and +, \nrespectively, are functions of the sensation level, which is svnonyvmous \nwith the term loudness as formerly used and is detined as\n\nwhere Pi is the r.m.s. pressure of the ith component when the com- \nplex sound is at the threshold of hearing. In case the components in \na narrow band of frequencies An are not resolved their energy must \nbe integrated to obtain the energy of the equivalent single component. \nThe root factor 7 is inversely proportional to the ratio of the minimum \nperceptible increase in energy to the total energy. For intensities \nnear the threshold, the weight factors are equal to the reciprocals of \nthe minimum audible pressures. Curves are given showing the \nvalues for HW\" for various frequencies at various sensation levels, also \nthe values of 7 as a function of S. As the intensity is increased the \nweight factors give greater weight to the lower frequencies; hence, \neven though the amplitude of the sound wave be increased without \ndistortion, the ear will perceive both an increase and a distortion. \nThis effect is due to the non-linearity of the ear.\n\near, Where NV is less than 5 or 6 eveles, two kinds of binaural beats \nare obtained. Objective binaural beats are heard for most values \nof f within the audible frequency range, provided there is the proper \ndifference in amplitude between the two tones. For telephone re- \nceivers as sound sources, this difference for best beats is about 55 Tl \nand for the same receivers supplied with sponge-rubber cushions \nabout 62 TU. These beats are heard because the louder tone is con- \nducted through the head to the ear of the weaker tone and the two \ntones there are about equally loud. Subjective binaural beats are \nheard for frequencies below S00 or 1,000 cycles when the tones at the\n\niwo cars have about the same amplitudes, differing by not more than \n25: Fi Data obtained with 22 observers are summarized. The \nevidence indicates that these beats are not due to cross conduction \nbut are of central origin and the result of the sense of binaural local- \nivation of sound by phase. If the beats are slow (ess than 1 per \nthey are generally recognized as an alternate right and left \nlocalization, though some observers may report one or more intensity \nmaxima during the beat eyvele. Such maxima are explained as the \nresult. of one\u2019s interpreting the sound as louder when localization \nis more definite. Fast beats (more than | per sec.) are generally \nrecognized as an intensity fluctuation. They are explained by assum- \ning that the sound appears louder when the phase relations are such \nthat it is normally best localized in\u2019 the position toward which the \nattention is directed. This explanation is supported by observations \nmade with a constant source rotating around the head of a listener.\n\nEffect of Tension Upon Magnetization and Magnetic Hysteresis in \nPermalloy. O. E. and L. McKEENAN.! Wires. ot \nfive nickel-iron alloys containing 45, 65, 78.5, Sl and S4 per cent. \nNi, 60: em. long and O.1 em. in diameter, were studied by a ballistic \nmethod, for tensions up to 10,000 Ib. per in\u2019 and fields up to satura- \ntion (10 to 20 gauss). Permalloy with SI per cent. Ni is nearly in- \ndifferent to tension in its magnetic behavior; permalloy with less \nnickel is more easily magnetized and has less hysteresis when under \ntension, While S4 per cent. permalloy is more dithicultly magnetized \nand has greater hysteresis when under tension. The saturation \nvalues are independent of the tension. In 78.5 per cent. permalloy,\n\nunder a tension of 3,560 Ib. per in, saturation is reached at only 2 \ngauss (and is practically complete at 0.2 gauss) and the hysteresis \nloss is only SQ ergs per cm.\u2019 per cycle, so small that it may be regarded \nas due to slight inhomogeneity rather than to any essential features \nof the magnetization process. Relation to crystal oricntation, X-ray \nexamination proves that this abnormally low loss is not due to any \npeculiar orientation of the crystal axes as the crystals are found to be \noriented at random. Miagnetostriction behavior can be deduced \nfrom these results. Above SI per cent. Ni, permalloy contracts like \nNi while below SL per cent. Ni, permalloy expands like Fe.\n\neter, Was determined experimentally and found to vary from a maxi\u201d \nmum of 1.6% 107\" to a low value, the changes being like these pre-\n\nA Contribution to the Theory of Ferromagnetism. McIWEEHAN \nRelation of permeability and hysteresis to atomic magnetostrictio \nIn permalloy, it has been found that magnetostriction changes sign \nabout Sl per cent. Ni, hysteresis losses can be made vanishingly \nsmall near this composition, and these effects are not due to the special \nalignment of crystals. It is suggested that in every ferromagneti \nmaterial the process of magnetization involves (1) intra-atomic \nchanges, presumably changes in the orientation of electron orbits, \ngoverned by quantum dynamics and independent of environment; \nand (2) inter-atomic changes (stresses and strains). The = inter- \ndependence of the inter-atomic changes and the intra-atomic changes \nis conveniently described as atomic magnetostriction. On this view, \nhysteresis loss and magnetic hardness are due to the energy required \nto produce, in succession, the local deformations associated with \nchanges in the magnetization of single atoms or small groups of atoms. \nHigh initial permeability and low hysteresis loss in) permalloy are \nexplained as resulting from locally compensatory atomic magneto- \nstrictions of the nickel and iron atoms in small groups. The tunda- \nmental differences in the magnetic behavior of Fe, Ni and Co are \nattributed to differences in their atomic magnetostrictions. Other \ndifferences are attributed to differences in the mechanical proper- \nties which alter the energy expended when atomic magnetostriction \ntakes place.\n\nInduction from Street Lighting Circuits: Effects on Telephone Circuits \nR. G. MeCurpy.\u00ae Synopsis. This paper discusses) series street \nlighting circuits from the point of view of their relations to nearby \ntelephone circuits. These lighting circuits often have a much greater \ninductive influence in proportion to the amount of power transmitted \nthan have most other types of power distribution or transmission \ncircuits. This is due to the relatively large distortion in wave shape ot \nvoltage and current on certain types of these lighting circuits, and to \nthe unbalanced voltages to ground which occur with series livouts. \nThree general types of lighting circuits are discussed. These are a ec, \nare circuits, d-\u00a2, are circuits supplied by mercury are rectifiers, and \nalternating-current incandescent circuits. Of these, the incandescent \ntype of circuit, in which the lamps are equipped with individual \nseries transformers or auto-transformers, is the most important in \nthis respect. Measures for reducing interference from these circuits \nare discussed.\n\nPower Distribution and Telephone Circuits. Inductive and Physical \nRelations. H. M. TRUEBLOOD and D. IT. Cone? Consideration of \nthe relation between power distribution and telephone systems is \nnaturally involved in the comprehensive review of the problems of \nthe rapidly expanding power distribution networks in this country. \nPending the completion of studies now being actively carried on in \nthis comprehensive review, a preliminary and qualitative discussion \nis given.\n\nSituations of exposure fall into three groups determined by the \ncharacter of the area served. (1) \u201cdowntown\u201d districts; (2) residential \nurban districts; (3) rural districts. The major problems arise in the \nsecond group. A wide variety of arrangements characterize both \nsystems, and require consideration.\n\nAmong technical features, coethcients of induction for close ex- \nposures, shielding action of metallic cable sheaths for both power \nand telephone circuits, and \u201cground potential\u201d effects, are distinctive \nproblems. Where both classes of circuits are in cable with suitable \nprecautions as to grounding, interference is rarely to be anticipated.\n\nNoise induction from power-distribution circuits is chiefly from \nresiduals, which occur on single-phase branches of polyphase circuits, \nor where triple harmonics or load-current unbalances are introduced \nby grounding neutrals, or where admittances to ground of phase \nWires are unequal. Residual currents are largest in systems having \nmultiple-grounded neutrals, both load currents and triple harmonics \noccurring. Approximate resonance at triple harmonic frequencies \nbetween the inductance of station apparatus and power cable capaci- \ntance has characterized several situations. Various single, two and \nthree-phase arrangements are compared from the induction standpoint.\n\nThe closely related matter of unbalances in the telephone plant is \nbrietly discussed.\n\nBANCROFT GHERARDI, M.E., M.M-.E.. Cornell Universitv. Engi \nneering assistant, 1895-09; traffic engineer, 1899, New York Tele- \nphone Company; chief engineer, New York and New Jersey Telephone \nCompany, 1900 06; assistant chief engineer, New York Te lephone \nCompany, and New York and New Jersey Telephone Company, \n1906-07; equipment engineer, American Telephone and Telegraph \nCompany, 1907-09; engineer of plant, 1909-18; acting chief engineer, \n1918-19; chief engineer, 1919-20; vice-president and chief engineer, \n1920-\u2014\u2014-. Mr. Gherardi\u2019s work in the field of telephony is too well \nknown to require comment.\n\nRoBERT W. Kina, A.B., Cornell University, 1912; Ph.D., 1915; \nassistant and instructor in physics, Cornell, 1913-17; Engineering \nDepartment of the Western Electric Company, 1917-20; Depart- \nment of Development and Research, American Telephone and Tele- \ngraph Company, 1920-21; Information Department, 192]\n\nWALTER A. SHEWHART, A.B., University of Hlinois, 1913; A.M., \n1914; Ph.D., University of California, 1917; Engineering Depart- \nment, Western Electric Company, 1918S 24; Bell Telephone Labora- \ntories, Inc., 1925 \u2014. Mr. Shewhart has been engaged in the study of \nthe relationship between the microphonic and physicochemical prop- \nerties of carbon.\n\nHARVEY FLETCHER, B.S., Brigham Young, 1907; Ph.D... Chicago, \n1911; instructor of physics, Brigham Young, 1907-08; Chicago, \n1909-10; Professor, Brigham Young, 1911-16; Engineering Depart- \nment, Western Electric Company, 1916 24; Bell Telephone Labora- \ntories, Inc., 1925\u2014. During recent vears, Dr. Fletcher has conducted \nextensive lavestigations in the fields of speech and audition.\n\nJoun R. Carson, B.S., Princeton, 1907; E.E., 1909; M.S., 1912: \nResearch Department, Westinghouse Electric and Manufacturing \nCompany, 1910-12; instructor of physics and electrical engineering, \nPrinceton, 1912-14; American Telephone and Telegraph Company, \nEngineering Department, 1914-15; Patent Department, 1916-17; \nEngineering Department, 1918; Department of Development and \nResearch, 1919 -.) Mr. Carson's work has been along theoretical \nlines and he has published many papers on theory of electric circuits\n\nKart K. Darrow, S.B., University of Chicago, 1911; University \nof Paris, 1911-12; University of Berlin, 1912; Ph.D., in physics and \nmathematics, University of Chicago, 1917; Engineering Depart \nment, Western Electric Company, 1917-24; Bell Telephone Labora \ntories, Inc., 1025. Mr. Darrow has been engaged largely in\u2019 pre \nparing studies and analyses of published research in various fields \nof physics.\n\nRatepu Bown, M.E., 1913, M.M.E., 1915, Ph.D., 1917, Cornell \nUniversity, Captain Signal Corps, U. S. Army, 1917-19; Depart- \nment of Development and Research, American Telephone and Tele- \ngraph Company, 1919\u2014. Mr. Bown has been in charge of work \nrelating to radio transmission development problems.\n\nDrLoss K. Martix, B.S., Polytechnical College of Engineering, \n1920; UL S. Navy, 1918S 1919; Department of Development and \nResearch, American Telephone and Telegraph Company, 1919 \nMr. Martin's work has related particularly to radio broadcast trans- \nmission.\n\nRatpn K. Porrer, B.S., Whitman College, 1917; E.E., Columbia \nUniversity, 1923; U.S. Army, 1917-19; Department of Development \nand Research, American Telephone and Telegraph Company, 1925 \nMr. Potter has been engaged in experimental work relating to radio \ntransmission phenomena.", "title": "The Bell System Technical Journal  1926-01: Vol 5 Iss 1", "trim_reasons": [], "year": 1926}
{"archive_ref": "sim_att-technical-journal_1961-03_40_2", "canonical_url": "https://archive.org/details/sim_att-technical-journal_1961-03_40_2", "char_count": 374008, "collection": "archive-org-bell-labs", "doc_id": 442, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc442", "record_count": 765, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/sim_att-technical-journal_1961-03_40_2", "split": "test", "text": "Pulse Transmission by AM, FM and PM in the Presence of Phase \nDistortion E. D. SUNDE 853\n\nFarther Results on the Detectability of Known Signals in \nGaussian Noise H. C. MARTEL AND M. V. MATHEWS 423\n\nConfocal Multimode Resonator for Millimeter Through Optical \nWavelength Masers G. D. BOYD AND J. P. GORDON 489 _\n\nRelation Between Surface Concentration and Average Conductivity \nin Diffused Layers in Germanium P. B, CUTTRISS 509\n\nA General Method of Applying Error Correction to Synchronous \nDigital Systems D. 4%. ARMSTRONG 577\n\nOn the Construction of Minimally Redundant Reliable System \nDesigns D. K. RAY-CHAUDHURI 595\n\nMode Conversion in Metallic and Helix Waveguide 4. a. UNGER 613 \nWinding Tolerances in Helix Waveguide H. G. UNGER 627\n\nH. 1. ROMNES, President, Wesiern Eleciric Company \nJ. B. Fisk, President, Bell Telephone Laboratories\n\nE. J. McNEELY, Executive Vice President, American \nTelephone and Telegraph Company\n\nG. E. SCHINDLER, jR., Editor \n. M. FOSTER, jR., Assistant Editor \n. POLOGE, Production Editor \n. T. MYSAK, Technical Illustrations \n. N. POPE, Circulation Manager\n\nTHE BELL SYSTEM TECHNICAL JOURNAL is published six times a year \nby the American Telephone and Telegraph Company, 195 Broadway, New York \n7, N. Y. F. R. Kappel, President; S. Whitney Landon, Secretary; L. Chester \nMay, Treasurer. Subscriptions are accepted at $5.00 per year. Single copies $1.25 \neach. Foreign postage is $1.08 per year or 18 cents per copy. Printed in U.S.A.\n\nIn pulse transmission systems, pulses modulated in various ways to carry \ninformation may be transmitted by amplitude, phase or frequency modula- \ntion of a carrier, and with each type of modulation various methods of de- \ntection are possible. An important consideration in many applications is \nthe performance of various modulation and detection methods in the pres- \nence of phase distortion or equivalent envelope delay distortion, which may \nbe appreciable in certain transmission facilities. The principal purpose of \nthis presentation ts a theoretical evaluation of transmission impairments \nresulting from certain representative types of delay distortion. These trans- \nmission impairments are reflected in the need for increased signal-to-noise \nratio at the detector input to compensate for the effect of delay distortion.\n\nThe performance in pulse transmission by various carrier modulation \nand detection methods can be formulated in terms of a basic function com- \nmon to all, known as the carrier pulse transmission characteristic, which is \nrelated by a Fourier integral to the amplitude and phase characteristics of \nthe channel. Numerical vaiues are given here for the carrier pulse transmis- \nston characteristic with linear and quadratic delay distortion, together with \nthe maximum transmission impairments caused by these fairly representa- \ntive forms of delay distortion with various methods of carrier modulation \nand signal detection. These include amplitude modulation with envelope \nand with synchronous detection, two-phase and four-phase modulation with \nsynchronous detection and with differential phase detection and binary fre- \nquency modulation.\n\nIn determining the effect of delay distortion, a raised cosine amplitude \nspectrum of the pulses at the detector input has been assumed in all. cases, \ntogether with the minimum pulse interval permitted with this spectrum and \nideal implementation of each modulation and detection method. Further- \nmore, optimum adjustments from the standpoint of slicing levels and sam- \npling instants at the detector output are assumed for each particular case of \ndelay d\u2019stortion. These idealizations insure that only the effect of delay dis- \ntortior. is evaluated and considered in comparing modulation methods, and \nthat this effect is minimized by appropriate system adjustments.\n\nSignal-to-Noise Ratios in Binary FM \n9 Slicing Levels and Noise Margins \n10 Evaluation of Transmission Impairments \nyvnehronous AM and PM \nGeneral \nSynchronous AM and Two-Phase Modulation \nQuadrature Carrier AM and Four-Phase Modulation \nEven Symmetry Pulse Spectrum and Delay Distortion \nRaised Cosine Spectrum and Quadratic Delay Distortion \nEven Symmetry Spectrum and Odd Symmetry Delay Distortion \nRaised Cosine Spectrum and Linear Delay Distortion \nVestigial-Sideband vs. Quadrature Double-Sideband AM \nEnvelope Detection vs. Synchronous Detection \nwith Differential Phase Detection \nGeneral \nBasic Expressions \nEven Symmetry Spectrum and Delay Distortion \nTwo-Phase Modulation \nFour-Phase Modulation \nRaised Cosine Spectrum and Quadratic Delay Distortion \nvary Frequency Modulation (FSK)\n\nRaised Cosine Spectrum and Linear Delay Distortion \nummary \n6.1 General \n3.2 Choice of Transmission Delay Parameters \n3 Double-Sideband AM \n3.4 Vestigial-Sideband AM and Quadrature Double-Sideband AM\n\nPM with Synchronous Detection \nPM with Differential Phase Detection \nBinary FM \n3.8 Comparisons of Carrier Modulation Methods \nVII. Acknowledgments \nAppendix. Determination of Carrier Pulse Transmission Characteristics. . \nReferences\n\nBinary pulse transmission by various methods of carrier modulation \nhas been dealt with elsewhere on the premise of ideal amplitude and \nphase characteristics of the carrier channels.\u2019 An important considera- \ntion in many applications is the performance in the presence of phase \ndistortion or equivalent envelope delay distortion, which may be ap- \npreciable in certain transmission facilities. An ideal amplitude spectrum \nof received pulses can be approached with the aid of appropriate terminal \nfilters with gradual cutoffs, such that the associated phase character- \nistic is virtually linear. Nevertheless, pronounced phase distortion may \nbe encountered in pulse transmission over channels with sharp cutoffs \noutside the pulse spectrum band, as in frequency division carrier system \nchannels designed primarily for voice transmission.\n\nThe principal purpose of the present analysis is a theoretical evalua- \ntion of transmission impairments resulting from certain representative \ntypes of delay distortion in pulse transmission by various methods of \ncarrier modulation and signal detection. These transmission impair- \nments are reflected in the need for increased signal-to-noise ratio at the \ndetector input to compensate for the effect of delay distortion.\n\nThe performance in pulse transmission by various carrier modulation \nand detection methods can be related to a basic function known as the \ncarrier pulse transmission characteristic. This basic function gives the \nshape of a single carrier pulse at the channel output, i.e., the detector \ninput, under ideal conditions or in the presence of the particular kind of \ntransmission distortion under consideration. From this basic function \ncan be determined the envelopes of carrier pulse trains at the detector \ninput, together with the phase of the carrier within the envelope. The \nshape of demodulated pulse trains with various methods of carrier \nmodulation and detection can, in turn, be determined for various com- \nbinations of transmitted pulses, together with the maximum transmis- \nsion impairment from a specified type of channel imperfection, such as \ndelay distortion dealt with here.\n\nThe carrier pulse transmission characteristic is related by a Fourier \nintegral to the amplitude and phase characteristic of the channel. It\n\nhas been determined elsewhere\u2019 for pulses with a raised cosine spectrum\n\nand cosine variation in transmission delay over the channel band, and \nfor pulses with a gaussian spectrum with linear variation in delay. A \ncosine variation in delay is approximated in some transmission facilities \nand has certain advantages from the standpoint of analysis, both as \nregards numerical evaluation and interpretation in terms of pulse echoes.\n\nA somewhat similar form of delay distortion that affords a satisfactory\n\napproximation in many cases is quadratic (or parabolic) delay distor- \ntion. Quadratic delay distortion is in theory approached near midband \nof a flat bandpass channel with sharp cutoffs, such as a carrier system \nvoice channel, and usually affords a satisfactory approximation over the \nmore important part of the transmission band of such channels. Linear \ndelay distortion is approximated when a bandpass channel with gradual \ncutoffs is established to one side of midband of a flat bandpass channel \nwith sharp cutoffs. These and other types of delay distortion do not \nlend themselves to convenient analytical evaluation of the Fourier in- \ntegrals for the pulse transmission characteristic. However, at present, \nthese integrals can be accurately evaluated by numerical integration \nwith the aid of digital computers for any specified pulse spectrum and \nphase distortion.\n\nNumerical values are given here for the carrier pulse transmission \ncharacteristics with linear and quadratic delay distortion, together with \nthe maximum transmission impairments caused by these limiting and \nfairly representative forms of delay distortion with various methods of \ncarrier modulation and signal detection. These include amplitude modu- \nlation with envelope and with synchronous detection, two-phase and \nfour-phase modulation with synchronous detection and with differential \nphase detection and binary frequency modulation. In determining the \neffect of delay distortion, a raised cosine amplitude spectrum of the \npulses at the detector input has been assumed in all cases, together with \nthe minimum pulse interval permitted with this spectrum, ideal im- \nplementation of each modulation and detection method and optimum \ndesign from the standpoint of slicing levels and sampling instants at the \ndetector output. These idealizations insure that only the effect of delay \ndistortion is evaluated and considered in comparing modulation methods, \na condition that is difficult to realize with experimental rather than \nanalytical comparisons.\n\nAs mentioned above, the present analysis involves a basic function \ncommon to all modulation methods, which in general would be deter- \nmined with the aid of digital computers. This approach has certain \nadvantages in comparison of modulation methods and from the stand- \npoint of optimum system design over direct computer simulation of each \nmodulation method. The latter direct approach may be preferable for \nany specified modulation method and type of transmission impairment \nand has been used in connection with a binary double-sideband AM \nsystem with envelope detection, for cosine and sine variations in trans-\n\nThe transmission impairments caused by linear and quadratic delay \ndistortion, and combinations thereof, have been determined experi-\n\nmentally for a binary vestigial-sideband amplitude modulation data \ntransmission system employing envelope detection.\"\n\nThe present analysis is concerned with certain \u2018\u2018coarse structure\u201d vari- \nations in transmission delay that ordinarily predominate over smaller \n\u201cfine structure\u201d variations, except in transmission facilities where \nelaborate phase equalization is used. Transmission impairments from \nsmall irregular fine structure gain and phase deviations over the channel \nband can be evaluated by methods discussed elsewhere\u201d and are not \nconsidered here.\n\nIn carrier pulse modulation systems the pulse trains at the transmit- \nting end modulate a carrier in amplitude, phase or frequency. In AM \nthe demodulated signal depends on the envelope of the received carrier \npulse train at sampling instants, in PM on the phase of the carrier \nwithin the envelope and in FM on the time derivative of the phase at \nsampling instants. To determine the performance of these various meth- \nods in the presence of transmission distortion it is necessary to formulate \nthe received carrier pulse trains.\n\nThe received carrier pulse trains at the channel output, i.e., the de- \ntector input, can in all cases be formulated in terms of the carrier pulse\n\ntransmission characteristic, that is, the received carrier pulse in response\n\nto a single transmitted pulse. This pulse transmission characteristic is \nrelated to the shape of the modulating pulses at the transmitting end, \nand to the amplitude and phase characteristic of the channel, by a \nFourier integral, as discussed and illustrated for special cases in the \nAppendix. The general formulation of the pulse trains at the detector \ninput and the resultant demodulated pulse trains with various methods \nof carrier modulation and signal detection is dealt with in the following\n\nIt will be assumed that a carrier pulse of reetangular or other suitable \nenvelope is applied at the transmitting end of a bandpass channel. The \nreceived pulse with carrier frequency w, can then be written in the \ngeneral form [Ref. 2, Equation (2.09)|\n\nquadrature components of the received carrier pulse and P. the resultant\n\nenvelope. The time \u00a2 is taken with respect to a conveniently chosen \norigin, for example the midpoint of a pulse interval or the instant at \nwhich RF. or P, reaches a maximum value. \nWith a carrier frequency w) rather than w,, relation (1) is modified\n\nPo(t) = cos (wot \u2014 Wo) Ro(t) + sin (wel \u2014 Wo) Qo(t) (7) \nand relations (2) through (6) are correspondingly modified by replac- \ning \u00a2 by the subscript 0.\n\na received pulse will change, provided the transmission-frequency \ncharacteristic of the channel remains fixed, except in the limiting case \nof a carrier pulse of infinitesimal duration having a flat spectrum. How- \never, by appropriate modification of the transmission-frequency charac- \nteristic the amplitude spectrum of a pulse at the channel output, i.e., \nthe detector input, can be made the same regardless of the carrier fre- \nquency. On the latter premise of equal amplitude spectra at carrier \nfrequencies wy and w,, the following relations apply |Ref. 2, Equation \n(2.18)]:\n\nPULSE TRANSMISSION WITH PHASE DISTORTION \ncarrier phase y, in (1) rather than the carrier phase Yo in (7). If a ear- \nrier phase yp is used as reference, \u00a5, = 0 in (8) and (9), and \nRoo = COS wlRo(t) \u2014 sin wlQo(t), (10) \nQ..o = COS wlQo(t) + sin w,tRo(t). (11) \nWith (8) and (9) in (3), or (10) and (11) in (3): \nP(t) = Py(t) = [Re(t) + Qe(t)}\u2019. (12)\n\nThe resultant envelope of a single pulse is thus the same regardless of \ncarrier frequency and phase, on the premise of a fixed pulse spectrum \nat the channel output as assumed above.\n\nLet carrier pulses be transmitted at intervals 7, and let \u00a2 be the time \nfrom the midpoint of a selected interval. The following designation will \nbe introduced for convenience\n\nwhere n is the time expressed in an integral number of pulse intervals \nof duration 7 and x the time in a fraction of a pulse interval.\n\nLet a(\u2014n) and y.(\u2014n) be the amplitude and phase of the carrier \npulse transmitted in the nth interval prior to the interval 0 under con- \nsideration, and a(n), \u00a5.(m) the corresponding quantities for the nth \nsubsequent interval. The received pulse train in the interval \u20147/2 < \nt < T/2 is then\n\nwhere the summations are between n = \u2014 x andn = @. \nDuring the next interval, 7 to 27, the received wave is obtained by \nreplacing a(n) and \u00a5,(n) by a(n + 1) and \u00a5,(nm + 1) and is thus\n\nIn pulse modulation systems as considered herein it is assumed that \nthe modulating pulses are rectangular in shape and of duration equal to \nthe pulse interval. For equal phases y.(n) = y, of all the modulating \npulses, (14) then becomes\n\nwhere A, is the amplitude of the transmission-frequency characteristic \nof the channel at w = w, and it is assumed in the determination of R, \nand @. that the phase characteristic is zero at w = w,. That is, a con- \nstant transmission delay is ignored, which is permissible without loss \nof generality.\n\nWhen F.(t) and Q.(t) are determined from the channel transmission- \nfrequency characteristic by the usual Fourier integral relations, in the \nform represented by (159) through (163) of the Appendix, the following \nrelations apply for rectangular modulating pulses of duration 7 equal \nto the pulse interval:\n\nWith synchronous detection, also referred to as homodyne and co- \nherent detection, the received wave is applied to a product demodulator \ntogether with a demodulating wave cos (wot \u2014 y.). After elimination of \nhigher frequency demodulation products by low-pass filtering the de- \nmodulated baseband output becomes, when a factor of one-half. is \nomitted for convenience\n\nIf w, is the bandwidth of the modulating signal, the high-frequency \noutput of the product demodulator will have a lowest frequency 20, \u2014 \nw, , Which can be separated from the modulating wave by low-pass fil- \ntering provided 2w, \u2014 w, = w,, or if w, = w, .\n\nAt sampling instants \u00ab = 0, the desired signal is a(0)R(0) and the \nremaining terms in (22) represent intersymbol interference in systems \nwhere R.(n) # Oforn = +1, +2, ete.\n\nOwing to elimination of the quadrature components, synchronous \ndetection is simpler from the standpoint of analysis than envelope de- \ntection, in which the demodulated signal depends on the envelope of \nthe received wave (21) as given by\n\nW(x) = (>> a(n)RAx \u2014 n)\\ + [> a(n)Q.(x - re\u2019. \u00a2@3) \nThe desired signal at sampling instants x = 0 is a(0)[R2(0) + Q2(0)|) \nand the remaining terms in (23) represent intersymbol interference. \n2.5 Phase Modulation with Synchronous Detection\n\nIn phase modulation systems the amplitude a(n) = a = constant \nand the phase y,(n) is varied from one pulse interval to the next. The \nreceived wave (15) then becomes\n\nIn a multiphase system, the received wave is in general applied to \nseveral product demodulators together with a demodulating wave cos \n(wt \u2014 w). In the particular case of two-phase modulation a single \ndemodulator suffices, and the demodulator output after elimination of \nhigh-frequency demodulation products by low-pass filtering and omitting \na factor of one-half is of the general form\n\nwhere as before the summation is between n = \u2014* andn = =. \nThe desired signal is represented by the term for n = 0 and is\n\nWhen the phase y of the demodulating wave is so chosen that 6 = 0, \nand if y.(0) = Oor as in two-phase modulation, then Uy(0) = +P.(0).\n\nIn four-phase modulation two product demodulators are required, \nwith the demodulating waves displaced 90\u00b0 in phase. The output of the \nsecond demodulator is then, in place of (26),\n\nThe preferable choice of the phase of the demodulating wave in the \nabove relations may depend on certain considerations in the imple- \nmentation of modulators and demodulators. In Table I are given the \nfour possible combined outputs as determined by the carrier phase \ny.(0) for the particular cases 6 = 0 and @ = 7/4. For convenience the \noutputs for 6 = 2/4 are normalized to unit amplitude, the actual ampli- \ntudes being +43+/ 2.\n\nIt will be noted that with @ = 0 the output Uy determines whether \none carrier is modulated in phase by y. = 0 or x, while the output V9\u2019 \ndetermines whether the quadrature carrier is modulated in phase by \nv. = Oor x. The two carriers can thus be modulated and demodulated \nindependently, without the need for circuitry to convert the two de- \nmodulator output to carrier phase, as would be required with 6 = 7/4. \nWith differential phase detection, to be discussed in the next section, \nsuch converters would be required both with 6 = 0 and @ = 2/4. In \nthis case @ = 2/4 may be preferable for the reason that only two states \n(\u20141,1) are possible for each demodulator, rather than three states \n(\u20141,0,1) with 6 = 0.\n\nTABLE [ DEMODULATOR OvutTPpUTS U,\u00ae AND V,\u00b0 IN Four-PHASE \nSysTEMS AS DETERMINED BY CARRIER PHASE y,(0) FoR DemMopu- \nLATING WAVES WITH PHASES 6 = 0 AND 7/4\n\nAn alternative method of demodulation that will be considered in con- \nnection with phase modulation is differential phase detection. With this \nmethod W(x) as given by (15) is applied to one pair of terminals of a \nproduct demodulator, and W,(.x) as given by (16) to the other pair \nwith a suitable phase shift 6. The demodulator output is then, with \na(n) constant as in phase modulation,\n\nAfter elimination of high-frequency components present in (31) by \nlow-pass filtering and omitting a factor of one-half, the resultant base- \nband output can be written\n\nIt will be recognized that this expression is of the same form as (27) \nexcept that y,(0) is replaced by the phase change y.(0) \u2014 y.(1) be- \ntween two successive sampling instants. In the particular case of four- \nphase transmission the phase @ may be chosen as, say, 6 = 0 or 7/4, in \nwhich case the outputs of the two product demodulators would be as \nindicated in Table I for phase modulation with synchronous detection, \nexcept that y,.(0) is replaced by the phase difference y.(0) \u2014 y.(1).\n\nWith a signal of bandwidth w, the high-frequency part of (31) will \nhave a lowest frequency 2(w,. \u2014 @,), since it is the product of two high-\n\nsignal represented by (32) will have a maximum frequency 2a, , so that \na flat low-pass filter of minimum bandwidth 2, is required to avoid \ndistortion of the baseband signal. In order that the filter also eliminate \nthe high-frequency components in the demodulator output, it is neces- \nsary that 2(w. \u2014 w,) 2 2w, or w, 2 2, . With synchronous detection it\n\nIn frequency modulation a(n) in (14) is constant and y.(n) varies \nwith time so that the time derivative of [w. \u2014 y.(n)| represents a \nvariable frequency. Pulse transmission without intersymbol interference \nover a channel of the same bandwidth as required for double-sideband \nAM is in this case possible for certain ideal amplitude and phase charac- \nteristics of the channels, as shown elsewhere [Ref. 1, Section 5]. The \nformulation is here modified to include any amplitude and phase char- \nacteristic of the channels.\n\nIt will be assumed that a space is represented by a frequency wo \u2014 @ \nand a mark by a frequency w) + @. Discontinuity in a transition from \nmark to space can then be avoided for rectangular modulating pulses \nof duration 7 provided,\n\nIn a system of minimum bandwidth k = 1, and in this case inter- \nsymbol interference can be avoided with a channel band no wider than \nrequired for double-sideband AM.\n\nWhen a mark is preceded and followed by a space during the nth \npulse interval, the envelope of the resultant carrier pulse is obtained \nwith f = (\u00a2 + nT) in Equation (23) of Ref. 1 and becomes\n\nP(t) = cos (wot + go)(\u20141)\"Ro(t \u2014 nT) \n(42) \noe sin ( wot + Lo) | \u2014| \"Qol(t \u2014_ ni\u2019). \nwhere \u00a2 is the time from the midpoint of interval 0. \nWhen \u00a5( \u2014@) is the phase distortion* at the frequency w) \u2014 @, Equa- \ntion (34) of Ref. 1 is modified into \nE(t) = \u2014cos (aot + go)[A(\u2014a@) cos y \u2014 (\u20141)\"Ro(t \u2014 nT)\n\ny = ot + \u00a5(\u2014@). (44) \nWhen a sequence of marks and spaces is transmitted, the resultant \nwave at the detector input becomes\n\nin which the notation is in accordance with (13), 2 = \u00a2t/T and y = \nmx + (\u2014w). \nThe phase of the wave (45) is given by \nsin y \u2014 uBo(.xr) \ntan Wo t) = y - : 9 \nCOS Y \u2014 Mao x) \nwhere \nmM 1/A(\u2014@o) (49) \nExpression (41) of Ref. 1 for a single pulse is replaced by the following \nfor the demodulated pulse train at \u00ab = \u00a2/T:\n\n* As in Ref. 1, the linear component of the phase characteristic is disregarded \n\u201cnee it only represents a fixed transmission delay.\n\nSince binary FM with frequency discriminator detection is a non- \nlinear modulation method, determination of the optimum signal-to-noise \nratio at the detector input for a given error probability presents a very \ndifficult analytical problem, at least when consideration is given to \nminimum bandwidth requirements together with appropriate shaping \nof bandpass and postdetection low-pass filters. In Ref. 1 these various \nfactors were taken into account, but the signal-to-noise ratios at sam- \npling instants were evaluated on the approximate basis of a steady \nstate carrier representing a continuing space or mark and a relatively \nhigh signal-to-noise ratio. On this basis it turned out that, in the absence \nof a postdetection low-pass filter, binary FM would have a disadvantage \nin signal-to-noise ratio of about 4.5 db compared to an optimum bipolar \nAM or phase reversal system. This would be reduced to about a 1.5-db \ndisadvantage by addition of an optimum low-pass filter. The analysis \nfurther indicated that, for a specified postdetection low-pass filter, there \nwould be an optimum division of shaping between the transmitting and \nreceiving bandpass filters that would give a slight advantage in signal- \nto-noise ratios over an optimum bipolar AM system. In view of the ap- \nproximations involved, the above analysis does not prove that such an \nadvantage exists. Rather, it is probable that optimum bipolar AM has \nsome advantage in signal-to-noise ratio over optimum binary FM. This \nis indicated by other analyses that do not assume a high signal-to-noise \nratio but introduce other approximations in that they do not consider \nfrequency discriminator detection or the shaping of band-pass filters or \nthe effect of a postdetection low-pass filter.\n\nIt is well known that an approximation is involved in assuming high \nsignal-to-noise ratios and thus ignoring the breaking phenomenon in \nFM. Moreover, even in the absence of intersymbol interference, it is an \napproximation to assume a steady state carrier over a short sampling \ninterval, regardless of the transmitted code, as shown below.* Referring \nto Equation (202) of Ref. 1, random noise at the detector input can be \nwritten in the form\n\nIn the absence of intersymbol interference at sampling instants, \na(0) = Oor 1, B(0) = 0, y = O and pz = 2. In this case appropriate \nmodification of (50) gives for the demodulated signal plus noise at \nsampling instants\n\nwhere r; = r:(0), gi = qi(O) and gq,\u2019 = dq,(t)/dt for t = 0. \nIf r; < 1, the last equation is approximated by\n\nwhere ()(0) = a(0) = 0 for space and 1 for mark, and the interfering \nvoltage after demodulation is\n\nThe first two terms represent the conventional approximation for a \ncontinuing mark or space and a high signal-to-noise ratio.\n\n0. This is not the case except for a continuing space, a continuing mark \nor a mark preceded and followed by a continuing space. For other \ncombinations of transmitted pulses there is some contribution from the \nthird term. In the particular case of a raised cosine pulse spectrum, as \nconsidered herein, the maximum effect for a random pulse train is less \nthan 0.15 db and can thus be ignored. For narrower pulse spectra the \neffect may be appreciably greater.\n\nIn the analysis that follows, transmission impairments from inter- \nsymbol interference owing to phase distortion will be evaluated on the \nsame basis for fM as for the other modulation methods, although the \napproximations involved may be somewhat greater.\n\nAs indicated by the preceding derivations, the demodulated wave is \nrelated to the received carrier wave Wo(.) in a manner that depends on \nthe carrier modulation and detection method. In general the demodu- \nlated wave at sampling instants may assume a number of different\n\namplitude or state a, of the transmitted signal and l the demodu- \nlated wave at a sampling instant for an adjacent amplitude or state\n\ning to intersymbol interference, and also a certain sequence resulting\n\na.4, and of positive and negative noise voltages, the optimum level for \nso > r(s r(s+1) -\n\nThe minimum margins are obtained with U' = Uys.\u201d and with \nr(s+l1 , s+1) - \u2014 ne nn 5 te \u2018 \n= Unin in (58) and (59). The minimum margins thus be-\n\nFor sequences of marks and spaces, or other signal patterns, such \nthat the minimum margins for distinction between adjacent signal \nstates are obtained, an error will occur if the noise voltage at the sam- \npling instant exceeds M,,i, in amplitude and has the appropriate \npolarity. (Polarity is immaterial except for the two extreme signal \nstates.) For other signal patterns the tolerable amplitude of the noise \nvoltage is greater. The value of 7,,;, relative to the value in the absence \nof intersymbol interference thus gives the maximum transmission im- \npairment. The average impairment obtained by considering various \npulse train patterns and the corresponding values of M\u201c\u2019 and M\u201c*\" \nas given by (58) and (59) will be less, as discussed below.\n\nBy way of illustration it will be assumed that all values of M between \nM min and a maximum value M,,,. are equally probable, and that the \nnoise has a gaussian amplitude distribution. With a given fixed value of \nM the probability of an error can be written as\n\nwhere erfe = 1 \u2014 erf is the error function complement and a is a factor \nthat depends on the ratio of signal power to noise power. \nConsidering all noise margins between the limits mentioned above. \nthe average error probability becomes \nl l \u00bbM max\n\nlor k = 1, the latter expression conforms with (61). \nThe maximum error probability would be obtained by considering a \nfixed noise margin equal to Win and would be\n\np. = 3 erfe A. (66) \nThe error committed in assuming .V/,,;, can be determined by writing \nLan) . aS \np. as given by (65) in the form \npe = 4 erfe (cA), (67) \nwhere \u00a2 2 1 is so chosen that (67) equals (65). \nThe average noise margin is then\n\nBy way of numerical illustration let A be so chosen that p, as given \nby (66) in one case is 10 \u2018 and in another case 10\u00b0\u00b0. The results given \nin Table II are then obtained from (65) and (67).\n\nTABLE II Ratio c = M/Myin FoR EQuaL PROBABILITY OF ALL \nNorsE MarGINs BETWEEN Myin AND Mmax = k Min FOR NOISE \nWITH A GAUSSIAN AMPLITUDE DISTRIBUTION\n\nmargin, whereas the actual impairment would be 1.4 db less for an error \na . 4 > ons Br \nprobability of 10\u00b0, about 1 db less for an error probability 10 \u00b0. For an\n\nerror probability of 10 \u00b0 or less the error committed in evaluating trans- \nmission impairments on the basis of the minimum noise margin can be \ndisregarded. This also applies for greater error probabilities when the \ntransmission impairment based on the minimum noise margin is small, \nin which case k < 2.\n\nAmplitude modulation can be used in conjunction with envelope de- \ntection and synchronous detection. The former method is simplest from \nthe standpoint of implementation, but synchronous detection, also re- \nferred to as homodyne and coherent detection, affords an improvement \nin signal-to-noise ratio. Since synchronous detection is also the simplest \nmethod from the standpoint of analysis, it will be considered here, ex- \ncept for a comparison of envelope and synchronous detection for binary \ndouble-sideband AM.\n\nAmplitude modulation in general implies several pulse amplitudes, \nand can be used with double-sideband and with vestigial-sideband trans- \nmission. The particular case of bipolar binary AM with synchronous \ndetection is equivalent to two-phase modulation.\n\nWith amplitude modulation and synchronous detection it is possible \nto transmit pulse trains on two carriers at quadrature with each other, \nand under certain idealized conditions to avoid mutual interference. The \nspecial case of bipolar binary AM on each of the two carriers is equiva- \nlent to four-phase modulation.\n\nThe signal-to-noise ratio as related to error probability is discussed \nelsewhere (Ref. 1, Section XVIII) for various optimized binary AM or\n\nPM systems on the premise of ideal synchronous detection. Ideal syn- \nchronous detection for AM or PM as assumed here can in principle be \napproached without penalty in signal-to-noise ratio, by various methods \nof implementation. For example, a demodulating wave for a product \ndemodulator can be derived with the aid of a resonator of sufficiently \nnarrow bandwidth (high Q) tuned to the carrier frequency, or the \nsecond or the fourth harmonic thereof, depending on the particular \nmethod and on whether two-phase or four-phase modulation is used. A \ndemodulating wave can also be supplied from an oscillator at the re- \nceiving end, the phase of which would be controlled by comparison with \nthat of the carrier of the received signal. Such phase-locked oscillator \nmethods have been devised for analog signal transmission by suppressed \ncarrier double-sideband AM?\u2019 and vestigial-sideband AM.\u00b0 With any one \nof the above methods, noise in the demodulating wave would be vir- \ntually absent, as would the effect of phase distortion in the channel. \nActually some penalty in signal-to-noise ratio as compared to ideal \nsynchronous detection would be incurred, owing to unavoidable fluctua- \ntions in the amplitude and phase of the demodulating wave, resulting \nfrom the finite bandwidth of the resonators and mistuning, or from im- \nperfect oscillator control. A common property of these methods is that \na rather long time, as measured in pulse intervals, is required to establish \na demodulating wave, if the above fluctuations in amplitude and phase \nare to be held within tolerable limits. This may be a disadvantage in \ncertain applications, which in the case of phase modulation can be over- \ncome by differential phase detection, in exchange for a penalty in signal- \nto-noise ratio resulting from the presence of both noise and phase dis- \ntortion in the demodulating wave, as discussed in Section IV.\n\nA general formulation is given here of intersymbol interference and \nresultant maximum transmission impairment as related to the carrier \npulse transmission characteristic, together with illustrative applications \nto the particular cases of linear and quadratic delay distortion. The \nformulation is, however, applicable to any given gain and phase devia- \ntion over the channel band, provided the carrier pulse transmission \ncharacteristic has been determined, which in general would entail Fourier \nintegral evaluation with the aid of computers.\n\nTHE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1961 \nThe following notation will be used:\n\nwhere FR.\u201d designates positive values of R. and R. absolute values \nwhen R, is negative.\n\nLet there be / different amplitude levels, between a minimum ampli- \ntude @min and a maximum amplitude dnax . When a pulse of amplitude \na, = a,(Q) is transmitted, the maximum value of (69) is\n\nry 7 . a . 0 - \nChe value of Myyin as given by (74) is smaller than /\u201d in the absence \nof intersymbol interference by the factor\n\nin which FR, designates the absolute value of R... \nThe factor R.(0)/R.'(0) represents the transmission impairment ow- \ning to reduction in pulse amplitude at sampling instants. The summation\n\nterm represents transmission impairments owing to intersymbol inter- \nference.\n\nRelation (77) applies regardless of the polarity of the transmitted \npulses and for both symmetrical (double sideband) and asymmetrical \n(vestigial sideband) systems. The special case 1 = 2 and Quin = \u20144max \nrepresents binary bipolar AM, which can also be regarded as two-phase \ntransmission,\n\nWith synchronous detection it is possible under certain ideal condi- \ntions to transmit signals on two carriers at quadrature without mutual \ninterference. In general, however, the quadrature component in (21) \nwill in this case give rise to interference and (69) is replaced by\n\nwhere b(n) are the pulse amplitudes in the quadrature system. \nlor equal differences between maximum and minimum amplitudes in \nthe two systems, i.\u20ac., dmax \u2014 Gmin = Vmax \u2014 Dmin, (77) is replaced by \nRO) t\u20141 \nR.(O) R.(O)\n\nIn general the phase of the demodulating carrier can be so chosen \nthat Q.(0) = 0, as is demonstrated later.\n\nExpression (80) applies regardless of pulse polarities in the two \nquadrature systems. The special case of two binary bipolar AM sys- \ntems, i.e., 2 = 2 and dyin = Dmin = \u2014Amax = \u2014bmax Can also be regarded \nas four-phase transmission.\n\nWhen the spectrum of a received pulse at the detector input has even \nsymmetry about the carrier frequency, and the phase characteristic has \nodd symmetry (even symmetry delay distortion), the quadrature com-\n\nponents Q.(n) vanish (see the Appendix). In this case (77) and (80) \nare identical, so that there is no mutual interference between pulse \ntrains transmitted on two carriers at quadrature. In this special case it \nis thus possible by quadrature carrier AM to realize a two-fold increase \nin pulse transmission rate, without increased intersymbol interference. \nAn alternative means to the same end is to use vestigial sideband trans- \nmission, as discussed below.\n\nLet T be the pulse interval in double-sideband AM, in which case the \npulse interval in vestigial-sideband AM would be 7\u201d = 7/2. Returning \nto (10) and (11) let Ro(t) be the in-phase component in double-side- \nband AM, and let Qo(\u00a2t) = 0 for an amplitude spectrum with even sym- \nmetry about wy and a phase characteristic with odd symmetry. Let w, \nbe the carrier frequency from midband in vestigial-sideband transmis- \nsion. By appropriate choice of w, it is possible to make w,7\"\u2019 = 7/2, in \nwhich case cos w,7T\u2019 = 0, sin wT\u2019 = 1. The following relations are thus \nobtained:\n\nIn accordance with the above relations, at even sampling points only \nthe in-phase components are present and are the same as in double-side- \nband AM. At odd sampling points the quadrature components are pres- \nent, but are eliminated with synchronous detection and need not be \nconsidered.\n\nIn summary, when the amplitude spectrum at the detector input has \neven symmetry about the midband frequency, and the phase character- \nistic has odd symmetry, relation (77) applies for double-sideband AM, \nquadrature double-sideband AM, vestigial-sideband AM, as well as \nspecial cases thereof, such as two-phase and four-phase modulation.\n\nIn the next section numerical results are given for the special case of \na raised cosine spectrum at the detector input with quadratic delay \ndistortion about the midband frequency.\n\nFig. 1 Raised cosine pulse spectrum and transmission-frequenecy character- \nistic of channel for double- and vestigial-sideband AM. Curve 1: Spectrum of car- \nrier pulse envelope at detector input (channel output); S(u)/S(0) = cos* (ru/4e). \nCurve 2: Transmission-frequency characteristic of channel with rec se trans \nmitted carrier pulses of duration 7\u2019 = r/o and carrier at midband; A(u)/A(O) = \n(ru/4m)/tan (ru/4o). Curve 3: Transmission frequency characteristic of channel \nwith rectangular transmitted carrier pulses of dur: ation T/2 and carrier at u =a; \nA(u)A(O) = cos? (ru/4@){[(u \u2014 &)/4@]/sin [x(u \u2014 @)/4a]}.\n\nNOTE: \nENVELOPES ARE SYMMETRICAL IN \nAMPLITUDE ABOUT ZERO LINE AND \nONLY ONE SIDE OF ENVELOPE IS SHOWN\n\nFig. 2 \u2014 Carrier pulse transmission characteristic for raised cosine spectrum \nas in Fig. 1 and quadratic delay distortion.\n\nThe shape of the transmission-frequency characteristic of the channel \nrequired to this end depends on the shape of the transmitted pulses. It \nis shown in Fig. 1 for rectangular modulating pulses, with the carrier at \nmidband and also with the carrier to one side of midband, as in vestigial- \nsideband transmission. These characteristics, together with the optimum \ndivision of shaping between transmitting and receiving filters, are dis- \ncussed in Section XIV of Ref. 1. Even though the amplitude characteris- \ntics of the detector input spectra are the same in double- and vestigial- \nsideband transmission, it is necessary to use different shaping of \ntransmitting filters, as indicated in Fig. 1, since the rectangular modulat- \ning pulses have different carrier frequencies and different durations.\n\nThe phase characteristic is assumed to contain a linear component, \ntogether with phase distortion component varying as the third power of \nfrequency from midband, which corresponds to delay distortion increas- \ning as the second power of frequency from midband, as indicated in Fig. \n2. The function Ro(t/T) = Ro(x + n) for this case has been determined \nby numerical integration, as discussed in the Appendix. It is given in \nTable XIX of the Appendix and shown in Fig. 2. The values for \u00ab = 0, \ni.e., integral values of t/T are given in Table ITI.\n\nTABLE III Function Ro(n) FoR RAISED CosINE SPECTRUM \nAND QuapRATIC DELAY DisToRTION\n\nWith exact evaluation of Ro(n) the summation }\u00b0R)(n) between \n= \u2014x andn = & should equal 1.\n\nWith the values of Ro(n) given in Table III, the values of 7, Fy ob- \ntained from (78), and of main as Obtained from (77) are given in Table \nIV.\n\nTABLE IV FACTOR nmin FOR RAISED COSINE SPECTRUM AND \nQuapraATIC DeLay Distortion FOR SYNCHRONOUS AM \nwith | AMPLITUDE LEVELS*\n\nsideband AM and for quadrature double-sideband AM, and special cases \nthereof, such as two-phase and four-phase transmission. Since the quadra-\n\nture component is absent the factors also apply for double-sideband AM \nwith envelope detection. It should be noted that 7 in all cases is the \npulse interval in double-sideband AM, which is twice the pulse interval\n\nFig. 3 Factor nmin for raised cosine pulse spectrum and quadratic delay dis \ntortion as in Fig. 2, for AM systems employing synchronous detection and / pulse \namplitudes. Factor applies for double- and vestigial-sideband AM and quadrature \ndouble-sideband AM.\n\nin vestigial-sideband AM or twice the combined pulse interval in quadra- \nture double-sideband AM.\n\nIn the above evaluation it was assumed that the pulses were sampled \nat \u00a2 = 0, which is not at the peak of a pulse except for d/T = 0. For \nd/T = 4 the pulse peak is nearly at (/T = x = 0.25. Sampling at \nx = 0.25 gives, for ] = 2, nuin = 0.356 rather than 0.347.\n\nThe factor nmin expressed in decibels, as in Fig. 3, indicates the maxi- \nmum transmission impairment, i.e., the maximum increase in signal-to- \nnoise ratio required at the detector input to compensate for the effect \nof delay distortion. This maximum impairment would be closely ap- \nproached for signal-to-noise ratios such that the error probability is suf- \nficiently small, say less than 10\u00b0\u2018. However, for error probabilities in \nthe range ordinarily considered the transmission impairment will be \nless, in accordance with the discussion in Section 2.10. For example, for \nd/T = 4, Ro(0) = 0.734 and (0) = 0.385. The maximum noise mar-\n\nFor k = Minax/Mmin = 3.2, the results given in Table II indicate that \nthe transmission impairments would be less than the maximum by \nabout 1 db and 1.4 db for error probabilities of 10\u00b0\u201d and 10 * respec- \ntively. Ford/T = 4and/ = 2 the maximum impairment indicated in \nFig. 3 is \u201420 logy Min & 9.2 db, whereas impairments of about 8.2 \nand 7.8 db would be expected for error probabilities of 10\u00b0 and 10\u201c, on \nthe premises underlying the evaluation in Section 2.10.\n\nWhen the pulse spectrum at the detector input has even symmetry \nabout the midband frequency and the phase characteristic has a com- \nponent of even symmetry, i.e., odd symmetry delay distortion, the in- \nphase and quadrature components both have even symmetry with re- \nspect to \u00a2. That is,\n\nWith synchronous detection the phase of the demodulating carrier \nwould preferably be so chosen that Qo(\u00a2) would vanish at \u00a2 = 0, since \nthis would give the maximum amplitude of the demodulated pulse at a\n\nit is therefore convenient to modify the phase angle so that the quadra- \nture component vanishes at \u00a2 = 0. The modified quantities are related \nto Ry and Qo by (see Appendix )\n\nIn the case of vestigial-sideband transmission at pulse intervals \ni T/2, the in-phase component referred to a carrier at frequency \nwy + @ is obtained from (10) and becomes\n\nFor the special case of a raised cosine spectrum and linear delay dis- \ntortion the functions Rp and Qp have been determined by numerical \nintegration, as discussed further in the Appendix. They are given in \nTable XX for certain ratios d/7, where d is the difference in delay be-\n\nNOTE: \nENVELOPES ARE SYMMETRICAL IN AMPLITUDE \nABOUT ZERO LINE AND ONLY ONE SIDE \na ENVELOPE IS SHOWN\n\nFig. 4 \u2014 Carrier pulse transmission characteristics for raised cosine spectrum \nas in Fig. 1 and linear delay distortion \ntween the midband frequency and maximum sideband frequency as il- \nlustrated in Fig. 4. The modified functions Ro and Qo are given in Ta- \nble X XI and are shown in Fig. 4. For negative values of t/ 7, Roo and Qoo \nare the same as shown in Fig. 4 for positive values.\n\nFor double-sideband transmission the factors qmin given in Table V \nare obtained from (77), with F. taken in accordance with (87). These \nfactors are shown in Fig. 5. The case | = 2 corresponds to two-phase \ntransmission.\n\nFig. 5 \u2014 Factor nmin for raised cosine pulse spectrum and linear delay distor \ntion as in Fig. 4 for double-sideband AM systems employing synchronous detec- \ntion and 7 pulse amplitudes.\n\nFor quadrature double-sideband AM the factors in Table VI are ob- \ntained from (80) with 7, and g, taken in accordance with (87) and (88). \nThese factors are shown in Fig. 6. The case 1 = 2 corresponds to four- \nphase transmission.\n\nlor vestigial-sideband transmission the factor nain is determined from \n(77), with #, taken in accordance with (92). The factors given in Table \nVII are thus obtained.\n\nTaBLE VI \u2014 Factor nmin FOR QUADRATURE DOUBLE-SIDEBAND \nAM For / = 2 anp 3 PuLSE AMPLITUDES\n\nFig. 6 Factor nmin for raised cosine pulse spectrum and linear delay distor \ntion as in Fig. 4 for quadrature double-sideband AM systems (solid lines) and \nvestigial-sideband AM systems (dashed lines) employing synchronous detection \nand 1 = 2 and 3 pulse amplitudes\n\nTaB_LeE VII FACTOR nmin FOR VESTIGIAL-SIDEBAND AM \nFoR 1 = 2 anp 3 PULSE AMPLITUDES\n\nThe factors nuin for vestigial sideband AM are compared in Fig. 6 \nwith the corresponding factors for quadrature double-sideband AM. \nWith ideal transmission-frequency characteristics and ideal synchronous \ndetection the two methods are equivalent from the standpoint of channel \nbandwidth requirements and optimum signal-to-noise ratio for a given \nerror probability. As shown in Section 3.4, this also applies for pulse \nspectra at the detector input with even symmetry about the midband \nfrequency, in the presence of delay distortion with even symmetry. The \nequations in Section 3.6 and the curves in Fig. 6 show that the above \ntwo methods are not equivalent in the presence of delay distortion with \nodd symmetry about the midband frequency. With linear delay distor- \ntion the factor nmin is, however, very nearly the same with both methods. \nFor practical purposes quadrature double-sideband AM and vestigial- \nsideband AM can be regarded as equivalent with any type of delay dis- \ntortion that would be expected in actual facilities. This equivalence \nwould apply on the premise of ideal synchronous detection but not neces- \nsarily with actual implementation of synchronous detection, for the rea- \nson that the penalty in signal-to-noise ratio incurred in deriving a de- \nmodulating wave may not be the same with both methods.\n\nIn the preceding analysis ideal synchronous detection was assumed, \nwhich permits the use of bipolar pulses. An alternative method that is \nsimpler in implementation is envelope detection, which, however, entails \nthe use of unipolar pulse transmission and for this reason has a certain \ndisadvantage in signal-to-noise ratio as compared to synchronous detec-\n\ncertain cases be greater with envelope than with synchronous detec- \ntion, as shown below.\n\nWhen both the pulse spectrum and delay distortion have even sym- \nmetry about the carrier frequency, so that the quadrature component is \nabsent, the effect of delay distortion is the same as with synchronous \ndetection. The results given in Table IV thus apply also for double-side- \nband AM with envelope detection.\n\nWhen a quadrature component is present in the carrier pulse trans- \nmission characteristic the resultant demodulated wave is in accordance\n\nOwing to the presence of both in-phase and quadrature components, \nit does not appear feasible to derive a simple general expression for \nUmax and Uyin*\u00ae*\u201d similar to (72) and (73). These values can, how- \never, be determined by examining several combinations of transmitted \npulses, as illustrated below for binary pulse transmission, a raised cosine \npulse spectrum at the detector input and linear delay distortion. Using \nvalues of Ro and Qo given in Table XXI of the Appendix, the results \nare as shown in Table VIII. Since both Ro(t) and Qo(t) in this case have \neven symmetry about \u00a2 = 0 the maximum effect of delay distortion is \nencountered for pulse trains with even symmetry about the sampling \npoint, i.e., a(\u2014n) = a(n). Hence, only pulse trains with this property \nneed to be considered.\n\nFrom Table VIII can be obtained W max\u201d and W min\u2019, as indicated by \nasterisks, together with the optimum slicing level given by (57) and the\n\nTABLE VIII VaLues oF (0) For RAIsED CosINE SPECTRU} \nLINEAR DeLay DisTorTION FOR VARIOUS COMBINATIONS \noF MARKS 0 AND SPACES = |\n\nTABLE LX \u2014 Factor nmin WITH BINARY AM AnD ENVELOPE \nDETECTION FoR RatsED CostNnE SPECTRUM AND LINEAR \nDeLAY DISTORTION\n\nIt will be recognized that synchronous detection has a significant ad- \nvantage over envelope detection as regards transmission impairments \ncaused by pronounced linear delay distortion, for the reason that the \neffect of the quadrature component is eliminated. In general, delay dis- \ntortion will have a component of even symmetry and a component of \nodd symmetry about the carrier frequency, in which case the quadrature \ncomponent will be smaller. The advantage of synchronous detection as \nregards transmission impairments caused by delay distortion will then \nbe less than indicated in Table [X. The principal advantage of syn- \nchronous detection is that it permits the use of bipolar transmission, \nwhich in the case of binary systems as considered above affords about \n3 db improvement in the ratio of average signal power to average noise \npower for a given error probability (Ref. 1, Table VIT).\n\nIn the case of vestigial-sideband transmission a pronounced quadrature \ncomponent is present even in the absence of phase distortion. The ad- \nvantage of synchronous detection over envelope detection is in this case \nsignificantly greater than for double-sideband transmission considered \nabove, for the reasons that bipolar transmission can be used and quadra- \nture component is eliminated. In the absence of phase distortion and \nwith a raised cosine pulse spectrum at the detection input, synchronous \ndetection has about a 9 db advantage over envelope detector in the ra- \ntio of average signal power to average noise power for a given error \nprobability (6 db owing to elimination of quadrature component and 3 \ndb owing to bipolar transmission).\n\nvaluation of transmission impairments from phase distortion is more \ncomplicated for envelope than for synchronous detection. These impair- \nments have been determined experimentally for a binary vestigial-side- \nband system with an approximately raised cosine spectrum at the de- \ntector input, for linear and quadratic delay distortion and combinations\n\n\u00bb A om . n \u00b0 \u00b0 \u00bb \nthereof.. They are significantly greater than determined herein for syn-\n\nchronous detection. Hence envelope detection entails more phase equali-\n\nzation than synchronous detection, unless a greater disparity in signal- \nto-noise ratio is accepted than the 9 db applying in the absence of phase \ndistortion.\n\nIn phase modulation with differential phase detection, the demodulator \noutput would under ideal conditions depend on changes in carrier phase \nbetween two successive pulse intervals of duration 7. In its simplest \nand ideal form, the signal with two-phase modulation would be applied \nto one pairof terminals of a product demodulator, while the signal de- \nlayed by a pulse interval 7 would be applied to the other pair. With four- \nphase modulation two product demodulators are required, each with a \ndelay network at one pair of terminals. In addition, a phase shift of 90\u00b0 \nmust be provided between all frequencies of the demodulating waves of \nthe two demodulators, as indicated in Fig. 7. Such a phase shift over a \nfrequency band can be realized in principle and closely approached with \nactual networks.\u2019 The modulator outputs would be applied to low-pass \nfilters of appropriate bandwidth for elimination of high-frequency de- \nmodulation products, and the output of these would be sampled at in- \nterval T. The phase of the carrier would be indicated by the combined \noutput as discussed in Sections 2.5 and 2.6.\n\nWith the above method it is possible with ideal channel characteristics \nto avoid intersymbol interference at sampling instants, without the need \nfor a wider channel band than required with synchronous detection. \nHowever, the two methods are not in all respects equivalent from the\n\nFig. 7 Basic demodulator arrangement for four-phase modulation with dif \nferential phase detection\n\nstandpoint of bandwidth utilization. As discussed in Sections 2.5 and \n2.6, with synchronous detection the carrier frequency must be at least \nequal to the maximum baseband signal frequency, whereas with differ- \nential phase detection it must exceed twice the maximum baseband \nsignal frequency. This requirement does not impose a limitation on \nbandwidth utilization with differential phase modulation provided the \nmidband frequency of the available channel is at least twice the lowest \nfrequency, or that this condition is realized through frequency transla- \ntion prior to demodulation.\n\nWith differential phase detection the demodulating wave is established \nwithout the need for a long delay (measured in pulse intervals) as re- \nquired with certain other methods inentioned in Section 3.1. Moreover, \na substantial fluctuation in carrier phase can be tolerated, since only the \ndifference in phase between adjacent pulses need be considered. These \nadvantages are realized in exchange for a penalty in signal-to-noise ratio \nas compared to ideal synchronous detection owing to the presence of \nnoise in the demodulating wave. For very small error probabilities, and \nassuming ideal implementation in all respects, this impairment is about \n1 db for two-phase and about 2.3 db for four-phase modulation.\u201d Com- \nparable penalties in signal-to-noise ratio as compared to ideal synchro- \nnous detection may be incurred with the other methods of providing a \ndemodulating wave mentioned in Section 3.1, owing to small unavoidable \namplitude and phase fluctuations in the demodulating wave resulting\n\ndetection, greater transmission impairments would be expected from \nphase distortion, since the effect of phase distortion, like that of noise, is \npresent in both the signal and the demodulating wave. The transmission \nimpairments resulting from quadratic delay distortion are determined \nhere, and compared with that encountered with ideal synchronous detec- \ntion.\n\nOther implementations of differential phase detection than assumed \nherein have been used, but in principle these entail a wider channel band \nthan with ideal synchronous detection. For example, the two demodu- \nlator inputs or outputs could be integrated over a pulse integral 7 with \nthe aid of a narrow-band resonator tuned to the carrier frequency, and \nthen be rapidly quenched before the next signal interval. When the \nchannel bandwidth is limited, the phase of the demodulating carrier will \nthen depend on the phases of the carrier during several pulse intervals. \nThus some intersymbol interference from bandwidth limitation is en- \ncountered even in the absence of phase distortion, and the effect of phase \ndistortion will be greater than that determined herein. However, exces-\n\nsive transmission impairments from bandwidth limitation and phase dis- \ntortion can be avoided by appropriate techniques, as when a large num- \nber of narrow channels are provided within a common band of much \ngreater bandwidth than that of the individual channels.\u201d\n\nIn differential phase modulation the carrier would be at midband, i.e., \nwe = w. With Um = V, expression (33) for the demodulated signal \nbecomes\n\nThe above expressions apply for the output of the single demodulator in \ntwo-phase systems, in which @ = 0. In four-phase systems the output of \none demodulator is obtained with 6 = 6, and the output of the other \nwith @ = 0 = 6 + 7/2.\n\nExamination of (97) shows that the term for m = n + 1 is independent \nof the phase difference yo( \u2014n) \u2014 Yo( \u2014m + 1) and is given by\n\n2 \nVo= > cos [gol \u2014n) \u2014 go( \u2014n \u2014 1) \u2014 6]Py(\u2014n)Py(\u2014n \u2014 1). (98) \nn=\u2014 \u00a9 \nDetermination of transmission impairments becomes rather difficult \nexcept for the special case in which go(n) = 0, which will be considered \nin further detail below.\n\nWhen the pulse spectrum has even symmetry about the midband \nfrequency, and the phase characteristic has odd symmetry (i.e., even \nsymmetry delay distortion) the quadrature component of P(t) van- \nishes, i.e., \u00a2(\u2014n) = 0. In this case, Py = Ry and (96) becomes\n\nWith synchronous detection, two-phase modulation could be used in \nconjunction with both double-sideband and _ vestigial-sideband trans- \nmission. With differential phase detection, however, vestigial-sideband\n\ntransmission is not practicable, since severe transmission impairments \nwould be incurred even in the absence of phase distortion, owing to the \npresence of the quadrature component. Hence only double-sideband \ntwo-phase modulation is considered here.\n\nAssume that in (107) a sequence of values of a,,(0) has been chosen, \nfor example a.sQ@) = |]. ad) = 1, asl(O) = 1, a0) 1, a(Q0) = \n\u20141, a,(0) = 1, ete. For any other value of n than n = 0, the sequence \nof a,(n) will either be identical with that for a,,(0), or all signs will \nbe reversed. This follows from (104), since Yo(\u2014n) will differ from \n\u00a5o(0) by 0 or x. Hence, for n # 0, the right-hand side of (106) can be\n\nIn the absence of intersymbol interference, V as given by (99) would \nbe \u20141 or 1. In the following, the minimum possible value of V will be \ndetermined, on the assumption that V 1 without intersymbol inter- \nference; 1.\u20ac., ao(0) -\n\nConsider first the term Sof (0) in (99). The minimum possible value \nof So is obtained from (107) by choosing a,,(0) \u20141 for Ry(\u2014m) > 0 \nand a,,(0) = 1 for Ro( \u2014m) < 0. The following relation is thus obtained \nfor the minimum possible value of Sp , on the above premise of ap(0) = 1:\n\nwhere Ry designates the absolute values. In the above expression a,(0) \nwould be taken as a,(0) 1 if Ro \u20141) < O and as a,(0) \u2014lif \nRo(\u20141) > 0. The term [1 a,(0)|Ro(\u2014 1) can therefore be written alter- \nnatively as Ro( \u20141) + Ro( \u20141), in which case (109) becomes\n\nof the demodulated voltage with synchronous detection, in the presence \nof a mark, as given in a somewhat more general form by (73).\n\nHaving thus determined So min it follows from (108) and (110) that \nthe two possible associated values of S, min are given by\n\nwhere the term [1 \u2014 a,4:(\u201d)|Ro(\u2014n \u2014 1) in (108) has been replaced \nby the equivalent representation by the first two terms in (112).\n\nTo obtain the minimum value of V as given by (99), each term in the \nseries must be made to have the maximum negative value. To this end \nthe negative sign in (112) for Uyin is chosen if Ro(n) is positive, and the \npositive sign if Ro(nm) is negative. The minimum possible value of V \nthus obtained with (110) and (112) in (99) is\n\nIn accordance with the discussion in Section 4.2, the demodulator \noutput contains a bias or pc component V\u00bb given by (103). Optimum \nperformance is obtained when the threshold level for distinction between \nV = 1 and V = \u20141 is made equal to Vo. When Vo is subtracted from \nboth sides of (114) the following expression is obtained for two-phase \nmodulation :\n\nWhen intersymbol interference is absent at sampling instants, \nRo(n) = 0 for n \u00a5 0, and for n = 0 is Ro (0). In this case Vinin? = \nUmin = [Ro (0)]\u2019. The voltage given by (115) is smaller than in the \nabsence of intersymbol interference by the factor\n\nThe basic difference between two-phase and four-phase modulation is \nthat relation (108) does not apply for four-phase modulation. Returning \nto the discussion following (107), if a sequence a,,(0) is chosen in four- \nphase transmission, the sequence a,,(m) can be chosen independently. \nThis follows from the (104), which shows that if a,,(0) has a given value, \nsay a,(0) = 1, it is possible to make each a,,(n) equal to +lor \u20141 by \nappropriate choice of \u00a5(\u2014n).\n\nFor this reason the minimum value (or maximum negative value) of \nthe right-hand side of (106) is now, for n # 0:\n\nBo ie : (118) \n\u2014R (0) \u2014 >> [Rol \u2014m) + Ro(m)]. \nm=1 \nThe right-hand side of (118) is smaller than for two-phase transmis- \nsion as given by (112) by \u20142R)(0). When this modification is intro- \nduced, the following expression is obtained for four-phase modulation, \nin place of (115) for two-phase modulation:\n\nwhere > is given by (116). \nThe voltage given by (120) is smaller than in the absence of inter- \nsymbol interference by the factor\n\nTABLE X MINIMUM AMPLITUDES OF DEMODULATED PULSE \nTRAINS IN Two-PHASE MopULATION WITH DIFFERENTIAL \nPHASE DETECTION\n\nThe function Ro(n) for this case is given in Table IIL. The values of \nUmin for synchronous detection are given in Table IV for / = 2. In \nTable X are given the various quantities appearing in expression (115) \nfor the minimum amplitudes of received pulse trains at sampling instants \nwith optimum slicing lead equal to the pc component V\u00bb. The values \nof Vinin. = Mmin. are shown in Fig. 8.\n\nFig. 8 Factor nmin for raised cosine spectrum and quadratic delay distortion \nas in Fig. 2 for synchronous detection and differential phase detection. Curve 1: \nIdeal syne hronous detection applies for two-phase and four-phase modulation \nwith carrier at midband and pulses at intervals 7\u2019, and for vestigial-sideband \ntransmission with pulses at intervals 7'/2. Curve 2: Ideal differential phase detec \ntion \u2014 two-phase modulation with pulse interval 7\u2019. Curve 3: Ideal differential \nphase detection \u2014 four-phase modulation with pulse interval 7\u2019.\n\nTABLE XI Minimum AMPLITUDES OF DEMODULATED PULSE \nTRAINS IN Four-PHASE MODULATION WITH DIFFERENTIAL \nPHASE DETECTION\n\n* Reversal of sign indicates a reversal in sign of the demodulated pulses.\n\nWith the accuracy used herein it turns out that 2 and Vo are numeri- \ncally equal but are not identical.\n\nIt will be noted that, when delay distortion is pronounced, the bias \ncomponent Vo is appreciable, and that a significant penalty can be in- \ncurred if the threshold or slicing level is taken as 0 rather than Vo. For \nexample, with d/T = 4 and 0 threshold level the minimum amplitude \nof a demodulated pulse for a carrier phase y = 0 would be 0.564, and \nthe minimum negative amplitude for a carrier phase y = x would be \n\u20140.12. With the optimum threshold level the minimum amplitudes are \n+0.342. Hence the tolerable peak noise amplitudes would be greater \nby a factor 0.342/0.12 = 2.85.\n\nWith four-phase modulation the values given in Table XI are obtained \nfrom (120). The values of Vinsin\u2019 = Mmin_ are shown in Fig. 8.\n\nIn the above illustrative examples it was assumed that pulses were \ntransmitted at the minimum interval 7\u2019 permitted if intersymbol inter- \nference is to be avoided in the absence of delay distortion. The effect of \ndelay distortion may or may not be reduced by increasing the pulse \ninterval, that is, in exchange for a slower transmission rate. By way of\n\nTABLE XII FuNcTION Ro(n) FoR RAISED CosINE SPECTRUM \nAND Quapratic DeLay Distortion with 50 PER CENT \nINCREASE IN PULSE INTERVAL\n\nTaBLE XIII MINIMUM AMPLITUDES OF DEMODULATED PULSES \nWITH 50 PER CENT INCREASE IN PULSE INTERVALS\n\nillustration it will be assumed that the pulse interval is increased by : \nfactor 1.5, in which case the values of Ry are as given in Table XII.\n\nWith this modification, the various quantities are as given in Table \nXIII.\n\nIn Fig. 9 values of Umin and Vinin\u2019 are Compared with those for the \nminimum interval between pulses. It will be noted that there is no signifi- \n\u2018ant difference in the case of two-phase or four-phase modulation with \nsynchronous detection. With differential synchronous detection some \nadvantage is realized for small delay distortion in exchange for a dis- \nadvantage with pronounced delay distortion.\n\nFig. 9 Effect of pulse interval on factor nmin for raised cosine spectrum with \nquadratic delay distortion (dashed curves: pulse interval 7\u2019, as in Fig. 8; solid \ncurves: pulse interval 1.57\u2019). Curves 1: Ideal synchronous detection \u2014 applies for \ntwo-phase and four-phase modulation with carrier at midband and pulses at in \ntervals 1.57 and for vestigial-sideband transmission with pulses at intervals 0.757. \nCurves 2: Ideal differential phase detection \u2014 two-phase modulation with pulse \nintervals 1.57\u2019. Curves 3: Ideal differential phase detection \u2014 four-phase modula- \ntion with pulse intervals 1.57\u2019.\n\n396 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1961 \nV. BINARY FREQUENCY MODULATION (FSK)\n\nshift keying requires the same minimum bandwidth as double-sideband \nAM. In the absence of transmission distortion from gain and phase \ndeviations, the optimum signal-to-noise ratio required at the detector \ninput for a given error probability is slightly greater than for two-phase \ntransmission with ideal synchronous detection, but would be expected \nto be about the same as for two-phase transmission with ideal differen- \ntial phase detection. Binary FM may be preferable to the latter method \nfrom the standpoint of implementation and has an advantage over the \nsimpler method of binary AM with envelope detection from the stand- \npoint of signal-to-noise ratio and performance during sudden transmis- \nsion level variations.\n\nThe performance of binary FM is determined here for channels with \nlinear and quadratic delay distortion and compared with that of the \nother methods mentioned above. In this analysis ideal frequency dis- \ncriminator detection is assumed, in which the demodulated signal is pro- \nportional to the time derivative of the phase of the received wave. This \ncondition may be closely approached with actual detectors when the \nchannel bandwidth is small in relation to the midband frequency. How- \never, when this is not the case, ideal FM detection is only approximated \nwith conventional frequency discriminators or zero \u2018crossing detectors.\n\nExpression (48) for the demodulated pulse train applies for any am- \nplitude and phase characteristic of the channel. In the case of a continu- \ning space, a(n) = 0 in (46) and (47) and Uo(t) = 0. With a continuing \nmark, a(n) = 1 in the above expressions and\n\nteturning to (21), it will be recognized that (122) and (123) repre- \nsent the in-phase and quadrature components in a binary amplitude \nmodulation system when pulses of duration 7 and alternating polarity \nare transmitted, i.e.,a(m) = (\u20141)\" in (21). The fundamental frequency \nof such a pulse train is @ = T/x. Let A(\u2014@) and \u00a5(\u2014@) be the ampli-\n\ntude and phase characteristic of the channel at the frequency \u2014a@ from \nw); A(@) and \u00a5(w) the corresponding quantities at the frequency @ \nfrom wo. Solution of (122) and (123) for the above steady state condi- \ntion of alternate marks and spaces in binary AM gives\n\nWith (124) and (125) in (50), it turns out by way of check that \nUo(a) = 1 for a continuing mark for any amplitude and phase charac- \nteristic of the channel.\n\nFor pulse trains other than continuing marks or spaces, intersymbol \ninterference will be encountered from amplitude and phase distortion. \nIn the following section special cases of phase distortion will be examined \nfurther. It will be assumed that the amplitude characteristic has the \nappropriate shape so that intersymbol interference can be avoided in the \nabsence of phase distortion. To this end it is necessary that A(\u2014@) \nA(@) = } or\u00bb = 2 as shown elsewhere (Ref. 1, Section V). In this case \n(50) becomes with Uy = U\n\nWhen the amplitude characteristic of the channel has even symmetry \nabout wy and the phase characteristic has odd symmetry Qo(.rx + n) = 0 \nand (126) simplifies to\n\nOptimum performance would be expected when a single pulse is sam- \npled at its peak, a condition which is at least closely approximated with \ny = 0. This condition is met when \u00a2t = \u00a2) is so chosen that\n\n2Zao( Xo) c=: l \nlor further analysis it is convenient to introduce the quantities\n\nwhere designates absolute values when (\u20141)\"Ro(.2x ) is nega- \ntive and \" when it is positive. \nIt will be recognized that\n\nwhere the last relations follow in view of (122), (124) and (128). \nDuring transmission of a space, delay distortion will have an adverse \neffect only if U as given by (133) is positive, since only in this case is \nthe tolerance to noise reduced. To obtain a positive value of UU, it is \nnecessary to have either a 2 3 or a < 0. For a space, a(0) 0 in \n(129) and a value of a 2 3 can then be excluded for any reasonable\n\ndelay distortion. It will, therefore, be assumed that ay < 0. The maxi- \nmum positive value of U\u2019, i.e., the maximum adverse effect of delay dis- \ntortion, is then obtained with the maximum possible negative value of \nag. This maximum value is obtained by choosing a(n) 0 in (129)\n\nwhenever (\u20141)\"Ro(2 \u2014 n) is positive and choosing a(n) = 1 whenever \n(\u20141)\"Ro(x \u2014 n) is negative. The maximum negative value of ao(z) thus \nobtained is given by (134). The corresponding maximum value of U (to) \nin the presence of a space and with sampling at \u00a2 = ft) as defined by \n(132) is\n\nDuring transmission of a mark delay distortion will have an adverse \neffect only if U(t)) < 1. This will be the case if ag > 1 or ay < 3 in \n(133). With a(0) = 1 in (129) for a mark, the condition ag < 3 will \nnot be encountered with any reasonable delay distortion and only the \ncase a > 1 needs to be considered. The minimum positive value of \nU(t) in the presence of a mark is obtained when ap is taken as the maxi- \nmum positive value given by a(x) = Ro(x) + ao (x), where ay\u2019 is \ngiven by (135). In view of (136) it follows that, for y = 0,\n\nWith (138) in (133) the minimum amplitude of a pulse train in the \npresence of a mark and at the sampling instant fy) defined by (132) be- \ncomes:\n\nThe optimum slicing level in the presence of delay distortion becomes \nfor conditions as discussed in Section 2.9,\n\nThe latter expression also applies for the difference between the slicing \nlevel and the maximum amplitude of a pulse train at a sampling point \nin the presence of a space.\n\nExpression (141) shows that the minimum amplitude at a sampling \npoint is smaller than in the absence of delay distortion (ao 0) by \nthe factor\n\nTABLE XIV FACTOR min FOR RAISED CosiINE SPECTRUM \nAND QuapRatTic DELAY DisTorTION\n\nIn the particular case of a raised cosine spectrum of the pulses at the \ndetector input, as shown in Fig. 1, and quadratic delay distortion, the \nfunction Ro(t/T) = Ro(x + n) is given in Table XIX of the Appendix. \nThe phase distortion \u00a5(\u2014q@) in this case is given by\n\nwhere d/T is defined as in Section 4.5. \nIn Table XIV are given \u00a5(\u2014@) together with x\u00bb as obtained from \n132), ao (2) as given by (134) and yin as obtained from (142). These \nfactors are shown in Fig. 10, together with the corresponding factor for \nbinary double-sideband AM as obtained from Table IV.\n\nWhen both the pulse spectrum at the detector input and phase dis- \ntortion has even symmetry about the frequency Wy , the following rela- \ntions apply (see Appendix\n\nThe maximum amplitude of a single pulse in this case is at \u00a2 0). \nOptimum performance is obtained with sampling at \u00a2 = 0, in which case \ny in (126) and (127) is given by\n\nQUADRATURE DELAY \nDISTORTION: \n--\u2014\u2014FM \nAM WITH \nL- SYNCHRONOUS OR \nENVELOPE DETECTION\n\nd/T =dfyax \nFig. 10 \u2014+ Factor nmin for binary pulse transmission by FM and double-side\n\nband AM, for quadratic delay distortion as in Fig. 2 and linear delay distortion \nas in Fig. 4\n\nBo\u2019 = Bi\u2019 (0) = >. (\u20141)\"a(n)Qi\u2019(n). (150) \nn=--2 \nFor the special case of a raised cosine spectrum and linear delay dis- \ntortion, the functions Ry and Q\u00bb are given in Table XX of the Appendix. \nThe functions Ro\u2019(n) and Qo\u2019(n) are related to the functions R; and Q, \ngiven in Table XXII by\n\nThe functions Ry and Qp are given in Table XV for integral values of \nt/T and the functions Po\u2019 /@ and Qo\u2019 /a@ are given in Table XVI. \nIn the above case of quadratic phase distortion of the form y(u)\n\nOwing to the several quantities ao, Bo, ao\u2019, Bo\u2019 , Bo\u2019ao, ao\u2019 Bo and Yo \ninvolved in (126), it does not appear feasible to derive simple relations \nfor Umax and U yin\u201d . However, it is possible to determine these by \nexamining several cases, as illustrated below ford/7T = landd/T = 2.\n\nTaBLeE XVII VALUES OF U(0O) ror RAtseEpD CosINE SPECTRUM \nAND LINEAR Deuay Distortion witH d/T = 1\n\nFor various combinations of marks and spaces, i.e., a(n) | and 0, \nthe results given in Table XVII are obtained.\n\nThe results in this case are given in Table XVIII. In this case U, \n0.57, Umin\u00ae = 0.42, Lo & 0.5 and qmin = Umin\u201d \u2014 Umax\u201d & \u20140.15.\n\nTaBLeE XVIII Vatues or U(O) ror Ratsep Cosine SPECTRUM \nAND LINEAR DeLay DisrorTion witu d/T' 2\n\nThe factors in Table XVIII are shown in Fig. 10, together with the \ncorresponding factors for binary double-sideband AM with synchronous \ndetection as given in Table V and with envelope detection as given in\n\nThe shape of pulse trains at the detector input and output in pulse \ntransmission by various methods of carrier modulation and detection \nhas been formulated in terms of a basic function common to all modula- \ntion methods: the carrier pulse transmission characteristic. This fune- \ntion is related to the amplitude and phase characteristics of the channel \nby a Fourier integral, which can be evaluated by numerical integration \nwith the aid of digital computers for any prescribed channel characteris- \ntic. In this way can be determined the effect of specified channel gain and \nphase deviations on the demodulated pulse train for any modulation \nmethod, together with the resultant maximum transmission impairment.\n\nThe carrier pulse transmission characteristics are given herein for the \nrepresentative case of pulses with a raised cosine spectrum at the detector \ninput, for two cases of envelope delay distortion over the channel band. \nIn one case delay distortion is assumed to vary linearly with frequency, \nand in the other case to vary as the second power of frequency from mid- \nband, as indicated in Fig. 11. The resultant maximum effect on the de- \nmodulated pulse trains at sampling instants has been determined for \nvarious carrier modulation and detection methods, together with the \ncorresponding maximum transmission impairment. The maximum trans- \nmission impairment is expressed as the maximum increase in signal-to- \nnoise ratio required at the detector input to compensate for the effect of \nphase distortion, or corresponding envelope delay distortion. The maxi- \nmum transmission impairments specified here apply as the error proba- \nbility approaches zero, and actual impairment will be somewhat smaller, \ndepending on error probability.\n\nIn evaluating the effect of phase distortion, idealized modulation and\n\nin other respects, such as instantaneous sampling of the appropriate \ninstants and optimum slicing levels.\n\nThe numerical results are given in various tables and curves, summa- \nrized in Fig. 12 and discussed briefly below.\n\nIn the expressions for the carrier pulse transmission characteristic the \nphase characteristic of the channel is a basie function. \u2018Transmission \nimpairments from phase distortion could be expressed in terms of some \nparameter or set of parameters that would define the type of phase dis- \ntortion under consideration. Alternatively, any type of phase distortion\n\nthat is, in terms of envelope delay distortion. From the standpoint of \nengineering applications the latter method is preferable, since variation \nin transmission delay over the channel band is more readily measured \nthan variation in phase, and it is ordinarily the quantity specified for \nvarious existing facilities.\n\nd = MAXIMUM DELAY DISTORTION OVER BAND fyax (CPS) \nT = PULSE INTERVAL IN DOUBLE SIDEBAND AM AND FM(SEC) \nT =1/fuax d/T=dfwsax\n\nFig. 11 Pulse spectrum at detector input and types of delay distortion as \nsumed in comparison of modulation methods.\n\nVESTIGIAL ANDO \nQUADRATURE \nDOUBLE - SIDEBAND \nAM WITH SYNCHRO- \nNOUS DETECTION \nAND L=2,3 \nPULSE AMPLITUDES\n\nAND DIFFERENTIAL \nPHASE DETECTION \n=\u2014=-=\u2014 SYNCHRONOUS \nDETECTION \n\u2014\u2014 DIFFERENTIAL \nPHASE \nDETECTION\n\nods for raised cosine pulse spectrum with linear and quadratic delay distortion\n\nLinear, quadratic or any other analytically specified delay distortion \ncan be expressed in terms of the difference in transmission delay between\n\nany two reference frequencies in the channel band. In the present analysis\n\nthe difference d in delay between the midband frequency and the maxi- \nmum frequency fmax from midband, as in Fig. 11, has been taken as a\n\nbasic parameter. The maximum transmission impairments with various \ncarrier modulation methods have been given in terms of the ratio d/T\u2019 = \nAfinax , Where T is the pulse interval in double-sideband AM.\n\ndifference dinax in transmission delay between any two frequencies in the \nchannel band. In the case of linear delay distortion dmax = 2d, while in \nthe case of quadratic delay distortion dnax = d, where d is defined in \nFig. 11. A third choice might have been the difference in delay d be- \ntween the midband frequency and the mean sideband frequency 4}fmax , \nin which case d = d/2 for linear and d = d/4 for quadratic delay dis- \ntortion.\n\nIt will be recognized that translation from one basic delay parameter \nto another can readily be made. Also, the question of whether linear or \nquadratic delay distortion causes greater transmission impairments will \ndepend significantly on the choice of transmission delay parameters.\n\nMaximum transmission impairments are shown in Fig. 12 for systems \nemploying / = 2, 3, 4 and 5 pulse amplitudes and ideal synchronous detec- \ntion. With envelope detection the transmission impairments are the same \nas with synchronous detection, for quadratic delay distortion and for \nany type of delay distortion with even symmetry about the channel mid- \nband (carrier) frequency. However, with envelope detection greater\n\ntransmission impairments are incurred in the case of linear delay dis- \ntortion, and for any type of delay distortion with odd symmetry about \nthe channel midband frequency. The difference between envelope and\n\nsynchronous detection in the presence of linear delay distortion is illus- \ntrated in Fig. 12 for 7 = 2 pulse amplitudes.\n\nAs noted previously, the maximum transmission impairments indi- \ncated in Fig. 12 would be encountered for extremely small error probabili- \nties. For error probabilities in the range normally considered, the maxi- \nmum impairments given in Fig. 12 would be rather closely approached \nwhen the impairments are fairly small, say less than 3 db. However, \nwhen the maximum impairments are rather high the actual impairments \nmay be significantly smaller. For example, with a maximum impairment \nof 10 db, the actual impairment would be expected to be about 1.5 db \nless for an error probability 10 \u00b0 and about 2 db less for an error proba- \nbility 10\u201c.\n\nVestigial-sideband AM and quadrature double-sideband AM _ with \nsynchronous detection are equivalent methods as regards channel band- \nwidth requirements and signal-to-noise ratios, in the absence of delay \ndistortion. Both methods may be used in preference to double-sideband \nAM either (a) to realize a two-fold increase in pulse transmission rate for\n\na given bandwidth in exchange for a 3 db penalty in signal-to-noise ratio \nor (b) to secure a two-fold reduction in bandwidth for a given pulse \ntransmission rate, without a penalty in signal-to-noise ratio.\n\nThe maximum transmission impairments shown in Fig. 12 are for the \nsame bandwidths as in double-sideband AM with a two-fold increase in \nthe pulse transmission rate. In this case transmission impairments from \nquadratic delay distortion are no greater than in double-sideband AM, \nand this applies for any type of delay distortion with even symmetry \nabout the channel midband frequency.\n\nWith linear delay distortion, or any delay distortion with odd sym- \nmetry about the channel midband frequency, transmission impairments \nare not identically the same for vestigial-sideband AM and quadrature \ndouble-sideband AM. However, the difference is not significant in the \nease of linear delay distortion, as indicated in Fig. 12. For practical \npurposes the two methods can be regarded as equivalent for any type of \ndelay distortion actually expected, as regards channel bandwidth require- \nments and signal-to-noise ratios for a given error probability, assuming \nideal synchronous detection.\n\nWith linear delay distortion the transmission impairments with the \nabove two methods are significantly greater than for double-sideband \nAM as indicated by comparison of the curves in Fig. 12 for the two meth- \nods for / = 2 and 3 pulse amplitudes. This assumes that the pulse trans- \nmission rate is twice as great as in double-sideband AM.\n\nWhen the pulse transmission rate is the same as in double sideband \nAM but the bandwidth is halved, delay distortion over the channel band \nis reduced. In this case vestigial-sideband AM or quadrature double- \nsideband AM affords an advantage over double-sideband AM in the \npresence of delay distortion with even symmetry about the channel mid- \nband frequency, but not necessarily when delay distortion has odd sym- \nmetry. With linear delay distortion the ratio d/T is halved, and in this \ncase there is a slight disadvantage compared to double-sideband AM, \nfor 1 = 2 pulse amplitudes. However, with the type of delay distortion \nordinarily encountered vestigial-sideband AM and quadrature double- \nsideband AM would afford some advantage in signal-to-noise ratio over \ndouble-sideband AM for equal pulse transmission rates and with ideal \nsynchronous detection.\n\nband AM with equal amplitudes but opposite polarities of the trans- \nmitted pulses. The curves in Fig. 12 for double-sideband AM and / = 2\n\npulse amplitudes apply also for two-phase transmission, for the reason \nthat the transmission impairments for a given peak-to-peak difference \nbetween pulse amplitudes is the same regardless of polarities.\n\nTwo-phase modulation can also be used in conjunction with vestigial- \nsideband transmission. The curves in Fig. 12 for vestigial-sideband AM \nand 1 = 2 pulse amplitudes also apply for two-phase vestigial-sideband \nmodulation.\n\nlour-phase modulation is equivalent to bipolar AM on each of two \ncarriers at quadrature with each other. The curves in Fig. 12 for quadra- \nture double sideband AM and / = 2 pulse amplitudes also apply for the \nspecial case of four-phase modulation.\n\nThe maximum transmission impairments with double-sideband two- \nphase and four-phase modulation and synchronous detection are shown \nseparately in Vig. 12 for comparison with PM with differential phase \ndetection.\n\nIn phase modulation systems differential phase modulation (described \nin Section 4.1) may be used in place of synchronous detection. Differen- \ntial phase detection has been implemented in various ways, which in gen- \neral involve some transmission impairments from channel bandwidth \nlimitations, even with a linear phase characteristic. Such transmission \nimpairments from channel bandwidth limitation is avoided with the \nimplementation assumed herein (Section 4.1), and only the effect of \nphase distortion is evaluated. Transmission impairments from delay dis- \ntortion will be greater with this method than with synchronous detection, \nas illustrated in Fig. 12 for double-sideband two-phase and four-phase \nquadrature systems and delay distortion. Transmission impairments from \nlinear delay distortion have not been determined for this case.\n\nrequires the same bandwidth for a given pulse transmission rate as binary \ndouble-sideband AM. Maximum transmission impairments with these \ntwo methods are compared in Fig. 12. It will be noted that with quad- \nratic delay distortion the impairments are smaller with FM than with \nAM employing either envelope or synchronous detection. In the case of \nlinear delay distortion, the transmission impairments are greater with \nFM than with synchronous AM, but are somewhat smaller than with \nAM employing envelope detection.\n\nThe transmission impairments given in Fig. 12 for kM apply without \na postdetection low-pass filter for noise reduction, and may involve \nsomewhat greater approximations than for the other modulation meth- \nods. Approximately the same impairments from phase distortion would \nbe expected with an appropriate low-pass filter.\n\nSignal-to-noise ratios at the detector input for a given error probability \nand various methods of carrier modulation are ordinarily compared on \nthe premise of ideal amplitude versus frequency characteristics of the \nchannels, and a linear phase characteristic. The curves in Fig. 12 indicate \nthat transmission impairments resulting from phase distortion depend \nsignificantly on the carrier modulation method. The optimum method as \nregards signal-to-noise ratio will thus depend on the type and degree of \nphase distortion encountered in a particular application. For example, \ntwo-phase modulation with synchronous or with differential phase detec- \ntion may have a slight advantage in signal-to-noise ratio over binary \nfrequency shift keying in the absence of delay distortion. However, the \nadvantage in signal-to-noise ratio would be expected to be with fre- \nquency shift keying in application to channels with pronounced quad- \nratie delay distortion or other types of delay distortion with essentially \neven symmetry about the carrier frequency.\n\nIn comparing the performance of various methods of carrier modula- \ntion it is necessary to consider other factors than signal-to-noise ratios \nand channel bandwidth requirements as discussed here. Among them \ncan be mentioned the adverse effects of sudden or gradual level and phase \nvariations and the complexity of instrumentation.\n\nThe writer is indebted to A. P. Stamboulis for pointing out some errors \nin the original equation (50) and for showing the presence of the third \nterm in (56) and to C. IF. Pease for numerical evaluation of integrals in \nthe Appendix with the aid of a 704 digital computer.\n\nof the carrier pulse transmission characteristics for any carrier frequency\n\ncan be determined from those for any other carrier frequency w\u00bb , for \nexample the midband frequency of the channel. Basic Fourier integrals \nare given here for the carrier pulse transmission characteristics for a \nreference or carrier frequency wo . In addition, special integrals are given, \napplying for a raised cosine pulse spectrum with linear delay distortion, \nquadratic delay distortion and the type of delay distortion introduced \nby flat bandpass filters with sharp cutoffs. For these three cases the \ncarrier pulse transmission characteristics have been determined by nu- \nmerical integration and are tabulated here.\n\nThe shape of Ro(t) and Qo(t) depends on the shape of the transmitted \ncarrier pulse and on the transmission-frequency characteristic of the \nchannel. If the carrier pulse is assumed of sufficiently short duration, the \nspectrum will be essentially flat over the channel band, so that the shape \nof the received spectrum is the same as that of the amplitud\u00e9 character- \nistic of the channel. The functions Ry and Qo are then obtained from ex- \npression given elsewhere (Ref. 2, Section 2) in terms of the amplitude \ncharacteristic A(u) of the channel, where u is the frequency measured \nfrom the carrier frequency wo , as indicated in Fig. 13. In the more general \ncase of carrier pulses of any shape and any channel transmission-fre-\n\nquency characteristic, the functions Ry and Qo are obtained by replacing \nin the above expressions A(uw) with the spectrum So(u) of the pulse\n\nAMPLITUDE \nCHARACTERISTIC OF \nSPECTRUM AT CHANNEL CARRIER FREQUENCY \nOUTPUT ~-\n\nFig. 13 \u2014 Amplitude characteristic So(u) and phase characteristic Wo(u) of \npulse spectrum at channel output (i.e., detector input) for carrier at frequency w\n\nThe various quantities in the above expressions are as shown in Fig. 13. \nIt will be recognized that the upper limit w in (160) and (162) can for \npractical purposes be replaced by ~, since So( \u2014wo) & 0.\n\n4.2 Even Symmetry Spectrum and Delay Distortion \nLet the spectrum at the detector input have even symmetry about \nwy and the phase distortion odd symmetry, in which case \nSof \u2014 a) So{u), (164) \nVo(\u2014u) = \u2014WVol(u). (165)\n\nWith (164) and (165) in (159) through (163), the following rela- \ntions are obtained when the upper limit wo is replaced by \u00ab :\n\nWhen the phase characteristic has a component with even symmetry \nabout the frequency wo , so that\n\nlor reasons discussed elsewhere (Ref. 2, Section 5) it is desirable in \npulse systems to employ raised cosine pulse spectra, as shown in Fig. 1 \nand given by \n\u2018yy\n\nwhere @ is the mean frequency from midband. \nThe corresponding carrier pulse transmission characteristic obtained \nfrom (159) through (163) with Wo(w) = 0 is\n\nPulses can in this case be transmitted without intersymbol interference \nat intervals 7\u2019 such that\n\nIt will be assumed that the phase characteristic contains a linear com- \nponent, which can be disregarded, and a distortion component given by\n\nwhere \u00a2 is a constant. The corresponding delay distortion is then quad- \nratic or parabolic, as given by\n\nNOMLMOLSI(] AVIAG] OLLVuaVvaA\u2019d aN \nWIUMLOAIS ASTAG ANISOY) GASIVY, UOA CL /))\"Yy ANV (f/))\u00b0Y SNOLLONOG XIX Wavy,\n\nwhere \n16 d \nb=\u2014-. (177) \n32\u00b0 7 \nThe ratio \u00a2/T is the time measured in pulse intervals and the ratio d/T \nthe maximum delay distortion measured in pulse intervals, with d de- \nfined as in Fig. 2 or Fig. 11. \nIn certain cases, as in connection with pulse transmission by frequency \nmodulation, the time derivative of Ro(t) is involved. This derivative is \ngiven by\n\nThe functions Ro(t) and R,(t) obtained by numerical integration of \n(176) and (179) are given in Table XIX. The function R)(\u00a2/T) is\n\nshown in Fig. 2. \n\\.6 Linear Delay Distortion \nIt will be assumed that the phase distortion component is given by \nWola) cu, (180) \nwhich corresponds to a linear delay distortion given by \nWo\u2019 (ua) Pou. \nIn this case expressions (169) and (170) give\n\nNOMLUOISI(] AVIA(] UVANITT ANY \nWOAUNLOAdS ASTAG ANISOD) aasivy woa (f/7)\u00b0%) axv (f/2)\u00b0Y SNOIONA A XX Wavy\n\nThe values of Rp and Qo obtained by numerical integration of (182) \nand (183) are given in Table XX.\n\nIt will be noted that Qo(0) # 0. From the standpoint of analysis, it \nmay be convenient to modify the phase such that Qo(0) = 0. The modi- \nfied values are given by\n\nThe modified values are given in Table XXI. The functions Ro(t/T) \nand Qo(t/7) are shown in Fig. 4.\n\nThe time derivatives of Ro(t) and Qo(\u00a2) are of interest in connection \nwith frequency modulation and given by\n\nThe functions R; and Q; obtained by numerical integration are given \nin Table XXII.\n\nNOLLUOLSI(Q] AVIAC] UVANITT GNV \nWAULIdG ASTAG ANISOD aasIvyY Yoda (,f,/2)\u00b0) GNV (L/2)Y SNOMONOY \u2014 [XX AMV\n\n(LAI) *) \u201cC(L/'Y\u2014 = CL/I-)'e \nNOILYOLSI(] AWIAC] UVANI'T aNV \nWAULIddg ASTAG ANISOD aasivy wod (7/7)'7) ANV (7/2) SNOMLONO GT WXX OMavVy\n\nLet a bandpass filter have an amplitude characteristic Ay between \n\u2014w. + wo and wo + w, and A, outside this band. When the bandwidth \n2w, is small in relation to the midband frequency wo , the phase charac- \nteristic is closely approximated by\n\nThe corresponding envelope delay distortion is D(w) = dyo(u)/du and \ndelay distortion relative to the midband frequency becomes\n\nIt will be noted that the first term in (196) represents quadratic delay \ndistortion, which is approximated for u/w, < 1.\n\nLet the pulse spectrum at the detector input have a raised cosine \nshape, as given by (171), in which case the maximum radian frequency \nto each side of midband is 2. With a phase characteristic as given by \n(193), the carrier pulse transmission characteristic is in this case ob- \ntained with (171) and (193) in (166) and becomes\n\nCarrier pulse transmission characteristics for raised cosine pulse \nspectrum and phase distortion resulting from flat filters with sharp cutoffs. \nTaste XXIII\u2014 Function R,(t/T) For Ratsep CosingE SpectruM \nAND PHASE Distortion RESULTING FROM FLAT FILTERS \nWITH SHARP CUTOFFS \nk = W/W, 1.05 \nAo/Ai\n\n. Sunde, E. D., Ideal Binary Pulse Transmission by AM and FM, B.S.T..J.. \n38, 1959, p. 1357. \n2. Sunde, E. D., Theoretical Fundamentals of Pulse Transmission, B.S.T.J., \n33, 1954, pp. 721; 987. \n3. Gibby, R. A., An Evaluation of AM Data System Performance by Computer \nSimulation, B.8S.T.J., 39, 1960, p. 675. \nFowler, A. D. and Gibby, R. A., Assessment of Effects of Delay Distortion in \nData Systems, Comm. & Elect., no. 49, 1959, p. 918. \nCostas, J. P., Synchronous Communications, Proc. I.R.E., 44, 1956, p. 1713. \n). Rieke, J. W. and Graham, R.S., The L-3 Coaxial System Television Termi \nnals, B.S.T.J., 32, 1953, p. 915 \nDarlington, 8., Realization of Constant Phase Difference, B.S.T.J., 29, \n1950, p. 94 \nCahn, C. R., Performance of Digital Phase Modulation Communication Sys- \ntems, Proc. I.R.E. Trans., CS-7, 1959, p. 3 \n. Cahn, C. R., Combined Digital Phase and Amplitude Modulation Communi \ncation Systems, I.R.E. Trans., CS-8, 1960, p. 150. \nMosier, R. R. and Clabaugh, R. G., Kineplex A Bandwidth Efficient Binary \nSystem, Comm. \u00ab Elect., no. 34, 1958, p. 723.\n\nFurther Results on the Detectability of \nKnown Signals in Gaussian Noise\n\nThe detection of a completely known signal which may or may not be \npresent in a finite sample of gaussian noise is considered from two points \nof view. The first examines the performance of a maximum likelihood de- \ntector operating on a finite set of discrete measurements of the stimulus as \nthe set becomes large. The stimulus is either signal plus noise or noise alone. \nExamples are presented for signals in bandlimited noise, using as measure- \nments either equispaced amplitude samples or derivatives at one instant in \ntime. For both, the detectability grows without bound as the number of meas- \nurements is increased. The second point of view bases detection on a con- \ntinuous measurement (linear integral operator) which maximizes the de- \ntectability. Solutions have been obtained when the noise has a rational power \nspectral density. The detector utilizes a cross-correlation between stimulus \nand signal which is well known and a mechanism, designated extrapolation \ndetection, which involves evaluation of derivatives of the stimulus. The con- \ntribution of the derivative measurements to the detectability is examined as \nthe noise approaches bandlimited noise and is found in many cases to \ngrow without bound.\n\nThe problem under consideration here is the detection of a completely \nknown signal which may or may not be present in a finite sample of \ngaussian noise. That is, we imagine a situation similar to Fig. 1 in which \na stimulus is made up of either signal plus noise or noise alone and we \nask, given 7\u2019 seconds of this stimulus, how accurately can we decide \nwhether or not the signal is present. The noise is thought of as having \nbeen produced by a stochastic process and thus the question is really \none of statistical hypothesis testing.\n\nThis particular problem has been treated rather extensively,\u20197 and \ncertain questions, even controversies, have arisen. These concern what\n\nconstitutes a proper description for the stimulus, under what circum- \nstances can the stimulus be characterized by a finite number of samples, \nand under what conditions is perfect detectability obtained, i.e., when \nis it always possible to detect the presence or absence of the signal. \nPeterson, Birdsall and Fox\u2019 have described the stimulus as being Fourier \nseries bandlimited and by so doing have obtained quite different results \nfrom the other authors, who for the most part consider stationary gauss- \nian noise. In many cases, finite-duration stimuli have been character- \nized by a finite number of samples usually chosen so they are independ- \nent, and maximum likelihood detectors operating on these samples have \nbeen developed. This has led to the equivalent of a correlation detection \nprocess in which the test statistic is the integral of the product of the \nstimulus and a function derived from the signal. Such detectors always\n\nout by an argument involving analytic continuation that many signals \ncan be perfectly separated from noise provided the noise is considered to \nhave a bandlimited spectrum. Clearly some mechanism in addition to \ncorrelation detection is inherent in Slepian\u2019s result, and indeed he points \nout one such detector.\n\nThe results of Peterson, Birdsall and Fox have been used extensively \nfor comparison with the performance achieved by humans and other \nanimals, and questions as to the validity of such comparisons originally \nmotivated this investigation. However, it seems very doubtful if the \nmechanisms which will be developed can have anything to do with per- \nception. In addition, we have chosen to work with stationary gaussian \nnoise rather than Fourier series bandlimited noise, the former being : \nmuch more satisfactory characterization of real noise.\n\nTwo different attempts to better understand the questions cited above \nhave been undertaken. The first examines the performance of a maxi- \nmum likelihood detector operating on a finite set of discrete measure- \nments of the stimulus as the set becomes very large. The results show \ncases where the detectability grows without bound. Thus, the charac-\n\nterization of the stimulus by a finite set of measurements is incomplete. \nHowever, in some cases, a law of diminishing returns operates so that \nthe rate of increase in detectability slows as the number of samples is \nincreased.\n\nThe second study bases detection on a continuous measurement \n(linear integral operator), which is the solution of an optimizing integral \nequation. The test statistic so obtained has two parts, one similar to \ncorrelation detection, the other based on measurements of the deriva- \ntives of the stimulus. The contribution of this latter term is usually the \nsmaller of the two, but, where the noise spectrum approaches a band- \nlimited form, it may grow without bound. In addition, it may be im- \nportant if the stimulus is very short.\n\nBoth maximum likelihood detection with a finite number of samples \nand the integral equation for the continuous statistic have been pre- \nviously presented. The new contributions arise from the more complete \nsolutions which have been obtained. The most significant result is un- \ndoubtedly the solution of the integral equation in closed form so that \nits characteristics and particularly its asymptotic properties for many- \npole noise can be seen. The derivative detector, which will be termed \nextrapolation detection, was apparent from this solution.\n\nIn this section we will derive the maximum likelihood detector for de- \ntecting a known signal in gaussian noise from a finite number of samples \nof the stimulus and apply this detector to two specific problems involv- \ning bandlimited noise. Each sample results from some linear operation \non the stimulus and the samples need not be independent. The deriva- \ntion of the detection equation differs only slightly from previously pub-\n\nwhich are the principal new results. In the problems the behavior of \nthe detector is studied as the number of samples becomes large, first \nwhen the samples consist simply of amplitude measurements of the \nstimulus and second when the samples are a set of derivatives at one \npoint in time.\n\n426 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1961 \nis either a gaussian noise N(\u00a2) or that noise plus a known signal S(t) \nand is observed for the interval 0 Ss \u00a2 < T. The n samples,\n\non which the detection is made are obtained by n linear operations \nL, , Le, -++,L, on the stimulus\n\nand N; will be gaussian random variables which may be completely \ncharacterized by their matrix 8 of correlation coefficients,\n\nBi; = EXN.N>, \nand by their means which for simplicity will be assumed to be zero, \nE<N > = 0.\n\nThe density function of the Y; samples when the stimulus is noise \nalone may then be written\n\nwhere ,\u00b0\u00b0*,Y\u00a5, are the dummy arguments of the density function \ncorresponding to Y,,---,Y, and | 8| is the determinant of 8, with all \nsums going over the range | to n unless especially indicated otherwise. \nThe density function of Y, for signal plus noise is simply\n\nA maximum likelihood detector says that signal is present if test \nstatistic L(Y,,---,Y,) is greater than some threshold a and will maxi- \nmize the conditional probability of detecting a signal when it is present \nfor a given conditional probability of indicating signal for noise alone.\n\nlor signal plus noise gsy is also gaussian with the same variance but \nwith mean \nExgsy> = >> Bij 'SiS;.\n\nThe density functions for \u00a2 are pictured on Fig. 2. The effectiveness of \nthis detector as indicated by the signal detection probability at a given \nfalse alarm rate can be characterized by a single number d, which is the \nratio of the squared mean of the signal plus noise distribution to the \nvariance of either distribution. The larger d is, the more completely \nseparated are the distributions on lig. 2 and the higher will be the de- \ntection probability. This number d is then\n\nThis form is usually preferable for computations since it involves the \nsolution of n linear equations rather than the inversion of an n X n \nmatrix. In addition this form more closely resembles the integrals which \nwill appear when continuous statistics are considered.\n\nTo summarize, a statistic \u00a2 which operates on a set of n correlated \nsamples and which is equivalent to a maximum likelihood statistic has \nbeen developed. Signal is indicated if \u00a2 is greater than some threshold. \n\u00a2 is formed as a linear sum of the samples, it has a gaussian distribution, \nand it has the same variance for both noise alone and signal plus noise \nvases. The performance of the detector may be characterized by a single \nnumber d = [E<\u00a2gsy>]\u2019'/E<gy\u2019>, the larger the d, the better the perform- \nance.\n\nThe argument presented by Slepian\u2019 indicates that theoretically, be- \ncause of the analytic nature of the noise, a sinusoid can always be de- \ntected in spectral bandlimited noise. However, this result says nothing \nabout how fast the detectability increases with the complexity of the \ndetector. In this section an example is examined in which the stimulus \nis time sampled with n samples equally spaced over the interval 0 S \nt < T and detectability is computed as a function of n. In addition to \nthe general behavior of this function, it is of special interest to note \nwhether any peculiarities occur at n = 2WT (the Nyquist rate), W \nbeing the noise bandwidth, since this is the maximum number of in- \ndependent samples which may be formed. The correlation function of \nthe noise is \n_ sin 2*Wr \n\u2014 QeWr \u2019\n\nno = 2WT. \nUnfortunately, no analytic way for either inverting this matrix or solv- \ning (5) is known, hence the detectability was computed numerically. \nThis computation was carried out on an IBM 704 machine for a signal \nwith frequency centered in the noise band\n\nA being the amplitude and an,/4n being an arbitrary phase chosen for \ncomputational convenience. The normalized results of a solution of (5)\n\nthe number of samples n/n, and of the stimulus duration in terms of \nthe number of independent samples n, . The curves exhibit a knee, not \nat n = n, but for n a bit larger than n, . Detectability continues to in- \ncrease but the rate of increase becomes imperceptible. The curves are \nall carried out to a matrix of size 128 X 128, which is the limit of the \ncapacity of the computer program. Double precision arithmetic and a \nsufficient error analysis were used to insure the accuracy of the results. \nThe increase in detectability beyond n = n, is essentially equivalent to \nthat which would be obtained by increasing T to T + 2/W and sampling \nat the Nyquist rate. Heuristically we can say that, by adding extra \npoints inside the interval, it is quite easy to predict N (\u00a2) two independent \nsample times beyond each end of the interval, but very hard to predict \nfurther. In an unpublished proof Slepian has shown that the quadratic \nform for d given by (3) does become infinite for bandlimited noise as n \nbecomes infinite. However, the present example indicates it increases \nat an exceedingly slow rate. Clearly a statistic which improves more \nrapidly is desirable, and such is evaluated in the next section.\n\nThe solution for the optimum integral operator detector carried out \nin the next section produced a statistic involving derivatives of the \nstimulus. This result suggests trying derivatives for bandlimited noise, \nparticularly since all derivatives of a bandlimited stimulus exist. Con- \nsequently, the detectability achieved by n samples, which are the stim- \nulus and its n \u2014 1 derivatives evaluated at one point in time, is studied. \nThis quantity, as will be seen, has the pleasant characteristics of being \nanalytically rather than only numerically determinable and of increasing \nuniformly with n rather than exhibiting the knee curves of the time \nsamples. A curious property is that the duration of the stimulus is no\n\nFig. 3 Normalized results of a solution of (5) and (6), with d/A2as a function \nof number of samples n/n, and of stimulus duration in terms of number of inde \npendent samples n, \nlonger a factor in detectability since, theoretically at least, any number \nof derivatives can be measured from as short a sample as desired.\n\nThe correlation coefficient may be written \na+ Oe \n| ; ers ree \nBr = G(w)(\u2014jw)\u2019 (gw) da, \nor . a \n_ $ oa \nwhere G(w) is the power spectrum of the noise\n\nIf bandlimited noise with a flat spectrum from \u20141 to +1 rad/second \nand unit rms amplitude is selected, then (7) yields\n\n0) if r + sis odd. \nA solution for (5) and (6) with these coefficients can be effected, since \nthe determinants involved are reducible to a form with a solution attri- \nbuted to Cauchy. The answer can probably be written on a large enough \nsheet of paper for signals having simple derivatives such as sinusoids,\n\nfor n odd. \nThe asymptotic behavior of d for large m can be seen by substituting \nStirling\u2019s approximation \na! & V/2n exp |\u2014a + (log a)(a + 3)}\n\nEquations (8) and (9) exhibit the behavior of a statistic in which d \nincreases linearly with the number of samples, each sample being a \nderivative. A similar behavior will be shown for rational noise where \none term in the detectability depends linearly on the number of deriva- \ntives which exist and form part of the statistic. The bandlimited noise \ndiffers from the rational noise in that all its derivatives theoretically \nexist and the detectability can be made, at least theoretically, as good \nas desired by making m large enough. Obviously, in any practical case, \nthe number of derivatives which can be estimated is limited. In addition, \nthe characterization of the random process as gaussian undoubtedly \nfails for high enough derivatives.\n\nEquations (8) and (9) are derived only for a signal which is a con- \nstant. However, a similar dependence on m would probably occur for \nsinusoidal signals.\n\nThe prominence of derivatives as an effective statistic for both band- \nlimited and rational noise gives a possible indication why detectability \nbased on equally spaced time samples increases so slowly. These, being \nuniformly distributed, give poor estimates of derivatives. A more effec- \ntive distribution might well be n, independent samples spaced uniformly \nover the interval and the rest of the samples clustered as closely as pos- \nsible about two points at each end of the interval. Such arrangement is \nsuggested by statistics for the rational noise case.\n\nThe preceding section discussed the detection of a known signal in \nbandlimited noise using a finite number of samples of the stimulus as a \nstatistic. In this section we consider the detection of a known signal in \ngaussian noise using as the statistic a continuous measure of the stimu- \nlus over an interval 7 in length. The noise is now taken to have a ra-\n\nsented at the ratio of two polynomials in w. Such noise can be thought \nof as resulting from the passage of ideal white gaussian noise through a \nfinite linear lumped-element filter, although it need not actually have \nbeen produced in this way. For the purposes of the analysis, it is con- \nvenient to think of the situation as shown in Fig. 4. White gaussian \nnoise is passed through a filter whose transfer function is H(s), (Laplace \ntransform of its impulse response ) and to this may or may not be added \nthe known signal S(t). 7 seconds of the combination form the stimulus \nY(t). The problem is to decide from an examination of the stimulus \nwhether or not the signal was present.\n\nThe detection scheme in this case is essentially an extension of the \nfinite sampling procedure. One asks for that linear integral operator \nwhich will extract from the stimulus a statistic giving the maximum \ndetectability. Thus, the statistic is obtained from\n\n- \ng= [ Y(t)Z(t) dt, (10) \n\u201c0 \nwhere Z(t) is that function of time which maximizes the detectability. \nBecause the noise is gaussian of zero mean and the signal (when present ) \nis simply added to the noise, the statistic g\u00a2 again has a gaussian proba- \nbility density function whose mean value is zero or not zero according \nto the absence or presence of the signal and whose variance is the same \nwith or without the signal. Thus it is reasonable to again define the de- \ntectability measure d as\n\nd= (11) \nThe optimization problem is thus to find Z(t) which maximizes d or, \nthat which is equivalent, to find Z(t) which minimizes E(gw) while \nholding E(\u00a2sy) constant. This latter form is a straightforward calculus \nof variation problem and its solution, the details of which are omitted, \nleads to the following integral equation for Z(t):\n\nThe discussion up to this point has not required that the noise have a \nrational spectral density. Unfortunately, it does not appear possible to \nearry (13) any further without actually solving (12) for Z(t), and this \nhas only been done in certain special cases. In particular, if the noise \nspectral density is the reciprocal of a polynomial, the solution for (12)\n\nan be exhibited in some detail; and furthermore if the signal is a sine \nwave, an exponential, or a constant the detectability can be expressed \nin a surprisingly simple form.\n\nIf the noise has a spectral density G(w), \no \nG(w) Rir)e *\" dr, \n\u00b0 x \nwhich is rational and contains only poles (2N in number), it can be \nwritten in the form \n. l \nG(w | = (14) \nay \u2014 Aw + ayw? \u2014 2.00 aie doy\" \nSuch a noise could have been produced by passing white noise of unit \nspectral density through a filter whose transfer function H(s) has N \npoles, \nl | - \nH(s) = . (15)\n\nand the poles can be placed in evidence by writing the denominator \npolynomial P(s) as\n\nwhere the y\u2019s are (possibly ) complex numbers giving the pole locations \nand each has a negative real part. In terms of H(s), the spectral density \ncan be written \nG(w) H (jw)\n\n-, yw and one constant by , or even the magnitude and phase of the \ntransfer function //(s) for real frequencies. The particular set of param- \neters to be used will be chosen to simplify the final answer.\n\nOne characteristic of N-pole noise is that its first NV \u2014 1 derivatives \nexist, while the Nth and higher do not. Because of this it is clear that \na necessary condition for finite detectability of a signal S(t) is that its \nfirst N \u2014 1 derivatives be continuous in the interval 0 to 7. If this con- \ndition is not satisfied; that is, if among the N \u2014 1 derivatives of S(t) a \ndiscontinuity occurs, then the detectability is infinite. This is clearly \ntrue, because one could simply differentiate the stimulus enough times \nto produce a step function in the interval and this could always be found \nby measuring the change in the differentiated stimulus just before and \njust after the time of the step.\n\nUsing this N-pole noise, it is possible to exhibit explicit solutions to \n(12) and (13). Unfortunately, strictly speaking, (12) does not have a \nsolution unless S(t) and its derivatives up to order N \u2014 1 satisfy a \ncertain set of boundary conditions (boundaries at 0 and 7\u2019). If S(t) does \nnot satisfy this set of boundary conditions, and in general for an arbi- \ntrary signal it will not, then (12) has a formal solution if Z(t) includes \ndelta functions and their derivatives to order N \u2014 1 at the end points \nof the interval (approached from inside the interval). The details of \nthis argument are presented in Appendix B, where it is shown that the \nsolution to (12) is\n\nwhere the superscript (n) indicates n-fold differentiation with respect \nto time, and the a\u2019s and \u00a7\u2019s are given by\n\nAmong the several other ways of writing d, one which is convenient \nis the following (partly operator notation ):\n\nN\u20141 \nP;( ) = 2\u00bb Dyes\" \nk=i \nand p is the derivative operator d/dt. The derivatives of S(t) and U(t) \nat 0 and T are to be interpreted as the limit of the value of the deriva- \ntives approached from inside the interval. \nThe form of Z(t) in (17) is quite interesting. The first part contributes \na function of time which is similar to the conventional cross-correlation \nresult. One simply multiplies the stimulus by this function and integrates \nthe product. In the second part, the delta functions, when used with (10) \nto form the statistic, represent evaluating the stimulus and its first \nN \u2014 1 derivatives at the ends of the interval. The derivatives at the \nends give information about the stimulus outside the interval. Essen- \ntially they allow prediction or estimation of the stimulus outside the in- \nterval, and this information is to be added to that from straight cross- \ncorrelation. As N becomes larger the noise spectrum drops off faster at \nhigh frequencies and more derivatives of the stimulus are used (more\n\nderivatives of the noise exist ); effectively, the stimulus can be predicted\n\nfurther outside the interval. Usually, this will mean that the signal can \nbe detected better (see examples below ). \n3.2 Damped Sinusoidal Signal\n\nAs a particular example, consider the case in which the signal is a \ndamped sine wave of arbitrary phase,\n\nSince the detectability is of primary interest, specific values for the co-\n\nefficients of the delta functions will not be calculated. The details of the \ncalculations are carried out in Appendix C, where it is shown that\n\nWith given signal parameters and noise filter, specific values of detect- \nability can be calculated from this expression.\n\nAs the number of poles in the noise filter increases, P\u201d( \u2014a)/P*(a) 0, \nassuming the poles are bounded away from the imaginary axis and that \na > 0. In this case d becomes\n\nIf as the number of poles is increased the pc gain of the filter is kept \nconstant (or allowed to increase), then P\u2019(a) increases without bound. \nThis can be seen by thinking of P(a@) in factored form, which for con- \nstant pe gain looks like\n\nand noting that | (a \u2014 y;)/y:| > 1. Thus, for fixed signal, more poles \nmean more detectability. A similar result obtains if a < 0.\n\nA noise filter of particular interest is a Butterworth filter, that is, one \nwhose poles are uniformly distributed on a semicircle in the left-half \nplane. Such a filter gives noise whose spectrum is maximally flat low- \npass and approaches ideal bandlimited noise as the number of poles \nincreases. In this case, the approximate behavior of d for large N can\n\nbe calculated by taking the poles as smeared out on a semicircle of radius\n\nA sketch of B versus @/ wo is shown in Fig. 5. Clearly B is greater than \none and the detectability grows exponentially for large N.\n\nFor an undamped sine wave (a = 0), (22) can be put in a more con- \nvenient form by using the magnitude and phase of the noise filter trans- \nfer function, H(s), which can be written\n\nThe angle @(w) then is the phase lag of the noise filter, a function of \nfrequency. In these terms (22) becomes \na ee \ni= T + 20(w) \u2014\n\nIf w7\u2019 > 1, that is, if the time is long so that there are many cycles \nof the sine wave in the interval, then the last term in (25) can be neg- \nlected. In conventional circuit analysis, 6 is generally considered the \ntime delay of a network; thus, the detectability includes a term pro- \nportional to twice the time delay of the noise filter. Roughly, this says \nthat the derivatives at the ends of the interval allow extension of the \nstimulus a distance equal to the time delay outside each end.\n\nIt is clear that the 6 term grows without bound as the number of \npoles bounded away from the imaginary axis is increased. In the par- \nticular case of noise with a maximally flat spectrum {Butterworth H(s)], \nthis growth can be shown more explicitly. The contribution to 6 from a \nsingle pair of poles located at \u2014aye*\u201d is\n\nwo At + 1 + 2A? cos 28\u2019 wo \nTo add up the contributions from N poles on a semicircle would lead to \na rather complicated expression, but an approximation for large NV can \nbe obtained by imagining the poles smeared out on the semicircle, so \nthat the sum can be evaluated as an integral. Then\n\nThis shows clearly that, for large NV, 6 increases directly in proportion \nto N. As a sidelight, the proportionality constant, plotted in Fig. 6, is \nlarger if the signal frequency is near the band edge. The apparent in- \nfinity for @ = a is a mathematical fiction; it resulted from smearing the \npoles. For any finite N, 6 is finite; thus, the curve in Fig. 6 really should\n\nW/W, \nFig. 6 \u2014 Proportionality constant. \nbe rounded over at the peak. For signal frequencies outside of the noise \nband, the detectability becomes large simply because the 1/G(w) term \nmultiplying everything in (25) becomes large. Even straight cross-\n\nshows clearly that the detectability increases as the number of poles \nbounded away from the imaginary axis is increased. \nFor N-pole Butterworth noise of bandwidth wo, (27) becomes (ex-\n\nHere again the detectability grows directly in proportion to N for \nlarge N.\n\nWe have presented solutions to some problems involving detection of \nthe presence of known signals in gaussian noise. Thus, we are concerned \nwith what a statistician would term hypothesis testing. Two general \nclasses of detectors are studied, the first a maximum likelihood detector \noperating on a finite number of samples of the stimulus, the second an \noptimum integral operator treating the stimulus as a continuous fune-\n\ntions, which differ little from ones previously given, but rather in the \nspecific solutions to these equations.\n\nIn the finite sampling case, detectability of a sinusoid or constant in \nbandlimited noise is computed for the cases where the samples are \nequally spaced time samples spread over a finite duration and where the \nsamples are measurements of successive derivatives at one point in time. \nAs the number of samples increases, detectability increases without \nbound for both cases. However, for the time samples the rate of increase \nis very slow for a large number of samples while for derivatives the rate \nbecomes a linear function of the number of samples.\n\nFor optimum linear integral detection a general solution is presented \nfor arbitrary signals in noise with a rational all-pole spectrum. The solu- \ntion in closed form is sufficiently tractable so that the asymptotic be- \nhavior of certain simple signals can be evaluated as the number of poles \nin the noise becomes very large. \u2018The solution puts in evidence two differ- \nent detection mechanisms, one involving integration of the product of \nthe stimulus with a function derived from the signal, the other involving \nmeasurement of the derivatives of the stimulus. The first is denoted \ncorrelation detection, the second extrapolation detection. Usually, the \nterm arising from correlation detection is the more important. However, \nif the stimulus is very short or if the noise spectrum has a great number \nof poles, the extrapolation term may become relatively large. For signals \nsuch as a sinusoid it grows without bound as the number of poles in- \ncreases.\n\nWhat are the implications of these solutions on previous detection \nresults? Probably they have very little bearing on the perception prob-\n\nlem which engendered the study, since it seems unlikely that animal \nsense organs embody the mechanisms implied by the solutions or that \nthe characterization of exactly known signals in gaussian noise is appro- \npriate. Both the solutions and the character of the stimuli differ signifi- \ncantly from the Fourier series bandlimited case treated by Peterson, \nBirdsall and Fox. In particular, the extrapolation detection does not \nappear in their universe. Also, we feel that the characterization of the \nnoise as described by a correlation function is, to say the least, more \nsuited to the present style of engineering and, to say the most, a much \nmore satisfactory model of most detection situations.\n\nThe practical impact, if any, of the detectors developed here would \nseem to inhere in situations where short pieces of valuable signals must \nbe detected and a great quantity of computing equipment is available. \nSuch might be the case for some space communication problems.\n\nA number of unsolved problems arise directly from the work. For a \nfinite number of time samples of the stimulus, the optimum distribution \nin time of these samples is unknown. Spectra with zeros as well as poles \nhave not been treated with anything near the elegance of the pure pole \nsituation. Only very specific classes of signals have been studied. It \nwould be of interest to establish which signals give unbounded and which \ngive bounded detectability as the number of poles in the noise increases. \nFinally, only the case of signals known exactly has been examined. The \nfar more difficult area involving signals with random parameters is al- \nmost untouched so far as practical solutions are concerned.\n\nThe authors would like to express their appreciation for the advice \nand assistance of D. Slepian and J. L. Kelly, Jr., which greatly furthered \nthis work.\n\nIn the main body of the paper it was shown that, for samples which \nare derivatives, detectability in terms of d can be determined from (5) \nand (6)\n\nS; is the 7 \u2014 1 derivative of the signal evaluated at \u00a2 = 0 and 8,;, the \ncorrelation coefficient of the noise derivatives, is\n\nfor flat bandlimited noise with unit rms amplitude. \nMquation (5) may be written out in matrix form for odd n as\n\nThis equation may be simplified by separating into two equations and \nmultiplying by minus one in appropriate places to remove minus signs. \nTwo forms occur, one for even n, the other for odd n. For\u2018n odd,\n\nThe determinants of these matrices can be evaluated by applying a rule \nattributed to Cauchy. In general, the rule says that a determinant whose \nijth element is\n\nIn addition, all cofactors of the matrices are also of Cauchy form. \nHence, it is possible to invert the matrices by the method of cofactors \nand thus solve the equations. Such solutions are quite complex for arbi- \ntrary signals. However, an especially simple answer can be obtained for\n\nZ, may be determined by the well-known method for solving equations \nas the ratio of two determinants,\n\nAPPENDIX B \nIn this appendix we give a general solution to the integral equation\n\nthe great advantage of being an explicit function of R(t) and S(t) rather \nthan involving the solution of a set of linear equations. In addition, it \npossesses the aesthetically pleasing property of not involving analytic \ncontinuation of S(\u00a2) outside the interval 0 < \u00a2 S 7. The noise spectral\n\n, | \\ 9 \nG(w) Q(p) >} Ay). (33) \n| P) \\p=ie k=0 \nIf we think of Q(p) as an operator with p interpreted as d/dt we see \nthat\n\npm I \nQ(p)[R(t)] ()(p) a _\u2014 6(t), \nLx (jw) 2r \nwhere 6(\u00a2) is the Dirac delta function. Operating formally on both sides \nof (32) with Q(p) vields \nN\n\nZAt Q(p){S(t)| >, GxS*\u201d (t), cick. (3) \nThe subscript has been added to Z to indicate that this may be only part \nof the answer and the superscript (mn) indicates n-fold differentiation \nwith respect to time. If (32) had a Z(t) solution which was continuous, \nthen (35) would be that solution. But the fact that (35) is continuous \n(as it would be if S(t) and its derivatives were continuous) does not \nprove that it is the complete solution. In fact, one can readily verify that \n(35) is not the complete solution by inserting it back in (32) and seeing \nif (32) is satisfied. It turns out that (35) is indeed part of the answer, \nand the remaining part is found by just this process of inserting (35) \nback in (32) and finding what is missing. If we imagine for the moment \nthat S(t) is extended in some arbitrary way outside the interval (so \nthat it is Fourier transformable and the function and its derivatives go\n\nwhat is equivalent, by finding an exact differential expression for the \nintegrand. We first note that Q(p) can always be factored,\n\nN \nU(t) = P(\u2014p)|S(0)] > (-1)\u2018o S(t). \nThe exact differential that we need is obtained by clairvoyance.\n\nNow, since P(p) has the left-half plane zeros of Q( p), the Fourier trans- \nform of P(p)[R(t)| will have only right-half plane poles and thus\n\nThe third integral on the right of (36) is evaluated in a similar way, \nusing now (39) with U. replaced by U, and by by (\u20141)\"b. , and noting \nthat P(\u2014p)R(t) 0 for? < 0. In this way we get\n\nIt is interesting to note that (40) and (41) depend only on values of \nS(t) inside the interval 0 < \u00a2 S 7, so that the way in which S(?\u00a2) was\n\nextended outside the interval does not matter. To summarize this, we \nfind\n\nIt is now clear that, for Z, to be the complete solution to (32), the double \nsum in (42) must be zero for all \u00a2 in the interval. This is equivalent to \nthe following boundary conditions on S(t):\n\nIf the signal is such that these conditions are not satisfied, then (32) \ng ; \nhas a solution only if Z(\u00a2) includes delta functions and their derivatives,\n\nIf this is used in (32), the delta functions bring out FR and its derivatives \nevaluated at \u00a2 and \u00a2 \u2014 7\u2019, and the a\u2019s and \u00a7\u2019s can be directly identified\n\nThis can be put in a slightly different form which may be more con- \nvenient by again partially integrating. Using another exact differential \nobtained by clairvoyance, which is\n\nand observing that when this is inserted in (46) the terms evaluated at \nT cancel, we get\n\nIn this form the summation only involves derivatives at / 0, which \nin some cases simplifies the algebra of a solution. \nAPPENDIX C\n\nAs a particular example, we calculate the detectability d for the case \nin which the signal is an exponentially damped sine wave,\n\nand the asterisk denotes complex conjugate. Using this in the expression\n\nfor detectability (20) or (49), we find that the second term\u2014eall it \nda \u2014 becomes \nN\u20141 \nds = : Be S'(O){[((\u20141)'Pi(p)P(\u2014p) + Pil \u2014p)P(p)|S(t)} 0\n\nThree special cases are now considered, the pure exponential, the \npure sine wave, and a constant (pc) signal. For a pure exponential \nsignal, \\ \u2014 \u2014a where a is real in (55), giving\n\nlor a pure sine wave signal, \\ \u2014 jw. The second term in (55) requires \na little special treatment, but it is easily shown that\n\nNow, (jw) is simply the reciprocal of the transfer function of the noise \nfilter at the frequency w; that is,\n\nP( jw) = 1/H( jw) \u2014\u2014\u2014., \nV G(w) \nwhere @(w) is the phase lag of the noise filter. Using this expression,\n\n1. Grenander, U., Stochastic Processes and Statistical Inference, Arkiv for Mate \nmatik, 1, 1950, p. 195. \n2. Zadeh, L. A. and Ragazzini, J. R., An Extension of Wiener\u2019s Theory of Pre \ndiction, J. Appl. Phys., 21, 1950, p. 645 \n3. Reich, E. and Swerling, P., The Detection of a Sine Wave in Gaussian Noise, \nJ. Appl. Phys., 24, 1953, p. 289. \n. Davenport, W. B., Jr. and Root, W. L., /ntroduction to the Theory of Random \nSignals and Noise, MeGraw-Hill, New York, 1958. \n. Middleton, D., An Introduction to the Theory of Statistical Communication, \nMeGraw-Hill, New York, 1960 \n). Peterson, W. W., Birdsall, T. G. and Fox, W. C., The Theory of Signal De \ntectability, I.R.E. Trans., PGIT4, 1954, p. 171. \n. Slepian, D., Some Comments on the Detection of Gaussian Signals in Gaussian\n\nG. E. Schindler, Jr., was appointed editor of the Bell System Tech- \nnical Journal, effective January 1, 1961. Mr. Schindler studied chemi- \ncal engineering at the Carnegie Institute of Technology, received the \nbachelor of science degree from the University of Chicago, and re- \nceived the master of arts degree in English literature and languages \nfrom the University of Pittsburgh. After additional graduate work at \nthe University of Chicago, Mr. Schindler joined Bell Telephone Labo- \nratories in 1953. He was editor of the Bell Laboratories Record from \n1957 to 1959, and most recently was with the Public Relations depart- \nment of the A. T. & T. Co.\n\nA_ theoretical investigation has been undertaken to study diffraction of \nelectromagnetic waves tn Fabry-Perot interferometers when they are used as \nresonators in optical masers. An electronic digital computer was programmed \nto compute the electromagnetic field across the mirrors of the interferometer \nwhere an initially launched wave is reflected back and forth between the \nmirrors.\n\nIt was found that after many reflections a state is reached in which the \nrelative field distribution does not vary from transit to transit and the ampli- \ntude of the field decays at an exponential rate. This steady-state field dis- \ntribution is regarded as a normal mode of the interferometer. Many such \nnormal modes are possible depending upon the initial wave distribution. \nThe lowest-order mode, which has the lowest diffraction loss, has a high \nintensity at the middle of the mirror and rather low intensities at the edges. \nTherefore, the diffraction loss is much lower than would be predicted for a \nuniform plane wave. Curves for field distribution and diffraction loss are \ngiven for different mirror geometries and different modes.\n\nSince each mode has a characteristic loss and phase shift per transit, a \nuniform plane wave which can be resolved into many modes cannot, properly \nspeaking, be resonated in an interferometer. In the usual optical inter- \nferometers, the resolution is too poor to resolve the individual mode resonances \nand the uniform plane wave distribution may be maintained approximately. \nHowever, in an oscillating maser, the lowest-order mode should dominate \nif the mirror spacing ts correct for resonance.\n\nA confocal spherical system has also been investigated and the losses are \nshown to be orders of magnitude less than for plane mirrors.\n\n\u2018 rm 1 . - . \nSchawlow and Townes have proposed infrared and optical masers \nusing Fabry-Perot interferometers as resonators. Very recently, Mai-\n\nbility of stimulated optical radiation in ruby. In these experiments two \nparallel faces of the ruby sample were polished and silvered so as to \nform an interferometer. The radiation due to stimulated emission \nresonates in the interferometer and emerges from a partially silvered \nface as a coherent beam of light.\n\nIn a maser using an interferometer for a resonator, a wave leaving one \nmirror and traveling toward the other will be amplified as it travels \nthrough the active medium. At the same time it will lose some power due \nto scattering by inhomogeneities in the medium. When the wave arrives \nat the second mirror some power will be lost in reflection due to the \nfinite conductivity of the mirror and some power will be lost by radia- \ntion around the edges of the mirror. For oscillation to occur, the total \nloss in power due to density scattering, diffractive spillover and reflection \nloss must be less than the power gained by travel through the active \nmedium. Thus diffraction loss is expected te be an important factor, \nboth in determining the start-oscillation condition, and in determining \nthe distribution of energy in the interferometer during oscillation.\n\nWhile it is common practice to regard a Fabry-Perot interferometer as \nbeing simultaneously resonant for uniform plane waves traveling parallel \nto the axis and at certain discrete angles from the axis, this picture is \nnot adequate for the computation of diffraction loss in a maser. It is true \nthat, when the interferometer is operated as a passive instrument with \nuniform plane waves continuously supplied from an external source, the \ninternal fields may be essentially those of uniform plane waves. In an \noscillating maser where power is supplied only from within the inter- \nferometer, the recurring loss of power from the edges of a wave due to \ndiffraction causes a marked departure from uniform amplitude and phase \nacross the mirror.\n\nThe purpose of our study is to investigate the effects of diffraction on \nthe electromagnetic field in a Fabry-Perot interferometer in free space. \nThe conclusions can be applied equally well to gaseous or solid state \nmasers provided the interferometer is immersed in the active medium, \ni.e., there are no side-wall discontinuities.\n\nOur approach is to consider a propagating wave which is reflected back \nand forth by two parallel plane mirrors, as shown in Fig. 1(a). [This is\n\nequivalent to the case of a transmission medium comprising a series of \ncollinear identical apertures cut into parallel and equally spaced black \n(perfectly absorbing) partitions of infinite extent, as in Fig. 1(b).}] We \nassume at first an arbitrary initial field distribution at the first mirror \nand proceed to compute the field produced at the second mirror as a \nresult of the first transit. The newly calculated field distribution is then \nused to compute the field produced at the first mirror as a result of the \nsecond transit. This computation is repeated over and over again for \nsubsequent successive transits. The questions we have in mind are: (a) \nwhether, after many transits, the relative field distribution approaches a \nsteady state; (b) whether, if a steady-state distribution results, there \nare any other steady-state solutions; and (\u00a2) what the losses associated \nwith these solutions would be. While it is by no means obvious that\n\ntem which has no side-wall boundaries, it will be shown that such solu- \ntions do indeed exist.*\n\nFig. 1 The Fabry-Perot interferometer and the transmission medium analog.\n\n* Schawlow and Townes! suggested the possibility that resonant modes for a \nparallel plate interferometer might be similar in form to those for a totally en \nclosed cavity\n\nWe shall use the scalar formulation of Huygens\u2019 principle to compute \nthe electromagnetic field at one of the mirrors in terms of an integral of \nthe field at the other. This is permissible if the dimensions of the mirror \nare large in terms of wavelength and if the field is very nearly transverse \nelectromagnetic and is uniformly polarized ir one direction. Later, we \nshall show that these assumptions are consistent with the results of our \nsolutions and therefore are justifiable. We shall also show that other \npolarization configurations can be constructed from the solutions of the \nscalar problem by linear superposition.\n\nwhere wu, is the aperture field, k is the propagation constant of the me- \ndium, F is the distance from a point on the aperture to the point of ob- \nservation and @ is the angle which R makes with the unit normal to the \naperture. We now assume that an initial wave of distribution u, is \nlaunched at one of the mirrors of the interferometer and is allowed to be \nreflected back and forth in the interferometer. After q transits the field \nat a mirror due to the reflected field at the other is simply given by (1) \nwith u, replaced by u,4: , Which is the field across the mirror under con- \nsideration and u, by u, , which is the reflected field across the opposite \nmirror giving rise to Ugii.\n\nIt is conceivable that after many transits the distribution of field at \nthe mirrors will undergo negligible change from reflection to reflection \nand will eventually settle down to a steady state. At this point the fields \nacross the mirrors become identical except for a complex constant;\n\nwhere v is a distribution function which does not vary from reflection to \nreflection and y is a complex constant independent of position coor- \ndinates. Substituting (2) in (1) we have the integral equation\n\nin which the kernel of the integral equation, K, is equal to (jk/4rR) \n(1 + cos @)e *\u201c. The distribution function v, which satisfies (3), can\n\nbe regarded as a normal mode of the interferometer defined at the mirror \nsurface, and the logarithm of y, which specifies the attenuation and the \nphase shift the wave suffers during each transit, can be regarded as the \npropagation constant associated with the normal mode.\n\nWe have studied and obtained numerical solutions for several geo- \nmetric configurations of the interferometer. These are (a) rectangular \nplane mirrors, (b) circular plane mirrors and (c) confocal spherical or \nparaboloidal mirrors.\n\nWhen the mirror separation is very much larger than the mirror di- \nmensions the problem of the rectangular mirrors reduces to a two- \ndimensional problem of infinite strip mirrors. This is shown in Appendix \nA. The integral equation for the problem of infinite strip mirrors, when \na\u2019/bd is much less than (b/a)\u2019, is\n\nquation (4) is a homogeneous linear integral equation of the second \nkind. Since the kerriel is continuous and symmetric [A (22, 21) \nK(x, , 22)|, its eigenfunctions v, corresponding to distinet eigenvalues \ny, are orthogonal in the interval (\u2014a,a); that is (Ref. 5, p. 413),\n\nIt should be noted that the eigenfunctions are in general complex and are \ndefined over the surface of the mirrors only. They are not orthogonal in \nthe power (Hermitian) sense as commonly encountered in lossless systems. \nHere, the system is basically a lossy one and the orthogonality relation is\n\none which is generally applicable to lossy systems, such as lossy-wall \nwaveguides.\n\nThe eigenfunctions are distribution functions of the field over mirror \nsurfaces and represent the various normal modes of the system. The nor- \nmal modes for rectangular plane mirrors are obtained by taking the \nproducts of the normal modes for infinite strip mirrors in \u00ab and y direc- \ntions; that is,\n\nWe designate this as the TEM,,,,, mode for the rectangular plane-mirror \ninterferometer. In view of (5) we see that the normal mode distribution \nfunctions v,,, are orthogonal over the surface of the rectangular mirror. \nThe logarithms of the eigenvalues represent propagation constants \nassociated with the normal modes. The propagation constant for the\n\nThe real part of the propagation constant specifies the loss per transit\n\nand the imaginary part the phase shift per transit, in addition to the \ngeometrical phase shift, for the normal modes.\n\nfor circular plane mirrors (lig. 3) when a\u00b0/b\\ is much less than (b/a)\u00b0, \nare given by \nvirg~) = R,(r)e \u2019\"* (n integer ), \nwhere P,,(7) satisfies the reduced integral equation \n2a\n\nproblem of infinite strip mirrors, (9) is a homogeneous linear integral \nequation of the second kind with a continuous and symmetric kernel. \nIts eigenfunctions corresponding to distinct eigenvalues are orthogonal \nin the interval (0,a); that is,\n\nTherefore, we see that the distribution functions v,,,(7,\u00a2) = Rin(re \ncorresponding to distinct eigenvalues y,, are orthogonal over the sur- \nface of the mirror; that is,\n\nA number of geometries other than plane parallel mirrors have been \nsuggested, and it is believed that most of these can be studied using the \nsame iterative technique. One of the geometries we investigated is that\n\nhave identical curvatures and their foci are coincident, as shown in \nFig. 4. One of the possible advantages of such a system is the relative \nease of adjustment, since the mirrors are no longer required to be parallel \nas in the case of the parallel plane system. Another is that the focusing \naction of the mirrors might give rise to lower diffraction losses.\n\nA spherical mirror with a small curvature approximates closely a \nparaboloidal mirror. In the case of confocal spherical mirrors, the condi- \ntions that its curvature be small is equivalent to saying that the separa- \ntion between mirrors is large compared to the dimensions of the mirrors. \nIt is shown in Appendix C that the solutions to the integral equation \nfor confocal paraboloidal mirrors, when a\u2019/bd\\ is much less than (b/a)\u2019, \nare given by\n\nAgain, we see that (13) is a homogeneous linear integral equation of \nthe second kind with a continuous and symmetric kernel. Therefore, \ngeneral remarks concerning the normal modes of circular plane mirrors \ngiven in the foregoing section are also applicable to confocal spherical \nor paraboloidal mirrors.\n\nAn IBM 704 computer was programmed to solve the integral equa- \ntions for the various geometries of the interferometer by the method of \nsuccessive approximations. As mentioned previously, this is analogous \nto the physical process of launching an initial distribution of wavefront \nin the interferometer and letting it bounce to and fro between the \nmirrors.\n\nwas employed for the computation, using an initial excitation of a uni- \nform plane wave at the first mirror. A total of one hundred increments \nwas used for the numerical integration. After the first transit the field \nintensity (electric or magnetic) had the amplitude and phase shown in \nFig. 5. In these and subsequent amplitude and phase distributions the \ncurves are normalized so that the maximum amplitude is unity, and the \nphase at that point is zero. The large ripples are due to the fact that the \ninitial wave front contains 6.25 Fresnel zones as seen from the center of \nthe second mirror. Therefore, in passing from the center to the edge of \nthe second mirror there is a change of 3 X 6.25 Fresnel zones, and this \nagrees with the number of reversals in curvature seen in the amplitude \ndistribution.\n\nWith subsequent transits, these ripples grow smaller, the amplitude \nat the edge of the mirror decreases, and the relative field distributions \napproach a steady state. By the time the wave had made three hundred \nbounces, the fluctuations occurring from bounce to bounce were less \nthan 0.03 per cent of the final average value. The amplitude and phase \nfor the 300th bounce are also shown in Fig. 5.\n\nWe regard this field distribution as an iterative normal mode of the \ninterferometer. In other words, if this distribution is introduced as an \ninitial wave at one mirror it will reproduce the same distribution at the \nother mirror. Indeed, this is what the computer is verifying when we \ncompute the 301st bounce.\n\nOnce the solutions have reached a steady state, we can pick any point \non the wavefront, say the center of the mirror, and examine how the \nabsolute phase and amplitude change from bounce to bounce. In this \nway we determined that the power loss of this mode is 0.688 per cent \nper transit and the phase shift per transit has a lead of 1.59 degrees.\n\nFig. 5 \u2014 Relative amplitude and phase distributions of field intensity for in \nfinite strip mirrors. (The initially launched wave has a uniform distribution.\n\nSince phase shift is measured relative to the free-space electrical length \nbetween the mirrors (360 b/ degrees), this means that the mode has an \neffective phase velocity which is slightly greater than the speed of light, \njust as for a metal tube waveguide.\n\nFig. 6 \u2014 Fluctuation of field amplitude at z = 0.5 as a function of number of \ntransits. (The initially launched wave has a uniform distribution.)\n\npoint (x = 0.5a) approaches its steady-state normalized value after a \nstart from a uniform plane wave. After the 100th transit the plot ap- \npears to be a damped sine wave. We interpret this damped oscillation \nas the beating between two normal modes having different. phase veloci- \nties. The mode with the lower attenuation, of course, survives the \nlongest, and this is the one shown in Fig. 5. We regard this as the dom- \ninant mode of the interferometer. We believe the other mode which \nbeats with the dominant mode to be the next-higher order, even-sym- \nmetric mode. Prior to the 100th transit, the curve is irregular, indicating \nthat a number of still higher order modes are present which are damped \nout rapidly.\n\nsmall compared to (b/a)*, the actual dimensions of the mirrors and their \nspacing are no longer important, the only parameter of importance being \nthe Fresnel number N a /b\\. This is approximately equal to the \nnumber of Fresnel zones seen in one mirror from the center of the other \nmirror, and as pointed out earlier, it determines the number of ripples \nin the field distributions. Amplitude and phase distributions for the\n\nRESONANT MODES IN A MASER INTERFEROMETER 465 \ndominant mode obtained by solving (27) are shown in Fig. 7 for differ- \nent values of N. The larger the NV, the weaker is the field intensity at the \nedge of the mirror, and the smaller is the power loss due to spill-over. \nThe plot of power loss per transit as a function of N is approximately a \nstraight line on log-log paper and is shown as the lowest line in Fig. 8. \nThe phase shift per transit as a function of N is given by the lowest \nline in Fig. 9.\n\nA uniform plane wave excitation can never give rise to a mode with \nodd symmetry. In order to investigate the possibility of modes of this\n\nFig. 7 \u2014 Relative amplitude and phase distributions of field intensity of the \nlowest order even-symmetric mode for infinite strip mirrors.\n\nFig. 8 \u2014 Power loss per transit vs. V = a?/bd for infinite strip and circular \nplane mirrors.\n\ntype, the problem was re-programmed for an initial wave for which the \nfield intensity over one-half the strip (0 to +a@) was equal but opposite \nin sign to the field intensity over the other half of the strip (0 to \u2014a). \nSteady-state solutions did indeed result, and odd-symmetric normal \nmodes therefore exist. The amplitude and phase distributions are shown \nin Fig. 10 for several values of NV. The amplitude is zero at the center, \nas expected. While shown for only one half of the strip, it is the same in \nthe other half, but with a reversal in sign. Note that for the same values \nof V, the amplitude at the edge is higher than for the dominant mode. \nThe spill-over loss should be higher and this is confirmed by the loss \ncurve in Fig. 8 labeled \u2018infinite strip odd-symmetric mode.\u201d\u2019 The cor- \nresponding phase shift curve is shown in Fig. 9.\n\nThe feasibility of obtaining the normal mode solutions for the infinite\n\ninvestigate the modes for plane circular mirrors. The first case considered \nwas that for uniform plane wave excitation of the system. Once again, \nthe polarization was assumed to be everywhere parallel to the same axis,\n\nwith n = O|. That is, the amplitude and phase of the field intensity is \nthe same for all points at the same radius from the center. The transverse \nfield distributions for the lowest order mode of this type are shown in \nFig. 11 for various values of NV. The loss and phase shift are shown in \nFigs. 8 and 9 under the title \u201ccircular dise (dominant mode, TEM).\u201d \nOne hundred increments along the radius were used for the numerical \nintegrations involved.\n\nNext we examined modes of the odd-symmetric type for circular \nplane mirrors. The equation we used was (9) with n = 1. Fig. 12 shows \namplitude and phase distributions for the lowest order mode of the odd- \nsymmetric type for circular plane mirrors. Again the loss and phase\n\nFig. 9 \u2014 Phase shift per transit (leading relative to geometrical phase shift) \nvs. N = a?/bd for infinite strip and circular plane mirrors.\n\nFig. 10 \u2014 Relative amplitude and phase distributions of field intensity of the \nlowest-order odd-symmetric mode for infinite strip mirrors.\n\nNormal modes with higher orders of angular variation (n 2 2) and \nradial variation (m 2 1) have greater losses and phase shifts than those \nof TEM and TEM, modes. The mode with the least attenuation is\n\ntherefore the lowest order, of TEMoo mode, which we designate as the \ndominant mode for circular plane mirrors. \n3.4 Confocal Spherical Mirrors\n\nBefore (13) was programmed for solutions on the computer a more \ngeneral method for solving the problem of the confocal spherical mirrors\n\nFig. 11 \u2014 Relative amplitude and phase distributions of field intensity of the \ndominant (TEMoo) mode for circular plane mirrors.\n\nFig. 12 telative amplitude and phase distributions of field intensity of the \nTEM)\u00bb mode for circular plane mirrors\n\nwas tried \u2014 a procedure that can be used to solve problems involving \nmirrors with rather arbitrary but small curvatures. In this method the \nfield at each mirror is caleulated using the equation for circular plane\n\nmirrors and then a phase distribution corresponding to the curvature of \nthe mirror is added to this field before it is used in the next iterative com- \nputation. The results from this general method of solution and from \nsolving (13) are in perfect agreement.\n\nThe problem of confocal spherical mirrors has also been solved by \nGoubau\u2019 and Boyd and Gordon.\u2019 The results of their analyses are in \ngood agreement with our computed results.\n\nAmplitude distributions of the field intensity for TEM) and TEM) \nmodes are shown in Figs. 13 and 14. The phase distributions are all \nuniform over the surface of the mirrors and therefore are not plotted. \nThe loss and phase shift per transit are given in Figs. 15 and 16. We \nnote some rather remarkable differences between these solutions and \nthose obtained for circular plane mirrors. First, the field is much more \ntightly concentrated near the axis of the reflector and falls to a much \nlower value at the edge than is true for plane mirrors; also the am- \nplitude distribution does not have ripples in it, but is smooth. Second,\n\nFig. 13 \u2014 Relative amplitude distribution of field intensity of the dominant \n(TEMoo) mode for confocal spherical mirrors. The relative phase distribution on \nthe surface of the mirror is uniform.\n\nFig. 14 \u2014 Relative amplitude distribution of field intensity of the TEMhio \nmode for confocal spherical mirrors. The relative phase distribution on the surface \nof the mirror is uniform.\n\nthe surface of the reflector coincides with the phase front of the wave, \nmaking it an equiphase surface. Third, the difference between the phase \nshifts for all the normal modes are integral multiples of 90 degrees. \nFourth, the losses may be orders of magnitude less than those for plane \nmirrors.\n\nThe result that the mirror surface is an equiphase surface should not \nbe surprising, but can be deduced from integral equation (13). If we\n\nMs \u00bb n+l . 7 . \nassociate the factor j\"\"\" with y, the kernel becomes real. Since the eigen-\n\nvalues and eigenfunctions of a real symmetric kernel are all real,\u2019 we \nsee that the field distribution is of uniform phase over the surface of the \nmirror. Furthermore, since (j\"*'y,) is real, the phase shift for the normal \nmodes belonging to a set of modes with a given angular variation must \nbe an integral multiple of 180 degrees and the difference between the \nphase shifts for the normal modes with different angular variations but \nthe same radial variation is an integral multiple of 90 degrees; that is, \nthe phase shift is equal to [180m + 90(n + 1)] degrees. Therefore, if \nthe mirrors are adjusted for the resonance of a particular normal mode,\n\nhalf of the totality of all the modes are also resonant. However, the \nresonant mode with the lowest loss would persist longest in the resonator. \nJust as in the case of plane parallel mirrors, the mode with the lowest loss \nis the TE Moo mode.\n\nThe results of machine computation have shown that a two-mirror \ninterferometer, whether of the plane or concave mirror type, can have \n100\n\nFig. 15 \u2014 Power loss \u2018tod transit vs. N = a?/bd for confocal spherical mirrors. \n(Dashed curves for circular plane mirrors are shown for comparison).\n\nFig. 16 \u2014 Phase shift per transit (leading relative to geometrical phase shift) \nvs. V = a?/b\\ for confocal spherical mirrors. (Dashed curves for circular plane \nmirrors are shown for comparison\n\nnormal modes of propagation which are self-perpetuating or self-repro- \nducing in the distance of one transit. We use the term mode of propaga- \ntion rather than mode of resonance to emphasize the fact that these \nsteady-state solutions are the result of multiple transits whether or not \nthe plate separation happens to be adjusted for resonance. An analog of \nthe plane mirror interferometer is a transmission medium consisting of a \nseries of periodic collinear apertures, as was shown in Fig. 1. The same\n\nsolutions apply, and here it is clear that the reproduction of a normal \nmode field at successive apertures does not depend on any critical rela- \ntion between b and X.\n\nIn Fig. 17 is shown the way in which a number of square-plate modes \ncan be synthesized from the infinite-strip modes. Diagram \u00aba shows \nschematically the field distribution for the dominant square-plate mode \nobtained as the product of the field distributions of two even-symmetric \nstrip modes crossed at right angles and with polarization as shown. \nSince the eigenvalue for the square plate is the product of the eigen- \nvalues for the two strips, the phase shift per transit is the sum of the\n\nphase shifts for the two strips and, if the loss is small, the loss per transit \nis essentially the sum of the losses for the two strips. Diagram B repre- \nsents an odd-symmetric square-plate mode formed by taking the prod- \nuct of an even- and an odd-symmetric strip mode; B\u2019 is the same mode \nbut with the polarization rotated 90\u00b0; c is a circular electric type of \nmode formed by adding two modes of the type B. This addition is per- \nmitted because the two components are degenerate. It follows that the \ncircular electric mode c is degenerate with B and has the same loss and \nphase shift per transit. By taking the difference between the same two \nB modes as shown, the mode pD is obtained, resembling the TE. mode in \ncircular waveguide. We give all the patterns B, B\u2019, c and p the same \ndesignation, TEM, (or TEM), since they are composites of the one \nbasic mode type. Similar syntheses can be performed for circular mirrors, \neither plane or concave. It is interesting that degeneracies of this type \nare common for the interferometer because the electric vector F is at \nliberty to be parallel or perpendicular to the mirror edges. In a metal \nwaveguide they are uncommon because the polarization of / at the \nboundaries is restricted.\n\nThe dominant mode and a number of higher-order modes for square \nand circular mirrors are depicted in Fig. 18, in which electric field vectors \nare shown. This classification of modes applies to plane as well as con- \nfocal spherical mirrors. In the case of rectangular mirrors, the x axis \nmay be taken along the longer dimension, in which case the first sub-\n\nscript always denotes the number of field reversals along the longer di- \nmension.\n\nIn formulating the problem we have assumed that the waves were \nalmost transverse electromagnetic. The solutions for the flat mirror are \nconsistent with this assumption. At the edges of the mirror there is a \nphase lag of approximately 45 degrees relative to the center, but this is \nonly one-eighth of a wavelength out of many wavelengths for the mirror \ndiameter. Thus the curvature of the wavefront away from the transverse \nplane is exceedingly small, and the assumption appears justified. For \nhigher-order modes such as Bb\u2019 of Fig. 17, it is clear that the field lines \nmust have longitudinal components. This is illustrated by an edge view \nin Fig. 19. However, provided the width of a cell c is much greater than a \nhalf-wavelength, the longitudinal field intensity should be negligible \ncompared to the transverse. Only for very high-order modes should this \napproximation begin to fail. Because the low-order modes of importance \nare essentially transverse electromagnetic, they are designated as TEM \nmodes.\n\nThe plane mirror modes have a phase which is not constant over the \nmirror. This does not mean that it is impossible to space the mirrors for \nresonance of the entire field pattern. Actually, the phase delay for one \ntransit is the same for every point on the wavefront. Therefore, if the \nplates are separated by the distance b plus an additional amount for the \nphase shift per transit of the mode desired, that mode should resonate in \n| \n|\n\nthe interferometer. Other modes should not be resonant for this separa- \ntion because they have different phase shifts per transit.\n\nSince the field configurations of many of the normal modes of the \ninterferometer are very similar to those of metal tube and parallel-plane \nwaveguides, it is not surprising to find that simple waveguide theory can \nbe used to predict certain characteristics of the interferometer modes. \nOne of these characteristics is phase shift per transit. For instance, the \nfield distributions of the normal modes for infinite strip mirrors are very \nsimilar to those of the TE modes of parallel-plane waveguide; also, by \nadding two orthogonally polarized TEM), modes for circular plane mir- \nrors, one obtains a field configuration which is very similar to that of the \ncircular electric (TE) mode of circular waveguide (Fig. 20). Thus the \namount of phase shift per transit computed for these modes of the \ninterferometer agrees well with the phase shifts obtained for TE modes \nof parallel-plane waveguide and TE\u00bb; mode of circular waveguide. This is \nillustrated in Fig. 21. We see that agreement becomes better for larger \nvalues of V. This is because the similarity between field configurations \nbecomes closer for larger values of NV.\n\nIf we regard a uniform plane wave as being resolvable into a set of \nnormal modes, there can be no such thing as a resonance for a uniform \nplane wave. Why then does it appear that there is such a resonance in \npassive optical interferometers? It is because for the usual optical case \na /bd is in the thousands. The phase shifts per transit are extremely \nsmall, hence the mode resonances lie very close together in frequency. \nAt the same time, the reflection coefficients of the best optical mirrors \nare so poor, and the Q of the interferometer is so low, that the resonance\n\nFig. 20 \u2014 TE modes in a parallel-plane waveguide and circular electric mode in \n1 circular waveguide\n\nFig. 21 \u2014 Comparison of computed phase shifts based on waveguide theory and \non interferometer theory.\n\nline width contains hundreds of normal mode resonances. Thus the uni- \nform plane wave undergoes very little decomposition when resonated. \nNevertheless, in the case of an active interferometer, the decomposition \nmay be complete.\n\nWe now make use of the formula for the Q of a resonant waveguide \ncavity to compute the Q of an interferometer system. The Q of a reso- \nnant waveguide cavity is given by\n\nwhere a@ is the attenuation constant of the waveguide and A, is the guide \nwavelength. For the interferometer we assume that @ is zero and that A, \nis equal to A, the free-space wavelength. The voltage reflection coeffi-\n\nwhere 6, is the power loss in reflection and 6g is the power loss due to \nspill-over. When these losses are very small, Q reduces to\n\nThe resonance line width at half-power points given as the change in \nelectrical length of the resonator, Ag, is\n\nLet us consider an interferometer having circular plane mirrors with \n2a = lem, b = 20cm, = 5 X 10 \u00b0cmand a reflection loss of 5, = 0.02. \nIn this case N = a'/b\\ = 250. Extrapolating the loss and phase shift \ncurves of Figs. 8 and 9, we obtain diffraction loss 6, = 9 X 10\u00b0 and \nphase shift for the dominant (TEM) mode gg = 0.11 degree. The dif- \nfraction loss is thus negligible compared to reflection loss, which limits \nthe Q to a value of 1.25 X 10\u00b0. The phase shift for the next higher order \n(TEM) mode is 0.30 degree and therefore it is separated from the \ndominant (TEM) mode by 0.19 degree or 0.0033 radian. The resonance \nline width, as given by (19), is 0.02 radian. Thus we see that TEMoo \nand TEM, modes are not resolved. As the mirror separation is reduced \nor mirror size increased, more and more normal modes will become \nunresolved and a uniform plane wave will suffer less decomposition when \nresonated.\n\nWhen an interferometer is filled with an active medium, the medium \ncan compensate for the mirror losses and yield an enormously increased \n(). Under these circumstances, the modes may be clearly resolved, and \ntheir Q\u2019s will be determined by the diffraction losses. If the gain of the\n\nmedium is increased until it compensates for mirror losses plus the dif- \nfraction loss of the lowest order mode, that mode will become unstable \nand oscillation can result. All higher-order modes will be stable and \nhave positive net loss. If the gain of the medium is further increased, \nthen many modes may become unstable. In starting from a quiescent\n\ncondition, spontaneous emission can initiate a large number of charac- \nteristic waves in the interferometer. These may then start to grow, but \nthe dominant mode will always grow faster and should saturate first. \nAt saturation the steady-state field distribution will be considerably \naltered. The relative field at the edges of the mirrors should increase, \nthereby increasing the relative power loss. This can be described as a \ncoupling of power into other modes as a result of the nonlinearity of the \nmedium. No attempt has yet been made to analyze this situation. The \nlinear theory is at present of most interest because it allows the com- \nputation of the starting conditions for oscillation.\n\nWith the development of the normal mode picture of interferometer \noperation and the computation of the losses for these modes, we may \nnow ask if there is an optimum geometry for a maser interferometer \nwhich will permit oscillation for the lowest possible gain in the medium. \nWe know that the power gained from the medium can be increased by \nincreasing length. For very great lengths corresponding to the far-field \nregion (NV < 0.1), the power gained from the medium increases more \nrapidly than transmission loss as length is increased, and there must \nalways be some length beyond which oscillations can occur. However, \nthese lengths are too great to be of practical interest. In the near-field \nregion (N > 1), represented by the curves of Fig. 8, the diffraction \nloss increases more rapidly than the medium gain. Therefore, if the \nreflection loss is sufficiently small, an optimum length may exist which is \nmost favorable for oscillation.\n\nTo be more specific, let us consider a circular plane mirror interferome- \nter. From Fig. 8 we find that the loss for the dominant mode may be \nrepresented by the expression\n\nIn order to find the optimum value of b to give a maximum Q, (20) and \n(18) are substituted in (17) and the resulting equation is differentiated \nwith respect to b. For the optimum ), the diffraction loss is 2.5 times the \nreflection loss, and not equal to it, as might be supposed. Moreover, this \nresult is general and holds for all modes and all shapes of plane mirrors \nrepresented in Fig. 8, provided the optimum falls on the straight-line\n\nportions of the loss curves. Since the power supplied by the medium is \nproportional to the stored energy in the interferometer, while the power \nloss of the passive interferometer is just w/Q times the stored energy, \noscillation is most likely to occur when Q is a maximum. Fig. 22 il- \nlustrates the way the interferometer dimensions affect Q. If a given mir- \nror diameter is chosen (as represented by the dashed line a), there is \nclearly an optimum distance 6 which will produce a maximum Q (inter- \nsection of lines A and B). However, if the distance b is held constant, \nthere is no optimum value for a. The larger a, the higher will be Q, \nalthough it will approach a limiting value beyond which there is nothing \nto be gained by further increase of a. \nAs an example, let us assume a case where\n\n5, = power reflection loss = 0.001. \nThe optimum proportions require that 6, be 0.0025, and for this, b is \n$35 em and the resulting Q is 7.8 X 10\u00b0. The length of 435 em is prob- \nably impractically large for a maser. If 6 is reduced to a more reasonable \nvalue of 50 cm, the Q will drop to 3.14 X 10\u00b0, which is the limiting value \ndue to reflection loss. (The value assumed here for 6, is already much \nlower than can be obtained from evaporated metal films and would\n\ncillate, the active medium would have to have a power amplification \nfactor in excess of 1.00002 per centimeter of path.\n\nIn the case of confocal paraboloidal mirrors of 2 em diameter, the \noptimum length turns out to be 8900 cm. If the diameter is reduced to \n0.5 em, the optimum length is still 530 cm, and for these proportions Q \nis 3.1 X 10\u201d. It is clear that with confocal mirrors the diffraction losses \nare negligible for any reasonable proportions of the interferometer.\n\nOne question of importance is whether there is an optimum set of \ndimensions which will discriminate against unwanted modes. It has \nsometimes been suggested that by making the mirror diameter small \nrelative to the mirror spacing, \u2018\u2018slant rays\u2019\u2019 will be more rapidly lost from \nthe system. However, from Fig. 8 it can be seen that the ratios of the \nlosses for the several modes is independent of N provided N is greater \nthan 1. Thus, if diffraction losses predominate, there is no way of dis- \ncriminating against unwanted modes by juggling dimensions. The limit- \ning amount of discrimination is merely governed by the ratio of the \nlosses for the different modes, which is independent of the dimensions. \nHowever, if reflection losses predominate, the discrimination between\n\nFig. 22 \u2014 Interferometer dimensions for constant Q. (Circular plane mirrors, \nreflection loss = 6, = 0.001.)\n\nlower-order modes would be almost nonexistent and it would be ad- \nvantageous to increase mirror separation and/or decrease mirror dimen- \nsions so as to make diffraction losses predominate. In the case of the \nconfocal mirrors, the loss ratios between modes are not constant (Fig. \n15) although, for values of N larger than those shown, they may become \nso. At any rate, for values of N close to unity, a small amount of in- \ncreased discrimination against higher order modes can be obtained by \nmaking the mirrors larger.\n\nDiffraction studies carried out on the IBM computer have led to the \nfollowing conclusions:\n\n1. Fabry-Perot interferometers, whether of the plane or concave \nmirror type, are characterized by a discrete set of normal modes which \ncan be defined on an iterative basis. The dominant mode has a field \nintensity which falls to low values at the edges of the mirrors, thereby\n\ncausing the power loss due to diffractive spillover to be much lower than \nwould be predicted on the assumption of uniform plane wave excitation.\n\n2. Uniform plane waves are not normal modes for a flat-plate inter- \nferometer. Consequently, interferometer resonances do not exist for \n\u201cslant rays,\u201d 1.e., plane waves traveling at an angle with respect to the \nlongitudinal axis.\n\nThe losses for the dominant mode of the plane mirror system are \nso low that for most practical geometries performance will be limited by \nreflection losses and scattering due to aberrations. For confocal mirrors \nthe diffraction losses are even lower.\n\nThere are no higher-order modes with losses lower than the dom- \ninant (lowest-order) mode.\n\nThe ratio of diffraction losses between the modes investigated for \nthe plane mirror system is independent of the interferometer dimensions \nin the range of interest. Therefore, if diffraction losses predominate, there \nis no way of proportioning the interferometer so as to favor any one \nmode.\n\nThe computer technique we employed is general and versatile. It can \nbe used for studying mirrors having rather arbitrary but small curva- \ntures. With little modification, the same technique can be used to study \nthe effects of aberration and misalignment.\n\nThe geometry for rectangular plane mirrors parallel to the xy plane is \nshown in Fig. 2. According to (1), the iterative equation for computing \nthe field at <i wartace of mirrors is\n\nActually, the stringency of this requirement can be relaxed somewhat for \nao sr-order modes in which field intensities near the edges of the mirror are rr ather \nlow. We have made check computations for the case a?/b\\ = 5 and (b/a)? = 25 \nand have found that the results based on the exact equation and on the smpeeni\n\nand the factor e \u201d\u201d is absorbed in y. \nHere, the kernel of the integral equation is separable in x and y. If\n\nit is possible to separate (23) into two equations, one involving \u00abx only \nand the other involving y only; that is,\n\nel\u00ae 4 iia \nr Kh ly y2)*/26b Qf \nK, = \u2014 \u00ab = . (25d ) \nav Nb \nThe product of the eigenvalues y, and y, is equal to the eigenvalue y in \n(23).\n\nIt remains to be shown that (25a) through (25d) represent integral \nequations for infinite strip mirrors. Let us consider a pair of infinite \nstrip mirrors of width 2a and separated by b. The iterative equation \nfor computing the field at the mirrors can be derived from (1).\n\n186 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1961 \nThe corresponding integral equation is\n\nAPPENDIX B \nCircular Plane Mirrors \nAssuming approximately plane waves propagating normally to the\n\ncircular plane mirrors (Fig. 3), the iterative equation for computing the \nsteady-state field distribution can be written as\n\n\u00a52\"\"r, dg, dr; , \nwhich is valid for (a\u00b0/b\\) \u00ab (b/a)\u201d.* \nThe integral equation corresponding to (30) is \nsa alr\n\n*Comments in Appendix A regarding the stringency of this requirement are \nalso applicable herein\n\nis absorbed in y. Making use of the relation \nTes) l [ et (rirelbeos(\u00a2y #2 \nQa \u201c0 \nand integrating (31) with respect to g; , it is seen that\n\nFor confocal spherical mirrors of circular cross section (Fig. 4), the \niterative equation corresponding to (29) is\n\nK(r2, \u00a22371,\u00a21 \nand where the factor e \u201d\u2019 is absorbed in y. Just as in the case of circular \nplane mirrors, it can be shown that\n\nCollins, R. J., Nelson, D. F., Schawlow, A. L., Bond, W., Garrett, C. G. B. \nand Kaiser, W., Coherence, Narrowing, Directionality and Relaxation \nOscillations in the Light Emission from Ruby, Phys. Rev. Letters, 5, \n1960, p. 303\n\nHildebrand, F. B., Methods of Applied Mathematics, Prentice-Hall, Englewood \nCliffs, N. J., 1952\n\nConnes, P., Increase of the Product of Luminosity and Resolving Power of \nInterferometers by Using a Path Difference Independent of the Angle of \nIncidence, Revue d'Optique, 35, 1956, p. 37\n\nGoubau, G., and Schwering, F., On the Guided Propagation of Electromag \nnetic Wave Beams, URSI-IRE Spring Meeting, May 1960, Washington, \nnp. GC\n\nBoyd, G. D., and Gordon, J. P., Confoeal Multimode Resonator for Millimeter \nthrough Optical Wavelength Masers, this issue, p. 489.\n\nMultimode resonators of high quality factor will very likely play a sig- \nnificant role in the development of devices, such as the maser, which operate \nin the millimeter through optical wavelength range. It has been suggested \nthat a plane-parallel Fabry-Perot interferometer could act as a suitable \nresonator. In this paper a resonator consisting of two identical concave \nspherical reflectors, separated by any distance up to twice their common \nradius of curvature, is considered.\n\nMode patterns and diffraction losses for the low-loss modes of such a \nresonator are obtained analytically, using an approximate method which \nwas suggested by W. D. Lewis. The results show that the diffraction losses \nare generally considerably lower for the curved surfaces than for the plane \nsurfaces. Diffraction losses and mode volume are a minimum when the \nreflector spacing equals the common radius of curvature of the reflectors. \nFor this case the resonator may be termed confocal. A further property of \nthe concave spherical resonator is that the optical alignment is not extremely \ncritical.\n\nSchawlow and Townes\u2019 proposed that coherent amplification could \nbe achieved in the infrared through optical regions of the frequency \nspectrum by maser techniques. At such frequencies multimode resonators \nare necessary to achieve reasonable dimensions and high Q. They and \nProkhorov\u2019 and Dicke\u2019 have suggested as a resonator two plane-parallel \nreflecting planes, known as a Fabry-Perot interferometer, or etalon.\"\n\nIn Fabry-Perot resonators the major factors contributing to the Q \n(i.e., resolving power) are reflection losses and diffraction losses. Reflec- \ntion losses result from absorption in the reflectors, and from transmission\n\nthrough them. At optical frequencies a very good layered dielectric \nreflector? can have a 993 per cent reflection coefficient. Diffraction \nlosses result from the finite aperture of the reflectors and from im- \nperfections in their \u201cflatness.\u201d\n\nFox and Li have shown in the accompanying paper\u2019 that modes, in \nthe sense of a self-reproducing field pattern, exist for an open structure \nsuch as a Fabry-Perot interferometer. They also have recognized that \nthe diffraction losses of a plane-parallel Fabry-Perot are very much less \nthan those obtained by assuming a uniform intensity distribution over \nthe reflector and the Fraunhofer far field diffraction angle. They have \nmade numerical self-consistent field calculations based on Huygens\u2019 \nprinciple to determine the actual diffraction losses and mode patterns.\n\nIn interferometry using a Fabry-Perot resonator, one normally ex- \ncites a system of plane waves traveling at certain discrete angles to the \naxis. Constructive interference at each of these discrete angles, as is \nappropriate to ring order, wavelength and spacing, results in a pattern \nof concentric bright rings. Schawlow and Townes indicated that each \nring of the interference pattern is not a pure mode of the resonator but \nan infinite sum of such modes, each representing a different field pattern \nover the reflector. This idea has been given much substance by the work \nof Fox and Li.\n\nThe plane-parallel Fabry-Perot is not necessarily ideal, however, as a \nhigh-frequency multimode resonator. A resonator formed by two \nspherical reflectors of equal curvature separated by their common \nradius of curvature is considered in detail in this paper. The focal length \nof a spherical mirror is one-half of its radius of curvature. Therefore the \nfocal points of the reflectors are coincident and the resonator is termed \nconfocal. G. W. Series, Fox and Li* and Lewis\u2019 have also suggested the \nconfocal resonator. Lewis has recognized that it would have lower \ndiffraction losses than the plane-parallel Fabry-Perot and has described \nthe analytic solution presented here.\n\nthey have an axis and thus lose the advantage of ease of adjustment. \nIl. RESONATOR QUALITY FACTOR \nResonator quality factor, or Q, is defined as\n\nConsider an interferometer consisting of two reflecting surfaces separated \nby a distance d which is large compared to the wavelength in the medium \n\\. By considering waves bouncing back and forth between the surfaces, \none may derive an approximate Q as\n\nwhere the power reflection coefficient per bounce is r = 1 \u2014 a. Resolving \npower is thus synonymous with Q within the small loss approximation \nof (2).\n\nIf diffraction losses are small compared with reflection losses, then \nresonator Q is proportional to the spacing bet ween the reflecting surfaces. \nlor a given reflector aperture size, the resonator Q will continue to \nincrease with the spacing d between the reflectors until the diffraction \nlosses become roughly comparable with the reflection losses. Further \nincrease in spacing then decreases the Q because of increasing diffraction\n\nAll resonator dimensions are assumed large compared to a wavelength; \nthe modes and diffraction losses of the confocal resonator are therefore \nobtainable from a self-consistent field analysis using Huygens\u2019 principle.\" \nA confocal resonator is considered, with identical spherical reflectors \nof radius b, as shown in Fig. 1. Assume the field to be linearly polarized \nover the 7\u201d surface in the y direction and given by Eofm(2\u2019)g,(y\u2019), \nwhere \u00a35 is a constant amplitude factor and f,,(.2\u201d) and g,(y\u2019) are the \nfield variations over the aperture. At point P?(.r,y) on the other surface, \none computes the electric field by summing over contributions from\n\nP\u2019P and the normal to the reflector surface at 2\u2019, and k is the propaga- \ntion constant of the medium between the reflectors. Note that /\n\n2a/d, where \\ is the wavelength in the medium. The electric field in the \nxz plane is approximately zero. The reflector is assumed square and of \ndimension 2a, which is small compared to the spacing 6 (since the con- \nfocal spacing is under consideration d = 6), and thus @ is very nearly \nzero. The medium is assumed to fill all space.\n\nThe normal modes or eigenfunctions of the confocal resonator are \nobtained by requiring that the field distribution over .x\u2019y\u2019 reproduce \nitself within a constant over the ry aperture, and thus F, Evfim( x) gn(y), \nwhere EF om0,/25. The proportionality factor \u00a2,0, is generally com- \nplex, giving both amplitude and phase changes. The resulting integral \nequation is\n\nThe distance p varies only a small amount for small apertures and thus \nmay be replaced by the separation b except in the exponential phase \nterm. For x and y small compared to b one can show that\n\nwhere w = 2\u00b0 + y\u2019. The third term makes a negligible contribution to \nthe phase when a\u2019/bA < b\u2019/a\u2019. Note that in this approximation one \ncannot distinguish between spherical and parabolic surfaces. In terms \nof some dimensionless variables\n\nve \n. \u00b0 11 \u00b0 . \u00b0 \u00b0 \u2018 \nSlepian and Pollak\u201d have considered the following integral equation:\n\nrus. . \u2018 . . . xx\u2019 \nrhis is a homogeneous Fredholm equation of the second kind with \u00ab \nas the kernel. It is often referred to as a finite Fourier transform. They\n\nwhere Son(e,n) and Rom (e,L) are respectively the angular and radial \nwave functions in prolate spheroidal coordinates as defined by Flammer,\u201d\u201d \nand where 9 = X//c = x/aand n = Y/Vc = y/a respectively for \nF\u2019,,(X) and G,(Y). There is an infinite number of eigenfunctions and \ncorresponding eigenvalue solutions to (9) for any value of c. Flammer\u2122 \ngives values of these functions for \u00a2 S 5 and Slepian and Pollak\" \nhave computed the eigenvalues x,, for the important region of \u00a2 > 5. \nThe eigenfunction solutions of (8) are thus the spheroidal wave \nfunctions So\u00bb,(\u00a2,v/a)NSo,(\u00a2e,y/a). The eigenfunctions are real; therefore, \nthe reflecting surfaces are of constant phase. The eigenvalues are \nOo mon XmXnt \u00a2 - (12)\n\nThe phase shift between the two reflecting confocal surfaces equals the \nphase angle of o,\u00a20,. For resonance the round-trip phase shift must\n\n494 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1961 \nequal an integer g times 27. From (11) and (12), one finds therefore\n\nThe confocal resonator is seen to have resonances only for integer values \nof the quantity 4b. If 4b, is odd, (m + n) must be even, likewise if \ntb AX is even, (m + n) must be odd. Note that considerable degeneracy \nexists in the spectrum; increasing (m + n) by two and decreasing q \nby unity gives the same frequency. The degenerate modes are or- \nthogonal over the reflector surface since they satisfy the integral (5) \nwith different eigenvalues. The modes have negligible axial electric \nand magnetic fields and thus will be designated by TEM,,,\u00a2, where \nm and n equal 0, 1, 2,--- , and refer to variations in the x and y di- \nrections, while g equals the number of half-guide wavelength variations \nin the z direction between reflectors.\n\nThe fractional energy loss per reflection due to diffraction effects is \ngiven by\n\nrhe function | \u2014 | x, |\u00b0 versus \u00a2 is shown in Fig. 2 for m OG: Boe: \nIt can be shown that Fig. 2 also gives the diffraction losses for an infinite \ncylindrical reflector strip of width 2a and radius of curvature b. The \ndiffraction losses for various TEM,,,, modes are shown in Fig. 3. Note \nthat TEM,,, and TEM,,, (u # v) have the same diffraction losses;\n\n, and TEM)\u00bb, are so nearly \nequal that they can be plotted as one curve. As indicated previously, \nthese last two types of modes cannot both be resonant at the same \nfrequency. Note that the losses are primarily determined by the higher\n\nof the transverse mode numbers m, n, regardless of the field polarization.\n\nwith circular reflectors are also shown. The diffraction losses for the \nconfocal resonator are seen to be orders of magnitude smaller than for \nthe plane parallel resonator. Fox and Li have also obtained numerical \nresults for the confocal resonator with circular cross section of radius a. \nThese are in good agreement with the results presented here, allowing \nfor the fact that in this paper the reflectors have a square cross section \nof width 2a.\n\nFig. 2 \u2014 Kigenvalues of integral equation; also the diffraction losses of an in \nfinitely long evlindrical reflector of width 2a.\n\nIf one approximates the diffraction loss curve by a function ap \nSS yaks \nflector radius a that the resonator Q is a maximum as a function of the\n\nconfocal spacing 6 when the reflection loss equals [2.30B(a?/bA) \u2014 1] \ntimes the diffraction loss. For the THMo9, mode, A 10.9 and B\n\nThe diffraction loss for the plane-parallel case assuming a uniform \nfield and phase distribution and a diffraction angle of 6 \\/2a is also \nshown. This diffraction angle corresponds to the first Fraunhofer mini- \nmum in far field theory. For a square (or circular) reflector of side 2a \nthe diffraction loss is approximately\n\nFig. 3 clearly demonstrates the inadequacy of the assumption of uniform \nintensity distribution.\n\nThough the eigenvalues given by (12) must be known accurately the \neigenfunctions are only of approximate interest. Flammer\u2019 shows that, \nin the approximation of n\u00b0 < 1 (near the center of the reflector), (10) \nbecomes\n\nThe inode shape is thus approximately a Gaussian times a Hermite \npolynomial H,,(X ). The gamma function is arbitrarily chosen as normal- \nization such that F,,(X = 0) = +1 for m even:\n\nF3(\u00a2,n) (2en> \u2014 l)e *\u201d\u2122 \nThe approximation involved in (17) fails away from the center of the \nreflector. For reasonably large values of c, however, the field is weak \nthere, and of little interest. The diffraction losses were previously ob- \ntained from (15). Curves representing (18) for various values of \u00a2 are \nshown in Fig. 4. The dotted curves for \u00a2 = 5 are the true eigenfunctions \n\u2019 : . \u2019 12 \nSon(\u00a2e,n) as obtained from llammer.\n\nThe exponential dependence of the electric field on en, which is \nindependent of the reflector half-width a, leads one to define a \u201cspot \nsize\u201d at the reflector of radius w w, , Where w ve + y, at which \nthe exponential term falls to \u00a2\n\nThe only effect of increasing the reflector width 2a is to reduce the \ndiffraction losses; the spot size is unaffected.\n\nIf one allows the reflectors to be somewhat lossy or partially trans- \nparent, then the resonator Q is reduced over that implied by diffraction \nlosses alone. The field distribution, i.e., the mode pattern, is not seriously \naffected so long as the losses are small and fairly uniform over the plates.\n\nFig. 4 Approximate field amplitude variation versus normalized radius for \nvarious modes. The exact dependence given by the angular prolate spheroidal \nfunction Som (ec, 7) is shown by dashed lines.\n\nThe electric field patterns derived thus far have all been linearly \npolarized. Fox and Li\u2019 have recognized that, by superimposing the \nTEM, mode linearly polarized in the x direction and the TEM, \nmode linearly polarized in the y direction, the lowest-order circular \nelectric mode can result, and it has the same diffraction losses as the \nlinearly polarized TEM, mode. Many other polarization configurations \ncan be obtained in this manner.\n\nThe field over the confocal aperture has been obtained in the preced- \ning section. The field over an arbitrary plane z = 29, as in Fig. 1, is also \nobtainable by a straightforward application of Huygens\u2019 principle as \nstated in (4). The arbitrary plane zo) may be placed outside the confocal \ngeometry as well as inside provided one takes into account the trans- \nmission loss of the reflector. The field distribution over the confocal \nsurface is given by F,,(c,7/a)G,(c,y/a). For large \u00a2 the spheroidal \nfunctions may be approximated by the Gaussian-Hermite functions. \nThe integral can be evaluated in the limit of c > \u00ab.\n\nWithin these approximations, the traveling wave field of the confocal \nresonator resulting from the field at one of the reflectors is given by\n\nWhen the reflecting surface is made partially transparent, as will be the\n\nwill be a traveling wave as given in (20) reduced by the transmission \ncoefficient of the reflector. Within the resonator, the field will be a\n\nstanding wave. The transverse standing wave is as given in (20) except \nthat the exponential phase function is replaced by the sine function.\n\nThe surface of constant phase which intersects the axis at 2) as ob- \ntained from (20) is given approximately by\n\nneglecting the small variation in \u00a2 due to variation in z. This surface is \nspherical, within the approximations of this paper, and has a radius of \ncurvature 6\u2019 given by\n\nAt & = +1 it coincides with the spherical reflector as expected. Also note \nthat the symmetry or focal plane ( = 0) is a surface of constant phase.\n\nThe field distribution throughout the resonator is given by the modulus \nof (20). The complete field distribution within the confocal resonator is \nshown schematically in Fig. 5 for the low-loss TEM oo, mode.\n\nThe field distribution over the focal plane is less spread out than over \nthe spherical reflectors. The field spot size over the spherical reflectors \nwas defined by (19). In any arbitrary plane zp the exponential term in \nthe field distribution falls to e * at a radius\n\nTo obtain the radiation pattern angular beam width of the TE Moo, \nmode spherical wave, one takes the ratio of the spot diameter from (20) \nor (24), as\u2014\u2014 ~, to the distance from the center of the resonator. The \nbeam width between the half-power points is given by\n\nSince the surfaces of constant phase of the confocal resonator are \nspherical, it is apparent that (20) also represents approximately the \nfield distribution between two spherical reflectors of arbitrary spacing. \nThat is, any two surfaces of constant phase may be replaced by re- \nflectors. The frequencies at which such a resonator will be resonant. will \nof course be determined by satisfaction of the phase condition.\n\nFig. 5 Field strength distribution within the confoeal resonator for the \nTEM oo, mode.\n\nConsider two identical spherical reflectors of radius of curvature b\u2019 \nspaced a distance d. The only restriction is that b\u2019 2 d/2. The confocal \ngeometry of spacing b of which this resonator is a part is [set & = d/b in\n\nThe spot size at the reflectors in the nonconfocal resonator may be \nimmediately obtained from (24) with & = +d/b. It is\n\nNote that the factor [2(d/b\u2019) \u2014 (d/b\u2019)*| achieves a maximum of unity, \nas a function of 6\u2019, when 6\u2019 d. Thus, for a given spacing between re- \nflectors, the spot size is a minimum for the confocal resonator.\n\n502 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1961 \nsection of dimension 2a\u2019 on the assumption that this loss is equal to that\n\nof its equivalent confocal resonator with reflector dimensions scaled up \nby the ratio of their spot sizes. The equivalent confocal resonator has\n\nThus the confocal geometry gives minimum spot size and minimum \nlosses for a given spacing. If one defines the mode volume as the spot \nsize at the reflector times the spacing, it is clear that the minimum mode \nvolume also results from the confocal geometry. The mode volume, so\n\nwhen the diffraction losses derived from the \u2018equivalent\u2019 confocal \ngeometry are small, that is, when the reflector dimension a\u2019 is somewhat \nlarger than the spot size. In an exact solution for the nonconfocal case \none should again start from the integral (4), and clearly the field dis- \ntribution and losses so derived will depart from that obtained from the \nequivalent confocal case if the confocal field is not substantially all \nintercepted by the nonconfocal reflectors. Conversely, so long as the \nspot size is small compared to the reflector dimension a\u2019, one expects the \nfield distribution and losses to be very nearly correctly given by the \nequivalent confocal solution.\n\nThe phase shift between the two reflecting nonconfocal surfaces may \nbe obtained from (20). The condition of resonance may then be shown \nto be\n\nIn the nonconfocal case 4\u00a2d/ is no longer necessarily an integer at \nresonance. It is more important, though, that the modes are no longer \ndegenerate in m + n. The spectral range or mode separation for the \nnonconfocal resonator is given by\n\nNote that in the confocal case the set of modes mnq 00g, Olq are \nmaximally split in frequency, whereas if the parameter in parenthesis \nequals 3 (when d/b = 0.414) then the mng = 00g, Olg, 11g, 12\u00a2 modes \nare maximally split in frequency.\n\nwhere m and n are small integers and g a large integer. In the confocal \ncase (b = d) note that equations (31) and (33) reduce to (14).\n\nThe theory of this section does not extend to the limit of plane-parallel \nreflectors, i.e., infinite radii of curvature. Let the spacing d remain fixed \nwhile b\u2019, and consequently [by (26)] the confocal radius b, approaches \ninfinity. The spot size, as seen from (27), keeps increasing with 6\u2019, and, \nas has been noted above, this results eventually in the breakdown of \nthe whole idea of an equivalent confocal resonator. The relations for \nthe nonconfocal resonator are valid as long as the reflector aperture \nradius a\u2019 is somewhat larger than the field spot size radius given by (27). \nThat is, one must require\n\nFor comparison purposes, consider the resonances of a rectangular \n\u00b0 . : \u2018 a bh ] \nconducting box in the manner of Schawlow and Townes. Let the \ndimensions be 2a X 2a X b:\n\nwhere q, r, and s are integers. Modes where q > r,s can be thought of \nphysically as waves bouncing predominantly back and forth between \nthe reflecting end plates of the rectangular box. The spectral range or \nmode separation is given by\n\nRemoving the conducting side walls causes large diffraction losses for\n\nthe r = 0 or s = O modes since they have a strong field at the edge of \nthe reflectors. Large r or s modes represent waves traveling at a con- \nsiderable angle to the normal between the reflectors and thus these \nmodes have such large diffraction losses that they are eliminated as \nresolvable resonant modes. Modes with r, s = 1,2,---, have small dif- \nfraction losses, and are approximations to the actual modes which can \nexist in the resonator without the conducting side walls. Fox and Li\u2019s*\u00ae \nwork shows that for a\u00b0/b\\ greater than unity the mode separations of a \nplane-parallel Fabry-Perot are given approximately by (36), the ap-\n\nThe mode separation corresponding to Ar or As = 1 has, to the writers\u2019 \nknowledge, never been resolved at optical frequencies due to the large \nvalues of a\u00b0/b\\ and low values of reflectance used. Calculations show, \nthough, that for reflectance coefficients of about 0.99 and a\u00b0\u2019/b\\ & 4, \nsuch that diffraction losses are comparable with reflector losses, the \nresonances should be resolvable.\n\nThe mode separation due to Ag = | is easily resolvable and is given by \nA(1/X) = 1/2b. This is the spectral range as normally stated for the \nplane parallel Fabry-Perot interferometer. It corresponds to changing \nthe number of half wavelengths between the reflecting surfaces by one.\n\nThe confocal resonator is resonant for integer values of 4b/A. The \nmode separation due to Aq l is A(1/X) | /2b. The modes are de- \ngenerate in frequency in that for a given integer 4b/\\ all TEM,,,,, modes \nare resonant such that m + n remains even or odd according to whether \n1b/X is odd or even. The modes of the plane-parallel Fabry-Perot are \nnot degenerate, except for rsq and srq. A possible advantage of this de- \ngeneracy of the confocal modes will be discussed in the next section.\n\nA type of solid state optical maser has recently been demonstrated by \nMaiman\u201d\u2122 and by Collins et al.\u2019 It consists of a fluorescent crystal ma- \nterial (ruby) a few centimeters in length and a few millimeters in diame- \nter. The crystal material should be optically homogeneous. The ends of \nthe crystal are optically flat and parallel. The ends are silver-coated for \nhigh reflectance. One of the reflecting surfaces must be slightly trans- \nparent, as the output of the optical maser is obtained through the \nreflecting surfaces. Thus far, silver has been used to provide the reflec- \ntion, but for ultimate performance multiple-layer dielectrics\u2019 should be \nused to obtain low transmission loss as well as high reflectance. The \npump power enters the fluorescent erystal from the side.\n\nflection losses, then the resonator Q is proportional to the spacing be- \ntween the reflecting surfaces. Consider a confocal resonator and a \nplane-parallel resonator each of spacing b and of equal Q. The energy \ndistribution in the former is more concentrated on the axis and thus the \nconfocal resonator has a smaller effective mode volume. The volume of \nmaser material required will thus be less for the confocal than for the \n; plane parallel resonator. For maser oscillation the required excess density \nof excited states depends only on the cavity Q and in no other way upon \nthe resonator shape.\u2019 The pump power is proportional to the volume of \nmaser material times the density of excited states divided by the natural \nlifetime of the excited state. Thus, assuming equal Q, the confocal \nresonator with its smaller volume of material requires less pump power \nthan the plane parallel resonator by the ratio of their cross-sectional\n\nand pump power with regard to the use of optical fibers in maser \napplications.\n\nThe minimum volume of maser material is limited by diffraction losses. \nIf diffraction losses are to be considerably less than | per cent for the \nlowest-order mode so as to be small compared to achievable reflection\n\nIfb = 4emand \\ = 10\u2018 em, then the rod of maser material should be \napproximately 0.4 mm in diameter. A rod of larger diameter would waste \npump power in that the field of the confocal resonator would be very \nweak outside this minimum diameter of material.\n\nThe analysis of the confocal resonator assumes a uniform dielectric \nmaterial between the spherical reflectors. For reasons of minimizing the \npump power, it is necessary to use a small diameter of maser material. \nTherefore, to prevent internal reflection of energy from the sides of the \nmaser material, it may be advisable to grind rough the sides of the rod \nof maser material or to immerse it in a surrounding medium of equal \ndielectric constant. If this is not done, the energy assumed lost due to \ndiffraction effects would not escape and the electric field pattern will not \nbe as computed herein. A more important effect of internal reflection \nfrom the side walls would be to increase the Q of the transverse modes \nwhich would increase the spontaneous and stimulated emission power \nto these undesired modes, and thus increase the over-all pump power \nrequired.\n\nThe natural linewidth of the material used in an optical maser will, \nfor reflector spacing d of a few centimeters, be large compared to the\n\nmode separation determined by integer changes in r,s for a_plane- \nparallel resonator. Hopefully, the natural linewidth of the maser ma- \nterial will be less than the mode separation corresponding to integer \nchanges in g. Thus, there is the possibility that a plane-parallel resonator \noptical maser may frequency wander between low-order r,s modes.\n\nIf the diffraction losses are comparable to or exceed the reflection \nlosses for the lowest mode then, as can be seen from Fig. 3, the ratio of \nthe Q\u2019s of the lowest two modes of the confocal resonator exceeds con- \nsiderably the ratio of the Q\u2019s of the lowest two modes of the plane- \nparallel resonator. By the lowest order mode is meant m = n = 0, and \nr = s = 1, respectively, for the confocal and plane-parallel resonator. \nTherefore, maser oscillation is more likely to take place in only the \nlowest-order mode of the confocal than of the plane-parallel resonator. \nThis greater loss discrimination between modes may be one of the \nsignificant advantages of the confocal resonator.\n\nIn the confocal resonator optical maser, if the maser oscillation wan- \nders between modes the output beam pattern will change, just as in the \nplane-parallel resonator, but the frequency will remain fixed due to the \nmode degeneracy. Thus, the observed linewidth of the maser output \nmay be narrower for the confocal resonator.\n\nThe required accuracy on the confocal condition to achieve degeneracy\n\nthe mode splitting of the near-confocal resonator equals the mode separ- \nation of the plane parallel resonator. To achieve a significantly smaller \nmode separation in the near-confocal resonator than the plane parallel \nresonator would require proportionately greater accuracy in the radius \nof curvature and spacing of the curved surfaces.\n\nThe plane Fabry-Perot requires accurately parallel reflecting surfaces. \nThe confocal resonator requires only that the axis of the confocal reso- \nnator approximately coincide with the axis of the rod of maser material. \nThe axis of the confocal resonator is the line passing through the two \ncenters of curvature. The resonator axis must intersect the two reflect- \ning surfaces near their center. Define the effective aperture radius as \nthe distance from the point of intersection of the axis of the confocal \nresonator with the reflector surface to the nearest edge of the aperture. \nThe diffraction losses will be approximately determined by this distance.\n\nIf the minimum diameter of maser material is used, then the axis of \nthe confocal resonator must coincide with the material axis. Increasing\n\nthe diameter of the maser material wastes pump power but relaxes the \ntolerance on the resonator axis.\n\nIt is well to note that a single spherical reflecting surface and a plane \nreflecting surface spaced by approximately half the radius of curvature \nwill have similar properties to the confocal resonator and may be ad- \nvantageous if it is desired to bring the output through a plane surface.\n\nA confocal multimode resonator formed by two spherical reflectors \nspaced by their common radii of curvature has been considered. The \nmode patterns and diffraction losses have been obtained. The confocal \nspacing of the reflectors is found to be optimum in the sense of minimum \ndiffraction losses and minimum mode volume.\n\nThe diffraction losses are found to be orders of magnitude smaller \nthan those of the plane-parallel Fabry-Perot, as obtained by Fox and Li.\u201d \nIt is more important, though, that a greater diffraction loss diserimina- \ntion between modes occurs, and thus oscillation in other than the lowest- \norder mode is less likely for the confocal resonator, assuming that \ndiffraction losses are comparable to reflection losses.\n\nThe modes of the confocal resonator are degenerate, in that one-half \nof all the possible field pattern variations over the aperture are resonant \nat any one time. This degeneracy is split if the resonator is nonconfocal. \nThe splitting is comparable with that of the plane-parallel resonator\n\ndifferent from the common radius. The mode volume and diffraction \nlosses are insensitive to the confoeal condition.\n\nThe required volume of maser material is smaller for the confocal \nresonator than for the plane-parallel resonator, and thus the required \npump power is less. The confocal resonator is relatively easy to adjust \nin that no strict parallelism is required between the reflectors. The only \nrequirement is that the axis of the confocal resonator intersect each \nreflector sufficiently far from its edge so that the diffraction losses are \nnot excessive.\n\nThe example of a confocal resonator mentioned here was taken at \ninfrared-optical wavelengths; however, such resonators may be useful \ndown to the millimeter wave range by virtue of their low loss. In this \nconnection, recent work of Culshaw\u2019\u2019 on the plane-parallel Fabry-Perot \nat millimeter wavelengths is of importance.\n\nThe writers have been informed that Goubau and Schwering\u201d\u201d have \nrecently investigated diffraction losses of parabolic reflectors and that \ntheir results agree with the work presented here.\n\nFruitful discussions with A. G. Fox, W. D. Lewis, T. Li, D. Marcuse, \nS. P. Morgan and G. W. Series are sincerely appreciated. Mrs. F. J. \nMacWilliams performed the computations.\n\nSchawlow, A. L. and Townes, C. H., Phys. Rev., 112, 1958, p. 1940 \nProkhorov, A. M., J.E.T.P., 34, 1958, p. 1658. \n3. Dieke, R. H., U.S. Patent 2,851,652, September 9, 1958. \n. Meissner, kK. W., J. Opt. Soe. Am., 31, 1941, p. 405; 32, 1942, p. 185. \n5. Heavens, O.S8., Optical Properties of Thin Films, Butterworths, London, 1955. \n). Fox, A. G. and Li, T., this issue, p. 453; Proce. LR.E., 48, 1960, p. 1904. \n. Lewis, W. D., private communication. \nConnes, P., Revue d'Optique, 35, 1956, p. 37; J. Phys. Radium, 19, 1958, p. \n262 \nJenkins, F. A. and White, H. E., Fundamentals of Optics, 3rd ed., MeGraw \nHill, New York, 1957 \n. Silver, S., Microwave Antenna Theory and Design, M.1.T. Radiation Labora- \ntory Series, Vol. 12, McGraw-Hill, New York, 1949. \n1. Slepian, D. and Pollak, H. O., B.S.T.J., 40, 1961, p. 43. \n12. Flammer, C., Spheroidal Wave Functions, Stanford Univ. Press, Palo Alto, \nCalif., 1957 \n3. Slepian, D. and Pollak, H. O., private communication \n. Maiman, T. H., Nature, 187, 1960, p. 493. \n5. Collins, R. J., Nelson, D. F., Schawlow, A. L., Bond, W., Garrett, \nand Kaiser, W., Phys. Rev. Letters, 5, 1960, p. 303 \nSnitzer, E., J. Appl. Phys., 32, 1961, p. 36. \n. Culshaw, W., I.R.E. Trans., MTT-7, 1959, p. 221; MTT-8, 1960, p. 182 \n. Goubau, G. and Schwering, F., U.R.S.1.-I.R.E. Spring Meeting, Washington, \nMay 1960\n\nRelation Between Surface Concentration \nand Average Conductivity in Diffused \nLayers in Germanium\n\nIn this paper an expression is derived for calculating the average con- \nductivity of a diffused layer in semiconductor material as a function of the \nsurface concentration of the diffused impurity and the background impurity \nconcentration. Curves are presented depicting the relationship among these \nparameters for the case of germanium. Included are curves for both diffused \nimpurity types for the complementary error function, gaussian, exponential \nand linear impurity distributions.\n\nIn the design of semiconductor devices in which junctions are pro- \nduced by solid state diffusion of impurities, it is of great value to know \nthe relationships which exist between the surface concentration of the \ndiffused impurity, Co, the background impurity concentration, Cg , and \nthe average conductivity of the diffused layer, \u00a2 These relationships \ncan be calculated from a knowledge of the resistivity as a function of \nimpurity concentration for material uniformly doped with a single \nimpurity. Such calculations are presented in this paper.\n\nFor convenience, assume initially that impurity atoms are 100 per \ncent ionized. Therefore, the conductivity at a point in a diffused layer \nin semiconductor material can be given by\n\nprimarily a function of the total number of ionized impurities present. \nIf it is assumed that both ionized impurity types scatter a majority \ncarrier identically, then the mobility in (1) may be considered to be a \nfunction of (C + Cz). \nThe conductivity of material doped with a single impurity can be \nexpressed, again assuming 100 per cent ionization of impurities, as \n*\n\nC-\u20ac\u00a2 \noO o* = . o ( | ) \nC + Cz, \nA log-log plot of the resistivity of single impurity doped material as a \nfunction of the impurity concentration can be approximated by a set of \nintersecting straight lines each having an equation of the form\n\nintegrating (6b) over values of \u00ab from the surface (2 = 0) to the june- \ntion (2 = x;) and dividing by the junction depth, x; . Thus,\n\n\u20180, t) + Cal \"O(Cy . xv) \u2014 ('p| dv. (7) \nThe values assigned to B and @ at any point on the interval are deter- \nmined by the value of [C(Cy, 7) + C,] at that point. Equation (7)\n\ngenerally requires numerical integration, using experimental values for \nB and a.\n\nlig. | shows the variation of resistivity, p, of single-impurity doped \ngermanium as a function of the impurity concentration, NV. Points in the \nrange of 10 < N < 2 X 10\" for n-type material and in the range of \n10'* < N s 6 X 10\" for p-type material were taken from Prince.\u2019 \nPoints in the range of 2 X 10'\"\u00b0 < N < 10\u201d for n-type material and \n6 X 10\u00b0\" < N S 10\u201d for p-type material were taken from Hall effect \nmeasurements of Tyler and Soltys.\u201d Hall effect measurements give the \nresistivity as a function of carrier concentration. However, direct meas- \nurements of resistivity as a function of impurity concentration by\n\na be) ri - 3 \u00a2 . . . \nlrrumbore and Tartaglia\u201d for p-type material agree with the results of\n\nTyler and Soltys, thus justifying the assumptions made, at least for the \ncase of p-type material. Five straight-line approximations were made to\n\nach curve giving equations of the form of (5). Values of B and @ and \nthe range of validity of each set of values are shown in Table I.\n\nFig. 2 shows data of resistivity as a function of electron concentration \nfor n-type germanium as reported by Tyler and Soltys,) Moody and \nStrauss,* Furukawa,\u2019 Zhurkin et al.\u00b0 and Spitzer.\u2019 Also shown in Fig. 2 \nis a portion of the n-type curve from Fig. 1. As can be seen, this curve \nrepresents a reasonable average of the arsenic data, for which case the \npresent calculations are intended.\n\nEquation (7) was evaluated on the IBM 704 computer for various \nimpurity distributions, and the results were checked by hand calculation \nof several points. Seven values of background concentration were used, \nand four points per decade of surface concentration were evaluated. The \nresults are shown graphically in Figs. 3 through 10 on pages 514 through \n521 for the various distributions as follows:\n\nSince the conductivity of perfectly compensated material is not zero, \n(7) will be in error by some small amount. However, for values of Cy \nand Cs, such that (Cy \u2014 C,) 2 10 n; this error will be negligible. All \nvalues of Cy and C\u2019, used in these calculations fulfill this requirement.\n\nThe author wishes to thank J. C. Irvin and W. H. Jackson for many \nhelpful and enlightening discussions, and special thanks are due Mrs. \nG. N. Alfandre whose able programming of the IBM 704 made solution \nof (7) possible.\n\n3. Trumbore, F. A. and Tartaglia, A. A., Resistivities and Hole Mobilities in Very \nHeavily Doped Germanium, J. Appl. Phys., 29, 1958, p. 1511.\n\n1. Moody, P. L. and Strauss, A. J., Electrochemical Society Meeting, Chicago, \nMay 2, 1960; also private communication.\n\nExpressions are presented for the magnetization and pull relations of the \nmating flat magnetic reeds used as contacting members in reed relays. The \nresults of an experimental study expressed in dimensionless form give the \nforce or pull between the reeds in terms of their dimensions, the gap between \nthem, and the flux density. Since the attainable flux density is limited by \nsaturation, the pull expression leads to conditions which must be satisfied \nby the reed and gap dimensions to provide desired levels of contact and \nretractile force.\n\nThe ampere-turn sensitivity of a relay using mating magnetic reeds \ndepends on the flux required and on the reluctance of the magnetic path \nthrough the reeds. Expressions are given for the reluctance in the case of \nan air return path and in that where the air return ts partially replaced by \na shielding member.\n\nExpressions are also given for the operate time of such reed relays. This \ndepends on the concurrent flux development and motion of the reeds. The \ntime of flux development varies inversely with the power input to the coil, \nbut the motion time cannot be less than that required when the flux density\n\nA sealed reed relay comprises one or more sealed contacts assembled \nwith a coil and shielding and supporting members. The distinctive fea- \nture is the construction of a sealed contact.\u2019 In its simplest form, as \nshown in Fig. 1, it consists of a glass envelope in which are sealed a pair \nof magnetic reeds, which serve as contacting members, actuated by the \nmagnetic field induced in them by the coil current.\n\nIn another form,\u2019 there is only one moving reed, the other magnetic \nmember being short and nearly rigid. This construction, using two of \nthese short members to provide both back and front contacts, has been\n\nused with mercury-wetted contact surfaces.\u00b0* Both forms have been \nused in single- and multiple-contact relays, both neutral and with \npermanent magnet bias used to give polar or locking performance.\u2019 \nRecent work has included development of a miniature sealed contact.\u201d\n\nAs an aid to the design of sealed contacts, and to the understanding \nof their characteristics, an analysis has been made of their performance \nrelations, including the dependence of the pull, contact force, speed and \nsensitivity on the dimensions of the reeds. This analysis is presented \nhere in the form applying to the simple case of equal mating reeds, but \nis, with some modification, applicable to other forms of magnetic reed \ncontacts.\n\nThe major controlling factor in the performance of sealed contacts is \nthe flux-carrying capacity of the reeds, as limited by the saturation \ndensity of the reed material. The treatment given here therefore starts \nfrom the relation between the reed flux and the attractive force at the \ngap, which must deflect the reeds and provide the contact force. This \nrelation is determined essentially by the reed dimensions alone. The \nsensitivity, as measured by the ampere turns for operation and release, \ndepends upon both the flux and the reluctance of the magnetic circuit, \nand hence upon the dimensions of the coil and shielding members, as \nwell as upon those of the reeds. Relations are given for the estimation \nof reluctance and hence of sensitivity. Finally, expressions are given \nfor the estimation of the speed of operation, which depends upon the \ntimes of field development and of reed motion, with the latter establish- \ning a lower limit to the attainable time of operation.\n\nUsing the notation of Fig. 1 for the reed dimensions, the attractive \nforce F between the reeds is given by Maxwell\u2019s law as \na \nRrab\u2019\n\nwhere \u00a2q is the flux in the gap where the reeds overlap. The total flux \n\u00ae is the sum of \u00a2g, and the fringing flux which passes from one reed to \nthe other by air paths around the gap. As shown below in the discussion \nof reluctance, approximate estimates may be made of the fringing flux, \nbut the relations involved are not adapted to a simple direct treatment \nof the relation between the pull / and the total flux &. A direct experi- \nmental study was therefore made to determine the approximate form \nof this relation.\n\nIn this study, pull and flux measurements were made of four different \nsets of reeds, having the dimensions listed in Table I. Each set of reeds \nwas assembled with overlap values, a, of 25, 50 and 100 milli-inches, and \nthe pull and flux measured for values of gap x in the range from 1 to \n10 milli-inches over the range of coil energization from 0 to 300 ampere \nturns.\n\nThese measurements were all made with the reeds supported in a \nbrass fixture which could be adjusted to make the overlap surfaces \nparallel and to set the overlap and the gap at desired values. The coil \nused had inside and outside diameters of 0.39 and 0.86 inch, and was \n1.64 inches long. It was centered over the gap in making the measure- \nments, and was provided with a central search coil located on its inner \ndiameter. Measurements were also made of the flux in the coil alone, \nand this air-core flux was subtracted from the flux readings to give the \nreed flux \u00ae.\n\nIf the reeds are long compared with the overlap, as in practice and in \nthese measurements, the pull is independent of the reed length. If the \npermeability of the reed material is high enough for the reeds to be\n\nessentially equipotential surfaces near the gap, the pull F is a function\n\nof &, x, a, b and h only. For dimensional consistency, therefore, the pull \nmust be given by an equation of the form \nSrabF of \nea f 4 (2) \n\u00a2? . \nThis is a convenient dimensionless form, since the left-hand term must, \nfrom (1), approach unity for \u00ab = 0. (In evaluating this term, consistent \nTABLE I \u2014 DIMENSIONS OF REEDS TESTED\n\nunits must be used: in C.G.S. units, \u00ae is in maxwells, a and b in centi- \nmeters, and F in dynes.) From direct plots of the measured values of \nF and # there were read values of F for & = 10,000 bh (i.e. for a flux \ndensity of 10,000 gauss). For each set of reeds and each overlap, the\n\nlustrated in Fig. 2, which shows the results for h = 10 milli-inches, b \n100 milli-inches and three values of a. As the points for all three values \nof a fall on the same curve, the right-hand side of (2) is independent of \na/b, and reduces to a function of 2/a and h/b only. The experimental \nrelation of Fig. 2 is linear, and similar linear plots, independent of a/b, \nwere obtained for the other three sets of reeds. Thus the functional \nrelation (2) was found to be of the form\n\nwhere k is a function of h/b. The observed values of k for the four sets \nof reeds were plotted against the corresponding values of h/b and a \nlinear plot was obtained, conforming to the equation\n\nof 10,000 gauss. Similar plots were made for other values of density, \nand substantially the same relation was found to apply for densities \nranging from 20 per cent to 80 per cent of the saturation density, which \nwas about 15,000 gauss for the material used in these tests.\n\nWithin this range of reed flux density, therefore, all the observed \nvalues of pull conformed approximately to (3), with k given by (4). \nThese expressions may therefore be used to estimate the pull of mating \nmagnetic reeds at densities up to 80 per cent of saturation, when each \nreed is long compared with the overlap. The expressions are approxi- \nmate, and minor deviations, particularly in the value of k, are produced \nby changes in coil length and in the return path and shielding con- \nfiguration.\n\nA further, and more important, deviation from these relations occurs \nin the closed gap condition, x 0, where the pull is given by (3) as \n\u00a9 /8rab. For release, this pull equals the retractile force of the reeds, \nshown in Fig. 3 as sX. Thus the flux at which release occurs should\n\nvariation, with the upper limit close to this computed value. This cor- \nresponds to the fact that surface irregularities or lack of parallelism of \nthe mating surfaces concentrate the closed gap flux over a smaller area \nthan ab, and increase the pull over that given by (3) for x = 0. The \nactual closed gap pull is therefore variable, and in general higher than\n\nlor operation, the pull curve of the reeds must exceed their stiffness \nload, shown in Fig. 3 as the force s(X \u2014 x). Here X is the open gap \nseparation, and s is the combined stiffness, or half the stiffness of one \nreed if the two are alike. (For unequal reeds, 1/s is the sum of the \ncompliances of the two reeds.) Let Fo\u2019 be the closed gap pull, &\u00b0/82ab, \nfor the minimum value of \u00ae required for operation. Then from (3) the \npull for this value of \u00ae is given by\n\nsX 4CX \nIf CX = 1, the tangency condition no longer applies, the pull and \nload are equal at \u00ab = 0, and Fy\u2019 = sX. \nAfter operation, the flux exceeds the just-operate value. Let Fo be\n\nthe contact force is equal to Fy \u2014 sX, and the retractile force, tending \nto open the contact on release, is sX. The contact behavior varies with \nthe contact force, so that one design requirement is that the contact \nforce should exceed some specified minimum value F,,\n\nA second requirement is that the ratio of retractile force to the contact \nforce should exceed some minimum value n (of the order of unity) to \nminimize the danger of sticking. The area of actual intimate contact, \nand hence the force required to break cold welds over this area, varies \ndirectly with the contact force. In formulating this requirement, the \noperated pull is taken as given by Fy. As stated above, the actual op-\n\nerated pull is variable and in general exceeds Fo. Hence the actual \ncontact force is usually in excess of its computed value, and the actual \nvalue of the retractile force ratio n less than its computed value.\n\nA satisfactory design must meet the requirements for operation, \ncontact force and retractile force through the range of dimensional \nvariation occurring in manufacture. The present discussion is confined \nto the case where only variations in the gap X need be considered. Let \nthe minimum and maximum gap values be X; and X,, and let \u00a2, \nX./X,. The final flux must be below saturation, and for purposes of \nestimation may be assumed to correspond to a density B\u201d, of the order \nof 90 per cent of the saturation density. For operation to occur without \nexcessive reed reluctance, the operating density B\u2019 should be of the \norder of 80 per cent of saturation. Let ec. = Fo/ Fo (B\u201d\u2019/B\u2019)*. Then \nthe operate condition (7) will be satisfied for all values of X if\n\nThe contact force requirement is satisfied for all values of X if \nFo F, + sXo \u2018\n\nThe retractile foree requirement that sX > n( Fo \u2014 sX) is satisfied \nfor all values of X if\n\nrom the preceding three equations, the design requirements can be \nwritten as \n1X, \n(1+ CX.)\n\nThe reed dimensions must be chosen to satisfy these three equations \nfor given values of X,, \u00a2, fe and n (ey being approximately fixed by \nthe reed material used). The dimensions to be selected are a, b, A and \nwhatever other dimension determines s: the free length / if the reeds \nare of uniform cross section. If the section ratio bh is tentatively taken\n\nas fixed by manufacturing considerations, the reed dimensions can be \nevaluated as follows: \nThe value of C is computed from (8), and since a k/C and k is \ngiven by (4), the required value of a is thereby determined. \nSince the value of \u00ae for Fy is bhB\u201d, \n\u2014 \nF, bh B . (11)\n\nThen, with Fo evaluated from (9), and a, B\u201d and b/h known, the thick- \nness h (and hence the width b) can be determined from (11). The stiff- \nness s is given by (10). For a reed of uniform section b & h the length / \ncan be determined from the required value of s. As (11) determines the\n\nrequirements will be that for the largest value of b/h consistent with \nmanufacturing considerations.\n\nThe relations outlined above allow for variations in the gap Y, but \nnot in the other dimensions. In practice, allowance must also be made \nfor variations in the reed thickness A. This can be done by applying the \ntreatment outlined above to the case where A has upper and lower \nlimits, leading to expressions similar to (8), (9) and (10).\n\nUsing the relations given above, reed dimensions have been computed \nto give the three values of minimum contact force F- shown as param- \neters for the curves of Fig. 4, over the range of gap dimensions shown. \nThese illustrative cases were computed for the values of b/h, Xo/X,, n \nand flux density shown in the legend of this figure. The values of flux \ndensity are those applying to iron-nickel alloys of about 50 per cent \nnickel. These have a Young\u2019s modulus value of about 25 X 10\u00b0 psi, \nwhich was taken as applying in computing the length required to meet \nthe stiffness requirement.\n\nThe computed dimensions shown in lig. 4 are the reed thickness h, \nlength /\u2019 and overlap a. The overlap varies inversely with the gap, as \nshown by (8) when bh, n, \u00a2 and c are fixed, as in these computed cases. \nThe thickness and length both increase as the gap and the required \ncontact force are increased. The gap is the major factor controlling the \nlength and hence the over-all size of the sealed contact. The choice of \ngap is fixed in part by voltage breakdown requirements, and in part by \nmanufacturing considerations which determine the variation in gap for \nwhich allowance must be made. The comparisons of lig. 4 are somewhat\n\nthe ratio \u00a2, is usually larger for small gaps than large ones, in which case \nthe reduction in reed thickness and length resulting from reducing the \ngap is less than that shown in Fig. 4.\n\nIn the computations for Fig. 4 the retractile force ratio n is taken as \nunity. This is the computed minimum ratio of retractile force to contact\n\nforce. As discussed previously, the actual contact force may be increased \nby flux concentration in the closed gap, and the actual ratio may be \nmaterially below the computed minimum, which serves therefore as a \nbasis for comparing one design with another, rather than as an absolute \nmeasure of the liability to sticking. From (9), increasing n increases Fy , \nand thus increases the thickness A and the required length /\u2019.\n\nThe reed length is denoted /' in Fig. 4, as that giving the desired stiff- \nness [2s from (10)] for a uniform reed of length /\u2019 and stiffness given \nby Ebh\u00ae/4l\u2019\u00b0. As shown in Fig. 1, convenient construction for the seal \nrequires the use of a cylindrical section extending from the seal for a \nlength cl, with the flat section extending beyond this for a length J. \nThe cylindrical and rectangular sections are of equal area. The stiffness \nof such two section cantilevers are given in Fig. 6-30 of Peek and Wagar,\u2019 \nfrom which can be derived the following expression for the ratio /'/l of \ntwo cantilevers of equal stiffness, one having the dimensions shown in\n\nThus if cl and / are known, /\u2019 can be evaluated and used to determine \nthe stiffness. Conversely, if /\u2019 is determined from the required stiffness, \nas in Fig. 4, the corresponding flat reed length / can be determined for a \ngiven value of cl.\n\nThe flux @ between the reeds required for operation can be estimated \nby means of the relations given in preceding sections. The corresponding \nvalue of coil magnetomotive force 5, or 44rNJ, is given by \u00ae&b, where \n\u00ae is the reluctance of the magnetic circuit. Estimating the sensitivity, \nor the value of \u00a5 for operation, therefore requires some method for \nestimating the value of the reluctance @.\n\nMost reed relays have essentially an air return path, whose reluctance \nis only slightly reduced by the can or shield placed over the core. For \nan air return, the flux through the reeds is analogous to the current \nthrough a leaky transmission line, as discussed in Section 9-2 of Peek \nand Wagar.\u2019 Referring to Fig. 5, let \u00a2 be the flux in the reed at a distance\n\nFig. 5 Notation for analysis of air return magnetie field. \nbetween the point x and the plane of symmetry. The flux dg leaving the \nreed over the length dv is given by \nde \n\u00a5 (13)\n\nwhere p is the permeance per unit length of reed of the air path from x \nto the plane of symmetry. The rate of change in f is given by\n\nlor continuity, the expressions for g and dg/dx at x L,/2 given \nby the two forms of (15) must be equal, from which expressions are \nobtained for the coefficients C and D in terms of A and B. The co- \nefficients A and B in turn are determined by the boundary conditions, \nwhich require that \u00a2 be zero at x I./2 and that for a gap flux \u00a2, \nat \u00ab = 0, the potential f \u2014ggRe/2 atx 0, where \u00ae, is the gap \nreluctance. Then, from (13), the boundary conditions become\n\nAs the boundary condition for x 0 shows, the flux \u00a2 in the reed \ninitially increases with x, and hence the point of maximum flux \u00ae is at \nsome positive value of \u00ab where dg dr 0. Applying this condition to \nthe expression for \u00a2 given by (15), the maximum flux \u00ae is given by\n\nThis maximum flux \u00ae is the total flux between the reeds. The reluc- \ntance \u00ae, or \u00a5 @, is given by gl, \u00ae&. Substituting from the preceding \nequations, this expression for \u00ae may be written in the form\n\nwhere Q RepLs , and r\u00a2,g q\u00a2 is given by (19). Thus the ratio \u00aepLe is a \nfunction of the three ratios Ly, Le, al, and Q, or Rypl. . For the limit-\n\ning case where the reed reluctance per unit length r is so small that a \napproaches zero, the hyperbolic functions in (19) may be replaced by \ntheir series expansions and the higher power in @ neglected. Then for\n\nFig. 6 shows RpLe for the case L,/L. = 0.5 plotted against Reple \nfor various values of al. . The increase of \u00ae with increasing al. measures \nthe effect of the reed reluctance. Aside from the reed reluctance, \u00ae& \ncomprises the air return reluctance in series with the gap reluctance as \nshunted by the parallel air path across the gap. The curves show how \nthis shunt path limits the increase in \u00ae with increasing Rez .\n\nThe effect of increasing L/L. is to increase \u00ae, as can be seen from (21 \nexcept for large values of \u00ae_ (and hence of Q). Thus the reluctance is \nreduced and the sensitivity increased by using a short coil concentrated \nover the gap.\n\nIn deriving the preceding equations, p and r have been taken as \nconstants. The estimation of the air path permeance of bar magnets is \ndiscussed in a classic paper by Evershed, who showed that satisfactory \nestimates can be obtained by using the constant value of p applying to \na uniformly magnetized ellipsoid. This is the function of the ratio L/D \nshown in Fig. 7. In applying this to mating rectangular reeds, L is taken \nas the over-all length L., and D as 2(b + h)/r.\n\nThe assumed constancy of r, the reluctance per unit length, is at best \nan approximation, since r is inversely proportional to the permeability of \nthe material, which varies with the flux density, so that r varies along \nthe length of the reeds. If, however, r is small, the error introduced by \nthis approximation is minor.\n\nThe effect of the permeability on the reluctance and sensitivity, as \nrelated to the size and capability of mating reed contacts, is illustrated \nby the computed results of Table II. Here the dimensions h, b, a and | \nhave been taken as those shown in Fig. 4 for minimum gap values of \nft and 10 milli-inches and contact force values of 2 and 10 grams. The \noperate flux values, as in Fig. 4, are for a density of 12,000 gauss. The \nover-all length L\u00bb. has been taken as 3.75 times the computed reed length. \nThe permeance p is taken from Fig. 7. The critical gap reluctance has \nbeen obtained by adding a 2 milli-inch air gap allowance for the closed \ngap to the critical gap x\u2019 computed from (6).\n\nWith \u00aegpLl, thus computed for each of these two cases, values of als \nhave been computed for two values of permeability u, 1000 and 5000. \nWith r given by 1 wbh, corresponding values of al, have been computed, \nand corresponding values of pl, read from Fig. 6, taking the coil\n\nContact force, Fc (grams) \nOperate flux, & (maxwells) \nLength, L\u00bb. (em) \nPermeance factor, p\n\nlength L, as half of L.. There were thus obtained the values of relue- \ntance \u00ae shown in Table II, and thus finally the operate values, given \n(in abampere-turns) by \u00ae/4r.\n\nIn the range of magnetic reed dimensions and contact and retractile \nforces illustrated by these computations, al, is small and the reed \nreluctance minor for \u00bb of the order of 5000. The materials used for these \nreeds have permeabilities of 5000 and higher at densities of the order \nof 12,000 gauss, or at about 80 per cent of the saturation density. The \npermeability falls rapidly as the density approaches saturation, and is \nof the order of 1000 at about 90 per cent of the saturation density. \nHence a design requiring an operating density over 80 per cent that for \nsaturation may have a materially increased reluctance and consequent \nloss of sensitivity.\n\nAs noted in the discussion of the pull relation, the computed value of \nflux at which release occurs is given by ~/ 8xabsX. Since this is necessarily \na lower value of flux than that for operation, the density is lower and \nthe permeability higher than in operation, so that in the release case the \nreed reluctance is minor, and the preceding equations can be used to \nestimate the reluctance for the closed gap condition. Since flat metallic \npole faces in contact have an effective air gap xo of about 2 milli-inches \n(0.005 em), \u00ae&g for this case is given by xo/ab. Then the abampere-\n\nturns for release are given by \u00ae&&/ 42, where \u00ae and \u00ae are the values \napplying in release. As previously stated, the actual release flux is \nvariable, and in general less than its computed value, and there is there- \nfore a corresponding variation in release sensitivity.\n\nThe analysis of the two preceding sections is useful in providing an \nunderstanding of the field distribution in the reeds and coil and of the \nway in which the leakage field shunting the gap varies with the gap, \nthe flux density, and the reed and coil dimensions. It is, however, limited \nto the case of an air return, and does not apply to a configuration such \nas that shown in Fig. 8, where sleeve members are used to couple the \nreeds to the return path provided by the cover. In such cases, the \nreluctance can be approximately estimated in terms of the lumped con- \nstants of the magnetic circuit shown in the figure. A similar treatment\n\nAs indicated in Fig. 8, the coil magnetomotive force is taken as de- \nveloping an air flux g, , which only affects the inductance, and a reed \nflux @, the maximum or total flux \u00ae of the preceding discussion. The path \nof @ comprises the reed reluctance R, in series with the sleeve gap re- \nluctance 2Rs and the parallel combination of the gap reluctance \u00ae, \nand the shunting leakage reluctance \u00ae,. The path, of course, includes \nthe shielding return members, but these are: negligible in reluctance \ncompared with the rest of the path.\n\nThe reed reluctance can be estimated from the reluctance per unit \nlength r = 1/ubh. The sleeve reluctance can be estimated by an ap- \nproximation obtained from Fig. 9-8 of Peek and Wagar,\u2019 from which\n\nAs shown in the air return case, the leakage reluctance \u00ae, varies with \nthe gap reluctance, and cannot readily be expressed in a simple form. \nAn approximate expression for the parallel combination of \u00ae,g and \u00ae, , \nhowever, can be derived from the experimentally observed pull relation \nof (3). The pull given by this relation must equal that given by the \ngeneral pull equation for a variable reluctance @&,\n\nIf Ry is the parallel combination of Rg and KR, , this expression for \nF'/\u00ae can be equated to that given by (3), and the resulting expression \nfor d\u00aey \u2018dx integrated to give the increase in \u00aey from its value for \u00ab = 0. \nThis zero gap value may be taken as .x\u00bb/ab, with x 0.005 em, as in \nthe Table II estimates of \u00ae, . Thus \u00ae,y is given approximately by\n\nwhere, as before, k is given by (4). \nThen the total reluctance \u00ae for the magnetic circuit of Fig. 8 is given by\n\nwith \u00aey given by (23) and @R\u00bb taken as 2Rs, as given by (22). This \nexpression can be used to estimate the reluctance for other relay con- \nfigurations, using expressions for @\u00bb corresponding to the return path \ncoupling and configuration. It may be applied as an approximation to \nthe air return path case, with \u00ae, taken as the air return reluctance. For \nnegligible reed reluctance, this is given by (21) for \u00ae, and hence Q \nequal to zero, so that \u00ae\u00bb is given by \nSs \np(2L. \u2014 Ly)\n\nWhile these expressions for the reluctance are approximate, they are \nsimple in form, and are convenient to use, in conjunction with operate \nand release flux estimates, in estimating operate and release sensitivity.\n\nThe operate time of magnetic reed contacts is that required for field \ndevelopment and reed motion. These two occur together, since reed\n\nmotion starts as soon as the field and the resultant pull start to develop.\n\nIf the power input is high enough for the field to develop rapidly, the \nreeds may be saturated before any significant motion has occurred. In \nthis case the times of field development \u00a2; and reed motion f. are additive \nin determining the operate time. The value of f: for this case constitutes \na lower limit to the operate time, since no shorter motion time is possible \nthan that for full magnetization of the reeds. An expression for this \nminimum motion time may be obtained by the following approximation.\n\nThe pull curve is given by (3), where @ has its maximum value, which \nmay be taken as about 90 per cent of saturation, as in the capability \nestimates. This pull exceeds the just-operate pull shown in Fig. 3, and \nhence always exceeds the stiffness load s(X\u00a5 \u2014 x). The area between \nthe pull curve and the load line represents the kinetic energy supplied \nto the reeds, which may be denoted 27, so that 7 is the kinetic energy \nsupplied to each of the equal reeds. Writing Fy , as before, for the closed \ngap pull for maximum flux, the kinetic energy is given by\n\nWhile the acceleration is variable, the motion time is nearly the same \nas that for uniform acceleration, for which the kinetic energy 7 for each\n\n27 \nwhere T is given by (26). To illustrate the magnitude of the quantities \ninvolved for mating magnetic reeds, Table IIT gives the computed values \nof t, for the two cases of Fig. 4 in which the minimum gaps are 4 and 10\n\ngap is taken as having its maximum value X\u00bb2, and Fy as corresponding \nto a density of 13,500 gauss, as before. As the reeds deflect as cantilever \nbeams, the effective mass m of each reed is taken as one quarter of its \nactual mass, assuming a density of 8.2 grams/cm.\u00b0 The values of C, or \nk/a, are those applying for the dimensions given in Fig. 4 for these two \ncases.\n\nThese estimates show that for the range of reed and gap dimensions \ncovered in Fig. 4, the motion times lie in a relatively narrow band. These \ntimes are small compared with those of most other devices which open \nand close contacts in a metallic circuit: they are large compared with \nthe switching times of many electronic devices.\n\nTo estimate the total operate time, an expression is required for the \ntime of flux development. As shown in Chapter 4 of Peek and Wagar,\u2019 \nan adequate approximation can be obtained from the exponential rela- \ntion for flux development with the constant reluctance (and hence \ninductance ) for the initial open gap condition. For magnetic reeds, eddy \ncurrent effects are negligible except for flux development so fast that 4 \nis negligible compared with &. The flux and coil current develop to- \ngether with a time constant L/R. The time \u00a2; at which the ampere- \nturns equal the just operate value (VJ )o is given by\n\nwhere v (NI)o/ (NJ) and L\u2019 is the single-turn inductance, or 42, @, \nwhere \u00ae is the open gap reluctance for the average flux linked by the\n\ncoil. If both numerator and denominator in this expression are multi- \nplied by J\u2019, NJ is written as (NJ )o/v, and t: added to \u00a2; to give the total \noperate time \u00a2, the resulting expression is\n\nAs shown in the text reference cited above, the function of v appearing \nin brackets has a minimum value of 2.5 for v = 0.715, and departs from \nthis minimum, by less than 5 per cent for 0.6 < v < 0.8. Thus for mini- \nmum operate time, (VJ) ) NJ should lie in this range, and (28) re- \nduces to\n\nwhere nL\u201d is written for L\u2019. If there are n sealed contacts (pairs of \nmating reeds) used in a single relay coil, the single-turn inductance is \nnL\u201d, where L\u201d is the single turn inductance of one sealed contact.\n\nIf the total time \u00a2 is materially larger than \u00a2, , the two terms of (29) \nare not strictly additive, but the approximation has been found to give \nadequate agreement with experimental observations.\n\nIn (28), the term L\u2019(NJ),\u00a2 is proportional to @(N/J)o (where \u00ae is \nthe operate flux) and hence to the field energy required for operation. \nThus the electrical energy input for operation, which is proportional\n\nto J*Rt, , is proportional to this field energy. As previously noted, (28) \nalso shows that f; varies with v, or /)/7, and that the time is a minimum \nfor a given power input if the coil circuit is designed to give Jo/J a \nvalue of about 0.7. It also follows from (29) that increasing the steady-\n\nstate power reduces the operate time until it approaches the lower \nlimit tf , the minimum time for reed motion.\n\nTo use (29) for preliminary estimates requires estimation of the \noperate ampere-turns (NJ), and the single-turn inductance nL\u201d. (NJ)o \ncan be estimated by the procedures described in the sections on relue- \ntance estimates. To estimate the single-turn inductance requires estimat- \ning the reluctance for the average field linked by the coil. In general, \nthis is a larger reluctance than that for the maximum reed flux, to which \nthe expressions given in preceding sections apply. The maximum flux is \nlimited by reed saturation, and hence controls the capability.\n\nFor a tightly coupled magnetic circuit, as in Fig. 8, the maximum \nand average flux are nearly the same, and the same reluctance expres- \nsions apply, except that some allowance may be made for the air field\n\nan air path return relay, the average flux linked by the coil is materially \nless than the maximum value. The analysis of this case given above can \nbe used to give expressions for the average flux, but these are complex \nin form, and are not included here.\n\nThe agreement of experimental observations with (28) and (29) is \nillustrated in Figs. 9 and 10. Fig. 9 shows direct measurements of operate \ntime plotted against the coil cireuit constant N\u00b0/R for various values \nof steady state power input. The dotted curves are for constant values \nof ampere-turns NJ. [(NJ)\u00b0 = ['R-(N*/R).| The sealed contacts in this \nrelay had a value of (NJ)o of about 100 ampere-turns, for which the \nminimum times should occur for a value of NJ of about 140, which is \nnear that for the observed minima.\n\nThe minimum times for this two-switch assembly are plotted against \nn FR in Fig. 10, together with those for three other cases with one, \nthree and four switches. In agreement with (29) the plots are linear, \nand have a common intercept &, the minimum motion time. These\n\ntimes were measured to the end of the short chatter interval following \ninitial closure, so that the indicated value of ft includes this chatter \ninterval. The slopes of the lines measure the relative value of the single\n\nto the considerable air core inductance in this coil, which had space for \nfour switches (sealed contacts). This results in a decrease in the ap- \nparent value of L\u201d as the number of switches increases.\n\nThe analysis given in this paper is intended for use in development \nstudies of sealed magnetic reed contacts. It can be used to estimate \nthe gap and reed dimensions needed to meet design objectives with\n\nsensitivity. In development studies, models based on the initial estimates \ncan be made and measured, and the measurement results used in con- \njunction with the analysis as a guide in further work.\n\nIn the performance of mating magnetic reeds, the major controlling \nquantity is the flux between the reeds. This flux, which determines the \npull, is limited by saturation and hence by the reed cross section. This \ntotal reed flux comprises the gap flux proper and the leakage flux shunt- \ning the gap. As shown in the general analysis of the air return reluctance, \nthis leakage field varies with the gap, reed and coil dimensions, and the \nrelation between the pull and the total reed flux has a corresponding \nvariation. Hence the simple pull equation derived from experiment ap- \nplies rigorously only for the experimental coil length and return path \nconditions used in deriving it, and minor deviations oceur in applying \nit to other cases.\n\nCorresponding deviations occur for the reluctance given by the ex- \npression derived from the pull equation, and for the other approximate \nreluctance equations. All these approximations, however, are adequate \nfor preliminary estimates, and may be applied with greater accuracy in \ndetailed studies by using values of the constants that are experimentally \ndetermined for the case in question. While the general analysis can be \nused to estimate the air return reluctance, and the treatment can be \nextended to give expressions for the pull and the inductance, it is limited \nto the air return case, and is too complex for convenient application. \nIt serves, however, to show the character and extent of the deviations \nfrom the simpler approximations resulting from the distributed char- \nacter of the leakage field.\n\nAs the times of flux development and reed motion depend respectively \non the inductance corresponding to the coil flux linkages and on the \npull for maximum flux, timing estimates based on the simple approxima- \ntions given here are also subject to some minor variation. This can be \nreduced in development studies by using experimentally determined \nvalues of the constants.\n\nsize, but are approximately applicable to unequal reeds provided both \nare long compared with the gap overlap. In such cases the effective \nstiffness of the reed combination must be taken as the reciprocal of the \nsum of the compliances of the two reeds.\n\nJ. Jones, using apparatus designed and constructed by J. L. Agterberg. \nThe author is also indebted to W. B. Ellwood, O. M. Hovgaard and \nother associates for information used in preparing this article.\n\n. Ellwood, W. B., U.S. Patent 2,289,830. \nEllwood, W. B., Glass-Enclosed Reed Relay, Elect. Engg., 66, 1947, p. 1104. \n3. Brown, J. T. L. and Pollard, C. E., Mercury Contact Relays, Elect. Engg., 66, \n1947, p. 1106. \nHovgaard, O. M., Capability of Sealed Relay Contacts, A.I.E.E. Trans., Pt. \nI, 75, 1956, p. 466 \nHovgaard, O. M. and Perreault, G. E., Development of Reed Switches and \nRelays, B.S.T.J., 34, 1955, p. 309. \nHovgaard, O. M. and Fontana, W. J., A Versatile Miniature Switching Capsule, \nProc. 1959 Electronic Components Conf., p. 32. \nPeek, R. L., Jr. and Wagar, H. N., Switching Relay Design, D. Van Nostrand \nCo., New York, 1955. \n. Evershed, 8., Permanent Magnets in Theory and Practice, J.I.E.E., 68, 1919, \np. 780.\n\nThe problem of protecting apparatus against lightning surges from con- \nnected transmission facilities has become more complex with the use of \nsolid state devices in apparatus design. Consideration of the protection re- \nquirements for such apparatus has indicated that existing information con- \ncerning the incidence and characteristics of lightning surges is insufficient \nto develop optimum protection measures. A recently completed field investi- \ngation provides additional information in this specific area.\n\nThe results of this field investigation and supplemental laboratory surge \ntests indicate that, in well-shielded underground cable pairs, electrical \nsurges do not exceed approximately 90 volts peak, and that transistorized \napparatus capable of withstanding such surge amplitudes needs no further \nprotection. In aerial and buried cable, however, transistorized apparatus \nrequires protection up to the full sparkover potential of 3-mil protector gaps, \nt.e., to about 600 volts peak. A firm basis for testing and evaluating tran- \nsistorized apparatus from the lightning surge voltage standpoint is presented.\n\nThe 3-mil air gap carbon block protector, which has a maximum spark- \nover value of 600 peak volts, is the basic protection device employed by \nthe Bell System for the protection of communication apparatus against \nextraneous potentials. Prior to the introduction of transistors and mini- \naturized circuitry, it was the general practice in apparatus design to \nprovide a withstand level for both metallic* and longitudinal? potentials \ngreater than 600 peak volts so as to coordinate directly with 3-mil \nprotector gaps. This customary design objective of providing an in- \nherent withstand capability exceeding the operating value of standard \nprotector gaps is not always feasible in the case of apparatus employing\n\nsolid state devices. Furthermore, lower voltage protection cannot be \nattained satisfactorily by simply reducing protector-gap spacing below \n3 mils, since excessive protector maintenance would be introduced. To \nmeet the lower voltage requirements of apparatus employing transistors, \ntherefore, it is necessary at present either to modify the circuitry so \nthat surge currents appearing in the more susceptible components are \nlimited in magnitude, or to introduce an additional stage of protection \nemploying semiconductor diodes supplementing the gaps. These pro- \ntection measures may introduce significant additional expense and, in \nsome cases, produce adverse effects on transmission characteristics.\n\nIt became apparent that selection of optimum protection measures \nto meet the exacting requirements of transistorized apparatus neces- \nsitated a more complete knowledge of the incidence and characteristics \nof lightning surges in the range below the operating level of standard \nprotector gaps. Recognizing this, a field investigation was undertaken \nto supplement existing information in this area. The results of this \nrecently completed field study and conclusions drawn from analysis of \nthe data are presented in this article.\n\nSince it appeared, at the time the investigation was undertaken, that \nthe present practice of employing 3-mil protector gaps as basic apparatus \nprotection would continue into the foreseeable future, all circuits used \nfor purposes of observation were equipped with such protectors. The \narea of study therefore was intentionally restricted to surges up to about \n600 peak volts as limited by protector operation.\n\nObservations of lightning surges appearing in trunk pairs in aerial \nand buried cable were recorded by means of automatic cathode ray \noscillographs. The plant locations selected were in areas known to ex- \nperience heavy thunderstorm activity. Surges were also monitored by \nmeans of peak amplitude recording devices in urban underground cables. \nBecause of the shielding provided by buildings and buried piping facili- \nties, the exposure of cables to lightning in this situation was relatively \nlow.\n\nInformation of engineering value secured includes the probability \ndistribution of voltage magnitudes and the rise and decay time char- \nacteristics of surges in the lower voltage range specifically under study. \nSuch data have been used as a basis for selecting waveshapes suitable \nfor laboratory testing of the energy and power handling capabilities of \ntransistorized apparatus.\n\nField data on lightning surge characteristics were obtained from types\n\n1. Low exposure, typified by underground plant in well-shielded \nurban areas.\n\n2. High exposure, typified by aerial and buried cables in suburban \nand rural areas.\n\nThe study of surge activity in underground cable was conducted on \nspare trunk pairs in Baltimore, Maryland; Pontiac, Michigan; and South \nOrange, New Jersey. Table I gives a brief description of the route and \nmake-up of these facilities.\n\nThe waveshapes of lightning surges in aerial and buried cable were \nstudied with cathode-ray oscillographs arranged to monitor continuously \nthe pair selected for observation and to record automatically on 16-mm\n\nmeasurements were made of open-circuit longitudinal surge voltages and \nany resultant metallic voltages appearing across a representative resis- \ntive termination. Spare H88-loaded trunk pairs in aerial cable were\n\nTABLE I DESCRIPTION OF UNDERGROUND TRUNKS MONITORED \nWITH SURGE COUNTERS\n\nSouth Orange to West Orange, 100-pair underground cable with \nN. J. aerial complements \nSouth Orange to Summit, N. J. 155-pair 100\u00b0) underground cable \nwith H8S8 loading\n\nPontiae to Birmingham, Mich. Underground cable with aerial \ncomplements and FI88 loading\n\nBaltimore to Towson, Md. 900-pair underground cable with \naerial complements and H88& \nloading\n\nTEST \n\u2018 LOCATION \u2122 5 MILES > peak eed \n(a) CENTRAL ENTRA \nOFFICE 22 GAGE, H88 LOADED, LEAD-SHEATHED AERIAL CABLE \n75 TRUNK PAIRS\n\nTEST \nUNDERGROUND | LOCATION \n, fo \"e: pain CENTRAL. \n51 PAIRS \n2 MILES OFFICE \nFAR END w\u00ab\u00ab 10.5 MILES > y \nCENTRAL \nOFFICE 100 PAIR AERIAL PIC-ALPETH CABLE \n75% OF CABLE iS OVERBUILT BY ONE \nCROSSARM OF OPEN WIRE ] \n5) PAIRS \nATLANTA\n\nCENTRAL \nOFFICE 19 GAGE, JUTE-PROTECTED, LEAD-SHEATHED, \nBURIED QUADDED CABLE \n(c) 27 QUADS\n\nFig. 1 Characteristics of test cable at (a) Mt. Freedom, N. J., (b) Buford, \nGa., (\u00a2) Griffin, Ga.; (d) arrangement of measuring equipment at test locations\n\nstudied in Buford, Georgia, and Mt. Freedom, New Jersey. The buried \ncable studies were conducted in Griffin, Georgia, on a spare nonloaded\n\nSupplemental laboratory surge tests were also conducted on a one- \nmile test cable to augment the information on the behavior of under- \nground cables with aerial complements and extensions.\n\nThe incidence and characteristics of the lightning surges recorded and \nthe resulting protection. considerations will be discussed in the order of \nthe degree of plant exposure.\n\nThe field study of surge activity in well-shielded urban underground \nplant, covering the first two categories, revealed that in no case did \nvoltages attain the 90-volt triggering value of the lowest stage of the \ncounters. During the five-month observation period, a total of 44 thun- \nderstorm days was reported by the U.S. Weather Bureau for these \nareas. Several of these storms were known to be quite severe, with \ntheir centers located over the test cables. The counters were tested \nperiodically during the study period to ensure proper operation. The \nlack of surge activity recorded during this study reveals the shielding \nbenefits enjoyed by urban underground plant. In such areas, buildings \nand buried metallic piping systems divert and dissipate lightning strokes \nthat otherwise might directly involve the telephone cables. Furthermore, \nduct runs usually contain two or more cables, the sheaths of which are \nbonded at each manhole. Surge current will, therefore, divide between \nthe cables, and the voltage induced in any one cable will be propor- \ntionately reduced.\n\nOn the basis of these field studies, it appears that apparatus capable \nof withstanding surges in the order of 90 to 100 volts peak will not \nrequire lightning protection when associated with all-underground cable \npairs (trunk or loops), whether such pairs are in an all-underground \ncable or one with well shielded aerial complements. The significant point\n\nis that this conclusion holds only for well-shielded plant, the shielding \nbeing provided by closely spaced buildings, extensive power distribution \nand buried metallic piping systems, and other telephone cables in the \nsame conduit run.\n\nThe question now arises as to the magnitude of surges that may ap- \npear in a 100 per cent underground trunk complement through coupling \nwith underground pairs extended aerially in cable having greater ex-\n\nInformation on the following specific points relative to such surge \ncoupling is useful in estimating the secondary exposure likely to be \nexperienced by apparatus connected to the 100 per cent underground \ntrunk pairs:\n\n1. The ratio between an open-circuit longitudinal voltage surge on a \ndisturbing circuit (underground pair associated with an aerial ex- \ntension) and the resultant voltage appearing in a disturbed circuit (100 \nper cent underground trunk pair).\n\n2. The resultant magnitude and waveshape of longitudinal current \nappearing in the disturbed circuit.\n\nTo secure this information, supplemental laboratory surge tests were \nconducted on a 50-pair, one-mile test cable. Longitudinal impulses \n(approximately 50 by 250 microseconds*) were applied to one or more \ncable pairs acting as the disturbing circuit. Several cases were investi- \ngated: energizing a single pair, then 5 pairs in parallel and finally 25 \npairs in parallel. Measurements were made of longitudinal open-circuit \nvoltage and short-circuit current in the disturbing circuit. Measurements \nwere then made of longitudinal open-circuit voltage and short-circuit \ncurrent in a disturbed pair. Pairs both adjacent and remote from the \ndisturbing circuit were investigated.\n\nThe ratio of longitudinal open-circuit voltage appearing in a disturbed \npair to the longitudinal open-circuit voltage in the disturbing circuit \nvaried, depending on test conditions, from about 0.47 to 0.85. The lower \nvalue was obtained when only a single pair was energized and the larger \nvalue when the surge was applied to 25 pairs paralleled at the generator \nend. Although the magnitude of the open-circuit surge voltage appear- \ning in the disturbed circuit was significantly lower than that in the \ndisturbing circuit, the waveshapes of the two were essentially the same. \nIn the cases investigated it was found that the longitudinal short-circuit \ncurrent in the disturbed pair was approximately 3 milliamperes per volt \nappearing in the disturbing circuit. The short-circuit current in the\n\n* That is, 50 microseconds rise time to crest and 250 microseconds from origin \nto point where wave has decayed to half of crest value\n\ndisturbed pair assumed the shape of a square pulse about 20 micro- \nseconds in duration. From a protection standpoint, this radical reduction \nin the duration of the induced current wave in the disturbed circuit is \nprobably the most significant bit of information secured in these labora-\n\nsurges, usually fail as a result of overheating of their junction or junc- \ntions \u2014 the heating effect being related to the magnitude and duration \nof the junction current. Therefore, the possibility of failure from longi- \ntudinal current of semiconductor components in apparatus associated \nwith a disturbed pair is much reduced because of the relatively short \nduration of the coupled surge.\n\nIt is of further interest to note the effect of grounding the disturbing \ncircuit at the far end, such as would occur with protector operation. \nThis condition was investigated by grounding one of the conductors \nconstituting the disturbing circuit, and it was found that the open- \ncircuit longitudinal voltage on the disturbed pair was reduced approxi- \nmately 50 per cent and the short circuit longitudinal current 30 per cent \nbelow the values that would obtain if the conductor of the disturbing \ncircuit had not been grounded. This indicates the order of beneficial \nshielding enjoyed by the disturbed circuit when protector blocks operate \non the disturbing circuit.\n\nAdditional laboratory surge studies were made employing a one-mile \ntest cable to determine the protection requirements for apparatus con- \nnected to underground pairs extended aerially. These tests revealed \nthat surges having typical waveshapes will propagate longitudinally on \na cable pair for one mile with little attenuation. Therefore, underground \ncable pairs extended aerially or in buried plant should be considered as \nexposed to lightning, unless protection is applied at the underground \njunction to limit surges from the exposed extensions.\n\nThe lightning exposure of buried and aerial cable is sufficiently severe \nte require supplementary protection for transistorized apparatus as- \nsociated with this type of plant. The most useful way of defining pro- \ntection requirements in this situation is in terms of a simulated lightning \nsurge which such apparatus must withstand in laboratory tests. Simu- \nlated surges, of course, must be based on surge conditions in the field. \nConsequently, derivation of suitable test surges requires knowledge of \nthe distribution of waveshapes and peak voltages of lightning surges \nappearing in the plant.\n\nThe data on longitudinal surges in aerial cable at Buford covered a \nperiod of six months, during which time six thunderstorm days occurred \nand 103 oscillograms were obtained. Additional data on aerial cable \nwere secured at Mt. Freedom, with 105 oscillograms being recorded \nduring three thunderstorm days. The buried cable studies conducted at \nGriffin covered a period of six months, during which 36 thunderstorm \ndays occurred and 1120 oscillograms of longitudinal surges were ob- \ntained. Since these cables involved very little unexposed plant, the \nsurge magnitudes and waveshapes recorded at the central offices should \nalso be reasonably representative of surge conditions along the cable \nroute.\n\nComparison of the recorded waveshapes of longitudinal surges in- \nduced in aerial and buried cables indicates that the two types of plant \ndo not differ significantly in their response to lightning surges. The data \nalso confirm that load coils have little or no effect on the waveshape of \nlongitudinal surges. This observation is based on the similarity of the \nsurges recorded on H88-loaded aerial cable and those recorded on non- \nloaded buried cable. These surges were found to be essentially impulses, \nas exemplified in Fig. 2. Longitudinal impulses can conveniently be \ncharacterized on the conventional basis of crest magnitude, time to crest \nand time from origin to the point at which the wave has decayed to \none-half of crest value.\n\nBoth the rise time to crest and decay time to half-crest value of \nlightning surges observed in cable exhibited log-normal distributions.\n\nfollow an exponential distribution similar to the lightning stroke cur- \nrents that produce these voltages.\n\nDistribution of surge rise times for the three test locations are pre- \nsented in Fig. 3. The median rise time measured at each location was \napproximately 100 microseconds, but the dispersion about the median \nvalue varied widely among locations, probably due to the difference in \ncable lengths involved. As a surge propagates along a cable pair, there \nis some modification in waveshape. Consequently, slightly greater dis- \npersion should be expected in longer cables. This is borne out by the \nmeasurements plotted in Fig. 3.\n\nThe peak voltages recorded in aerial and buried cable in the range \nbelow protector block operation (voltages less than 400 volts) were \nfound to be exponentially distributed. This distribution provides a basis \nfor determining the probability of any given surge exceeding a particular \nvalue. The derivations of these probability functions for aerial and\n\nburied cable are given in the Appendix. These probability functions, \nused in conjunction with the average number of surges induced in the \ntest cables per thunderstorm day, provide an order-of-magnitude esti- \nmate of the number of surges per thunderstorm day exceeding any given \namplitude. The similarity between the peak voltage distributions for \naerial and buried cable makes it feasible to develop a single plot of the \nestimated number of surges per thunderstorm day as a function of \nvoltage.\n\nFig. 4 presents both the distribution that would be expected if no \nprotector blocks were associated with the test pair and the modified \ndistribution reflecting the operating characteristics of standard 3-mil \nair gap carbon protectors. This information is useful in the design of \nreliability tests for apparatus vulnerable to repeated low-amplitude \nvoltage surges, and is used in the selection of suitable test surges, as \ndiscussed later.\n\nB S | \nMT. FREEDOM oe OA \\ GRIFFIN \nS5-MILE AERIAL CABLE } 8-MILE BURIED CABLE \n3-SIGMA LIMITS ~._ |__| 3-SIGMA LIMITS \n70 {\u00a3LSEC ;2900 KLSEC \u2018 : = 69 SEC ;2300 \u201cSEC\n\nThe distributions of decay times of lightning surges appearing in \ncable plant are presented in Fig. 5. The median values and dispersions \ndiffered slightly among the three test locations. Median values ranged \nfrom 350 to 450 microseconds. Three-sigma limits ranged between 40 \nto 70 microseconds on the low side and between 2300 and 2900 micro- \nseconds on the high side. The variation of median values is possibly due \nto differences in the relative exposure of the test cables.\n\nIn Buford and Griffin, the cables traverse relatively level terrain and, \ntherefore, should have fairly uniform exposure along their lengths. Close \ncorrelation is noted between the median decay time values for these \ncables. In Mt. Freedom, however, the cable is placed on hilly terrain, \nwhich results in a higher exposure for some sections of the cable. With \nthe majority of the surges being induced into one section of the cable, a \nshift in the median value results.\n\nMetallic surges were recorded on the same trunk cable pairs used at \nBuford and Griffin to record longitudinal surges. Measurements were \nmade across a 940-ohm resistive termination. Simultaneous operation\n\nof the automatic recording oscillographs observing longitudinal and \nmetallic voltages was arranged in order to permit correlation between \neach longitudinal surge and any resultant metallic surge.\n\nIn the absence of conductor insulation breakdown or protector opera- \ntion, the circuit balance of cable pairs is sufficiently good that metallic \nsurge voltages should be much lower than longitudinal surge levels, \nwhich is confirmed by the measurements secured in this study. Of all \nlongitudinal surges exceeding 60 volts* only about 10 per cent produced \nmetallic voltages exceeding 10 volts peak. However, when protector \noperation occurred, metallic surge potentials of significant magnitudes \nwere produced. An approximate breakdown of those metallic surges \nwhich exceeded 10 volts peak is presented below on the basis of wave- \nshape and magnitude:\n\n1. Twenty per cent were low-amplitude, high-frequency oscillations \nhaving maximum peaks of about 35 volts and frequency components \nranging from 10 ke to approximately 50 ke.\n\n2. Twenty-five per cent were impulses ranging in amplitude up to \nabout 60 volts peak, which were probably caused by protector operation \non adjacent pairs.\n\n3. Fifteen per cent ranged in amplitude from 120 to 200 volts peak. \nThe oscillograms of the associated longitudinal surges definitely indi- \ncated protector operation on the test pair although the surge amplitudes\n\nwere considerably lower than the level normally required for protector\n\nblock operation. The protector blocks on the test pair were changed peri- \nodically during the study period, but these low-voltage operations (16 in \nall) occurred with the same set of blocks. However, visual inspection of \nthese protector blocks by local personnel did not reveal anything un- \nusual in their appearance.\n\n4. Forty per cent of the metallic surges exceeded 350 volts peak. \nThese surges were all associated with protector operation on the test \npair.\n\nExamination of the oscillograms of these high-amplitude metallic \nsurges and the corresponding longitudinal surges illustrates some interest- \ning aspects of protector block operation. Fig. 6(a) shows one type of \nsingle-block operation where the discharge is continuous for the duration \nof the surge. Fig. 6(b) illustrates another type of single-block operation \nin which clearing and restriking of the are discharge occurs. This situa- \ntion is the result of circuit regulation when the longitudinal surge po- \ntential is just sufficient to initiate gap sparkover.\n\n* Threshold value of automatic recording oscillographs measuring longitudinal \nsurges\n\nFig. 6 Lightning impulse voltages on a nonloaded buried cable pair: (a) \nwith protector operation on one conductor; (b) with multiple protector block \noperation on one conductor.\n\nA brief explanation of surge voltage relationships during single-block \noperation as indicated by the oscillograms follows:\n\nAt point a [Fig. 6(a)], the gap associated with one conductor operated \nand remained operated for the duration of the surge. The longitudinal \nvoltage dropped to about one-half of peak value as the oscilloscope is\n\n560 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1961 \nconnected to the midpoint of a 940-ohm termination (see Fig. 1).* \nThe metallic surge shows the true voltage unbalance, which is the dif- \nference between the are discharge voltage on one conductor (approxi- \nmately 50 volts) and the longitudinal surge voltage on the other con- \nductor.\n\nAt the first point a [Fig. 6(b)], the gap associated with one conductor \noperated and remained operated for approximately 350 microseconds. \nThe metallic surge began at point A and continued for the same dura- \ntion. At the second point a the protector gap \u2018\u2018cleared\u201d and then operated \nagain. When it cleared, the longitudinal scope recorded the true longi- \ntudinal voltage* and the metallic voltage dropped to zero. When the \nblock restruck, the longitudinal scope again read one-half true longi- \ntudinal voltage and once more there was a metallic voltage.\n\nMetallic surges resulting from operation of both blocks on a pair are \nshown in Figs. 7(a) and 7(b). In the first case, one block operated \ninitially, then cleared, and then the other block operated. This sequence \nof operation produced a metallic surge having impulses of both polarities. \nIn the second case, both blocks operated, but unbalances were produced \nby nonsimultaneous clearing and restriking of the two gaps.\n\nFurther explanation of surge voltage relationships resulting from a \nlongitudinal surge of sufficient magnitude to operate both blocks is as \nfollows:\n\nAt the first point a [Fig. 7(a)], the protector on one conductor op- \nerated, in this case, the one on the \u201ctip\u201d conductor. The recorded \nlongitudinal voltage then dropped to one-half of the actual value and \nthe metallic surge began. At the second point a, the protector on the \n\u201cring\u201d conductor operated and the longitudinal surge voltage dropped \nto essentially zero. At the third point a, the protector on the \u201ctip\u201d \nconductor cleared. This is evidenced by the reappearance of metallic \nvoltage of reversed polarity.\n\nFig. 7(b) illustrates a phenomenon that commonly occurs during the \noperation of protector gaps on telephone circuits. The discharge is not a \ncontinuous process but is punctuated by a random restriking of the are \ndischarge. In this case there was sufficient longitudinal potential to \noperate both blocks, but apparently some differences in electrode con- \nditions caused nonsimultaneous clearing and restriking of the are dis- \ncharge which produced metallic potentials. It is only during the brief\n\n* In effect, the longitudinal scope always reads the true open circuit voltage \nuntil one block operates; then it reads one-half the true longitudinal voltage. The \nmetallic oscilloscope always reads the total voltage difference between the two \nconductors of the pairs.\n\nFig. 7 Lightning impulse voltages on a nonloaded buried cable pair: (a) with \nprotector block operation first on one conductor, then on the other; (b) with \nmultiple striking and restriking of protector blocks.\n\nperiods when the two gaps are either both discharging or not discharging \nthat metallic potentials are reduced to negligible values. \nIV. SELECTION OF SUITABLE WAVEFORMS FOR LABORATORY TESTING\n\nThe similarity of longitudinal waveforms observed in buried and aerial \nplant under loaded and nonloaded conditions makes it practicable to\n\nemploy common waveforms for laboratory testing of apparatus intended\n\nfor use with all of these types of plant. In the past, a 10 by 600-micro- \nsecond surge was selected on the basis of limited field data. The supple- \nmental surge data obtained in this study makes it possible to select \nwaveshapes for laboratory testing which more closely simulate surges \nproduced in the telephone plant by natural lightning.\n\nAnalysis of the recorded data obtained during this study provided \nthe distributions of rise times, peak voltages and decay times presented \nin Figs. 3, 4 and 5. Using these distributions as a basis, it is possible to \nselect suitable laboratory test surges.\n\nSince the severity of a surge is dependent on its peak voltage and its \ndecay time, it is necessary to establish whether decay time is independent \nof peak voltage before computing the joint probability of occurrence of \na surge with a given amplitude and decay time. Accordingly, new decay \ntime distributions were developed from the recorded data for two voltage \nranges; voltages below 225 volts and voltages above 225 volts. Fig. 8 \npresents the results of this analysis. It will be observed that there is a \nslight correlation between the decay time and voltage. This is the result \nof the manner in which the surge current in the sheath induces voltage \nonto the cable pairs. The capacitive coupling between the sheath and \nthe core results in an integration of the sheath current. Thus, surges of \nlonger duration will tend to produce higher voltages on the cable pairs. \nIn determining the joint probability of exceeding a given amplitude and \na given decay time, this correlation, as indicated by the field data, \nshould be included by using the probability distribution of decay times \nassociated with higher voltages (upper curve in Fig. 8).\n\nThe parameters of laboratory test surges should be so selected as to \nevaluate apparatus properly for its dielectric strength and its energy \nand power handling capabilities. The effect of these factors on the \nparameters of test surges is discussed below.\n\nSurges suitable for test purposes should have a peak amplitude of at \nleast 600 volts to provide a minimum test of apparatus dielectric strength, \nsince 3-mil protector gaps associated with apparatus assure protection \nonly against surges in excess of this value. Furthermore, when the\n\nmore detrimental than energy content and, given two surges of equal \ncontent, more power will be delivered by the surge with the higher \namplitude.\n\nThe energy content of a surge is dependent on its peak voltage and \nwaveshape (i.e., rise time and decay time). Test surges should have \nshort rise times for two reasons. A short rise time will provide a more \nsevere test of inductive circuit elements and, for a given decay time, the \nshorter the rise the higher will be the total energy of the surge. Ac-\n\nmately the lower 3-sigma limit of the rise times recorded in the field. \nIn those instances where the energy-handling capability of apparatus \nis the controlling factor, the reliability of a surge testing program will \ndepend on the degree of assurance that the energy content of the test \nsurge will not be exceeded in the field. The total energy of an impulse \nwith a short rise time is proportional to the decay time and the square \nof the voltage. The energy content of any arbitrarily assumed test surge \ncan be exceeded in the field in two ways: surges of lower amplitudes but \nappreciably longer decay times, or surges of higher amplitude and only \nsomewhat shorter decay times. This second classification can be elimi- \nnated, however, by selecting test surges having peak amplitudes of 600 \nvolts or greater, as the standard 3-mil air gap protectors associated with \nequipment will not permit surges in excess of 600 peak volts. To deter-\n\nmine the probability of exceeding the energy of a particular test surge, \nthe joint probability of two factors must be calculated: (a) the prob- \nability of obtaining a particular voltage and (b) the probability that a \nsurge of this voltage has a sufficiently long decay time such that its \ntotal energy exceeds the energy content of the test surge. These joint \nprobabilities must be summed for all surges with realizable combinations \nof voltages and decay times which exceed the energy of the test surge.\n\nMathematically the procedure is as follows: The energy content of a \nlightning impulse* is:\n\nAssuming a test surge of amplitude V\u00bb and decay time dy, the prob- \nability [P|] that a surge of somewhat lower amplitude V, will have\n\nImpulse = rapid rise to crest implitude followed by a slow exponential fall\n\nhaving greater energy content than the test surge requires the summa- \ntion of [P] for all possible values of V. These probabilities need be \nsummed for voltages only above 400 volts, since lower voltages would re- \nquire associated decay times far longer than observed or expected in \ncable plant. This follows for two reasons: In order for a small amplitude \nsurge to have an energy content equal to that of the test surge, its decay \ntime must increase as the square of the voltage ratio of these surges. \nThis would require existence of decay times much longer than 2,000 \nmicroseconds, a condition generally contrary to all test observations, \nand contrary to the observed correlation between peak voltages and \ndecay times which indicated that lower voltage surges have smaller \ndecay times.\n\nTherefore, the total probability (27) of exceeding the energy of a \nparticular test surge is:\n\n[P(590 \u2014 600)|[P(d = d,)] A, \n( Pr) 2 \nThe probability of obtaining a voltage in the range Av may be deter- \nmined from Fig. 4. For example:\n\nThe probability of obtaining a decay time greater than d; may be \nread directly from the upper curve of Fig. 8.\n\nThis type of calculation must be repeated for each test surge desired. \nIn this study, test surges with peak amplitudes of 600, 700, 800 volts \nand a number of different decay times were examined. By providing \nseveral waveshapes with equivalent energy content: (a) a better in- \ndication may be obtained as to the adequacy of the thermal time con- \nstants of vulnerable apparatus components, (b) more latitude is provided \nfor determining the dielectric strength margin of apparatus, and (c) \nflexibility is provided in the selection of laboratory surge forming circuit \nconstants. The above calculation indicates the probability, per thunder- \nstorm, of exceeding the energy content of a particular test surge. For \nengineering purposes, however, it is more practical to develop curves \nof apparatus energy-handling capabilities versus trouble expectancy in \nyears rather than in thunderstorm days. Reference to isoceraunic charts \nindicates thunderstorm incidences varying from 5 to 90 thunderstorms \nper year for various sections of the country, with the higher incidences \nin the Southeast. However, even when a thunderstorm is reported in \nthe general area of a particular cable it is not necessarily close enough \nto induce surges having magnitudes of the order under discussion. It is \nfelt that an average of only 25 thunderstorm days per year in the higher \nstorm incidence areas are likely to produce significant surges in the cable \nplant. This factor, together with the calculated probability per thunder- \nstorm of exceeding various energy levels, provides the basis for the curves \npresented in Fig. 9. These curves indicate the probable surge trouble \nrate of apparatus tested with surges having parameters that will just \ncause failure.\n\nThe curves in Fig. 9 provide a means of determining the probable \nlightning surge-handling capabilities of apparatus in two ways. [irst, \nwhere the acceptable lightning trouble rate has been established by \nsystem requirements, the parameters of the appropriate test surge may \nbe read from the curves. A second approach may be employed in the \ncase where it is desired to determine the probable trouble rate with re- \ngard to a specific piece of apparatus. The procedure would be to estab- \nlish, by tests, the withstand level of the apparatus in question by em-\n\nthe withstand point, the corresponding estimated trouble rate may be \nread from the curves.\n\nThe waveshapes presented in Fig. 9 will concurrently test apparatus \nfor its dielectric strength, its energy-handling capabilities and its ability\n\nFig. 9 \u2014 Relationship of probable trouble rates to test surges having param \neters that will just cause equipment failure\n\nbasis of longitudinal surge data, it is felt that they will provide a reason- \nable test of the surge-handling capabilities of apparatus, both longi- \ntudinally and metallically. Since the total energy contained in a metallic \nsurge must be somewhat less than the total energy of the associated \nlongitudinal surge, the recommended test surges, when applied metal- \nlically, will provide an added safety factor. However, this additional\n\nsafety factor cannot be specifically evaluated from the limited metallic\n\nThe main detrimental effect of lightning surges on semiconductor \ndevices is excessive heating of their junction or junctions, the heating \neffect being related to the energy content of the surge. Since semicon- \nductor devices are used in circuits having different frequency response \nranges, it is desirable to determine the energy versus frequency distribu- \ntion of lightning surges.\n\nAnalysis of the energy versus frequency distribution was performed \non a 25 by 160-microsecond surge, as it represented one of the shortest \nduration surges observed on cable plant in this study, and therefore con-\n\ncent of the total energy of this wave is contained in the frequency band \nup to 3,500 cycles per second. For comparison with a longer duration \nsurge, a similar analysis was performed on a 10 by 1000-microsecond \nsurge, which indicated that 90 per cent of the energy of this surge is \ncontained in the frequency band up to 660 cycles per second. Plots of \nthe cumulative per cent energy as a function of frequency for both of \nthese surges are given in ig. 10.\n\nIn view of the limited frequency spectrum of the energy content of \nlongitudinal surges, it is desirable to determine the energy versus fre- \nquency distribution of those metallic surges which are oscillatory. The \noscillograms of metallic surges showed the shortest time interval be-\n\nFig. 10 \u2014 Cumulative distribution of energy vs. frequency for two sample surges.\n\ntween polarity reversals to be 80 microseconds. An energy-frequency \nspectrum analysis for such a wave indicates that 90 per cent of the en- \nergy is contained in frequencies below 7 ke. It therefore appears that \nmost of the energy of metallic surges appearing across a resistive termina- \ntion is likely to be in the frequency band below 7 ke. This does not, how- \never, preclude the need for lightning protection of apparatus operating \nat carrier frequencies. If reactive components are present in the metallic\n\ntermination, the resulting metallic surges may have considerable energy\n\nLightning surges were recorded in trunk pairs in aerial and buried \ncable at several locations known to experience heavy thunderstorm ac- \ntivity. Surges were also monitored in trunks in well-shielded under- \nground cables in urban areas. Observations included measurement of \nlongitudinal surge voltages (from conductor to sheath or ground) and \nmetallic surge voltages (between conductors of a pair). Supplemental \nlaboratory measurements of surge characteristics in simulated under- \nground plant were made using test cable.\n\nIn the underground cables monitored, no surges appeared in excess \nof 90 volts (the minimum sensitivity of measuring equipment ) although \na total of 44 thunderstorm days occurred during the observation period \nat the three test locations. Based on this field experience and supple- \nmental measurements made on test cable, it was concluded that appara- \ntus capable of withstanding surges up to 90 volts peak should not re- \nquire lightning protection if connected to well-shielded all-underground \ncable pairs. This category includes underground trunks in cables with \nwell-shielded, aerial subscriber complements such as block cable. How- \never, underground pairs extended aerially, or in buried plant, should be \nconsidered as exposed to lightning, unless protection is applied at the \nunderground junction to suppress surges from the exposed extensions.\n\nIn the aerial and buried cable plant studied, about 1400 surges were \nrecorded, ranging in peak amplitude from 60 volts (minimum sensitivity \nof equipment) up to 450 volts, the value at which carbon block pro- \ntectors operated. About 90 per cent of the recorded surges were longi- \ntudinal; the remainder were metallic. Analysis of this data helped es- \ntablish the relationship between the parameters of specific test surges \nand the probable lightning trouble rate of exposed transistorized appa- \nratus. This information will facilitate appropriate laboratory testing of \napparatus for specific levels of reliability.\n\nWith the recording technique used in this study, only those surges \nin the range above 60 volts and less than the operating value of carbon \nblock protectors were registered. The 3-mil gap, carbon protector blocks\n\nnominal value of 500 volts. Due to manufacturing and field duty varia- \ntions, the operating value of these carbon blocks may vary from approxi- \nmately 400 to 600 volts. Protector block operation, therefore, affected \nthe distribution of peak voltages above 400 volts. The recorded data \nestablished the distribution of surge voltages between the limits of 60 \nvolts and approximately 400 volts. This distribution was extended to \n600 volts by considering the effects of block operation in the 400 to 600 \nvolt range as discussed later.\n\nIn the aerial cable plant at Buford, Georgia, a total of 103 surges was \nrecorded during the six-month study period. In the buried cable plant \nat Griffin, Georgia, a total of 1120 surges was recorded for the same \nperiod of time. Histograms of the distribution of these surges as a fune- \ntion of voltage are presented in Figs. 11 and 12 for the two types of\n\nFig. 12 \u2014 Amplitude distribution on buried cable. \nplant. These distributions are of an exponential form and can be closely \napproximated by\n\nSmoothing of the raw data (fitting of an exponential by the method \nof least squares) resulted in the plots on Fig. 13. The constants for the \nexponential distribution on aerial cable are a = 0.012 and \u00a2 = 40. The \nconstants for the buried cable are a = 0.0094 and \u00a2 = 330.\n\nThe large difference in the calculated values of \u00a2 for the two types of \nplant merely reflect the difference in the sample sizes and will not affect \ntheir probability distributions. It will be noted that the values caleu- \nlated for the constant term a for the two types of plant varies over 20 \nper cent, but it is recommended that the smaller value, calculated for \nburied cable, be used for both types of cable installation for two reasons.\n\nwhich permitted a more accurate determination of the shape of the ex- \nponential distribution. Also, since our objective is to provide suitable \nsurge voltages for laboratory testing of apparatus, it is safer to use the\n\ndistribution with the smaller value of a which predicts a greater per- \ncentage of higher voltage surges.\n\nHaving determined the distribution of surge voltages, it is possible to \nestimate the probability that V is less than or equal to some value 7\u2019:\n\nUsing the previously established value of a, the probability distribu- \ntion of surge voltages was then plotted (Fig. 14). This, however, gives\n\nthe distribution of all surges to the cable plant, while the voltages re- \ncorded in the field were truncated, with only values above 60 volts be- \ning recorded. The desired distribution is, therefore, the conditional \nprobability that a voltage peak exceeds T volts given that it is greater \nthan 60 volts. This may be accomplished by shifting the distribution \nfunction to the left until the probability of exceeding 60 volts is equal \nto one.\n\nThis expression provides the probability distribution presented in \nVig. 15. \nTo estimate the number of surges per thunderstorm day which exceed\n\nPROBABILITY (PER CENT) THAT SURGES EXCEEDING 60 VOLTS \nWILL ALSO EXCEED ABSCISSA\n\nFig. 15 \u2014 Probability distribution of all surges exceeding 60 volts in buried \nand aerial cable\n\nber of surges induced in a cable per thunderstorm day. An estimated \nvalue was computed by counting the total number of recorded surges \nover the entire study period and dividing by the number of thunder- \nstorm days occurring in the area. Very good correlation was established \nbetween observed thunderstorm days (on film) and those reported by \nlocal weather stations.\n\nA total of 36 thunderstorm days was recorded in the vicinity of the \nburied cable at Griffin and six thunderstorm days were recorded in the \nvicinity of the aerial cable at Buford. These 42 thunderstorm days re- \ncorded for the two test locations accounted for a total of 1220 surges, \nfor an average of 29 surges per thunderstorm day.\n\nThe actual number of surges induced in a cable, however, is approxi- \nmately 2.5 times this value or 73 surges per storm. This results from the \nfact that multiple surges occur in approximately half of all lightning \nstrokes, and that the average number of surges in a multiple discharge\n\nsecond.\u201d The sweep speed of the test oscillographs (full scale deflection\n\nFig. 17 \u2014 Sparkover distribution characteristics of 3-mil carbon block protectors\n\nA General Method of Applying \nError Correction to Synchronous \nDigital Systems\n\nA general method is presented for applying error correction to synchronous \nbinary digital systems to improve reliability. It includes the familiar scheme \nof triplication and \u201c\u2018vote taking\u2019\u2019 as a special case. In principle, the method \npermits the system to operate continuously, even when a fault is present or \nmaintenance is being performed. An efficient maintenance routine, including \nrapid repair of faults, is an essential adiunct to the scheme if the potentially \nlarge increase in reliability made possible by error correction is to be realized.\n\nThe percentage redundancy needed to realize the scheme decreases as the \ncomplexity of the system to which it is applied increases, but may amount \nto triplication of equipment even for moderately large systems. The paper \ndescribes some error-correcting codes to implement the scheme, discusses \nerror-correcting circuits in a general way, indicates how to estimate the \nredundancy, and presents a formula for determining the reliability improve- \nment obtainable with a particular maintenance routine. In a companion\n\nThis paper describes a general method of applying error correction to \nsvnchronous digital data systems. It includes, as a special case, the well- \nknown scheme of triplication with vote taking.? Since the scheme employs \nerror-correcting codes, it is capable of detecting errors as well as correct- \ning them. Hence, maintenance personnel can be alerted as soon as a \nfault occurs. Also, it has the property of enabling the system to which \nit is applied to continue to function correctly even when faults are \npresent and maintenance is being performed, provided all the faults \nare confined to any one of the several subunits which comprise the sys-\n\nthe error-correcting capabilities of the scheme, the system may be kept \nin continuous operation for much longer periods than could the equiva- \nlent system without error detection and correction.\n\nIn this regard, it is estimated that the \u201cmean life\u2019\u2019* of the system \nwith error correction can be made several thousand times as long as \nthat of the equivalent nonredundant system, provided faults are re- \npaired sufficiently soon after their occurrence. Such potentially vast \nincreases in reliability depend of course on the availability of rapid \ndiagnostic and fault-repair facilities. Conversely, in the absence of \nmaintenance the mean life of the redundant system will in general be \nless than that of the nonredundant system. Hence the scheme is not \nusefully applicable to a system which must operate in an environment \nwhere rapid fault repair is impossible \u2014 in such situations some other \nmethod of building in reliability, such as microlevel redundancy,\u2019 would \nbe necessary.\n\nIn comparison with triplication and vote taking, our procedure will \npermit more precise localization of faults. Also, for large systems it \nshould result in less over-all equipment redundancy. For small systems, \nhowever, an equipment advantage may not always be realized. Since \nthe triplication scheme is fairly well known, we shall start by deserib- \ning it, but from a slightly different point of view, which shows how it \nappears as a special case of our procedure.\n\nFig. 1 shows a system, A, with m inputs and n outputs, and two exact \nreplicas of the system, B and c. Corresponding output wires from a, B \nand \u00a2 are fed to \u2018\u201c\u2018majority\u201d\u2019 circuits, or vote takers, each of whose out-\n\n* The mean life of a svstem is here defined as its mean time to failure, assum \ning it is in perfect condition at the start\n\nputs agrees with the majority, i.e., with any two or all three, of the \ninputs which are in agreement. Thus, the system corrects for errors \nwhich are confined to the outputs of any one of systems A, B or c.\n\nWe may consider the outputs from A to carry information bits, and \nthose from B and \u00a2 to carry check bits, which generate Hamming? \nsingle error-correcting codes, in the following manner. The matrix be- \nlow displays the output bits from a, B and \u00a2 in a matrix consisting of \nthree rows, each row having n entries:\n\nAlternatively, the matrix may be thought of as displaying n columns, \neach with three entries consisting of one information bit and two check \nbits. For example, the first column contains the information bit A; , and \ncheck bits B; and C;. Two parity checks are constructed from this \ncolumn; bits A; and B, satisfy the parity relation\n\nwhere \u00ae represents the sum modulo 2. Bits A; and C; satisfy the parity \nrelation\n\nThese relations merely state that, when the complete system is operating \ncorrectly, both B; and C; will have the same value as A; .\n\nThis coding has the ability to detect any single error in column 1, \nand moreover tells us which bit is in error, so that corrections can be \nperformed. Therefore, in particular, this scheme permits the correction \nof any pattern of errors which is confined to a single row of the matrix. \nSince faults which are confined to one of the systems A, B, or C can \ncause errors on the outputs of that system alone, this error-correcting \nscheme will permit the over-all system to operate correctly even when \nany one of the three systems comprising it is faulty, or is disabled for \nmaintenance purposes.\n\nObviously, this particular coding is inefficient, because two check bits\n\nSuppose system a of Fig. 1 is designed so that it breaks down into a \nnumber, say r, of electrically independent subunits, each subunit carry- \ning not more than p of the n system outputs, as shown in Fig. 2.\n\nA fault or faults that is confined to any one subunit can at most cause \nerrors on the outputs of that subunit. Therefore, consider the following \nmatrix, in which the outputs of each subunit are displayed in a separate \nrow, with p entries per row. There are (q \u2014 r) additional subunits shown \nin Fig. 2; these provide k check bit outputs:\n\nnore a Care? outputs from subunit r \n| aes oe oe \nrows X X --- X outputs from subunit gq\n\nSince faults in a single subunit affect only a single row of the matrix, \nwe may, for example, apply Hamming single error-correcting codes on a\n\nper-column basis. Thus, if r = 4, we need only three check bits per \ncolumn to provide Hamming single-error correction, and the code re- \ndundancy is much less than in the triplication scheme (+ instead of $).\n\nWe now wish to show how it is possible to break down a system into \nelectrically independent subunits. Digital systems may be classified into \ntwo types: those which perform only combinational logic (have no \nmemory), and those which perform sequential logic (have memory). \nThe latter type is of more interest, but it is useful to deal with the former \nfirst. We assume throughout that the data on the input wires are not \nin error, and that faults in the system do not cause errors on the input \nwires.\n\nSuppose then that the r subunits in Fig. 2, which produce the 7 \nsystem outputs, consist entirely of combinational logic. It is evident \nthat the system can be broken down into such subunits because, for \nexample, each output can be realized by designing a separate combina- \ntional logic circuit which generates the appropriate Boolean function of \nthe m input variables. Alternatively, some savings in logic elements \nmay be possible by designing multifunctional logic circuits, each generat- \ning only the p outputs of a single subunit.\n\nTo provide check outputs, additional subunits are needed, and are \ndesignated (r + 1) through gq in Fig. 2. To design these, it is necessary \nto be able to express each check output as a Boolean function of the m \ninput variables. This can be done because the structure of the error- \ncorrecting code will specify each check output to be the sum modulo 2 \nof some set of information outputs, and since the latter are known \nfunctions of the inputs, we can therefore express the check outputs \ndirectly as functions of the inputs. We may, of course, work with truth \ntables instead of functional representations.\n\nIn the case of sequential logic, a complication is introduced which \nmay be explained with the aid of Fig. 3. In this figure, a sequential\n\n* Some authors replace the memory elements by unit delay elements. See for \nexample, Fig. 1 of Unger.\u00ae His paper deals with asynchronous circuits, whereas \nwe are treating synchronous circuits of the type designated \u2018\u2018PP\u2019\u2019 by Cadden.\u00ae\n\n(MEMORY ) \nFig. 3 \u2014 A possible configuration for a finite-state sequential system. \nThe combinational logic generates two sets of outputs:\n\nThe s outputs of the memory unit, in conjunction with the m system \ninputs, comprise the inputs to the combinational logic unit.\n\nSuppose that unit | is designed as r electrically independent subunits. \nIn general, the (m + s) inputs to unit 1 will feed all r subunits. A fault \nin a single subunit will cause errors on the output of that subunit, and \nthese will feed back via the memory unit to the inputs of some or all \nof the other subunits. Hence, in a few cycles of operation it is possible \nthat the outputs of all subunits will be in error because of a fault in \njust one subunit. This situation can be remedied by applying error \ncorrection to some or all of the s feedback wires in addition to the n \nsystem output wires. These additional corrections should be made be- \ntween the outputs of unit 2 and the inputs of unit 1, in order to correct \nerrors caused by faults in unit 2 as well as in unit 1.\n\nAlternatively, it is possible to design the system so as to avoid cor- \nrecting the internal feedback wires, and yet insure that a fault affects \nnot more than p of the n system outputs. For example, instead of break- \ning down the system into r subunits, one could replicate the system r \ntimes and utilize only outputs 1, 2, --- , p, from the first replica, out- \nputs p + 1, p + 2,--- , 2p, from the second replica --- and outputs \nn\u2014p+1,n\u2014 p+ 2,---,n, from the rth replica.\n\nNo doubt this alternative realization could be achieved without using \nr complete replicas of the system. However, the necessary design pro- \ncedures are not well formulated and the resulting equipment redundancy \nis difficult to estimate. In contrast, the design procedure for the first \nmentioned method is straightforward, its redundancy is easier to esti- \nmate and, at least with present devices and techniques, it appears to \nresult in considerably less over-all redundancy. Therefore, in the re- \nmainder of the paper we shall assume that the first method is to be used.\n\nAccordingly, the correction of a sequential system requires that unit | \nof Fig. 3 be designed as r independent subunits and that it be augmented \nby a combinational logic unit which generates k check outputs, where \nk is large enough to provide the necessary parity checks for correcting \n(n + s) wires (we assume that all s feedback wires may require correc- \ntion). As in the purely combinational case, each check bit can be ex- \npressed as a sum modulo 2 of an appropriate subset of the n outputs of \nunit | and the s outputs of unit 2, and since these are known Boolean \nfunctions of the (m + s) inputs to unit 1, each check bit output ean \nlikewise be expressed as a Boolean function of these same inputs.\n\nIt is of course necessary that the check bit logic circuits also be de- \nsigned as independent subunits with not more than p outputs per sub- \nunit.\n\nBefore discussing specific codes, we wish to establish lower bounds on \nthe number of check bits, /, needed to fulfill our error-correcting re- \nquirements. Specifically, referring to the matrix above for the outputs \nfrom r subunits of system A, we ask what minimum value of / is required \nto permit correction of every possible pattern of errors in any single \nrow of the q rows.\n\nActually, two lower bounds are applicable. The first bound, which is \nalso the larger of the two when q > (2? + 1), p being the number of en- \ntries per row, is easily derived as follows: Observe that the number of \npossible error patterns in a single row is (2? \u2014 1), if we exclude the no-er- \nror pattern. Therefore, the total number of error patterns in all g rows is \nq(2\u201d \u2014 1). Obviously, & must be large enough to permit as many \u2018\u2018parity \nfailure\u201d? patterns as there are error patterns. This requires that / satisfy \nthe inequality\n\nwhere the square bracket denotes the smallest integer which is \n= logs (q2\u201d \u2014 q + 1). \nThe second lower bound, which is larger than the first when \n(2\u201d + 1), and which is therefore of greater practical significance, | \nk 2 2p.\n\ncan be chosen out of a set of 2\u201d binary words of lengths pg, and which \nfulfill the specified error-correcting requirements. Its derivation \u00a7 is \nrelegated to the Appendix.\n\nSurprisingly, it was found not too difficult to construct codes for \nmost values of p and q in the range 2 < p < 10,3 < q < 9 which \nachieved the appropriate lower bound.\n\nSubsequent to the work described here, Ray-Chaudhuri! developed a \ngeneral theory of minimally redundant codes for this application. How- \never, it will not be out of place to exhibit here some of the codes pre- \nviously derived, since they are also minimally redundant, and since the \nerror-correcting equipment required to implement either them or the \nRay-Chaudhuri codes is of the same general character and complexity. \nThree families of codes* are exhibited below in matrix form, correspond- \ning to three values of p, as follows:\n\n* These codes were constructed by George Allen at Bell Telephone Labora \ntories in Summer, 1959\n\nThe set of digits beside each bit position indicates the parity groups \nwhich that bit enters. For example, in family 1, the digit set attached \nto the last bit in the last row is 123, indicating that this bit enters parity \ncheck groups 1, 2 and 3. As a further illustration, the bits in family 1 \nthat are labeled 1, 14, 13, 124 and 123 enter parity group 1; their sum \nmodulo 2 is zero when there are no errors. The bit positions which carry \nonly a single digit are check bit positions. They are denoted by X\u2019s and \nthe information bits are denoted by O\u2019s. The matrix displayed for family \nlisa 2 X 5 matrix; however, if a 2 * 4 or a2 X 3 matrix is desired, \none omits respectively the last row or the last two rows. Similar remarks \napply to families 2 and 3.\n\nSeveral remarks should be made at this point. First, observe that in \nthe original matrix we represented the check bits as being located en- \ntirely in the last (\u00a2 \u2014 r) rows. However, this is not necessary; the check \nbits may appear along with information bits in some or all rows, as is \nthe case in the three matrices above. The arrangement is dictated by \nthe structure of the code, but bits may be permuted within each row \nwith impunity.\n\nSecondly, it is possible to delete information bits from any row with- \nout destroying the utility of a code; in such cases, the deleted bits will \nbe omitted from the parity checks in which they would normally par- \nticipate.\n\nFinally, we observe that the three matrices above provide only for \nvalues of p S 4andq S 9. If for any reason we wish to form matrices \nwith p > 4, or q > 9, this may be done by building up the over-all \nmatrix, either vertically or horizontally, or both, from several of the \nabove matrices. Thus a wide variety of equipment arrangements can be \naccommodated. However, if we build up vertically, we will sacrifice \nminimal code redundancy. For example, if we form a matrix with 3 \ncolumns and 18 rows by using two matrices of family 2, we shall have\n\nA discussion of error-correcting circuits is included here to indicate \nroughly the amount of equipment involved in error correction, and to \nprovide a basis for a maintenance routine which is proposed later.\n\nIt was explained in Section II that, with the present scheme, it is \nnecessary to apply error correction either to the nm system outputs of a \npurely combinational system, or to the (nm + s) outputs and feedback\n\nconnections of a sequential system. To do this, k parity check bits must \nbe generated, where k is determined by the structure of the error-cor- \nrecting code employed. Therefore, the inputs to the error-correcting \ncircuits consist of (V + k) wires, where VN = n+ s (s 0 for a purely \ncombinational system) and / wires carry the check bits. The output of \nthe error-correcting circuits consists of NV wires which carry the cor- \nrected versions of the corresponding NV inputs.\n\nWe shall want to distinguish between the correcting circuits and the \ncircuits which are corrected or correctable. The latter comprise the set \nof g subunits which generate the (V + k) outputs; for example the\n\nThe error correcting-circuits may be considered to perform the fol- \nlowing three functions:\n\n2. From the pattern of parity check failures, determine which of the \n(NV + k) input wires are carrying erroneous bits.\n\n3. Correct that subset of the V wires which are carrying erroneous \nbits. Erroneous check bits do not require correction.\n\nCircuits to perform the above tasks may be realized in several ways, \nand with varying degrees of redundancy to assure reliability. At one \nextreme, the error-correcting circuits could be nonredundant, in which \ncase an efficient preventive maintenance routine would be required to \ninsure that they perform for long periods without error. Alternatively \nthey could be built with microlevel redundancy, in which case preventive \nmaintenance would again be necessary but would be appliec less fre- \nquently. A third alternative would be to make some or all of the error- \ncorrecting circuits \u2018\u201c\u2018self-error-detecting.\u201d\u2019 Those parts which were self- \nerror-detecting would be subjected to maintenance only when a fault \nwas detected; the parts which were not self-error-detecting would require \npreventive maintenance.\n\nAs a fourth alternative, it might be attempted to make the error- \ncorrecting circuits completely self-error-correcting. However, a simple \nheuristic argument can be given which indicates that it is impossible to \nachieve this goal.\n\nFig. 4 shows a block diagram of a proposed error-correcting circuit, \ndesigned according to the third alternative above. Box 1 in Fig. 4 con- \ntains the units which perform functions | and 2 above. Box 2 performs \nfunction 3 above, and also an error-sensing and alarm function. For\n\nsimplicity, sets of wires in this figure are represented by single directed\n\nlines with an associated symbol indicating the number of wires. Box | \nis designed to be self-error-detecting, but box 2 is not. A detailed deserip- \ntion of the operation of the circuits in Fig. 4 does not contribute sig- \nnificantly to an understanding of the over-all scheme, and so is omitted.\n\nThe maintenance routine is as follows. If a fault occurs in some sub- \nunit of the system, unit bp under control of box | corrects the resulting \nerrors. Simultaneously, box | operates alarm 1. The faulty subunit is \nthen located and repaired as quickly as possible. In principle, the svstem \ncan continue to operate correctly even when the faulty subunit is being \nreplaced or repaired, provided a fault does not develop in another sub- \nunit or in the error-correcting circuits during the repair of the original \nfault.\n\nIf box 1 fails, alarm 2 is operated, and possibly alarm 1 also, and \nsimultaneously relay E is switched to the bypass position, thus causing \nessentially no interruption in system operation. Box | must also be \nrepaired quickly, because if a fault occurs in the system during the \nrepair of box 1, the system will fail, since it is not now being error cor-\n\nrected. If units c or p fail, they do not operate an alarm since they are \nnot provided with error detection. The probability of failure of these \nunits must, therefore, be minimized by a preventive maintenance routine. \nTo minimize the outage times of units \u00a2 and pb, two copies of each could \nbe provided, one operating and one standby. They would be interchanged\n\nto the units currently in the standby condition. In this way, the system \nwould not lose its error correcting capability during preventive main- \ntenance. Relay E must be switched to the bypass position to permit \ncontinuous operation while unit \u00a2 or b is being replaced. We assume that \nthe relay switching time is short enough so as not to interrupt system \noperation.\n\nWe now wish to determine a quantitative measure of the reliability \nimprovement of the maintained error-correcting system over that of the \nnonredundant system. A useful measure is the ratio of their respective \nmean lives; namely, R, = L,/L;, where L, is the mean life of the re- \ndundant system and L; is the mean life of the nonredundant system. \nIn general, the derivation of Ry, is quite complicated, but by making \nsuitable simplifying assumptions we can obtain an approximate formula \nwhich is useful. These assumptions are:\n\n1. All components have an exponential survival probability function \nand the same mean life, which is taken to be the time unit. Therefore, \nthe survival probability of any component is exponential (\u2014?).\n\n3. Failure of any component in the nonredundant system causes that \nsystem to fail.\n\n5. The times taken to repair faulty circuits are assumed constant. \nThese and other parameters are now defined:\n\nA; = time that unit c\u00a2 or unit D is removed from circuit when \nbeing exchanged with its standby,\n\nT = time interval between successive replacements of unit \u00a2 or D \nby its standby\n\nERROR CORRECTION FOR SYNCHRONOUS DIGITAL SYSTEMS \u2014 589 \nvalues of the above parameters, we can obtain the following formulas:\n\nAs an application of (3) and (4), consider a system having parameter \nvalues as follows: \ntime unit = mean component life 1000 years, \nA, = As = one-half hour } X 10-7 units (time to repair a \nsystem subunit or box 1), \nA; = 8 seconds = 2 X 10\u00b0'\u00b0 units (time to replace unit \u00a2 or \nunit D by its standby), \nT = one month = 10~* units (maintenance interval for units \ncand pb), \nnz = 333 components, \n666 components, \n6 \nr= 4, \nSubstituting these values in (3) and (4) results in Ry equal to 2900.\n\nThat is, the mean life of the redundant system is 2900 times that of the \nnonredundant system. Actually, this figure could be improved if we \nreduced the repair times A; and A, from one-half hour to, say, five min- \nutes. Such a reduction would be possible if the system were built of small \nmodular packages and a highly automated diagnostic routine were \navailable to locate faults in the order of a minute or less. Thus, a fault \ncould be pinpointed to one or two particular packages, and these pack- \nages could be replaced immediately by good standby packages, thus \npermitting correction of the fault in minutes. The faulty packages could \nthen be tested and repaired in more leisurely fashion, and this latter \ntime would not be chargeable to A; or Ae .\n\nTherefore, it appears that, with an efficient maintenance routine, the \nmean life of the error-correcting system can be several thousand times \nthat of the nonredundant system.\n\nin simple terms, but the contributing sources can be delineated and \nroughly evaluated. They are:\n\n(b) the cireuit redundancy which results from designing the system \nas q independent subunits;\n\nThe redundancy contribution of item (a) can reasonably be estimated \nto be in the ratio k N to the amount of equipment in the original non- \nredundant system. That is, if the amount of equipment in the non- \nredundant system is treated as one \u201cunit\u201d? of equipment, the amount \nof equipment needed to generate the check bits would be roughly k/N \n\u201cunits.\u201d\u2019 As indicated by the codes in Section II], values of kN as \nsmall as 3 are achievable for V = 18. The assumption underlying this \nestimate is that the amount of circuitry required to generate each check \noutput is the same as the amount required to produce each original \nsystem output.\n\nThe redundancy contribution of item (b) is believed to be insignificant \ncompared to the other contributions, especially for large systems, pro- \nvided optimal design techniques are employed. The contribution of \nitem (c) is by far the largest, and is the most difficult one to estimate. \nIt depends on the type of logic technology employed, on the amount of \ntime delay that the error-correcting function is permitted to introduce, \nand to some extent on the particular error-correcting code used. The \nfollowing estimates may suggest an order of magnitude for item (\u00a2), in \nthe particular case of the correcting circuits proposed in Fig. 4, and \nassuming the use of diode logic. Based on \u201cpaper\u201d designs of these \ncircuits, the author estimated that the correcting circuits might require \nroughly 60 to 70 \u201cequivalent\u201d? diodes per wire corrected (the number of \nwires corrected is V). Transistors were counted as equivalent to two \ndiodes, resistors etc. were not counted.\n\nThis estimate assumes that units A and a\u2019 of Fig. 4 both employ \nc \u2014 1) EXCLUSIVE OR circuits per parity check over \u00a2 bits, and that \nunit B of Fig. 4 is realized with two-stage logic. Thus, if these error- \ncorrecting circuits were to be applied to a system which, in its non- \nredundant form, was realizable with 30 to 35 \u201cequivalent\u201d? diodes per\n\nwire corrected, it is evident that the correcting circuits would comprise\n\nIn general, therefore, it is to be expected that the scheme in question \nmay introduce an amount of redundancy equivalent to at least triplica- \ntion of the original equipment. However, it has the potential of being\n\ngeneration of the check bits alone results in triplication and the correct-\n\nIn principle, it appears that the redundancy penalty might be made \nto decrease monotonically as the systems to which error correction is \napplied becomes increasingly complex, provided the following two as- \nsumptions are valid:\n\n(a) as the number N of corrections is increased, the coding efficiency \nalso increases; that is, k/N becomes smaller;\n\n(b) the amount of equipment per correction in the original system \nincreases faster than the amount of equipment per correction in the \ncorrecting circuits.\n\nAssumption (a) is realizable, but (b) cannot be verified. Indeed, (b) \nmay be plausible only provided the correcting circuits use an increasing \nnumber of logic stages, which can be expected to result in an increase \nin time taken to perform corrections; that is, an increasingly severe \nspeed penalty is imposed.\n\nIn this regard, the parity check circuits referred to earlier in this \nsection require 2 X [logec] logical stages. (The square bracket denotes \nthe smallest integer which is equal to or greater than logec.) For the \nspecial codes described in Section III, the number of bits per parity \ncheck, \u00a2, is typically equal to g, the number of subunits in a system, \nand gq must increase in order to increase the coding efficiency. It there- \nfore follows that greater coding efficiency can be achieved only at the \nexpense of greater delay in the corrector, or more complex correcting \ncircuits, or both, and a compromise must be reached.\n\nFinally, a remark should be made concerning the impact of this scheme \non the over-all design of a sequential system when the method which \nrequires both feedback and output corrections is used. To minimize \nthe number of corrections necessary, the number of feedback and output \nwires should be kept to a minimum. The designer usually is able to \nexercise some control over both. In particular, there are roughly as \nmany feedback connections in a sequential system as there are binary \nmemory elements; therefore it would be desirable to minimize the \nnumber of memory elements. At present, large systems are frequently\n\ndesigned with many more memory elements than necessary, presumably \nbecause this results in simpler design procedures. It may, therefore, be \ndesirable to find methods which lead to designs having nearly minimal \nnumbers of memory elements, in order to make our error-correcting \nprocedure more attractive.\n\napplicable to synchronous digital systems, and which includes the \ntriplication and vote-taking scheme as a special case. It permits systems \nto which it is applied to operate continuously even when faults are \npresent and maintenance is being performed. The scheme can lead to \nvery large increases in system reliability, but only if augmented by a \nmaintenance routine which effects rapid repair of faults.\n\nTwo types of error-correcting codes have been discussed, Hamming \ncodes and special codes. The Hamming codes are universally applicable, \nbut are not minimally redundant in this application. The special codes \nare minimally redundant but not universally applicable, in that they \nhave not been developed for a large range of values of p and q.\n\nThe equipment redundancy required to implement the scheme may \nbe equivalent to at least triplication for moderately large systems, but \nshould be less for more complex systems. It is not specifiable in simple \nterms and can be determined accurately only by carrying through the \ndetailed design of the specific systems. Such detailed applications have \nnot yet been made.\n\nWe shall derive this bound by showing that an upper bound on the \nnumber of code words of length pg which satisfy our error-correcting \ncriterion is 2\u00b0?\u201d words. This implies that the maximum number of \nbits which can be assigned values arbitrarily is (\u00a2 \u2014 2)p bits. The re- \nmaining bits must be check bits; therefore, a lower bound on the number \nof check bits is gp \u2014 (q \u2014 2)p = 2p bits.\n\n[t is useful to think of the pq bits which comprise a code word as being \narranged in a single row, with each successive block of p bits being re- \nplaced by a single symbol, D;, which can take on any one of the 2\u201d\n\ndifferent values. In this alternative representation, a typical g-symbol \nword would be\n\na: \nIn terms of this representation, our error-correcting criterion requires \nthat an error in any one of the g symbols be correctable. This implies \nthat admissible code words must differ in more than two symbol posi- \ntions. For, consider the following two words which differ only in the first \ntwo symbol positions:\n\nIt is possible for an error in the first symbol of word | and in the second \nsymbol of word 2 to cause both to become, for example, the word\n\nHence, we cannot determine whether word 1 or 2 was the correct word: \ntherefore, admissible code words must differ in more than two symbol \npositions.\n\ntion S into disjoint subsets S; ,7 = 1, 2,3, --- , such that two elements \nof S belong to the same subset if they are identical in the last (\u00a2 \u2014 2)\n\nsymbol positions. Thus there are as many subsets as there are trun- \ncated words D,D, --+ D \ntains 2?\u201d elements.\n\nNow arbitrarily choose a (q-symbol) word from subset S; to be a code \nword. Then no other words from subset S; can be chosen as code words, \nbecause any two words from S; differ only in the first two-symbol positions. \nBy the same argument, at most one word can be selected as a code \nword from S., ete. Therefore, there cannot be more code words than\n\n2. Von Neumann, J., Probabilistic Logies and the Synthesis of Reliable Organ \nisms from Unreliable Components, in Shannon, C. Ek. and MeCarthy, J., eds., \nAutomata Studies, Annals of Math. Studies, No. 34, Princeton Univ. Press, \nPrinceton, N. J., 1956, pp. 44-98 (particularly Section 8.3).\n\n3. Moore, E. F. and Shannon, C. E., Reliable Cireuits Using Less Reliable Re \nlays, J. Frank. Inst., 262, 1956, pp. 191; 281.\n\n5. Unger, H., Hazards and Delays in Asynchronous Sequential Switching Cir \ncuits, I.R.E. Trans., CT-6, 1959, p. 12.\n\nSeveral authors have considered the possibility of increasing the reli- \nability of large and complex binary digital systems by introducing some \nredundancy in the system. In a companion paper, Armstrong\u2019 proposes a \nscheme for applying error correction to a synchronous digital system. In \nthis paper we develop a general mathematical theory for generating mini- \nmally redundant error-correcting codes for the scheme in question. This \nresults in what are called *\u2018minimally redundant reliable systems.\u201d The \nproblem of constructing minimally redundant reliable systems whose out- \nput is free of error when there is a fault in at most one block of the system \nis completely solved. An example is considered in detail showing how the \nmathematical theory can be actually applied.\n\nIn complex binary digital systems employing a large number of blocks \nof electrical equipment it often is difficult to ensure a sufficient level of \nreliability of each single block of equipment. An attempt to attain the \ndesired degree of reliability by improving the reliability of each block \nmay prove to be uneconomical. On the other hand, by introducing some \nredundancy in the system, it is possible to construct highly reliable \ncomplex systems, even though each single block is not as highly reliable. \nMoore and Shannon,\u201d Tryon,\u2019 Von Neumann,\u2019 Lofgren\u2019 and Armstrong! \nhave considered the problem of constructing reliable system designs. \nIn this paper a general mathematical theory has been developed for the \nconstruction of minimally redundant reliable system designs, based on \nthe scheme outlined by Armstrong.\u2019 This theory is closely related to the \ntheory of error-correcting codes. The problem of constructing minimally \nredundant system designs whose outputs will be free of error whenever \nthere is fault in at most one block of the system is completely solved in \nthis paper.\n\nSuppose there are m binary input variables X, , X., ---, X,,. Let B,, \ndenote the set of 2\u201d m-place binary sequences. Every set of values of \nthe m binary input variables will be regarded as an element of B,, . Any \nmapping of B,, into B, will be called a Boolean function of the m input \nvariables X,, X2,---,X,,. For the sake of brevity, the collection of m \ninput variables will be denoted by X. Let\n\nbe pk Boolean functions of the m binary variables X,, X.,---,X,. \nOur problem is to construct a system which will synthesize the pk \nBoolean functions with a high degree of reliability. Thesystem uses blocks \nof electrical equipments each of which can synthesize p Boolean func- \ntions. For the sake of brevity, a collection of p Boolean functions, will \nbe called a Boolean p-function. Thus f; = (fia, fie, +++, fip) isa Boolean \np-function. Any Boolean p-function is a mapping of B,, into B,. Each \nblock of our system synthesizes a Boolean p-function. Fig. 1 is a sche- \nmatic diagram for the original nonredundant system.\n\nThe blocks act as units in the system. If there is a fault in a block, \nthen some or all the p outputs of the block are erroneous. In other words, \nin the case of a fault a block will synthesize the corresponding Boolean \np-function wrongly. Let V,' denote the set of 2\u201d binary p-tuples. Then\n\nt= 1,2---,k. Let f = (fi, fo, --:,f,). Then f can be regarded as a \nmapping of B,, onto V,*. We shall define the addition of p-tuples as the \nusual mod 2 addition. For example, if p = 3, a; (OO1) and as (101), \nthen a; + a (100). Let @ and a\u2019 be two elements of V,* given by\n\nis defined to be the element (a; + ay\u2019, --+, a + a\u2019). The p-tuple \n(00 --- 0) will be called the null element of V,'. The weight w(a) of \nthe k-vector @ is defined to be the number of nonnull elements among \na, @,-**:, a. For any particular value X\u2019 of the input variables \nf(X\u2019) is a vector in V,\". Suppose there are faults in \u00a2 (t < k) blocks. \nThen \u00a2 of the functions fi; , fo, -- +, will be synthesized wrongly. Hence \nthe output will be the vector f( X\u2019) + \u00ab, where \u00ab = (6, @,\u00b0\u00b0:,e) isa \nvector in V,* with weight \u00a2. While designing a system to synthesize the \nBoolean function f, one might require that whenever the number of\n\nfaulty blocks is \u00a2 or less, the output is error-free. One can achieve this by \nintroducing some redundancy in the system, i.e., by synthesizing (/ + r) \nBoolean p-functions and adding a logical corrector unit to the system.\n\nSuppose \u00a2; , \u00a22, \u00b0**,\u00a2, are n Boolean p-functions and C is a mapping \nof V,\" onto V,. We shall consider \u00a2 (g1,\u00a22,\u00b0**,\u00a2n) asa function\n\nfrom B,, to V,\". For every value X\u2019 of X, \u00a2(X) is an element of V,\u201d. \nSuppose the functions g and (' possess the property P\u2019 stated below:\n\nlor every vector \u00a2 belonging to V,\" with w(e) not exceeding \u00a2 and \nevery value X\u2019 of the input variable X,\n\nThe functions \u00a2 and C enable us to construct a system which will \nsynthesize the k Boolean p-functions f; , fe, \u00ab++, f, free of error whenever \nthe number of faulty blocks in the system is \u00a2 or less. The n Boolean \np-functions \u00a21 , g2, +++, \u00a2, can be considered as a collection of np Boolean \nfunctions of m input variables. Therefore we can easily obtain the \nlogical design of a system which will synthesize these np Boolean func- \ntions. This system will be called the encoder subsystem. Similarly, the \nfunction C can be considered as a collection of pk Boolean functions of \nnp binary input variables, and therefore we can obtain a system which \nwill synthesize these pk functions. This system will be called the cor- \nrector subsystem. The np outputs of the encoder subsystem will be the \ninputs of the corrector subsystem. Now it is easily seen that, because \nof the property P of the functions \u00a2 and C, whenever the number of \nfaulty blocks in the encoder subsystem is \u00a2 or less and the corrector unit \nis free of error, the pk outputs of the corrector subsystem will be\n\nthe above discussion, the following two definitions given below are \nmeaningful.\n\nDefinition 1: The functions \u00a2 ,\u00a22,\u00b0-:,\u00a2, and C possessing the \nproperty P stated in (1) will be called a reliable system design of order \ntand redundancy r = n \u2014 k for thek Boolean p-functions f; , fo, \u00ab++, fi.\n\nDefinition 2: A reliable system design of order \u00a2 and redundancy 7\u00bb \nfor the & Boolean p-functions f; , fo, \u00ab++, fe will be called minimally re- \ndundant if the redundancy r of any other reliable system design of order \nt for the same functions is not less than ro .\n\nIn the present paper we have given a method of obtaining a minimally \nredundant system design of order | for any set of k Boolean p-functions \nfor arbitrary / and p. System designs of higher order will be given in a \nsubsequent paper.\n\nWe have used the redundancy r as a measure of the extra amount of \nequipment which has to be used for making the system reliable. And \nhence we seek the system which has minimum possible value of the re- \ndundancy r. It should be pointed out that we assumed that the corrector \nsubsystem does not make any error at all. Therefore, to make the whole \ndevelopment practically feasible, it is imperative that either the amount \nof equipment necessary for the corrector subsystem is small in com- \nparison to the amount of equipment necessary for the whole system, or \nthat other steps be taken, such as are suggested in Ref. 1, to ensure \nreliability of the corrector system. We have not used any mathematical \ncriterion to incorporate this requirement in the development of the \ntheory.\n\n(00,01,10) and d(a,e\u2019) = 2. It can be easily checked that the distance \ndefined above satisfies the three conditions of a distance function. We \nhave seen that Boolean p-functions as defined in Section II can be con-\n\nshall assume that all Boolean p-functions appearing in our discussion \nare nondegenerate. In the following we have s = 2\u201d andn = k + r.\n\nTheorem 1: A necessary and sufficient condition that there exists a \nreliable system design of order \u00a2 and redundancy r for the k Boolean \np-functions f, , fe, \u00ab++, f; is that there exists a subset A of V,,\" contain- \ning s* elements such that d(a,a\u2019) = 2+ 1; aa\u2019 \u00a9 A\u2019, a # a\u2019.\n\nProof: Necessity. Suppose there exists a reliable system design of \norder \u00a2. Let the encoder functions be \u00a2 = (\u00a2),\u00a20,: + -,\u00a2g, ) and the corrector \nfunction be C. For every value X\u2019 of the input variable Y, \u00a2g( YX\u2019) is a \nvector of V,,\". Consider the set\n\nUsing the fact that the Boolean p-functions f; , fo, +-+,f, are nonde- \ngenerate functions, it follows easily that the set A contains at least s\u00b0 \nvectors of V,\". Consider two distinct vectors a and a\u2019 of the set A. If \npossible, suppose d(a,a\u2019) S 2t. Since d(a,a\u2019) S 2t, we can find a vector \neof V,\u201d\" such that a + \u00ab\u20ac = a\u2019 + cand w(e) S \u00a2. Since w(e) S t, we \nhave\n\nMquation (3) contradicts the assumption that @ and a\u2019 are distinct \nvectors of A. This completes the proof of necessity.\n\nSufficiency. Suppose A is a subset of V,\" containing s\u00b0 elements and \nhaving the property that d(a,a\u2019) 2 2t+ 1; aa\u2019 \u00a9 A, a \u00a5 a\u2019. We set \nup a one-to-one correspondence between the s* vectors of V,* and the\n\n\u2018 . a . . . \u2018 a ; , \n[fi NX\u2019) fol X\"),- ++ fe X\u2019)| is a vector of V,\u00b0 and there is a corresponding \nvectora of V,\" belongingto A. The encoder function \u00a2 5025 \u00b0**, Gn) \nis defined by\n\nwhere a is the vector of V,\" belonging to A and corresponding to the \nvector f( X\u2019) of V,\u00b0. The corrector function C is defined in the following \nmanner. Let y = (y1,\u00a52,\u00b0**, Yn) bean arbitrary vector of V,\u201d\". First\n\nwe choose a vector @ belonging to A such that d(y,a) S d(y,a\u2019), \naa\u2019 \u00a9 A, Let 8 = (8), 8, +++, Be) be the vector of V,\" which corresponds\n\nThus C is a mapping of V,\" onto V,\". It is easy to check that the en- \ncoder function \u00a2 and the corrector function C defined above possess the \nproperty P stated in Section II. This completes the proof of sufficiency.\n\nTheorem 2: If there exists a reliable system design of order \u00a2 and re- \ndundancy r for k Boolean p-functions, then\n\nwhere n = k + rands = 2\u201d. \nProof: From Theorem 1, it is necessary that there exists a subset A of \nV,,\u201d with the property that\n\nd(y,a) & t. It follows easily from (7) that, for any two distinct vectors \na and a\u2019 of A, the sets S, and S,\u2019 do not have any common element. \nLet S, | denote the set of elements of V,\u201d\" which have distance k from \na, k = 0,1,2,---,\u00a2. Obviously S, is the union of the (\u00a2 + 1) sets ig .\n\nTheorem 2 follows from (8). \nTheorem 2 gives a lower bound on the redundancy r of a reliable\n\nsystem design of order \u00a2 for k Boolean p-functions. Theorem 2 is actually \na generalization of a result of Hamming.\u00b0\n\nLet n.(r) denote the maximum integer n for which there exists a \nreliable system design of order \u00a2 and redundancy r for k = n \u2014 r Boolean \np-functions. For \u00a2 = 1, the inequality (6) becomes\n\nIf there exists a reliable system design of order \u00a2 and redundancy r for k \nBoolean p-functions, then n,(r) 2 k + r. \nLemma 1:\n\nProof: Suppose n,(r) = n. Then there exists a reliable system design \nof order \u00a2 and redundancy r for k = n \u2014 r Boolean p-functions. Hence \nby Theorem | there exists a subset A of V,\" containing s* elements \nwith the property that d(a,a\u2019) 2 2\u00a2 + 1; aw\u2019 \u00a9 A, a # a\u2019. To every\n\nof V,\"*'. Thus we have a subset A of V,\"\"' containing s* elements and \nalso possessing the property that d(a@,a@) = 2\u00a2+ 1; aa \u00a9 Aja \u00a5 &. \nHence, by Theorem 1, we can obtain a reliable system design of order \u00a2 \nand redundancy n + 1 \u2014 k = r + 1. It follows that\n\nn(ir+t1)2k+r4+1.=n4+1 = nr) +1. \nTheorem 3: If for a reliable system design of order \u00a2 and redundancy \nr for k Boolean p-functions we have \nn(r\u20141)\u2014-(r-1) <k Snir) \u2014 7, (9) \nthen the design is minimally redundant.\n\nProof: If possible, suppose the system is not minimally redundant. \nThen there exists a reliable system design of order \u00a2 and redundancy\n\nThe inequality (10) contradicts the inequality (9); hence the theorem \nis established.\n\nIn this section we shall consider a particular subclass of system de- \nsigns called the linear system designs. To define the linear system de- \nsigns, we have to use the theory of finite fields. Let K be the finite field \nof characteristic 2 containing s = 2\u201d elements and x denote a primitive \nelement of AK. Any binary p-tuple (apo, a, +++, @,-;) will be made to \ncorrespond to the element ajo + qx +--+ + 4a, w\u2019*' of K and \nvice versa. An element a = (a, a@:,\u00b0*:,a@,) of V,\" now will be con- \nsidered as an n-vector with elements in K. The weight w(a@) of a@ is\n\nObviously V,\" is a vector space over K. Consider a system design for k \nBoolean p-functions. Suppose the encoder function is\n\nDefinition 3: A system design for k Boolean p-functions is said to be \na linear system design if the subset A of V,\" defined by (11) is a vector \nspace over K.\n\nLemma 2: A necessary and sufficient condition that a reliable linear \nsystem design for k Boolean p-functions is of order \u00a2 is that the weight \nof any nonnull vector of the set A defined in (11) is not less than \n(2 + 1).\n\nBy definition d(a,a\u2019) = w(a + a\u2019). Since A is a vector space, a + a\u2019 \nis an element of A and also a + a\u2019 is a nonnull element of A. Hence it \nfollows that (12) will hold if and only if w(a@) 2 2\u00a2 + 1 for every non- \nnull element @ of A.\n\nDefinition 4: A matrix M with elements in K will be said to have the \n(P,)-property if no \u00a2 rows of the matrix are linearly dependent.\n\nTheorem 4: A necessary and sufficient condition for the existence of a \nreliable linear system design of order \u00a2 and redundancy r for k Boolean \np-functions is that there exists a matrix WV with n = (k + r) rows and r \ncolumns with elements in A which possesses (/\u2019,,)-property.\n\nLet A denote the vector space orthogonal to the vector space generated \nby the r rolumn vectors of W/. A contains at least s\u2018 elements. It would \nbe sufficient to show that the weight of any nonnull vector @ of A is at \nleast (24 + 1). If possible, suppose A contains a nonnull vector with \nweight less than (24 + 1). For simplicity of writing assume that the \nvectora = (a, a2, +++, a@,0,---,0) belongsto A where a; , a2, +++, ax,\n\nEquation (14) implies that the first 2\u00a2 vectors of the matrix 1/7 are \nlinearly dependent which is a contradiction. This completes the proof \nof sufficiency. Necessity can be proved by exactly similar arguments. \nThe reader acquainted with the literature on error-correcting linear \ncodes would recognize from Theorem 4 that a reliable linear system de- \nsign of order \u00a2 for k Boolean p-functions exists if and only if a (error \ncorrecting linear code in s( = 2\u201d) symbols with n places and /& informa- \ntion places exists. Lemma | and Theorem 4 given above are not new \nresults; they were proved by Bose\u2018 and Zierler\u00ae in a different form. We \nhave included short proofs for these results for the sake of completeness.\n\nIn this section we shall give methods for constructing minimally re- \ndundant linear system designs of order | for / Boolean p-functions for \nany arbitrary value of k and p.\n\nTo prove (15) we shall construct a matrix 7 with r columns and n = \n(s\" \u2014 1)/(s \u2014 1) rows which has (P2)-property. We shall denote the ele- \nments of K by 0,l,a2, ++ ,a,-1, where 0 is the null element and 1 is the \nmultiplicative identity. Consider the matrix 7 given by \nM, \nM = (16) \nI, \nwhere /, is the identity matrix with r rows and r columns and J, is a\n\nIt can be easily checked that the matrix V7 has (/?.)-property; 1.e., no \ntwo rows of VV are linearly dependent. This completes the proof of \nTheorem 5.\n\nIt should be observed that Theorem 5 enables us to construct mini- \nmally redundant system designs of order | for any arbitrary values of k \nand p. For given k, we find out the integer r for which\n\nmir \u20141)\u2014(r\u20141) <k Snr) \u2014--.. \nIf n,(r) \u2014 r = k, we construct the matrix M with r columns and n,(r) \nrows as defined in (16) and then obtain the system design as illustrated\n\nin the proof of Theorem 4. If n(7) \u2014 r > k, then we delete n(r) \u2014 \n(k + r) rows from M, and thus obtain a matrix WM with (P2)-property \nwhich has r columns and (k + r) rows. From Theorem 3, is follows that \nthe resulting system design will be minimally redundant. Now we shall \ngive explicitly the encoder function of the minimally redundant system \ndesigns of order 1. Let\n\nwhere /, is the identity matrix with & rows and & columns and J/)\u2019 is \nthe transpose of the matrix /,. It can be verified that the k column \nvectors of B are orthogonal to each of the r column vectors of 4/7. The \nk column vectors of B generate the vector space A and every nonnull \nvector of A has weight greater than 2/. For the sake of convenience of \ndescription, we write\n\nIf there is error in one block of the encoder subsystem, \u00a2 will have one \nnonzero element among its coordinates. Suppose the /th coordinate of \nis a nonzero element A of A. Then e (Biss \u00ab2 Xe.* 2). 2 OS H, \nLet us denote the /th row vector of the matrix VW by \nR; Uy Cl2,y \u00b0\u00b0 *, Cir). \nThen the following equations hold: \n6 4 AR, \n(21) \nNC) \u00bb ] \nConversely, if for a given output vector y the vector 6 computed by \n20) satisfies (21) and there is one block of the encoder subsystem in \nerror, then the error vector \u00a2 will have \\ as its /th coordinate and zero \nas the other coordinates. \nrom the above discussion, it is clear that a corrector subsystem which\n\nobserves the rules given below will accomplish the job of correcting \nerrors in one block of the encoder subsystem and produce the k Boolean \np-function fi( X ),fo( X),+++.fe( X) as its output. The rules are\n\nil. If 6 is not the null vector, find out the integer / for which the \nvector AR, for some A K has maximum number of common coordi- \nnates with 6. If / > k, the k outputs are 8B; = y;,7 = 1,2,---,k. Ifl < k, \nthe k outputs are 8B; = y;, Be Y2,\u00b0\u00b0', Bi = yr tA,-\u00b0+, Be Vk.\n\nIn this section we shall give an example to show how the theory de- \nveloped in this paper can be applied.\n\nSuppose m = 3, p = 2,k = 3 and \u00a2t = 1. From Section V, we can \nsee that for the minimally redundant system r 2. Suppose the three \nBoolean two-functions to be synthesized are\n\nfi X) [ful X), fie( X)] \n(X,-X.,X1 @ X2), \n1,(X) = [for(X), foo( X)] \na ee ee \nSs(X) = [fa X), foo X)] \n(X,-X;,X1 \u00a9 X;) \nwhere the symbols \u00ae and - are respectively used to denote the Boolean\n\noperations of additions (or) and multiplication (AND) between two \nbinary variables. Let k denote the field containing four elements. Let \u00a2\n\nminimum function and every element of the field satisfies the equation \nx 1. The four elements are shown below in terms of the primitive \nelement \u00a2, and their correspondence with binary 2-vectors is also pointed\n\nIn view of the correspondence between the binary 2-vectors and the \nelements of K, any particular value of a Boolean 2-function will be con- \nsidered as an element of A. For example, if\n\nAddition and multiplication between the elements of K are shown in \nthe tables given below:\n\n3 2 Q@ a3, 3 \nThe sum of two elements a; and a; is obtained by adding the cor- \nresponding polynomials in \u00a2 modulo 2 (7, 7 = 0, 1, 2, 3). The product \nof two elements is obtained by multiplying the corresponding poly- \nnomials modulo 2 and modulo (\u00a2 + \u00a2 + 1). From Section V we have\n\nM1 = Mo = M3 = a, and my = ay, Me = ae and Mo; a;. There- \nfore the five encoder Boolean 2-functions are given by\n\ntion (AND) between two binary variables. \nThe truth table for the two Boolean functions \u00a25; and \u00a25\u00bb is given\n\nThe computation of this table will be illustrated by one example. Sup- \npose X, = 0, X2 = 1, X3; = 1. Then fi(X) = (0,1) = ae, fo(X) \n(1,1) = az and f3(X) = (0,1) = ae. So gs(X) = ae + avaz + azar = \na. = (0,1). And so \u00a23(X) = O and \u00a2g5.(X) = 1. The corrector sub- \nsystem uses the outputs y; = (ya,v2), 7 = 1, 2, 3, 4, 5, of the encoder \nsubsystem as inputs. The final outputs 8; = (8a, Bi), 7 = 1, 2, 3 of \nthe corrector subsystem will be built up in several stages. From Section \nVib) = va + vn t+ v2 + \u00a53 and db: = ys + \u00a51 + avy + ayy;. At the \nfirst stage the corrector subsystem synthesizes n; = avy,,7 = 2, 3. The \ntruth tables for (na, 72), 7 = 2,3 are given below:\n\nThe addition between the binary variables in (22) is modulo 2 addi- \ntion. At the third stage, the three binary 2-tuples, \u00ab , \u00ab& and e,, which \nare the first three coordinates of the error vector \u00a2 are synthesized as \nBoolean functions of the 6\u2019s. The part of the truth table in which at \nleast one of the e\u2019s takes the value | is given below:\n\nThe truth table given below is computed from the rules given in \nSection V. In the case of our example,\n\nSuppose 6); 0, db: 1, do; 1 and 602 1. Then 6, a\u00bb) and bo 3 \nSince 6s aoa, and 6, a: , the vector 6 is a scalar multiple of the \nsecond row vector of 1/7. Therefore it follows that \u00ab ao, \u20ac a. and \n6; = ay. Hence, ey QO, \u20ac12 O, \u20ac2) QO, \u20ac20 1, \u20ac3) Q and eg\u00bb l.\n\nsmall and the Boolean functions required to be synthesized were chosen \nto be very simple. Therefore the corrector subsystem would probably \nrequire more equipment than the encoder subsystem. However, it \nshould be noted that the design of the corrector subsystem is independent \nof the number of binary input variables and the nature of the original \nBoolean functions. This design depends only on p and k. Therefore \nwhen the number of input variables is large and the Boolean functions \nrequired to be synthesized are complicated, the amount of equipment \nrequired for the corrector subsystem may be small in comparison to \nthat required for the whole system. This is very desirable, since we \nassume that the corrector subsystem is highly reliable. The example \nshows how we can build up the logical design of the corrector sub- \nsystem in any general case. However, the author believes that it. is \npossible to build up much more economical corrector subsystems using\n\nlonger time to correct the errors. Such economical corrector subsystems\n\nliable systems which correct faults of more than one block are also given. \nVil. ACKNOWLEDGMENTS\n\nThe author wishes to thank T. H. Crowley, D. B. Armstrong, J. P. \nRunyon and B. A. Tague for many stimulating and useful discussions.\n\n1. Armstrong, D. B., this issue, p. 577 \nMoore, kK. F. andShannon, C. E., Reliable Circuits Using Less Reliable Relays, \nJ. Frank. Inst., 262, 1956, pp. 191; 281 \n3. Tryon, J. G., Redundant Logic Circuitry, U.S. Patent No. 2,942,193 \nVon Neumann, J., Probabilistic Logics and the Synthesis of Reliable Organisms \nfrom Unreliable Components, Automata Studies, Annals of Math. Studies \nNo. 34, Princeton Univ. Press, Princeton, N. J., 1956, pp. 44-98 \nLofgren, L., Automata of High Complexity and Methods of Increasing Their \nteliabilitv by Redundancy, Inf. & Cont., 1, 1958, p. 127 \nHamming, R. W., Error Detecting and Error Correeting Codes, B.S.T.J., 29, \n1950, p. 147 \nBose, R. C., Mathematical Theory of the Symmetrical Factorial Design, Sank \nhva, 8, 1947, p. 155 \nZierler, N., A Class of Cyelic, Linear, Krror-Correcting Codes in p\u2122 Symbols \nGroup Report 55-19, Lincoln Laboratory\n\nHelix waveguide consisting of closely wound insulated copper wire \ncovered with an electrically absorptive or reactive jacket (Fig. 1) is a \ngood transmission medium for circular electric waves.\u2019 In long distance \ncommunication with waveguides it is useful as a mode filter, for nego- \ntiating bends or particularly as the transmission line proper instead of a \nmetallic waveguide.\u201d\n\nThe loss of circular electric waves decreases steadily with frequency \nonly in a perfect metallic waveguide.\u201d\" A similar situation prevails for \nhelix waveguide. Any deviations from a round and straight guide will \nadd to the transmission loss. At such imperfections power is converted \nfrom the circular electric wave into other propagating modes and re- \nconverted. The mode conversion-reconversion effects increase the loss \nand degrade the transmission characteristics.\n\n* Parts of this paper were presented at the I.R.E. Professional Group on Mi \ncrowave Theory and Techniques Symposium, San Diego, California, 1960\n\nwhere A, are the amplitudes of the TE,, or TM,, modes normalized \nwith respect to power and x, are their propagation constants. z is the ax- \nial coordinate. A square matrix with only the diagonal terms ky, = jha,\n\nThe perfect helix waveguide may be considered a perturbed metallic \nwaveguide. The boundary conditions for the tangential electric field\n\nIt is now expedient to transform from the normal modes of metallic \nwaveguide A, to the normal modes of helix waveguide L,\n\nIts transpose is equal to its inverse. The diagonal terms y,,, of T are the \npropagation constants of normal modes in helix waveguide. The circular \nelectric wave remains unaffected by this transformation :\n\nIt is now possible to consider an imperfect helix waveguide, perturbed \nby curvature or cross-sectional deformations in the same terms. The \nmatrix AK\u2019 then has off-diagonal elements also in the row and column \nassociated with the circular electric wave:\n\nThese new off-diagonal elements represent coupling between the circular \nelectric wave and other modes in an imperfect helix waveguide. They are \nthe coupling coefficients of generalized telegraphist\u2019s equations. When \nMaxwell\u2019s equations for the imperfect helix waveguide are converted \ninto generalized telegraphist\u2019s equations, it is found that the coupling \ncoefficients xo, in K\u2019 for curvature and cross-sectional deformation are \nindependent of the wall impedance and the same as in metallic wave- \nguide. Generalized telegraphist\u2019s equations for deformed helix waveguide \nare found in the Appendix.\n\nThe independence of the xo,\u2019s from the wall impedance has an impor- \ntant bearing on the coupling coefficients between the normal modes in \nhelix waveguide. If the new K\u2019 for the imperfect helix waveguide is trans- \nformed by the previous modal matrix L, there results\n\nThe new matrix I\u201d has off-diagonal elements cp, which are caused by the \nimperfection. They are the coefficients of coupling between the normal \nmodes of the perfect helix waveguide. In terms of the elements of K\u2019 \nand L they are:\n\nm=1 \nIf this expression is squared and summed over all n, then, since the l\u2019s \nare elements of an orthonormal matrix, the resulting expression does not \ncontain any l\u2019s:\n\nIt is recalled that the \u00abxo, are coupling coefficients between normal \nmodes in an imperfect metallic waveguide. Consequently, the sum of the \nsquares of the coupling coefficients co, in an imperfect helix waveguide is \nindependent of the wall impedance and the same as in metallic wave-\n\nThe following statement will be proved: The average mode conver- \nsion loss of certain kinds of random imperfections in helix waveguide \ndepends only on the sum of the square of all coupling coefficients.\n\nTo determine mode conversion, first of all, generalized telegraphist\u2019s \nequations have to be solved. With the elements of the perturbed IT \nmatrix the coupled line equations are\n\nWhen only a circular electric wave of unit amplitude, /(0) \nincident and the imperfections are small, a first-order solution at z \naL eL\u2014u\n\nThe coupling coefficients are proportional to the geometric imperfec- \ntion 6:\n\nCon( 2) = C'o,6(z). ( 15) \nLet the imperfections be a stationary random process with covariance \np(u) = (6(z)\u00e96(z2 + u)) (16)\n\nIf, furthermore, the correlation between imperfections any appreciable \ndistance apart is small the covariance drops off very rapidly with in- \ncreasing argument. Then in the expression for (| /o(L) | ) the exponential \nand the factor (lL. \u2014 uw) are constant for any weight of p(w), and one \nobtains:\n\nthe spectral density S of the random imperfections 6 and the coupling \nfactors Co,\n\nWith (12) and (15), it may be concluded from (19) that the average loss \nfor circular electric waves in an imperfect helix waveguide is independent \nof the wall impedance and the same as in metallic waveguide that has \nthe same geometrical imperfections.\n\nThe above derivation has assumed the covariance to drop off fast or \nthe correlation distance to be small. This is the case for any imperfection \ncreated in the manufacturing process of the waveguide. Any manufac- \nturing imperfections some reasonable distance apart are hardly corre- \nlated to each other. The effects of manufacturing imperfections for helix \nwaveguide are therefore the same as in metallic waveguide. It is rela- \ntively easy to determine tolerances for metallic waveguide.\u2019 The above \nrule lets these tolerances be valid for helix waveguide and the very in- \nvolved calculations for helix waveguide are not necessary.\n\nBefore accepting this rule the range of correlation distance for which \nit is valid must be examined. As a typical example, the covariance has \nbeen assumed exponential:\n\nThe average TE, loss at 55 kme has then been calculated for various \nhelix waveguides of 2-inch inside diameter as a function of the correla-\n\ntion distance Ly.\u2019 Fig. 2 shows for deformed helix waveguide the rms of \nelliptical diameter differences which increase the TEo; loss by 10 per cent \nof the loss in a perfect copper pipe. Up to a correlation distance of one \nfoot the curves almost coincide and indicate independence of the wall \nimpedance.\n\nFor a curved helix waveguide the range is even larger. Fig. 3 shows for \na curved helix waveguide the rms curvature under the same conditions. \nRandom curvature of up to a 10-foot correlation distance causes nearly \nthe same average TE, loss in helix waveguide and in metallic waveguide.\n\nFor a correlation distance larger than 10 feet there is an ever growing \ndependence of curvature loss on wall impedance. But such curvature \ndistributions do not occur in the manufacturing process. They are, how- \never, representative of laying tolerances when the waveguide is installed \nwith long bows to follow right of ways or the contour of the landscape. \nA properly designed helix waveguide can tolerate much more laying\n\nFig. 3 \u2014 TE: loss in round waveguide with random curvature, 2-inch inside \ndiameter, at 55.5 kme.\n\nGeneralized Telegraphist\u2019s Equation for Noncylindrical Helix Waveguide\n\nMaxwell\u2019s equations have been converted into generalized teleg- \nraphist\u2019s equations for curved helix waveguide elsewhere.\u201d They have \nbeen represented in terms of normal modes of the metallic waveguide as \nwell as in terms of normal modes of helix waveguide. In the former repre- \nsentation the coefficients of curvature coupling between circular electric \nwaves and the modes of metallic waveguide were independent of the wall \nimpedance and the same as in metallic waveguide.\n\nThe same is true for any cross-sectional deformation in helix wave- \nguide. To prove this, Maxwell\u2019s equations will be converted into gen- \neralized telegraphist\u2019s equations in terms of modes of metallic waveguide \nfor the deformed helix waveguide. This representation is different from \nanother analysis, where the equations are written in terms of the normal \nmodes of helix waveguide.\u201d\n\nwhere a is the nominal radius and the deformation 6 is for the moment \nassumed to be only a function of \u00a2 of cylindrical coordinates (rgz), then \nthe boundary conditions at r = a, are\n\nug \nThe fields at r = a; can by series expansion be written in terms of the \nfields at r = a. The boundary conditions then take the approximate \nform: \nOK \n_\u2014_ - \u00a5 ao \nor\n\nMaxwell\u2019s equations in cylindrical coordinates for exponential time de- \npendence e\u201d*' are\n\nwhere \u00bb and \u00a2 are permeability and permittivity of the waveguide \ninterior. \nThe electromagnetic field is derived from two sets of wave functions,\n\n| 0 oT 0 faT \u2014 \n\u2014{r\u2014|]+. ; ea \nr Lor or dg \\rdg \nwhere x is a separation constant, which takes on discrete values for the \nvarious normal modes. The 7'-functions are normalized so that\n\n| (flux T) (flux 77) dS = x | T\u2019 dS = 1, \nwhere S is the nominal cross section of the guide and the gradient and \nflux vectors of 7 are defined by: \na. isc \ngrad, 7\u2019 = \u2014, grad, T = \u2014,\n\nTo transform Maxwell\u2019s equations into generalized telegraphist\u2019s \nequations the series expansions (32) are substituted for the field com- \nponents in (26) through (31). Certain combinations of the equations \nare integrated over the nominal cross section, and advantage is taken of \nthe orthogonality relations (36) and (37).\n\nMODE CONVERSION IN METALLIC AND HELIX WAVEGUIDE 625 \n(25) is substituted for EF. . In the second term (31) is substituted for F \nMaas tes ; ot IT (x) OT (im\n\n2r \u201crn \u201crr? alr ao aber \ng 07 n| OF tn\u2019 2 Ok, 07 m) \n+I | (1 \u2014 6) \u2014 pe dy a | 6\u2014 \u2014 de, \n0 0g or a 0 or or a \nwhere h\u201d = wpe \u2014 x\u2019. \nAdd 07;,)/dr times (26) and 07\u2019, \nover the nominal cross section:\n\nAdd \u20140T (\u00bb)/rdg times (29) and d7;,,,;/dr times (30) and integrate over \nthe nominal cross section:\n\ndl m] . , o f ry \u2019 \n\u2014* Jwe Vim Xim] | H, Tm) aS. (43) \ndz Js \nlor the right-hand side of (43) integrate 7\u2019,,,,; times (28) over the nom-\n\nAL im \ndz \ndL tm Rim) , . a r 2 2 \u2122 rr rry \nJ } ay = =z J (n)JX{m] X{[n] / 6 T' tm F ke dg. \ndz Wu Wu on 0 \nAll the integrals are taken along the nominal circumference. \nAlternatively, in terms of voltages and currents the equations are \nmore conveniently written in terms of amplitudes A of forward- and B \nof backward-traveling waves. The mode current and voltage are related \nto the mode amplitudes by \nV, = VK, (A, + B,), \nI \n(A, \u2014 B,), \nwhere K,, is the wave impedance: \n. he, \u2019 \nKin) = : K,, (52) \nwe Ryn)\n\nIf the currents and voltages in the generalized telegraphist\u2019s equations\n\nMODE CONVERSION IN METALLIC AND HELIX WAVEGUIDE \u2014 625 \nare represented in terms of the traveling wave amplitudes, the system of \ncoupled equations can be written in matrix notation as\n\nrepresents the amplitudes of metallic waveguide modes. The square \nmatrix\n\n626 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1961 \nLet the first element of A be the amplitude Ao, of a circular electric \nwave. The propagation constant of this wave is\n\nMorgan, 8S. P., Mode Conversion Losses in Transmission of Circular Electric \nWaves through Slightly Noneylindrical Guides, J. Appl. Phys. 21, 1950, p. \n329.\n\ntowe, H. EK. and Warters, W. D)., Transmission Deviations in Waveguide Due \nto Mode Conversion: Theory and Experiment, Proce. I..E., 106, Pt. B, \nSuppl. 18, 1959, p. 30\n\nSchelkunoff, S. A., Conversion of Maxwell\u2019s Equations into Generalized Teleg \nraphist\u2019s Equations, B.S.T.J., 34, 1955, p. 995\n\nIn a perfect helix waveguide the circular electric wave loss is increased \nby eddy currents, finite pitch of the helix, radiation through the wire spac- \ning and effects of the wire coating. Only the contributions from eddy cur- \nrents and pitch are large enough to limit wire size and spacing.\n\nExperimental helix waveguides have tilted turns. These tilts cause cou- \npling between circular electric and unwanted modes. From the coupling \nbetween modes in curved and in offset helix waveguide, the coupling in a \nhelix waveguide with tilted turns is found. For helix waveguide with slightly \nirregular winding of arlitrary form, generalized telegraphist\u2019s equations \nare derived.\n\nTilts and other irregularities in the winding increase the circular elec- \ntric wave loss. The average increase is a function of the covariance of irregu- \nlarities. Winding tilts with an exponential covariance and an rms value of \n0.6\u00b0 increase the TK, loss in 2-inch inside diameter wavequide at 55 kme \nat the most by 10 per cent of the loss in a perfect copper pipe with smooth \nwalls. Present fabrication procedures insure a smaller wire tilt than this.\n\nHelix waveguide consisting of closely wound insulated copper wire \ncovered with an electrically absorbing or reactive jacket is a good trans- \nmission medium for circular electric waves.\u2019 In long distance communica \ntion with circular electric waves it is useful as a mode filter, for negotiat \ning bends or particularly as transmission line proper instead of a plain \nmetallic waveguide.\n\nAs in metallic waveguide, the loss of circular electric waves decreases \nsteadily with frequency only in a perfect helix waveguide. Any devia- \ntions from a round and straight guide and from a uniform and low pitch \nwill add to the loss of circular electric waves.\n\nDeviations from straightness and deformations of the cross section of \nhelix waveguide have been analyzed before and their effect on circular \nelectric wave transmission has been determined.\u2019 When these imperfee-\n\ntions are caused in the manufacturing process they are statistically dis- \ntributed over the guide length with a small correlation distance. Then \nthey add nearly the same average loss to the circular electric wave\n\nstraightness and for cross-sectional deformations are therefore the same \nfor helix waveguide as they are for metallic waveguide.\n\nDeviations of the winding from a low-pitch uniform spiral are imper- \nfections peculiar to the helix waveguide. Their effect on circular electric \nwave transmission will be analyzed here and tolerances on the winding \nof the helix waveguide for low-loss transmission will be determined.\n\nHelix waveguide is usually wound from round wire with an insulating \nlayer. Even when such a helix is perfectly accurate and uniform its differ- \nences from a smooth metallic waveguide add to the circular electric \nwave loss. The various effects can be listed as follows:\n\nThe circumferential wall currents of circular electric waves are uni- \nformly distributed in a smooth wall. In the spaced wires of the helix \nwaveguide their distribution is nonuniform. The heat loss is therefore \nincreased over the smooth wall loss. In Fig. 1 this loss increase is plotted \nover the spacing for a wire size small compared to the wavelength, using \nMorrison\u2019s calculations.\n\n2.2 Pitch! \nIf the helix of radius a is closely wound from a single wire of insulation \ndiameter D, then the pitch angle y is given by \nD \ntany = 5\u2014. \n2a \nIf for a faster manufacturing process n wires are wound simultaneously, \nnD\n\nThe wall currents of circular electric waves are strictly circumferential. \nIn the helix waveguide their path is disturbed by the finite pitch. Power \nof circular electric waves is dissipated into the wall impedance Z, which \nthe surrounding jacket presents to the waveguide interior. The added \ncircular electric wave loss due to finite pitch is\n\nwhere Zy = VWyu/e is the wave impedance and k = w/e the propaga- \ntion constant of free space. k,, is the mth root of J;(.c) = 0, and h,,\n\nA reactive jacket will not dissipate any power. A helix waveguide, \ndesigned for transmitting the circular electric wave around bends, has a \nquarter wave jacket with a very large wall impedance.\u201d In this case, to \nkeep a, low, \u00a5 has to be chosen small.\n\nEven though the helix is closely wound the wires are spaced by the \nwire insulation. With the electric field of circular electric waves parallel \nto the wires, the space between acts as waveguide below cutoff. Being \nshort, this cutoff waveguide will transmit some circular electric wave \npower, which is then absorbed by the jacket. The circular electric wave \nloss caused by this power absorption has been investigated for various \nforms of wire cross section.\u2019 For round helix wires this loss is so small \ncompared to the eddy current losses of Fig. 1 that it may be entirely \nneglected for any wire spacing. Consequently the increase in eddy current \nlosses, rather than the power dissipation through the gaps, limits the \nwire spacing.\n\nThe insulating layer of the helix wires adds to the circular electric \nwave loss in two different ways. Its dielectric constant tends to concen-\n\ntrate the electric field into the layer. Thus the wall currents and the wall \ncurrent losses are increased. In addition, the finite loss factor of any \ninsulating material causes dielectric losses in the small but finite electric \nfield of the circular electric wave. Both of these effects can be calculated \nwith a sufficient approximation from attenuation formulas for the round \nwaveguide with a dielectric lining.\u201d\n\nAgain, it is found that the effects of the insulating layer are so small \ncompared to the eddy current losses that they may be neglected.\n\nNumber of wires, wire size, and wire spacing through insulation are \ntherefore determined by the pitch effect of (2) and the eddy current loss \nof Fig. 1. To speed the winding the numbers of wires should be large. \nTo increase the effects of a reactive or resistive jacket on unwanted \nmodes the wires should be widely spaced.\u201d The increase in circular elec- \ntric wave loss from Fig. | and equation (2) sets a limit, however, to num-\n\nThe preceding discussion has considered only the loss in a perfectly \nwound helix waveguide. A practical helix will not have perfectly uniform \nwindings. One imperfection in particular has been most notable in re- \nsearch models of helix waveguide made by Bell Telephone Laboratories. \nThis imperfection is tilts in the winding.\n\nAside from the finite pitch, a single turn of the helix is usually not in a \ntransverse plane, but is slightly inclined and forms a small angle @ with \nthe axis. Even an improved winding method with an automatic feed \ncontrol has not entirely eliminated this inclination.\u201d\n\nSuch inclined helix turns give rise to mode conversion. There is a simple \nway to analyze circular electric wave propagation in helix waveguide \nwith nonuniformly tilted winding, in which the results of previous cal- \nculations are used. Consider a perfect helix waveguide, a section of \nwhich between 2 0 and z L has been deflected in an arbitrary \nmanner by 2x(z), as shown in Fig. 2. With this deflection is associated a \nchange of guide direction dx/dz and for gentle deflections a curvature \n1/R dx (dz.\n\nOne way to calculate propagation through this deflected section is to \nuse the formulas for wave propagation in the curved helix waveguide\u2019 \nand evaluate them for the curvature distribution d\u00b0r/dz\u00b0. Thus, when\n\nis curvature coupling between a circular electric mode m and the modes n\n\nFig. 2 \u2014 Curved helix waveguide as superposition of offset and tilted winding\n\nof first order circumferential dependence (i.e., TE), and TM,,). The \ncoupling coefficient is\n\nwhere k, k,, and h,, have the same meanings as in (2). k, is the radial \npropagation constant of the coupled mode n and h, Vk? \u2014 (k,?/a?) \nis the axial propagation constant of the coupled mode n. NV, is a normali-\n\nA mode m of unit amplitude incident at z = 0 converts power in the \ndeflected section to the coupled modes n. For gentle deflection the ampli- \ntude and phase of mode n at z = L is given by\n\nAnother way to calculate propagation through the deflected section \nis to consider it as a continuously offset waveguide with a continuously \nvarying tilt 6 of the winding. Both the offset x and the tilt 6 = dxr/dz \nwill then cause coupling between a circular electric mode m and modes n \nof first-order circumferential dependence. The coupling coefficient for \noffset has been calculated\u00ae before:\n\nEquation (10) should give the same result as (6). Integrating by parts \nbrings (10) into a form that can be directly compared with (6):\n\nEquations (11) and (6) can only given identical results when \nCo ; C, \ni me @ \n(hyn \u2014 h,)? hn \u2014 h, \nHence the coupling coefficient for tilted winding is \nCo \nMi, = (h, \u2014h,)Ce+ . \nh,, a hin \nSubstituting from (4) and (8) into (12), and from (12) into (9):\n\nWith these coupling coefficients, generalized telegraphist\u2019s equations\n\nThe summation in (14) is to be extended not only over all the modes n \nbut also over their two polarizations according to the orientation of 6; \n6 may not only be in the plane of Fig. 2, it can also be perpendicular to \nthat plane.\n\nFrom (14) the loss which is added to the mode m by a tilted winding \nmay be calculated. With\n\nThe obliquely wound helix of the preceding section is just a special \ncase of a general irregular winding. In Fig. 3 a turn of such an oblique \nhelix has been drawn in more detail. Aside from a small pitch the wire \nfollows the curve\n\naround the circumference. Its direction deviates by y from the trans- \nverse direction, where\n\n\u00bb SIN pe. (18) \nA summation of cos pg would only add identical terms with different \npolarization; it has been omitted from (18). The boundary conditions\n\nwhere Z is the wall impedance which the outside jacket presents through \nthe helix to the waveguide interior. \nWave propagation in such an irregular structure is best analyzed by\n\ninto generalized telegraphist\u2019s equations\u2019 for the boundary conditions \n(19) and (20).\n\nAn appropriate form of representation is in terms of normal modes of \nthe perfect helix waveguide. With two sets of wave functions\n\nthe normal mode fields of the perfect helix waveguide are the individual\n\nEquation (32) is the characteristic equation for the perfect helix wave- \nguide. Its roots k, = x,a determine the propagation constants \nh, =k \u2014 Yn\n\nof the normal modes of the perfect helix waveguide. \nThe transverse fields of the normal modes are orthonormal in that\n\nThe integral in (34) extends over the cross section of the guide; 6,,,, is \nthe Kronecker symbol.\n\nThe z-dependence of the voltage and current coefficients in (29) is \nfound by substituting the sums of (29) for the transverse field com- \nponents into Maxwell\u2019s equations.\n\nIn the line integral along the boundary, F, from the boundary condition \n(20) is substituted. In the surface integral over the cross section, (30) is \nsubstituted for F, . Subsequently the boundary conditions of the perfect \nhelix waveguide may be used to simplify (38) to first order in y:\n\nExpression (30) for H, holds only for the normal modes of the perfect \nwaveguide. It has been obtained by differentiating the sum (29) for EF, \nin (23). The individual terms of this sum vanish at r = a, while in the \npresent case according to (19) EF, has a finite value there. Hence the \nsum (29) for \u00a3, is nonuniformly convergent and differentiation makes it \ndiverge. To replace H, in (41), substitute in (23) FE, from (29), multiply\n\nThese equations represent an infinite set of coupled transmission lines. \nFor the present purpose it is more convenient to write these transmision-\n\ntraveling wave amplitudes, the following equations for coupled traveling\n\nlor a tilted winding with y from (17), \new jet. \nIn this case circular electric waves interact only with modes of first \ncircumferential order (p 1). Helix irregularities of higher order in \nwill cause coupling to modes of correspondingly higher circumferer tial\n\nThe design of a helix waveguide is started by selecting wire size and \nspacing. A tolerable amount of added TE, loss is specified, and with \nlig. 1 and equation (2) wire size and spacing are determined.\n\nIf, for example, eddy current losses in the helix should not be more \nthan 10 per cent of the loss in a guide with smooth walls, then from Fig. 1 \nthe ratio of wire diameter to insulation diameter should be\n\nTo determine the actual wire size with (2) the wall impedance has to \nbe specified. Different applications of the helix waveguide require differ- \nent values for the wall impedance. A typical and also very critical ex-\n\nample is the helix waveguide for intentional bends. In this case, by sur- \nrounding the helix with a quarter-wave jacket and a metallic shield the \nwall impedance is made very high.\n\nZ=j Kn tan k,, 6, (49) \nWe A \nwhere \u00a2, isthe permittivity of the jacket, h,\u00b0 = avV/ wye, \u2014 h,? the radial \npropagation constant in it, and 6 the relative thickness. \nFor a quarter-wave jacket of a low-loss material the wall impedance is \nreal and approximately\n\nFiber glass laminated with epoxy resin has a relative permittivity at \nmillimeter wavelengths of \u00a2./e9 = 4 \u2014 j(0.04). The relative wall imped- \nance from (49) is then Z/Z\u00bb $1.4. In 2-inch inside diameter wave- \nguide with smooth walls, the TE, loss at 55.5 kme is aga = 2.77 X 10\u00b0 \nLess than 10 per cent of this figure is added to the TE loss in the \npresent example when the pitch is\n\nsoe <4x 10\u00b0. (51) \na \nNo. 37 wire (AWG) with a heavy Formex coat has d 0.0045 and \nD = 0.0054. It very nearly satisfies conditions (48) and (51) when the \nhelix is wound from one wire only (n = 1). Lower wall impedance values \nsuch as are used for helix mode filters or all helix guide would not re- \nquire as low a pitch as (51).\n\nFor the winding process, tolerances for irregularities must \u2018be specified. \nIn (15) the added loss is expressed in terms of the 6 of a tilted winding. \nThe loss caused by higher-order irregularities can also be determined by \n(15) when the corresponding coupling coefficients (47) are substituted.\n\nIn the present problem, however, the irregularities are not known, but \nat best some of their statistical properties are known. Equation (15) \ncan then be used to express the statistics of the loss in terms of the sta-\n\nFor an oblique winding @ is assumed to be a stationary random process \nwith covariance R(u) and spectral distribution S(\u00a3):\n\nIn (52), <2> is the expected value of xr. \nTaking the expected value on both sides of (15) the average added \nloss is obtained in terms of the covariance R(u):\n\nlor a mere estimate the covariance is assumed to be exponential to \nsimplify the calculation: \nTS\n\nS(\u00a3) So (56) \nAS : aaa? 8) \nl + Let \nwith S(\u00a3) nearly flat with spectral density So for mechanical frequencies \nsmaller than & = 1/L,. Lo may be regarded as the cutoff mechanical \nwavelength according to (56) or as a correlation distance according to \n(55). \nThe average added loss is for L >> Lo \n\u2019 P,(2r \u2014 Aa,Lo) \u2014 Q, AB Lo = \n<a,> TiN > 7 93 = \u00b0 (4) \nn AB,? Le + (47 = Aa, lL)?\n\nTo evaluate (57) the characteristic equation (32) of the perfect helix \nwaveguide has to be solved for all the coupled modes n, and propagation \nconstants and coupling coefficients have to be calculated.\n\nThe helix waveguide for intentional bends is again a typical and critical \nexample. In this case Z x at the design frequency and the characteris- \ntic equation (32) simplifies to\n\nThe roots kyo of JpyiCc) = Oand J,4(2) = 0 are good approximations \nfor the roots of (58). With &, Knog & w, where\n\nthe approximations can be sufficiently improved. The coupling coeffi- \ncients in this case are given by\n\nWith these relations, (57) has been evaluated for a helix waveguide \nwith a nonuniformly tilted winding (p = 1).\n\nFig. 4 shows the rms value of @ as a function of the correlation distance \nLy for an added average TE ; loss of 10 per cent of the loss in a copper \npipe with smooth walls. The waveguide diameter is 2 inches and the \nfrequency 55.5 kme. The tolerance is most critical for a correlation dis- \ntance of 1 inch. But even then an rms tilt of 0.6\u00b0 can be tolerated. In \nexperimental models of helix waveguide the maximum tilt has, with \nsome care, been kept below 0.3\u00b0.\n\nIna perfectly wound helix waveguide the circular electric wave loss is \nsignificantly increased only by the eddy current losses in spaced wires. \nThe finite pitch contributes to the circular electric wave loss only when \nthe wall impedance is very high or when the helix is wound from more \nthan one wire.\n\nFig 4. \u2014 TE\u00bb, loss in helix waveguide with random tilt of winding, 2-inch inside \ndiameter, at 55.5 kme. Design for intentional bends with infinite wall impedance.\n\nincrease in circular electric wave loss can be calculated. In a 2-ineh \ninside diameter helix waveguide at 55.5 kme an rms tilt angle of 0.6 \nadds at the most 10 per cent of the loss in a perfect copper pipe to the \naverage TH loss. In experimental models of helix waveguide the maxi- \nmum tilt deviation has been kept below 0.3\u00b0.\n\n. Morgan, 8. P., and Young, J. A., Helix Waveguide, B.S.T.J., 35, 1956, p. 1347 \n2. Unger, H. G., Helix Waveguide Theory and Application, B.S.T.J., 37, 1958, \np. 1599. \n3. Unger, H. G., Noneylindrical Helix Waveguide, B.S.T.J., 40, 1961, p. 233.\n\n}. Morrison, J. A., Heat Loss of Circular Electric Waves in Helix Waveguides, \nI.R.E., Trans., MTT-6, 1958, p. 173\n\n. Katsenelenbaum, B. Z., Attenuation of Ho. Modes in a Helical Waveguide, \nRadiotekhnika i Elektronika 4, 1959, p. 428.\n\n3. Unger, H. G., Circular Electric Wave Transmission in a Dielectric-Coated \nWaveguide, B.S.T.J., 36, 1957, p. 1253\n\n. Beck, A. C. and Rose, C. F. P., Waveguide for Circular Electric Mode Trans \nmission, Proc. I.f.E., 106, Pt. B, Suppl. 13, 1959, p. 159.\n\n. Schelkunoff, S. A., Conversion of Maxwell\u2019s Equations into Generalized \nTelegraphist\u2019s Equations, B.S.T.J., 34, 1955, p. 995.\n\n. Rowe, H. E. and Warters, W. D)., Transmission Deviations in Waveguide Due \nto Mode Conversion: Theory and Experiment, Proc. I.E.E., 106, Pt. B, \nSuppl. 13, 1959, p. 3.\n\nD. B. Armstrrona, B.A., 1940, University of Toronto; M.S., 1951, and \nSe.D., 1955, Massachusetts Institute of Technology; Bell Telephone \nLaboratories, 1954\u2014. He has been engaged in research in switching prob- \nlems, which have included simulation and economic studies of telephone \nsystems, studies of central office systems and reliability studies of digital \nsystems. Member Sigma Xi.\n\nD. W. Bonus, B.S. in E.E., 1937, New York University; Bell Tele- \nphone Laboratories, 1935\u2014. He has been concerned with problems of \nelectric shock and the protection of communication facilities from for- \neign potentials. This has included field investigations of lightning be- \nhavior, design of surge measuring devices and laboratory surge testing of \napparatus. Member A.I.E.E.\n\nG. D. Boyp, B.S., 1954, M.S., 1955, and Ph.D., 1959, California \nInstitute of Technology; Bell Telephone Laboratories, 1959\u2014. He is en- \ngaged in optical maser research.\n\nDonaLp B. Currriss, B.S. in E.E., 1959, Newark College of Engineer- \ning; Bell Telephone Laboratories, 1951\u2014-. Until 1959 Mr. Cuttriss was \na member of the component development department working on design \nand development of voice-frequency laminated-core inductors and s\u00e9mi- \nconductor field-effect devices. In 1959 he transferred to transistor de- \nvelopment and he has been concerned with development of diffused-base \ngermanium transistors and with methods of producing high-quality gal- \nlium arsenide. Member Tau Beta Pi.\n\nA. GARDNER Fox, 8.B., 1934, and 8.M., 1935, Massachusetts Institute \nof Technology; Bell Telephone Laboratories, 1936\u2014. His early work was \nin development of mobile radio transmitters, early radar development \nand general waveguide research. Since 1944 he has been engaged in de- \nsign of radio frequency amplifiers and in research on millimeter waves. \nHe heads a group specializing in microwave physics. Fellow I.R.E.\n\nJames P. Gorvon, B.S., 1949, Massachusetts Institute of Technology, \nM.A., 1951, and Ph.D., 1955, Columbia University; Bell Telephone Lab- \noratories, 1955\u2014. His research in quantum electronics has involved work \non molecular beam masers, paramagnetic resonance and solid state ma- \nsers. Member American Association for the Advancemens of Science, \nAmerican Physical Society, Sigma Xi.\n\nPuitip A. Gres, B.S.E.E., 1956, Carnegie Institute of Technology; \nBell Telephone Laboratories, 1956\u2014-. His work in Systems Engineering \nhas included statistical and economic optimization studies of outside \nplant facilities. Member Eta Kappa Nu, Phi Kappa Phi, Tau Beta Pi.\n\nTinGYE Lt, B.Sc., 1953, University of Witwatersrand (South Africa) ; \nM.S., 1955, and Ph.D., 1958, Northwestern University; Bell Telephone \nLaboratories, 1957\u2014. He has been engaged in studies of microwave an- \ntennas and microwave propagation. Recently he has been primarily con- \ncerned with work on optical masers. Member I.R.E., Eta Kappa Nu, \nSigma Ni.\n\nH. C. Marre, B.S., 1949, and Ph.D., 1956, California Institute of \nTechnology; M.S., 1950, Massachusetts Institute of Technology; Bell \nTelephone Laboratories, 1959-60; associate professor of electrical en- \ngineering, California Institute of Technology, 1953\u2014-. While at Bell Lab- \noratories during a year\u2019s leave of absence from Cal Tech, Mr. Martel was \nengaged in visual and acoustics research related to signal coding and \ndetection. Member I.R.E., Sigma Ni, Tau Beta Pi.\n\nMax V. Maruews, B.S., 1950, California Institute of Technology; \nM.S., 1952, and Se.D., 1954, Massachusetts Institute of Technology; \nBell Telephone Laboratories, 1955-\u2014-. He has specialized in acoustics re- \nsearch in speech transmission and has been especially concerned with \nsimulating speech experiments on a digital computer. Member Acoustical \nSociety of America, I.R.E., Sigma Xi.\n\nR. L. Peek, Jr., A.B., 1921, Columbia College; Met.I\u00e9., 1923, Colum- \nbia School of Mines; Bell Telephone Laboratories, 1924\u2014\u2014. His early \nwork was in materials testing. In 1936 he turned to apparatus develop- \nment involving coin collectors, electromagnets, relays and switches. Dur-\n\nspring relay development. Since 1951 he has been in charge of a group \nstudying new electromagnetic devices. Member A.1.E.E.\n\nD. K. Ray-Cuavupuurt, B.Se., 1953, Presidency College, University of \nCaleutta (India); M.Se., 1955, University of Caleutta; Ph.D., 1959, Uni- \nversity of North Carolina; instructor in statistics, Case Institute of\n\nTechnology, 1959-60; Bell Telephone Laboratories, summer, 1960; visit- \ning assistant professor of statistics, University of North Carolina, 1960 \nHe has been engaged in research on problems of constructing experi-\n\nmental designs in statistics, sampling theory and error-correcting codes. \nMember Institute of Mathematical Statistics.\n\nERLING D. Sunpe, Dipl. Ing., 1926, Technische Hochschule, Darmstadt, \nGermany; American Telephone & Telegraph Co., 1927-34; Bell Tele- \nphone Laboratories, 1934\u2014. He has made theoretical and experimental \nstudies of inductive interference from railway and power systems, light- \nning protection of the telephone plant, and fundamental transmission \nstudies in connection with the use of pulse modulation systems. Author \nof Karth Conduction Effects in Transmission Systems, a Bell Laboratories \nSeries book. Senior member I.R..; member American Association for \nthe Advancement of Science, A.I.E.E., American Mathematical Society.\n\nHans-GrorG UnGer, Dipl. Ing., 1951 and Dr. Ing., 1954, Technische \nHochschule, Braunschweig (Germany); Siemens and Halske (Germany), \n1951-55; Bell Telephone Laboratories, 1956\u2014. His work at Bell Labora- \ntories has been in research in waveguides, especially circular electric wave \ntransmission. He is now on leave of absence from Bell Laboratories as \nprofessor of electrical engineering at the Technische Hochschule in \nBraunschweig. Senior member I.R.I.; member German Communication \nengineering Society.", "title": "The Bell System Technical Journal  1961-03: Vol 40 Iss 2", "trim_reasons": [], "year": 1961}
{"archive_ref": "bitsavers_BellSystemJV31N05195209_10493720", "canonical_url": "https://archive.org/details/bitsavers_BellSystemJV31N05195209_10493720", "char_count": 189113, "collection": "archive-org-bell-labs", "doc_id": 483, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc483", "record_count": 341, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bitsavers_BellSystemJV31N05195209_10493720", "split": "test", "text": "Auotmatic Switching for Nationwide Telephone Service \nA. B. CLARK AND H. S. OSBORNE 823\n\nFundamental Plans for Toll Telephone Plant J. J. PILLIOD 832 \nNationwide Numbering Plan W. H. NUNN 85] \nAutomatic Toll Switching Systems F. F. SHIPLEY 860\n\nMathematical Theory of Laminated Transmission Lines\u2014Part I \nSAMUEL P. MORGAN, JR. 883\n\nImpedance Bridges for the Megacycle Range H. T. WILHELM 999 \nAbstracts of Bell System Papers Not Published in this Journal 1013\n\nTHE BELL SYSTEM TECHNICAL JOURNAL is published six times \na year by the American Telephone and Telegraph Company, 195 Broadway, \nNew York 7, N. Y. Cleo F. Craig, President; Carroll O. Bickelhaupt, Secretary; \nDonald R. Belcher, Treasurer. Subscriptions are accepted at $3.00 per year. \nSingle copies are 75 cents each. The foreign postage is 65 cents per year or 11\n\ncess of the No. 4 installation at Philadelphia, led to studies of the \nservice and operating results which might be expected from a nationwide \nextension of automatic switching. The conclusion was reached that this \nwould be a desirable objective of the Bell System companies and would \nresult in a very substantial further improvement in the speed and ac- \ncuracy of handling of long distance messages. Accordingly, during the \nnext few years, a national plan was prepared and was adopted by the \ntelephone companies.\n\nThe features of this nationwide plan and the present status of its \napplication form the subject of the three technical papers which accom- \npany this introductory paper.\u201d \u201c\u00b0 The basic requirements to be met* \nin the development of this plan included the following:\n\n1. It should be suitable for the nationwide extension of automatic \nswitching both by originating toll operators and by the customers direct.\n\nWhen this work was commenced it was clear that a program leading \ntoward general nationwide operator dialing was desirable. Subsequent \ndevelopments have confirmed the wisdom of making the basic plan \nconsistent with general nationwide customer dialing as well since it now \nappears that a very wide extension of this form of service will become \ndesirable.\n\n2. The plan must provide for satisfactory overall service between \nany two telephones in this country and Canada.\n\nUnder manual operation satisfactory overall service was provided \nfor by the general toll switching plan in use since about 1930. This plan \nis modified to recognize the far greater speed and accuracy of automatic. \nswitching compared with manual switching. This involves also modifi- \ncations of transmission design standards so that the overall connections \nwill continue to be satisfactory.\n\n3. The system must be designed for instantaneous service, so that \ndelays due to lack of circuits or equipment would be very infrequent. \nThis is necessary, both from the standpoints of service and the avoidance \nof tieups, particularly of the automatic switching machinery.\n\nA trunking system must therefore be devised which will most economi- \ncally meet this requirement, considering overall costs of lines, switch- \ning equipment and operation.\n\n4. Machines must be designed for use at strategic points in the net- \nwork, called \u201ccontrol switching points\u2019, to perform automatically the \nvarious tasks required to make the overall plan operative and economical.\n\ntype of local central office equipment, called No. 5 crossbar, which was \ndesigned with this in view.\u2019 For older types of equipment, the job is \nmore difficult.\n\n2. The switching equipment must be provided with automatic means \nfor recording all of the information necessary for charging the call. In \nthe case of operator dialing this is now done manually by the operator.\n\nGreat advances have been made in recent years in the development of \nautomatic message recording equipment. In 1944 there was placed in \nservice in California the first installation in this country of automatic \nticketing equipment.\u201d This equipment is associated with step-by-step \nlocal switching equipment and automatically prints for each call a \nticket similar to that prepared by the operator with manual operation. \nIn 1948 there was installed in Media, near Philadelphia, a greatly im- \nproved type of message recording equipment in which the information \nappears in the form of punched holes in a tape.\u2019 This equipment is \nmuch more economical than the earher system and also lends itself to \nthe automatic preparation of toll statements or bills.\n\nThe present forms of equipment have been designed to be associated \nwith local central offices. A careful study has been made of their field \nof application and cf the basic plan necessary to provide for a general \nnationwide extension of customer dialing. This indicates that there will \nbe a large field for automatic message accounting equipment associated \nwith the toll network and arranged to receive orders for toll messages \nfrom a number of local dial offices. This centralized AMA equipment, \nas it is called, is under development and an initial installation will be \nmade next year in Washington, D. C. In this installation the range of \ncustomer dialing will be limited and certain service features will be \nlacking, which it is planned to add later.\n\nRecognizing that the best way to develop these questions is a trial, \narrangements were made to open such a trial last fall at Englewood, \nN. J. This office is equipped with a No. 5 crossbar system so that arrange- \nments for such a trial could readily be made there. The Englewood \ncustomers are able to dial directly any of about eleven million telephones \nin ten metropolitan areas scattered throughout the country, including\n\nAll of these plans depend upon the successful development of striking \ninnovations in toll switching and automatic message accounting equip- \nments. The plans in turn react upon the features to be incorporated \nin such equipments and upon the schedule of their development. Mr. \nShipley\u2019s paper, pages 860 to 882, tells about the more important fea- \ntures of these equipments and the problems which are involved in their \ndevelopment.\n\nExperience with operator toll dialing shows clearly that it provides a \nmarked improvement in toll service. This improvement will increase as \nprogress is made toward the full application of the nationwide automatic \nswitching plan. j\n\nThe development of long distance dialing by customers is at an early \nstage. The results of recent trials, however, indicate that nationwide \ncustomer dialing has service advantages and will generally be received \nwith enthusiasm by telephone users. It is anticipated, therefore, that \ncustomer dialing will rapidly expand both on a regional and on a nation- \nwide basis.\n\nSwitching plans providing for the systematic routing of toll telephone \ntraffic have been employed by the communication industry for many \nyears. These plans have contributed directly to the high quality of long \ndistance telephone service enjoyed by the public in the United States \nand Canada. This generally excellent service is the result of the coopera- \ntive work of many organizations including the Bell Operating Companies, \nmany independent connecting Companies and others in the United \nStates as well as in adjoining countries. The techniques employed today \nreflect a great amount of research and engineering and improvements in \nmanufacturing skill and in construction, maintenance and operating \nmethods developed over a period of many years.\n\nThroughout the United States and Canada there are approximately \n20,000 different places \u2014 cities, towns, and villages \u2014 that serve as toll\n\nIn order to illustrate the problem a specific example may be useful. \nFig. 1 is a map of Wisconsin and Minnesota on which nearly 1200 circles \nindicate points at which exchange facilities may\u2019 be connected to the \ntoll network. The extent of the coverage in this area is typical of that \nfound throughout the country.\n\nMore than 1,000 smaller circles on the map represent \u201ctributaries\u201d \u2014 \nthat is, towns where little or no toll operating is done. Toll connections \nto and from these points are completed at the toll centers which in gen- \neral do the toll operating required.\n\nIn the United States and Canada as a whole, there are approximately \n2,600 toll centers. The remainder of the toll connecting points\u2014about \n17,500\u2014are tributaries.\n\nFig. 2 gives an idea of the variety and complexity of the network of \ncircuit groups required to interconnect the toll centers in one area. Here \neach line represents a group of circuits, known as \u201cintertoll trunks,\u201d \nbetween two toll centers. Each group may contain anywhere from one \nto several dozen trunks. The location of the lines on the map is unrelated \nto the geographical routing of the trunks, and only a part of the circuit \ngroups are shown. To get a complete picture one should visualize that a \ncluster of relatively short circuit groups radiates from each toll center \nto its tributaries, of which there may be up to 15 or more.\n\nPhysically, the plant consists of a network of open wire lines, cables \nand radio systems. On these, voice frequency or carrier operation is \nemployed in each section as required to provide the necessary intertoll \ntrunks. The routes of the lines in Minnesota and Wisconsin are shown \nby Fig. 3. In this area there are no radio routes carrying telephone cir- \ncuits, but a radio system between Chicago and Minneapolis is in the \nplanning stage.\n\nAreas like Wisconsin and Minnesota must, of course, be connected to- \ngether, and Fig. 4 shows the major Bell System toll routes that accom- \nplish this. On a map of this kind it is not possible to include anything \nlike the detail shown in Fig. 3. One must visualize, therefore, that each \nstate contains a network of routes generally comparable to those shown \nfor Wisconsin and Minnesota.\n\nThis then represents the interconnection problem to be met by an \norderly switching plan that will provide efficient, reliable and fast toll \ntelephone service between any two points.\n\nextensive use of carrier than would have been practicable with a slower \nrate of growth.\n\n- Consideration of these factors which offer an opportunity to improve \nservice has led to the gradual reorientation of the fundamental plans \nfor the intertoll trunk plant which is now under way.\n\nin turn one segment of the Chicago \u201cregion\u201d which serves a somewhat \nlarger area than shown by Fig. 5.\n\nUnder this arrangement, toll calls between two tributaries in the Hib- \nbing toll center area can be completed by switching at the toll center. \nIn a similar manner, any two points within the Duluth tandem outlet \narea can be served by switching at Duluth. The same treatment also \napplies for connections between any two points in the same sectional \ncenter area or in the same regional center area. For example, a connection \nfrom Hibbing to any point within the Chicago region (which involves \nmore than six states as shown in Fig. 7) requires no more intertoll links \nthan Hibbing to Duluth, Duluth to Minneapolis and Minneapolis to \nChicago, and a corresponding number of links on through to another \nsectional center, and primary or tandem outlet to the toll center desti- \nnation. Circuits between the toll center and tributaries are not referred\n\nFig. 5\u2014Intertoll trunks between Davenport, Iowa and Hibbing, Minnesota, \nshowing alternate routing possibilities.\n\n\\ , \n\u2014, \\ peceann \u2018 \n-.. 8B \\ \nSEATTLE \u201c7 Pes i \\ \nf ie ' IN QUEBEC - \\ Gay \n\\ \u00b0\u00b0 erences 5 5 SASK, i ACS? . ( yn \n\\ i oe n 2 \u201csl z \\ \nwi ae | \nae ; y \nPORTLAND S=~~2-~__ : MONTREAL f \npices | * {\\ \ni ee \u201c\u2014\u2014_\u2014 \u2018 QVe:.. \u2018 \n| BISMARCK ae) Bee \n| oe ! \n\u2018 \nBILLINGS t \\ { \n\\ f en ee \\A WHITE RIVER JCTS \na ae het 4 \\ MS IN \nee ! * 1 \nPo | ensue . y aan 9) ee \nfs ss | TORONTO ae \nee 1 i MINNEAPOLIS \\ STEVENS POINT UZ\u201d ALBANY}-7~~ \neae [ | RAPID CITY \\ N \\ SYRACUSE \" Let \n! ae I ~ ~NEW \\ \n/ - i | \\ BUFFALO oe | HAVEN \n! 7 ee \n/ AGASER L lea eee face 13 ae gene \u201c\\ MADISON _---TSCRANTON, | \nwee LL 7 i \u2014 nad ~S \ni ' ~--~J{@ Sioux ciTY SINAN EEN NE \u00e9s | / \ni | a !__-\u20ac$ \u2018 L \u2019 NEWARK NEW YORK \nI \u2019 \\ DES MOINES pAvENPoRT SouTH [TOLEDO \u00a2, - ; / , \nSACRAMENTO Toe SIDNEY \\ =< \u2014e/ c peo ) \\2eNo CLEVELAND\\ > a aa PHILADELPHIA \nPI SALT LAKE CITY ~ OF \u2014 \\ \u201c7 oy A a? \u00a7 4 \n( ~~Le OMAHA Ae ce : 7 FQRy WAYNE oA \u2014_\u2014- \n\\ \u2018 Lon - \n\\ ! ) ) 1 | 1@ oe \n\u2018 H 'GRAND ISLANDe~ \\ / | Eee, ie BALTIMORE \nOAKLAND \\\\ cnet ory CHAMPAIGN | \\ NDIANAPOLISscouumaus/ \u00a2 ms, \\ \n\u2018 Pes ry | 77\u2019 &@P WASHINGTON \nre oo eee oe i] 7 \no\u2014 MDENVER = 4 ef \n\\ \\ GRAND JUNCTION \u2019 ? \\(SPRINGFIELD Je\u201cenne \u2018 | ACEINCINNATI Se ay \nFRESNO \\, | SALINA \\ . KANSAS \u2018 -\u2014\"-{\\ . BOOTE [> SERN ae Pa Ne . \na ; = #~ city i | 1 Yo { \u2014 fricHMOND \ni LY \n\\ | ! ae ane ae ie ahs /+_j@ LOUISVILLE CHARLESTON ate \n\u2014- ; ee ROA \n_- : oA evaNSviLte | NAS as \n_\u2014e \\ / ono \n\u2014- | PuiARE | i \u00a2 nN ' / \u2014\u2014\u2014 = \n\u2014aP P emere e = 1 \\ Not oe a a a\" Set \n: a | JOPLIN < PPADUCAH Ye ; \net See |, See See \nSAN BERNARDINO Y Lo Cees a Wee So a: Soha 5 a ew ne , KNOXVILLE...\u201d \nyy aK \nLOS ANGELES i TULSAp | re a ee, Cae A CHARLOTTE \n1 AMARILLO 1 4 @ CHATTANOOGA oe oa Be ay \npee pfs | Xvempnis BO oo Raat Bee F \" \n/ e PHOENIX ! i OKLAHOMA cITY ; haan of \\ oe | \u2018 \n\\ J \\ Fs a li \\ iN COLUMBIA \n: ) I ae | LITTLE ROCK | ae H | i he \na 7 ; Te ; [~~ WEST POINT ae \\ATLANTA Ne \nWICHITA FALLS TREO BIRMINGHAM 4 a \nrn 1 NWOOD J nN \n7 | aS nh GREENWOO if PA \\ \\a MACON \\ \nt \\ \n! ) ~~ = 1 \\ \\ \neo f DALLAS HI te \\ i \\ (@} NATIONAL CENTER \nSee | ae a of. lee | SHREVEPORT ! \u201cA 1 MONTGOMERY ; \n. EL PASO assy eee \n3 MIDLAND SWEETWATER ! 1 / JACKSON l : ALBANY WM REGIONAL CENTER \nN ! \\ \ndex anceco NCO ~\\ ; i eissees ~ oe A SECTIONAL CENTER: \nNOTE: THE SHORT LINE FROM EACH PO AND \\ Prep aay ad ee Se JACKSONVILLE \nSC POINTS TOWARD ITS HOME CSP \u2019 \u2018 @ PRIMARY OUTLET \n\u2014-\u2014 REGIONAL SWITCHING AREA BOUNDARIES \nNEW ORLEANS & ee ss \nORLANDO\n\nFig. 7\u2014Tentative locations of control switching points in United States and Canada.\n\ntically all overflows from the high usage groups during the heavy traffic \nperiods. The \u201chigh usage\u2019\u2019 and \u201cfinal\u201d groups which could be used for \nrouting calls between Hibbing, Minnesota and Davenport, Iowa are \nshown by Fig. 5.\n\nTHe expanded general toll switching plan for nationwide dialing con- \ntemplates a degree of alternate routing far in excess of that used with \nthe former switching plan designed for manual operation. This change \nalong with the reduction in toll centers will have a marked effect on the \nnormal flow of many traffic items through the intertoll network. As a \nresult the arrangement of the present intertoll trunks will be significantly \nmodified both in number, routing and terminating points. It is necessary \nto take these facts into account in engineering toll plant additions so \nthat they will lead toward an advantageous layout for future nationwide \ndialing as well as meet the needs of the more immediate future. Fortun- \nately, the effect is in the direction of greater concentration of circuits \nin main routes so that with the new cable and radio facilities available, \nover-all economy and better service should result.\n\nFig. 8\u2014Terminal equipment of type-N1 cable carrier system. Provides twelve \nmessage channels with self contained signaling equipment over two pairs of cable \nconductors in same sheath.\n\nFic. 10\u2014Microwave radio relay tower at Cotoctin Mountain, Maryland, on a \nNew York-Washington radio route. There are 300 message circuits in service with \nmore planned.\n\non the type of carrier system (Fig. 8). Coaxial cable transmission systems \ncurrently provide up to 600 telephone channels per pair of coaxials \n(Fig. 9). A new coaxial system, under development, is expected to pro- \nduce about 1,800 telephone channels per pair of coaxials.\n\nMost of the applications of radio for toll telephone service now contem- \nplated, involve the use of point-to-point microwave systems. By employ-\n\nFig. 11\u2014Growth in Bell System intertoll trunk mileage showing trend toward \nmore extensive use of carrier type facilities.\n\nvantages. They are inherently of the \u2018four-wire\u201d type which minimizes \nthe number of possible singing and echo paths on a circuit. Also, the \nspeeds of propagation over carrier systems are generally higher than over \nvoice frequency systems thereby further minimizing the echo problem. \nThese features are of great advantage in reducing limitations on circuit \ndesign and layouts of the general toll switching plan.\n\nIn addition to the ability to carry messages, intertoll trunks must be \nprovided with suitable signaling facilities.\u201d \u2019 These must provide a means \nof: first, attracting the attention of the distant point, either an operator \nor automatic equipment, to the fact that a connection is to be established ; \nand second, in the case of dial operation, transmitting coded information \nin the form of pulses for establishing the connection; and third, trans- \nmitting a general class of supervisory signals including connect and \ndisconnect signals, on and off switch hook signals, recall signals and\n\nWith dial operation, the number of trunks in tandem in a given toll \nconnection may vary on successive calls. To avoid undesirable trans- \nmission contrasts and other adverse effects, it is important that every \ntrunk be designed to operate as closely as possible to the theoretically \ncorrect transmission loss. The problem is complicated by the fact that \nthe extent to which the echo, noise and crosstalk will limit the perform- \nance of an individual link is not directly proportional to the length of the \ncircuit. In fact, the minimum loss at which a particular circuit used \nsingly or in various built-up combinations can theoretically be operated \ndepends on the number, length and characteristics of the other circuits \nconnected in tandem with it. Arrangements for precisely adjusting the \nloss in the individual trunks for each call would be complicated. Adequate \nperformance can be achieved however by compromise methods which \nprovide for automatic adjustments in the loss of each trunk in accordance \nwith the following:\n\n1. When a trunk is switched to other intertoll trunks at both ends it \nis operated at the minimum loss practicable. This loss is known as \u201c\u2018via \nnet loss.\u2019\u201d? (VNL)\n\n2. When the trunk is switched to another intertoll trunk at one end \nonly, the loss is increased two db.\n\nend a further loss of two db is added. This loss which is four db greater \nthan the via net loss is known as \u201cterminal net loss.\u201d? (TNL)\n\nThe data and methods used in the derivation of the via net loss are \nrather complex and not within the scope of this paper.\n\nThe definite routing patterns established for the toll machine switching \noperation impose more severe transmission conditions on certain classes\n\nof circuits than on others. For example, a trunk in a \u201cfinal\u201d group be- \n~ tween a TC and a PO can become involved in an eight-link connection, \nwhereas a trunk in a \u201chigh usage\u201d group, say, between a PO and another \nPO will not be involved in more than a three-link connection.\n\nThis creates a need and provides an opportunity for allocation of the \navailable facilities among the various trunk groups in a way that will \nprovide the best overall service. For example, to the extent practicable \nit is desirable to assign carrier grade facilities to trunks in \u201c\u2018final\u2019\u2019 groups \nthat may be involved in connections with the maximum number of links. \nFacilities with less favorable transmission characteristics may then be \nreserved for trunks in groups that are used for connections involving \nfewer links.\n\n1. H. 8. Osborne, \u2018\u2018The General Switching Plan for Telephone Toll Service,\u201d\u2019 \nA.I.E.E. Transactions, 49, pp. 1549-1557, 1930.\n\n. C. A. Dahlbom, A. W. Horton, Jr. and D. L. Moody, \u201cApplication of Multi- \nfrequency Pulsing in Switching,\u201d A.J.Z.E. Transactions, 68, pp. 392-396, \n1949.\n\n7. N. A. Newell and A. Weaver, \u2018\u2018Single-frequency Signaling System for Super- \nvision and Dialing over Long Distance Telephone Trunks,\u201d A.J.E.E. Trans- \nactions, 70, (7 pages), 1951.\n\n8. Articles prepared by American Telephone and Telegraph Company for infor- \nmation of Dial Interexchange Committee of the United States Independent \nTelephone Association. Published in Telephony on dates indicated.\n\nIn telephone language a numbering, plan gives each telephone in a city, \na town, or a geographical area an identity or designation different from \nthat given any other telephone in the same area. There is a wide variation \nin the types of numbering arrangements in use today in the Bell System, \nand this paper gives the reasons for this diversity, and examples of the \nvarious numbering plans now in use. With the introduction of modern toll \nswitching facilities and the extension of toll dialing to nationwide scope, \nit was realized that an improvement in the method of dialing toll calls to \ndistant cities was essential in order to realize the maximum speed and \naccuracy inherent in toll dialing. A nationwide numbering plan covering the \nUnited States and Canada has been designed. Each of the more than 20,000 \ncentral offices in the two countries are to be given a distinctive designation \nwhich identifies that particular office. This designation is to consist of a \nregional or area code and a central office code The new switching equipment \nfor the key points in the toll network is being designed so that any toll opera- \ntor, wherever located, will use the same designation or code for reaching a \ngiven office. The combination involved in laying out these areas and the \ncomposition of the area codes are presented. A total of 152 codes are available \nof which approximately 90 are assigned to the present numbering plan areas. \nUltimately each central office will be given a type of number consisting \nof an office name and five numerical digits, such as LOcust 4-5678, in \nwhich the first two letters of the office name become the two letters of the \ncentral office code. The entire program will take a considerable number of \nyears to realize, but 1s one which must be accomplished in order to achieve \nthe best results in operator toll dialing and the ultimate goal of nationwide \ncustomer toll dialing.\n\nIn telephone language a numbering plan is exactly what the name im- \nplies, a plan or system of giving each telephone in a city, a town or any \ngeographical area an identity or designation which is different from that \ngiven every other telephone in this same area. This designation is the\n\nIn many places local service areas were changed so that customers \ncould call into contiguous exchanges at local rates. To enable customers \nto dial into these neary-by places the original numbering plans required \nexpansion to include this increased number of offices. In addition, with \nthe advance in the telephone art many cities introduced equipment for \nautomatic charging on multi-unit and short haul toll calls so that cus- \ntomers could dial such calls directly instead of placing them with an \noperator for completion. In order to enable customers to dial these calls, \nit was necessary to expand the original city numbering plans to encom- \npass wider and wider geographical areas.\n\nIn expanding the various types of numbering plans to serve a larger \nnumber of central offices than were originally anticipated, various ex- \npedients were resorted to. In the largest cities having three-letter office \ncodes a numeral was substituted for the third letter thus very materially \nincreasing the code capacity from about 325 to about 500 and making it \npossible to form a number of codes using the same office name. The name \nCANal for example, instead of serving but one office may serve a number \nof offices, CAnal 2, CAnal 3, CAnal 4, etc. In the medium size cities \nhaving two-letter codes, expansion meant adding a digit to the code to \nall or in some cases to only a part of the offices in the city.\n\nThe five-digit places were usually expanded by adding a digit to \nsome of the numbers so that some of the telephones had five digits and \nothers six digits in their numbers.\n\nAs a result of choosing originally a numbering plan which at the time \nseemed adequate and most suitable for the city involved and in many \ncases being forced to expand to meet changing needs, we now have in the \nBell System a considerable variety of different numbering plans. These \nare given in Table I. The numbering plans given are all adequate to \nserve the present local dialing needs for the cities in which they appear.\n\nHaving reviewed the numbering plan situation as it exists today in \nthe various cities and towns, let us turn to the problem of handling toll \ncalls. Under ringdown operation there is an operator at the outward toll \ncenter where the call originates and another operator at the terminating \nor inward toll center. On built-up toll connections there are additional \noperators at each intermediate toll switching point. The inward toll \noperators, who are familiar with the numbering plans in the offices served \nby their particular toll center, can be relied upon to connect to the de- \nsired station even though there is uncertainty on the part of the calling \ncustomer or the outward toll operator regarding the precise pronuncia- \ntion or spelling of the name of the called office or the particular form of \nnumbering system used at the called city.\n\nPlace Directory Listing Customer Dials Ordinarily Referred to as \nPhiladelphia, Pa. LOcust 4-5678 LO 4-5678 Two-five \nLos Angeles, Cal. PArkway 2345 and | PA 2345 and | Combined two-four \nREpublic 2-3456 RE 2-3456 and two-five \nIndianapolis, Ind. MArket 6789 MA 6789 Two-four \nEl Paso, Texas PRospect 2-3456 | PR 2-3456 and | Combined two-five \nand 5-5678 5-5678 and five digit \nSan Diego, Cal. Franklin 9-2345 F 9-2345 One letter, four and \nFranklin 6789 F 6789 five digit \nDes Moines, Iowa 4-1234 and 62-2345 | 4-1234 and 62- | Combined five and \n2345 six digit \nBinghamton, N. Y. | 2-5678 2-5678 Five digit \nManchester, Conn. | 5678 and 2-2345 5678 and 2-2345| Combined four and \nfive digit \nWinchester, Va. 3456 3456 Four digit \nAyer, Mass. 629 and 2345 629 and 2345 Combined three and \nfour digit\n\nbut the Z is never used in a central office code), hence any office code \nwill always avoid a 1 or a 0 in the first two places. The digits 1 and 0 \ncan therefore be used in area codes to distinguish these from office codes. \nIt is not practical to use them as initial digits of area codes since custo- \nmers dial 0 to reach operators and the local dial equipment is arranged to \nignore an initial 1 for technical reasons. A 1 or 0 in the second place, \nhowever, can be employed in an area code without conflicting with any \ncentral office codes or interfering with any existing practices. Accord- \ningly the area codes will consist of three digits with either a 1 or a 0 as \nthe middle digit, 516, 201, etc. A few codes of this type are now in use, \nleaving a practical total of 152 of these area codes available as compared \nto approximately 90 assigned to our present numbering plan areas. This \nwill provide a comfortable spare for additional future numbering plan \nareas or possibly for reaching overseas points which may later be in- \ncorporated into the toll dialing network.\n\nAs shown in Fig. 1, states and provinces such as Montana or Alberta \nwhich are contained in a single numbering plan area will have area \ncodes with a 0 as the middle digit to distinguish them from areas in \ndivided states such as Texas where the middle digit will be a 1. This is \nto enable toll operators to differentiate between the two classifications \nof areas. On calls to single area states the operators will always know that \nevery call to the state in question uses the one area code, whereas on \ncalls to subdivided states additional information will be required to de- \ntermine which of the several area codes should be employed to reach the \nparticular destination. It is proposed to show on the operator position \nbulletin the codes of all single area states and the codes of all frequently \ncalled cities in multi-area states. The area codes of the less frequently \ncalled places in the multi-area will be obtained from a routing operator.\n\nWithin each numbering plan area each of the 590 or fewer offices \nare to be given a three-digit office code which will be different from that \nof any other office code in that same area. Ultimately each central office \nwill be given a 2-5 type of number consisting of an office name and five \nnumerical digits, such as LOcust 4-5678, illustrated for Philadelphia. \nIn the larger cities customers will dial seven digits, LO 4\u20145678, on local \ncalls to numbers in the same exchange. In many of the smaller places \nthe customers on local calls will dial only the numerical digits, the office \nname being employed for toll dialing purposes only.\n\nConsidering the thousands of central offices which now have numbers \nother than the 2-5 type and the fact that to change existing numbering \nsystems is a difficult and often costly procedure, it will be a number of \nyears before this ultimate objective is realized. As a practical measure,\n\ni , 902 \n= l \\ \naes | 204 L \n= ce lee en and \u201d \n~ BANGOR \nMONTREAL e \ne \n701 i\u201d \ne FARGO Za \nal win 603 PORTLAND \nJct. \nst, CLOUD i x \noA | goston \n} \u00a5 518 \nAPOLIS \n\u2014 : ee A \u2018 XZ sinacvst lye \nPIERRE 1 612 : \n\\ APPLETON soe ; Ae \ne \n\u2018 i ee PROVIDENCE \nj MANKATO / AVA Caan wR AVEN \nUc MADISON \\ an \n{ \u2018 * VAUKEE \\, yy ARK \nMIL f ; \n| ware U! O HIGHLAND @ KALAMAZOO, am ys NeW ORK \n\\ KFORD\u00ae \\ PARK rane ae \n\u2014 (ncoSS a6 \\aserrel & \n\\ as =e A\\e\\ @\\\u00b0 pre, 201 \n@ \\ ae) \u2018 ae \\ e \\ pvr? Ny \neS MOINES aoe \nWw. Pee Wied \nOTTUMWA\u00ae \\ ok ag \n\\ 217 CHAMPAIGN \n\u2018 . \n\\MOBERLY SPRINGFIELD \n\\ @ e \n\u2018 ee \nKANSAS CITY \\ s \n\\ SE LOUNS CENTRALIA \nercue -\u2014\u2014~ \\ \n~---.\\ NA 618 \nNY CAPE @ \n\\_GIRARDEAD\n\ndigit local dialing and for offices in the larger places served by certain \ntypes of dial equipment, as they are arranged today, it will be necessary \nto prefix the dialing of toll calls by a transfer or directing code to permit \nthe customer getting from the local office into the toll network.\n\nIndependent of the advantages of a universal 2-5 numbering plan \nfor nationwide operator and customer toll dialing, the Bell System has \nmade considerable progress in this direction over the past several years. \nNew York and Northern New Jersey adopted 2-5 numbering in 1930 \nin order to take advantage of the flexibility of office code assignments \nand the large code capacity which this type of local numbering provides. \nSince World War II many cities and their environs such as Chicago, \nBoston, Philadelphia, San Francisco, Oakland, Pittsburgh, Milwaukee, \nProvidence and a number of smaller cities have followed suit. Presently \nabout 12 million telephones are in areas which have 2\u20145 numbering \nexclusively in addition to perhaps two million telephones with 2-5 \nnumbers in mixed 2-4 and 2-5 areas. Another five million telephones \nare already planned for conversion to 2-5 numbers within the next \nseveral years.\n\nThe entire program will take many years to realize but it is one which \nmust be accomplished in order to achieve the best results in operator \ntoll dialing and make it possible for a customer at any telephone in the \nUnited States and Canada to reach a telephone anywhere in the two \ncountries by dialing without the assistance of an operator.\n\nswitching points is based on the Toll Crossbar System? now in service \nand has been constructed by the addition of the necessary CSP features \nto the basic structure of that system.\n\nThe necessity of providing this feature in CSP switching systems \narises from the nature of the numbering and switching plans. The num- \nbering plan* is constructed with the objective of using a minimum \nnumber of digits to give each telephone user in the country a distinctive \nnumber.\n\nNumbers delivered to the CSP equipment are in the form ABX- \nXXXX if the called place is in the same numbering area as the CSP. \nAB represents the first two letters of any office name and X represents \nany numeral. If the called place is in another numbering area this set of \ndigits will be preceded. by XOX or X1X. XOX or X1X is the area code, \nABX the local office code, and these are the digits used for routing \npurposes. Regardless of the number of switches required to complete the \ncall, these two sets of code digits are all that will be supplied. They are \nuniversal codes in that they identify specific destinations \u2014 any place\n\ndigits forward by MF and leaves the connection to accept another call. \nThis information is relayed from the decoder to the sender by way of \nthe marker. The work time of the decoder has been in the order of a half \nsecond.\n\nThe marker determines the identity of the frames on which the in- \ncoming and outgoing circuits are located, finds an idle path between the \ntwo circuits and sets up the connection. After checking the path through \nthe switches to be sure that there are no troubles it notifies the sender \nthat its task has been completed and then leaves the connection. Its \nwork time has also been in the order of a half second.\n\nIn the meantime other digits have been coming in to the sender but \nit does not wait for all of them to arrive before advancing the call. When \nthe marker selected the circuit to New York a signal was immediately \nsent forward to summon a sender in the New York switching system. \nThe process of attaching the sender in New York was carried on con- \ncurrently with the establishment of the connection through the switches \nin Atlanta.\n\nWhen the New York sender is attached a signal is sent to the Atlanta \nsender to advise it that pulsing may proceed. It immediately sends the \narea code 207 to New York by MF pulsing and follows it with the \nremaining digits of the called number, AC4\u20142345, as they are received \nfrom the operator, ending with a start pulse, and then leaves the connec- \ntion. All common control equipment in Atlanta is now free.\n\nIn New York, as soon as the Maine area code is received it is submitted \nto the decoder. Upon examination of the code the decoder finds that it is \ninsufficient for routing purposes. New York has a direct circuit group \nto Portland over which traffic to some offices in Maine is routed, but \nother offices are reached through Bangor by way of Boston. In order to \ndetermine which route to take the decoder must know what office is \ndesired. It, therefore, gives the sender a signal saying \u2018\u201c\u2018come again when \nyou have six digits\u2019? and leaves the connection. When the sixth digit \narrives the sender again calls for a decoder and gives it the complete \ncode 207-AC4.\n\nThe decoder again translates the area code, which now directs it to the \nforeign area translator which serves the Maine area, and submits the \ncomplete code to that translator. From the ensuing translation it learns \nthat the route.is by way of Boston and that all digits should be sent \nforward by MF. It then calls for a marker and releases the foreign area \ntranslator.\n\nboth codes are again translated since Boston also has a choice of routes \nto Maine, and the route to Bangor is selected. The translating equipment \nin Boston knows that Bangor is in the Maine area and that the area \ncode will, therefore, not be needed. However, since Bangor is a TO \nhaving no senders, the Boston sender must pulse forward all of the \ndigits needed to complete the call through switches in Bangor, Houlton \nand Monticello. It is assumed that Houlton is arranged to route the call \nto Monticello on receipt of the digits AC4. Numerical digits 2345 will \nroute the call through the Monticello switches to the called customer\u2019s \nline. These digits are all registered in the Boston sender but the digits \nrequired to switch the call through Bangor are not and must be supplied. \nAn arbitrary set of digits beginning with \u2018\u20181\u201d\u2019 can be used for this purpose \nsince no office code begins with \u201c1\u201d and there will, therefore, be no \nconflict.\n\nIn setting up this call all of the characteristic CSP features were em- \nployed, automatic alternate routing in Atlanta, six-digit translation in \nNew York and Boston, digit storing and variable spilling at all CSP\u2019s \nwith substitution of arbitrary digits for the area code at Boston.\n\n* All talking connections through the CSP system are made on a four- \nwire basis, that is, separate pairs of conductors are provided for trans- \nmission in the two directions. This is done in order to simplify the \nproblem of maintaining satisfactory balance so that the loss introduced \nby extra links in a connection can be held to a minimum value. The \nimportance of this feature is emphasized by the fact that the switching \nplan permits as many as eight intertoll trunks to be connected in tandem \nfor the completion of a call.\n\nThe advantages of four-wire switching were fully explained in the \npaper\u2019 on the toll crossbar system now in service.\n\nTwo separate groups of incoming senders are provided, one to receive \nDP and the other MF pulsing. Whether the system is installed in a \nstep-by-step or a panel-crossbar area both groups of senders will always \nbe needed. MF will be received from senders in other CSP\u2019s and from \nswitchboard positions. DP will be received from switchboard positions\n\npattern of the CSP network for the country as seen from the CSP \nconcerned and are interconnected in a pattern of progression correspond- \ning to the fixed order of alternate route selection. Group busy leads from \nthe toll line groups are connected to the contacts of the relays in such a \nmanner that if a group is busy the relay corresponding to the next choice \nroute in the chain will be operated. In this way the lowest choice route \nhaving an idle circuit will be speedily selected without testing individual \ntrunks of separate groups. The decoder learns from the translator which \nrelay in the array to operate first and the choice of the best route avail- \nable follows automatically. The principle will be readily understood by \nreference to the simplified sketch in Fig. 3. Contacts not shown on the \nrclays cause the translator to select the route corresponding to the last \nrelay operated in the chain.\n\nThe magnitude of the translating job for nationwide dialing led to the \ndecision to develop a new translator operating on a principle radically \ndifferent from that employed in other crossbar systems. In previous \nsystems translation is done by relays. The code digits \u2014 never more than \nthree \u2014 operate a group of relays which cause a single terminal corre- \nsponding to the code to be selected. A cross-connection is made between\n\nHIGH-USAGE ROUTES FINAL ROUTES \nOISTANT DISTANT DISTANT HOME ALL \nPRIMARY SECTIONAL REGIONAL NATIONAL REGIONAL TRUNKS \nOUTLETS CENTERS CENTERS CENTER CENTER BUSY\n\n1-INDICATION FROM TRANSLATION OF THE FIRST ALTERNATE ROUTE \n2-INDICATION THAT ALL TRUNKS IN THE GROUP ARE BUSY\n\nthe code point and a route relay associated with the trunk group to be \nselected. The route relay has a number of contacts:which are cross- \nconnected to supply the information required for proper routing of the \ncall. When changes in routing or equipment location of trunks within \nthe office are made it is necessary to change cross-connections.\n\nWith the nationwide dialing plan in operation routing changes or \nopening of new offices in one part of the country will necessitate trans- \nlator changes in many offices, some of them far removed from the scene \nof the event that forces them to be made. The changes in any one CSP \nwill, therefore, be frequent and to make them by running cross-connec- \ntions would be cumbersome and expensive. The new translator uses \npunched cards instead of relays, making it possible to effect changes by \nthe simple process of removing old cards and inserting new ones in a \nmachine. This can be done in a very short time and not only saves labor \nbut requires less out-of-service time for the equipment. Fig. 4 is a photo- \ngraph of the machine.\n\nA metal card about 5 by 102 inches is provided for each area code \nand also one for each office code that must be translated in a particu-\n\nTRANSLATOR BOX NUMBER \nLGe SAR ORE oA ODIkG. \"RIS Gein Ublicn. \nINO! HB BTO BT1 BUO BU! BU2 Bu4 BU7\n\nROUTING INSTRUCTIONS CONT & DIGIT CONTROL \nRIO RI1 RI2 RI4 RI7\u00b0 \u201cCOCO CcDdC1 CDC2 CDC4\n\n$ \ni SOURCE OF RECORD\u201d Sao \nOM C OT MT Tv CT TRI \u201cre TRI \u201cTre Q ! 2 6 7 8 9 10 uZ are mer <4 ! 16 7 \nTSE OF RECORD MARRERSTT OR COE : \ns? FIF MFT CEROSTIMSTIATRETST SOT __oGt] 9 1 2 93 4 1 8 9/0 1 29. ari 6 or) \nSOR, FA TENS SOR FR UNITS \n3.4 5 6 7 8 910 2.3.4 5 6 \u2018 5 0. ii - \nSoaucrten bustets See SST Ca \nWms oe see eile ee a he \nDECODER Pur COOL \nMiao 1 2 4 a7 90.) 2 4 07 coy 2 4 C? 00 1 2 4 9 \u00a30 1 2 4 FO) 2 477 \nCODE BARS \nAO 1? 4 a7 60 1 24 7 co 1 204 \u00a27 poy 2 4 om oy 2 4 ee OP a \nOCCODER InpuT\n\nS21 3p 6D 6DA vO _NVO NRO CKICFM PF _TSA TSB TSC SBD} LI U2 U3 1a \nDECODER ROUTE ADVANCE \nRA RAI RAZ RA GSO GS! _GS2 653. G54 GS5 GO _GI_G2_G3 GB RLS MB_RO_AOIT ROTC \nTar\n\nsie \nTRANSMITTEO-MMRKER TO SENDER \nR31 a _DC_MF_SxD LPD XOD.XSG DLC SXR20C___ODG 4DG 5DG_NSK SK3 SK6\n\n59 \nDECODER CAOSS Ix} GECODER Time-OUT MARKER TIME -OUT \nCAREC IK CRK ODT TRA CF TRE ORL ALT WT_FT) MO reol \"Tie Twi TH2 Tus 38s SOFFICE \nMaanER CROSSIN=) OATE \nx15 IPs 38 St MS OK OT JP WS OLS JS _SM_SMISMO TL RCK TKS TR TRL STRMRL TIF TOF\n\n2: ie G.  Khinhend A. J. Busch and F. F. Shipley, \u2018\u2018Crossbar Toll Switching \nSystem,\u201d A.J.H.H. Transactions, 68, June Section, pp. 302-309, 1944.\n\n3. C. A. Dahlbom, A. W. Horton, Jr. and D. L. Moody, \u201cApplication of Multi- \nfrequency Pulsing in Switching,\u201d AI.E.E. Transactions, 68, June Section, \npp. 505-510, June 1949.\n\nF. J. Scudder and J. N. Reynolds, \u2018\u2018Crossbar Dial Telephone Switching Sys- \ntem,\u201d A.I.E.E. Transactions, 58, May Section, pp. 179-192, 1939.\n\nN. A. Newell and A. Weaver, \u2018\u2018Single Frequency Signaling for Telephone \nTrunks,\u2019\u2019 Presented at Winter General Meeting of A.I.E.l., Jan. 31, 1951.\n\nF. A. Korn and J. G. Ferguson, \u2018\u2018The Number 5 Crossbar Dial Telephone \npaverne System,\u201d A.J.H.E. Tranasctions, 69, First Section, pp. 233-254,\n\n. J. Meszar, \u2018\u2018Fundamentals of the Automatic Telephone Message Accounting \nSystem, \u00bb Presented at the Winter General Meeting of A.I.E.E., Jan. 31, \n1951.\n\nA mathematical analysis is given of the low-loss, broad-band, laminated \ntransmission lines proposed by A. M. Clogston, including both idealized \nparallel-plane lines and coaxial cables. Part I deals with \u2018\u201cClogston 1\u201d \nlines, which have laminated conductors with a dielectric, chosen to provide \nthe proper phase velocity for waves on the line, filling the space between the \nconductors. Part II will treat lines having an arbitrary fraction of their \ntotal volume filled with laminations and the rest with dielectric, and will be \nconcerned in particular with \u201cClogston 2\u201d lines, in which the entire propaga- \ntion space is occupied by laminated material.\n\nThe electromagnetic problem is first formulated in general terms, and then \nspecialized to yield detailed results. The major theoretical questions treated \ninclude the determination of the propagation constants and the fields of the \nprincipal mode and the higher modes in laminated transmission lines, the \nchoice of optimum proportions for these lines, the calculation of the fre- \nquency dependence of attenuation due to the finite thickness of the laminae, \nthe increase in loss caused by improper phase velocity (dielectric mismatch) \nin Clogston 1 lines and by nonuniformity of the laminated material in \nClogston 2 lines, and the effects of dielectric and magnetic dissipation.\n\nTy, VODROMUCHION sin fire ot enact Gxwh dkeaad enced pam band chek daypebek 884 \nII. Wave Propagation Between Plane and Cylindrical Impedance Sheets. 887 \nIII. Surface Impedance of a Laminated Boundary..................... 896\n\nIV. Principal Mode in Clogston 1 Lines with Infinitesimally Thin Laminae 908 \nV. Effect of Finite Lamina Thickness. Frequency Dependence of Attenu-\n\nVI. Effect of Dielectric Mismatch. .......... 0.0... c cece teens 931 \nVII. Dielectric and Magnetic Losses in Clogston 1 Lines................ 940 \nAppendix I: Bessel\u2019 Function Expansions...................000005 944 \nWabble-of Symi bls 22.55: c sees oes asa yt de eeaneasie antenatal lathe yak at ae kedes 946\n\nA recent theoretical paper! by A. M. Clogston presents the very \ninteresting discovery that under certain conditions skin effect losses in \nthe conductors of a transmission line at elevated frequencies can be \nmuch reduced by laminating the conducting surfaces, parallel to the \ndirection of current flow, with alternate thin layers of conducting and \ninsulating material. The requirements are that the thickness of each \nconducting layer must be considerably smaller than the skin depth in \nthe conductor, and the phase velocity of waves on the transmission line \nmust be held very close to a certain critical value, which depends on the \nrelative thicknesses and the electrical properties of the conducting and \ninsulating layers: Under these conditions the \u2018\u2018effective skin depth\u201d of \nthe laminated surface is greatly increased; in other words, the eddy cur- \nrents induced by a high-frequency alternating field will penetrate much \nfarther into such a laminated structure than into a solid conductor, with \nconsequent marked reduction of ohmic losses in the metal. The metal \nlosses can also be made to vary much less with frequency, over a fixed \nband, than the ordinary skin effect losses, which are known to be very \nnearly proportional to the square root of frequency.\n\nClogston goes on to show that a laminated material composed of \nalternate thin conducting and insulating layers may itself be regarded \nas a transmission medium. Tor example, if the space in a coaxial cable \nwhich is ordinarily occupied by air or other dielectric be filled with a \nlarge number of coaxial cylindrical tubes which are alternately conduct- \ning and insulating, the cable will propagate various transmission modes, \nand under the proper circumstances some of these modes will exhibit \nlower attenuation constants than the transmission mode in a conven- \ntional coaxial cable of the same size at the same frequency.\n\nExperimental verification of Clogston\u2019s theory of laminated conductors \nhas been obtained? at the Bell Telephone Laboratories, and the trans- \nmission properties of a line filled with laminated material have also been \nmeasured at these Laboratories and found in reasonable agreement with \ntheory. However experiments with structures as complex as those pro- \nposed by Clogston are by no means simple, and the experimental work \non laminated conductors is still in an early, exploratory stage. Inasmuch \nas the experiments are necessarily time-consuming, it has been thought\n\n1A.M. Clogston, Proc. Inst. Radio Engrs., 39, 767 (1951), and Bell System Tech. \nJ., 30, 491 (1951). References will be to the Bell System Technical Journal article, \nalthough ocr for equation numbers the two papers are identical. \n2H ack, C. O. Mallinckrodt, and 8. P. Morgan, Jr., Proc. Inst. Radio \nEngrs., 40, p. 362 (1952).\n\nwe discuss losses due to imperfect dielectrics and lossy magnetic ma- \nterials.\n\nPart II will be largely devoted to transmission lines of the so-called \n\u201cClogston 2\u201d type, in which the entire propagation space is filled with \nthe laminated medium, though to a lesser extent we shall also consider \ntransmission lines having an arbitrary fraction of their total volume \nfilled with laminations and the rest with dielectric. We shall first derive \nexpressions for the propagation constant and the fields of the lowest \nClogston 2 mode assuming infinitesimally thin laminae, so that the \nattenuation constant is essentially independent of frequency, and then \ngo on to investigate the transition of the lowest Clogston 1 mode into \nthe lowest Clogston 2 mode as the space occupied by the main dielectric \nis gradually filled with laminations. We shall also discuss the higher \nmodes which can exist in Clogston 1 and Clogston 2 lines with infinitesi- \nmally thin laminae. Next the effect of finite lamina thickness on the \nvariation of attenuation with frequency in a Clogston 2 will be investi- \ngated, and then the important question of the influence of nonuni- \nformity of the laminated medium on the transmission properties of the \nline. We shall conclude with a short section on dielectric and magnetic \nlosses.\n\nInsofar as possible, plane and coaxial lines will be treated together \nthroughout the paper. Since however Bessel functions are not so easy \nto manipulate as hyperbolic functions, there will be a few cases where \nexplicit formulas are not yet available for the cylindrical geometry. In \nthese cases the formulas derived for the parallel-plane geometry usually \nprovide reasonably good approximations, or if greater accuracy is desired \nspecific examples may be worked out numerically from the fundamental \nequations in cylindrical coordinates.\n\nThe purpose of the present paper is to set up a general mathematical \nframework for the analysis of laminated transmission lines, and to treat \nthe major theoretical questions which arise in connection with these \nlines. In view of the length of the mathematical analysis, we have not \ndevoted much space to numerical examples, although a large number of \nspecific formulas are given which may be used to calculate the theoretical \nperformance of almost any Clogston-type line that happens to be of \ninterest. A considerable part of our work is directed toward evaluating \nthe effects of deviations from the ideal Clogston structure. Both theoreti- \ncal and experimental results suggest that the limitations on the ultimate \napplications of the Clogston cable are likely to be imposed by practical \nproblems of manufacture. These limitations, however, depend upon \nengineering questions which we shall not consider here.\n\nThe quantity o is called the mtrinsic propagation constant and 7 the \nintrinsic impedance of the medium.\n\nWe begin by considering structures bounded by infinite planes parallel \nto the x-z coordinate plane, and we confine our attention to transverse \nmagnetic waves propagating in the z-direction. We assume that the \nonly non-vanishing component of magnetic field is H; , and that all the \nfields are independent of x. Then the non-zero field components, written \nto indicate their dependence on the spatial coordinates, are H,(y, 2), \nE,(y, 2) and E,(y, z), the time dependence e\u201c\u2019\u2018 being understood through- \nout. The field components are shown in Fig. 1. .\n\nwhere g is the intrinsic propagation constant defined above. It is easy \nto see that (4) is satisfied by a wave function of exponential form, say\n\nponents corresponding to any particular H, are easily obtained from \nequations (2). i\n\nA concept important in what follows is that of wave impedances\u2019 at \na point. For a wave whose field components are H,, EF, , E,, the wave \nimpedances looking in the positive and negative y- and z-directions at a\n\nFor waves of the type that we consider, Z; and Z, are functions of \ny only, so that if two media having different electrical properties are \nseparated by the plane y = yo, the continuity of the tangential compo-\n\nis quite important. The sheets are located at y = +4), as shown in Fig. \n1, and the space between them is filled with a medium whose electrical \nconstants are \u20ac0 , Ho , Jo (Or oo , 0 , if we wish to use the derived constants). \nFrom the symmetry of the boundary conditions it is evident that for \nany particular mode H, must be either an even function or an odd func- \ntion of y about the plane y = 0. Taking the even case first, we have\n\nIf we replace go + iwes by oo/m and xo by (a9 \u2014 7)', the boundary con- \ndition at y=3b, namely\n\n2no \nThe transcendental equations (11) and (18) are satisfied by the propa- \ngation constants of the various even and odd modes; presumably each \nhas an infinite number of roots, which we could find, at least in principle, \nif we knew the explicit form of the function Z(y). We shall confine our- \nselves here to deriving an approximate expression for the propagation\n\nconstant of the principal mode (lowest even mode) when the walls are \nvery good conductors.\n\nIf the walls were perfectly conducting we should have Z(y) = 0, \nand the lowest root yo of (11) would be given by\n\nThe principal mode between perfectly conducting sheets is just an un- \ndisturbed slice of the plane TEM wave which could propagate in an - \nunbounded medium. If Z(yo) is not rigorously zero, but still so small that\n\nand if Z(y) does not vary rapidly with y in the neighborhood of yo, \nthen the lowest root of (11) is given approximately by\n\nwhere the second term is the first-order change in y due to the finite \nimpedance of the walls. If we formally set go = 0 (this does not actually \nrestrict us to perfect dielectrics since we could still assume eo or po to be \ncomplex), we have\n\nAs an example of the use of (20) and (21), suppose that the impedance \nsheets in Fig. 1 are electrically thick metal walls of permeability uw and \n(high) conductivity gi. Then to a very good approximation at all en- \ngineering frequencies and for all ordinary dielectrics between the walls, \nthe surface impedance is\n\nis the skin depth in the metal. We thus obtain from (20) and (21) the \nfamiliar formulas\n\nand \nOE ,/dp \u2014 dF ,/dz = iwpnH, , (28) \nfrom which we can eliminate Hp and EF, to obtain \nHd _ 10H, Hz , @Hs 2 \nEEA ATOR ak cel ee = 2 \nae ag ge ge eo\n\nIf we assume a wave traveling in the positive z-direction with propaga- \ntion constant y and write\n\nwhere \u00abx is given by (6) as before. But (31) is just the equation satisfied \nby modified Bessel functions of order one and argument xp, so\n\nwhere A and B are arbitrary constants. The other field components can \nbe obtained from H, using (27); the results are\n\nFor cylindrical fields of the type that we are considering, the wave- \nimpedances looking in the positive and negative p- and z-directions at a. \ntypical point are defined to be, respectively,\n\nWe shall now discuss the propagation of circular transverse magnetic \nwaves in a homogeneous region of space whose electrical constants are \n\u20ac) , Mo, Jo (OF oo, m0), and which is bounded by coaxial cylinders of radii \np1 and p2, where p2 > p1, aS Shown in Fig. 2. We suppose that the radial \nimpedances looking from the main dielectric into the inner and outer \ncylinders are, respectively,\n\nIf equations (36) are to be satisfied by values of A and B which are not \nboth zero, it is easily shown that a necessary and sufficient condition is\n\nnop o(Kop1) ae Z1(y) Kx (kop) a nopL& o(Kop2) = Za(y) Ki (Kops) (38) \nnoolo(Kopr) \u2014 Zi(y)L1(Kopi) nopLo(Kop2) + Ze(y)L1(Kop2)\n\nand (88) is a transcendental equation for the determination of the propa- \ngation constants of all the circular magnetic modes in the coaxial line.\n\nAs in the discussion of the parallel-plane line, we shall confine our \nattention to the principal mode and shall assume forthwith that the wall \nlosses are small.\u2019 Since for the principal mode we expect that y will be \nnearly equal to oo , we may write yo for oo and evaluate Z; and Z:2 at yo ; \nand we may replace the modified Bessel functions in (88) by their ap-\n\n5 J. A. Stratton, Electromagnetic Theory, McGraw-Hill, New York, 1941, pp. \n551-554, gives a similar treatment of the principal mode in an ordinary coaxial \ncable with solid metal walls.\n\nproximate values for small argument. From the series given in Dwight* \n813.1, 813.2, 815.1, and 815.2, we have\n\nfor |\u00ab| < 1, where log represents the natural logarithm. If we put \nthese approximations into (88) and if we suppose that the wal] im- \npedances are so small that\n\n2 2 2 \n= _ = : 41 \nieee cae no log (p2/p1) oe \nNow further assuming that \n1 | Zi(yo)/p1 + Ze(yo)/pe | \n= | \u2014\u2014 | <1 42 \n8 | ~ cum log (on/o1) 7 ae \nwe get by the binomial theorem \nZ \neee (yo) /pi + Zelyo)/pe (43)\n\nIf we formally set go = 0, we find that the attenuation and phase con- \nstants of the principal mode in a coaxial line with low-loss walls and no \ndissipation in the main dielectric are\n\nAs before, these approximations for a and 8 will ultimately break down \nas the frequency approaches zero, but they will certainly be valid over \nthe frequency range in which we are interested in the present paper.\n\ntion, Macmillan, New York, 1947. We shall refer to Dwight for a number of stand- \nard series expansions.\n\nThe magnetic field lines of the principal mode will of course be circles \nand the electric field will be largely radial, but with a small longitudinal \ncomponent unless the wall impedances are rigorously zero. The general \nexpressions (83) for the fields may be reduced to simple approximate \nformulas if we use the fact that xo is given by (41) and xp is small com- \npared to unity. The ratio A/B may be obtained from either of equations \n(36). Introducing the approximations (39) for the Bessel functions and \ncarrying out a little algebra, we get the following approximate expres- \nsions for the fields:\n\nwhere the amplitude factor J is equal to the total current flowing in the \ninner cylinder. Incidentally we note that the above results might have \nbeen derived from more elementary arguments if we had started with the \nfields in a coaxial line with perfectly conducting walls and treated the \neffect of finite wall impedance as a small perturbation.\n\nIf we consider an ordinary coaxial cable with solid metal walls at a \nfrequency high enough so that there is a well-developed skin effect on \nboth conductors, then to a good approximation\n\nIf necessary we may take account of dissipation in the main dielectric \nof either a plane or a coaxial transmission line by assigning complex \n7 \nvalues\u2019 to \u00ab and po, say\n\n7 See, for example, C. G. Montgomery, Principles of Microwave Circuits, M. I. T. \nRad. Lab. Series, 8, McGraw-Hill, New York, 1948, pp. 365-369 and 382-385.\n\ntenuation due to dielectric and magnetic losses, \nag = Reo = Re toW nei(1 \u2014 7 tan do) (1 \u2014 7 tan $0) \nLieya/ ule, (tan go + tan \u00a30), \nprovided that tan \u00a2 and tan {o are both smal] compared to unity, as they \nwill always be in practice. We shall neglect second-order effects and so\n\nregard the dielectric losses, the magnetic losses, and the wall losses as \nadditive.\n\nin a layer of homogeneous, isotropic material whose electrical constants \nare \u00a2, wu, g (or o, 7), and which is bounded by planes perpendicular to \nthe y-axis. Henceforth we shall always assume that the z-dependence of \nevery field component is given by the factor e \u201d*, where the complex \nquantity y, whose value may or may not be known a priori, is the propa- \ngation constant of the wave in the z-direction. Then the first of Maxwell\u2019s \nequations (2) yields\n\nEy = \u2014ly/Qg + twe)|Hz , (52) \nand on eliminating EH, from the other Maxwell equations, we get \n0H,/dy = \u2014(g + wwe)E.,\n\nNow if we formally identify H, with \u201ccurrent\u201d and FE, with \n\u201cvoltage\u201d, equations (53) are just the equations of a uniform one-dimen- \nsional transmission line extending in the y-direction, with series im- \npedance \u00ab\u2019/(g ++ iwe) per unit length and shunt admittance (g + iwe) \nper unit length; in other words a transmission line whose propagation \nconstant is x and whose characteristic impedance is 7, , where\n\nHence we can apply the whole theory of one-dimensional transmission \nlines with the assurance that in so doing we shall not violate the field \nequations. For example, if \u00a3(0), H(O) and E(d), H(t) represent the \ntangential field components E,, Hz at two planes separated by a dis-\n\ntance t, these fields are related by the general circuit parameter matrix \nof a uniform line, namely\n\nWe are now in a position to determine the surface impedance normal \nto a laminated plane structure composed of layers of which every other \none has thickness 4, and electrical constants o,, m, while the inter- \nvening layers each have thickness \u00a2, and electrical constants o2, m2. \nFig. 3 shows the cross section of such a stack in which the total number \nof double layers is n (2n single layers), while Fig. 4 represents the corre- \nsponding coaxial stack. Ultimately we shall assume the layers of thickness \ni: to be good conductors and those of thickness f to be good insulators, \nbut these assumptions need not be brought in immediately.\n\nIf the fields in the plane stack all vary with z according to e \u201d, then \nwhen we look in the direction of increasing y each double layer may be \nregarded as a four-terminal network formed by two sections of uniform \ntransmission line of lengths 4; and &, the propagation constants and \ncharacteristic impedances of the two sections being given respectively by\n\nThe matrix of the double layer is the product of the matrices of the two \nsingle layers in the proper order. Thus if the tangential field components \nare Ey , Hy at the lower surface of the first layer and EH, , H; at the upper \nsurface of the second layer, we have\n\nThe stack of double layers may be regarded as a chain of iterated four- \npoles; such chains have an extensive literature.\u2019 The relation between \nthe tangential fields Z, , H, at the upper surface of the nth double layer \nand Hy, Ho at the lower surface of the first double layer is\n\nwhere M is the @@CD-matrix appearing in equation (57). However there \nis a simple expression\u201d for the nth power of a square matrix of order \ntwo, namely\n\nI, (60) \nwhere I is the unit matrix of order two, I is the propagation constant \nper section of the chain of four-poles, defined by\n\nThe determinant of the matrix whose elements are given by (58) is unity, \nas may easily be verified; but this may not be the case for all the matrices \nwhich occur in our study of cylindrical structures. M will therefore be \ncarried explicitly in the following equations.\n\nKy, is the impedance seen when we look into a semi-infinite stack of \ndouble layers if the first layer is of type 1, while K\u00bb is the impedance seen \nif the first layer is of type 2. In calculations relating to Clogston 1 lines \u00a9 \nwith dissipative walls, the real parts of Ki and Ke will both be positive. \nBy a straightforward procedure we may express the matrix elements \n@, \u00ae, \u00a9, D in terms of Ki, Ke, T, and M, and then transform equation\n\n10 F. Abel\u00e9s, Comptes Rendus, 226, 1872 (1948). This result was called to the \nauthor\u2019s attention by Mr. J. G. Kreer.\n\nFinally we obtain from (59) and (64) an expression for the impedance \nZo looking into a plane stack of n double layers when the nth layer is \nbacked by a surface whose impedance is Z,, , namely \ngz, - Eo 17,(Kye\u2122 + Kee \u2122\") + KiKe sh nI 65) \n\u00b0 Ay Z, sh nt + 4(Kie * + Ke\")\n\nFor the cylindrical geometry, matters are a good deal more compli- \ncated. If we consider waves having field components H,, E,, Zz in a \nhomogeneous, isotropic shell bounded by coaxial cylindrical surfaces, \nand assume a propagation factor e \u201d, Maxwell\u2019s equations (27) and (28) \nmay be written .\n\nIf desired, we might identify /, with \u2018\u2018voltage\u201d and \u2014 pH, with \u2018\u201c\u2018current\u201d\u2019 \nand regard equations (67) as describing a nonuniform radial transmis- \nsion line, having series impedance \u00ab\u2019/(g + iwe)p per unit length and shunt \nadmittance (g + twe)p per unit length. Since, however, in equations (34) \nwe have already defined the radial wave impedance to be a field ratio \nwithout the extra factor of p, we shall carry out the analysis of the \npresent paper directly in terms of the field components H, and \u2014H,.\n\nFrom the general expressions (83) for the fields in cylindrical co- \nordinates, we can show that the matrix relation between the tangential \nfield components HL, , \u2014H, at two radii p: and pe is given by\n\nIt may be verified that the determinant M of the square matrix ap- \npearing in (68) is simply\n\nwhere terms of the order of \u00a2/p; represent the first-order curvature cor- \nrections. If we use the same value of p; , say p, for both parts of a double \nlayer, then up to first order the elements of the matrix of the double \nlayer become\n\nAs in the analogous equations (58) for a plane double layer, the sub- \nscripts 1 and 2 refer to the first and second layers respectively.\n\nIf we have a stack of double layers in which all the layers of the same \nkind have the same thickness and same electrical constants, then the \nonly term in (73) which varies from one double layer to the next is the \nmean radius p. Depending on the circumstances, we may wish to use a \nsingle value of p for the whole stack, or a few different values, or even, \nif high-speed computing machinery is available to carry out the matrix \nmultiplications, a different value of j for each double layer. The matrix \nof the whole stack then becomes a product of powers of as many different \nmatrices as we have chosen values of p. Obviously this method is better \nadapted to the numerical analysis of special cases than to the general \ntheoretical treatment of a stack whose ratio of outer radius to inner \n_ radius is unspecified.\n\nIn principle we are now able to compute the normal surface impedance \nof any laminated plane or coaxial stack at a given frequency provided \nthat we know the electrical constants and the thickness of each layer, \nthe number of layers, the propagation constant y in the z-direction, and \nthe normal impedance Z,, of the material behind the last layer. Since the \ngeneral formulas even for plane stacks are quite complicated, however, \nwe shall introduce at this point some very good approximations which \nwill be valid for all of the following work.\n\nHenceforth we shall take the layers of thickness \u00e9, to be such good \nconductors that the ratio we:/g: of displacement current to conduction \ncurrent is negligible in comparison with unity. For metals like copper \nthis is an excellent approximation at even \u2018the highest engineering fre- \nquencies. Then on introducing the characteristic skin thickness 6; , we \nhave for the conducting layers,\n\na= Viowg = (1+ 1)/4, i \nm= Vion/g = (1+ )/qd, \nwhere \nb= V2/omgn - (75) \nFor pure copper the permeability and conductivity are\n\nwhere fye is the frequency in Mc-sec\u201d. Referring to equations (56) \nand (69) and bearing in mind the above numerical values, we see that \nfor the conducting layers we have\n\nto a very good approximation, since in our applications the quantity y \nwill always be of the order of 27i/d, , where the vacuum wavelength \ndy is at least a few meters, while the skin thickness 6; will be at most a \nsmall fraction of a centimeter.\n\nFor the insulating layers of thickness f we shall set the conductivity \ngz equal to zero, so that\n\nWe denote the relative dielectric constant and permeability by e, and \nber respectively; dissipation in the insulating layers may be included\n\nif necessary by making e, and/or 2, complex. In MKS units we have \n\u20ac2 = \u20ac2r\u20acy M2 = Herby y (81) \nwhere the electrical constants of vacuum are\n\n= a (82) \nMy = 1.257 X 10 ~ henrys-meter ~. \nIt follows that \nQrt \u20acor 2WUf ye WV par \u20acoy us \n02 = CoV pore = a Sr ee Vee f Sone. \u201cr meters (93)\n\nwhere as usual the subscript v refers to vacuum. It is clear that unless \nwe deal with ferromagnetics, the quantities c2 and 7 will be of roughly \nthe same order of magnitude as o, and yn, . From (56) and (69) we have\n\nwhere since o2 and y are both of the same order of magnitude as 2z7/), , \nin general no further approximations can be made.\n\nIn all of what follows we shall assume that the thickness t of each \ninsulating layer is very small compared to the vacuum wavelength at \nthe highest operating frequency; in practice ft, will be at most a few mils \nand X, at least a few meters. Then the quantity | kete |, which is of the \norder of 27t2/d, , will be so small that to an excellent approximation we \nmay set sh kote = kolo and ch kof, = 1. Using this simplification, together \nwith the fact that m, <\u00ab m2, for all frequencies which may conceivably \nbe of interest, it is not difficult to show from (58) that the matrix ele- \nments of the plane double layer reduce to\n\nDOD = Maytals sh Kyly + ch Kil . \nMy \nThe determinant of the matrix is unity, and from (61) the propagation \nconstant per section is defined by\n\nKy = = pMaykale + -V (Snoykote)? + miynoykele coth \u00abti + iy \u00bb (87) \nKo = + omaykate + V (Sneykete)? + mameykete coth mtr + ni, -\n\nIf we make the same simplifications in the approximate expressions \n(73) for the matrix elements of a coaxial double layer, we obtain\n\nIn the preceding equations no restrictions have been laid on the \nthicknesses t; and \u00a2: except the trivial requirement that t, shall be small \ncompared to a wavelength. We shall now consider the limiting case in \nwhich both \u00a2; and \u00a2\u00e9, are infinitesimally small. When we make this last \nand most drastic approximation we do not expect that the idealized \nstructure thus obtained will show all of the features which are of interest \nin a physical transmission line with finite layers; but the results of the \nsimplified analysis will be useful in some cases nevertheless. It need \nscarcely be pointed out that we are dealing here only with a mathematical \nlimiting process, in which we assume that each layer, no matter how \nthin, always exhibits the same electrical properties as the bulk material. \nIf this assumption be regarded as unrealistic, it may be observed that \nthe quantity which we actually allow to tend to zero is the ratio of layer \nthickness to skin depth. The skin depth may be made as large as desired \nby lowering the frequency, so that the formulas which we derive by\n\nletting t, and t, approach zero at a finite frequency will also hold for finite \nthicknesses if the frequency is sufficiently low.\n\nWe shall let .@ denote the fraction of the stack which is occupied by \nconducting material, so that\n\nwhere at present t; and \u00a2; are both infinitesimal. Then the stack may be \nregarded as a homogeneous, anisotropic medium, characterized by an \naverage dielectric constant \u20ac perpendicular to the layers, an average \npermeability @ parallel to the layers, and an average conductivity @ \nparallel to the layers. Sakurai\u201d has treated such an artificial anisotropic \nmedium, and from his formulas we find that when the layers are al- \nternately conductors and insulators, the average electrical constants are, \nto a very good approximation,\n\nSakurai has also shown that the average values of the electrical con- \nstants may be used in Maxwell\u2019s equations for the average (macroscopic) \nfields, due regard being paid to the orientations of the field vectors with \nrespect to the laminae.\n\nwhere s is the thickness of the stack. The propagation constant TI, per \nunit distance normal to the stack and the characteristic impedance K \nof the stack are given by\n\nI, and K may also be derived from equations (86) and (87) by limiting \nprocesses; we have\n\nIt should perhaps be noted that terms of the order of we/g: and weo/gi \ncompared to unity were omitted in the expressions (90) for @ and g, \nand in the derivations of IT, and K. Since, however, under all practical \ncircumstances the omitted terms appear to be insignificant, we shall \nnot take space to write out the formally more complicated results which \nwould be obtained by keeping them.\u201d\n\nIn a cylindrical stack of infinitesimal layers, the average fields satisfy\n\n0H,/dz2 = \u2014iw\u00e9E, , \nO(pH5)/dp = GoL. , (97) \ndE./dp \u2014 dE, /dz = iwpH, . \nThe relation between the tangential field components L,, \u2014H, at two\n\n12 Tn Reference 1, equations (II-17) through (II-26) give examples of equations \nin which these small terms have been retained.\n\nIV. PRINCIPAL MODE IN CLOGSTON | LINES WITH INFINITESIMALLY THIN \nLAMINAE\n\nAn idealized parallel-plane Clogston 1 transmission line is shown \nschematically in Fig. 5. It consists of a slab of dielectric of thickness b, \nwith electrical constants yo , \u20ac , bounded above and below by laminated \nstacks each of thickness s. Outside each stack there may be an insulating \nor a conducting sheath, of which nothing more will be assumed at present \nthan that its normal surface impedance Z,(y) is known. The total dis- \ntance between the sheaths will be denoted by a, where a = b + 2s.\n\nThe corresponding Clogston 1 coaxial line is shown in Fig. 6. We de- \nnote the thickness of the inner and outer stacks by s; and s2 respectively, \nwhile a is the radius of the inner core (if any), and b is the inner radius \nof the sheath around the outer stack. The inner and outer radii of the \nmain dielectric are p; = a + s, and p. = b \u2014 82, respectively. In practice \nthe core may be a dielectric rod and the sheath may be a conducting \nshield, but in the present theoretical analysis we shall merely assume that \nthe radial impedances Z,(y) and Z,(y) looking into the core and the \nsheath are known.\n\nIn Part I of this paper we shall deal with \u201cextreme\u201d Clogston 1 lines, \nin which the space occupied by the stacks is small compared to the \nspace occupied by the main dielectric. We may then regard the laminated \nboundaries as impedance sheets guiding waves whose phase velocity is\n\ndetermined by the properties of the main dielectric, as discussed in Sec- \ntion II, and we may use the intrinsic propagation constant of the main \ndielectric in calculating the surface impedance of the boundaries. This \napproximation simplifies the analysis of Clogston 1 lines a great deal. \nWe shall treat the general case, in which an arbitrary fraction of the \ntotal space is filled with laminations, in Section IX of Part II, as a part \nof our study of Clogston 2 lines.\n\nIn this section we shall assume that the laminae are infinitesimally \nthin, so that the stacks may be completely characterized by their average \nproperties \u00e9, 7, and g. The case of finite laminae will be taken up in the \nnext section. We shall also assume throughout that dielectric and mag- \nnetic dissipation may be neglected except, as in Section VII, where the \ncontrary is explicitly stated.\n\nIn general the current density and the other field quantities in a plane \nstack of infinitesimally thin layers will be linear combinations of the \nfunctions sh T',y and ch Ty, where y is distance measured into the stack, \nand I, is the propagation constant per unit distance, as given by (98). \nThe qualitative behavior of the fields in a cylindrical stack will be similar. \nIn particular, if the stack is thick enough the current density and the \nfields will fall off as \u00a2 \", and we can define an \u201ceffective skin depth\u201d A by\n\nohmic losses in a stack carrying a fixed total current the current density \nshould be uniform across the stack, and that we can achieve uniform cur- \nrent density by adjusting the poe product of the main dielectric so as to \nmake I, equal to zero. If in equation (93) we set\n\nEquation (102) will be referred to henceforth as Clogston\u2019s condition\u2019 \nIf the permeabilities of the various materials are all equal, the condition \nreduces to\n\nWhen Clogston\u2019s condition is satisfied, Tz = 0 and the effective skin \ndepth of the stack is infinite; that is, the current density is uniform in \nany stack of finite total thickness. The quantities [, and K vanish \nsimultaneously, but the limiting value of their ratio is finite; and the \nmatrix of the plane stack, as given by (92), takes the form\n\nwhich is, as might have been expected, just the impedance between \nopposite edges of.a unit square of material of conductivity g and thick- \nness s through which the current density is uniform, in parallel with the \nsheath impedance Z,(yo). It follows from equations (20) and (21) of \nSection II that the attenuation and phase constants of the principal mode \nin a plane Clogston 1 line with infinitesimally thin laminae, Clogston\u2019s \ncondition being satisfied exactly, are\n\n3 This statement is certainly accurate enough for all practical purposes, al- \nthough an exact calculation which takes into account the small terms that were \nneglected in the approximate formula (93) for fT, shows that the effective skin \ndepth is \\o/276, where Xo is the length of a free wave in the main dielectric. The \nexact result is derived by Clogston in Reference 1, equation (II-26). In practice, \nfinite lamina thickness will restrict us to effective skin depths much smaller than \nthis theoretical limit.\n\nwhereas 6s will usually be several times the skin thickness 6; in the fre- \nquency range of interest. If the sheath is free space, its impedance is a \nfortiori much greater than 1/gs, since then it may be shown that\n\nwhere 7,\u00bb = 376.7 ohms is the intrinsic impedance of free space, and \nMor and \u20ac, are the relative permeability and relative dielectric constant \nof the main dielectric. Under most circumstances, therefore, we may \nneglect 1/Zn(yo) in comparison with gs, and obtain the very simple \nresults,\n\nTo this approximation the line exhibits neither amplitude nor phase \ndistortion.\n\nFor a coaxial stack of infinitesimally thin layers with Clogston\u2019s con- \ndition satisfied, the stack matrix given in (98) reduces to\n\nwhere po and p, denote the inner and outer radii of the stack. It follows \nfrom (112) that\n\nwhere Z;(yo) and Z2(yo) are the radial impedances looking into the stacks \nat pi and pe respectively, and Z,(yo) and Zs(yo) are the radial impedances \nlooking into the core and the outer sheath. From equations (44) and \n(45) of Section II, the attenuation and phase constants of a coaxial Clog- \nston 1 cable with infinitesimally thin layers, Clogston\u2019s condition being \nsatisfied exactly, are\n\nZ1(yo)/p1 + Zo(-yo)/p2 \nI 115 \npea tla teal ca \nwhere Z1(yo)/pi and Ze(yo)/p2 are given by (118). \nThe impedances Z,(yo) and Z,(yo) may be computed if we know ihe \nstructure of the core and the sheath. For a solid, homogeneous core and \na homogeneous sheath of effectively infinite thickness, we have\n\nbut of course the intrinsic propagation constant o and the intrinsic im- \npedance 7 need not be the same for the core and the sheath. If the sheath \nis of finite electrical thickness or has a laminated structure (alternate \nlayers of copper and iron, for example, to provide effective shielding), \nits surface impedance may be calculated by a straightforward but longer \nprocedure. We shall not go into this matter here, but shall merely observe \nthat in many cases of interest Za(yo) and Z(yo) are so large that we \nmay neglect the terms containing their reciprocals in (118). This means \nthat we neglect the total conduction and displacement currents flowing \nin the core and the sheath, compared to the conduction currents in the \nstacks. Then the expressions for the attenuation and phase constants \nbecome\n\nand again to this approximation there is neither amplitude nor phase \ndistortion. \nThe formulas which have just been derived on the assumption of\n\nfor \u20143b S y S 4b, where H is an arbitrary amplitude factor and \nZo(yo) is given by (105). In the stacks the fields are \nHe & Hill + gZo(vo)(3b F y)le\u2122,\n\nfor 36 S$ |y| S $a, where in cases of ambiguous sign the upper sign \nrefers to the upper stack (y > 0) and the lower sign to the lower stack \n(y <0). It should be noted that whereas the tangential field components | \nH, and E, are continuous through the stack, the normal field component \nE, is discontinuous at layer boundaries. From equation (52) we have, in \nthe conducting layers,\n\nTo our approximation, therefore, the only contributions to the average \nfield #, come from the insulating layers. \nThe average current density J, in either stack is uniform, being\n\nThe total current per unit width carried by the stack is just J.s, where \ns is the thickness of the stack; there will also be small currents in the \nsheaths unless we assume the sheath impedance to be infinite. The \npotential difference between any two points y: and 72 in the same trans- \nverse plane may easily be found from\n\nFor a Clogston 1 line of the proportions which we have been considering, \nthe potential difference across the stacks will be small compared to the \npotential difference across the main dielectric.\n\nIn a coaxial Clogston 1 with infinitesimally thin laminae, the fields in \nthe main dielectric are given to a good approximation by equations \n(46) of Section II, namely\n\nwhere J is an arbitrary amplitude factor and Z,(yo) and Zo(yo) are ex- \npressed by (113). In the inner stack we have\n\nThe average current density in either stack is uniform and is given by \nJ, = gE. , (129)\n\nthough in general the current density will not be the same in the two \nstacks because of the difference in cross-sectional areas. The potential \ndifference between the surface of the inner core and any other point in \nthe same transverse plane is\n\nIf the stacks are thin compared to the thickness of the main dielectric, \nas we are assuming throughout Part I, then the potential difference \nacross the stacks will be small compared to the potential difference \nacross the main dielectric, and the characteristic impedance Z; of the \nClogston 1 cable will be approximately the same as the characteristic \nimpedance of an ideal coaxial cable with perfect conductors of radii \npi and p2 and the same main dielectric, nome,\n\nZ, = 60 ~ log = = ohms. (131) \nOr \nWe shall defer making any field plots i Clogston-type transmission \nlines until Section IX of Part II, when we shall discuss the transition \nfrom Clogston 1 to Clogston 2 as the space originally occupied by the \nmain dielectric is gradually filled with laminations. Our present results \nwill then appear as the limiting case in which the thickness of the stacks \nis small compared to the thickness of the main dielectric. \nIn conclusion we shall mention briefly the question of how to dispose \na given amount of laminated material in a Clogston 1 coaxial cable so as \nto achieve the minimum attenuation constant. The whole problem of \noptimum proportions for Clogston cables is a complicated one of which \nan adequate treatment would require a separate paper in itself, with \nthe results depending to a great extent on engineering considerations \nwhich limit the ranges of the parameters that we can vary in any practical \ncase. Here we shall discuss only the following rather highly idealized \nproblem:\n\nGiven a coaxial Clogston 1 with infinitesimally thin laminae, having a \nhigh-impedance core and a high-impedance sheath of fixed radius b, \nand in which the total thickness s; + se of both stacks is a fixed constant \n2s. Assuming that 2s is small compared to b, what should be the radius \na of the core, and how should the total stack thickness be divided between \nthe outer and inner stacks so as to minimize the attenuation constant of \nthe line? Finally, what should be the fraction @ of conducting material \nin the stacks to minimize the attenuation constant, if the electrical \nconstants of the conducting and insulating layers are fixed, but the \nproperties of the main dielectric are at our disposal?\n\nand vary s; and s2 in accordance with this relation while holding a and b \nconstant, it is easy to show that the expression on the right side of \n(133) is a minimum when\n\nVa + Vo\u2019 Vat Vb \nThese equations tell us the most efficient way to divide the stacks in a \nClogston 1 when the radii of the core and the outer sheath are a and b re- \nspectively, still assuming of course that the thickness of each stack is \nsmall compared to its mean radius. \nIf we introduce the optimum values of s; and s2 into (183), we get\n\nIf b is fixed, the last expression is a minimum, considered as a function \nof a, when\n\nattenuation constant of a Clogston 1 cable with infinitesimally thin \nlaminae and high-impedance boundaries may be written in the form\n\nwhere the first factor depends on the electrical constants of the com- \nponents of the cable, while f(a, b, s: , s2) isa function only of the geometry. \nBy (110) the attenuation constant of a plane Clogston 1 has the same \nform, only with a different dependence on the geometrical factors. Now \nassume that the geometrical proportions of the line are fixed, and that \nthe electrical constants 41, 91, we, and e of the conducting and insula- \nting layers are given, but that the constants yo, \u00ab of the main di- \nelectric and the fraction of space @ occupied by conducting layers in \nthe stacks are at our disposal. The woe product of the main dielectric \nis to be codetermined with 6 so that Clogston\u2019s condition (102) is always \nsatisfied. Solving (102) for 6 gives\n\nHo\u20aco \u2014 [2\u20ac2 \n=. 142 \nHo\u00e9o +- (yu = M2) \u20ac2 \nHence the first factor in the expression (141) for a may be written\n\nIf we minimize the right side of (143) with respect to 9 , all other quanti- \nties being held constant, by equating to zero the derivative with respect\n\nG= (145) \nSui i (uj ae Surya)?\u2019 \nand the corresponding attenuation constant is proportional to \n(\u20aco/ U0)\" = (\u20aco/ uo)? Bur + (ui + Sprue)? (146)\n\nIt will be observed that so far we have determined only the optimum \nvalue of the product poe) , and so we are still free to alter the ratio of \nMo to \u20ac while holding the product of these two quantities constant. For \ngiven values of 4 and pe , we obtain the lowest attenuation constant by \nmaking \u00a2\u20ac as small as possible and yo as large as possible, subject of course \nto the practical restriction that \u00ab cannot be lower than the dielectric \nconstant of free space. However if we permit ye and po to be simul- \ntaneously increased, as by magnetic loading of both the insulating layers \nand the main dielectric, we find from (146) that on paper it is possible \nto decrease the attenuation constant without any definite limit. This \nobservation is in accord with the fact that the attenuation constant of \nan ordinary coaxial cable may be decreased indefinitely, with a corre- \nsponding decrease in the velocity of propagation along the cable, if we \nare willing to assume an unlimited amount of lossless magnetic loading.\n\nWhen po = pe, corresponding to no magnetic loading, we must take \n\u20ac\u00e9) = 3e,, and (148) reduces to\n\nmagnetic material in the stacks (u. = we = p,), and the optimum propor- \ntions given by (189) and (147), we have \n4.857\n\nimpedance of the main dielectric, which cannot be air in a Clogston cable.\n\nattenuation constant a, of an air-filled standard coaxial of the same size, \nmade of the same conducting material, is\n\nwhere por and \u00a2\u20ac; refer to the main dielectric of the Clogston cable. \nSince the attenuation constant of a standard coaxial cable is propor- \ntional to the square root of frequency in the range we are considering, \nwhile the attenuation constant of the ideal Clogston cable is independent \nof frequency in this range, there will be a crossover frequency above \nwhich the Clogston cable has a lower attenuation constant than a con- \nventional coaxial cable of the same size. If we are dealing with copper \nconductors and if frequencies are measured in Me-sec and linear \ndimensions in mils, then from equations (78) and (153) we find that the \ncrossover frequency is given approximately by |\n\nme, 49.50 (\u20ac0,/ Mor) \n\u00b0\u2122 (st + 82)inits \nFor example, let us take an ideal Clogston 1 cable of outer diameter\n\n0.375 inches, excluding the sheath, with no magnetic loading, and assume \nthe following values:\n\nThis cable has a lower attenuation constant than a standard air-filled \ncoaxial of the same size at frequencies above about 1 Me-sec \u2019, the ap- \nproximate formula (154) yielding 0.955 Me-sec\u2122 for the crossover \nfrequency and the exact equation (118), taken in conjunction with (151), \nyielding 1.251 Mc-sec\u2122.\n\nThe reader is cautioned that the comparison given by (153) between \nClogston and conventional cables is based upon certain highly idealized \nassumptions. In the first place we have neglected the finite thickness of \nthe laminae, which will in fact cause the attenuation constant of a \n_ physical Clogston cable to increase with increasing frequency, and \nultimately to cross over again and become higher than the attenuation \nconstant of a conventional air-filled coaxial. We have also neglected \ndielectric and magnetic losses, which are likely to be directly propor- \ntional to frequency and by no means negligible at the upper end of the\n\nfrequency band. In practice, too, the poco product of the main dielectric \nmust be held very close to the Clogston value or the benefit of the large \neffective skin depth is lost; and the stacks must be extremely uniform or \nagain the depth of penetration is greatly reduced. We shall take up all \nthese matters in later sections, and shall see that while the results just \ngiven represent ultimate limits of performance, the practical improve- \nments which can be achieved over conventional cables depend upon the \ndegree to which one can solve the manufacturing problems that tend to \nmake every actual Clogston cable differ more or less from the ideal struc- \nture considered above.\n\nV. EFFECT OF FINITE LAMINA THICKNESS. FREQUENCY DEPENDENCE OF \nATTENUATION IN CLOGSTON 1 LINES\n\nThe principal effect of finite lamina thickness in a Clogston cable is to \nintroduce a frequency dependence into the propagation constant, and \nin particular to cause the attenuation constant to increase, with increas- \ning frequency, above the value which we have found for infinitesimally \nthin laminae (or for finite laminae at low frequencies). The increased \nlosses are associated with the fact that the penetration depth in a lami- \nnated stack decreases with increasing frequency, even when Clogston\u2019s \ncondition is exactly satisfied, if the laminae are of finite thickness. We \nshall now obtain expressions for the surface impedance of a plane lami- \nnated stack of n double layers, such as is shown in Fig. 3, when Clogston\u2019s \ncondition is satisfied but the individual layers are of finite thickness.\n\nwhere in the last step we have used the fact that in the conducting \nlayers m1y is equal to 7 and x, is equal to o; to a very good approximation. \nWe now introduce the dimensionless parameter\n\nwhich may be regarded as a measure of the electrical thickness of the \nindividual conducting layers. From (86) and (156) we have, for the prop- \nagation constant per double layer,\n\nKy = a [\u2014 20 + 6? \u2014 \u00a9 coth \u00a9 + 1)4, \nsince My = Ki/Ga = O/giti \nIf the thickness \u00a2, of each conducting layer is moderately small com- \npared to the skin depth 6, at the highest frequency of interest, the quanti- \nties T, Ky, and Ky may conveniently be expanded in powers of \u00a9. The \nidentity\n\nafter we expand sh \u00a9 and ch 9 by Dwight 657.1 and 657.2 and collect \nterms. Taking the square root by the binomial theorem gives\n\nthe negative sign being introduced because from (157) \u00a9\u201d is a positive \nimaginary number and we want Re T > 0. Then\n\ner ae 170! > \nPaes | - zal?\u2019 +4 c+: \ni fe. & Q\u00b0 | \n--3.|5+% + met aa \nprovided that we expand the sh\u2019 function by Dwight 706. From (159) \nwe get \na. ee. @? + iVv3 @! iv\u20193 ef + |,\n\nwhere we have expanded coth \u00a9 by Dwight 657.5 and chosen the sign \nof the square root to make Re K, and Re K; both positive.\n\nOur first observation is that when the lamina thickness is finite the \neffective skin depth of the stack is also finite. We have, from (157) and\n\n(163), \n1 E it} Sti | \n=\u2014~|/54+\u20144-\u201457-\"'' (165) \n| V3 Le 158; 52561 : \nand the average propagation constant per unit distance into the stack is\n\nRe I, ti mug: fte tt \na result also given by Clogston.\u201d The number N of double layers in one \neffective skin depth is\n\nT, is essentially the thickness of conducting material in each stack which \nis effectively carrying current; it is evident that for small values of 4/6 \nthis effective thickness is inversely proportional to the frequency f and \nto the thickness t; of the individual conducting layers, but independent \nof the thekness 2 of the insulating layers, provided that tf is very small \ncompared to the length of a free wave in the insulating material.\n\nIn the general case, still assuming of course that Clogston\u2019s condition \nis satisfied, the surface impedance Zo(yo) of a plane Clogston stack is \ngiven by equation (65) of Section III, namely\n\n924 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1952 \nwhere Z,,(yo) is the impedance of the surface behind the stack. If \u00a9 = 0, \n(170) reduces to (105) of Section IV, that is,\n\n1 1 \nGs + 1/Zn(yo.) = Ti + 1/Zn(y0)\u2019 \nwhere 7; is the total thickness of conducting material in the stack. If \nZn(Yo) is infinite, then for all values of \u00a9 and n we have\n\nand if Re nT is large, corresponding to a stack many effective skin depths \nthick, then for any Zn(yo) we have\n\nOnce Zo(yo) has been computed for a particular frequency, the at- \ntenuation and phase constants of the plane Clogston 1 line at that fre- \nquency are given, as in Section II, by\n\nExplicit expressions for the surface impedance of a coaxial stack of finite \nlayers have not been derived. However, if in a coaxial Clogston 1 the \nthickness of each stack is small compared to its mean radius, or if the \ndepth of penetration given by (167) is small compared to the radius of \nthe surface near which the currents flow, then the parallel-plane formula \n(170) may be used for the stack impedances Zi(yo) and Ze(yo) which are \nto be substituted into the equations of Section II for the attenuation \nand phase constants, namely\n\nWV Moe + Im Ona log palo) . (177) \nIf the plane approximations are regarded as insufficiently accurate, one \ncan compute the surface impedance of a cylindrical stack by repeated \nmultiplication of matrices similar to the one given by equations (88) of \nSection III. This procedure would obviously involve considerable numeri- \ncal computation, but we can hardly expect that it would reveal anything \nqualitatively new for Clogston cables of the proportions considered in \nPart I.\n\nIt will be instructive to compare the impedance of a laminated plane \nstack with the impedance of a solid metal plate over the full frequency \nrange from zero to very high frequencies.\u201d If the stack contains n con- \nducting layers, each of thickness \u00a2, , and the metal plate is of thickness \nT, = nt, , the impedances of the plate and of the stack will be equal at \nzero frequency, and also at very high frequencies where the first layer \nof the stack is already many skin depths thick. For simplicity we assume \nthat both the plate and the stack are backed by infinite-impedance \nsurfaces at all frequencies.\n\nTo orient ourselves we shall define three critical frequencies, for which \nrespectively the thickness of the solid plate is equal to one skin depth in \nthe metal, the thickness of the stack is equal to one \u201ceffective skin depth\u201d\u2019, \nand the thickness of a single conducting layer is equal to ~/3 skin depths \nin the metal. These frequencies are\n\nThe approximate forms of the surface impedance functions of the plate \nand the stack in the various frequency ranges are then quite simple.\n\nIn the range 0 S f S fi, the surface impedance of the solid plate is \napproximately constant and given by\n\nwhile in the range f = f; we see approximately the surface impedance of \nan infinite plate,\n\nwhich is proportional to +/f. The surface impedance of the stack is \napproximately constant in the range 0 S f S fe, where\n\nwhile in the range fo S f S f; it is approximately equal to the impedance \nK, of an infinitely deep stack of moderately thin layers as given by the \nfirst of equations (164), namely\n\n18 In this connection see also Reference 1, Fig. 2, p. 494. Clogston compares \na laminated stack with a solid plate of the same total thickness as the stack, hence \na plate which contains more conducting material than the stack.\n\nwhich is directly proportional to frequency (and independent of con- \nductivity). For f 2 f3 the stack acts much like an infinitely thick solid \nplate, for which\n\nBy substituting the above series into equation (170), we can obtain \nthe variation of the stack impedance with frequency so long as ;/6; is \nsufficiently small. Although in principle there would be no difficulty in \ntaking into account an arbitrary sheath impedance Z,(yo), for brevity \nwe shall restrict ourselves here to the case in which the sheath impedance \nis so high that at all frequencies of interest the current in the sheath may \nbe neglected. Then we have equation (191) (see next page).\n\nFig. 7\u2014Surface resistance R of solid plates and laminated stacks versus \nfeoqueney f on log-log scale.\n\n= V3/(amgitiTs), (196) \nat which frequency the approximate formulas yield \nAR/Ro = 1/3, AX/Ry = V3. (197)\n\nWe may now answer the question: What must be the thickness \u00a2, of \nthe individual conducting layers in a plane stack which contains a given \ntotal thickness 7, of conducting material, if at a specified top frequency \nfm the resistance of the stack is not to have increased by more than a \nspecified small fraction of its de value? We find that the permissible\n\nand we note that this value of \u00a2, is inversely proportional both to fm and \nto 7, . If we measure \u00a2; and 7, in mils and f,, in Me-sec , then on putting\n\nFor a plane Clogston 1 with stacks of equal thickness, the attenuation \nconstant is given by (174), and the fractional change in attenuation \nwith frequency is equal to the fractional change in resistance of either \nstack, as calculated from (193). For a coaxial Clogston 1 with stacks \nthin enough so that the plane approximation is valid we may also use \n(193), but the fractional changes in resistance will be different for the \ntwo stacks if these are of different thicknesses, and the fractional change \nin the attenuation constant must be calculated from equation (176). \nTf Rio and Rs are the de resistances \u201cper square\u201d of the two stacks, and \nAR, and AR; their increments as obtained from (193), then the fractional \nincrease in attenuation is given approximately by\n\nFor either plane or cylindrical geometry we find that if we scale up a \nparticular Clogston line by multiplying the thicknesses of the stacks and \nthe main dielectric by the same factor, then the low-frequency at- \ntenuation constant will be divided by the square of the scale factor. \nHowever, the permissible thickness of the individual conducting layers, \nif we are to have the attenuation flat to a specified degree up to a fixed \nfrequency, is inversely proportional to the scale factor. Thus if we double \nthe overall dimensions of the line and double the amount of conducting \nmaterial in the stacks, we shall divide the low-frequency attenuation \nconstant by four, but we shall have to make the individual layers half as \nthick in order to maintain the same relative increase in attenuation \nconstant at the same top frequency f,, . In addition it is clear that if we \ndouble the top frequency while maintaining the same requirement on \nAa/a for a line of given dimensions, we shall also have to cut the thick- \nness of the individual layers in half.\n\nAs a numerical example, let us return to the cable whose specifications \nwere given by (155) at the end of Section IV. For this cable we have:\n\nIf the conducting layers are copper, we find that equation (200) for the \nfractional increase in attenuation becomes, numerically,\n\nIf for example the copper layers are 0.1 mil thick and the polyethylene \nlayers 0.05 mil thick, since we are assuming 6 = 2/3, then the attenuation \nconstant has increased by 10 per cent of its \u201cflat\u201d\u2019 value at a frequency of \nabout 9.1 Mc-sec\u2122.\n\nWe may also ask for the upper crossover frequency, above which the \nClogston cable will have a higher attenuation constant than a standard \nair-filled coaxial of the same size. Such a crossover frequency must exist \nbecause the dielectric loading of the Clogston cable (in our case @, = \n6.78) introduces a factor ~/e, into the asymptotic expression for the \nattenuation constant at extremely high frequencies when the stacks look \nlike solid metal walls; in addition there will be slight differences due to \nthe fact that the geometric proportions of the conventional and Clogston \ncables are not exactly the same.\n\nWe assume, subject to a posteriori verification, that the upper cross- \nover frequency lies between the critical frequencies f2 and f; , defined by \n(178), for each stack. Then we have in effect infinitely deep stacks of \nmoderately thin laminae, whose surface resistances are equal and are \ngiven by (182) to be\n\nThe attenuation constants of the conventional and Clogston cables are \nobtained from (151) and (176) respectively, where for the conventional \ncoaxial we set nm = 7,. After a little arithmetic we find for the upper \ncrossover frequency in this particular case,\n\nThus if the copper layers are 0.1 mil thick, the upper crossover frequency \nis about 280 Mc-sec\u2122\u2019, which turns out to lie well inside the interval \nbetween the critical frequencies fz and f; for both stacks.\n\nComparing this result with the result at the end of Section IV, we see \nthat a 0.375-inch Clogston 1 cable with 0.1-mil copper conductors and\u2019 \nthe other specifications given by (155) is nominally better than a con- \nventional air-filled coaxial cable of the same size in the frequency range \nfrom about 1 Mc-sec\u2122 to 280 Me-sec\u2019. We are still neglecting the effect \nof failure to satisfy Clogston\u2019s condition exactly, the effect of stack non- \nuniformity, and dielectric losses. All of these factors will be present to a \ngreater or less degree in any physical embodiment of a Clogston cable,\n\nIn terms of k, the general expressions for T, K; , and Ke in a plane stack \nof finite layers take a relatively simple form. We have\n\nafter a little rearrangement, where the only approximation that has \nbeen made so far is to set my % m and m  o . Substituting (207) into \n(86) and (87) gives\n\nIf k = 0, equations (208) and (209) evidently reduce to (158) and (159) \nof the preceding section. For a stack of infinitesimally thin layers, the \nconstants IT, and K are given by equations (93) and (94) of Section \nIIT, namely\n\nUp to this point we have set no restrictions on the magnitude of k, \nand we have not even assumed that k is necessarily real. Throughout \nthe rest of this section, however, we shall assume that k is a positive or \nnegative real number, as it must be if there is no dielectric or magnetic \ndissipation.\n\nIn practice both the lamina thickness and the amount of dielectric \nmismatch will be as small as it is feasible to make them. It will be useful, \ntherefore, to obtain approximate expressions for I, K,, and Ke under \nthe assumptions\n\nIf |k| \u00ab 1 we can neglect 2k compared to unity in the coefficient of \n\u00a9*, but since we have made no assumptions as to the relative magni- \ntudes of |@| and |k|, we cannot drop either the term in kO* or the \nterm in \u00a9*. If we replace sh I by 3 in (214), we get\n\nwhere we have taken the square root of the complex quantity by Dwight \n58.2, and\n\nThe effective skin depth of a stack of moderately thin layers in the \npresence of slight dielectric mismatch is, from (215),\n\nAn equation essentially equivalent to this was given by Clogston, in \nsomewhat different notation.\u201d It is clear from (211) or (218) that if the \nlayers are infinitesimally thin, we have \nA = 6/0|k i, (219) \nand the effective skin depth in the stack is proportional to the skin \n17 Reference 1, equation (III-42).\n\ndepth 6, in the conducting material at the operating frequency, although \nif the mismatch parameter k is small, the proportionality constant mul- \ntiplying 6; will be large. In the general case, the number of double layers \nin one effective skin depth is\n\nIt is instructive to plot the effective skin depth of a given stack at a \nfixed frequency as a function of dielectric mismatch. If\n\nAo = V(t + &)51/ti (222) \ndenotes the effective skin thickness when there is no mismatch, then the \nrelative skin thickness when the mismatch parameter is k is just\n\nThis ratio is plotted against k in Fig. 8, a universal curve being obtained \nby measuring k in units of (4/6:)\u201d. It is worth noting that when k = \n(:/6:), the effective skin thickness is only 53 per cent of the skin \nthickness with perfect dielectric match.\n\nThe surface impedance Zo(yo) of a laminated plane stack at any fre- \nquency and with any amount of dielectric mismatch is given by equa- \ntion (65),\n\nZa(yo) sh mE + 3(Kie\"\"? + Kee\"?) \u00a9 \nFor a stack with infinitesimally thin layers and total thickness s, the \nequation becomes\n\nSaye) a Z(yo) sh Tes + K ch Tus\u2019 (225) \nwhere I, and K are given by (211) and (212). At zero frequency, \n| 1 \nDi ae (226)\n\nZo(yo) = K coth Tys. (228) \nIf the stack is many effective skin depths thick, we have \nZ(yo) = Ki, (229) \nwhile if the individual layers are infinitesimally thin, | \nZ(vo) = K, (230)\n\nwhere K, and K are given by (209) and (211), respectively. \nWhen Z(yo) is known, the attenuation and phase constants of the \nparallel-plane Clogston 1 are given as usual by\n\n' but the impedances of the cylindrical stacks are easy to compute only if \nwe can employ the parallel-plane approximation for each stack. To take\n\nFig. 8\u2014Relative skin depth A/A\u00bb in a stack of finite layers versus dielectric \nmismatch parameter k, measured in units of (t1/61)?.\n\ncurvature effects into account would require a considerable amount of \nnumerical calculation. Equation (98) of Section III provides an explicit \nexpression for the surface impedance of a cylindrical stack of infini- \ntesimally thin layers in the presence of dielectric mismatch, in terms of \nBessel functions of complex argument; but if the layers are of finite \nthickness we can at present do nothing better than multiply out the \nmatrices of the individual layers step by step.\n\nThe variation of the surface impedance of a laminated stack with \nfrequency over the full frequency range is not quite so simple in the \npresence of dielectric mismatch as when Clogston\u2019s condition is exactly \nsatisfied, but a somewhat analogous discussion may be given. As in the \npreceding section, we consider a plane stack of n conducting layers \neach of thickness \u00a2; , where nt; = 7',, and backed by an infinite-imped- \nance surface. When the mismatch parameter is k, the three critical \nfrequencies are:\n\nIn the range 0 S f S fe, the surface impedance of the stack is ap- \nproximately constant, being given by\n\nwhere Ky is given by (217) provided that k is small compared to unity. \nTor infinitesimally thin layers the upper critical frequency f; is infinite, \nand we have for f = fo,\n\nZolyo) & | k- \u2014 isgn k)/gds \n= (1 \u2014tsenk)Vam|k| f/m, | \nwhich is proportional to ~/f. If the layers are of finite thickness but \nk = 0, we have the result obtained in the preceding section,\n\nwhich is proportional to f up to the critical frequency fo . If neither the \nmismatch parameter k nor the layer thickness tf; is zero, then the surface\n\nimpedance Zo(yo) cannot be represented by a simple power of f in the \nrange fe < f S f;. At frequencies above f;, if the layer thickness is \nfinite, the impedance is approximately that of a solid conductor, namely\n\nSince in general the surface resistance depends upon the two param- \neters 4/6; and k, it is not possible to plot a single curve which shows the \nvariation of resistance with frequency under all possible conditions of \ndielectric mismatch. However if we compare a matched stack of finite \nlayers with a similar mismatched stack, we see that the asymptotic \nbehavior of Zo(yo) is the same for both stacks at very low and very \nhigh frequencies. A numerical study of the exact equation for Zo(yo) \nshows that in the neighborhood of the critical frequency fe, the resist- \nance of the mismatched stack is higher than the resistance of the matched \nstack. (The critical frequency fe as defined in (235) is a function of the \nmismatch parameter k, but will be of the same order of magnitude for \na slightly mismatched stack as for a perfectly matched stack.) The \nresistance of the mismatched stack exhibits relatively small fluctuations \nabove and below the resistance of the matched stack in the neighborhood \nof the upper critical frequency fs , but this region is not of as much prac- \ntical interest as the region near fo, where the stack resistance is defi- \nnitely increased by the effect of dielectric mismatch.\n\nAn explicit expression for the rate at which the surface impedance of \na mismatched stack begins to depart from its de value as the frequency \nis increased has been worked out only for the ideal case of infinitesimally \nthin layers. For a plane stack of infinitesimal layers backed by an in- \nfinite-impedance surface, equation (228) gives, at moderately low fre- \nquencies,\n\n2 Bie cc \ngly 36? 456+ : \nfrom which the fractional changes in resistance and reactance are\n\nThe admissible value of | & |, if the fractional change in resistance is not \nto exceed a specified value AR/Rp at a given top frequency fn, is\n\nwhich is inversely proportional both to fm and to the square of the total \nthickness of conducting material in the stack. If we express 7\u2019; in mils, \nfm in Mc-sec\u2122\u2019, and assume the conducting layers to be copper, we get\n\nfor the three values 4/6: = 0, \u00e9/6; = 0.1, and 4/6 = 0.2. For an elec- \ntrically thick solid conductor we have simply \nRe gio, = Ts (247)\n\nhence to get any benefit from the laminated stack we must have \nRe g:6:K1 smaller than unity. Actually, if we meet Clogston\u2019s condition by\n\nFig. 9\u2014Normalized stack resistance Re gi\u00e9:kK, versus dielectric mismatch \nparameter k, for different values of t1/61.\n\nraising the dielectric constant and thus lowering the impedance of the \nmain dielectric, then since the attenuation constant of the line is pro- \nportional to the ratio of stack resistance to dielectric impedance, we \nmust have Re gi6:4, considerably smaller than unity to obtain a lower \nattenuation with the Clogston line than with an ordinary air-filled line \nhaving solid metal walls.\n\nFor a plane Clogston 1 line with stacks of equal thickness, the frac- \ntional change in the attenuation constant with frequency is equal to \u2014 \nthe fractional change in the resistance of either stack, whether this \nchange arises from the effects of finite lamina thickness or from di- \nelectric mismatch or both. The fractional change in the attenuation con- \nstant of a coaxial Clogston 1 depends not only on the change in resist- \nance of each stack, but also on the geometric proportions of the cable, \nin the manner expressed by equation (200) of Section V.\n\nThe effect of dielectric mismatch on the overall attenuation versus \nfrequency characteristic of a Clogston cable is in general to reduce the \ntotal frequency range (in Mc-sec\u2122) over which the Clogston cable has \na smaller attenuation constant than a conventional air-filled coaxial \ncable of the same size. To calculate the lower crossover frequency we \nmay ordinarily neglect finite lamina thickness effects and use equation \n(241) for the stack impedances, while at the upper crossover frequency \nthe stack impedances are very nearly equal to K1, as given by (217).\n\nIt should be remembered that mismatch of the ye product of the main \ndielectric will usually be accompanied by a change in the dielectric \nimpedance ~/ po /e\u00e9 . Thus under certain conditions the lower crossover \nfrequency may even be reduced by choosing 6 slightly below the Clog- \nston value, inasmuch as the increase in dielectric impedance may more \nthan compensate for the increase in stack resistance at low frequencies; \nbut it appears that this will be paid for in a steeper slope of the attenu- \nation versus frequency curve and a consequent greater reduction of the \n- upper crossover frequency.\n\nIt would be very useful to make a numerical study of the effects of \ndielectric mismatch in Clogston cables having a variety of different pro- \nportions; but in the present paper space limitations restrict us to a few \nobservations concerning orders of magnitude. For the cable which we \nconsidered at the end of the preceding section, it turns out than an \nincrease or decrease of 1 per cent in the value of \u00a2) makes a change of \nat most a very few per cent in either crossover frequency; with a matched \ndielectric, we recall, these crossover frequencies were about 1 Mc-sec\u2122 \nand about 280 Me-sec\u2122 respectively. However if we had designed a \nlaminated cable with thicker stacks or thinner laminae or both, so as to \nincrease the theoretical factor of improvement over a conventional cable \nin the working frequency range, we should have found that the tolerable \ndeviation of \u00a2 from Clogston\u2019s value, instead of being of the order of \n1 per cent, was more nearly of the order of 0.1 per cent or even smaller; \nand the greater the improvement striven for, the more stringent the re- \nquirement of accurate dielectric match.\n\nDielectric and magnetic dissipation in the main dielectric and in the \nstacks can be taken into account by introducing complex dielectric con- \nstants and permeabilities for the lossy materials. Thus we may write\n\n2 = Wy \u2014 tue = we (1 \u2014 7 tan f), \nwhere in the most general case the loss tangents may all be different, \nthough it will be assumed that they are all small compared to unity, so \nthat the problem may be treated by first-order perturbation methods.\n\nThe average rate of energy dissipation per unit volume in a lossy di- \nelectric by a harmonically varying electric field of maximum amplitude \nE is just 4we\u2019 i\u201d, since the imaginary part e\u201d of the complex dielectric \n_ constant corresponds to a conductivity g = we\u201d. Similarly the average \nrate of energy dissipation per unit volume in a lossy magnetic material \nby a harmonically varying magnetic field of maximum amplitude H is \n1wu\u201dH\u2019. The part of the attenuation constant which arises from di- \nelectric and magnetic dissipation is one-half the ratio of power dissipated \nper unit length of line to total transmitted power, provided of course \nthat the attenuation per wavelength is small. Since the loss tangents \nof the various materials are assumed small, we can use the fields found \nfor the lossless case to calculate the transmitted and dissipated power.\n\nIf the volume occupied by currents in the stacks is small compared - \nto the volume of the main dielectric, so that we can neglect the power \nflow in the stacks in the direction of wave propagation compared to the \npower flow in the main dielectric, then the part of the attenuation \nconstant which is due to dielectric and magnetic dissipation is given by \nequation (51) of Section II, namely\n\nwhere X, is the vacuum wavelength and lor , \u20acor are the real parts of the \nrelative permeability and relative dielectric constant of the main di- \nelectric. This equation will be derived from energy considerations pres- \nently. It should be noted that the part of the attenuation constant \ngiven by (249) is directly proportional to frequency, provided that the \nloss tangents are independent of frequency; but it is the same for both \nplane and coaxial geometry and is independent of all the geometrical \nfactors which describe the size and the relative proportions of the line. \nEquation (249) will probably be sufficiently accurate for all Clogston \n1 lines having the proportions (stacks thin compared to main dielectric) \nwhich we have considered in Part I. As an example wherein we also take \ninto account the power flow in the stacks, however, we shall treat a \nparallel-plane line with infinitesimally thin laminae backed by high- \nimpedance walls. Then, according to equations (120) and (121) of \nSection IV, the principal field components in the main dielectric are\n\nthe propagation factor \u00a2 \u201d*\"*' being understood throughout. To take . \naccount of dielectric and magnetic dissipation in the stacks, we write\n\nThe average power Pp transmitted through the main dielectric is \nobtained by integrating the real part of the z-component of the com- \nplex Poynting vector 3E X H* over unit width of the line; thus\n\nSimilarly, the average power P; transmitted per unit width of either \nstack is\n\nThe average power AP, dissipated in the main dielectric per unit length \nand width of the line is\n\nwhile the average power AP, dissipated per unit length and width of \neither stack is\n\nwhich reduces to (249) if we neglect the terms in s/b. The total attenua- \ntion is the sum of the metal losses, given by equation (110), and the \ndielectric and magnetic losses.\n\nFor a coaxial Clogston 1 cable with infinitesimally thin laminae and \nhigh-impedance boundaries, the principal field components are given by \nequations (126)\u2014-(128) of Section IV. In the main dielectric we have\n\nI(b? \u2014 p\u2019) \n2mp(b\u201d \u2014 ps3)\u2019 \nfa ft JOP) \nme @\u2019 Qap(b\u201d \u2014 p2)* \nThe average power transmitted through the main dielectric is\n\nwhile for the average power transmitted through the inner and outer \nstacks it will be sufficient to replace the exact expressions by the follow- \ning simple approximations,\n\nFor the average power dissipated per unit length of line in the main \ndielectric and the inner and outer stacks we have, respectively,\n\nThe part of the attenuation constant which is due to dielectric and \nmagnetic dissipation is therefore\n\nWe need scarcely point out that if the loss tangents are not small \ncompared to unity, it may be impossible to satisfy Clogston\u2019s condition \n(102) very closely with a real value of 6, and the resulting mismatch \nmay reduce:\u2018the depth of penetration and increase the metal losses in \nthe stacks. In practice, however, the loss tangents will be of the order \nof 0.001 or even 0.0001, and matching the imaginary parts of woe) and \nj\u00e9 will be much less of a practical problem than matching the real parts.\n\nLet pi and pe be the inner and outer radii of a cylindrical shell and \nlet the thickness t, given by\n\nbe less than p; . Then, following Schelkunoff,\u2019 we may replace the Bessel \nfunctions appearing in equation (68) of Section III by their Taylor ex- \npansions, namely\n\nex (A2) \nIulvps) = Tolees) = Yo \"19 (ens, \nad of ! iat = (xt)\u201d (n+1) \nKx(kpo) a Ko(xpe) = a nl Ko (kp1). \nIt follows that , \noo i n \nKo(kpi)I 1(kpe) ai Kx(kpo) Lo(xp1) eae a a Bryilkpi), \n0o i n \nKaeor)Iu(ees) \u2014 Kaleoa)Io(ses) = \u2014 2  Bylups), \nca : (A3) \n(x\u00e9)\u201d \nKilxpr)Likp2) \u2014 Ki(kp2)Li(kp1) = a ae Ansi(kpi), \nKu(xp1) Lo(kp2) Bo Ko(kp2)L1(xp1) = dX uo\" A, (xpi), \nwhere \nA,(x) = Ip(a)K\u00a7\u201d (x) \u2014 Ko(x)I8 (2), (M4)\n\nThe quantities A,(#) and B,(x) turn out to be finite polynomials in \n1/z, the general expressions for the coefficients having been derived in \na rather inaccessible monograph by Pleijel.\u201d When x is large, however, \nthe leading terms are quite simple. From Pleijel\u2019s analysis, or directly \nby substituting the asymptotic series for Jo(x) and Ko(x) into (A4), we \nfind\n\nIf we substitute these approximations into the first of equations \n(A38), we obtain\n\n2H. Pleijel, Berdkning af Motstand och Sjdalfinduktion, K. L. Beckmans Bok- \ntryckeri, Stockholm, 1906.\n\nThe other three equations may be treated similarly. Doing so, and \nremembering that\n\np2o/p. = 1 + t/pi, (A7) \nwe obtain the results which were quoted in Section . namely \nKp2( Kol 12 + Kyl) & 1+ st ch xt \u2014 oa sh Kt, \nxpx(Kulo: \u2014 Koeln) | 1+ \u2014 x sh xt,\n\nNote: Rationalized MKS units are employed throughout. The sub- \nscripts 0, 1, 2 applied to symbols representing material constants, such \nas \u20ac, uw, g, , and 7, have the significance that 0 refers to the main dielec- \ntric in a Clogston line, while 1 refers to the conducting layers and 2 re- \nfers to the insulating layers in the stacks. Bars denote average values. \nSubscripts not included in the present table are explained in the context \nwhere they are used.\n\na: Distance between outer sheaths of plane Clogston line. \nRadius of inner core of coaxial Clogston line.\n\nCharacteristic or iterative impedances of laminated stack \n(introduced in Section ITI).\n\nA parameter related to dielectric mismatch in a Clogston \n1 line (Section VI).\n\nRectangular coordinate in the direction of magnetic field \nina plane Clogston line.\n\nRectangular coordinate in the direction normal to the \nstacks in a plane Clogston line.\n\nSurface impedance of a plane or cylindrical boundary; \nratio of tangential components of the electric and mag- \nnetic fields (subscripts explained in context).\n\nI'/(i + t); average propagation constant per unit dis- \ntance normal to laminated stack.\n\nEffective skin depth; the depth at which the current den- \nsity in an infinite plane stack has fallen to 1/e of its value \nat the surface. A small change in a quantity.\n\ntan\u201d '(u\u201d/u\u2019); phase angle of complex permeability. \niwp/(g + twe); intrinsic impedance of medium.\n\nn(1 \u2014 y/o)\u2019; characteristic impedance looking in the \ny- or p-direction in a homogeneous medium.\n\n(1 + 1)t/6: ; a parameter related to the electrical thick- \nness of a.conducting layer.\n\n(o*? \u2014 7\u2019); transverse propagation constant in the y- or \np-direction in a homogeneous medium.\n\nA parameter related to the propagation constant in a \nClogston 2 (Section XII).\n\ny/a + 4; normalized coordinate transverse to a plane \nClogston 2 line (Section XII).\n\nAngular coordinate in cylindrical system. Phase angle, \ntan\u2019 \u2018(e\u2019/e\u2019), of complex dielectric constant.\n\nBessel (Neumann) functions of the second kind. \nModified Bessel functions of the first kind. ~ \nModified Bessel functions of the second kind.\n\nis the spectrum. The spectral density (power per unit bandwidth) varies \ninversely as the frequency, according to the relation\n\nwhere the exponent lies between 1 and 1.5 with an average about 1.2. \nThis type of spectrum will be referred to as a 1/f spectrum. Measure- \nments of the spectra of silicon point contact diodes have been reported \nby P. H. Miller\u2019 for the frequency range 20 cycles to 300 kilocycles. \nSpectra of point contact transistors measured by the author have been \nreported elsewhere\u2019\u2019 * for the range 20 to 15,000 cycles. Typical spectra \nfor p-n junction type diodes and transistors are shown in Fig. 1. Almost \nwithout exception, our measurements and those reported in the litera- \nture have shown the 1/f spectrum over most of the frequency range \ncovered. There is some evidence from the related fields of flicker noise \nand carbon microphone noise that the 1/f spectrum may extend to fre- \nquencies well below 0.1 cycle per second. Some departures from this \ntype of spectrum have been noted in the neighborhood of 100 ke, as \nshown in the curves.\n\nFig. 1\u2014The spectrum of noise in n-p-n transistors varies inversely with fre- \nquency.\n\nFig. 2\u2014The short-circuit noise current from a point contact or junction diode \ngenerally increases with de bias current.\n\ni 0.2 a 0.6 0.8 1.0 40 \nCOLLECTOR BIAS nN voLTs \nFig. 3\u2014The noise figure of an n-p-n transistor depends in a fairly simple way \non emitter current and collector voltage.\n\nture on noise behavior. Such experiments have been rather unsatisfac- \ntory because the changes in impedance and gain characteristics as a \nfunction of frequency are of the same order as the changes in noise prop- \nerties. This makes the interpretation ambiguous. By and large, such \nexperiments suggest that changes in noise with temperature are rather \nsmall, perhaps of the order of the change in absolute temperature, and \nnot at all like the exponential changes associated with a diffusion process. \nThis observation does not necessarily rule out a diffusion-like noise \nprocess; it might indicate merely that we are not looking at the right \npart of the spectrum to observe exponential changes with temperature.\n\nIt has been observed. that in many semiconductor structures the \nnoise voltage is approximately proportional to the de bias current. \nThis relation suggests that the noise is the result of fluctuations of the \nconductivity of the material, which modulate the bias current and pro- \nduce a fluctuating voltage across the specimen. Such fluctuations in \nconductivity could result from variations in concentration of the mi- \nnority carrier (holes in n-type material, electrons in p-type). The mag- \nnitude of the observed noise and the type of spectrum seem to demand \nthat the fluctuation be coarse-grained in time to a much greater extent \nthan could be accounted for by random statistical fluctuations of carrier \ndensity. Experiments of Haynes\u201d on lifetime and transit of injected car- - \nriers in rods of germanium have occasionally indicated finite sources of \nminority carriers in the material. Our hypothesis is that such sources \nof carriers are rather generally distributed over the material (although \nmostly too small to be noticed in experiments of the Haynes type), and \nthat their activity is being modified at a slow rate by some unspecified \nlocal influence in a suitable way to agree with the observed noise spec- \ntrum. ;\n\nSamples, referred to as \u201cbridges\u201d, have been cut from thin slabs of \nsingle crystal germanium, by a technique devised by. W. L. Bond,\u201d \noften of a form shown in Fig. 4. Side arms for both the current and \nthe noise measuring electrodes have been found necessary to avoid \nspurious noise at the electrodes. A large inductance in the bias circuit \ngreatly reduces the effect of any noise voltage generated at the bias \nelectrodes. The spurious noise power from this source is seldom more \nthan a few per cent of that being measured. It should be noted that the \ncontact area for the noise measuring electrodes should not be on a por- \ntion of the specimen carrying bias current, otherwise spurious noise may \nbe generated at these electrodes. Typical dimensions for the straight \ncentral filmanent of the bridge are 0.05 x 0.05 x 0.7 cm. The side arms \nhave sandblasted surfaces to suppress holes or electrons injected at the \nelectrodes. The central portion may be etched, sandblasted, or other- \nwise treated at will. The enlarged circular areas are rhodium plated to \nprovide good contacts to each side arm.\n\nMeasurements of the noise spectrum in such bridges with several dif- \nferent etching treatments and with sandblasted surfaces are charac- \nterized by the 1/f spectrum over a wide frequency range.* Fairly ex- \ntensive measurements have been made in the audio frequency range, \nand a few covering the range from 20 cycles to 1 megacycle. A typical \nspectrum is shown in Fig. 5.\n\nThe current dependence of the noise is shown in Fig. 6 for a number \nof samples, mostly n-type, one p-type, and with various resistivities. \nThe outstanding feature is that noise voltage always increases with de \nbias voltage. In many cases there is direct proportionality at the lower \nbias values, increasing to a square law at higher biases. There are some\n\nLENGTH \nABOUT 7MM \nFig. 4\u2014Filament with side arms cut out of a single crystal of germanium.\n\n* Departures from the 1/f spectrum at frequencies of the order of 100 kilo- \ncycles and above were first discovered by G. B. Herzog and A. Van der Ziel. \nSee Reference 13.\n\nFig. 5\u2014Typical spectra of noise in single crystal filaments carrying a de cur- \nrent.\n\nexceptions to this trend. Also, there are large variations in the magni- \ntude of the noise. An average unit shows a noise voltage about three \ntimes Johnson noise at a bias of-10 volts per centimeter.\n\nThe noise behavior at reduced temperatures has been investigated. \nResults on \u2018three different bridges are shown in Fig. 7. The open circuit \nnoise voltage is shown as a function of temperature for constant bias \nvoltage. Although the curves show rather large irregularities, there \nseems to be no general trend for noise to decrease with decreasing tem- \nperature over the range covered, from \u2014200\u00b0C to room temperature.\n\nThe surface treatment applied to a bridge may affect the noise very \nsubstantially. A sandblasted surface usually gives the lowest noise. \n_ Etching the surface may raise the noise voltage by a factor of ten or \nmore, though the de resistance changes only a few per cent. The tech- \nnique of washing and drying the surface may have an important effect\n\nBIAS FIELD IN VouTs PER CENTIMETER. \nFig. 6\u2014Variation of noise with de bias in single crystal filaments.\n\non the noise. Some of these processes also affect the lifetime of carriers \nin the bridge to a large extent. However, there seems to be no direct \nand simple relation between the two effects, since treatments have been \nfound which change the noise by a large factor with almost no effect on \nlifetime, and vice versa.\n\nFig. 8 shows measurements of noise voltage on several dozen bridges \nat a uniform bias of 10 volts per centimeter, all having sandblasted sur- \nfaces, mostly of n-type but a few of p-type germanium, and with widely \ndifferent values of resistivity, produced by varying impurity concentra- \ntions. There is considerable scatter in the results, but there is a fairly \nobvious tendency for noise voltage to increase in proportion to resis- \ntivity. Since Johnson noise also increases in proportion to resistivity in \na structure of fixed dimensions, the conclusion is that with constant bias \nvoltage the ratio of current induced noise to Johnson noise tends to be\n\nindependent of the resistivity of the material. From the data it also \nappears that there is no consistent difference between n- and p-type \nmaterial. .\n\nNoise does not appear to depend on orientation of the filament with \nrespect to the crystal axes. Filaments orientated along the 100, 110, and \n111 directions and rotated in several ways about these directions showed\n\nAn important role for the minority carrier in the noise mechanism \nwas first clearly indicated in experiments on the effect of a magnetic \nfield on noise in germanium filaments. It has been found experimentally \nthat the noise in a single crystal filament may change by a substantial \nfactor when the filament is subjected to a steady transverse magnetic \nfield. The following discussion will show that this behavior is in har- \nmony with the hypothesis of noisy injection of minority carriers, as set \nforth in a preceding section.*\n\nThe physical picture on which this treatment is based involves the \nrandom injection of holes into an n-type filament by hole sources which \nmay be either in the interior or on the surface of the filament.t It is \nassumed that the spectrum of the noise arises from the fluctuating na- \nture of the noise source. The effect which any source has will depend \nupon the lifetime of the holes which it emits. If these holes remain in \nthe filament for a long time, they will produce more noise than if they \nremain in the filament for a short time. We shall be concerned with \nthe effect of magnetic fields upon these lengths of time and shall not \ndeal in this paper with the fluctuations of the noise sources themselves. \nIf a transverse magnetic field is applied to an n-type germanium fila- \nment, a Hall effect voltage is set up and the holes will be deflected to- \nwards one surface of the filament. Since recombination takes place prin- \ncipally at the surfaces, this may cause a substantial change in the lifetime \nof the holes. In order to determine the effect of the magnetic field on \nthe noise we proceed along the following lines.\n\n(a) We assume that the observed noise is due to fluctuations in the \nconductivity of the filament produced by fluctuations in the hole con- \ncentration. Since these fluctuations are small, we may take the change \nin conducitivity to be proportional to the change in average hole den-\n\n* The following semi-quantitative theory of the dependence of noise on mag- \nnetic field is taken with some modification from unpublished work of W. Shock- \nley and H. Suhl, on the basis of which the calculations leading to the curves of \nFigs. 10 and 11 were carried out. It is hoped that this work may be published in \nthe near future.\n\n+ To simplify the terminology, the discussion is based on n-type material with \nholes as minority carrier. An exactly similar argument could be made for p-type \nmaterial with electrons as the minority carrier. There is some experimental evi-\n\ndence of the similarity of behavior of n- and p-type germanium, though most of \nthe experimental work has been done with n-type.\n\ncussion which follows, and also in the calculation of the curves of Figs. \n10 and 11.\n\n* It should be pointed out that a consequence of the hole injection theory of \nnoise in a filament is that marked frequency dispersion should occur when the \nfrequency being studied is high enough so that a period is short compared to the \nlifetime of holes in the filament. However, we shall neglect this important and \ninteresting aspect of the problem.\n\n(a) (b) \nFig. 9\u2014Excess hole density across the thickness dimension, (a) with no magnetic \nfield, (b) with moderate magnetic field.\n\nIf we suppose a steady hole current Jo emitted from the left-hand \nsurface of the filament, then a relatively high concentration 7; of holes \nwill appear directly in front of the surface. Some of these holes will re- \ncombine upon the surface, the rate J; being given by\n\nwhere S is the recombination constant for the surface. The balance of \nthe holes will diffuse through the filament to recombine upon the right \nsurface at a rate\n\nand we note that J; + Jz. = J. Because of the high recombination rate, \n2 will be very small; hence, J2 will be much smaller than J; . In the ab- \nsence of a magnetic field the gradient is uniform, and the concentrations \nwill be linear, as shown in part (a) of the figure. An identical argument\n\nThe constant may be derived by noting that k7'/q is 1/40 volt at room \ntemperature, and that the effective transverse field, Hy , may be ex- \npressed as follows. (See Reference 14, Section 8.8.)\n\nwhere 6 = Hall angle . \nHall mobility for electrons (2800 em\u2019/volt-sec) \nUy\u00bb = Hall mobility for holes (1500 em?/volt-sec).\n\nThe other dimensionless parameter is proportional to the rate of surface \nrecombination, and is defined as the ratio of the surface recombination \nvelocity to the diffusion velocity from the center:\n\nThe numerical constant is given for holes at room temperature. The \nnoise changes are expressed in decibles, that is, ten times the common \nlogarithm of the ratio of noise powers with and without the magnetic \nfield.\n\nA second case is that in which generation and recombination are on \nthe surfaces, but the two surfaces have unequal absorption properties. \nIt might be expected that rather large increases in noise would result \nwhen the magnetic field was poled to pull holes away from the surface \nwith high absorption properties, and this turns out to be the case when \nthe calculations are carried out. The results are shown in Fig. 11 for a\n\nSURFACE GENERATION \na~\u2014\u2014 VOLUME GENERATION \nyp SURFACE RECOMBINATION \nCONSTANT\n\n-16 -14 -12 -10 -8 -6 -4 -2 0 2 4 6 8.10 12 14 16 18 20 \nMAGNETIC FIELD PARAMETER, \u00ae \nFig. 11\u2014Calculated magnetic effect for dissimilar surfaces. Curve 1 of each \npair is for the contribution from the surface having the lower recombination \nconstant.\n\n-4 Oo 4 8 \nMAGNETIC FIELD PARAMETER, 6 \nFig. 12\u2014Experimental magnetic effect for dissimilar surfaces.\n\ncase where the two recombination parameters are 0.1 and 10, and.for \na second case where the parameters are 0.5 and 50. In this figure, sepa- \nrate curves have been shown for noise due to holes generated on each \nof the two surfaces. The total noise would be gotten by adding the noise \npowers represented by the two curves after appropriate weighting for \nthe contributions of the two surfaces. At present we do not see any way \nof determining the weighting factor.\n\nA third case is that in which it is assumed: that the noisy generation \nof holes is uniform throughout the body of the filament, but that re- \ncombination takes place on the surfaces only. These assumptions seem \nat first sight to be in contradiction to the statistical mechanical prin- \nciple of detailed balancing, which states that under equilibrium condi- \ntions all processes occur with equal frequency in the forward and reverse \ndirections. Thus it would seem that if holes are generated in the interior, \nwe must consider recombination in the interior also. Actually this is \nnot necessary under the non-equilibrium conditions which prevail \nduring noise measurements. There is no necessity for the noise generated \nby a source and a sink for holes to be simply related to the strength of. \nthis source. Thus we may suppose there are relatively weak sources and \nsinks for holes in the interior, but that the hole absorption and genera- \ntion of the sources is very noisy compared to the recombination and \ngeneration processes on the surfaces. If this is the state of affairs, most\n\nwhere EF is the applied field in volts per centimeter and uw is the drift \nmobility of holes. Hence, by varying the biasing voltage a large range \nof life path values can be obtained.\n\nThe correlation is measured by carrying the noise voltages through \nseparate amplifying channels having identical pass bands extending\n\nfrom 800 to 1300 cycles per second. A switching arrangement makes it \npossible to apply either of the output voltages or their sum or difference \nto a rectifier-meter combination. From the readings of the meter the \ncorrelation can be computed according to the relation \u2018:\n\nVY, and V2 are rms values of the individual noise voltages, and S and D \nare the rms values of their sum and difference. The equivalence to \nexpression (5) can be seen by noting that\n\nResults of correlation measurements on three bridges are shown in \nFigs. 14-16. In each case the calculated curve is shown for reference. \nThe values of \u00a2 were calculated from decay measurements on optically \ninjected holes, as described by J. R. Haynes,\u201d using a value for mobility \nof 1700 cm\u2019/volt-sec. In Fig. 14 the agreement with the theoretical model \nis very good. The scatter in the points is due to fluctuations in the \nnoise, which are quite large in the band used for these measurements. \nIn Fig. 15 the agreement could be made quite good with a lateral shift\n\nFig. 14\u2014Noise correlation. The solid curve is calculated, the points experi- \nmental.\n\nFig. 15\u2014Noise correlation. The dotted curve includes allowance for losses at \nthe side arms.\n\nby a factor of two. In Fig. 16 the form of the experimental curve seems \ndifferent from that calculated. In particular, the slope is steeper, and \nthe curve tends to level off at a correlation of about 0.8. It seems pos- \nsible to explain the discrepancies between the experimental data and \nthe calculations on the basis of two considerations which were not in- \ncluded in the model. (a) The pair of side arms separating the two seg- \nments of the filament serve to drain off some holes which would other- \nwise contribute to the correlation. The dashed curve in Fig. 15 shows \nthe calculated effect, on the assumption that the absorption in the side \narms is equivalent to an extra section of filament equal in length to half \na segment. The actual distance across the side arms is only 20 per cent \nof a segment, but it is not hard to believe that the decay rate in this \nregion might increase by a factor of two or three due to the reduced elec- \ntric field and loss of holes down the side arms. (b) The model assumed a \nuniform distribution of noise sources along the filament. There is ex- \nperimental evidence that the distribution may be quite spotty. This \ncan have a substantial effect on the form of the correlation curve. For \nexample, the dashed curve in Fig. 16 shows the curve calculated for \nnoise sources lumped at the mid-point of each segment. Other assumed \npositions might shift the curve considerably along the horizontal axis.\n\nFig. 16\u2014Noise correlation. The dotted curve is calculated for lumped noise \nsources.\n\nIn view of these considerations there seems to be very satisfactory agree- \nment between the experimental results and the model.\n\nAnother type of experiment involves noise measurements at frequen- \ncies high enough so that the transit time of a hole across a segment is \nan appreciable fraction of a cycle. In this case the correlation between \nnoise voltages from adjacent segments can be improved by putting a \ntime delay in one channel of the measuring circuit. Measurements were \nmade by taking the noise voltages from the two segments through sepa- \nrate amplifying channels having identical pass bands extending from 17 \nto 24 kilocyles. The outputs of the two channels were put on the vertical \nand horizontal plates of a cathode ray oscilloscope, forming a sort of \nLissajous pattern. The patterns differ from those obtained with sinus- \noidal voltages in that the elliptical figures are filled in solid, due to the \ncontinual variation in amplitude of the noise. A phase shifting device is \nincluded in one channel, and as the phase is shifted to give optimum \ncorrelation, the elliptical pattern narrows down and approaches a, line \ninclined at 45\u00b0. For a quadrature phase shift, the pattern becomes cir- \ncular, and in practice this setting can be determined more precisely \nthan the in-phase setting, largely because the background noise in the \ncircuit is less troublesome. With the phase shift for optimum correlation \ndetermined, the delay at the center of the pass band is easily calculated,\n\nData for a bridge of n-type germanium of resistivity about 20 ohm- \ncm are given in Table I. The transit distance, L, after a small correction\n\nfor reduced field across the side arm, was taken as 0.305 cm. As noted \nin the table, the bridge temperature rose somewhat at the higher bias \nvalues, and the assumed values of mobility have been modified accord- \ning to the inverse three-halves power of the absolute temperature. The \ndelay required for optimum correlation is shown in the second column \nof the table, and the calculated transit time between segments in the \nlast column. It is seen that the two are in reasonably good agreement, \nespecially at low fields. When the direction of the field is reversed, an \nequal delay is required, but in the opposite channel of the measuring \ncircuit, as would be expected. Here, again, we have experimental evi- \ndence supporting the noisy hole injection hypothesis. The cause of the \ndiscrepancy shown in the table at higher fields is not understood. It is \npossible that trapping phenomena increase the transit time over that \ncalculated from the mobility. There is some evidence for this sort of \nbehavior in lifetime experiments, but to date there does not seem to be \nenough information for any estimate of magnitude of such an effect.\n\nSuppose that a source of holes located at a point 2\u00bb in a filament pro- \nduces a fluctuating current of holes of rms value J; in a specified fre- \nquency band. The hole current is swept down the filament by a field # \nand is assumed to decay exponentially according to the relation\n\nwhere the life path \u00a3 may be expressed in terms of drift velocity v, hole \nmobility yw, and lifetime 7 \nLor = pr.\n\nAssuming that the frequency of measurement is low enough to justify \nneglecting the hole transit time, the noise voltage due to holes from a \nsingle source is proportional to the number of holes present in the seg- \nment. This is obtained by integrating (1) over an appropriate range\n\nUnder the assumption that the sources are statistically independent, \nthe total voltage squared is obtained by integrating the square of (2) \nover all the sources.\n\n| re an \n= K,{1 \u2014F Tze a, \nSimilarly, the cross product of voltages in two segments, extending say \nfrom 0 to L and L to 2, is\n\nwhich is the desired relation, from which the solid curves of Figs. 14-16 \nwere calculated.\n\nRelays are produced by a large number of manufacturers in this country. \nWhen we survey their product, we find that there are many kinds and vari- \nettes. They differ widely as to their size, shapes and configurations. Many \nof these differences are dictated by the requirements of the task they must \nperform and by the environments under which they must work. Other differ- \nences are brought about from considerations of cost and by the design and \nfabrication techniques the particular manufacturer employs. However, all \nrelays have a common. objective. For whatever use they are employed, it 1s \nhighly desirable that they be reliable. They are expected to function each time \nthe, are called upon without failure and over the expected life of the equip- \nments in which they are used. This paper deals with the more important \ndesign factors which all relays have in common that greatly influence their \nreliability of performance. Contact spring pile-up stability and the impor- \ntance of strength of screws, insulating materials with low cold flow and mois- \nture absorption, and manufacturing procedures and controls to achieve this \nend are discussed. Coil construction so as to minimize the occurrence of open \nwindings due to corrosion of the wire and breakage of the lead-out wires 1s \ndwelt upon. Contact reliability and how it 1s affected by the material used, \nits size and shape, the method of actuation, the presence of contaminating \nvapors, and single versus twin contacts are discussed. The degree by which \nmagnetic materials change their magnetic properties with age and treatments \nfor alleviating this effect are described. The importance of adequate structural \ndesign so that the relay will be rugged and remain stable so that its perform- \nance 1s substantially unaffected by wear, shock and vibration 1s stressed. \nMethods of test to determine how well the relay meets these objectives are \ndescribed.\n\nAlthough a relay is conceptually a simple device, the wide range of \nconditions under which relays are required to operate, the many different \ncharacteristics they must have, and the complete dependence placed\n\ncycles of alternate humid and dry environments over the years. Humid \nconditions exist during the summer season, followed by a dry exposure \nduring the winter months when the central offices are heated. Experience \nhas shown that a relay exposed for six days to 90 per cent relative hu- \nmidity at 85\u00b0F will be comparable to those observed in service in humid \nlocalities. Although the six day exposure is admittedly an accelerated \ntest, the dimensional changes produced are approximately the same as \nthose caused by the accumulating effect of fluctuating humidity during \nthe entire season. Similarly, a period of six days exposure to 120\u00b0F pro- \nduces the same effect of drying as that which occurs during the heating \nseason.\n\nBy making careful measurements on an adjusted relay of such impor- \ntant parameters as operate current, releasing current, contact separation, \ncontact spring tension, armature back tension, stud gaps, etc., and then \nsubjecting the relay to repeated cycles of humid and dry conditions, \nrepeating the measurements after each exposure and noting the changes, \na good appraisal of the relay can be made. A repetition of the test will \nreproduce the same pattern of results as the first cycle unless permanent \ndeformation of the materials in the relay has taken place.\n\nCOIL COVER COIL COVER \nWAXED_VARNISHED KRAFT PAPER INTERLEAVING \nINTERWINDING INTERWINDING\n\nDAYS ON TEST DAYS ON TEST \nAT 90% RELATIVE HUMIDITY AND 85\u00b0F AT 90% RELATIVE HUMIDITY AND 85\u00b0F\n\nCELLULOSE ACETATE INSULATED SPOOL- \nWOUND CGIL VINCELLATATE MUSLIN \nCOVER AND RED ROPE PAPER WASHERS \nIN CONTACT WITH WINDING WIRE (NOT \nIMPREGNATED).\n\nITA CELLULOSE ACETATE FILLED COIL \nHAVING LEAD WIRES INSULATED WITH \nCOTTON BRAID,(NOT IMPREGNATED).\n\nDURING EXPOSURE TO 95% RELATIVE HUMIDITY \n+100 VOLTS DC WAS APPLIED TO THE WINDING \nWITH NEGATIVE POTENTIAL ON CORE.\n\nae 4\u2014Corrosion comparison of impregnated and non-impregnated relay \ncoils.\n\nwinding washers. Likewise, one group was impregnated while the other \nwas not. Fifth and sixth groups of coils having cellulose acetate insulation \nthroughout with and without impregnation were exposed and there were \nno failures at the end of the test. This shows that the corrosive effects \nof impure materials can be retarded, but not overcome by resorting to \nimpregnation. In general, impregnation of relay coils is not desirable \nbecause of the risk of contaminating the vital working surfaces of the \nrelay.\n\nIn the normal operation of a relay when the circuit through its winding \nis opened, a transient voltage, which may reach hundreds of volts is \ngenerated across the winding terminals by the collapsing flux. If the \ninsulation between the lead-out wires or between the lead-out wires and \nthe end turns of the winding is not adequate, electrical breakdown causes \narcing and repeated operation of the relay may cause ultimate disintegra- \ntion of the wire and consequent failure of the relay. It is important, there- \nfore, to design the coil so that lead-out wires under all conditions are \nproperly spaced, and to provide adequate insulation between those por- \ntions of the winding where high voltages can exist. A test has been de-\n\nFig. 5\u2014Relay coil employing a motion limiting washer to prevent lead breakage.\n\nSince the opening and closing of contacts are the prime objectives of \na relay, it is extremely important that the contacts themselves are made \nreliable. To realize these objectives, several factors should be taken into \nconsideration. First is the contact material. While much could be said \nregarding the behavior of contact metals, space does not permit more \nthan a brief treatment. The contact should maintain reasonably low re- \nsistance and under the environment in which it is used, be able to with- \nstand the erosion. Electrical resistivity of most metals is low enough to \nbe satisfactory from the resistance standpoint, but unfortunately, most \nof them develop tarnish or corrosion films when exposed to the atmos- \nphere, thus increasing the contact resistance and rendering them unsuit- \nable for a contact. These metals are sometimes referred to as \u2018\u2018base\u201d\u2019 \nmetals, and include aluminum, brass, bronze, copper, chromium, nickel \nand stainless steel. There is a much smaller group of metals known as \n\u201cnoble\u201d or precious metals, such as platinum, palladium, gold and irid- \njum. These are relatively free from the tendency to tarnish and will \nmaintain low contact resistance. Alloys of these metals and certain alloys \nin which silver is included are widely used in the telephone plant. Pure \nsilver is also used and is attractive because of its low cost; however, it \nhas a tendency to form high resistance tarnish films and therefore has \nlimitations in its use. It is employed in signaling circuits where the con- \ntact makes or breaks current. Its contact resistance remains low because \nthe films that form on the silver are broken down or destroyed by the \nare. It is not employed in circuits carrying voice currents on account of \nits tendency to introduce noise.\n\nEnough metal must be provided to give satisfactory life. Each time a \ncontact makes and breaks an electrical circuit, a small part of the metal \nmay be lost, so that life may be considered roughly proportional to the \nvolume of metal available for erosion. The pair of contacts must have \nsufficient height to provide enough contact spring clearance to allow for \nspring adjustment and to insure that the springs will not touch during \nthe normal operation of the relay. At least one contact of a pair must \nbe large enough, that is, present a sufficiently large target area, to insure \nfull registration of the contacts with normal manufacturing variations \nof the position of the contacts on the springs and with the variations in \nalignment of the springs during assembly.\u2019\n\nFig. 6\u2014Two methods of contact spring actuation and their influence on contact \nlocking.\n\noperation is \u2018\u2018open\u2019\u201d\u2019 contacts due to small insulating particles present \nin the atmosphere becoming trapped between the contacts. This causes \nhigh resistance or open circuit and consequent circuit failure. Many at- \ntempts have been and are being made to reduce \u201copen\u201d contact troubles. \nExamples are filtering the air supply to the central office, enclosing the \nrelay equipments in closed cabinets, pressurizing the enclosing cabinets, \ncovering smaller groups of contacts by independent covers, employing \ntwin contacts rather than single contacts, and enclosing the relay or its \ncontacts in a hermetically sealed chamber. Even going to the extreme \nof completely isolating the relay from its surroundings is not a complete \nanswer. There is always the possibility of failure by wear particles gen- \nerated within the enclosure by the relay actuation.\n\nThe most widely employed method to reduce dust failures in the tele- \nphone plant is the use of twin contacts in combination with some of the \nabove mentioned types of enclosures. If the incidence of dust failures fol- \nlowed the laws of probability, then elementary considerations would lead \nus to predict that if single contacts failed at the rate of once in 1,000 oper- \nations, the simultaneous failure of the two such contacts comprising the \ntwin would be once in 1,000,000 operations. This is the so-called \u201csquare\u201d\u2019 \nlaw.. However there are a number of reasons why this is not realized \nin practice and why the figure of merit for the twin contact is very much \nless than that indicated by the \u2018square\u2019 law. In the first place, when \nforeign matter becomes lodged on a contact, it seldom falls out on the \nfirst subsequent operation, but will require a number of operations before \nit cleans itself; in fact, it may remain inoperative indefinitely. When this \nhappens to one of the twin contacts, during this period of time, the twin \ncontact is no better than the single contact. In practice, twin contacts \nare generally used with the same total force as the single contact, being \nnominally divided equally between the two contacts. This reduction in \nforce per contact on the twin contacts is of considerable importance in \nreducing its effectiveness. Relay designs employing twin contacts that \nhave been used in the past do not have complete mechanical inde- \npendence of the two members to which the contacts are attached. \nForeign material or protrusions under one contact can adversely influence \nthe performance of its mate. In a new design of relay which is about to go \ninto production, the design criterion that twin contacts to be most \neffective should be completely and mutually independent has been met. \nLaboratory tests and field experience obtained to date show a marked \nimprovement over the former designs in regard to the incidence of open \ncontact failures:\n\nOPERATE AMPERE TURNS \nFig. 8\u2014Improvement in relay performance by hydrogen anneal.\n\nARMATURE MOTION IN INCHES \nFig. 9\u2014Mechanical load of a relay and its friction component.\n\nthe displacement and hence the friction is large, aside from the fact \nthat it is indicative of a rapid rate of wear, the relay would be unstable.\n\nWhile the wear of the relay parts can be minimized by good design, \nit cannot be eliminated entirely, especially for relays required to operate \na very large number of times during their life. For telephone relays the \ndesign objective is for a 40-year life. The effects of wear on performance \nto a great extent can ofttimes be counteracted by ingenious design. \nFig. 10 is an illustration of such a case.> The diagram on the left shows \na moving system of a relay in which the contact springs are stud ac- \ntuated. The moving springs are tensioned toward the armature and exert \na force tending to open the contacts. When the armature operates, the \nstud presses the moving springs into engagement with the stationary \nsprings. There is no contact force when engagement is first made and \nfurther flexing of the spring is necessary to build up the contact force \nto the desired value when the armature reaches its fully operated posi- \ntion. As the contacts and studs wear, it is apparent that the contact \nforce and consequently the load on the armature decreases rapidly. The \nstud wear becomes cumulative in its effect on the outside pair of springs \nas more springs are added to the pile-up.\n\nThe diagram to the right shows a moving system of a relay using \nwhat is called \u201clift-off\u201d card actuation. The moving springs are ten-\n\nFig. 10\u2014Two moving systems of relays in relation to the effects of wear on \ntheir performance.\n\nFig. 13\u2014Comparison of flat and embossed pole surfaces and their magnetic \nclosed circuit reluctance with misalignment.\n\ndone with the ordinary quick-to-release relays, for slow release relays \nthe armature is allowed to contact the core, finish to finish. When plane \nflat pole face surfaces are provided, it is expensive and difficult to insure \nin commercial practice that precise and uniform alignment of the pole \nface surfaces will obtain. Variations in the alignment of these two sur- \nfaces will cause variations in the closed magnetic circuit reluctance and \nconsequently on the release time of the relay.\n\nIn Fig. 12 is shown a design where the necessity for holding the align- \nment of core and armature so precisely is not so great.\u00ae A spherical surface \nof rather large radius is embossed on the front end of the armature, so \nthat with commercial variations in alignment, the armature always pre- \nsents a point on the surface of a sphere for contacting the flat surface of \nthe core. Similarly, the legs of the armature where they pivot on the \nfront ends of the hinge bracket are likewise embossed. The results of \nthe effects of these structural differences on the closed circuit re- \nluctance are shown in Fig. 13 for a design with flat surfaces and one \nwith embossed surfaces. While it is true that with perfect alignment the \nrelay with flat surfaces will give longer release times, it is apparent that \nas variations in alignment occur from time to time and from relay to \nrelay, it will have larger variations in performance than the relay with \nthe embossed surfaces. This is a feature which has proven of great value \nin the manufacture of slow release relays of reasonable time precision.\n\niif Oa ponueicet \u201cCellulose Acetate Filled Coils,\u2019 Bell Labs. Record, 29, p. 514, \nov., 1951. \n2. W. C. Slauson, \u2018\u2018Improved U, UA and Y Type Relays,\u2019 Bell Labs. Record, 29, \np. 466, Oct., 1951.\n\nFor measuring impedance and admittance parameters, that is 2, L, \nC and G, suitable ac bridges, ordinarily simply designated as impedance \nbridges, have long held a high place in the Bell System because of their \ninherent reliability and precision, and their ability to cover a wide \nrange of values. The development of many of the original bridges\u2019 \u201d *' * \nfor frequencies above the audio range stemmed from the needs of the \nearlier carrier systems. With this development came also analysis of \nshielding technique,\u2019 standardization of capacitance,\u2019 and a syste- \nmatic classification of bridge methods\u00ae by J. G. Ferguson in 1933, in \nwhich bridges were grouped into two major types designated as ratio- \narm and product-arm, respectively. Following this classification, com- \nbined impedance and admittance bridges were developed,\u201d *\u00b0 utiliz- \ning a single set of bridge standards for both kinds of parameters by \nchanging the configuration of the bridge network. There have also \nbeen special purpose bridges\u2019 \u201d'*'\u201c for use at audio and the lower \ncarrier frequencies. More recently, coaxial impedance standards\u201d having \nvalues calculable from physical dimensions have been developed.\n\nBridges for frequencies above one-half megacycle were used in the \nBell System as early as 1919,\u2019\u00b0 but relatively few bridges were built \nuntil the mid 1930\u2019s when new carrier systems required bridges in the\n\nfor their best performance, even though both bridges are useable at \nhigher frequencies. It will be observed that while there is some over- \nlapping of the three ranges, all three methods are necessary to obtain \nthe impedance coverage shown. It should be emphasized that all the \nranges shown cover both capacitive and inductive reactances. In the \ncase of the admittance and series-reactance bridges, inductive imped- \nances are measured by using a resonating capacitor, in parallel or \nseries, respectively, with the apparatus being measured. In the Maxwell \ninductance bridge, capacitive impedances are measured by using a \nfixed resonating inductor in series with the impedance under test. A \ncomplete accuracy statement for these bridges is necessarily complex, \nbut in general accuracies of +0.25 per cent for the major component\n\nFig. 2\u2014One-megacycle Maxwell inductance bridge, shown schematically in \nFig. lc, designed for relay-rack mounting.\n\nFig. 3\u2014Reactance/frequency chart showing the measurement range of the \nbridges shown in Fig. 1.\n\nfrequency of twenty megacycles was decided upon as a design objective \nwith a basic accuracy of +0.5 per cent for the major component. The \nimmediate need was for a general-purpose bridge, but it was expected \nthat special-purpose bridges having better accuracy would be required \nlater.\n\nFig. 4\u2014Reactance/frequency chart applying to the general-purpose bridge \nshown in Fig. 5.\n\nThe successful use of a center-tapped transformer for ratio arms in a \n465-KKC direct capacitance bridge\u2019 indicated that the resistance ratio \narms RI, r2 of Fig. 1 might be omitted if a suitable transformer could be \ndeveloped for higher frequencies. The transformer group of the Labora- \ntories succeeded in producing a transformer with a deviation from unity \nratio of less than 0.1 per cent over a frequency range from 0.5 to 20 \nmegacycles. This was made possible by precise location of the windings \nin fine milled grooves in the form of reversed helices, cut on a longitudin- \nally-split brass cylinder for the inner winding, and ona surrounding phenol\n\nFig. 5\u2014Schematic of the 20-megacycle general-purpose bridge showing both \nthe series (impedance) and parallel (admittance) bridge circuits combined in a \nsingle unit.\n\nfibre cylinder for the bifilar outer winding which serves as the bridge \nratio arms. Electrostatic shielding limits the direct capacitance be- \ntween primary and secondary to less than 0.01 uyf. The core material is \ncompressed powdered molybdenum permalloy. This transformer was \nthe nucleus around which the general purpose bridge was built, and the \nresulting bridge is shown schematically in Fig. 5.\n\nCapacitors cP and cs are worm-driven air capacitors with a range of \nabout 220 puyf, and were specially designed for this bridge. In the case of \ncs, any direct conductance between rotor and stator would result in an \neffective series resistance which would vary both with frequency and \ncapacitor setting, and therefore require laborious correction. This was \navoided by arranging the construction so that the rotor and stator are \nmounted on independent insulating supports to the ground panel, thereby \ncompletely eliminating direct conductance from rotor to stator. While \nthis results in some conductance from test terminal x1 to ground, the \namount is small and its effect is negligible because of the relatively low \n\u2018impedance values measured. In the case of cp, on the other hand, it is \nimportant to minimize series resistance and inductance to avoid con- \nductance and capacitance corrections which would change both with \nfrequency and capacitor setting. This was accomplished by careful design \nof the rotor brush using silver contact surfaces and center-fed connections \nto both rotor and stator.\n\nThe conductance standard, cp, and resistance standard, rs, were de- \nsigned to emphasize high-frequency performance. Deposited carbon resis- \ntors\u2019 on ceramic rods 3\u201d in diameter and 2\u201d long mounted on small \ndecade rotors were used, so arranged that only one resistor on a rotor is \nin the circuit at any time, and that adjacent resistors are short-circuited \nby means of auxiliary shorting brushes to eliminate shunting admittance \nwhich might vary with frequency. For er the resistance values are such \nthat the two lower decades and the slide-wire rheostat each have a \nresidual conductance of 333 micromhos, thereby avoiding the use of \nresistors exceeding 3,000 ohms in value which would be more likely to \u2014 \nvary with frequency. The structure is designed to minimize series in- \nductance and to maintain constant capacitance for all settings. For Rs, \non the other hand, it is necessary to maintain constant inductance for \nall settings. This was accomplished by adding small wire-loop compen- \nsating inductors in series with individual resistors in the 10-ohm and \n100-ohm decades when necessary. To minimize the over-all inductance, \nthe resistor rotors are placed very close together and are driven by gear- \ning from the corresponding dials.\n\nrotor switch on which are mounted five uncalibrated mica capacitors \nwhich enable cs to measure both positive and negative reactance values \nup to 10,000 uuf without additional switching. The 20 uyf cr capacitor \ncovers capacitance measurements up to 60 yuf; the 40 uyf capacitor \ncovers up to 150 uf; the 80 uuf up to 600 uyf; the 140 uuf up to 10,000 \npuf; and the 200 yyf capacitor covers all the positive series reactance \nmeasurements. Since the cr capacitor permits the bridge to be balanced \nwith the test leads short-circuited, the value of the effective resistance \nunder test is simply equal to the difference between rs readings for the \nmeasurement balance and the short-circuit balance, and the reactance \nunder test is determined from a computation of the two readings of cs.\n\nA front view of the general purpose bridge is shown in Fig. 6. The four \nlower dials are for Gp; above them are the four Rs dials; and above them \nis the cr dial. The capacitors cs and cp are located adjacent to the test \nterminals, but are operated remotely by the dial knobs at the extreme \nright end of the bridge. This was done to remove the operator\u2019s hands as \nfar as possible from the test terminals. Near the test terminals is a coaxial \nconnector engraved a. This allows plug-in capacitors (cu in Fig. 5) to \nbe added in parallel with cp for extending the capacitance range. Com- \npact silvered mica capacitors in steps of 200 uf are used. Fig. 7 shows \nthe interior of the same bridge with cp and cs in the lower foreground, \nap at the left and Rs in the upper right.\n\nFig. 6\u2014Front view of the general-purpose bridge shown in Fig. 5. The bridge \nis approximately 103 inches high and 19 inches wide.\n\nFig. 7\u2014Interior view of the general-purpose bridge. The panel edge shown in \nthe foreground is the left edge of the bridge shown in Fig. 6.\n\nFig. 8\u2014The five-megacycle Maxwell inductance bridge is approximately 12} \ninches high and 19 inches wide. Test terminals are at upper left, and the three \nknobs at lower right are zero-balance adjusters.\n\nFig. 9\u2014Interior of bridge of Fig. 8 showing the shielding for test terminal X1 \nin foreground; at the left is the calibrated air capacitor; at the right are the con- \nductance decades using glass-sealed deposited-carbon resistors.\n\nDevelopment of capacitors for the L3 coaxial system has required a \nnew ten-megacycle admittance bridge. Intended especially for determin- \ning temperature coefficient and frequency characteristics of small capa- \ncitors, the bridge is capable of measuring capacitance values up to 200 \nuuf with a precision of +0.01 uyf, and a wide range of conductance values. \nUnlike the other two bridges described which make grounded measure- \nments only, this bridge is arranged for direct and balanced-to-ground \nmeasurements as well. This is accomplished by using the ratio-arm trans- \nformer already described in combination with a simple grounding circuit \nusing a three-position key, as shown in the bridge schematic of Fig. 10.\n\nFig. 10\u2014Ten-megacycle admittance bridge with three-position key for shift- \ning ground to B for measuring direct admittance, to junction of Cl and C2 for \nbalanced admittance, and_to D for grounded admittance. Unknowns may be \nconnected from A to D or Cto D.\n\nBridges have been developed for the measurement of impedance and \nadmittance parameters at megacycle frequencies with accuracies hereto- \nfore possible only at much lower frequencies. Several of the twenty- \nmegacycle general-purpose bridges have been built and are furnishing \nuseful measurements of networks and components. Experience with these \nbridges has indicated ranges for which supplementary special-purpose \nbridges would be desirable, and two such bridges have been built: a \nMaxwell bridge for low-valued inductors, and an admittance bridge for \nlow-valued capacitors. One feature of all of these bridges not generally \navailable in commercial measuring instruments for megacycle frequen- \ncies is the provision of standards having a range of several decades. These \nallow balances to be made with greater precision over a wider range of \nphase angles in the apparatus under test, and assure that the absolute \naccuracy will not be limited by readability. This added precision is very \nuseful in comparing similar components or in measuring characteristics \nsuch as temperature coefficient.\n\n1. W. J. Shackelton, \u2018\u2018A Shielded Bridge for Inductive Impedance Measure- \nments,\u2019\u201d\u2019 Bell System Tech. J., 6, pp. 142-171, Jan., 1927.\n\n2. W. J. Shackelton and J. G. Ferguson, \u2018\u2018Electrical Measurement of Communi- \ncation Apparatus,\u2019 Bell System Tech. J., 7, pp. 70-89, Jan., 1928.\n\n3. J. G. Ferguson, \u2018\u201cMeasurement of Inductance by the Shielded Owen Bridge,\u201d\u2019 \nBell System Tech. J., 6, pp. 375-386, July, 1927.\n\n5. J. G. Ferguson, \u2018\u2018Shielding in High-Frequency Measurement,\u201d\u2019 Bell System \nTech. J., 8, pp. 560-575, Aug., 1929.\n\nAbstracts of Bell System Technical Papers\u201d \nNot Published in This Journal\n\nA Full Automatic Private-Line Teletypewriter Switching System. W. M. \nBacon! and G. A. Locke\u2019. Trans. A.I.E.E., 70, Part 1, pp. 473-480, \n1951. (Monograph 1837).\n\nThis paper describes a full automatic teletypewriter message switching system \nfor use in private-line networks involving one or more switching centers and a \nmultiplicity of local or long-distance lines, each of which may have one or more \nstations. This system provides fast teletypewriter communication from any \nstation to any other station or group of stations in the network. At its point of \norigin a message first is perforated in tape accompanied by suitable directing \nand end-of-message characters, thereafter it is transmitted automatically, stored \ntemporarily in perforated tape at a switching office, and then routed at high \nspeed to its point or points of destination. Important features are the arrange- \nments provided to permit efficient use of long full duplex transmission lines, the \nfull automatic handling of multiple-address messages with only a single originat- \ning transmission, and the various guards and alarms which.are provided to \nprotect against loss of messages in case of trouble.\n\nOperational Study of a Highway Mobile Telephone System. L. A. \nDorrr. Trans. A.I.E.E., 70, Part 1, pp. 31-37, 1951. (Monograph \n1838).\n\nThe Dynamics of the Middle Ear and Its Relation to the Acuity of Hear- \ning. H. Furrcumr\u2019. J. Acoust. Soc. Am., 24, pp. 129-131, March, 1952.\n\nThe transformer action of the middle ear as measured by Bek\u00e9sy is shown to \nbe the principal cause for the low acuity of hearing for low frequencies. Because \nof the very low mechanical impedance across the basilar membrane at low fre- \nquencies, large acoustical pressures in front of the ear drum produce appreciable \nacoustical pressures across the basilar membrane. For example, at 100 cps this \npressure is thirty times and at 6000 cps it is one-tenth that created across the \nbasilar membrane.\n\nDiffusion of Donor and Acceptor Elements Into Germanium. C. \u00a7. \nFuutuer\u2019. Phys. Rev., 86, pp. 136-137, April 1, 1952.\n\n* Certain of these papers are available as Bell System Monographs and may be \nobtained on request to the Publication Department, Bell Telephone Laboratories, \nInc., 463 West Street, New York 14, N. Y. For papers available in this form, the \nmonograph number is given in parentheses following the date of publication, and \nthis number should be given in all requests.\n\nA Submarine Telephone Cable with Submerged Repeaters. J. J. GILBERT\u2019. \nTrans. A.J.E.E., 170, Part 1, pp. 564-572, 1951. (Monograph 1815).\n\nPhysical Structure and Magnetic Anisotropy of Alnico 5. Part I. R. D. \nHemenrercu and E. A. Nessirr\u2019. Jl. Appl. Phys., 23, pp. 352-371, \nMarch, 1952. (Monograph 1976).\n\nIt is concluded from electron metallographic results that the high coercive \nforce and anisotropy of Alnico 5 are caused by a very finely divided precipitate \nproduced by the permanent magnet heat treatment. This precipitate is a transi- \ntion structure rich in cobalt and is face-centered cubic with a9 = 10A and ap- \npears as rods growing along the [100] directions of the matrix crystal when no \nmagnetic field is applied during heat treatment. The size of the precipitate rods \nat optimum properties is approximately 75-100A by 400A long. The spacing \nbetween rows of rods is about 200A. The rods are not distinctly resolved in the \nelectron images unless they are grown by aging at 800\u00b0C. Their orientation and \nstructure is clearly evident in the electron diffraction patterns at all stages of \ngrowth. The precipitate responds to a magnetic field applied during heat-treat- \nment both by suppression of nuclei making an angle greater than about 70\u00b0 with \nthe field and by the forcing of the rods off the [100] direction into that of the \nfield. The precipitate rods tend to scatter in direction about the field vector when \nthe field is off the [100] but are aligned accurately when the field is along [100].\n\nIt is shown that the band or itinerant electron model of a solid is capable of \naccounting for the \u201cexchange stiffness\u2019? which determines the properties of the \ntransition region, known as the Bloch wall, which separates adjacent ferromag- \nnetic domains with different directions of magnetization. In this treatment the \nconstant spin function usually assigned to each running electron wave is replaced \nby a variable spin function. At each point of space the spin of a moving electron \nis inclined at a small velocity-dependent angle to the mean spin direction of the \nother electrons, and this gives rise to an exchange torque which makes the spin \ndirection of the given electron precess as it moves through the transition region, \nthe precession rate being just sufficient to keep it in approximate alignment with \nthe macroscopic magnetization. Physical insight into the mechanisms involved \nis provided by a rigorous solution of the wall problem for a ferromagnetic free \nelectron gas in the Slater-Fock approximation, although it is known that the \nfree electron gas is not likely to be ferromagnetic in higher approximations. \nRough upper limits to the exchange stiffness constants for actual ferromagnetic \nmetals can be calculated without using any empirical constants other than the \nsaturation moment and the lattice constant. The results are only a few times \nlarger than the observed values.\n\nElastic and Plastic Properties of Very Small Metal Specimens. C. \nHerrinG and J. K. Gaur\u2019. Phys. Rev., 85, pp. 1060-1061, March 15, \n1952. (Monograph 1977).\n\nA Scanner for Rapid Measurement of Envelope Delay Distortion. L. E. \nHunt\u2019 and W. J. Aupersuetm\u2019. Proc. I.R.E., 40, pp. 454-459, April, \n1952. (Monograph 1967).\n\npossible mechanisms of the current multiplication process in the Type-A transistor \nare discussed. One of the mechanisms is based on trapping holes in the collector \nbarrier of the semiconductor. By means of this trapping model, the effect of \nemitter current and temperature on the current multiplication is predicted. It \nis shown that these predictions are in reasonable accord with experiment. \nFurthermore, assuming this model to hold, the trap density and activation energy \n(produced by forming) may be evaluated.\n\nTransistor Forming Effects in n-Type Germanium. L. B. Vaupzs\u2019. \nProc. I.R.E., 40, pp. 445-448, April, 1952. (Monograph 1969).\n\nSome of the effects of electrical forming of the collector of an n-type germanium \ntransistor are discussed. Evidence is presented for the existence of a region of \np-type germanium underneath the formed electrode, together with some indica- \ntion of the size of the formed region. These experiments lend support to the p-n \nhook mechanism in that they explain the observed high values of alpha in tran- \nsistors. This relation is discussed.\n\nDomain Structure of Perminvar Having a Rectangular Hysteresis Loop. \nH. J. Wiuitams\u2019 and M. Gorrrz\u2019. Jl. Appl. Phys., 23, pp. 316-323, \nMarch, 1952. (Monograph 1985).\n\nAn investigation has been made of the magnetic domain structure of Permin- \nvar (43 per cent Ni, 34 per cent Fe, 23 per cent Co) ring specimens having rec- \ntangular hysteresis loops after heat-treatment in a magnetic field. Domain \npatterns obtained with colloidal magnetite showed curved domain boundaries \nextending completely around the rings, forming circles concentric with them. \nChanges in magnetization occur when an applied field causes the circular bound- \naries either to expand or contract so that there is a change in the relative values \nof clockwise and counter-clockwise flux. A nucleus of reversed magnetization \nwas formed by making a small notch in a specimen, and this decreased the co- \nercive force and hysteresis loss by a factor of two. It was found that in a 180\u00b0 \ndomain boundary it was possible to make the change in spin orientations, which \noccurs in going from one side of the boundary to the other, have either a right- \nor left-hand screw relation, by the application of a field of appropriate sign per- \npendicular to the surface. The effect of superposing an applied alternating field \nwas also investigated, and an effective permeability of 4,000,000 was obtained.\n\nMeasuring Techniques for Broad-Band. Long-Distance Radio Relay \nSystems. W. J. ALBERSHEIM . Proc. I.R.E., 40, pp. 548-551, May, 1952. \n(Monograph 1971).\n\nLine-up and maintenance of radio relay systems require sensitive yet rapid \nmeasurements. These are obtained by scanning the systems response as func- \ntions of time, frequency, and amplitude. Parameters thus scanned include the\n\ntransient response to step functions; frequency characteristics of gain, phase, \nimpedance and their frequency derivatives; and amplitude characteristics of \noutput nonlinearity and of intermodulation products.\n\nAluminum Die Castings\u2014The Effect of Process Variables on Their \nProperties. W. Bastncron\u2019 and D. H. Kueprincer!. Proc. A.S.T.M., \n51, pp. 169-197, 1951.\n\nDiffusion in Alloys and the Kirkendall Effect. J. BARDEEN\u2019 and C. \nHERRING!. pp. 261-288 of Imperfections in Nearly Perfect Crystals, Wiley \nN. Y., 1952, 490 p. Edited by W. Shockley, J. H., Hollomon, R. Maurer \nand F. Seitz. Symposium held at Pocono Manor, Oct. 12-14, 1950, by \nCommittee on Solids, National Research Council.\n\nLightning Protection for Fixed Radio Stations. D. W. Bopur\u2019. Tele- \nTech, 11, pp. 58-60, 126+, June, 1952.\n\nCommon grounds, parallel conducting paths, and discharge gaps provide \nthree important means for avoiding equipment damage from high current surges. \nProtection of connecting facilities must also be considered to preserve service.\n\nCompression Tests on Lead Alloys at Extrusion Temperatures. G. M. \nBouton\u2019 and G. 8. Pures\u2019. Proc. A.S.T.M., v. 51, pp. 761-770, 1951.\n\nLoad-deflection measurements made during compression tests on lead and \nlead-alloy cylinders at various temperatures show the effects of alloying in- \ngredients on the force required to produce deformation. The curves also furnish \nclues as to changes taking place in the materials during the course of the test. \nThe load, P, to produce definite small deformation in pure lead at various tem- \nperatures, T, are shown to follow the relationship P = Ae~\u00ae7, where A and B \nare constants for the material. This is the same relationship found by others in \nextrusion studies. The elements added to lead were those most commonly used \nin the manufacture of cable sheath, namely, antimony, arsenic, bismuth, silver, \ntellurium, and tin. The results show that the stronger alloys now used for cable \nsheathing deform less readily at extrusion temperatures than pure lead or the \nweaker alloys.\n\nLiffect of Prior Strain at Low Temperatures on the Properties of Some \nClose-Packed Metals at Room Temperature. W. C. Exuis\u2019 and E. 8. \nGreiner\u2019. J. Metals, 4, pp. 648-651, June, 1952. (Monograph 1966).\n\nThe Fatigue Test as Applied to Lead Cable Sheath. G. R. Goun\u2019 and \nW. C. Exits\u2019. Proc. A.S.T.M., 51, pp. 721-740, 1951.\n\nThis paper discusses the more important factors affecting the design of labora- \ntory test methods suitable for obtaining significant fatigue data from reversed \nbending tests on cantilever-beam specimens of lead cable sheathing alloys. \nData are presented to show the effect of cycling rate, temperature, shape of \nspecimen, alloy additions, and aging on fatigue life. The close correlation be- \ntween bending fatigue tests on strip specimens and full size sections of cable is \ndemonstrated. The fatigue data are analyzed in terms of (1) cycle life versus \ndeflection, (2) cycle life versus strain, and (3) cycle life versus stress. Photo- \nmicrographs illustrating representative laboratory and field failures are included.\n\nThermal Conductivity of Germanium. A. Grizco\u2019 and H. C. Monr- \ngomErRY. Phys. Rev., 86, p. 570, May 15, 1952. .\n\nBell System Cable Sheath Problems and Designs. F. W. Horn\u2019 and R. B. \nRamsry\u2019. Trans. A.J.E.E., 10, Part 2, pp. 1811-1816, 1951. (Monograph \n1917).\n\nPowdered Standards for Spectrochemical Analysis. E. K. Jaycox\u2019. \nApplied Spectroscopy, 6, pp. 17-19, May, 1952. (Monograph 1978).\n\nLingineering for Low Product Cost and High Product Quality at the \nWestern Electric Company. A. C. Jonus\u00ae. Ind. Quality Control, 8, pp. \n53-59, May, 1952.\n\nA successive-approximations method is applied to the selection of network \nfunctions having desired magnitude and phase variation with frequency. The \nfirst approximation, the first set of pole and zero locations, can be selected on \nthe basis of known solutions to similar problems or through use of a set of curves. \nIn succeeding approximations the pole and zero locations are adjusted to decrease \nthe deviation of the earlier approximations from the desired characteristics. The \nprocess adjusts the magnitude and phase characteristics simultaneously. Its \nflexibility permits accommodation of practical constraints not possible with \nother methods.\n\nThe Magnetic Structure of Alnico 5. E. A. Nessrrr\u2019 and R. D. Hewen- \nrEIcH. Elec. Eng., 71, pp. 530-534, June, 1952. (Monograph 1981).\n\nIn the investigation of Alnico 5, two problems arose. What is the mechanism \nwhich enables the alloy to respond to heat treatment in a magnetic field? What \ncauses the alloy to have a high coercive force of 600 oersteds? The first problem \nhas been solved and progress has been made toward solving the second.\n\nSingle-Frequency Signaling System for Supervision and Dialing Over \nLong-Distance Telephone Trunks. N. A. Newett' and A. Weaver\u2019. \nTrans. A.I.E.E., 70, Part 1, pp. 489-494, 1951. (Monograph 1841).\n\nThe single-frequency signaling system for long-distance telephone trunks \nfrees dial calls from the range and other limitations imposed by de signaling \nmethods. It uses alternating currents in the voice range as the signaling medium \nand so can be used with any trunk of any length or type of line facility which \nmeets voice-transmission requirements. The signaling requirements, design \nproblems, main features of the circuit and equipment arrangements, and the \noperation of this system are outlined in this paper. The system described is the \nfirst practical arrangement of its type satisfactorily to meet all the conditions of \ntelephone service in the Bell Telephone System.\n\nExperimental Information on Slip Lines. W. T. Reap, Jr\u2019. pp. 129- \n1p1 of Imperfections in Nearly Perfect Crystals, Wiley, N. Y., 1952, \n490 p. Edited by W. Shockley, J. H. Hollomon, R. Maurer and F. Seitz. \nSymposium held at Pocono Manor, Oct. 12-14, 1950, by Committee \non Solids, National Research Council.\n\nOn the Geometry of Dislocations. W. T. Reap, Jr.\u2019 and W. SHockuey\u2019. \npp. 77-94 of Imperfections in Nearly Perfect Crystals, Wiley, N. Y., \n1952, 490 p. Edited by W. Shockley, J. H. Hollomon, R. Maurer and \n\u00a5. Seitz. Symposium held at Pocono Manor, Oct. 12-14, 1950, by Com- \nmittee on Solids, National Research Council.\n\nA Servo System for Heterodyne Oscillators. T. Suoncznwsxr. Trans. \nA.J.E.E., 70, Part 1, pp. 1070-1072, 1951. (Monograph 1883).\n\nA constant rate of progression of frequency of a motor-driven heterodyne \noscillator is obtained by comparing its output with a frequency standard. The \nresult is fed into a servo loop which drives the motor at the proper speed. When \nused in connection with a level recorder a linear frequency scale is obtained which \nis more accurate than the static calibration of the oscillator.\n\nMetallic Rectifiers in Telephone Power Plants. D. E. Trucxssss\u2019. \nTrans. A.I.E.E., 70, Part 2, pp 1464-1467, 1951. (Monograph 1987).\n\nMetallic rectifiers are a comparatively new means of converting power from \nalternating current to direct current. Most of the component apparatus used \nin the Telephone Systems operates with direct current while the normal power \nsource is alternating current. Therefore a static device without expendable parts \nwhich is obtainable in small and large current capacity lends itself as a means ox \npower conversion in telephone power plants.\n\nA. B. Cuark, B.E.E., University of Michigan, 1911. A. T. & T. Co., \n1911-34; Bell Telephone Laboratories, 1934-. Toll Transmission De- \nvelopment Engineer, 1929; Toll Transmission Development Director, \n1934; Director of Transmission Development, 1935; Director of Systems \nDevelopment, 1940; Vice President, 1944. Bell System Chairman of \nJoint Subcommittee on Development and Research of the Edison Elec- \ntric Institute and Bell System since 1938. Since June, 1951, Mr. Clark \nhas been in charge of coordinating all Bell System programs at the \nLaboratories. During World War II he served both as a consultant to \nand a member of various divisions of the Office of Scientific Research \nand Development. In 1944 he was appointed Consultant to the Secretary \nof War, and in connection with this work made trips to the European \nand Mediterranean theaters of operation. Member of I.R.E., Tau Beta \nPi, Sigma Xi, and A.A.A.S. and Fellow of A.I.E.E. and the Acoustical \nSociety of America.\n\nJ.R. Fry, M.E., Cornell University, 1915. Western Electric Company, \n1915-25. Bell Telephone Laboratories 1925-. Mr. Fry has been Assistant \nSwitching Apparatus Engineer in the Switching Apparatus Development \nDepartment since 1946. Except for the years 1941-45, when he worked \non military projects, most of Mr. Fry\u2019s Bell System service has been \ndevoted to the design and development of electromagnetically operated \nswitching apparatus such as relays, switches, registers, and selectors. \nMember of Eta Kappa Nu.\n\nH. C. Monreomery, A.B., University of Southern California, 1929; \nM.A., Columbia University, 1933. Bell Telephone Laboratories, 1929-. \nPrior to the war, Mr. Montgomery was engaged in studies of hearing \nacuity and the analysis of speech sounds. His recent work in the transis- \ntor physics group has been concerned with fluctuation phenomena in \nsemiconductors.\n\nSAMUEL P. Moraan, Jr., B.S., California Institute of Technology, \n1943; M.S., California Institute of Technology, 1944; Ph.D., California \nInstitute of Technology, 1947. Bell Telephone Laboratories, 1947-. A\n\nment System, 1944-1945; member, Hoover Medal Board of Award, \n1945-1951; and Chairman, Board of Trustees of Volta Memorial Fund, \n1949-. He also has been active in the American Standards Association. \u2014 \nHe was long a member of the Board of Directors, and was Chairman of \nthe Standards Council from 1942-1945 and Vice President 1948-1951. \nSince 1949 he has been President of the U. 8. National Committee of the \nInternational Electrotechnical Commission. He is a member of the Joint \nConference Committee on Standards of the Department of Commerce \nand ASA, and Chairman of the U. 8. N. C. Executive Council Subcom- \nmittee. He is Fellow of the American Institute of Electrical Engineers, \nAcoustical Society of America, American Physical Society, American \nAssociation for the Advancement of Science, and of the Institute of \nRadio Engineers; and is a member of the American Society for Engineer- \ning Education and of Tau Beta Pi.\n\nJ.J. Prturop, E.E. 1908, D.E. (Hon.) 1939, Ohio Northern University ; \nA. T. & T. Co., 1908-. From 1910 until 1943 Mr. Pilliod was associated \nwith the Long Lines Department and the General Engineering Depart- \nment of the American. Telephone and Telegraph Company. From 1914 \nto 1918 he was Division Plant Engineer in Chicago; 1918-1920, Engineer \nof Transmission, New York City; 1920-1941, Engineer in charge of \nLong Lines Engineering Department; and 1941-1948, General Manager \nof the Long Lines Department. In 1943 he assumed his present position \nas Assistant Chief Engineer of the American Telephone and Telegraph \nCompany. From October 1942 to April 1943 he was Chief of Signal \nSection, Production Division, Army Service Force. He is a Fellow of the \nA.I.E.E. and is a Trustee of Ohio Northern University and of Vassar \nCollege.\n\nF. F. Surpuey, B.S. in E.E., Purdue University, 1925. A. T. & T. Co., \n1925-34; Bell Telephone Laboratories, 1934\u2014. Since 1948, Mr. Shipley \nhas been switching engineer in charge of planning large automatic \nswitching systems, both local and toll. This includes panel, crossbar, \nand large step-by-step epee Member of the A.I.E.E., Tau Beta Pi, \nand Eta Kappa Nu.\n\nH. T. Wituetm, B.S. in E.E., Cooper Union, 1927; E.E., Cooper \nUnion, 1936. Western Electric Company, 1922-24; Bell Telephone Lab- \noratories, 1925-. Since joining the Laboratories Mr. Wilhelm\u2019s work has \nbeen with the Transmission Apparatus Development Department, where \nhe has designed electrical measurement apparatus and developed test \nmethods. Member of A.I.E.E. and Tau Beta Pi.", "title": "magazine :: Bell System Technical Journal :: BSTJ V31N05 195209", "trim_reasons": [], "year": 1952}
{"archive_ref": "bitsavers_BellSystemJV46N05196705_8738393", "canonical_url": "https://archive.org/details/bitsavers_BellSystemJV46N05196705_8738393", "char_count": 287730, "collection": "archive-org-bell-labs", "doc_id": 555, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc555", "record_count": 572, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bitsavers_BellSystemJV46N05196705_8738393", "split": "test", "text": "Combinatorial Solution to the Problem of Optimal Routing in \nProgressive \u201ctradings V. E. BENES 865\n\nIntegral Equation for Simultaneous Diagonalization of Two \nCovariance Kernels T.T. KADOTA 883\n\nPrinciples of Design of Magnetic Devices for Attitude Control \nof Satellites M.S. GLASS 893\n\nA High-Capacity Digital Light Deflector Using Wollaston \nPrisms W. J. TABOR 957\n\nTransistor Distortion Analysis Using Volterra Series Repre- \nsentation S. NARAYANAN 991\n\nB.S.T.J. Briefs: Realizability Conditions for the Impedance \nFunction of the Lossless Tapered Transmission Line\u2014A Critique \nE. N. PROTONOTARIOS 1047\n\nTHE BELL SYSTEM TECHNICAL JOURNAL is published ten times a \nyear by the American Telephone and Telegraph Company, B. 8. Gilmer, \nPresident, C. &. Wampler, Vice President and Secretary, J. J. Scanlon, Vice \nPresident and Treasurer. Checks for subscriptions should be made payable \nto American Telephone and Telegraph Company and should be addressed \nto the Treasury Department, Room 2312C, 195 Broadway, New York, \nN. Y. 10007. Subscriptions $5.00 per year; single copies $1.25 each. Foreign \npostage $1.08 per year; 18 cents per copy. Printed in U.S.A.\n\nA simple analysis shows that the unit-cube conductance is a figure of \nmerit in semiconductor device design theory. The unit-cube conductance, \nG, ws gwen by 2Kv, where K is the permittivity of the semiconductor and \nvq 18 the limiting drift velocity.\n\nThe space-charge resistance, h,, , due to carrier generated under ava- \nlanche condition is derived for p-n junctions. It 1s found that for parallel- \nplane structure, R,. = 1/GN, where N is the number of unit cubes in the \ndepletion region with cube edge equal to the depletion width or N = A/W* \nwhere W is the depletion width and A the junction area. The disturbance \nin voltage caused by the space-charge effect is given by I/GN = JW\u2019*/G \nwhere I and J are the current and current density, respectively. Similar \nresults are obtained for p-n junctions with coaxial-cylinder and concentric- \nsphere structures.\n\nFor silicon, the value of G is approximately 40 umhos. The transconduct- \nance of a silicon surface-controlled avalanche transistor in terms of the \nunit-cube expression 1s about 12.5 N wmhos.\n\nA simple analysis of \u2018\u201c\u2018avalanche resistance\u2019 can be given for the \nlimiting case in which carriers are generated at one boundary surface \nof the depletion region of a p-n junction and travel across the depletion \nregion with a limiting drift velocity v,. Structures satisfying these \nconditions can be of the n\u00b0 pp\u201d form. It will be shown that the quantity\n\n2Kv,z (where K is the permittivity of the semiconductor) is a figure of \nmerit in semiconductor device design theory which limits the perform- \nance of space-charge-limited devices. This quantity is a combination \nof \u2018material constants\u201d similar to F',v, (where F\u2019; is \u2018\u2018breakdown field\u2019\u2019) \nwhich limits the frequency-power performance of transistors\u2019\u2019\u2019 and \nK/c (o is the conductivity) which limits the gain-bandwidth product\u00ae \nof solid-state devices. |\n\nTor a structure in which the space-charge layer is bounded by parallel \nplanes of area A and spacing W, it will be shown that the effective space- \ncharge resistance can be interpreted as due to N unit-cube conductances \nin parallel, where the unit-cube conductance G is given by\n\nand N is number of unit cubes in the depletion layer with cube edge \nequal to the depletion width, or\n\nFor coaxial-cylinder and concentric-sphere structures, similar results \nare obtained for the R,,. The number of unit cubes (or curvilinear \ncubes), however, depends on the radius of the surface upon which ava- \nlanche occurs and the length of the cylinder (for the coaxial-cylinder \nstructure). These functional dependences are derived below.\n\nAn interesting application of the space-charge resistance and the \nunit-cube expression is given for a surface-controlled avalanche tran- \nsistor (SCAT).\u00b0\n\nAs represented in Fig. 1(a) the depletion layer of an n*pp\u201d structure \nextends through the p layer with a doping of NV,, and is bounded by \nthe planes at x = 0 and x = W. When the applied voltage V is equal \nto the breakdown voltage V, the electric field E(x) has its maximum \nabsolute value fF\u2019, at \u00ab = 0 and decreases to Fz \u2014 (qN.W/K) atx = W. \nThis insures breakdown at x = 0. Furthermore, if gV,W/K < 0.9 Fz, \nthen the field is everywhere = F',,/10, so that holes have their limiting \ndrift velocity vz all across W.\n\nFig. 1\u2014 p-n junction geometry of (a) parallel plane, (b) coaxial cylinder, \nand (ce) concentric sphere structures.\n\nwhere p is the carrier-charge density and A the area. Since &# at z = 0 \nis assumed to be equal to F,, the disturbance A(x) in the electric \nfield due to p is\n\nso that the disturbance in voltage caused by the carriers (1.e., the average \nfield times W) is obtained by integrating A(z)\n\nConsider first that the maximum field occurs at the mner surface. \nAs shown in Fig. 1(b) the depletion layer extends through the intrinsic \nregion of an nip\u201d coaxial-cylinder structure and is bounded by the \ncylinders of radii r = a and r = b. When V = V,, the electric field \nE(r) has its maximum absolute value F', at r = a, and decreases to \nFza/b atr = b.\n\nso that p varies as 1/r. Integrating Poisson\u2019s equation leads to a dis- \nturbance AK(r) in the electric field and AV, in the voltage due to p \ngiven by\n\nA, is the area on the outer cylinder surface that corresponds to one \nunit-cube conductance G.\n\nEquation (10) may be interpreted as the resistance of N unit curvilinear \ncubes in parallel. These cubes are formed by intersection of equipotential \nsurfaces with the orthogonal family of electric field lines. Each cube has \na conductance (2Kv,), and the number of cubes N is given by (11). \nThe area A, approaches (b \u2014 a)\u201d when a \u2014 b, and approaches 2(b \u2014 a)\u2019 \nwhen a \u2014 0, and consequently remains finite even as the inner cylinder \napproaches a line.\n\nThe maximum field may be caused to occur on the outer surface \nr = b by adjusting the chemical charges in the depletion layer appro- \npriately, such as a p pn\u201d structure with the pn\u201d junction at r = 8, \nthe p*p boundary at r = a. In this case, the area A, approaches (b \u2014 a)\u2019 \nwhen a \u2014 b, and approaches 20\u00b0 In (b/a) when a \u2014 0. Hence, the space- \ncharge resistance has the same value as given by (10) and (11) when \na \u2014 b, but approaches infinity as a \u2014 0.\n\nAs shown in Fig. 1(c), the depletion layer extends through the in- \ntrinsic region of an n\u201cip~ concentric sphere structure and is bounded by \nthe spheres of radii r = a andr = b. When V = Vz, the electric field \nE(r) has its maximum value F\u2019, at the inner surface r = a, and decreases \nto Fsa\u2019/b\u2019 at r = b.\n\nand A, is quantity of area on the outer sphere surface that corresponds \nto one curvilinear unit-cube conductance G. When a is finite or a \u2014 Bb, \n(15) may be interpreted as the resistance of N unit curvilinear cubes \nin parallel, where each cube has a conductance (2Kv,) and the number \nof cubes N is given by (16). Unlike the n*ip\u201d* coaxial-cylinder case, \nR,, of the concentric-sphere structure approaches infinite resistance as \nthe inner sphere approaches a point.\n\nFs can occur on the outer sphere for a p*pn\u201d structure, for example. \nThe results of A, are the same for both limiting cases as given in (17).\n\nAn interesting application of the unit-cube expression is that for a \nsurface-controlled avalanche transistor (SCAT)* with a total junction \nperimeter of P and a space-charge layer W. Because the space-charge \nresistance is finite for an n\u201cip* coaxial cylinder structure as the inner \ncylinder approaches a line [see (12)], it is feasible to make calculations \nfor SCAT on the basis of an avalanche line source. There are N = P/W \nsuch unit cubes around the edge, and the total transconductance, g,, ,\n\nTABLE I\u2014 Unit-CusE Expressions (W = b \u2014 a) \nStructures Parallel Coaxial Concentric \nplane cylinders spheres \nN A a\u2014b a\u20140 a\u2014>b \n(No. of unit cubes) Ws or Lb Lh And? \nW2 b Ww? \nCube edge W W Vb W \n(cm) \nR \n8c 1 1 \nconn) GN ~ (2Kva)N \nor \nGm = (Lex) = : : (19) \n0 TR oe\n\nFor a silicon SCAT with a device geometry of P = 1000 \u00bb and W = \n0.5 uw, there are 2000 unit cubes with cube edge 0.5 nw. The space-charge \nresistance is 12.5 ohms, and the transconductance is 25,400 wmhos.\n\nAnother application is to calculate the voltage disturbance AV, in \na Read diode.\u2019 For a silicon Read diode with drift region of 10 um and \nan operating current density of 1000 amp/cm\u2019, the value of AV;, as \nobtained from (6), is approximately 25 volts.\n\nA summary of the number of unit cubes and other pertinent quantities \nis presented in Table I. It has been shown that the unit-cube conductance \n(2Kv,) is a figure of merit in semiconductor device design theory. The \nunit-cube expressions are shown to be useful for calculation of the space- \ncharge resistance.\n\n1. Johnson, E. O., Physical Limitation on Frequency and Power Parameters of \nTransistors, IEEE International Convention Record, Part 5, March, 1965, \npp. 27-34. Also appearing in RCA Review, June, 1965, pp. 163-177.\n\n2. De Loach, B. C., Recent Advances in Solid State Microwave Generators, a \nchapter of Advances in Microwaves, to be published by Academic Press \nand edited by Leo Young.\n\n3. Rose, A., An Analysis of the Gain-Bandwidth Limitations of Solid State \nTriodes, RCA Review, December, 1963, pp. 627-640.\n\n4. Shockley, W. and Hooper, W. W., The Surface Controlled Avalanche Transis- \ntors, Wescon Meeting, Los Angeles, August, 1964.\n\n5. Read, W. T., A Proposed High-Frequency, Negative-Resistance Diode, \nBS.T.J., 37, March, 1958, pp. 401.\n\nConsideration of large alphabet digital communication systems ts of both \ntheoretical and practical interest. Although performance bounds on optimum \nsystems for the Gaussian channel are available, constructive methods for \napproaching these bounds are unknown, except in a few very special cases. \nSpecific systems have been proposed and evaluated relative to these bounds, \nbut exact evaluation of error probability is generally a difficult numerical \ntask. It 1s of interest to consider simpler performance criteria which permit: \ncomparison of various systems without extensive computation.\n\nAn easily evaluated criterion (based on the alphabet size and minimum \ndistance between signal vectors) 1s shown to yield a simple sufficient condi- \ntion for one system to be better than another (smaller error probability \nfor the same energy-per-bit). The criterion is applied to orthogonal, bi- \northogonal, simplex, and more general permutation modulation systems. \nIn addition to comparing the various systems, we consider ways of obtain- \ning good special cases of permutation modulation. Finally, we assess a \nrecently proposed system (\u2018\u2018N-orthogonal phase modulation\u2019\u2019) and show \nthat zt 1s generally inferior to more conventional techniques.\n\nThe choice of waveforms for communicating over the Gaussian \nadditive noise channel is a classic problem in communication theory. \nOrthogonal modulation systems (i.e., digital communications in which \nthe alphabet consists of orthogonal waveforms) are known to result \nin good power efficiency at the expense of poor bandwidth utilization.\u2019 \nAs the alphabet size M is increased, the energy-per-bit # required to \nachieve a given error probability P, diminishes, but the information \nrate to bandwidth ratio (#/W) diminishes even more rapidly. Bi- \northogonal and simplex modulation afford somewhat improved per- \nformance, but are likewise restricted to low values of R/W.\n\nThere is considerable interest in finding large alphabet systems \nwhich have both good power efficiency and good bandwidth utilization.\n\nSlepian\u00ae has given bounds on what can be achieved, but constructive \ntechniques for approaching these bounds are generally unknown.\n\nAlthough computer evaluation is ultimately required for precise \nknowledge of error probability, it is of interest to consider simpler \nperformance criteria which permit at least a qualitative comparison \nof various systems without extensive computation. It is the purpose \nof this paper to demonstrate the utility of the latter approach.\n\nAfter defining the problem more precisely in Section IJ, some well- \nknown bounds on the error probability are employed in Section III \nto obtain a simple analytic criterion for comparing systems in the limit \nof low P,. In Section IV this criterion is applied to systems (PSK, \nFSK, biorthogonal, and simplex) for which extensive exact computa- \ntions are available and for which the conclusions drawn are already \nwell-known. After these illustrative examples, permutation modula- \ntion* is considered in Section V and N-orthogonal phase modulation\u2019\u2019\u00ae \nin Section VI. It is shown that the former can yield better performance \nthan conventional techniques, but that the latter is generally inferior. \nIinally, in Section VII limits on our performance criterion, obtained \nfrom sphere-packing arguments, are presented.\n\nWe consider an M/-ary modulation system of equienergy waveforms \nS,(t),2 = 1, --- , M, on (0,7), having the correlation matrix\n\nE, is the energy of each waveform so that p;; = 1, and \u20141 S p,, $ 1. \nIt is conventional\u00ae\u2019\u2019 to define a normalized information rate, \n(2 log. M7)/n, where n S M is the rank of the correlation matrix (di- \nmensionality of the signal space). We choose to call this normalized \nrate the \u2018information to bandwidth ratio, R/W\u201d\u2019 motivated by the\n\nwhere the second equality follows if we set n = 27W, which is at \nleast partially justified for large n by the work of Pollak and Landau.* \n_ For our purposes, the right-hand side of (2) may be considered as the \ndefinition of R/W.\n\ncorrelation matrix can be written as a permutation of the first row. \nConsidering the waveforms as vectors in an n-dimensional linear vector \nspace, this means that each waveform sees an identical environment \nof neighboring waveforms. This restriction is a desirable one if it is \ndesired to transmit each waveform with equal a prior? probability. \nThe restriction is satisfied by the various modulation systems mentioned \nin the introduction.* Slepian\u2019 has termed such systems \u2018\u2018group codes \nfor the Gaussian channel.\u201d\u2019\n\nwhere n(\u00a2) is a sample function from a white Gaussian noise process \nof spectral density N, ; 1.e.,\n\nOn the basis of this observation we wish to decide with minimum prob- \nability of error (P,) which of the 7 waveforms was transmitted. The \noptimum (minimum P,) receiver is known\u2019\u2019 to consist of 14 matched \nfilters which give\n\nand decision that the kth waveform was transmitted is made if 2, > 2; \nfor allj # k;i.e., the decision is made on the basis of the largest matched \nfilter output.\n\n* The only commonly employed M-ary system (known to the author) which does \nnot satisfy this restriction is M-level amplitude modulation.\n\n846 THE BELL SYSTEM TECHNICAL JOURNAL, MAY-JUNE 1967 \nThe error probability of this system is given by* \n-1-| ef dey ++ dew plas, -** 24), (8)\n\nwhere p(a,, \u00b0*- , Yar) is the multi-variate zero-mean Gaussian dis- \ntribution with covariance given by (7), and the region of integration \nQ,; 1s defined by the condition\n\nLandau and Slepian\u2019* have proved the long-conjectured result that \nP, is minimized for a given M (but n unrestricted) by the simplex \nconfiguration in which the correlation matrix has the formy\n\nFor this case the expression for P, may be reduced to a single integral\u2019\u00ae \nand numerical results are readily obtained.\u201d\n\nWeber\u2019* has derived locally optimum configurations when M/2 < \nn <= M \u2014 1. Forn = M/2 a local optimum is the biorthogonal con- \nfiguration in which the signal vectors are located along the coordinate \naxes (+ and \u2014) of the n-dimensional vector space such that\n\n* Actually this is the error probability assuming the 7th signal is transmitted. \nHowever, under the assumption of equal a priori transmission of all signals, and the \npermutation property assumed for the correlation matrix, this probability is inde- \npendent of z and is equal to the system probability of error.\n\n+ The \u201clocal optimality\u201d of the simplex configuration (viz., that P, has a local \nminimum) had been proved previously by Balakrishnan.\u201d\n\nIn this case P, may also be expressed as a single integral which is \nreadily evaluated by machine techniques. Although for a given value \nof M, biorthogonal modulation requires slightly more energy-per-bit \nto achieve a given P, than simplex,* it is noted that (for large M) \nR/W for biorthogonal is essentially twice that of simplex. Further- \nmore, for biorthogonal half of the waveforms are the negatives of the \nremaining half; consequently, M/2 rather than M matched filters are \nrequired. For these reasons biorthogonal is generally preferred to sim- \nplex, and indeed has been employed for deep-space communications.\u201d\n\nThe disadvantage of both simplex and biothogonal modulation is \nthat good power efficiency is associated with large values of M (as it \nmust be for any modulation system) which from (10) and (12) imply \nsmall values of R/W. Weber\u2019s** results indicate locally optimum sys- \ntems with R/W between simplex and biorthogonal, but these are then \nalso restricted to relatively small R/W.\n\nOptimum systems (in the sense of minimum P,) are not known for \nn < M/2. However, bounds on the error probability of optimum sys- \ntems have been obtained\u2019 and evaluated.* (The upper bound is ob- \ntained by random coding arguments, and the lower bound by sphere \npacking arguments.) These bounds are extremely useful in assessing \nthe performance of specific systems; however, to do so involves ex- \nplicit evaluation of P, for the specific systems of interest. This is at \nbest a difficult numerical task. Furthermore, we may find in comparing \ntwo systems that one is better if we are interested in P, } 10%, \nwhereas the reverse is true when P, = 10-\u00b0. Also, in comparing sys- \ntems with different values of M it may be unrealistic to compare P.,, \nsince P, is the word error probability, and the systems contain a dif- \nferent number of bits per word. Comparison on the basis of bit error \nprobability involves a difficult conversion from word to bit error \nprobability which involve coding arguments separate from the modula- \ntion system performance.** For all of these reasons it 1s desirable to \nfind a simpler criterion than P, which permits at least a gross com- \nparison of modulation systems. :\n\nrequires less energy per bit. The simple unqualified statement that simplex is \nthe optimum modulation system is misleading.\n\nOne approach to comparing modulation systems is to obtain lower \nand upper bounds on the true error probability*\n\nIf two systems are close in performance, the above procedure may \nnot enable us to determine which is better unless the bounds are close. \nOn the other hand, close bounds may be difficult to evaluate and may \nnot lead to a simple performance criterion. We adopt the viewpoint \nhere that it is desirable to have bounds, which although quite loose, \nlead to a simple sufficient condition for determining when one system \nis better than another.\n\nThat is, p is the largest non-diagonal entry of the correlation matrix. \nIt is readily established thati\n\nThe lower bound is obtained by observing that P, for an M-ary system \ncan be no less than that of the binary system containing nearest \nneighbor waveforms. The upper bound follows from\n\nwhere the first inequality in (17) 1s a consequence of the symmetry \nproperty of the system and the fact that the probability of a union of \nevents is less than the sum of the probabilities of the events. The\n\n* We consider here bounds on word error probability, which, however, may be \neasily converted to bounds on bit error probability. For example, if a word is in \nerror at least one bit is in error, and at most all bits are in error. Hence, \nP./log:M and P. are lower and upper bounds on the bit error probability.\n\n\u2018~ These bounds are generally well known and appear widely in the literature; \neg., Refs. 7, 17, 18, 19. Also, as noted in the previous footnote, these bounds on \nword error probability are readily converted into bounds on bit error probability.\n\nsecond inequality in (17) follows simply by observing that a sum of \n(MZ \u2014 1) terms is no greater than (M \u2014 1) times the largest term.\n\nIn comparing modulation systems with different alphabet size it is \nmore appropriate to consider the energy per bit # rather than the \nsignal energy E,, where\n\nIndeed, the parameter #/N, is an appropriate measure of the power \nefficiency of a modulation system. The Shannon channel capacity \nformula requires that E/N, > log.2 to achieve arbitrarily small P., \nconventional systems generally require values of E/N, at least 4 times \nthe Shannon minimum.? |\n\nIn terms of the parameter L/N, the error probability bounds may \nbe rewritten\n\n(2) In the limit of large H/N,, Ay > Ke is sufficient to ensure that \nsystem 1 is better than system 2. (If K, > Ke we will say that sys- \ntem 1 is \u201casymptotically better\u201d than system 2.)\n\n(22) If system I is asymptotically better than system 2, then there \nexists a valuc of H/N, above which system 1 is better than system 2. \nBelow this value of H/N, our formalism is generally inadequate to \ndetermine which system is better. (The critical value of E/N, may \nbe obtained by replacing the inequality in (21) by an equality.)\n\n(22) A binary system that is asymptotically better than an M-ary \nsystem is always better than the M-ary system.\n\nThus, we can always determine quite simply which of two systems is \nasymptotically better, and may, in many special cases, be able to make \ncomparisons at specific H/N, of interest.\n\nof the P, obtained with the two systems when operated at the same \naverage power and information rate. To complete the comparison, the \nbandwidth requirements of the two systems should also be considered. \nThus, the parameter R/W, as well as K should be used in comparing \nsystems.\n\nIn the following sections of this paper specific systems will be con- \nsidered and represented by points on a K, R/W plot. This will enable \nan immediate comparison of the asymptotic performance of systems \nhaving the same R/W. It should be noted that Gilbert*\u201d used a simi- \nlar plot in his 1952 paper which addressed the same subject con- \nsidered here. Gilbert employed a (SNR, R/W) plot in which the effec- \ntive signal-to-noise ratio (SNR) was obtained for a given P, by using \nthe upper bound in (19). Since the SNR is related to our 1/K, better \nsystems correspond to smaller SNR. Our purpose in writing this paper \nis not to argue that our plot is a better way to present the results \nthan Gilbert\u2019s. (Indeed, since in general, P, is much closer to the upper \nthan to the lower bound, his method of comparison is somewhat bet- \nter, although somewhat less convenient to use.) Our purpose rather is \nto resurrect these old methods which have been largely discarded \nsince the advent of high-speed computation, and to illustrate their \napplicability to recently proposed modulation systems.\n\nNote that K = 2 for both M = 2 and M = 4* and falls off thereafter. \nSince the dimensionality of the signal space is n = 1 for M = 2 and \nn= 2 for M > 2, it follows that h/W is given by\n\nR 2 for M = 2, \nvo : (23) \nlog, M for M > 2. \n* K is maximized (for integer M) when M = 3. In practice, it is generally\n\ndesirable to consider only those values of M which are integer powers of 2 (ie., \neach symbol conveys an integer number of bits). We shall restrict our numerical \nexamples to such cases.\n\nTable I lists the K and R/W values for phase-shift modulation, and \nthese are denoted by dots in Fig. 1. It is apparent that M' = 2 and \nM = 4 are asymptotically better than the higher-order systems, and \nfrom our previous results this implies that the binary system is always \nbetter than the general M-ary case with M > 4.* Recall that we are\n\nFig. 1\u2014 (K,R/IW) plot for phase-shift, orthogonal, biorthogonal, and simplex \nmodulations.\n\nconsistently using the term \u201cbetter\u201d to mean smaller P, for a given \nE/N,. Large alphabet phase modulation may still be desirable because \nof the larger R/W.\n\n4.2 Frequency Shift (Orthogonal) Modulation \nFor M orthogonal signals (e.g., frequency-shifted signals with es- \nsentially non-overlapping spectra), p = 0 and \nK = log. M. (24) \nThe dimensionality of the signal space is the number of orthogonal \nvectors, nm = M, so that\n\nTable II lists the K and R/W values for orthogonal modulation, and \nthese are denoted by circles in Fig. 1. Larger values of 1/7 correspond to\n\nsystems which are asymptotically better, at the expense, however, of \nsmaller values of R/W. It is clear that binary orthogonal is inferior \nto binary and quarternary PSK both in terms of a smaller K and \nsmaller h/W.7\n\nA biorthogonal system consists of 11/2 orthogonal waveforms and \ntheir negatives. The maximum correlation coefficient 1s p = 0 for M 2 \n4, but p = \u20141 for M = 2. Therefore,\n\n* This conclusion is confirmed by the exact calculations of P. for M-ary PSK \nby C. R. Cahn.?\u00b0\n\n+ Binary FSK may still be employed, of course, for simplicity reasons or \nbecause the channel phase coherence may not be consistent with phase-shift \nmodulation.\n\nVv mM | on \nTable III lists the K and R/W values for biorthogonal modulation, \nand these are denoted by []\u2019s in Fig. 1. Note that M = 2 and M = 4 \nbiorthogonal are equivalent, respectively, to binary and quarternary \nPSK.\n\nClearly, for fixed R/W, biorthogonal is asymptotically better than \northogonal. For example, consider M = 4 orthogonal and M = 16 \nbiorthogonal, both of which have R/W = 1. From (21), the biortho- \ngonal system is better than the orthogonal system for all E/N, > 2.8, \nwhich corresponds to all P, of practical interest. (P, < 3(10)-*).\n\nIn simplex modulation, the M code vectors form a regular simplex \nin M \u2014 1 dimensions. (All vectors are equally spaced from all other \nvectors. This corresponds to an equilateral triangle in two dimensions, \nand a regular tetrahedron in three dimensions.) All correlation coef- \nficients are equal and are given by?%??3 \u00bb = \u20141/(M \u2014 1). Therefore,\n\nWwW M\u20141 29) \nComparison of (28), (29) with (24), (25) indicates that for large M \nsimplex modulation is essentially identical to orthogonal modulation. \nTable IV lists the K and R/W values for simplex modulation, and \nthese are denoted by /\\\u2019s in Fig. 1. A quick glance at Fig. 1 indicates\n\nthat depending on the R/W of interest, biorthogonal or PSK modula- \ntion offers the best asymptotic performance of the systems considered \nso far. (The dashed line in Fig. 1 is drawn through these \u201cbest\u201d \npoints.) Note that although simplex provides the largest K for a fixed \nvalue of M, it does not do so for fixed R/W.*\n\nSlepian* has recently described an exceedingly general modulation \nsystem (permutation modulation) for which all of the systems con- \nsidered in the previous section are special cases. The optimum de- \nmodulation algorithm is particularly simple, but the actual evaluation \nof P,, and the finding of good special cases is somewhat more complex. \nWe restrict ourselves here to a special subclass of permutation modula- \ntion. This subclass is suggested both as the simplest generalization of \nbiorthogonal systems, and because perusal of Slepian\u2019s results indicate \nthat systems taken from this subclass are amongst the better of the \nmoderate-sized alphabet examples which he considers.\n\nFollowing Slepian we define an (n,m) permutation modulation sys- \ntem as follows. The time interval T is divided into n subintervals (n = \n2TW). The first waveform of the alphabet consists of a signal with \namplitude unity in the first m subintervals (m < n), and zero ampli- \ntude in the remaining subintervals. The remainder of the waveforms \nconsist of all possible permutations of the subintervals, allowing also \nall combinations of plus and minus amplitudes. For example, the \n(3,2) system contains twelve waveforms which we may represent as\n\nIt is also noted that the special case (n,1) corresponds to biorthogonal \nmodulation.*\n\nThis (n,m) modulation clearly satisfies the symmetry requirements \nof our theory. All members of the alphabet have equal energy? and the \ncorrelation matrix has the desired permutation property. It is readily \nseen that the maximum correlation coefficient is given by\n\nm\u2014 1 \npre (31) \nThus, \nK = (log, M)(1 \u2014 p) (32) \n=[+ Z logs te ) \nm m \nso that (n,m) modulation always achieves K > 1. Also \nfk _ 2 log, M \nW n (33) \nSeo \nn\n\nSince m = 1 leads to biorthogonal modulation which has many de- \nsirable properties, it is natural to look next at the special case m = 2. \nFrom (32) and (33) it follows that for (n,2),\n\n+ With the normalization employed above, the signal energy is m. However, \nall code words may be multiplied by a constant to achieve any desired \u00a3E,.\n\nvalues of the K and R/W are given in Table V and are plotted as M\u2019s \nin Fig. 2. (For reference, Fig. 2 also contains the biorthogonal and \nPSK results from Fig. 1.) Thus, similar to biorthogonal, as n becomes \nlarge KC increases but R/W decreases. It is seen from Fig. 2 that (n,2) \nmodulation gives better performance (larger K for a given R/W) than \nbiorthogonal or PSK.*\n\n(2,1) corresponds to M = 4 biorthogonal, which from our earlier \nresults gives K = 2, R/W = 2. It is seen from Table V that (4,2) \ngives K = 2.30, R/W = 2.30 which corresponds to both better \nasymptotic performance and better bandwidth utilization. It is ap-\n\nparent from (33) that whenever n = 2m, R/W = K, and an im- \nmediate question 1s how large can we make these two quantities. \nWith n = 2m, it follows from (32) that\n\n2m\\ 1 re \nso that for large m, K \u2014 3. It is easily shown that K increases mono- \ntonically towards this asymptotic value as m is increased. Thus, \n(2m, m) modulation does not permit attainment of arbitrarily large \nvalues of K, and hence cannot attain arbitrarily low P, with finite \nH/N,. This is consistent with Slepian\u2019s statement* that permutation \nmodulation cannot approach channel capacity arbitrarily closely at \nnon-zero h/W.\n\n* This is, of course, achieved only at the expense of a larger alphabet size. It \nmay also be noted from Table V that the alphabet size is not generally, a power \nof 2 which may also be a practical disadvantage.\n\nAs an immediate generalization of the above, consider the more \ngeneral case n = km where k > 1.* Then, from (83)\n\nFor large k, the right-hand-side of (41) increases as log k; however, \nas seen from (88) R/W diminishes as k-. The locus of K, R/W values \nobtained with different values of k (but large m so that the approxima- \ntion (41) applies) is shown by the solid curve in Fig. 2. In the limit \nas k > 1 (but m always sufficiently large such that (kK \u2014 1) m > 1), \nK > 1 and R/W \u2014 2. As k increases, both R/W and K increase until \nk =~ 1.5 at which point R/W ~% 3.2 and K = 2.8. Further increases in \nk result in a reduction in R/W but continued increase in K.\n\nThe above results indicate that (n,m) codes can be found with \nF/W as large as octary PSK (R/W = 8) and with a better \nasymptotic performance.\n\nIn the previous examples we have compared by approximate meth- \nods modulation systems which have already been analyzed exactly. \nAlthough perhaps additional insight into the relative performance of \nthese systems has been obtained, many of our conclusions may be \ninferred from existing exact calculations. We now wish to consider a \nnew system, recently proposed by Reed and Scholtz,\u00b0:\u00ae which (to our \nknowledge) has not yet been evaluated numerically.\n\nConsider an alphabet M divided into M; groups, each group con- \ntaining M, members. Thus,\n\nThe different groups may be considered to be sufficiently separated in \nfrequency so that waveforms from different groups are orthogonal. \nWithin a group the waveforms have the correlation properties as- \nsociated with phase-shift modulation. Thus for M, 2 4 the maximum \ncorrelation coefficient is p = cos (27/M,), and\n\nM, M, \nIn the special case of M; = 1 it is apparent that this system reduces to \nsimple phase-shift modulation (Section 4.1). In the special case of \nM, = 4 it reduces to the biorthogonal case (Section 4.38). A question \nof interest then is whether choices of M, > 4, M; > 1 lead to better \nperformance than either phase-shift or biorthogonal modulation.*\n\nshown for the combined phase orthogonal modulation. The solid curves \nare for constant values of M; (noted on the curve); the uppermost \npoint on each such curve corresponds to M, = 4, and each lower point \ncorresponds to M, increased by a factor of two. The dashed curve \ngoes through the M, = 4 points (biorthogonal). It is apparent from \nthis figure that in this class of systems, for R/W = 2, the M, = 4 \nbiorthogonal systems give the largest value of K. For R/W = 2, the \nM; = 1 phase-shift systems give the largest value of K. Thus, in terms\n\ngenerating waveforms with the above correlation properties, rather than in a \ncomparative evaluation of performance.\n\nof asymptotic performance, choice of M, > 4, M; > 1 always gives \npoorer performance than systems which achieve the same R/W with \neither M, = 4 (biorthogonal) or M,; = 1 (simple phase shift).\n\nFor example, consider M; = 2, M, = 8. This yields R/W = 2 and \nK = 1.17. However, R/W = 2 is also achieved with M; = 1, M, = 4 \n(quaternary PSK), and for this case K = 2. From (21) we can con- \nclude that the latter system is better than the former for all #/N, > \n2.5, which includes all P, of interest. The significance of these results \nis that we can make this comparison with only a simple slide-rule \ncalculation.\n\nIn the above comparison we considered only M, = 4. The case of \nM, = 1, M; > 1 is the orthogonal modulation previously considered. \nThe case of M, = 2, M; > 1 gives the same performance as biorthog- \nonal but achieves only 4 the R/W and consequently is of little in- \nterest. The case of M, = 3, M; > 1 consists of orthogonal combina- \ntions of two-dimensional simplexes (equilateral triangles). Reed and \nScholtz\u00ae conjecture that for M = 3M,;, the three-phase orthogonal \nsystem gives a smaller P, than any other collection of 3M; signal func- \ntions in a space of dimensionality 247;. Although this conjecture may \nwell be true, we wish to point out that if the comparison is made on \nthe basis of fixed R/W (rather than fixed M) then biorthogonal is \nasymptotically better than three-phase orthogonal. One way of seeing \nthis is by noting that three-phase orthogonal has the same K but \nsmaller R/W than the four-phase (biorthogonal) system of the same \ndimensionality. To increase the R/W of the three-phase system re- \nquires a reduction in K which makes it asymptotically poorer than \nthe corresponding biorthogonal system.\n\nIt has been shown that the (K,R/W) plot provides a useful tech- \nnique for comparing the performance of various modulation systems. \nAlthough our main concern here is in the comparison of specific sys- \ntems, it is still natural to ask whether there are bounds on what may \nbe achieved in the (K,R/W) plane.\n\nthat if no constraint is placed on alphabet size or signal space \ndimensionality, K can, in principle, be made arbitrarily large for any\n\nR/W. This corresponds to the fact that the Shannon channel capacity \nformula implies that arbitrarily small P, may be achieved at all \n(finite) #/W with finite H/N>.\n\nIf M is held fixed but n is unconstrained, then the maximum J is \nachieved by the simplex modulation\u2019? (Section 4.4) for which case \nKk = [M/(M \u2014 1)] logeM and R/W = (2 loge\u2019T)/M \u2014 1.\n\nPerhaps of more practical interest is the opposite case where the \nsignal space dimensionality n is fixed, but M is unconstrained. Here, \nsphere-packing arguments may be used to show that?\n\nM$ Tonal s+, 4), (47) \nwhere I,,(p,q) is the incomplete beta-function which is extensively \ntabulated.\u201d Thus, for a given p-and n, an upper bound to M may be \ncalculated from (47). Since I,(p,q) is monotonic increasing in 2, this \nalso gives a lower bound on p for fixed M and n. Considered in this \nlatter context we can then determine an upper bound on K with which \nis associated a given value of R/W = (2/n) logsM. This upper bound, \nK,, is plotted in Fig. 4 as a function of R/W for n = 5 and n = 10. \nBoth curves indicate that K, achieves a maximum value. This is un- \nderstandable since for large R/W, 1 \u2014 p decreases more rapidly than \nlog,M increases. On the other hand, as R/W decreases, logeM keeps \ndecreasing, whereas 1 \u2014 p is of course always less than 2. Thus, it is \nnot surprising that there exists an R/W at which K, is a maximum.\n\nIt should be noted, however, that K, is an upper bound which likely \ncannot be achieved. For example, when R/W = (2/n) loge2n, cor- \nresponding to M = 2n, the optimum configuration is widely con- \njectured to be the biorthogonal case.2* The corresponding K and R/W \nvalues for biorthogonal with n = 10 and n = 5 are shown by the \npoints marked (10,1) and (5,1) on the dashed curves of Fig. 4. These \npoints lie well below the upper bounds.\n\nBiorthogonal is a special case (m = 1) of the (n,m) permutation \nmodulation considered in Section V. Fig. 4 (dashed curves) shows the \nK and &/W values for the (10,m) and (5,m) cases. As must be, \nthese curves lie below the upper bounds given by the solid curves.\n\nFinally, we note from Fig. 4 that (n,m) permutation modulation \npossesses the interesting feature that as m is increased (for a fixed n) \na maximum f/W is achieved. Both the properties of the maxima of \nK, and the maxima of the R/W of (n,m) modulation are probably \nworthy of further study.\n\nTig. 4\u2014 Bounds on (K,R/W) for fixed n and comparison with permutation \nmodulation.\n\nThe main conclusion to be drawn is that the K, R/W plot provides \nan exceedingly useful technique for comparing modulation systems. \nWe have restricted ourselves to modulation systems in which the \nsignal alphabet consists of equienergy waveforms for which all rows \nof the correlation matrix are permutations of a given row. (Geo- \nmetrically, the alphabet consists of M points on the surface of an n- \ndimensional sphere such that all points see exactly the same environ- \nment.) This class of systems, although somewhat limited, is sufficiently \nbroad to cover most systems of theoretical and practical interest. \nGiven two systems in this class such that K, > Ko; then in the limit \nof large E/N, (low P,) Per < Peo for the same H/N,. Furthermore, \nwe have obtained a simple sufficient condition on the E/N, above \nwhich this inequality is valid. These results are in reality not new.\n\nThey are implicit in the results of Shannon\u2019 and in many other works.*\u00ae \nWhat is perhaps new is that many interesting results and comparisons \ncan be obtained by such simple techniques.\n\nConsiderably more precise comparisons can of course be made by \nexact computation of P, rather than by comparison of K. The latter \nprocedure however is considerably quicker and allows ready con- \nsideration of entire classes of systems (e.g., the (n,m) permutation \nmodulation and the combined phase-shift orthogonal modulations \nconsidered in the previous sections). The comparisons discussed here \nare not meant to supplant exact evaluation, but rather as a coarse sieve \nfor delineating systems worthy of more extensive calculation.\n\n1. Viterbi, A. J., On Coded Phase-Coherent Communications, IRE Trans. Space \nElec. Tel., SE7\u2019-7, 1, March, 1961, pp. 3-14 (see also Chapter 7 of Digztal \nCommumecations with Space Applications, by S. W. Colomb, et al., \nPrentice Hall, 1964).\n\n2. Jacobs, I., Theoretical and Practical Limitations of Weak-Signal Processing \neo Space Research II, North-Holland Publishing Co., 1961, pp.\n\n5. Reed, I. 8. and Scholtz, R. A., N-Orthogonal Phase-Modulated Codes, Univer- \nsity of Southern California USCEE Report 134, May, 1965, (Presented at \nthe First IEEE Annual Communications Convention, Boulder, Colorado, \nJune 7-9, 1965.)\n\n6. Reed, I. 8. and Scholtz, R. A., N-Orthogonal Phase-Modulated Codes, IEEE \nTrans. Inform. Theor., 7-12, July, 1966, pp. 388-395.\n\n7. Shannon, C. E., Probability of Error for Optimal Codes in a Gaussian Chan- \nnel, B.8.T.J., 38, May, 1959, pp. 611-656.\n\n8. Landau, H. J. and Pollak, H. O., Prolate Spheroidal Wave Functions, \nFourier Analysis and Uncertainty-III: The Dimension of the Space of \nKssentially Time- and Band Limited Signals, B.S.T.J., 47, July, 1962, pp. \n1295-1336.\n\n9. Slepian, D., Group Codes for the Gaussian Channel, Notes prepared for use \nat Summer School on Coding, Royan, France, August-September 1965.\n\n10. Kotelnikov, V. A., The Theory of Optimum Noise Immunity, Translated by \nR. A. Silverman, McGraw-Hill Book Co., Ine., New York, 1959.\n\n11. Landau, H. J. and Slepian, D., On the Optimality of the Regular Simplex \nCode, B.S.T.J. 45, October, 1966, pp. 1247-1272.\n\n12. Balakrishnan, A. V., Signal Selection Theory for Space Communication Chan- \nnels, Advances in Communication Systems: Theory and Application, Vol. \n1, Chapter 1, Academic Press, 1965.\n\n13. Nuttall, A. H., Error Probabilities for Equi-Correlated M-ary Signals Under \nPhase-Coherent and Phase-Incoherent Reception, IRE Trans. Inform. \nTheor., JT-8, July, 1962, pp. 305-314.\n\n14. Weber, C. L., Signal Design for Space Communication Channels, Part I, Uni- \nversity of Southern California USCEE Report 129, February, 1965.\n\n15. Sanders, R. W., The Digilock Orthogonal Modulation System, Advances in \nCommunication Systems: Theory and Application, ed. by A. V. Balakrish- \nnan, Vol. 1, Chapter 3, Academic Press, 1965.\n\n18. Arthurs, E. and Dym, H., On the Optimum Detection of Digital Signals in \nthe Presence of White Gaussian Noise, IRE Trans. Commun. Syst., CS-10, \nDecember, 1962, pp. 336-372.\n\n19. Wozencraft, J. M. and Jacobs, I. M., Principles of Communication Engineer- \nang, John Wiley & Sons, Inc., New York, 1965, Chapters 4 and 5.\n\n20. Cahn, C. R., Performance of Digital Phase-Modulation Communication Sys- \ntems, IRE Trans. Commun. Syst., CS-7, May, 1959, pp. 3-6.\n\n21. Pearson, K., Tables of the Incomplete Beta-Function, University Press, \nCambridge, England, 1934.\n\nCombinatorial Solution to the Problem of \nOptimal Routing in Progressive Gradings\n\nThe grading or graded multiple proposed by E. A. Gray ts a certain \nkind of one-stage, two-sided, partial access telephone connecting network \nfor switching customers\u2019 lines to trunks all having the same destination. \nIts essential feature is that traffic from lines not having identical access \npatterns can be offered to a common trunk, and so pooled. In a progressive \ngrading the trunk groups are partially ordered in a hierarchy, 7.e., some \nprovide primary routes, others function as secondary routes which handle \ntraffic overflowing from primary routes, as well as originating traffic, etc., \nup to final routes.\n\nA call which zs using an overflow or \u201c\u2018later\u2019\u201d\u2019 trunk when it could be using \na primary or \u2018earlier\u2019 group 1s said to make a \u201chole in the multiple\u2019. \nIt was recognized early in the development of gradings that such holes were \nundesirable.\n\nThe problem of optimal routing in telephone networks, considered in \ngeneral in the author\u2019s earlier work, 1s here specialized to progressive grad- \nings. It had been shown that for networks with certain combinatorial \nproperties the optimal choices of routes for accepted calls (so as to minimize \nthe loss under perfect information) could be described in a simple and in- \ntuitive way in terms of these properties. The present paper gives a proof \nthat all progressive gradings have such a combinatorial property, associated \nwith the hierarchical nature of the grading. The optimal policy for routing \naccepted calls 1s related to the phenomenon of \u201c\u2018holes in the multiple\u2019, \nand can be paraphrased in the traditional telephone terminology thus: \nfilling a hole in the multiple is preferable to using a final route, and filling \nan earlier hole is preferable to filling a later one.\n\nIn this paper, we consider some ways in which the concept of a \nhierarchy of routes is relevant to the problem of optimal routing as \nformulated in previous work.? Naturally, such a hierarchy can be \nrelevant to routing only if it is in a suitable way related to those \ncombinatorial properties of the network which distinguish the \u2018good\u2019 \nfrom the \u2018poor\u2019 ways of completing calls. (Examples of such properties \nwere given in Ref. 2.) It shall be shown that natural hierarchies as- \nsociated with certain gradings hold the key to the routing problem in \nthese one-stage networks.\n\nIt is now known? that if a network possesses one of certain com- \nbinatorial properties, then this property can be used to describe in a \nsimple way the optimal choices of routes for accepted calls so as to \nminimize the loss under perfect information. The next natural ques- \ntion is, then, what networks possess some of these properties? We \nshall prove that the members of an important subclass of connecting \nnetworks, that of progressive gradings, all have a combinatorial prop- \nerty similar to the strongest of those of Ref. 2; this property is as- \nsociated with a natural hierarchy of routes, and leads to a solution of \nthe routing problem for accepted calls.\n\nWe first discuss and clarify some of the usage and terminology as- \nsociated with gradings. Since about* 1905 the noun \u2018grading\u2019 and the \nadjective \u2018graded\u2019 have been used in telephony to describe a certain \nkind of one-stage two-sided network for connecting customers\u2019 lines \nto trunks all having the same destination. Roughly speaking, a grading \nhas this property: some trunk is such that two lines have access to \nit which do not have access to the same trunks. The essential feature \nis that traffic from distinguishable lines (i.e., ones not having identical \naccess patterns) can be offered to a common trunk.\n\n*. A. Gray proposed the \u201cgraded multiple\u201d in 1905, and was granted a \npatent for it (No. 1002388) in 1911.\n\nIt appears, though, that the word \u2018grading\u2019 has been used in a wider \nsense in Europe than in the United States. In particular, the American \nusage\u00ae implies a certain order in the pattern of access that the lines \nhave to the trunks, whereas in the European meaning this implication \nis absent. The order implicit in the American usage amounts to this: \nthe trunks are partitioned into groups which are so partially ordered \nthat no group has more than one successor in the ordering; a line that \nhas access to one group has access to all groups that follow it in the \nordering. (This ordering usually determines the order in which the \nlines hunt over the trunks.) Thus, e.g., a trunk group with no predeces- \nsors in the ordering can be used by exactly one group of lines, for \nwhich it is the \u201cprimary\u201d route. In one European sense of \u201cgrading,\u201d \nhowever, a trunk group which is the first one hunted over by one line \ngroup may be the nth one (n > 1) hunted over by some other line \ngroup.* The distinction drawn here is of some importance, inasmuch \nas the order structure implicit in the American usage gives rise to a \nnatural hierarchy of routes that is directly relevant to routing, whereas \nin the more general case this hierarchy is not necessarily present.\n\nRecently, in an effort to establish a uniform terminology, the \nnomenclature committee of the International Teletraffic Congress de- \ncided\u00ae that the terms \u2018grading\u2019 and \u2018graded multiple\u2019 should be in- \nterchangeable, and the structures described in R. I. Wilkinson\u2019s paper? \nas graded multiples be called, more specifically, progressive graded \nmultiples or progressive gradings, the word \u2018progressive\u2019 here re- \nferrmg to the order structure we have described as characteristic of \nthe American usage. The usage recommended by this committee is \nadopted herein.\n\nSince the present work can be viewed as a continuation of Ref. 2, \nwe take the liberty of assuming familiarity with the notations and \nconcepts used there, and we include only occasional reminders of the \nmeanings of important notions.\n\nIt will be convenient to have a notation for routes. A route r for a \neall c is just a way in which c can be put up or realized in a network \u00bb, \nand so it can be identified with the state in which the only call in pro- \negress is c using route r. Thus, a route for c is any element of y~*(c).* \nWe use the variables g and r (over the set L, of states with one call in \nprogress) to denote routes.\n\n* We recall that if xis a state, y(z)is the assignment of inlets to outlets realized by z.\n\nBy a hierarchy of routes we mean a partial ordering ~ contained in \nCae ao le\n\nIt is apparent that > can hold only between alternative routes for \nthe same call. (Of course, not every hierarchy of routes is relevant to \nrouting; only those that have a suitable relation to the ways in which \ncalls in progress block new calls will be of interest. The problem is \nto clarify the meaning of \u2018suitable\u2019.)\n\nA hierarchy of routes, being a partial ordering of the states with one \ncall in progress, can be extended to, or can induce, a partial ordering \nof the whole set S of states in several natural ways. Since > can hold \nonly between alternative routes for the same call, it is reasonable to \nconfine attention to extensions which hold only between states that are \nequivalent in the sense of ~ in Ref. 2, i.e., are (possibly) different ways \nof realizing the same assignment. An obvious first candidate for such \nan extension is given by the condition\n\nx~y and rSuaqsyr~q imply rDgq. (1) \nHowever, we eschew this definition in favor of a stronger one: let us set\n\nx Dy = \u00ab is reachable from y by sequentially moving calls in pro- \ngress from routes that are lower (later) (in the sense of \n\u2014 on L,) to routes that are higher (earlier) .*\n\nIt is intended here not merely that, as in (1), each call have a higher \nroute in x than in y, but that it should be possible to pass from y to z by \na sequence of equivalent states each differing from the previous one \nin that one call has been rerouted on a higher route. This stronger \ncondition is rendered formally by first defining\n\n* In an attitude prejudiced and justified by the principal results (Theorems 1 and \n2) we are working toward, we use the words \u2018lower\u2019, \u2018earlier\u2019, and their antonyms so \nas to suggest consistently that lower routes are less desirable than higher, earlier ones \nare preferable to later, etc.\n\nIn a one-stage connecting network vy = (G,J,Q,S), with J the set of \ncustomers\u2019 lines (inlets) and \u00a9 that of trunks (outlets), the graph G \ngiving network structure is determined entirely by the access relation \nA such that\n\nThe set S of states of \u00bby can be represented by the set of all subsets of \nA which are one-to-one correspondences. The range of x, rng (2), is \nthe set of trunks which are busy in 2.\n\nThe access relation A can be used to give a simple definition of a \nprogressive grading. We use X X Y for the Cartesian product of X \nand Y, i.e., the set of pairs (x,y) with \u00ab e X and y \u00a2 Y. If X is a set, \n| X | denotes the number of elements of X.\n\nDefinition: v is a progressive grading if and only if it is a one-stage \nnetwork for which there exist partitions II and Z of Q and J, respectively, \nand a partial ordering = of II, such that for 7, U, V eM andleaz\n\n@) (2 XT) an A # @ implies (L X T) CA, \n(ii) (L X U) CA, V = U imply (LX V) CA, \n(iii) U= T,V = Timply U S$ VorV s$ U \n@) |L|2z| U TI.\n\nThe first condition simply says that if a line has access to some trunk \nfrom a group 7\u2019, then all lines in its line group have access to every trunk \nin 7. The second condition says (roughly) that a line with access to \na trunk group 7\u2019 has access to all groups that are later than 7' in the \npartial ordering. The third condition says that a trunk group 1s followed \n(in the partial ordering) by at most one other group; if the \u2018\u201c\u2018later\u2019\u2019 \ngroups are thought of as overflow groups, this means that each group \nhas at most one group to which to overflow traffic. Finally, the fourth \ncondition rules out the relatively uninteresting cases in which some line \ngroup has access to more trunks zn toto than there are lines in the group.\n\nIt is apparent that if a trunk group 7\u2019, is later than one T\u2019, , then every \nline with access to 7\u2019, has access to 7\u2019, . This is the \u201cprogressive\u201d prop- \nerty. In analogy with the intuitions expressed in Ref. 2, it should be \nbetter to use an earlier trunk group than a later one, if both are available. \nThus, the structure of a progressive grading at once suggests the con- \njecture that optimal routing will consist of using the early routes in\n\npreference to the later or (to anticipate a bit) overflow groups. This \nconjecture is true and follows from Theorem 2. In traditional telephone \nterminology (see E. C. Molina\u2019s appendix in Ref. 3) it states that filling \na hole in the multiple is preferable to using a final route, and that \nfilling an earlier hole is preferable to filling a later one.\n\nA line group L is said to be a bye if it has access only to \u2018\u2018overflow\u201d\u2019 \ntrunk groups, i.e., if\n\nIt is easily seen that in a progressive grading a hierarchy of routes \ncan be defined by this rule: r D gq if and only if r ~ q and g(q) 2 g(r), \nwhere g(r) is the trunk growp used by route r.* This is the natural \nhierarchy of routes associated with a progressive grading; here r _) q \nif and only if 7 ~ q and r is on an \u201cearlier\u201d trunk group than g. In this \ninstance, = is also a simple ordering on each g(y~*(y(r))). These simple \norderings forming the hierarchy of course correspond exactly to the \npreference relation among routes suggested by the natural intuition \n(already mentioned) that there is no point in using a later or \u201c\u2018overflow\u201d\u2019 \ntrunk when an earlier one is available, because possibly fewer lines have \naccess to the latter. The relation > defined above on L, extends by \n(2) to all of S.\n\nIn a proof to be given later we shall use the fact that the set of pro- \ngressive gradings can be partially ordered by a relation D according \nto the following definition of covering: \u00bb, covers v, if and only if v2 is \nobtained from \u00bb, by removing, for some line group L, either (case 1) a \ntrunk from the first (in S,) trunk group to which L has access together \nwith one line of L if L has access to more than one trunk, or (case 2) \nthe trunk to which L has access together with L itself if L has access \nto exactly one trunk. That is, if \u00bb, is defined by partitions TI, , 4,, a \npartial ordering =, of I], , and an access relation A, , then v, covers v2 \nprovided that there exist t \u00a2 II, andl \u00a2 L \u00a2 Z. with\n\nFor practical purposes a network in which some line group has access \nto no trunks is in all respects equivalent to the same network with those \nlines omitted. For this reason the definition of covering was divided into \ncases 1 and 2, so as to build this equivalence right into the definition.\n\nAs we have said, \u00bb, covers v, if and only if \u00bb, results from \u00bb, by ripping \nout (7) some trunk from a \u201c\u2018primary\u201d\u2019 group, (27) a line with access to \nit, and (222) all crosspoints associated with these terminals, with the \nproviso that if this leaves some lines with access to no trunks, then these \nlines are also to be removed. Because of this, there exists a natural or \ncanonical map yu of the states S(v,) of \u00bb, into those S(v.) of \u00bb., defined \nroughly by the condition that ux is what is left of x after the line and \ntrunk that define the covering of v2. by \u00bb, have been ripped out. The \ncanonical map can be defined formally very simply, as follows: A state \nx of v, is representable as a subset of A, which is also a one-to-one cor- \nrespondence; similarly, a state of v. is just a one-to-one map contained \nin A; what is left of x after the ripping-out process is Just\n\nLe = 7 fr A\u00bb . \nThus, if \u00bb corresponds to ripping out line / and trunk \u00a2, and \u00ab = {(1,t)}, \nthen uz = 6 = zero state. If z = {(1,t,)} ora = {(1,, H} with, #1 \nand t, ~ t, then again wx = zero state. Ifa = {(,,4)} v y, withl, = 1 \nort, = t, then wx = py. It is easy to see that if u rips out J and \u00a2, then \nuwS 18 isomorphic with the \u2018\u2018cone\u201d\u2019 \n{ve Six = {(l,D}},\n\nbecause it does not make any difference whether / and \u00a2 are present in \nthe system and connected to each other, or are just absent. That is,\n\nuwS is essentially the set of states of S that remain available if | is con- \nnected to \u00a2 with a holding-time of +o. \nThis notion of a canonical map provides many useful notations. \nIt is convenient to extend the p-notation as follows: For 7 \u00a2 II\n\nClearly, u7' is what is left of the trunk group T after the line J and the \ntrunk \u00a2 associated with \u00bb have been ripped out. Also, we set\n\n{(ux uy): D y, pe \u00a5 Oorxe = O, and py + Gory = 4}. \nThe relation \u00bb 2 can be seen to be identical with 2, ; it is a useful \nmnemonic; it defines the hierarchy of routes in the \u201creduced\u201d system \nv2; the partial ordering induced in S(\u00bb,)[= wS(r,)] by this hierarchy is \nprecisely yp 2.\n\nVI. PRELIMINARY RESULTS \nIn Ref. 2, for a general partial ordering A, the notation \nsup A,, \nR\n\nwhenever this set was nonempty. The notation was chosen to denote \na set of R-maximal elements of A,,, rather than an actual R-maximal \nelement itself, so as not to prejudge the question as to how many there \nwere. It will be shown that if the network \u00bb under study is a progressive \ngrading, and R = 2 = natural hierarchy, then unless c is blocked in \nx (and A,, is empty) A., always has a C-maximal element which is \nunique to within equivalence under permutations of lines within their \nline groups and trunks within their trunk groups. |\n\nLet now x be a state and let \u00a2 \u00a2 x be a call which is not blocked in z. \nIt is apparent that for y, 2 \u00a2 A., we have either g(y \u2014 x) 2 g(z \u2014 2) \nor g(y \u2014 x) S&S g(e \u2014 x). Hence, there is a yp \u20ac A,, such that\n\nfor all w \u00a2 A,, , and yo is unique to within equivalence. (Recall the con- \nstruction of Q in Section III, and the fact that D is J u Q.) Hence,\n\nexists, and equals 7(yo), 7(-) being the natural homomorphism of S \ninto the quotient S/(2 n C). (See Ref. 2.) \nWe now consider policies g(-,-) such that \nea x\u2014h if eisa hangup h, \nbe sup A,, if eis a new call c not blocked in z.\n\nSuch a policy expresses the routing rule of always choosing the earliest \navailable trunk in the natural hierarchy characteristic of a progressive \ngrading.\n\nThe relation B (for \u2018\u2018better\u2019\u2019) was defined in Ref. 2 by the condition \nxBy ifandonlyif \u00ab2 ~ y and every call blocked in x 1s also blocked \nin y.\n\nBy Theorem 1, to be proved shortly, it will follow that x D y implies \nx B y, which in turn implies s(x) 2 s(y). Thus, the policies \u00a2(-,-) coin- \ncide with the \u2018\u2018maximum s(-)\u2019 policies suggested in Ref. 2. (See Ref. \n2 for notations.)\n\nLet / and t be the line and trunk, respectively, associated with yu. There \nexists a trunk \u00a2* such that\n\nLet T* denote the set (trunk group) achieving the infimum on the right. \nSince t is busy in x and \u00a2* is not, t \u00a5 \u00a2*. Thus, 7* \u00a5 {t}, and w(7*) + 8.\n\nWe next show that \u00bb7\u2019 \u20ac rng (ux) implies 7 \u20ac rng (x). If not, then \nthere exists t, \u00a2 uJ\u2019, hence \u00a2 T such that \u00a2, \u00a2 rng (ux) and \u00a2, \u20ac rng (2). \nBut\n\nNow 7* S T for every T such that 7 \u20ac rng (x) and [*A7. From the \ntwo previous paragraphs, it follows that \n(uT*)(u S)pT \nfor every 7 such that u7 \u20ac rng (ux), *AuT, uT \u00a5 0. That is,\n\nwy ESUP Agius \npS \nand since y was arbitrary within sup A,, , the lemma is proved. \n>\n\nNow let \u00a2 be a call from line / which is blocked in x but not in y. Then \nc is not blocked in z either. The only trunk which is busy in x and not \nin z is that used by the call y(a \u2014 z). Thus, since c is blocked in x and \nnot in z, g(x \u2014 2) is a trunk group usable for the call c. However, by \nproperty (22) of progressive gradings, {l} & g(y \u2014 z) C A, Le, J has \naccess to the group g(y \u2014 z) as well. Hence, some trunk of g(y \u2014 2) \nis idle in x, since the call y(2 \u2014 z) has a choice of routes in state z, one \nof these being on g(y \u2014 z). Thus, c is not blocked in x, and zx B y.\n\nTheorem 1: In a progressive grading, the partial ordering > induced by \nthe natural hierarchy of routes 1s contained in B.\n\nProof: Immediate from Lemma 2 and the facts that > is the transitive \nclosure of J u Q, and that B is transitive.\n\nLemma 8: If x D y, then x 1s obtainable from y by moving calls to earlier \nroutes in such a way that each call is moved at most once.\n\nProof: The result is true if only one move is made. Suppose it to hold \nif nm moves 2n toto are made. Let x be obtainable from y by sequence of \n(n + 1) moves. The trunk groups available for a given call c form a \nset simply ordered by S, and so can be indexed 1, 2, --- , the S-earlier \nreceiving the lower integer. For c S y(z), let n(c,x) be the index of the \ngroup used by \u00a2c in x. Some call c that is moved in obtaining x from y \nachieves\n\nStarting in state y it is possible to move such a call (once) directly to \nits route in xz, to get a state 2 in which it is still possible to carry out \nexactly each of the moves that take y into x except those involving c. \nThese are at most 7 in number, so each call involved need be moved at \nmost once.\n\nfor every event e that is either a hangup or a new call not blocked in \neither x or y. It has been shown in Ref. 3 for a general network that if \ngy preserves B then it embodies the optimal routing policy for accepted \ncalls.\n\nThe main theorem we prove (Section VII) states that a sup A., \npolicy, i.e., one satisfying (3), preserves >. The method to be used in \nthe proof of this result is illustrated in part by the following remarks: \nconsider linear arrays 2, y, 2, --- each of m urns, n = 2, each urn con- \ntaining at most one ball, with fewer than n nonempty urns per array. \nLet x D y mean that x is obtainable from y by moving balls to the left. \nLet yx denote the result of adding a ball in the leftmost empty urn.\n\nProof: The result is obviously true for n = 2 by enumeration. Let it \nhold for a given value n = 2, and consider arrays x, y of n urns satisfying \nthe hypotheses. Let yz denote the result of removing the leftmost urn \nfrom z, and bz that of adding an urn containing one ball at the left of \nz. There are two cases: (2) the leftmost urns are empty in both x and\n\nCase (iz): In obtaining x from y some ball moved into the leftmost \nurn. Obtain z from y by moving just this one ball to the leftmost urn. \nThen + D zD y, ot D v2, gy = blz, oz = bez. Since gz is obtained \nfrom wy by removing some ball, and replacing it in the leftmost empty \nurn of the resulting array, we have gyz D Wy, and so gz D gy.\n\nIn cases 3 and 4 of the proof of the next theorem, the analog of the \ninductive index n will be the partial ordering of the set of progressive \neradings.\n\nTheorem 2: In a progressive grading v let D be the partial ordering in- \nduced by the natural hierarchy of routes in v, and let \u00a2 be a policy with the \nproperty that\n\nProof: The proof is by induction over the partial ordering > of the \nset of progressive gradings which is defined by the definition of covering \ngiven earlier. A grading v that is minimal in D has no \u201c\u2018overflow groups\u2019\u2019, \n1e.. 2 = identity relation, so that no trunk group has a successor in \nthe order 2 characteristic of \u00bbv. Thus, v consists entirely of trunk groups \nserving line groups on a one-to-one basis, so that for some n\n\nIn this minimal case > is the identity relation, and \u00a2 obviously preserves \nit.\n\nAs a hypotheses of induction, we now suppose that every progressive \ngrading covered by v has the property that any sup A,, -policy preserves \n\u2014. Let now x D y inv and let e \u00a2 x. The induction argument will have \nfour cases, the last two of which are analogous to the observation made \nearlier.\n\ne987 jeder yn\u2014 1. \nThis sequence indicates how one would get y from x by moving calls \nto \u201cpreferred\u201d routes. By Lemma 3 it is no restriction to assume that \nno call is rerouted more than once. Let the route of h be rin x and q in y.\n\nIf h is one of the calls whose route is changed in the above sequence, say \nto take z, into 2,,, by changing the route of h from r to q, then\n\nis a sequence which shows that (2 \u2014 r) D (y \u2014 q). If the route of h is \nnot changed, then r = g and the same conclusion follows.\n\nCase 2: x D y, and e \u20ac x is a new call c blocked in x. By Theorem 1, \nx B y, so c is also blocked in y. Then,\n\nCase 3: x D y, eis a new call c not blocked in either x or y, and the line \ngroup L of c is not a bye. Let\n\nSubcase 3.1: T is full in neither x nor y. Then there exist routes r, q\u00a2 \nsuch that g(r) = g(q) = T,\n\n21,22, \u00b0\u00b0 \u00bb 2, = Y IS a Sequence with \n2; O 2341 g=1,\u00b0-\u00b0: ,n\u2014I, \nshowing that x > y, then \nLUr=2, UU yk UrT=YuUrY Y gis a sequence which \nshows that\n\nThis is because we can assume without loss of generality that the trans- \nformations which change y into x reroute a call at most once, and thus \nmove no calls onto 7\u2019. (Lemma 3.)\n\nSubcase 3.2: T is full in both x and y. Since L is not a bye, there exist \nl_.melLandt, ue T with\n\nBecause 1, m and \u00e9, wu are respectively interchangeable, 1.e., since lines \nand trunks are permutable within their respective groups, no loss of \ngenerality is incurred if it is supposed that 1 = m andt = n. Let pu be \nthe canonical map corresponding to ripping out / and Z.\n\nin \u2014< in case 1 \nI \u2014 {7} in case 2, \nume \u2018\u2014 in case 1 \nm\u2014 {D} in case 2, \n> with 7 \u2014 {t} replacing 7\u2019 throughout, in case 1 \nw= = 2, \n2 Sy Xt) in case 2, \nth A cane in case 1 \nA\u2014 (UX {t}) - ZX Q) in case 2, \nwith\n\npo(c,x) Sd A eur) \nwe(c,y) Esup Agu) . \nre) \nSince x D y, and either both wx = 0, py = 0, or neither, we have \n(ux my) \u20ac wD. (4) \nLet ~ be a policy for \u00bb, with \nE(d uz) \u20ac hes A atue) ? Vd E pe. (5)\n\nBut \u00a2(c,x) differs from ye(c,x) and \u00a2g(c,y) from py(c,y), only in having \nan additional line 1 and an additional trunk \u00a2 connected to each other. \nHence,\n\nThe argument of subcase 3.2, basic to Theorem 2, can be appreciated \nby looking at it thus: x D y means 4z,, \u00b0\u00b0: , & with 2; Q 241,7 = \nlee, n\u201414=2,2 = y. Since\n\nusing r is not moved as y is transformed into zx by moving calls. Thus, \n(2; \u20147r) Q @is1 \u2014 1), ~=1,\u00b0---,n-1 \n@\u2014rnQdy >).\n\nBut the \u2018\u2018cone\u201d {z:z = r} is isomorphic to the states of a grading (\u00bb, of \nthe proof) covered by v and the isomorphism, 22z., uw restricted to the \ncone, has the basic property, for x, y in the cone\n\nSubcase 3.3: T is full in x, but not in y. Since ZL is not a bye it is S- \nminimal, and hence there exists a call d with d S y(x) nm y(y) such that \ndis on T in x is not on T in y, and can be moved to T in state y to give \nrise to a new state z without rendering impossible any the remaining \nmoves which transform y into x. Thus, 7 D z D y. Since x n z \u00a5 8, \nsubcase 3.1 gives v(c,x) D (cz). Further, the route of c in \u00a2(c,z) is \nno higher (later) in $ than the one in y left by d as it was moved to \nT to give rise to state z. Hence, to within equivalence\n\nCase 4: x D y, \u20ac is a new call c not blocked in either of x or y, and the \nline group of c is a bye. There is at least one other line group LZ which\n\nSubcase 4.1: L XT nxw4#0LXTanyxroboorLxTnz= 8, \nLXToay = 6. Since L is not a bye, there exist 1,m eLandt,u eT \nwith\n\n(In the second instance, property (2) of the definition of a progressive \ngrading has been used to conclude that there must be idle lines on L \nif there are idle trunks on 7.)\n\nAs in subcase 3.2, no loss of generality is incurred if it is supposed \nthat 1 = m and \u00a2 = wu. Let uw be the canonical map corresponding to \nripping out J and \u00a2. The argument now continues as in subcase 3.2.\n\nSubcase 4.2: (L XT) nx4#90,(LXT) ny = 6.Since Lis S-minimal, \nthere exists a call d with d = y(x) n y(y) such that d is on T in @, is \nnot on 7 in y, and can be moved to 7 in state y to give rise to a new \nstate z without rendering impossible any of the remaining moves which \ntransform y into x. Thus, x D z D y. Since x n z \u00a5 6, subcase 3.1 gives\n\nT,. = gQnf {S:*AS,S \u20ac rng (y)}). \nHere 7\u2019, is the earliest group c could be put on in y. Let also y denote \nthe operation of moving d from T, to T, and for any call f \nA; = tor): = v(r)} \n= {S:the line of f has access to S}.\n\nCase (1): Tz \u20ac A, N Ag, Ta S T,. Then moving d from T,; to T means \nthat c can use 7', in Zz, so o(c,z) > o(c,y), because \u00a2(c,z) results from \ny(c,y) by moving first d to J, and then \u00a2 to T, , so actually\n\n1. Weber, J. H., Some Traffic Characteristics of Communication Networks \nwith Automatic Alternate Routing, B.S.TJ., 41, July, 1962, pp. 1201-1247.\n\n2. Benes, V. E., Programming and Control Problems Arising from Optimal \nRouting in Telephone Networks, B.S.T.J., 45, November, 1966, pp. 1373- \n1438; Abstract in SIAM J. Control, 4, 1966, pp. 6-18.\n\n3. Wilkinson, R. J., The Interconnection of Telephone Systems\u2014Graded Multi- \nples, B.S.T.J., 10, October, 1931, pp. 531-564.\n\n4. Syski, R., Introduction to Congestion Theory in Telephone Systems, Oliver \nand Boyd, Edinburgh and London, 1960.\n\nIntegral Equation for Simultaneous \nDiagonalization of \u2018Two \nCovariance Kernels\n\nLet K,(s,t) and K,(s,t), \u2014T S s,t S T, be real, symmetric, continuous \nand strictly positiwe-definite kernels, and denote by K, and Kz the cor- \nresponding integral operators. Let x(t) be a sample function of erther of \ntwo zero-mean processes with covariances K,(s,t) and K.(s,t). We prove \na generalized version of the following: If the integral equation\n\nhas formal solutions \\; and W;(t) which may contain 6-functions, and \naf {K,;} forms a complete set in &.[\u2014T,T], then (2) the two kernels have \nthe following sumultaneous diagonalization:\n\nin the stochastic mean, uniformly in t, and the coefficients are simul- \ntaneously orthogonal, 7.e.,\n\nEt (w,wi)(@v;)} = 6s , Eni (x,vi)(a,W;)} = rz bi; , \nwhere (x,;) ts obtained by formally integrating y(t) against x(t).\n\nKk, the integral operators with kernels K,(s,t) and K.(s,t). We have \npreviously\u2019 established that, if Ky7?K.K7? is a densely defined and \nbounded operator on \u00a3, (the space of all square-integrable functions \non [\u2014T7,T]) and if its extension to the whole of \u00a3, has eigenvalues \n\\; and complete orthonormal eigenfunctions \u00a2,(t), 7 = 0, 1, --+ , then \nthe two kernels have the following simultaneous diagonalization:\n\nuniformly on [\u2014T7,7] X [\u2014T7,T]. In addition, if x() is a sample func- \ntion of either of two (separable and measurable) zero-mean processes\n\na(t) = \u00bb ni(2)(Kig,) (t) | (2) \nin the stochastic mean, uniformly in \u00a2t. Moreover,* \nEy {ni(a)nj(a)} = 43; , Lat n(x) ni(ax)} = rx be \nwheret \nn(x) = lim (2,K7*y;,) (3)\n\nin the stochastic mean, and {\u00a2;,} 1s any sequence of functions in the \ndomain of Kj? such that lim || 9; \u2014 \u00a2;, || = 0.\u00b0\u00b0? Furthermore, if the \ntwo kernels have continuous 27th derivatives (0\u00b0\u2019/ds\u2019dt\u2019)K,(s,t), p = \n1, 2, then (1) and (2) can be differentiated term-by-term r times while \nretaining the same senses of convergence.\u2019\n\nWe remarked in Ref. 1 that, if y; is in the domain of K;?, \u00a5; = Ky\u00bb; \nsatisfies the integral equation\n\nSlepian (private communication) has long conjectured that, if (4) \nadmits formal solutions \\; and y; , 7 = 0,1, --- , where y; may contain\n\n* Ey, p = 1, 2, denotes the expectation with respect to P \n+ For any f, g eL2 , (f,g) denotes the inner product of f and g, and || f || the norm \nof f.\n\n6-functions and their derivatives, then the expansion coefficients and \nfunctions of (2) are given by formally substituting such y, into (5).* \nThis conjecture, proved here, is significant since it provides a concrete \nmeans of obtaining the expansions (1) and (2). To illustrate the point,\n\n(iid)n; = c,(z,b;) as.,* Kip; = c;Ky);, which verifies Slepian\u2019s \nconjecture for this example. Here c; is a normalization constant given by\n\nCox+1 = Cox lon\u2014dx ; \n(that is, Co.4; is obtained by replacing @, with 6, in co,), up; and fp; , \np = 1,2,7 =0,1,--- , are the eigenvalues and orthonormal eigenfunc- \ntions of K,, and (y; , f:;) 1s defined analogously to (9). \nIn this paper we prove the generalization of (2), (27), and (222), starting \nwith abstract kernels K,(s,t) and K,(s,t) and a generalized version of \nthe integral equation (4).\n\nII. MAIN RESULT \nTheorem: Let K,(s,t), p = 1, 2, -\u2014T S s,t S T, be real, symmetric, \nstrictly positive-definite kernels with continuous 2 rth derivatives \n(0\u00b0\"/dsdt\u2019)K,(s,t). If there exist sequences of real numbers {Qitm},\n\nfor some constants b, and bz, , and sequences of square-integrable functions \n{W.1}, which satisfy the equation\n\nsuch that the right-hand side of (12) forms a complete set in \u00a32 , then \n(() K7?K.K;? is a densely defined and bounded operator on \u00a32 , \n(22) ats extension to the whole of &2 has eigenvalues and complete ortho- \nnormal eigenfunctions, which are the \\; and\n\nand by the right-hand side of (12) without \\;. Here, K?,,, p = 1, 2, \ndenotes an integral operator whose kernel is defined as\n\n2 Mil pi (fri Q) = a \nuniformly in (s,t).\u201d It follows from this that (15) converges in the \nmean in (s,f) as well. Hence, from Fubini\u2019s theorem, K?,,(s,t) is a \nsquare-integrable function of t for almost every s. Thus, \u00a2;(s) of (18) \nis well defined. We assume without loss of generality that 9;, 7 = \n0,1, --- , are normalized.\n\nthe right-hand side of (12) without \\; is given by (10), and completeness \nof {cos 6,t, sin 6,\u00a2} follows from (18) and a gap-and-density theorem.\u00b0\n\nFor notational simplicity, we write K,.1, p = 1, 2, for the integral \noperator whose kernel is\n\nwhere the second equality follows from the mean convergence of (15) \nand the third from the uniform convergence of (16). To prove (18), \nconsider\n\nwhich vanishes as 7 \u2014 \u00a9 since (16) converges uniformly in (s,t). \n(7) K;?Ki and K;?K? are densely defined and bounded on &, . \nTo prove this, apply Kz? on both sides of (12) and use (18) to obtain\n\nwhere the second equality follows from (17) and (18), the third from \nk time differentiation of (12) and from (17) and (18), and the last \nfrom (13). Hence, with 9; being normalized,\n\nNow {y;} is complete since the right-hand side of (12) without ), , \nwhich forms a complete set by hypothesis, is equal to Kiy;, and K? \nis strictly positive-definite. Hence, from (11), Ky?K? is densely defined \nand bounded.\n\nTo prove that K;?K3 is also densely defined and bounded, define \n\u00a2,; as the normalized right-hand side of (18) with the subscript 1 re- \nplaced by 2. Completeness of {\u00a2;} is similarly deduced via (12). Now, \nby following the same procedure with the roles of K, and K, inter- \nchanged, we obtain\n\n|| KriK3\u00a2, \nThen, the assertion follows immediately from (11). \n(vit) The ranges of K? and K? are equal, namely, \nKi(\u00a32) = Rie):\n\nTo prove this, denote by ZL and M the extensions to the whole of \n&, of K>?K? and Ky'K? respectively, which exist as a result of (it). \nSince the domains of K?Z and KiM are &, , which is also the domains \nof K? and K?, we have\n\nKi = KiL, K} = Kim. \nThen, from the first equality, Ki(\u00a3.) C Ki(\u00a3,), while, from the second, \nK3(\u00a32) C Ki(\u00a32). Hence, the assertion holds.\n\nTo prove (19), note first that fi; , 7 = 0,1, --- , are in the domain \nof Kz? as a result of (i772) and also that (Ky*f1;, K3f:;) = 6;; from \northonormality of {f,;}. Thus, {K7#/,;} and {K3f,;} form a pair of \nmutually reciprocal bases of \u00a3, . Hence,\n\nBut, according to (22), the right-hand side converges uniformly. Hence, \nthe above partial sum must converge uniformly to f,; . Suppose for \nsome k,0 Sk <1,\n\nAPO = > (hi fedf PO) \nuniformly in \u00a2.\u201d Hence, by induction, (23) holds for every k,O Sk Sr \nTherefore, from (22), \n(Ke*hi; , Kin()) = fP?@, b=0,1, +--+ ,7. \nThen, (19) follows from (21) and the above. \nTo prove (20), we expand K3,, g relative to {K?},,}:\n\n(v) To prove (2) of the theorem, we note from (727) and (72) that \nKk7?K? is everywhere-defined and bounded on \u00a32. Hence, its adjoint \n(K;?K3)* is also everywhere-defined and bounded. Now, for any\n\nf \u00ab & and g \u00ab D(K;'), the domain of K;?, we have (Kj?K3f,g) = \n(f,K3Ky*g). Thus, KiKy*g = (Ky*K3)*g, g \u00ab D(Ky*). Hence, KiK;y} \nis bounded. Since (Kj?) is dense in \u00a3,, we conclude that Ky7?K.K;? \nis densely defined and bounded.\n\nwhere the second equality follows from (19), (20), (15) and (18), and \nthird from (12) and (13). Now denote by Q the extension of K7?K,Ky7? \nto the whole of \u00a3, . Then,\n\nwhich vanishes as nm \u2014 \u00a9. Therefore, Qy; = X,9,. Lastly, since {g;} \nis complete in \u00a3,, {d;} constitutes the entire spectrum of Q. \n(viz) To prove (272) of the theorem, note from (8), (24) and (vz) that\n\nboth of which vanish as n \u2014 \u00a9 by virtue of (19) and (20). Therefore, \nupon combination of the above results, (14) is proved.\n\n1. Kadota, T. T., Simultaneous Diagonalization of Two Covariance Kernels \nand Application to Second-order Stochastic Processes, submitted for publi- \ncation in SIAM J. Appl. Math.\n\n2. Root, W. L., Singular Gaussian Measures in Detection Theory, Proc. Symp. \non Time Series Analysis, edited by M. Rosenblatt, John Wiley & Sons, \nNew York, 1963, pp. 292-315.\n\n3. Yaglom, A. M., On the Equivalence and Perpendicularity of Two Gaussian \nMeasures in Function Space, Proc. Symp. on Time Series Analysis, edited \nby M. Rosenblatt, John Wiley & Sons, New York, 1968, pp. 327-348.\n\n4. Huang, R. Y. and Johnson, R. A., Information Capacity of Time Continuous \nChannels, IRE Trans. Inform. Theory, 77-8, September, 1962, pp. 191-205.\n\n5. Kadota, T. T., Simultaneously Orthogonal Expansion of Two Stationary \nGaussian Processes-Examples, B.S.T.J., 45, September, 1966, pp. 1071-1096.\n\n6. Kadota, T. T. and Shepp, L. A., On the Best Finite Set of Linear Observables \nfor Discriminating Two Gaussian Signals, IEEE Trans. Inform. Theory, \nApril, 1967.\n\n7. Kadota, T. T., Term-by-term Differentiability of Mercer\u2019s Expansion, Proc. \nAm. Math. Soc., 78, February, 1967, pp. 69-72.\n\n8. Levinson, N., Gap and Density Theorems, Am. Math. Soc. Colloquium Pub- \nlications, 26, Am. Math. Soc., Providence, Rhode Island, 1940, p. 3, \nTheorem II.\n\n9. Apostol, T. M., Mathematical Analysis, Addison-Wesley, Reading, Massachu- \nsetts, 1957, p. 403.\n\nPrinciples of Design of Magnetic Devices for \nAttitude Control of Satellites\n\nMagnetic devices mounted within an orbiting satellite interact with \nthe earth\u2019s magnetic field and produce torque to modify the attitude or \nangular adjustment of the satellite axis of spin. The satellite environment \ndictates that these devices be designed for minimum weight or minimum \npower consumption, or a suitable compromise between these two minima. \nPrinciples of design of magnetic devices to satisfy these requirements are \ndeveloped in this paper. The resulting design equations and charts enable \nthe ready optimization of design and selection of preferred materials. \nWhile most of this work was directed initially at the Telstar\u201d satellite \nproject, the design charts and formulas are found useful in other areas \nof magnet design. Methods of magnetic measurement devised for the satellite \nare discussed.\n\nSatellites with directional instrumentation, such as the antenna sys- \ntem of the communications satellites, require attitude control to keep \nthis instrumentation properly on target. For example, a spin imparted \nto the satellite at time of launch gives it a sort of gyroscopic stability. \nHowever, complete attitude control requires some available torque to \ncorrect the direction of the spin axis.\n\nIn the orbiting satellite the earth\u2019s gravitational field is balanced by \ncentrifugal force, leaving the earth\u2019s magnetic field as a convenient \nmeans for interaction torque. Suitable interaction with the earth\u2019s \nmagnetic field can be set up by electromagnets, or by air-core coils of \nlarge area, either of which can be turned on or off at will to provide \nattitude correction as needed. Small permanent magnets can be de- \nsigned and installed to cancel out residual magnetic moment in the \nsatellite, which if permitted to interact with the earth\u2019s field could \ncause precession of the spin axis. Other miscellaneous torque applica-\n\nLimitations of payload and of available power in the satellite gen- \nerally make it necessary to design with quantitative accuracy and to \noptimize the factors which control weight and power consumption. To \nthis end, the magnet designer may select from various geometries of \nmagnet and coil and from various available materials. This selection \nand optimization is facilitated by the use of suitable design formulas \nand charts. In this paper, we review the derivation and illustrative \nuse of such formulas and charts. While the work reported here has \nbeen aimed specifically at certain problems of the Telstar\u00ae satellite, \nit is evident that the technique of magnet design presented here is ap- \nplicable to any similar set of problems.\n\nFor the convenience of the magnet designer who buys magnets and \nmagnet wire by the pound, measures them in feet, inches, or mils, \nand measures torque in pound-inches, all of the derived design \nformulae and graphs are built around the practical units (inches, \npounds, oersteds, gauss, etc.). This avoids the necessity of converting \nunits, which is time consuming and can lead to costly errors. There \nis included for convenience a table of the most frequently used con- \nversion factors (Table I).\n\nThe torque characteristic of the air-core coil is derived from the \ngalvanometer formula which, in some textbooks, is written in MKS \nunits:\n\nCopper Aluminum \nRom | aaa ace \nE(volts) 0.75 X SERED 1.21 x Wa) \nW(watts) 2.18 x 1.21 X as (NI)? \nWet.(Ibs.) 0.321 AP 0.0983 AP\n\nNI: Required ampere turns \nN: Number of turns used \nA: Cross section area of winding (inch?) \n(N times the section area of a single turn) \nP: Average length of turn in winding (inch)\n\nHere 7\u2019, is the maximum torque exerted on the coil when its axis 1s \nperpendicular to the field H,, and NIA is the required product of am- \npere turns and area enclosed by the coil to deliver that amount of \ntorque.\n\nIt is convenient to set up a table of formulas from which one may \ntranslate the geometry and ampere-turn characteristics of the coil into \npower and weight requirements. The power and weight will depend \nupon the winding material used, but practical considerations usually \nlimit this to copper or aluminum. So one may take the weight and \nresistivity characteristics of copper and aluminum from handbook \ntables and with the aid of Ohm\u2019s Law derive the formulas of Table II. \nUsing (2) and Table II, one may estimate readily the power and \nweight of an air-core coil to satisfy specified torque requirements. It is \nevident that copper has the advantage in lower power consumption, \nbut that aluminum offers a greater advantage in weight reduction. If \npower and weight are of about equal importance, then the power- \nweight product should be minimized. It is evident that aluminum has a \nfactor-of-two advantage over copper in this characteristic.\n\nA magnetized bar, either a permanent magnet or the core of an \nelectromagnet, displays a moment, or normalized torque, proportional \nto the product of the volume of the bar by the intrinsic induction \nwithin the bar. The magnetic moment, M,,, is identified as normalized \ntorque in the familar equation\n\nand the relation between magnetic moment, intrinsic induction in the \nbar, and the geometry of the bar is given by another familiar equation\n\nin which B \u2014 AH is the intrinsic induction, A is the cross-section area \nof the bar at the median plane, and S is the effective distance between \npoles. For a magnet of length / and diameter d, one may define a \nshortening factor, Rg = S/l, which evaluates the effect of recession of \nthe poles, and rewrite (4) as\n\nThe shortening factor, Rs, has been evaluated by Okoshi.t Okoshi\u2019s \nvalues are plotted as a function of lJ/d of the bar in Fig. 1, with a\n\nbroken line extrapolation guided by experimental data. If one com- \nbines (3) and (5) after conversion to practical units, the result is\n\nand the required volume of bar to produce a specified magnetic mo- \nment is given by rearrangement of (6)\n\nHere H, is the applied magnetizing field, and B and H describe the \ncondition of magnetization within the bar. The demagnetizing factor, \nDz\u00bb, which is partially defined by (8) is used to express the dependence \nof the intrinsic induction within the bar upon the aspect ratio (l/d) of \nthe bar. It has been tabulated and charted as a function of l/d by \nOkoshi,? Bozorth and Chapin,? and others. These sources agree upon \nthe value of Dz for long slender magnets. For shorter magnets, where \nthere is some disagreement, we find the Okoshi data to be in agreement \nwith experiment.\n\nIn plotting the characteristics of magnetic materials we normally \nplot the intrinsic induction (B \u2014 H) or the flux density (6) as the \ndependent variable, and the field strength (H, or AH,) as the in- \ndependent variable. Hence, the ratio\n\nbecomes the slope of the generalized load line of the magnetized bar. \nThis reciprocal of the demagnetizing factor is found useful in numerous \nmagnetic calculations and possibly deserves a name and symbol of its \nown. We have elected to call it the loading factor, with the symbol U,\n\nand have plotted it as a function of //d in Fig. 2. In the electromagnet \ncore operating below saturation, H is generally negligibly small compared \nwith either B or H, , and the expression for the loading factor reduces \nto U = B/H,. In the permanent magnet, H, disappears and the ex- \npression for loading factor becomes U = (B \u2014 H)/(\u20144H). For long \nmagnets, (I/d > 5) H is negligibly small compared with B, and the \nloading factor is further simplified to U = B/(\u20144H). In this restricted \nform the loading factor is identified with the \u201c\u2018permeance coefficient\u2019 \nand similar terms used in the literature of permanent magnets.\n\nIn Fig. 3 we illustrate the application of the loading factor to the \nanalysis of permanent magnets and electromagnets. For this illustra- \ntion each is assumed to have I/d = 33 so that the loading factor, U & \n400. In the permanent magnet (Remendur) the magnetomotive force\n\nFig. 3\u2014 Application of load factor to magnet design: (a) permanent magnet \n(Remendur 38), (b) electromagnet (Permalloy 45 core).\n\nis generated within the magnet and varies with the loading of the \nmagnet as indicated by the B, H curve. (Here 4H is sufficiently small \nso that B is indistinguishable from B \u2014 #H). The operating point is \ndetermined by the intersection of the B, H curve with the load line \nof slope U. This is a fixed point for a particular magnet with a par- \nticular condition of magnetization. The conditions of Fig. 3 were chosen \nto match the characteristics of Remendur. In the electromagnet when \noperated below saturation, H is negligibly small so the load line rep- \nresents the relation between flux B in the bar, and applied field, H, , \nup to the region (around B = 10,000 for Permalloy 45) where the core \nmaterial starts saturation, and the characteristic starts deviating from \nthe straight load line.\n\n3.2 Design of Permanent Magnets for High Torque-W eight Ratio \nThe weight of the magnet in pounds is\n\nin which p is density of magnet material in Ibs/in\u00b0. One may combine \n(10) with (6) to obtain\n\nHere the left-hand side of (11) is the normalized torque-weight ratio. \nThe design objective is to maximize this ratio.\n\nThe dependence of the operating point of the permanent magnet \nupon the load factor, U, has been illustrated in Fig. 3. On similar charts \none may plot intrinsic induction (B \u2014 H) as a function of field (A) . \nfor various magnet materials as in Fig. 4. Values of B and # for these \nplots may be derived readily from the regular demagnetization curves \nsupplied by magnet manufacturers. Then, for a particular value of\n\nFig. 4\u2014JIntrinsic induction of permanent magnets; (a) directional grain \nceramic, (b) Alnico 9, (c) Alnico 5.\n\nFig. 5\u2014 Variation of torque-weight ratio with aspect ratio; (a) directional \ngrain ceramic, (b) Alnico 9, (c) Alnico 5, (d) Remendur 38.\n\nl/d one may pick off the corresponding value of U from ITig. 2, and \nusing this as the slope of the load line may find the value of (B \u2014 H) \nfor a particular magnet material at the point of intersection. This \nvalue of (B \u2014 H) and the value of p appropriate to the material may \nbe inserted in (11) to give the normalized torque-weight ratio. For \nexample, at l/d = 4, U & 17.5. The load line of that slope intersects \nthe intrinsic induction curve for Alnico 5 at (B \u2014 H) = 10,000. In- \nserting this value and the value of Rs in (11) and using p = 0.26 for \nAlnico, one obtains\n\nWH, \nRepeating this procedure for various values of l/d and for various \nmaterials, one can assemble the necessary data to plot the curves of \nFig. 5.\n\nIt is evident that for each magnet material there is a value of l/d \nabove which the torque-weight ratio is essentially constant, and below \nwhich the torque-weight ratio falls off rapidly with decreasing [/d. \nThis follows the shape of the demagnetization curves of Fig. 4. This \nvalue of I/d is large for magnets having low coercivity, and small for \nmagnets having high coercivity. One would normally design the mag- \nnet to operate on the flat part of the characteristic to obtain high \ntorque-weight ratio.\n\nThe electromagnet is assumed to consist of a cylindrical core of \nferromagnetic material with a solenoid wound around it. The design \nformula for the core is the same as that for the permanent magnet and\n\nis given in (7). This gives the required volume to produce a specified \nmoment, operating the core at a specified value of flux density. The \nampere-turn requirements are derived as follows.\n\nNI \n2.021 \nIf, as is usually the case in the electromagnet, H is negligibly small \ncompared with H,, then one may combine (8) and (138) to obtain\n\nSince Dy, and Rg are functions of ee ratio (lJ/d) one may define \nan aspect ratio factor,\n\nFor any proposed geometry of an Seeeauane with specified value \nof magnetic moment (7',/H,) the ampere-turn requirement may be \ndetermined from (19) and then translated into power and weight re- \nquirements by reference to Table II.\n\nA permanent magnet material with low coercivity and high rema- \nnence, as exhibited by Remendur in Fig. 3, offers the inviting possi-\n\nbility of easy magnetization and reversal of field by means of short \npulses of current through a winding. Between pulses it acts as a perma- \nnent magnet, with polarity determined by the direction of the preced- \ning pulse. Thus, it provides a switchable field with very low expendi- \nture of power. It requires, however, for complete demagnetization, or \n\u201cknock-down,\u201d much more sophisticated circuitry. Also, the certainty \nof complete demagnetization from an applied pulse or series of pulses, \ndepends to a considerable extent upon the preceding history of the \nmagnet. For this reason, it is not likely to replace readily the simple \nair-core coil or electromagnet unless the available power is so \nseverely limited as to justify the added circuit development effort.\n\nIn the preceding sections we have developed design formulas and \ndesign graphs which enable us to estimate with fair quantitative ac- \ncuracy the size and weight of various magnetic structures to satisfy \ntorque requirements as specified. Intercomparison of the air-core coil \nand electromagnet offers an interesting illustration of the use of these \ntechniques. We consider a typical example which assumes a spherical \nsatellite of 45 inches effective diameter in which there is required an \navailable magnetic moment of 0.2 lb-in per oersted which can be\n\nturned on or off at will. It is further assumed that an upper limit of \nnine pounds weight and twelve watts power consumption is to be \nimposed upon the magnetic circuitry.\n\nFirst, we assume that an air-core coil is laid out around the equator \nof the satellite to enclose maximum area, and that this area is\n\nformula for aluminum from Table II we can show that the power X \nweight product is\n\npower X weight = 0.119 * 10\u00b0\u00b0(141)\u00b0(210)\u201d = 105. \nIf we use the total weight allowance of nine pounds for the winding \nthen the required power is\n\nThis is within the permitted 12 watts, so we have shown that it is \nfeasible to use an equatorial coil.\n\nTurning now to the design of the electromagnet, one inserts T,/H, = \n0.2 and (B \u2014 H) = 10,000 in (7) to show that the required volume of \ncore material is\n\nThe characteristics of the winding, however, are closely dependent \nupon the aspect ratio of the core. To illustrate this point we consider \ntwo shapes, one to be 10 inches long, the other to be 45 inches long \nto just fit in along the spin axis of the satellite. For the 10-inch core,\n\nIf we assume the average diameter of the winding is 1.6 inches then \nthe average length of turn, P = 5.05 inches. We insert these numbers \ninto the power X weight formula for aluminum in Table IT and obtain\n\nIf we assume that the average diameter of the winding is 0.85 inch, \nthen the average length of turn, P = 2.67 inches, and\n\nSo we may use 0.5 pound of winding to bring the total weight only to \nfive pounds, and the required power will be only 0.78 watt. This \nillustrates the advantage of the long slender electromagnet over the \nshort one for purposes of producing torque.\n\nIt is evident that the specified conditions of the example can be \nsatisfied by the equatorial coil or by the long slender electromagnet. \nOn the basis of the calculated results one might well prefer the \nelectromagnet, which satisfies the requirements with a substantial \nsaving of power and weight. However, other factors must be con- \nsidered. It is not likely to be convenient to mount the magnet full \nlength along the spin axis because of interference with other equipment. \nIn a core of this length, a very small amount of residual magnetization \nafter removal of current will result in a considerable magnetic mo- \nment, rather than the desired zero magnetic moment which is charac- \nteristic of the de-energized air-core coil. The weight distribution of \nthe electromagnet along the spin axis decreases the spin stability, \nwhile the weight distribution of the equatorial coil enhances the spin\n\nstability of the satellite. For these and other reasons the equatorial \ncoil remains a favored method of attitude control in the communica- \ntions satellite.\n\nThere have been various proposals to provide friction damping of \nroll or precession of the spin axis by mounting a small magnet within \na hollow spherical enclosure attached to the satellite. The magnet \nwould tend to maintain its alignment in the earth\u2019s field and to pro- \nvide damping through friction contact with the interior of the sphere. \nFor this application, if it exists, or for any similar application, one \nwould wish to design for maximum normalized torque and minimum \nnormalized period of oscillation in the field and within the confines \nof the sphere.\n\n5.1 Design for Maximum Moment within Limiting Spheres \nReferring to (6) and dividing through by D* where D is diameter \nof sphere in inches, and D? = (d?+/?)%/?, \ni \nH,D\u00b0 \nThe relation expressed by (20) is displayed in Fig. 7 for the same\n\ng. 7\u2014 Normalized moment within limiting sphere; (a) ceramic, (b) Alnico \n9, co \u201cAlnico 5. :\n\nThis normalized period of oscillation is displayed graphically as a \nfunction of l/d in Fig. 8. In designing a magnet for friction damping\n\nFig. 8 \u2014 Period of oscillation within limiting sphere; (a) ceramic, (b) Alnico 9, \n(c) Alnico 5. ,\n\none would probably select the best compromise between maximum \ntorque displayed in Fig. 7 and minimum period of oscillation as shown \nin Fig. 8. This would suggest the use of Alnico 9 and design for J/d & 1.5.\n\nThe spheroids are a family of solids the surfaces of which are \ngenerated by ellipses revolving around an axis. Revolution around a \nmajor axis generates a prolate spheroid for which //d > 1. Revolution \naround a minor axis generates an oblate spheroid for which l/d < 1. \nRevolution of a circle around a diameter generates a sphere for \nwhich l/d = 1.\n\nValues of load factor, U, for spheroids are plotted in Fig. 1. The \nvolume of the spheroid is only two thirds that of a cylinder having \nthe same I and d, so (11) becomes, for spheroids,\n\nFrom solutions of (26) one may plot curves for normalized torque- \nweight ratio. In Fig. 9 we show a curve for spheroids of Alnico to- \ngether with a curve for cylinders of Alnico borrowed from Fig. 5. \nWhile the spheroids show a somewhat better torque-weight ratio \nthan the cylinders, it is doubtful whether the advantage is sufficient \nto offset the added cost of shaping and mounting.\n\nThe sphere might have unique advantages mounted in a spherical \nenclosure for friction damping. For the sphere of diameter D, one may \nrewrite (6)\n\nThis is an expression for the total normalized torque that can be \npacked into a specified spherical enclosure. Solutions of (28) for \nvarious magnet materials are collected in Table III. The moment of \ninertia of the sphere is\n\nEquations (26), (28), and (31) define for the sphere the same \nnormalized quantities which are plotted for the cylinder in Figs. 5, 7, \nand 8. Solutions of these equations for various magnet materials are \nlisted in Table III. The combination of the table and the three figures \ngives all the information required to select the preferred material and \ngeometry for a specified application and to arrive at a quantitative \ndesign of the magnet.\n\nSatellites with spin stabilization introduce two magnetic measuring \nproblems\u2014measurement of \u201cdrag\u201d and measurement of residual mo- \nment. The \u201cdrag\u201d results from eddy currents induced in the rotating \nmetal shell of the satellite by the earth\u2019s magnetic field. The energy \ndissipated in these eddy currents must be derived from the rotational \nenergy of the satellite, and there results a decay of the spin rate. \nOne wishes to evaluate the rate of this decay to forecast when the \nspin rate will fall below the minimum required for stability. The \nmoment measurement is to detect any residual magnetic moment \nperpendicular to the axis of spin which will interact with the earth\u2019s \nmagnetic field to induce precession of the spin axis. After an accurate \nmeasurement this moment is canceled out by mounting in the satellite \na small permanent magnet of equal moment and opposite polarity. \nBoth measurements\u2014drag and moment\u2014can be made conveniently \nwith a specially designed coil array.\n\nThe drag test requires a reasonably uniform field over the volume \nof the satellite. A paper analysis reveals that this can be provided \nby an array of coils of reasonable size with a particular distribution \nof ampere turns. Two coils, each of radius r, and of N turns are spaced \n-+r/4 from an assumed zero point on a common axis. Two other coils, \neach of radius r, and 7N/3 turns, are spaced +r from the assumed\n\nzero along the common axis. The coils are connected in series to run \nat the same current so that the outer coils have effectively 7N/3 times \nas many ampere turns as the inner pair. The arrangement of coils and \nthe resulting distribution of field along the axis are shown in Fig. 10.\n\nIt was established by measurements that the region of uniform \nfield extended out radially from the axis to include a spherical volume, \nthe radius of which is roughly two thirds the radius of the coils. \nHence, an array of coils five feet in diameter easily provided uniform \nfield over the volume of the satellite. (If a conventional Helmholtz \narray were used the coils would have to be about 10 feet in diameter \nto achieve reasonably uniform field over the same volume.) This \narray was mounted on a turntable and rotated around the satellite \nwhich was supported by a calibrated torque suspension. From the \nresult of this drag measurement it was possible to calculate the rate \nof decay of satellite spin in the earth\u2019s magnetic field.\n\nThe magnetic moment perpendicular to the spin axis of the satellite \nwas measured by rotating 1t within a coil array similar to the one \nused for drag tests except that the windings were connected to an inte- \ngrating fluxmeter. One reasons intuitively that if a magnetic object is \naligned parallel to the axis of the coils and rotated 180\u00b0, it will give a \ndeflection of the integrating fluxmeter proportional to the moment, \nand that the proportionality constant will be unaffected by the position\n\nof the magnet in the array as long as it is within the volume in which \nthe array produces uniform field. This intuitive reasoning has been \nconfirmed by various measurements. The proportionality constant \nfor the array is established by calibration with a small air-coil, of \nknown NIA for which the moment can be calculated from (2). A \ntwo-to-one scale down of the array has proved to be convenient for \nbench measurements of magnetic moment of small magnetic objects.\n\nVIII. ACKNOWLEDGMENT \nThe author is indebted to Mr. L. Rongved for stimulating discus- \nsions of the dynamics of the orbiting satellite.\n\nAPPENDIX \nDefinition of Symbols \nThe following letter symbols have been adopted for use in this paper.\n\nRs = shortening ratio, from recession of poles. \nLT = torque between magnet and perpendicular field. \nT = period of mechanical oscillation.\n\n1. Okoshi, Takanori, Demagnetizing Factors of Rods and Tubes Computed from \nAnalog Measurements, J.A.P., 36, August, 1965, pp. 2382-2387.\n\nWe consider, from a number of different viewpoints, the tensor coefficients \nwhich describe second harmonic generation, optical rectification, and the \nPockels or linear electro-optic effect in acentric crystals. Stationary per- \nturbation theory 1s used to calculate the low-frequency limit of the intrinsic \nelectronic nonlinearity neglecting all effects due to local fields or lattice \npolarization. Solid methane is used as an example and the result used to \nestimate the coefficient in hexamethylene tetramine. The calculated result \nis within a factor of 2 of the expervmental figure. The method 1s susceptible \nlo further refinement and, since rt requires only a knowledge of ground \nstate wave functions, and is essentially very simple, it appears to offer a \nuseful approach to the calculation of the coefficients.\n\nThe classical anharmonic oscillator model 1s briefly covered and the model \nis related to a quantal treatment. We find that the anharmonic potential \nused in the model is directly related to the actual crystalline potential. It \ncan also be related to the charge distribution in the electronic ground state.\n\nLocal field corrections and the effects of lattice polarization are presented. \nThese alter the nonlinear properties in a sumple and obvious way, but one \nwhich has been misunderstood in some of the literature.\n\nOur results form a theoretical background to Muller's empirical rule \nrelating the nonlinear coefficients to the linear susceptibilities. An extensive \ntable of Muller-reduced tensor coefficients collated from the published litera- \nture is presented.\n\nFinally, we draw together some of the threads of the previous sections. \nAn appendix deals with the vexing question of definitions.\n\nSecond harmonic generation, optical rectification, and the linear \nelectro-optic effect are particular aspects of a process in which two \nfields, H\u2019e\u2019* and E7ve'\u2122', generate a polarization\n\nOur concern is with the tensor coefficients d%\".\" which (Nye\u2019) necessarily \nvanish in centric (centrosymmetric) crystals and which, in acentric \ncrystals, are subject to symmetry restrictions, (Kleinman\u2019) which often \nleave only one or two independent components.\n\nIixperimentally, the values of the allowed components of d in different \nmaterials and at different frequencies range from about 2.10\u00b0*\u00b0 esu \n(cm/stat volt) to about 6.10\u00b0\u00b0 esu. This range may be contrasted with \nthe linear optical susceptibility ~ which is between 0.1 and 0.3 for the \nvast majority of materials and only quite exceptionally exceeds unity. \nThere 1s, however, a connection between the tensor d and x which is \nexpressed by an important empirical rule due to Miller.* If we write \ndh\u201d as\n\nty = XeexGXeAsa \u2019 (2) \nwhere x,\u00b0 1s the 227 component of x at a frequency a, and if we have chosen \n& principal axis system for x, then the allowed components of A;;, \nfor all effects and all materials are similar in magnitude. We shall see \nin a later section that for very many materials in both the visible and \n10 w region of the spectrum (Patel*), A;;, is near 3 X 10\u00b0\u00b0 esu. No \nmaterials with A above 20 X 10\u00b0\u00b0 esu have yet been found and very \nfew are known to have a value below 0.2 X 10\u00b0\u00b0 esu. In the case of \nNH,H,PO, where the best measurements of s.h.g., optical rectification \nand the electro-optic effect are available (Francois,\u2019 Ward,\u2019 Carpenter\u2019) \nthe value of A,.; from all three effects is 3 X 10\u00b0\u00b0 esu within the experi- \nmental error of 15 percent. The fact that s.h.g., a purely optical effect, \nleads to the same value of A as rectification and the electro-optic effect \nindicates quite clearly that the basic mechanism of the nonlinearities \nis common to all three effects and must therefore reside in the electronic \nmotion of the system. In the next section, we shall concentrate on this \naspect of the problem and neglect the effects of local fields and lattice \npolarization.\n\nA number of authors (see Section IV for references) have given quantal \ntreatments of the optical nonlinearities whose end result 1s an expression \nfor the coefficients d*?,\u201d in terms of sums of rather inaccessible matrix \nelements. Useful as these expressions are, in establishing some of the \ngeneral properties of the coefficients, they are not a practical step on the \nroad to calculating the coefficients from other empirical quantities. At \nthe other extreme, the classical anharmonic oscillator model has been \nused to demonstrate some of the qualitative features of nonlinear be- \nhavior (see Section III). This treatment, though simple and appealing, \nsuffers from the defect that the relation between the parameters of\n\nthe model and those of the real system is obscure. In Section IV, we \nshall remedy this defect and show that the two approaches are closely \nrelated.\n\nFirst, however, we give an approximate method of calculating the \nlow-frequency limit of the coefficients from stationary perturbation \ntheory in a form in which it has been successfully applied to the linear \nproperties of n electron atoms (see, e.g., Dalgarno\u2019).\n\nAt low frequencies, 1.e., well below any electronic resonance we can \nuse stationary perturbation theory to calculate, to arbitrary order in \nthe applied field # the energy W of the ground state. The polarization \nis then given by\n\now \nP,;=- al, (3) \nWeshall assume that we are dealing with a crystal containing N identical \natoms or molecules in unit volume whose individual ground state energies \nare w, so that W = Nw.\n\nm=1 \nis the dipole moment operator of the molecule, and the sum extends over \nall n valence electrons. We can neglect the core electrons because of \ntheir high binding energies. If we expand w in increasing order in # as\n\nthe term w, gives the permanent dipole moment of the molecule, w, \ngives the linear susceptibility and w; gives a polarization quadratic in \nFE which leads to the desired nonlinear coefficients. The electric field \nwill perturb the state function y and we shall write the perturbed func- \ntion as either\n\nKnowledge of y or | ) to first order in # is sufficient to determine w, , wz , \nand ws; for\n\nMoreover, the correct value of y, or | 1) is determined by the require- \nment that it minimize w. Thus, we can obtain | 1) by a variational \nprocedure and the only element of choice left to us is that of the trial \nwave function.\n\nMinimizing w is equivalent (see Dalgarno and Lewis\u2019) to the simpler \nproblem of minimizing\n\n| ABA |0) ~~~ (0| AAA {0) \nThe unperturbed Hamiltonian of the system is of the form \n__l Sy: \nA, 2 Im pe Wa, + V. (13) \nand so, in the denominator of (12), \nHh = hl, + eB ie a (14) \nThus, \n\u201c eh\u2019 \n(0| bith 0) = \u2014\u2014= fy, DO Btw B-DVadedr, \u2014 (15)\n\nwhere dr is an element of configuration space and we have used H | 0) = \n0. Equation (15) can be written as\n\nwhere the discarded first integration part vanishes at the limits, if \nthese are infinite, or if they are the boundary of a cell in periodic lattice, \nprovided only that H does not vary appreciably within a cell (dipole \napproximation).\n\n(0| AHA |0) = I \na somewhat unfamiliar form of the sum rule. If, on the other hand, \nwe are dealing with overlapping molecules in a periodic lattice, the \nvariational problem is to minimize the contribution to w from a single \ncell of the lattice. Thus, in (11) and all succeeding equations, the inte- \ngrals implied by the expectation values are to be taken only over the \ninterior of a cell. This will also apply to all integrals involved in evalua- \nting w, = (1 |h| 0) and w, = (1 | # | 1). In this case (18) remains un- \nchanged. This can be shown to be a general consequence of time reversal \ninvariance and the commutation rule\n\n4 (r2\\2 \name (As (24) \nFor the H atom, this gives a = 4a} instead of the correct value 4.5a;, \nwhile for the helium atom, taking an effective nuclear charge Z = \n27/16 gives 1.8 X 10\u00b0*\u00b0 ces. The experimental value is 2.1 X 10~*\u00b0 ces. \nIn general, (24) is a lower limit to a, if we evaluate (X\u2019) correctly as the \nexpectation value of the mean square moment of all the electrons. If \nthe electrons are uncorrelated\n\nwhere (\u00a3\u2019) refers to one electron. We used this procedure in helium since \nthe two electrons are in orthogonal spin states and are automatically \nuncorrelated. In more complicated atoms correlation exists and almost \nalways results in\n\n(X*) < n(\u00e9\u2019) (26) \nsince electrons repel each other. Thus, while (24) is a lower limit we \ncannot say anything about the sign of the error in\n\nWe note, in passing, that, in a solid with overlapping molecules, a the \npolarizability is large. This leads to an element of instability in the situa- \ntion for as a increases the screening of the coulomb potential becomes \nmore effective and the electrons less localized leading to a further in- \ncrease in a and eventually metallic behavior. For this reason, most \nmaterials, which are not regular insulators, are metals. Those rare \nmaterials which have values of Na appreciably greater than 0.3 (n > 2.2) \nowe their existence to a rather delicate balance of forces. \nThe third-order energy is\n\nNONLINEAR OPTICAL COEFFICIENTS 919 \nWith N molecules in unit volume this gives a nonlinear coefficient \n2S SS ey (31)\n\nEquation (31) is the central result of this section. It expresses d;;, in \nterms of the linear (corrected for local fields) susceptibility x and a \ncubic moment (third-order semi-invariant) of the electronic distribution \nin the ground state.\n\nIf we neglect overlap and, for simplicity, also assume that the electrons \nare uncorrelated so that 7';;, = nt,;;, where \u00a2;;, refers to a single electron \nwe have\n\nand 7';;, 1s now apart from numerical factors the octupole moment of \nthe charge distribution. If the electron density in the molecule is p(r)\n\nIf we account for local fields through a Lorentz correction the correct \nvalue of x to insert in (33) is obtained from\n\nAt first sight (83) seems to imply that d is proportional to x in conflict \nwith Miller\u2019s rule. However, x depends on N/n(R?Y + nN\u2019 and N is \ninversely proportional to (r\u2019)? so that x = nr, while d = r*. Thus, d is \nin fact more nearly proportional to x* than x.\n\nWe now consider as an example, the tetrahedral molecule methane \nCH, , which crystallizes in the tetrahedral space group F43m with a \nlattice parameter ~6 A and a molar volume of 32 ces. If we take Car- \ntesian axes along the sides of the cubic cell, the bonds point in the 111,\n\nand tetrahedrally related, 111, 111, 111, directions. From symmetry, \nthere is only one independent component of d;;, , in which all the sub- \nscripts are unequal.\n\nThe shortest c-c distance is 4.2 A and from the size of the free molecule \nwe conclude that overlap is unimportant.\n\nTurner, Saturno, Hauk and Parr\u2019\u2019 haveused one center wave functions \nto calculate the electronic density p in the molecule. F'rom this we can \nobtain f,.3 using (34).\n\nand this is not very sensitive to the limits of integration. The experi- \nmental molar susceptibility of CH, is 1.6 ces and so x = 0.05. Thus, if \nwe neglect correlations between the eight valence electrons we have \nfrom (33)\n\nIn a similar way, neglecting correlations, we can calculate the molar \nsusceptibility from (27). Turner et al\u2019s charge density leads to\n\nand so, with cight valence electrons, we obtain a = 6.5 X 10\u00b0\u201d and \nx = a molar susceptibility of 3.9 ccs, rather over twice the experimental \nvalue.\n\nThis is a clear indication that the electrons are correlated. However, \nthe correlation enters twice in x but only once in d (since we have ex- \npressed d in terms of the experimental x). Thus, d problaby lies between \n2x 10\u00b0 and 3 X 107\u00b0 esu.\n\nTo see whether 3 X 107\u00b0 esu is a reasonable value for d,23 we compute \nthe Miller reduced tensor d/x* = Ayo3 = 24 X 107\u00b0 esu.\n\nThis is quite exceptionally high. Most materials have allowed com- \nponents of A near 3 X 107\u00b0 esu and only one coefficient in LiNbO; \n(9 X 10\u00b0) and Ajo; in hexamethylene tetramine (15 X 107\u00b0) approach \nthis value.\n\nHowever, we believe it is in fact not far wrong. In most materials \ngeometric factors conspire to reduce d by various factors of cos @ and \nthe atomic groups are in the first instance less aspherical than CH, . \nIn CH, the effects of every electron are directly additive.\n\nHexamethylene tetramine (HMT), the other exception to Muiller\u2019s \nrule, is, like methane, a tetrahedral molecule N,(CH.), in a tetrahedral \n143m crystal. The 4 nitrogen atoms form the 111 1II, 111, 111, corners\n\nof a regular tetrahedron and the CH, groups occupy the edges but the \nN-C-N bonds are bent outwards in such a way that all the angles are \nvery nearly tetrahedral. The carbon atoms occupy the six sites 2, 0, 0 \netc. (see Kitaigorodskii\u2019\u2019).\n\nThe refractivity of the molecule as a whole can be very satisfactorily \naccounted for by a system of additive bond refractions. (See Lelevre\u2019\u201d \nfor a review of bond refractions.) The three basic units are 12 C-\u2014H \nbonds pointing in the 1, 1, 1, and related directions, 4 nonbonding orbitals \non the nitrogen atoms pointing in the 111 and related directions, and \n12 N-C bonds in the 111 directions.\n\nSince the refractivities are additive, these units appear to act inde- \npendently in determining the molar refractivity. Le Fevre (loc. cit.) gives \nvalues of R = (47/3)La (where L is Avagadro\u2019s number) of 2.8 ces for \neach unbonded nitrogen pair, 1.65 ces for each C\u2014H bond and 0.62 ces \nfor each N-C bond. Thus, the N\u2014C bonds make a rather small con- \ntribution to x, and probably even less to d since they have an approxi- \nmate inversion centre at the centre of the bond (C and N are similar \natoms as compared with C and H). We therefore neglect them.\n\nThe 12 CH bonds in the 111 direction are roughly equivalent to 3 \nmethane molecules in the molecular volume 105 ccs, and further the \nelectrons will be less correlated than in methane. Thus, their contribution \nto dy23 18\n\nTo calculate the effect of the nonbonding nitrogen electrons we assume \nthat they occupy SP* hybrid orbitals directed along 111 etc. with Slater \nradial wave functions Ar exp (\u20142.5r/2a,). It is then straightforward \nto show that for one electron\n\nThus, the total value of dis; is \u20144.5 X 10\u00b0-\u00b0 esu. This could be slightly \nincreased by the effects of atomic overlap, and possibly by contributions \nfrom the N-C bonds. It could be either increased or decreased by elec- \ntron correlations on individual CH, groups. The experimental values\n\nfor the electro-optic effect Heilmeyer\u2019\u00ae and second harmonic generation, \nHeilmeyer, Ockman, Braunstein, and Kramer,\u2019* when corrected for local \nfield effects using a Lorentz factor, both give d = +8.2 X 10\u00b0\u201d esu. \nThus, our calculation is within a factor 2 of the observed value.\n\nThis method is therefore capable, in simple cases, of predicting the \nmagnitude of d rather successfully. Moreover, the experimental value \nof d for HMT suggests that we were correct in assuming that CH, \nwill have an anomalously large reduced tensor Aj); .\n\nThe fact that the division of a complex molecule such as HMT into \nsimple components leads to a reasonable value for d leads us to hope \nthat a similar procedure will be possible in other cases. It might then \nbe possible to assign empirical values of d to basic components such as \nthe C-H bond or the N: nonbonding pair, and to combine these ad- \nditively (with a proper attention to geometry) to predict the values of \nd for even more complex molecules. This would not be surprising since \na similar procedure (see LeFevre loc. cit.) works very satisfactorily for \nthe linear susceptibilities.\n\nIt is then obvious that large nonlinear effects will only result, if the \nmolecule contains polarizable groups disposed in an arrangement which \nresults in a ground state of, far from inversion, symmetry. The large \nvalue of A in HMT results from the fortunate coincidence that the \nmost polarizable components are themselves strongly asymmetric and \nso oriented that their effects are additive. The much smaller values of \nA commonly observed can then be explained as partly due to no group \nin the crystal being quite so asymmetric as N: or CH. in HMT and \npartly due to unfavorable geometric relations between the groups. I\u2019or \nexample, if our approach is correct we should expect the analogous \ncompound adamantane (CH),(CH:), in which the nitrogens are re- \nplaced by CH groups with the CH bond along 111, etc. to have a djo3 \nappropriate to 2 (=3 \u2014 1) CH, molecules in 105 ccs, ie., di23 2 2 X 107\u00b0 \nesu or about half the value for HMT.\n\nExceptionally small values of A will occur in materials where most \nof the molecule possesses local inversion symmetry, so that only a \nfraction of the molecule contributes to d, while the whole molecule \ncontributes to x. We shall consider an example of this in a later section.\n\nOverlap between adjacent molecules is neccessarily bound to lend \nfurther uncertainty to the calculation in materials with a pronounced \nband structure, but it seems possible that rough approximations should \nbe obtainable from, for example, the relation between bandgap and \nthe corresponding separation in the isolated atoms. In fact, since what \nwe actually require is 7',;;,/n, which, if the electrons are uncorrelated,\n\nFinally, we may remark that very much better approximations to \nd;;, can obviously be made if we know the ground state wave function \nexplicitly and also use more sophisticated trial wave functions in the \nvariational calculation. It is at first sight surprising that a knowledge \nof the ground state wave function alone is sufficient to determine x and \nd which, in the more usual treatments involve the properties of excited \nstates. However, we should remember that a knowledge of the exact \nground state wave function is, except in pathological circumstances, \nsufficient to determine the unperturbed Hamiltonian; thus, the whole \nspectrum of states.\n\nAlthough the considerations of the preceding section are sufficient \nto determine the magnitude of d at low frequencies, they offer little \nguide to the variation of d with frequency and, if recast in terms of time \ndependent perturbation theory they tend to lose their attractive sim- \nplicity. In the next section we shall show that a more familiar form of \ntime dependent theory leads to results which can be represented in \nterms of a classical anharmonic oscillator model. Here, we discuss the \nproperties of the model itself.\n\nWe assume that unit volume of the material contains N optical elec- \ntrons which move in a potential\n\nV = 3mQja; + V i jnUsXjXy, , (43) \nwhere a sum over repeated subscripts is implied. The potential V;;,\n\nThere will be a similar response to a field H%e'\u201d\u2019 and, if we introduce \nthese responses back into the nonlinear term in (44) we obtain a re- \nsponse at the sum frequency a = 6 + \u00a5y given by\n\n3V 55,Ne\" 1 1 1 \noe faeearaae | \nThus, the symmetry of d;;, mimics that of V;;, if we neglect the res- \nonance denominators. |\n\nThe linear susceptibility obtained from (45) is the familiar expression \n\u00bb Ne 1\n\nIf we assume that V;;, is electrostatic in origin its order of magnitude \nwill be e\u2019/d* where d is an atomic spacing and we shall also have Nd\u2019 ~ 1. \nThus,\n\nWith d equal to 2 A this is 2.5 X 107\u00b0 esu, about the mean value of A \nfor most materials. In a later section, we shall give another estimate of A.\n\nThe potential V;;,v;7;2, distorts the shape of the ground state of \nthe harmonic oscillator and as a result the system acquires a cubic \nmoment \u00a2;;, [defined in (82)] which we now calculate.\n\nLet | 0) represent the unperturbed ground state wave function in \nthe absence of the anharmonic term and | 7p) be an excited state, then \nthe perturbed wave function is\n\nThe expectation values of even operators such as (27), (v;a7;) are un- \naltered by V, while the expectation value of an odd operator such as \nXU; OY \u00a9;2;X, 18 given by\n\nDp hw, \nIt will suffice if we calculate t,;;, with \u00a2 # 7 # k. Since (;7;) = 0 if \nt ~ i we only require (7;%;x7,) and contributions to this come only from \nthe 6 = 3! terms in V with2 + 7 \u00a5 k. The only state which contributes \nto the sum is |p) = | 1, 1, 1) with an energy A(Q, + Q, + Q;). The matrix\n\nholds for all the components of \u00a2;,; . \nIf we substitute this relation in (47) and take the limit as afy \u2014 0 \nwe obtain\n\nThis is twice the value obtained in (83) because there we treated {,;, \nas a fixed property of the ground state which was then perturbed by E; \nwhereas here we have considered an even ground state perturbed by # \nand a fixed potential.\n\nThus, if the real system has a cubic moment #\u00a2;;, in the ground state, \nthe equivalent anharmonic oscillator model requires an anharmonic \npotential\n\n3 a4 \ntik = ee bs in (56) \nand this will result in a cubic moment \u00a2/,, = 3\u00a2;;, in the oscillator ground \nstate.\n\nin molecular crystals of strongly covalent compounds such as CH,. \nBut, in ionic crystals it may be more sensible to consider the ions as \nspheres perturbed by a crystal potential V\u2018,;,. In a later section we \nshall see that there is a simple relation between the model potential \nand V\u00a2;, -\n\nThe classical anharmonic oscillator model has previously been used \nby Bloembergen,\u2019\u2019 Garrett and Robinson\u201d\u00ae and Kurtz\u2019\u2019 to give a qualita- \ntive account of nonlinear phenomena. The latter authors also discuss \nin some detail its relation to Miller\u2019s rule.\n\nObviously, the model is the nonlinear analogue of the classical har- \nmonic oscillator model used with such success for the last 60 years in \nthe discussion of linear behavior such as dispersion, and, just as the \nharmonic model is directly related to the results of a quantum mechan- \nical treatment, we may expect the anharmonic oscillator to have a \nsimilar basis. In the next section we explore this relation.\n\nDucuing and Pershan,\u201d Butcher and McLean,\u2019\u00ae Kelley,\u2019\u00ae Cheng and \nMiller\u201d\u2019 and Ward\u201d have given rigorous quantal treatments of optical \nnonlinearities in solids. We select an expression due to Armstrong, \net al (loc. cit.) which expresses the nonlinear coefficients in terms of the \nenergies fw, of excited states and the matrix elements (0 | 2; | 7), \n(p | x; | q), ete. of the dipole operator between states. The ground state \nis (0 |.\n\nThis expression is valid, cither for an assembly of N isolated atoms \nin unit volume or, in the dipole approximation, for a real solid where \nthe wave functions overlap. In the latter case, the solid must be divided \ninto cells of the periodic lattice, and N is then the density of cells, while \nthe matrix elements are to be evaluated only over the interior of a cell. \nThe periodicity of the lattice ensures that contributions from parts of \nthe wave function outside a cell cancel in the crystal as a whole.\n\nTo avoid a plethora of subscripts we let each of x, y, and z serve to \nrepresent any one of the components and we can then write the expres- \nsion for d as\n\nThis expression vanishes if the states | 0), | p), etc. have adefinite parity, \nits value therefore depends on the existence of matrix elements whose \npresence is contingent on the absence of inversion symmetry. For this \nreason, it 1s almost impossible to make an informed guess about its \nmagnitude or behavior.\n\nA familiar approximation to x is obtained if we note that in (58) the \nvariation of the summand with | p) is almost exclusively due to the \nmatrix elements. These not only obey selection rules, but also decrease \nrapidly in magnitude as the state | \u00bb) increases in energy, and therefore \noverlaps the ground state less and less. For example, in the H atom with \na 1S ground state the matrix element x,, vanishes unless p is one of \nthe states 2P, 3P, etc. Moreover, as we go from the 2P state to the 8P \nstate x,,7,, decreases by over a hundredfold. At the same time, w, \nchanges by less than 30 percent. Thus, except near a resonance, we can \ntreat w, as a constant 2, somewhere near the first allowed transition and \nwrite (3.2) as\n\nwhere the primed sum excludes p = 0. Now \nDe\u2019 ZosYv0 = Dy Lose \u2014 LocYoo = (LY)oo \u2014 LooYoo \n= ((\u00ab \u2014 (x))(y \u2014 (y)) (60) \nwhere a ( ) denotes a ground state expectation value. Thus,\n\nWe shall not pursue the further manipulations of (61) using the sum \nrule which lead back to (27) but we remark that in many cases a form \nsuch as (61) for x, involving a single Sellmeier or classical oscillator \nterm, gives an excellent account of optical dispersion, and that when\n\napplied to the hydrogen atom, with iQ set equal to 2e\u2019/a, the 1S \u2014 2P \nenergy, it leads to a value of x at low frequencies\n\nBefore we can adopt a similar procedure with the nonlinear coefficient \nwe must first satisfy ourselves that there is no essential difference be- \ntween a sum with three matrix clements and one with two. In x all \nmatrix elements terminate on | 0) but in d it is quite possible in a term \nsuch a8 %o,Ype2ao \u00bb With p gq corresponding to highly excited states, of \ngreat spatial extent, that the term y,, may be large enough to compensate \nfor the smallness of x,,2,. . If this were the case it would be possible for \nthe exact value of the sum to depend critically on cancellations between \nlarge terms involving highly excited states. The removal of the frequen- \ncles w, , etc. as a single average would then have disastrous results on \nthe sum.\n\nWe will advance three arguments why this is unlikely. Consider first \nan even higher-order calculation, that of the fourth-order Stark shift \nof the ground state of atomic hydrogen due to a field F. In atomic \nunits this is given by an exact calculation (Dalgarno\u2019\u2019) as\n\nW? = \u201433359* ~ \u201456F\". (62) \nWe can also express W\u201c (Dalgarno, loc. cit.) as \n? \nWw? er \u2014 > ye 4 L opt pnglar& ro a by Lon po a come Pe. \np q r WWW> p Wp Wy\n\nThis is close to the correct result (36), but despite the fact that we have \ntaken the lowest possible value of \u00a9 it is too small. This is a clear indica- \ntion that some cancellation of higher terms, which we have aggravated \nby our cavalier treatment of w, , etc., is occurring. This is not surprising \nfor, if in the triple sum we consider the lowest possible sequence of levels \n1S 2P 2S 2P 18 for which wo, = o, = w, = Q the product of the matrix \nelements is 5 a.u. while for the sequence 1S 8P 8S 8P 1S the product\n\nis 10 a.u. made up of a contribution of 0.0033 from the two 1S 8P \nelements and 3000 from the 8S 8P elements.\n\nHowever, this is not quite so serious as it appears, for in a real solid \nno matrix element can exceed the linear dimensions of a cell say 5 a.u. \nso that the product in the low transition would remain at 5 a.u. while \nthe product for the upper transition would be reduced to 0.08.\n\nOur final argument is empirical. If cancellations between large terms \nare critically important, the relevant feature of our procedure is the \nchange in the ratio w,/w, it causes for highly excited neighboring states. \nIn hydrogen the ratio of the 1S \u2014 8P energy to the 1S \u2014 7P energy \nis 1.005 and we replace this by unity. In a time dependent theory res- \nonance denominators appear, and if the sum is really so critically \nbalanced, we expect the observed quantity, in this case the hyper- \npolarizability, to vary rapidly with frequency when \u00bb ~ 0.005 0\u201d, \ni.e., at a frequency 10 times lower than the first absorption edge. In \nnonlinear optics, no such variation is observed until one of the frequen- \ncles approaches much more closely (about 70 percent) to the absorption \nedge (Chang, Ducuing, and Bloembergen\u201d*).\n\nTaken together these arguments give us reasonable grounds for \nhoping that the sums will not bite us if we remove w, , etc. from under \nthe summation sign.\n\nIn the sum in (57) there is no restriction on 7 or q, in particular terms \nwith either p = 0 or g = 0 occur. These will lead to trouble if we attempt \nto approximate the sums as they stand. We therefore first segregate \nall such terms. Let { } denote the entire summand in (57), then\n\nW, \u2014~ VY Yor Wr \u2014 Q@ \na Zortro _ B oe ATi ra \nYoo B pie: ws __ eine vidos 5 2 \u2014 Bp \n8 Ure ye cseoeeee : \nZoo 7 00 i eee 65 \na yea pt ued Ei fet (65)\n\nThree single sums remain unprimed, but because a = 8 + vy the terms \nwith r = 0 cancel and so we may regard all the sums as primed.\n\nWe can now remove w, , w, and w, as a single average Q, and this leads \nto an expression containing terms such as\n\nEach of the sums on the right is now a ground state expectation value. \nWhen all the terms are collected together we obtain\n\nin terms of the, by now, familiar cubic moment 7\u2019,,,. This expression \nbears an obvious resemblance to (61) for x.\n\nOur expression (66) or (67) would be very nearly exact if all the optical \nlevels had very nearly the same energy. It would then correspond to \nthe fictitious two level system (see Refs. 15, 16, 18) often used to oblit- \nerate some of the intractable features of (57). Unlike this model, how- \never, our expression retains the geometry of the system implicit in the \nselection and sum rules.\n\nEquation (66) is possibly valid up to a frequency where one of a, 8, \nor y approaches the first allowed transition frequency. At somewhat \nlower frequencies, it is legitimate to drop the term By \u2014 a\u2019 in the numer- \nator. This then allows us to make a further generalization at no increase \nin complexity.\n\nBy removing w, and w, from (57) as a single average we have tacitly \nneglected the possibility that the system might be birefringent. We \ncan remedy this by noting that in (57) each frequency , or a, \nis uniquely associated with a matrix element such as x,, or 2,. which \nterminates on the ground state | 0) and therefore also appears in x. \nThus, we can consistently introduce three averages 2, , 2, , and Q, as- \nsociated with correspondingly polarized transitions. If we follow this \nprocess through all its tedious ramifications, we find that, except in \nthe term By \u2014 a\u2019 which we are omitting, it leads to the surprisingly \nsimple result that D(Q,a,8,7) is replaced by\n\nIf we compare this with the result for the classical anharmonic oscillator \nobtained by combining (54) with (47) we see that they are identical \nexcept for a factor 2 which once again arises because in one case we \nassumed that T,,, was a fixed parameter while in the other it was V,,, .\n\nWe now see that the classical model is equivalent to the quantal \ntreatment, except near a resonance, in the following sense.\n\nIf we construct the model, by choosing Q, , 2, , and 2, to give the \ncorrect linear properties then we must choose the anharmonic term in \nthe potential to produce a cubic moment in the ground state of the model \nequal to $ the corresponding moment in the real system. The dynamical \nproperties of the two systems are then equivalent and the model can \nbe used to treat more complicated systems where the quantal treatment \nis too difficult.\n\nWe now consider the relation of V/,, to the actual potential respon- \nsible for the existence of 7',,,. Obviously, the relation is obtained by \nrequiring that both potentials yield the same cubic moment, one in the \nmodel, the other in the real system. In this case, there will be no factor \nof 2.\n\nFor simplicity, we consider only a system which is isotropic before \nthe application of the anharmonic potential. Further, we restrict our- \nselves to atoms in which there is only one valence electron. Our results \nwill, however, be directly applicable to atoms with more electrons if \nwe can neglect electron correlations.\n\nWe already have an expression for the oscillator (54) which we can \nwrite as \nVou\n\nV = Vi;.0;2;2,, the crystal potential, satisfies Laplace\u2019s equation. \nWe will consider a more general potential of the form\n\nwhere P\u2019?} is an associated Legendre polynomial normalized to unity. \nThis potential satisfies Laplace\u2019s equation only if n = lL.\n\nIf the unperturbed ground state wave function is Y the first-order \ncorrection y, due to V satisfies\n\nSince 7 is an odd moment we need only consider odd terms in V (in \nfact only / = 1 andl = 3) and for these /, vanishes since yp has definite\n\nparity. \nWe let \nvi = flo (73) \nand then \n(Hy \u2014 Eo)fvo = \u2014Vo \nbut, since | \nHo -* Wa \nthis leads to \nVif + 2VP-V log vo = V. (7-4)\n\nand in this new ground state we can easily evaluate expectation values \nsuch as\n\nIn evaluating 7;;, we shall need (z;), (x7), and (x;7,;2,). The even \nmoments are unchanged by V and we obtain the odd moments by ex- \npanding x; and x;2;x, in terms of Legendre polynomials and powers \nof r.\n\nWe omit most of the gruesome details of the calculation, and further \nrestrict V to contain only terms of the type\n\nThe term in V;} which does not satisfy Laplace\u2019s equation is necessary \nto obtain the most general form of the cubic part of the potential \nV ;;42%,2;2, . This contains 10 independent parameters while P\u2019} has only \n7. The missing 3 are supplied by P\u2019.\n\nand it is easy to check that the coefficients of S; and X,; vanish, so that \nwe recover (70).\n\nIf V satisfies Laplace\u2019s equation S; = O and in the absence of an \ninternal field X; every component is given by\n\nThus, in this case 7';;, and the reduced Miller tensor A;;, have the same \nsymmetry as V;,;,. Therefore, since S; = 0 we have\n\nAss3 = \u20142Ag11 . (94) \nThis relation is rather well obeyed by the coefficients for the 6-mm\n\ncrystals listed in the Table I. Signs are available only for the electro-optic\n\ncoefficients and so the s.h.g. results represent moduli only. References \nto the experimental data are given in conjunction with later tables. \nIn the case of the electro-optic data, the experimental figure is for Aj13 \nand we have assumed that Kleinman\u2019s rule (Kleinman\u2019) holds and that \nthis 1s equal to A;,,. Except for s.h.g. in ZnO the ratio is \u20142 within \nthe experimental error.\n\nand the expected ratio is +3. In crystals where both terms occur in \nV with arbitrary strength, any value of the ratio is possible. This is \nobserved in the ferro-electric crystals BaTiO, ratio +4 and LiNbO, \nwhere it is +1.7 for the electro-optic effect and +11 for s.h.g. It is \nperhaps somewhat surprising that the ratio is so exactly \u20142 in the \n6-mm crystals since this is a polar point group and an internal field \nX3 1s not forbidden by symmetry.\n\nIf V does not satisfy Laplace\u2019s equation, (it need only satisfy Poisson\u2019s \nequation) there is no direct relation between the components of T,;, \nand those of V;;, even in the absence of a field X;, although, since \nxyz 1s a spherical harmonic, we still have\n\nT 123 = 358333 Vins : (96) \nHowever, since the coefficients of S; vanish for the harmonic oscillator\n\nwe may expect them to be small in other cases. We gain some support \nfor this view by considering the hydrogen atom for which\n\nThe coefficient of V;;, in each term of 7';;, is then \u2014315/16(a/ha,) \nwhile the coefficient of S; in 7';;; is a factor 23/200 smaller. In 7;;; it \nis 23/600 smaller. Thus, in the absence of X; the non-Laplacean terms \nin V cause no more than an 11 percent departure from the relation\n\nSince we expect the dominant terms in V to satisfy Laplace\u2019s equation \nit appears that 7';;,, A.;, and the model potential V/,, will be very \nnearly proportional to the corresponding terms in V.\n\nThe potential V\u2019 required in the model is related to the crystal \npotential by\n\nTor a hydrogen atom this gives V/,, ~ 5V;;, thus, insofar as real atoms \nbehave like hydrogen atoms, a model with the same spatial extent \na = a, and the same first allowed transition w,  w, will require a \npotential roughly five times as strong as the actual potential. This \nreflects the obvious fact that a harmonic oscillator is a stiffer system \nwith more sharply localized (Y ~ e7\"\u2019) wave functions than an atom \n(y ye\").\n\nWe have now shown that, with an appropriate choice of parameters \na classical anharmonic oscillator model is a very good approximation \nto the intrinsic electronic nonlinearities of real systems.\n\nIn the next section, we use the model to consider the effect of lattice \npolarizability which we have so far neglected.\n\nWe have already remarked in the introduction that the seat of the \nnonlinearities resides in the electronic motion. It is, however, consider- \nably modified by local field corrections and in the case of optical rectifica- \ntion and the linear electro-optic effect by lattice polarization.\n\nMiller\u2019s rule states that d?{\" is proportional to the product of the \nobserved linear susceptibilities X 4 , etc. at the appropriate frequencies. \nIf one of these is de we are to take the actual de susceptibility and not \nthe extrapolated long wavelength limit of the optical susceptibility.\n\nAt first sight, 1t seems plausible that this is simply the effect of in- \nternal fields, which cause the local field experienced by an atom to be \ngreater than the applied field. We now examine this hypothesis and \nshow that it is inadequate.\n\nMicroscopic calculations yield the polarization of single atoms due to \nlocal fields. In the linear case, if we have N atoms per unit volume of \npolarizability a\n\nIn some cases the Lorentz value of ! = 47/3 is applicable and we then \nobtain the well-known relation between the refractive index n, or the \ndielectric constant \u00ab = n\u2019 and a.\n\nIn nonlinear optics the two driving fields Ef and E% are obviously \nmodified according to (105) but, as Armstrong, Bloembergen, Ducuing \nand Pershan ** have shown, there is a further factor in P. This arises \nbecause the nonlinear polarization\n\nproduced directly on the atoms, further polarizes the surrounding \nmedium. \nWe have \nPY = py + IT NoP?, (107) \nso that \na Dp; a a \nP= 1 \u2014 TNa = (1 + Ixii)pe . (108)\n\n\u2018Therefore, even if d does not vary with x, D will do so. \u2018This is, however, \nnot enough to explain the observed variation of D with x. For example, \nin semiconductors it is very likely that IT\u2019 is small if not zero and yet\n\nthe measured values of D appear to obey Miller\u2019s rule and be propor- \ntional to x\u00b0. Thus, the intrinsic coefficient d itself must have a similar \ndependence on_X,\n\nin terms of the measured susceptibilities (i.e., n\u00b0 \u2014 1), which is the con- \ntent of Miller\u2019s rule, and then use (104) to express D in terms of the \natomic polarizabilities we obtain\n\nThus, the reduced tensor is the same whether or not we apply local \nfield corrections as long as we do it consistently. To obtain a more or \nless constant value of A we must have d varying as a\u2019.\n\nWe repeat that optical nonlinearities have an electronic origin. \nElectrons in atoms do not move in a harmonic potential. Second har- \nmonic generation, which can only involve electronic motion, is much \nthe same in covalent organic materials, ionic crystals and ferro-electrics. \nLarge values of d\u2019* are associated exclusively with large refractive in- \ndices. Thus, nonlinearities in the ionic motion play a secondary role \nin nonlinear optics; however important they may be in determining \nthe ferro-electric properties.\n\nWe shall attempt to construct a model, just sufficiently general to \nexhibit the gross features of ferro-electric behavior, and show that it \nmodifies the nonlinear optical behavior exactly as predicted by Miller\u2019s \nrule. The model is not put forward as an explanation of ferro-electricity \nalthough it has a venerable past in that connection, but as a demonstra- \ntion that a simple system with singular dielectric properties behaves \nin a way consistent with Miller\u2019s rule.\n\nIn Fig. 1, we illustrate a moderately realistic one-dimensional model \nin which electrons of mass m are coupled to ions of mass JM in a lattice. \nForces act between like and unlike particles and of these by far the\n\nstrongest 1s K,,,,, which is responsible for the electronic optical spectrum. \nThe remaining forces determine the lattice spectrum. The important \nnonlinearities are associated with K,,,,. The linear behavior of this \nmodel is formidably complicated and we therefore assume that its \nsalient features are already evident in the much simpler model of Fig. 2.\n\nThe electron of mass m, = m is coupled to the ion of mass mz = M \nby the force constant k,. which replaces K,,. . It is anharmonic. The \nelectron and the ion are also coupled to rigid supports representing the \nrest of the crystal by forces k, and k, . It is as though we had gone di- \nrectly from the Born-Von Karman theory of specific heats to the Kin- \nstein theory without mentioning Debye.\n\nLet x, be the displacement of the electron of charge e, and z, that of \nthe ion of charge e, . We shall assume that the potential energy is\n\nso that the anharmonic term is exclusively associated with the \u2018\u201c\u2018atomic\u201d\u2019 \nbinding of the electron to its parent ion. It will be convenient to define \nVo, = \u2014Vi2. The equation of motion in a field H\u2019e*\u201d\u2019 is then\n\nTo obtain the sum frequency polarization due to two fields E\u2019e \nand E7e*\u2019\u2019\u2019 we substitute the linear responses back into the nonlinear \nterm in (114). The result is a nonlinear coefficient\n\nWe note first that if, as seems most reasonable, e, = \u2014e, then g(a) = \ng(8) = g(y) = 1. In any case at optical frequencies g(w) ~ 1 for all \nreasonable values of e./e, and at de\n\nke \u201cSe = ky \n(0) = 2 (123) \nke -++ (2) ky -+F kaa ae ) \n\u20acy C1 \nwhich is also near unity if e. \u00a9 \u2014e,. Thus, to all intents \nAm \u2014 Sos (124)\n\nThus, A is an intrinsic electronic property and the effect of ionic \nmotion is entirely contained in its effect on x. We note, however, that \nin some ferro-electrics, where the departure from inversion symmetry \nis both small and temperature dependent, A will also be temperature \ndependent.\n\nIf k,, ki2, and k, are all positive, the de susceptibility is greater \nthan the low frequency limit of x\u00b0 but not dramatically so. There is, \nhowever, no reason why one of these constants should not be negative. \nNegative compliances are familiar in classical mechanics, a well-known \nexample is the common automatic door stop which exhibits a positive \ncompliance as the door is first opened but a negative compliance when \nthe door is almost fully open. The force between atoms as a whole in a \nlattice exhibits a positive compliance but if we separate this force into \nnuclear-nuclear and electron-electron repulsion and nuclear-electron \nattraction, it is quite reasonable to assume that at the equilibrium \ndistance the latter component has a negative compliance.\n\nIt is immaterial which term in (113) we take as negative although \non physical grounds it seems most suitable to take k, and this is also \na convenient choice.\n\nthe system remains in stable equilibrium at x, = a = 0. \nThe natural resonance w, and we, of the system satisfy \nmmwiws = kiko + kiki + kokis (127)\n\nand so as 7 \u2014 1 one of these frequencies \u20140. At the same time the de \nsusceptibility (for simplicity we take \u2014e, = e, = e)\n\nbecomes infinite, while the low-frequency limit of the optical suscepti- \nbility remains finite.\n\nIf \u00bb exceeds unity there is a spontaneous polarization limited only \nby terms such as wx} which we have failed to include in \u00a2.\n\nAll this is reminiscent of ferro-electric behavior if \u00bb is temperature \ndependent and the Curie point corresponds to y = 1.\n\nThe inclusion of a term wx; in \u00a2 will, in fact, make 7 temperature \ndependent, for the effect of this term is to replace k. by an effective value \nfor low-frequency displacements\n\nwhere 2? is the mean square thermal displacement. As a result if 7, is \nthe value at 7 = 0 we have\n\nif we define 7\u2019, as the temperature at which y = 1. This is of course a \ncrude approximation to a Curie-Weiss Law.\n\nBy ascribing all the temperature dependence to changes in k,, it \nis obvious from (117) that X\u00b0 is temperature independent. For 7 to \nbe equal to unity, it is not necessary for \u2014k, to be of the same magnitude \nas ky. , all we require [see (126)] is that \u2014k, be near k, . Thus, from (117),\n\nwe do not expect any very anomalous values of x\u00b0 in ferro-electrics, \nexcept in so far as materials with a high electronic polarizability are \nmore likely to be ferro-electric.\n\nWe have now shown that it is possible to incorporate in our model \nfeatures which lead to quite different behavior for the optical and de \ndielectric constants without either invalidating Miller\u2019s rule or even \nchanging the value of A which is essentially a purely electronic property.\n\nWe should, therefore, expect the temperature variation of D%A\u201d to \ncorrespond to that of x,<x,x,, and this is well borne out by the measure- \nments of Zwicker and Scherrer\u2019 of the electro-optic coefficients and \nBass, Franken, and Ward\u2019 of the optical rectification coefficients in \nthe dihydrogen phosphates. Both coefficients are directly proportional \nto the de dielectric constant which obeys a Curie Weiss Law.\n\nIn KDP there is almost no change of the s.h.g. coefficient (Van de \nZiel and Bloembergen*\u2019) with temperature above or below the Curie \npoint, in accord with our expectations, but at the Curie point there is \na small discontinuous change. In an orthorhombic coordinate system \nd2* and dZ% are equal above 7\u2019, but below T,, d,4 increases and d;,3 \ndecreases while at the same time x,; \u2014 x33 decreases and x22 \u2014 X33 \nincreases. With a constant A this is not compatible with Miller\u2019s rule.\n\nHowever, at the transition there is a change in crystal class in which \na, increases and a, decreases, Jona and Shirane.\u201d\u201d It is not unreasonable \nto assume that this increases 73,, and decreases 7'32.. by more than \nenough to compensate for the changes in x,, and x22.\n\nThe classical anharmonic oscillator model, which we have shown to \nbe a good approximation to the behavior of a real system, leads directly \nto that part of Miller\u2019s rule which refers to the geometric properties \nand frequency dependence of the nonlinear coefficients in a single \nmaterial. We have also in (51) advanced a crude argument to show that \nA will not vary much from material to material.\n\nWhen we examine the experimental data we shall see that the allowed \ncomponents of A are between 1 X 107\u00b0 and 6 X 107\u00b0 esu for most mate- \nrials but that there are a few materials with significantly higher values \nand a number with values as low as 0.1 X 10\u00b0 esu.\n\nIn most cases, these exceptional values have a rather simple explana- \ntion and we have therefore to explain a constancy of A to within a \nfactor of about 10.\n\nNeglecting the effects of lattice polarization and local field corrections,\n\nwhich we have shown are irrelevant, the results for the classical an- \nharmonic oscillator model are, from (50) and (56),\n\nThis is also the result from the static perturbation treatment of Section \nIT.\n\nIf we use \n= wy an \u00a9) \nx = 4N aes gN ar, (133) \nwe arrive at \n_ 243 do Vin. \nAi, = 16 e N27?) (134) \nNow the volume occupied by the oscillator is both 1/N and 8 \u00a2r\u2019)' and \nSO \n4/ 23 T 5% \nAijr wd, 10 (r \u00bb (?)3 esu (135)\n\nwhere we have inserted numerical values for a, and e. This expresses \nA;;, as the product of a scale factor (r\u2019)? and a dimensionless shape \nfactor T\u2019/r\u2019. Whether we assign to each oscillator the volume per valence \nelectron, per atom or per group of atoms, (r\u2019)? is likely to be between \n0.75 and 3 A; so that A will be sensibly constant near 3 X 10\u00b0\u00b0 esu, if \nthe shape factor is of the order of 0.01 to 0.05. We have from Turner, \nSaturno, Hank and Parr\u2019s\u2019\u2019 results for CH, a shape factor of 0.05, \nand so this range of shape factors is not unreasonable. It corresponds \nto a linear distortion 0.02? ~ 25 percent. It is also not unreasonable \nthat the distortion should be of this general order, wherever it is al- \nlowed by symmetry. We may speculate that much smaller values of 7'/r\u00b0 \nwould imply very weak interatomic forces and much larger values would \nlead to a structure unstable relative to a more symmetric arrangement.\n\nThus, qualitatively, the relative constancy of A reflects relatively \nconstant shape factors, although we can hardly claim that this is more \nthan a sophisticated form of dimensional analysis. It does, however, \nsuggest that A is determined primarily by the geometric properties of \nthe molecular and crystal structure.\n\nLarge values of A will occur only when the molecules themselves \ndepart considerably from inversion symmetry and are arranged in the \ncrystal in such a way that the effects of individual parts of the molecule\n\nare additive. Small values of A will occur when sections of the molecule \nhave local near inversion symmetry or when their disposition in the \ncrystal favors the cancellation of effects from different atomic groupings. \nHowever the molecules are arranged in the lattice, A will be small if \nthe molecules themselves have near inversion symmetry, or consist \nof uncoupled parts with the same property.\n\npublished s.h.g. data, 7 values from optical rectification data and 50 \nfrom electro-optic data. Definitions and conventions are discussed in \nthe appendix and a separate list of references is given for the data in \nthe appendix. Probable errors vary from measurement to measurement. \nIt is probably safe to say that no measurement has an accuracy better \nthan +10 percent and in many cases the probable error is greater. That \nfor the s.h.g. data at 10.6 u is 30 percent and except for ADP the rectifica- \ntion data is only good to a factor of 3. In addition, a few materials have \ndiscordant results reported by different groups and this suggests that, \nespecially in the case of crystals which are difficult to grow, the data \nshould be regarded rather critically. In the case of CuCl, Sterzer, \nBlattner and Miniter\u2019? have constructed a modulator whose behavior \nis consistent with the higher value of the electro-optic coefficient. This \ncasts some doubt on the low value for CuBr reported in conjunction \nwith CuCl. In the case of the linear e-o coefficient in HMT, Heilmeyer\u2019s\u2122 \nvalue dios = 32 X 10\u00b0\u00b0 esu is the most recent and reliable.\n\nThe average of all the s.h.g. data is A = 3.3 X 10\u00b0\u00b0 esu, and only \ntwo coefficients d,.. in HMT and dz33 in LiNbO; exceed 6 X 107\u00b0 esu \nby more than the probable error. One coefficient dj... in LiNbO is \nunambiguously less than 1 X 107\u00b0 esu. The sole accurate rectification \ncoefficient, Ajo; in NH,HPO,, is 3 X 107\u00b0 esu which is remarkably \nclose to the values 3.15 and 3.2 X 10\u00b0\u00b0 esu for s.h.g. and the electro- \noptic effect.\n\nWhereas the s.h.g. data have a rather compact distribution about \n3 X 10\u00b0\u00b0 esu the linear e-o data are more straggled. The mean value \nis 2.3 * 10\u00b0\u00b0 esu but there are a considerable number of materials \nwith A < 1 X 10\u00b0\u00b0 esu. The difference between the averages A,,, and \nA,., i8 not due to the different materials in the two lists, it persists if we\n\nMaterial Class | A 123 A | aie) A Ref. \nKH2POs \u201cJam |0.55| +65 4.0 | \u201450 |1.7 constant stress 29 \nIWH2PO. 42m 60 aad constant strain 30 \nKD2POs 42m 160 4.0 constant stress 31 \nKHeAsOa 42m 77 3.7 84 1.7 constant stress 32 \nRbHeAsO, 42m 92 3.5 constant stress 32 \nNH4HePO, 42m +55 4.4 | ~\u2014146 |3.4 constant stress 32,29 \nNHaHe2PO, 42m 36 3.9 constant strain\n\neliminate all materials not common to both lists and may therefore, be \neither a real effect or a systematic error.\n\nA few materials [e.g., SrS.0,-H.0, Cs;H,,0;NaBr-H,O and K.Mg, \n(SO,)3] have very low values of A. The latter is especially interesting \nsince the isomorphous (NH,).Cd.(SO.)3 and (NH,).Mn,.(SOx,)3 salts \nhave somewhat larger values. The ammonium cadmium salt 1s known \nto be ferro-electric at very low temperatures and the ammonium man- \nganese salt is also suspected of ferro-electricity (Jona and Shirane\u2019*). \nMore significant, perhaps, is the fact that the divalent ions have very \nnearly regular octahedral coordination (Zemann and Zemann\u2019*\u2019) and \nso form a unit with near inversion symmetry and contribute little to d. \nThe main contribution comes from the monovalent ions and their \nirregularly placed neighbors. The difference between the potassium and \nammonium salts would then be due to the difference in the polariza- \nbility of the two ions. For K\u201d the refractivity is 2.45 ccs and for NH4 \nit is 4.05 ces (see Le Fevre\u2019\u2019). If this enters d as a cube the expected \nratio of the d coefficients would be 4.5. The observed value is greater \nthan about 5. Note also that NH{ itself lacks inversion symmetry.\n\nThe tabulated values of A show that Miller\u2019s rule is an excellent rough \nguide to the probable value of d. If the component is allowed by sym- \nmetry\n\nHowever, the rule by itself is not infallible. Occasionally, a value of \nd much higher than that predicted by (136) will occur. In some cases, \n(e.g., d333 in LiNbO3) this is accompanied by a very low value of another \ncoefficient and it is then plausible to assume that this is due to a partic- \nularly critical geometric configuration. In other cases, (e.g., HMT) it \nis quite clearly due to the coincidence of a number of favorable factors. \nThe atoms, the molecule and the crystal all have the same symmetry \nand moreover, as we saw in Section 1, all the separate contributions to \nd have the same sign. Thus, it is likely that the value of A ~ 15 X 10\u00b0 \nesu for HMT represents something of an upper limit to what is possible.\n\nMore often (186) will overestimate d. This is especially hkely to \noccur if the molecules themselves, or large sections of the molecule have \nnear inversion symmetry, but it may also occur if the crystal structure \nitself departs only very slightly from a centro-symmetric structure.\n\nIf reasonably good ground state wave functions are available, the \ndirect perturbation method of Section II seems most suitable as a\n\nbasis for calculating the magnitudes of the coefficients. It gives the \nintrinsic nonlinear coefficient \n38x\" T esx \nSep eee EE 13 \nin = (137) \nin terms of the intrinsic low-frequency limit of the optical suscepti- \nbility x* and a cubic moment in the ground state. If the electrons are \nuncorrelated, this can be replaced by \n3x\u201d \neo 7 best (188) \nand t,;;, can be obtained from the charge distribution. From (138) we \nobtain the reduced tensor \n3 beck \nace (x\u00b0)\u201d\n\nAsin, = (139) \nand we can then incorporate this directly in Miller\u2019s rule using the \nobserved susceptibilities x,4, etc. to obtain dfh7 .\n\nThis continuation of the basic perturbation calculation with Miller\u2019s \nrule appears to be the most straight-forward approach to the coefficients. \nApart from the cubic moment \u00a2,;;, it involves only experimental \nquantities.\n\nThe analogy with the classical anharmonic oscillator established in \nSection IV seems most likely to be fruitful in qualitative discussions \nof the general behavior of the coefficients. It appears to have both \nempirical and theoretical justification.\n\nA calculation based on fourth-order perturbation theory and a lavish \nuse of sum rules leads to\n\nwhere (r\u2019) is the mean square radius of the charge distribution and \nQ;.. 18 the semi-invariant\n\n950 THE BELL SYSTEM TECHNICAL JOURNAL, MAY-JUNE 1967 \n\u00b0 e \u00b0 \nIf we take (r\u2019)? as 1 A this gives \nAve 3 xX 10\u00b0 ase esu. (143)\n\nWe have seen that in the lower-order processes 7'/r* is of the order of \n2 X 10\u00b0. This does not imply that Q/r* is of the order (2 X 107*)*? ~ \n5 X 10\u00b0\u00b0 for whereas T is nonzero only because of asymmetric molecular \nand intermolecular forces, Q is nonzero even for free atoms or ions. For \nexample, in the hydrogen atom Q,;;;/(r\u00b0)\u2019 = 5/18 and Q;,;;/(r\u00b0)\u2019 = \n1/9 so that we expect A to be of the order of 3 X 10\u00b0\"\u2019 to 10\u00b0\u201d esu. \nFor calcite with x ptiecay = 0.1 and xa, = 0.55 this gives a value of d \nbetween 3 X 10\u00b0\" and 10\u00b0\u201d esu. Bjorkholm and Siegman*\u2019 have meas- \nured 3 X 10\u00b0\u201c esu.\n\nWe have seen that the reduced tensor A,;, is proportional to the cubic \nmoment,\n\nand it is therefore clearly symmetric in all its indices. This is in agree- \nment with Kleinman\u2019s\u2019 hypothesis and follows from the origin of the \nnonlinear behavior in the electronic motion.\n\nFinally, we remark that nothing increases d like large values of the \nlinear susceptibilities yet, although, the values of most allowed reduced \ntensor components A,;;, are near 3 X 10~\u00b0 esu they can vary by a factor \n100:1. The molecular geometry will often indicate which end of the \nrange is likely to apply to a particular material.\n\nWe have throughout adopted a notation, originally introduced by \nBloembergen*\u201d and his colleagues, in which two fields with complex \ntime dependence H%e*** and E%e'\u2019' produce a polarization P%e'*' at \nthe algebraic sum frequency a = 8 + y according to\n\nIf the actual fields vary as cos w,\u00a2 and cos wet there will be terms in \nP at w, + w. and w, \u2014 w, obtained from (144) by letting 8 = a, , a, , \netc. where @, = \u2014,.\n\nThis notation has several advantages in theoretical calculations for \nmuch the same reason that the use of complex numbers simplifies ac \ncircuit theory, and for much the same reason it has a number of dis- \nadvantages in calculating numerical values. For this reason, it has not \ngained general acceptance by experimentalists who tend to use a number \nof different notations, some of which, especially in electro-optics, are \nof respectable antiquity. The difference between the two notations \nintroduces various factors of 2. These are independent of the general \nreluctance of physicists to state unequivocally whether they are using \npeak or rms fields. In particularly fertile ground, these various factors \ncan luxuriate and blossom as factors of 8 in the final answer. We use \npeak fields in all our definitions.\n\nwhere we have used d%23 = d%? and suppressed the first superscript, \nwhich 1s always the algebraic sum of the second and third superscripts. \nWe can also write (146) as\n\nP,(t) = 4{dtos + diss} FoF cos e + QO)t \n+ 4fdt3 + dis}F2Fs cos (m \u2014 Ot. (147) \nIf w = Q, this gives \nPit) = 3fdies*\u00ae + diss\u2019 } Fol\u2019; cos Qu\u00e9 + g{diss\u2019 + dige\u2019 } Fos . (148) \nNow the usual experimental definition would be \nPi) = (dig, + dizz)FoF's cos 2wt + (dtog + d)g2) FoF\u2019; (149) \nand so we see that \nsh.g. dif, = 4d79\u00b0\u00b0 (150) \nrectification d{,, = 4473. (151)\n\nP,(t) = iss\u2019 + diss\u2019) FFs cos wt. (152) \nThe experimental definition reads \nP(t) = dyo3F2F'; cos wt (153) \n1.\u20ac., \n6x12 = Aish\u2019, , (154)\n\nso that it is possible to contract the last two suffices according to the \nscheme\n\n(di23 + d332)F'oFs cos Qut \n2d?3,F.F cos 2uwt = d\u00b0.4-2F.F cos 2wt. (156) \nIt is therefore common to define the \u201cvector\u201d \u00a7\n\n6 \nP; = 2 dipSp \u00b0 (158) \np=l \nWith this notation d\u00b04 = dij = df ~ djs + d28 but also ds = dj. \nIn the electro-optic case, the subscripts referring to optical fields \ncan be contracted\n\nThe alternative ordering (155) leads to df, .. \nNote that in this case, since d operates on two distinct fields, one \noptical the other dc, there is no possibility of constructing a \u2018\u2018vector\u201d\u2019\n\nElectro-optic data are often presented as coefficients r,, in the sus- \nceptibility ellipsoid. If n is the refractive index (assumed isotropic),\n\nIn the MKS system, the units are meters per volt, in the cgs system \nthey are centimeters per stat-volt. One MKS unit is 3 X 10\u00b0 esu and \nso numerical values of d in esu are the larger numbers.\n\nWe have not discussed the influence of a mixed use of rms and peak \nfields but we note that if rms fields are used throughout the values of \nthe coefficients will all be v2 times larger than if peak fields are used \nthroughout. No one is, however, likely to use an rms de field.\n\nExperimental values of the electro-optic coefficients are usually \nexpressed in absolute units and the only ambiguity that can occur is \nassociated with whether the measurements were made at constant stress \n(unclamped) or constant strain (clamped). It is safe to assume that \nconstant stress is implied by the absence of any definite statement to \nthe contrary.\n\nSecond harmonic coefficients are sometimes given in absolute units \nbut more often relative to the coefficient d;., in KH,PO,. An absolute \nmeasurement of this by Ashkin, Boyd, and Diedzic\u00ae\u201d gave\n\nbut this is now believed to be too large. The most recent measurements, \nFrancois\u2019, Bjorkholm,\u201d\u2019 give\n\nfor the coefficient in NH,H,PO, . Relative measurements show that it is \nidentical in KDP and ADP. We have used a rounded off, compromise \nvalue\n\nin compiling the tables. It affects all values of d\u00b0* at optical frequencies \nbut not at 10.6 pu.\n\nIt will be apparent that in comparing theory or experiment with \nexperiment, considerable care is needed to be sure that like definitions \nare being compared with like.\n\n2. Kleinman, D. A., Nonlinear Dielectric Polarization in Optical Media, Phys. \nRev., 126, 1962, p. 1977.\n\n3. Miller, R. C., Optical Second Harmonic Generation in Piezo-Electric Crystals, \nAppl. Phys. Lett., 5, 1964, p. 17.\n\n. Dalgarno, A., Atomic Polarisabilities and Shielding Factors, Adv. in Physics,\n\nKitaigorodskii, A. I., Organic Chemical Crystallography, Consultants Bureau, \nNew York, 1957.\n\nLe Fevre, R. J. W., Molecular Refractivity and Polarizability, Adv. in Phys. \nOrganic Chem., 2. 1964, p. 1.\n\nHeilmeyer, G. H., Dieleetne and Electro-optic Properties of Hexamine, Appl. \nOptics, 3, 1964, D. 1281.\n\nHeilmeyer, G. H, Ockman, N., Braunstein, R., and Kramer, D. A., Second \nHarmonie Generation and the Electro-optic Effect in Hexamine, Appl. \nPhys. Lett., 5, 1964, p. 229.\n\nGarrett, C. G. B. and Robinson, F. N. J., Miller\u2019s Phenomenological Rule \nfor Computing Nonlinear Susceptibilities, J. Quantum Electronics, QE-2, \n1966, p. 328.\n\nArmstrong, J. A. , Bloembergen, N.,, Ducuing, J., and Pershan, P., Light Waves \nin a Nonlinear Dielectric, Phys. Rev., 127, 1962, p. 1918.\n\nButcher, P. N. and McLean, Reel Ss The Nonlinear Constitutive Relation in \nSolids at Optical Frequencies, Proc. Phys. Soc., 81, 1963, p. 219.\n\nWard, J. F., Calculation of Nonlinear Optical Susceptabilities, Rev. Mod. \nPhys., 37, 1965, p. 1.\n\nChang, R. K., Ducuing, J., and Bloembergen, N., Dispersion of the Optical \nNonlinearitv in Semiconductors, Phys. Rev. Lett., 14, 1965, p. 415.\n\nVan der Ziel, J. P. and Bloembergen, N., Temperature Dependence of Optical \nHarmonic Generation in Ferro-Electrics, Phys. Rev., 135, 1964, p. A1662.\n\nSterzer, I\u2019., Blattner, D., and Miniter, S. J., Coprous Chloride Light Modula- \ntors, Opt. Soc. Am., 44, 1964, p. 62.\n\nSecond-Harmonic Generation in Ammonium Dihvdrogen Phosphate and \nCalcite, Phys. Rev., 154, February, 1967, p. 851.\n\n. Ashkin, A., Boyd, G. D., and Diedzic, J. M., Phys. Rev. Lett., 17, 1963. p. 14. \n. Bjorkholm, J. E., Optical Second Harmonic Generation Using a Focused\n\nHeilmeyer, G. H., Ockman, N., Braunstein, R., and Kramer, D. A., Second \nHarmonic Generation and the Electro-optic Effect in Hexamine, Appl. Phys. \nLett., 5, 1964, p. 229.\n\n. Miller, R. C., Optical Second Harmonic Generation in Piezo-electric Crystals,\n\n\u2014An Efficient Phase Matchable Nonlinear Material: Appl. Phys. Lett., 9, \n1964, p. 234; Miller, R. C. and Savage, A., Temperature Dependence of \nthe Optical \u2018Properties of Ferro-electric LiNbO, and LiTaO:, Appl. Phys. \nLett., 9, 1966, p. 169.\n\n. Shiker, T. R., Linear Electro-optic Effects, J. Opt. Soc. Am., 54, 1964, p. 1281. \n: McQuaid, R. W., Pockels Effect of Hexamethylene-tetramine, Appl. Opt, 2 Z,\n\n: Nitsche, R., Crystal Growth and Electro-optic Effect of Bismuth Germanate,\n\nk McQuaid, R. W., Proc. IRE 50, 1962, p. 2484; ibed. 61, 1963, p. 470. \n. Sliker, T. R. and Jost, J. M., Electro-optic Effect and Refractive Index of\n\n. Thornber, K. K., Kurtzig, A. Je and Turner, E. H., unpublished work. \n. Turner, E. H. and Kaminow, I. P, J. Opt. Soc. Am., 63, 1963, p. 528. \nLandolt-Bornstein, Zahlenwerte und Funktionen II Band 8 Teil, Optische\n\n. Thornber, K. K. and Turner, E. H., unpublished work. \n. Buhrer, C. F. and Ho, L., Electro-optic Effect in (NH:)2 Cde(SO.)s and\n\n. Turner, E. H., unpublished work. \n. Carpenter, R. O\u2019B., Electro-optic Sound on Film Modulator, J. Opt. Soc. Am.,\n\nSliker, T. R. and Burlage, 8S. R., Dielectric and Optical Properties of K Dz \nPQ,, J. Appl. Phys., 34, 1963, p. 1837.\n\nOtt, J. H. and Sliker, T. R., Electro-optic Effects in KH-sPO,; and its \nIsomorphs, J. Opt. Soc. Am., 54, 1964, p. 1442.\n\nKaminow, I. P., Barium Titanate Light Phase Modulator, Appl. Phys. Lett., \n7, 1965, p. 123; zbid., 8, 1966, p. 54.\n\nJohnston, A. R. and Weingart, J. M., Low-Frequency Electro-optie Effect in \nBa Ti Os, J. Opt. Soc. Am., 55, 1965, p. 828.\n\nTurner, E. H., High-Frequency Electro-optic Coefficients of Lithium Niobate, \nAppl. Phys. Lett., 8, 1966, p. 303.\n\nBloch, O. G., Zheludev, I. S., and Shamburov, U. A., Electro-optic Effect in \nPenta-erythritol, Sov. Phys. Crystallog., 8, 1963, p. 37.\n\nThe problems associated with \u00abimperfect modulators are discussed and \nthe characteristics of three different schemes of operation are calculated. \nResults from experiments with one such scheme, the reflection mode of \noperation, are given. They compare favorably with the calculations.\n\nA digital light deflector (DLD) is a device that can switch a light \nbeam to a number of distinguishable positions and has been previously \ndescribed by a number of authors.+-* Such a device can be made from \na number of modulators and passive deflectors. The modulator, for \nthis application, 1s one that is capable of switching the sense of \npolarization, and the deflector unit is a passive element which has \ndifferent optical paths corresponding to the two senses of polarization. \nA basic unit of a DLD is shown in Fig. 1. It has been previously\n\nshown\u2019 that such units in series can generate 2\u201d distinguishable \npositions.\n\nA modulator can be made from any material which can become \nbirefringent with the application of an external signal. A minimum of \n7 retardation is needed in order to switch the sense of polarization. \nExamples of modulators that have been considered for this applica- \ntion are Kerr cells,\u00ae stressed plate shutters,\u00b0 and crystals such as \nKDP? and KTN7\u00ae which exhibit an electro-optic effect. The Kerr \ncells, although very fast, cannot be used at high repetition rates be- \ncause of heating difficulties. Stressed plate shutters, since they depend \non mechanical strain, are limited to lower frequencies. The most at- \ntractive modulator materials are the electro-optic crystals and are the \nones that are being seriously considered for the DLD.\n\nThe passive deflectors that have been considered for this application \nare uniformly thick sections of properly oriented uniaxial crystals \nsuch as calcite,2.? prisms of the same materials) and Wollaston \nprisms. The uniformly thick pieces of calcite are used with con- \nverging light for the maximum number of resolvable positions\u00ae *? but \nwith such use suffer from aberrations that are caused by the variation \nof angle in the converging beam. A converging beam of light passing \nthrough a thick piece of calcite oriented for a displaced beam shows \naberrations which for the most part appear like astigmatism. Prisms, \nwhen used with plane waves, can deviate the angle of the plane wave \nwithout distortion, and therefore, a DLD using prisms can give results \nthat are limited only by diffraction theory. A DLD using prisms \nalso uses much less birefringent material than one based on uniformly \nthick pieces of the same material. A disadvantage of simple prisms is \nthat the difference in angle between the two oppositely polarized \nbeams is only a small variation superimposed on the much larger \nnormal type prism deflection. This difficulty can be minimized if the \nprisms are immersed in an oil whose index of refraction is near that of \nthe prism. Wollaston prisms do not have the deflection associated with \nsimple prisms\u2014instead, the only deviation is that between the two \noppositely polarized beams\u2014and are therefore well suited for DLD \nuse.\n\nIn this paper, the design of a high-capacity DLD using Wollaston \nprisms will be discussed along with experimental results which will \nshow that this system does lead to densities limited primarily by \ndiffraction effects. The problem of an imperfect modulator is also \ndiscussed, and calculations on several systems are given which relate \nthe signal to background ratio to the modulation efficiency.\n\nSince a Wollaston prism is used to deflect a beam in angle, it is \nconvenient to think of the operation of a DLD in terms of angular \nspace. Later it will be convenient to place a lens after the DLD \nwhich will focus the beams of light, each with a different angle with \nrespect to the axis of the lens, into corresponding points on the image \nplane.\n\nThe capacity of a DLD is determined by the values of two angles. \nOne is the largest angle that is allowed in the system and the other \nis the minimum angular separation between adjacent positions. The \ncapacity of the DLD is then just the square of this ratio. The \nminimum value is determined by either diffraction effects or imperfec- \ntions in the optical system, and the largest value is determined by the \nmaximum angular aperture of the system. First, let us consider the \nlower limit set only by diffraction theory. The light emerging from a \ncircular aperture illuminated by a uniformly intense plane wave will \nhave a spread of angles that is caused by diffraction. The intensity of \nthe light as a function of angle is given by the well-known Airy func- \ntion (Fig. 2).\u00b0 The smallest deflection angle in a DLD must be suffi- \nciently large so that the deflected beam must be resolved from the \nundeflected one. If we set the criterion that the two beams should be \nseparated in angle such that in the far field the first dark ring of \neach beam overlap, then the minimum angle is given by 2.44 \u00bb/D \nwhere A is the wavelength of light and D is the diameter of the circular \naperture (Fig. 2).\n\nIt is possible to estimate the crosstalk, e.g., the ratio of light within \nthe first dark ring to the light within a circle of equal size as the first \ndark ring but displaced, by examining the Airy function. The light \nwithin the first dark ring contains 84 percent of the total energy,\u00ae and \na ring displaced by one diameter (corresponding to a separation of the \ntwo directions of 2.44 A/D) falls within an annulus of 1.22 A/D to 3.66 \nA/D which contains 10.6 percent of the total energy.\u00ae Since a circle \ncan be surrounded by six circles of the same diameter, this displaced \nring contains somewhat less than 3 (10.6 percent) = 1.77 percent. \nThe crosstalk ratio 1s then 84/1.77 = 47 or 16.7 dB. For a separation \nbetween the beams of twice the above, i.e., 4.88 A/D, the crosstalk \nratio can be estimated to be 210 or 23.2 dB.\n\nIf, in addition to diffraction effects, the wavefront is distorted further \nby some aberration, then the focused spot size will increase and thereby \ndecrease the capacity of the DLD. The amount of wavefront distortion \ndepends somewhat on the type of aberration, but for wavefront distor-\n\ntion of \\/4 or less the increase in the focal spot size is not very significant.\u201d\u00b0 \nThe aberrations in the DLD will result from inhomogeneities in the \nmaterial and from poorly worked surfaces. It should be emphasized \nthat the value of \\/4 is the maximum variation allowed after passing \nthrough the entire DLD, and therefore, the maximum variation for \nany individual unit is much less. For a DLD with 10\u00b0 resolvable posi- \ntions, the total number of units would be 20, i.e., 2\u00b0\u00b0 is approximately \n10\u00b0; and therefore the maximum wavefront error in any unit should \nbe less than \\/4+/20 & /20 where by using the square root we have\n\nassumed random irregularities. The requirement that a modulator \nhave an optical distortion of less than /20, not allowing for any im- \nperfections in the remainder of the optics, is extremely difficult and will \nprobably represent a serious problem for some time to come. The high \nrequirements placed on individual components is a direct result of the \nlarge number of such elements that must be placed in series for the \ncomplete DLD.\n\nThe maximum angular aperture of a DLD is limited by a number \nof effects: (7) the response of the Wollaston prisms, (2) the walk-off \nof the beam as it is deflected to larger and larger angles, (22) the \nangular aperture of the modulators, and (iv) the angular aperture of \nthe output lens. These limitations will now be considered in more \ndetail.\n\nThe deviation angle of a Wollaston prism is not constant but is a \nfunction of the incident angle (see Appendix A) and at some angle \nthe deviation will vary sufficiently such that the array of angles is no \nlonger uniformly spaced. Calculations based on the equations in \nAppendix A indicate that if the Wollaston prism with the smallest \ndeviation is placed first in the DLD and the next largest second, etc., \nfor a total of 20 stages and a maximum angle of deviation of 8\u00b0, the \narray 18 uniformly spaced to within 10 percent.\n\nAs the beam traverses through the DLD, the deflection angle can \nbecome larger and larger and unless the apertures of the prisms and \nmodulators are very large, the beam will eventually strike the sides of \nthe apparatus. It is clear that the Wollaston prism with the largest \ndeviation angle should be placed last in the DLD in order to minimize \nthe spreading of the beams. With this arrangement approximately 14 \nof the beam is intercepted by the apparatus for a 20-stage deflector \nwith a maximum deviation angle of 8\u00b0. This latter figure is not a \nresult applicable in all cases because it depends on the specific lengths \nof the elements in a DLD.\n\nThe relative retardation in a modulator is also a function of the \nincident angle. How rapidly this function varies depends on the type \nof modulator. In KDP and similar crystals the angular aperture is \nvery small (less than 1 minute of are for 30 dB extinction between \ncrossed polarizers for a crystal thickness of 0.089 inches!) since these \ncrystals are uniaxial, with a large birefringence, and will, therefore, \ngive large relative retardation for even small angles away from the \ncrystal axis. Techniques are reported that compensate the birefrin- \ngence?!2 in KDP but the degree of success is not made clear. Because \nof the limited angular aperture, crystals in the class with KDP are\n\nnot considered attractive for the DLD. In CuCl and other crystals in \nthis class, the angular aperture can be very large because they are \ncubic crystals in the field-free case and are therefore optically iso- \ntropic. Angular apertures of +25\u00b0 have been reported'* for this \nmaterial. KTN is also a cubic crystal, but the electro-optic effect in \nthis crystal is quadratic in contrast to the linear effect in most other \nmaterials useful as modulators. Therefore, KTN is usually biased by a \nde voltage in order to reduce the value of the modulation voltage, and \nthis bias reduces the angular aperture of the modulator; however, as \nshown in Appendix B, the angular aperture can still be +10\u00b0 for \nreasonable bias fields.\n\nThe lens at the output of the DLD will focus the beams to points \non the image plane. If this lens is not perfect, the spots will be larger \nthan that calculated by diffraction theory, and the capacity of the \nDLD will be reduced. Since the choice of lens will depend on the \napplication of the DLD, it is not possible to state very precisely what \nthe angular aperture could be; however, +10\u00b0 seems reasonable for \nmost applications.\n\nThe calculations on the capacity of the DLD have been based on a \nplane wave of uniform intensity resulting in an Airy pattern in the \nfar field. By placing a filter in such a beam which attenuates the light \nas a function of the radial distance, it is possible to greatly reduce the \nenergy in the rings at an expense of slightly increasing the size of the \ncentral disk.1* If such a filter could be effectively incorporated in the \nDLD, the crosstalk between resolvable positions could be greatly \nreduced without significantly reducing the overall capacity.\n\nAs an illustrative example, we will calculate the number of re- \nsolvable positions assuming that the minimum angle is set by diffrac- \ntion theory which corresponds to 2.44 A/D for 16.7-dB crosstalk and \nto 4.88 4/D for 23.2-dB crosstalk, and that the maximum angle is \n+10\u00b0. For a wavelength of 6000A and an aperture of 1 cm, this cor- \nresponds to a two-dimensional array of approximately 4(10)\u00ae re- \nsolvable positions with a crosstalk ratio of 16.7 dB and (10)\u00ae positions \nwith a crosstalk of 23.2 dB. These two cases imply 22 basic units in the \nfirst case and 20 units in the latter. If one can learn to use much \nlarger angles, then the number of positions will increase; however, on \nthe other hand, if components are optically imperfect so that diffrac- \ntion-limited performance is not possible, then these numbers will be \nreduced.\n\nThus far the DLD has only been considered in conjunction with a \ndiffraction-limited beam; however, images may also be transmitted\n\nthrough the deflector. Since it takes a number of diffraction-limited \npoints to make up an image, the capacity of a DLD in terms of images \nwill obviously be less.\n\nAt the present time it is not possible to construct a large capacity \nDLD using electro-optic modulators since these materials are not \navailable in the quantity and quality required. In order to check the \nperformance of this system, it is necessary to replace the electro-optic \nmodulators with half-wave plates and thereby replace electronic acti- \nvation with mechanical rotation.\n\nA system as shown in Fig. 3 was constructed consisting of 18 mica \nhalf-wave plates, 7 pair of quartz Wollaston prisms, and 2 pair made \nfrom calcite. The aperture of the system was 18 mm and the wave- \nlength was 6328 A. The smallest deviation angle in the system was 1 \nminute and the largest was 4\u00b0; the smallest deflection angle cor- \nresponds to 8.38 A/D, which is a separation somewhat larger than that \nconsidered earlier in this paper. The aperture of the pinhole was (0.001 \ninches which is larger, by a factor of approximately 4, than that re- \nquired to give a diffraction-limited divergence to the wave emerging \nfrom lens 1. This system when used with an aperture of this size \nshould be considered to be deflecting an image rather than operating \nwith a diffraction-limited beam. The purpose of using a spot of this \nsize is that the ratio of light in the central disk to that in the diffrac- \ntion rings is much higher than when a diffraction-limited beam is \nused, hence the crosstalk between adjacent positions should decrease \nshen compared to the diffraction-limited case.\n\nFig. 4 is a picture of a focal plane taken with this apparatus. It \nshows the 218 214(10)\u00ae resolvable positions. This picture was taken \nby setting each half-wave plate in the halfway position so that light\n\nWOLL ASTON. WOLLASTON \n\u2018 ca PRISM 2 \neet aes \nEe On HHH \nAPERTURE LENS1 LENS 2 FOCAL \nHALF WAVE / \\ HALE WAVE PLANE \nPLATE | \u201cPLATE 2\n\nwas divided equally into both polarizations. In this way all 2** po- \nsitions are simultaneously illuminated. Fig. 5 is an enlargement of an \narbitrarily-selected subsection of Fig. 4 and shows the resolution \nmuch more clearly.\n\nFig. 6 is an enlargement of a single position taken under two cases: \n(1) the pattern on the top illustrates the focused beam with the DLD\n\nremoved from the system, and (72) the pattern on the bottom is the \nsame focused beam with the DLD in the system. The degradation of \nthe pattern on the bottom is a result of the optical imperfections in \nthe many elements that make up the DLD. A comparison of the two \npatterns shows that the DLD did not increase the size of the central \ndisk by an appreciable factor but did make it much more irregular. \nFig. 6 was overexposed in order to show the weaker diffraction rings \nmuch more clearly.\n\nThe crosstalk ratio between adjacent positions was measured by \nfirst setting the DLD so that only one position was present at the \nfocal plane. A 0.001-inch aperture was placed at the focal point, ad- \njusted in position for maximum light transmission, and the amount \nof light was measured by a detector. The aperture was then moved \nto an adjacent position, and the amount of light passing through \nthe opening was again measured. The ratio of these two numbers \nis the crosstalk. The measurements were made for various settings \nof the DLD and the values ranged from 20 to 28 dB. The range in the \nmeasurements is presumably due to imperfections in the optical \nsystem which cause the focused spot to be irregular and unsymmetric \nin shape. The irregular shape is also evident from Fig. 6.\n\nFig. 6\u2014 The degradation of the focused beam pattern by the DLD. (The \npattern on the top was taken without the DLD in the system and that on the \nbottom with the DLD in the system.)\n\nThe performance of this system is probably worse than the A/4 \ntolerance discussed previously in the paper but is probably not much \nworse than a wave or so; this latter figure was not measured directly \nbut was estimated from diagrams which show spot patterns as a \nfunction of various aberrations.*\u00b0\u00b0\n\nIt is anticipated that an electronically activated modulator will be \nthe weak link in DLD performance for some time to come; it therefore \nis important to know how an inefficient modulator will affect the \nperformance of a DLD. In this study, we assume that the Wollaston \nprisms in the DLD are perfect and that the modulator can be charac- \nterized by a single term, #, which is defined as\n\n7 = 2 .. light intensity in the desired polarization _ \nb \u2014\u2014 light intensity in the undesired polarization (1) \nwith\n\nWith perfect modulators the image plane would have one bright \nspot at the desired position and the remaining 2\u201d-1 positions would \nbe completely dark. With imperfect modulators, and for simplicity \nwe assume that they are all imperfect to the same degree, some light \nwill fall on every position. The resulting intensity distribution on the \nfocal plane has been studied by otherst\u00ae?* and is also given in \nAppendix C. It is shown there that the intensities in the focal plane \nof an n unit DLD can be generated by the expansion of a\u201d[1 \n+ (1/E)]|\" where the first term a\u201d gives the intensity of the desired \nposition, the second term n(a\"/H) implies n positions with intensity \na\"/E, the third term [n(n-1) /2](a\"/H?) implies n(n-1)/2 positions \nof intensity a\"/H?, etc. The sum of all the coefficients in the expansion \nof (1 \u2014 1/E)\u201d is equal to 2\u201d so that each position in the focal plane can \nbe assigned to one of these terms.\n\nIt can also be established that the polarization of the even powers of \nKE, i.e, a\", a\"/E?, a\"/E*, etc. have the opposite sense of polarization \nfrom that of the odd powers of \u00a3, i.e., a\"/E, a\"/E?, etc. This result \ncan be determined from the basic definition of the efficiency (1).\n\nThe requirement on the modulator efficiency is determined by the \nparticular application of the DLD. If the deflector will be used to \naccomplish localized heating or welding, to supply energy for a switch, \nor to be used as a printout or display, then the ratio of the intensity \nat the desired location to that at the next highest position is impor- \ntant. This ratio must be sufficiently large so that the intensity at the \ndesired location must be great enough to cause the reaction, and yet \nthe intensity at the next brightest position must be less than that to \ncause the reaction. For example, it is possible that for processes that \ndepend on heating 10 dB is a sufficient ratio, whereas for a visual \ndisplay 30 dB or greater may be necessary so that the eye will not be \nconfused by multiple images. The intensity in the desired position, \nas previously stated, is a\u201d and that in the next highest case is a\u201d/E \nfor the opposite polarization and a\"/EH? for the same polarization. \nTherefore, this ratio is 1/H when there is no polarization selection and \n1/E? when polarization selection is used. In general, it should always \nbe possible to eliminate the opposite polarization so that the first \ntroublesome term will be a\u201d/E?.\n\nThe DLD can also be used as a memory device.\u201d In this application \n& memory is placed in the focal plane which is constructed such that \nat each of the focused points an opaque or transparent spot is present. \nThis code is suitable for a binary organized memory where, for \nexample, the opaque position can represent 0 and the transparent\n\nposition can represent 1. The nature of the position can be read by \nplacing a detector behind the memory and then directing a light beam \nto the desired location. The presence of an opaque position is de- \ntermined by no response at the detector, and similarly a transparent \nposition will result in a positive response.\n\nWhen the modulators are imperfect, the undesired beams of light \nwill also strike the memory plane and have some chance of reaching \nthe detector with the possibility of causing erroneous results. For \nerror-free operation the detector must receive more light when the \nDLD is addressed to a transparent position than when it is addressed \nto an opaque position, and for future reference let us call the ratio of \nthese two intensities the signal-to-background ratio, R. The least light \nthat can reach the detector when the DLD is addressed to a trans- \nparent position is a\u201d, i.e., the main beam alone, while the most light \nthat can reach the detector when the DLD is addressed to an opaque \nposition is Its. \u2014 a\u201d, ie., all the light except for the main beam. The \nminimum signal-to-background ratio is then\n\nThe Rmin ratio defined by (2) is not an unreasonable minimum in \nthat a memory plane can be designed to give ratios very close to the \nvalues calculated by this equation.\n\nThree different ways of interrogating the memory will be discussed, \nand for each case calculations will be made for the signal-to-back- \nground ratios. The first case is where all of the light is allowed to \nstrike the memory plane; the second uses a polarization selection \nbefore the memory so that only light polarized in the same sense as \nthe main beam will reach the memory plane; and the third is the \nreflection mode of operation. To prevent confusion the subscripts 1, 2, \n3 will be used on the Ruin ratios for the three cases mentioned.\n\n4.1 Case 1\u2014All Light From The DLD Allowed to Reach The Memory \nIn this case J... = 1 since a + 6 = 1. Therefore, \na\u201d 1\n\nThe signal-to-background ratio for this case is plotted as a function of \nmodulator efficiency, HZ, for several values of n in Fig. 7. It shows that \nfor an nm = 20 DLD with an (Ryin)i ratio of 5 dB, a modulator \nefficiency of 18.6 dB is required.\n\nIf polarization selection is used after the DLD, which would require \nan additional modulator and polarizer, the odd powers of H can be \ncancelled from (2) and the (Ryin)o ratio becomes\n\nThis ratio, (Rmin)2, is plotted as a function of efficiency in Fig. 8. It \nshows that for n = 20 and (Rmin)o = 5 dB, a modulator with an \nefficiency of 14.0 dB is required. This case represents an improvement \nof 4.6 dB over the first case.\n\nFig. 7\u2014 Minimum detector ratio versus modulator efficiency for a DLD where \nthe detector is placed behind the focal plane.\n\nAn alternate way\" of reading the memory plane is shown in Fig. 9. \nIn this case, the light is reflected from a mirror located just behind \nthe memory and is redirected through the DLD to be detected \nafter passing through a second aperture. The second aperture elimin- \nates a large part of the background and therefore the ratio, Ruin, for \nthe same modulator efficiency is considerably improved. The deriva- \ntion of the Ryin ratio for this reflecting mode of operation, (Rmin)3, is \ngiven in Appendix D and only the result 1s shown here:\n\n(Rmin)3 18 plotted as a function of # in Fig. 10. With this mode of \noperation a modulator with an efficiency of only 9.1 dB is required to \ngive an (Rmin)g ratio of 5 dB or better. The reflection mode of opera- \ntion therefore represents an improvement, when measured by reduced \nrequirements on the modulator, of 9.5 dB over the first case and 4.9 dB\n\nFig. 8\u2014 Minimum detector ratio vs modulator efficiency for a DLD with \nthe detector placed behind the focal plane and with polarization selection.\n\nFig. 9\u2014 Arrangement of elements for the DLD using the reflection mode of \noperation.\n\nover the second case. It should be emphasized that the improvement \nratios given here are strongly dependent on the choice of the Amin \nratio, which was taken to be 5 dB in this paper. For larger Rin \nratios the improvement in the HE ratio would be even greater and vice \nversa.\n\nA disadvantage of the reflection mode is that a low f-number lens \nmust be used at the output of the DLD. The reason for this is that \nthe light reflected from the plane mirror must enter the output end of \nthe DLD, and this requires that the focal plane be approximately 14 \nthe linear dimension of the aperture of the DLD. One can show that \nthis requires a lens of f:1.5 or so if the angular spread of the DLD 1s \n+10\u00b0. If a lens is designed to have a spherical focal surface and a \nspherical mirror is used as the reflector, then the lens can have any f \nnumber.\n\nAny light reflected from the surfaces in the DLD when used in the \nreflection mode can be prevented from entering the aperture near \nthe detector by giving a slight tilt to the elements that make up the \nDLD. It is possible to choose an angle such that no reflection is \ncentered on the aperture.\n\nV. EXPERIMENTS WITH THE MANUALLY-OPERATED DLD EQUIPPED WITH \nPOOR MODULATORS\n\nFig. 10-\u2014- Minimum detector ratio versus modulator efficiency using the re- \nflection mode of operation.\n\nIdeal half-wave plates have the property that if the angle between \nthe polarization direction and the axis of the half-wave plate is 6 \nthen the plane of polarization emerging from the plate will be rotated \nby 26 from its original direction. For maximum efficiency, the half- \nwave plates are oriented at 6 = 0 if no switching action is desired and \nat 6 = 45\u00b0 if the other polarization is desired. The practical maximum \nefficiency for the split mica plates used in this experiment ranged \nfrom 30 to 40 dB, which is high enough to give almost perfect DLD \nbehavior. In order to simulate poor modulators, the wave plates are \nset at 6 = e for the predominantly unswitched case and at 6 = 45\u00b0 \u2014 \u00ab \nfor the predominantly switched case. The angle \u00ab can then be adjusted \nto achieve any degree of modulator efficiency.\n\nTo illustrate the behavior of a DLD under the influence of poor \nmodulators, the half-wave plates were set for H = 10 dB, and a \npicture, which is shown in Fig. 11, was taken at the focal plane. This \npicture shows some characteristics which will now be enumerated:\n\n(1) The desired spot is shown as the brightest point in the upper \nleft-hand quadrant and is vertically polarized. 7\n\n(72) Some of the 18 points whose intensity is 1/H of the main \nbeam are shown in the lower right-hand quadrant and are horizontally \npolarized.\n\n(122) Some of the 153 points whose intensity is 1/H? of the main \nbeam are shown in the upper left-hand quadrant and are vertically \npolarized.\n\n(iv) Some of the 816 points whose intensity is 1/H? of the main \nbeam are shown in the lower right-hand quadrant and are horizontally \npolarized.\n\n(v) The points in the upper left-hand quadrant are vertically \npolarized and those in the lower right-hand quadrant are horizontally \npolarized. A polarization selector in this case would eliminate the \nentire lower side. It will always eliminate the side that is opposite to \nthe one that contains the main beam.\n\n(v1) There are no points with significant intensity (> 1/EH\"/?) in \neither the upper-right or lower-left quadrant.\n\nAs all of the focal plane is not shown in this figure, some of the \nbackground positions are missing in order that an enlargement could \nbe presented. In this exposure there is a total of approximately .4.6 \ntimes more light energy in the background than in the main beam.\n\nThe performance of the reflection mode of operation was compared \nto the theoretical calculation by making use of the properly mis- \noriented half wave plates. For two reasons (5) cannot be directly \nused: (1) In this experiment a memory was used that contained only \none opaque position and for such a case the signal-to-background ratio \nusing this equation is not accurate, and (iw) equation (5) assumes \nthat the opaque positions are also perfectly absorbing which is not \nvalid \u2018for this experiment in that the opaque position reflected 4 \npercent of the incident power. A signal-to-background ratio for this \nparticular experiment can be calculated as follows. When the DLD is \naddressed to a transparent position the light reaching the mirror is \nTyo and the light striking the mirror when the DLD is addressed to \nthe opaque position is Ij \u2014 (1 \u2014 T)a\" where T is the power reflec- \ntion coefficient. The ratio of these two values is\n\nThe calculated and experimental values of (R.x\u00bb)3 are summarized \nin Table I. The first column lists the modulator efficiency, the second \ncolumn contains the calculated (Rexp)3 and the third column lists the \nmeasured values.\n\nThe agreement between the calculated and measured quantities \nagree very well and indicate that the behavior of the reflection mode \nof operation is adequately understood.\n\nIn the DLD thus far discussed, only one memory in the output \nfocal plane is used (see Fig. 3), and therefore the memory is read one \nbit at a time. For some applications it may be advantageous to \nparallel the output as shown in Fig. 12 in order to increase the bit \ncapacity of the DLD. With this scheme the number of bits read with \neach setting of the DLD is equal to the number of memory planes; \nthe memory now corresponds to one with word organization. Fig. 12 \nshows four such memory planes but by going to a three-dimensional \narray it is possible to parallel 30 to 40 such planes and still use only \none output lens providing the maximum angular aperture of the DLD \nis limited to a total angle of 12\u00b0 or so. If additional lenses are em- \nployed, then any number of memory planes can be incorporated.\n\nThis scheme of paralleling is directly applicable to cases 1 and 2, \nwhich were discussed in Section IV, but will not work for the reflection \nmode since there is no way to distinguish between different memory \nplanes when the light is redirected through the DLD. In order to \nparallel the output of the reflection mode, three different schemes have \nbeen devised: (2) to use different wavelengths for each memory plane \nand separate the colors before and after the DLD,\u201d (i) to modulate \na monochromatic beam at each memory plane with a different fre- \nquency and then to separate the different frequencies after reflection \nthrough the DLD,?\u00b0 and (12) to arrange the memory planes to have \ndifferent distances between the output of the DLD and the memory \nplane and to use a short pulse of light; the different planes can now \nbe read since each plane will return the pulse to the detector at a \ndifferent time.**\n\nOne difficulty that can arise in the paralleling schemes is that very \nlow f-number lenses must be used to refocus the output plane into \nrepeated images (Fig. 13). The first lense placed after the DLD can\n\nhave a reasonable f number since the light beams are still confined to \nthe aperture of the DLD. The second lens must have f:1 or so if the \noutput of the DLD has a total angle of approximately 12\u00b0. Additional \nlenses must have even lower f numbers (Fig. 13). The practical solu- \ntion to this problem is to perform all of the paralleling within the \nfirst focal length. This scheme will not work with the reflection mode \nwhere the time of flight varies for each memory plane since one lens \nimplies only one distance between it and the various memory planes.\n\nThe problem of reading a memory has been discussed in earlier \nsections of this paper; we will now consider the problem, which is \nagain primarily the result of imperfect modulators, of using the DLD \nto write into a memory. Two general types of materials will be \nconsidered for use as a memory medium. One is a medium where the \nprocess is linear in terms of total exposure, i.e., the effect on the \nmedium of n pulses of light of intensity I/n, each lasting for a time \nAT, is the same for any value of n; and the second is one which has a \nthreshold in terms of the light intensity. An example of the first type \nis a photographic film and of the second is a memory based on a \ntransparent ferrimagnetic garnet at its compensation temperature.\u201d*\n\nA linear medium has the disadvantage, for this application, in that \nit integrates the light striking its surface. Therefore, when the photo- \ngraphic film is being exposed by a DLD with poor modulators, the \nproblem of the background light must be considered. This problem \nis different from that considered in Section IV because in that case \nthe DLD was set for one address and the question was asked what is \nthe light intensity distribution over the whole focal plane. In this \ncase, we ask what is the total amount of light energy striking one \nposition on the memory plane when the DLD is addressed to all of \nthe positions. When the DLD is set for one address, the total exposure \nover the whole plane is a\u201d[1 + (1/E)]\"AT where AT is the duration \nof the exposure. This result is evident from the discussion in Section \nIV. When one sits at a position and the DLD is addressed to all \npositions and dwells at each position for the same time AT\u2019, the total \nexposure at that position is also a\"[1 + (1/H)]\"AT. This latter result \nhas been previously published?\u2019 and is also proven in Appendix C.\n\nThe following calculations, which represent worst cases for writing, \ncan be performed. The first case that will be considered is the situation \nwhere all of the light is allowed to strike the memory plane and the \nsecond case makes use of polarization selection. In both cases it will be \nassumed that all points will be addressed, except for one.\n\nIn this case the light striking any of the addressed positions will \nhave an exposure of nearly a*(1 + 1/H)\"AT and the exposure at the \none position that was not addressed will be [a\"(1 + 1/4)\u201d \u2014 a\u2019JAT, \nand the ratio of these two values is\n\nThe exposure ratios M, and Mz are plotted as a function of modula- \ntor efficiency in Figs. 14 and 15. These plots can be used to determine \nthe minimum modulator efficiency required for a certain exposure \nratio.\n\nFrom the exposure ratio and the properties of the medium, e.g., \nphotographic film, the density ratios of the positions can be calculated. \nThese two ratios do not have to be the same, as a material such as \nphotographic film can be very linear in terms of exposure but the ex- \nposure vs film density can be very nonlinear.\n\nFor a theshold medium the problem is much simpler. The only re- \nquirement is that the most intense beam must be greater than threshold \nand that the next highest position be less than threshold. As men- \ntioned in Section IV, this ratio is 1/H when no polarization selection \nis used and 1/H? when polarization selection is used.\n\n_ Fig. 14\u2014 Exposure ratio vs modulator efficiency for a DLD where all of the \nlight is allowed to reach the memory plane.\n\n2(10)*x along the Y bank for a total of 4(10)\u00b02 total retardation. With \na DLD, a total retardation of 207 can accomplish the same number \nof resolvable positions. Therefore, it is evident that the DLD makes \nvery efficient use of the variable retardation. The reason for this ef- \nficiency is that the DLD makes use of the fixed retardation in the \npassive elements whereas the analog deflector must generate all of the \nretardation. In addition, the DLD can be designed for any separation \nbetween the beams and still not require any more than the 207 varia- \nble retardation. The analog reflector, on the other hand, cannot \nseparate the beams any further without supplying additional retarda- \ntion.\n\nFig. 15 \u2014 Exposure ratio vs modulator efficiency for a DLD where polarization \nselection is used.\n\nresolvable positions that can be attained is reasonably close to that \nallowed by diffraction theory. The effect of imperfect modulators on \nthe performance of the DLD has also been discussed.\n\nThe discussion presented here does not mention the problems as- \nsociated with the high-speed switching of an electro-optic modulator, \nwhich is a problem that must be solved if the DLD is to have broad \napplication. This problem has been studied by 8S. K. Kurtz.?*\n\nThe author is greatly indebted to H. E. D. Scovil and K. D. Bowers \nfor valuable discussions involving the applications and overall per- \nformance requirements of the digital-light deflector.\n\nThe author is also indebted to J. T. Sibilia and J. E. Geusic for their \ncontributions to many of the solutions associated with the reflection \nmode of operation. He also wishes to acknowledge discussions with \nR. G. Smith on some of the problems associated with imperfect \nmodulators and to J. G. Skinner on some of the optical problems.\n\nThe formulas necessary to trace the two wave normals through a \nWollaston prism in a direction as shown in Fig. 16 are given below:\n\nA useful approximate formula for calculating the total deviation angle \nof a Wollaston prism, A, (A = a + 6b) for perpendicular incidence, \ny = Oils\n\nThe variation of A with respect to a variation in y at perpendicular \nincidence, (dA/dy),-0 , can be calculated from (9) and (10) to be\n\nThis calculation is valid for materials that are cubic, and therefore \noptically isotropic, in the absence of an electric field and become \nuniaxial, with the optic axis parallel to the electric field, in the pres- \nence of an electric field.\n\nFig. 17 describes the placement of the crystal with respect to the \nincident radiation. The vy plane is the first surface of the modulator, \nand the second surface is parallel to the first and passes through the \npoint Z equals \u2014T\u2019. The induced C axis of the crystal is parallel to the \ny axis. The light ray makes an angle y with the z axis, and the inter- \nsection of the plane of incidence with the zy plane makes an angle a \nwith the z axis. The relative retardation between the extraordinary \nand ordinary ray can be calculated to be\n\nreduces the familiar (np \u2014 ,) (2x7'/d,) for perpendicular incidence. \nSince n, and n, are nearly the same in this case, we will expand\n\nno)? and higher. We will also expand sin y using a power series in y. \nThe result of these substitutions is\n\nFig. 17 \u2014 Coordinate axes showing the placement of the uniaxial crystal with \nrespect to the incident radiation.\n\nIn order that a modulator switch the sense of polarization, it is nec- \nessary that the retardation be changed by x. In general, the retarda- \ntion will be changed from Nx to (N + 1) with the application of an \nelectric field. The retardation as a function of incident angle for the \nlargest retardation, (NV + 1)z, using terms only up to y?, becomes\n\nThe angular aperture of a modulator is determined by the value of \ny where the change in retardation from + becomes serious enough to \ncause unwanted behavior. If we call this change in retardation AR, \nthen the value of y that corresponds to this AR can be calculated from \n(16), 1.e.,\n\nn.AR | \n_ lax + 1)r(cos\u2019 a \u2014 4) ]\u00b0 uo) \nV\u2019rom (17) one can see that the angular aperture for a given material \nis inversely proportional to ~/N + 1 so that we can write\n\nYbiased = \u2014tonblased \u2019 (18) \nVN+1 \ni.e., to the same degree of performance the angular aperture of a mod- \nulator biased to Nz is decreased by the factor 1/WN +1 of the \nunbiased case.\n\nFor a material with a linear electro-optic effect, it is most sensible \nto use N = 0 in order to use the lowest voltages; this will result in the \nmaximum angular aperture. For a quadratic material like KTN, it is \nsometimes more efficient to use a biasing de voltage in order to reduce \nthe modulation voltage. In that case, an N of 10 or 20 might be used. \nIf we decide that the maximum retardation error is AR = 0.02067, \ncorresponding to a minimum extinction of 30 dB between polarizers, \nthen the angular aperture of KTN is +26\u00b0 in the unbiased case, +-7.8\u00b0 \nfor N = 10, and +5.7\u00b0 for N = 20.\n\nThe angular aperture of biased KT'N can be increased by placing \na properly oriented positive uniaxial crystal such as quartz in series \nwith the biased KTN modulator. This technique can be used to elim- \ninate the terms in y* from the total retardation and thereby increase \nthe angular aperture to approximately that of the unbiased case.\n\nAssume that we have a DLD consisting of n stages made up of \nmodulators with an efficiency, EH = a/b, a + b = 1. We again assume \nthat only the modulators are imperfect, and that every modulator can \nbe characterized by the same efficiency.\n\nAny light beam incident to a modulator is broken up into two \nbeams, an \u201ca\u201d beam for the desired polarization and a \u201cb\u201d beam for \nthe oppositely polarized position. A table can be made up which lists \nthe total number of paths through the DLD (Table II). Since we have \na choice at each modulator as to whether an a or b path is taken, the \ndifferent paths are characterized by all possible combinations of the\n\na and 6 terms. Let us consider the paths containing an r number of 6 \nterms and an (n \u2014 r) number of a terms. The total number of pos- \nsible ways of grouping these terms is given by\n\nn! \nseal aa a Tae 19 \n(n \u2014 r)!r! 1) \nand the associated intensity of these beams is a~\"b\". The total \nnumber of all paths is then given by summing r through its range\n\nwhere the intensity terms have been included. These terms can also be \ngenerated by\n\nTo derive (22), the DLD was set for one address and the intensity \nat each point was determined. We now ask what are the different in- \ntensities that arrive at a particular position when the DLD is ad- \ndressed to all possible positions.\n\nIn order to address the DLD to every position, the state of the \nmodulators can be arranged according to Table III. In this table, 0 \nmeans no change in the state of polarization and 1 refers to a change\n\nto the opposite sense of polarization. Table III is a partial listing. The \ncomplete table is made by first writing the nth column which con- \nsists of alternating 0\u2019s and 1\u2019s for a total number of 2\u201d entries; the \n(n \u2014 1)th column is written by entering pairs (2') of the 0\u2019s and 1\u2019s \nfor a total of 2\u201d; the (n \u2014 2)th column by entering 2? of 0\u2019s and 1\u2019s, \netc. The sequence of addresses in Table III would place the main beam \nonce at each location on the focal plane. The 0\u2019s and 1\u2019s that appear \nin any horizontal row is the address of that beam.\n\nWe must now be able to compare intensities between that of the \nmain beam, which we shall call A,, and some arbitrary position, which \nwe shall call A,. Table IV illustrates the technique. Table IV was \nconstructed by using a rule that sets the intensity ratio at 1 for \nmodulators that have the same setting and b/a at modulators that \nhave different settings.\n\nWe now wish to determine all of the intensities x some arbitrary \nposition, say A, = ...010 while the DLD is addressed to all positions. \nUsing Table III which lists all of the addresses and Table IV which \nillustrates the comparison rule, we can construct a table (Table V) \nwhich lists the intensity ratios at A, = ...010.\n\nTable V is similar to Table III in appearance in that for any \nvertical column a 0, 1 in Table III is changed into 1, b/a or b/a, 1 to \nmake Table V.\n\nTable V is one that lists all possible combinations of the entries 1, \nb/a and therefore is calculable from the same general formula as that \ndeduced for Table IT, 1.e.,\n\nIt is evident that Table V would not change, except for a different \nordering of the horizontal rows, no matter what the particular address \nof A,. Therefore, (23) is the same for all points A,.\n\nEquation (28) will list the intensity ratios between A, and some \npoint A,. If we multiply though by the intensity of A, = a\u201d, then (22) \nbecomes\n\nTABLE V\u2014 List or INTENSITY Ratios at Point ---010 WHEN \nDLD 1s ADDRESSED TO ALL POSITIONS\n\nA beam of light traveling through the DLD breaks up into 2\u201d exit \nbeams due to the imperfect modulators. These intensities are given by \nthe terms in the expansion of (a + b)\u201d as shown in Appendix C. We \nnow need to ask how much of the light comes back through the \nsecond aperture after being reflected from the focal plane (see Fig. 9).\n\nLet us consider one term of the expansion of (a + b)\", say a\u201d-\"b\u2019. \nWe state that this system is reciprocal and that if a\u201d-\"b\" of the \nincident beam exits the DLD, then if unity power were directed \nthrough the DLD in exactly the opposite direction the same fraction \nof power, i.e., a\u00ae~\"b\", will pass through the aperture.\n\nThus, for an n unit DLD there will be 2\u201d exit terms and each of \nthese terms, for example a\u201db\", will generate one term that contributes\n\n* This appendix represents the results of calculations performed jointly by \nJ.T. Sibilia and the author.\n\nto the intensity at the second aperture; and for the example above that \ncorresponds to (a*-\"b\")*. The total number of terms that exit through \nthe second aperture is then the sum of the squares of the exit terms \nand can be generated by (a? + b?)\u201d.\n\nBefore we can add up the 2\u201d terms in the second aperture, we must \nknow something about the relative phases of the terms. Each of the \n2\u201d exit terms in the expansion of (a + 6b)\u201d traverses through a dif- \nferent optical path length in the DLD. The reason for this is because \nlight traverses through some of the prisms as an ordinary ray and \nothers as an extraordinary ray, and the combinations of such paths \nare different for each of the 2\u201d exit beams. Thus, unless the DLD has \nbeen specifically designed to the contrary, each path has a different \nphase delay in passing through the DLD.\n\nA term in the expansion (a? + b?)\u201d such as a?(-\")b* represents an \nE field of [a?-\"b*\"]? and a phase factor yg. Consider the sum of all \nthe n!/(n \u2014 r) tr! terms of the type a?-\u201d 6?\"\n\n[arb F@, Rs ee aie): (25) \nThe intensity of the sum of all a?-\"b?\" terms is given by the \nsquare of (25)\n\nf 2 \n= ae ; for the same phase. \n(near \nAs explained earlier, all of the phases are, in general, different, and so \nwe will use the random phase addition. The intensity at the second \naperture is then the sum of all of the terms in (a? + 67)\u201d. \nThe \u00a3# ratio, the ratio of the light from the main beam, a?\u201d, to the\n\n. Schmidt, U. J., Optical Processing of Information, Spartan Books, Ince., Balti-\n\n. Nelson, T. J., Digital Light Deflection, B.S.T.J., 43, May, 1964, p. 821. \n. Kulcke, W., Harris, T. J., Kosanke, K., and Max, E., IBM J. Res. Develop., &,\n\n. Born, M. and Wolf, E., Principles of Optics, Pergamon Press, 1959, p. 394. \n. Cagnet, M., Francon, M., and Thrierr, J. C., Atlas of Optical Phenomena,\n\nKulcke, W., Kosanke, K., Max, E., and Fleisher, H., Use of Optical Masers \nin Displays and Printers, Third Quarterly Report, February 24 through May \n23, 1964, Contract No. DA36-039-AMC-00118(E).\n\nSkinner, J. G., Increasing the Memory Capacity of the Digital Light Deflector \nby \u201cColor Coding,\u201d B.S.T.J., 48, April, 1966, pp. 597-608. \nSeidel, H., unpublished work.\n\nKurtz, 8. K., Design of an Electro-Optic Polarization Switch for a High- \nCapacity High-Speed Digital Light Deflection System, B.S.TWJ., 45, Octo- \nber, 1966, pp. 1209-1246.\n\nTransistor Distortion Analysis Using \nVolterra Series Representation\n\nIntermodulation distortion due to nonlinear elements in transistors is \nanalyzed using Volterra series representation. It is shown that this technique \n1s well suited for the analysis of transistor distortzon where the nonlinearities \nare small but frequency dependent. An ac transistor model incorporating \nfour nonlinearities 1s briefly described. The nonlinear nodal equations of \nthe model are successively solved by expressing nodal voltages in terms of \nthe Volterra series expansion of the input voltage. Based on this analysis, \na digital computer program has been developed which computes the second \nand the third harmonic distortion for a given set of input frequencies and \ntransistor parameters. The results compare favorably with measured values. \nThis method also enables the derwation of closed form ac expressions for a \nsimplified model; these expressions show the dependence of distortion on \nfrequency, load and source wmpedances, bias currents and voltages, and \nthe parameters of the transistor. The technique 1s also extended to cascaded \ntransistors, and simplified expressions for the overall distortion in terms \nof the distortion and gain of individual transistors are derived. Finally, a \nfew pertinent practical applications are discussed.\n\nSolid-state long-haul analog communication systems are being de- \nsigned for higher frequencies to meet the growth in demand. One of \nthe more critical and significant problems facing the system designer \nis intermodulation noise arising from transistor nonlinearities. Thus, \nan analysis of transistor distortion at higher frequencies is a practical \nproblem; this paper investigates the transistor distortion using the \nVolterra series as an analysis tool.\n\nTransistor distortion has been investigated in some detail previously. \nMany authors have considered the exponential nonlinear relation be- \ntween emitter current and emitter-to-base voltage which is important\n\nat low currents.''\u2019\u2019***\"? The effect of frequency on this nonlinear source \nalone has been reported.\u201d Three nonlinearities (exponential, avalanche, \nand hr, at de) have been examined by Riva, Beneteau and Dallavolta.\u00b0 \nFor currents up to 20 mA and frequencies up to 100 kHz, Meyer\u2019\u2019*\u2019\u00ae \nhas developed a more accurate and complex model obtaining the non- \nlinearities from h-parameters. However, he takes into account the fre- \nquency dependence by assuming that the h-parameters can be written \nas h' +- jwh\u2019\u2019. Moreover, he does not take into account avalanche dis- \ntortion, nor has he extended the model to higher currents (100 mA) and \nfrequencies (20 MHz). The model described here considers four non- \nlinearities; they are, exponential, avalanche, hy, , and collector capaci- \ntance nonlinearities. These nonlinearities are superimposed on a linear \nac equivalent circuit.\u2019\u00b0\u2019\u2019* Much of the initial development of the model \nwith three nonlinearities was done by Thomas.\u2019\u00b0\n\nThe transistor model is analyzed using a Volterra series representa- \ntion; this series is a generalization of the power series. In a now classic \nreport, Wiener applied this analysis technique\u201d to find the response \nof a nonlinear device to noise.\"* Bose has carried the theory further.\u2019 \nFollowing a series of lectures by Wiener,\u2019\u2019 the theoretical framework, \nhigher-dimensional transforms, and optimization with Gaussian inputs \nwere considered by Brilliant,\u2019\u00b0 George,\u2019\u2019 and Chesler,\u2019\u00ae respectively. \nBarrett\u2019\u201d has treated statistical inputs. The synthesis problem has been \nexamined by Van Trees,\u2019 who also applied the method to phase-locked \nloops.\u201d The technique has been extended to discrete systems,\u201d\u2019\u201d\u2019\u201d* and \na class of time-variant systems.\u201d*\u2019\u2019\u2019 More recently the theory of the \nconvergence of the series has been treated.\u201d\u00ae This work relies more on \nGeorge\u2019s work on the higher-dimensional transform theory.\u201d\n\nEven though much work has been done in this area, the Volterra \nseries has not found a wide application in solving nonlinear system \nproblems due to several reasons; if the rate of convergence is not rapid, \nthe higher-degree terms, which are cumbersome to handle, cannot be \nneglected; hence, it cannot conveniently represent gross nonlinearities. \nIt is not simple to invert the multidimensional transforms to the time \ndomain, and it is not a useful technique to determine the stability of a \nnonlinear differential equation.\n\nThe Volterra series method does, however, offer certain distinct \nadvantages in analyzing transistor distortion. Since transistor distortion \nis frequency dependent, the power series is inadequate to characterize \nit; the Volterra series does indeed represent frequency dependent sys- \ntems. The nonlinearities in the transistors under consideration are \nextremely small so that the second- and third-degree terms suffice to\n\ncharacterize them. Since the output corresponding to sinusoidal input \nsignals is of interest, there is no need to find the inverse of the higher- \ndimensional transforms; the output can be expressed in terms of the \ntransform of the kernel. The higher-dimensional transforms of the \nkernel are complex numbers when s; = jw; , where s; is the complex \nvariable in the transform domain; hence, these kernels can be numeri- \ncally evaluated using the computer (see Section IV). Moreover, for a \nslightly simpler model closed form ac expressions can be derived. Since \nthe kernels retain phase information, this approach will be useful for \nthe AM-to-PM conversion problem at IF frequencies. Finally, in an \namplifier two or more transistors are cascaded; the nonlinear behavior \nof such cascaded transistors is a significant problem. The Volterra series \napproach can be easily extended to study such cascaded transistors.\n\nA brief exposition of Volterra series with pertinent reference to the \nproblem under consideration is presented below. For further details \nthe reader is referred to the references cited.\n\nwhere \u00a2, , C2 , C3 are constants. For a time-invariant system with memory \n(capacitors and inductors in an electrical network), the linear term \n(\u00a2,2(t)} is replaced by the convolution integral (a(t) = 0; 4 < 0)\n\nThis transform domain representation of the system [C,(s)] has been \nan invaluable aid to the communication engineers since it brings into \nfocus the frequency behavior of the system.\n\nA generalization of the second-degree term, c.[x(t)]\u2019, is the double \nconvolution integral\n\nThe output depends on the past values of the input; the above expression \ninvolves a product of the input with itself, thus representing a quadratic \nsystem. \u00a22(\u00a2 \u2014 7; , \u00a3 \u2014 72) is known as the second-degree Volterra kernel.\n\nA two-dimensional Laplace transform can be defined for (4) after \nintroducing dummy variables \u00a2, and \u00a2,. As shown in Appendix A, (4) \nbecomes\n\nWhen two sinusoidal signals at frequencies f, and f, are applied \n(Appendix A), the output at the harmonic frequency f, + f, is given \nby {| Co(fa + fo) | cos (27(fa & fo)\u00e9 + dazs)]. Since in general C2(f. , fz) \nwill not be equal to C.(f, , \u2014f,), different values of distortion at different \nharmonic frequencies are directly reflected in the kernel. Moreover, as \nin the power series case, the 2f product is less by a factor of two.\n\nLikewise, the third-degree term [c3(x(r))*] can be generalized to a \ntriple convolution integral;\n\nThe magnitude of the signal at the harmonic frequency f, + f, \u2014 f. \ndue to the three fundamental signals at f, , f, and f, is given by | Cs(f, , \nfs, \u2014f-) |. The constants like 1/4 for a \u20188f,\u2019 product are the same as \nobtained from the power series approach. |\n\nLater in the paper (in Section IV) the cascade relations in the trans- \nform domain are frequently used; their physical significance is discussed \nin detail in Section VI. (See also Fig. 1.) The cascade formulae and the \nprocedure for deriving them are given in Appendix A.\n\nThe second and third harmonic distortion are defined as the second \nand third harmonic power in dBm, respectively, when the fundamental \npower at the output of the transistor is at zero dBm (one milliwatt). \nIn the analysis of the model in Section IV, the output voltage is ex- \npressed in terms of a Volterra series of the input voltage. Thus, the \nkernels C,(s:), Co(s:, S2), and C3(s,, S, Ss) are the voltage transfer \nratios; for a given load R&, , the second and the third harmonic distortion \nin dBm are given by the following expressions:\n\nA model is a simple but realistic representation of a physical phe- \nnomenon in terms of measurable parameters such that the phenomenon \ncan be analyzed, and controlled if possible. The linear equivalent circuit \nof a transistor is one such example. In reality, several elements of the \ntransistor equivalent circuit are not linear but are linearized versions \nof nonlinear functions; they are the first-degree terms of the Taylor\u2019s \nseries expansion of the nonlinear functions. Hence, a logical way to \ndevelop the nonlinear model is to consider the second- and third-degree \nterms of the Taylor\u2019s series expansion; thus, the emitter resistance \n(exponential nonlinearity), current gain (hp, and avalanche nonlin- \nearity), and the collector capacitance (collector capacitance nonlin- \nearity) have been represented by nonlinear voltage dependent current \ngenerators whose parameters are higher-degree Taylor\u2019s series terms. \nThis approach has another advantage in that it is difficult to measure \nthe nonlinearities since they are small; but, it is not too difficult to \nmeasure the overall functions and to curve fit with the known theoretical\n\nThe emitter current, J; , is related to the emitter voltage, V., by \nthe exponential relation\n\nand A and B are constants which depend on the transistor parameters \n(Ref. 27; p. 181, p. 249). An experimental curve of the emitter current \nI, and the emitter-to-base voltage V,. 1s shown in Fig. 3. This non- \nlinearity is expressed as a voltage-dependent current generator by a \nTaylor\u2019s series expansion of (10) as follows:\n\nwhere the Taylor\u2019s series coefficients are derived in terms of known \nparameters, the emitter resistance r,, and the emitter bias current J; ; \n1.\u20ac.,\n\nThe collector current is a nonlinear function of the emitter current \nat higher values of current (hr, nonlinearity) and of the collector-to-base \nvoltage at higher values of voltage (avalanche nonlinearity).\u201d\u2019 hy, , \nthe ratio of Ig to Ig, is plotted as a function of collector current J, \nin Fig. 4. It is seen that the following empirical relation\u201d matches the \nexperimental result (Fig. 4):\n\nwhere herp max 18 the maximum value of hrz , Ic max 18 the value of Ig \nat which hyg max occurs, and a is a constant.\n\nThe avalanche nonlinearity is due to avalanche multiplication which \noccurs at higher collector-to-base voltage. It is determined from the \ncollector characteristic which is a plot of collector current (J,) and \ncollector-to-emitter voltage (Vex), (Fig. 5). The empirical Miller\u2019s \navalanche multiplication factor is given by\n\nwhere V\u00a2zo is the sustained voltage, and the exponent n is determined \nby experiment. From expressions (13) and (14), the ratio IgusIz is \ngiven by\n\nThe ac 7, can be expressed in terms of 2, and v,,[v3 \u2014 v,] by a Taylor\u2019s \nseries expansion of (15). Since 7, is a function of emitter voltage v, , 7, \nis represented by a current generator g(v2 , v3 \u2014 v1); for convenience in \nnotation it is separated into a linear term g,(v2., v3 \u2014 v,), a second- \ndegree term g2(v2, v3 \u2014 0) and a third-degree term g3(v2, v3 \u2014 2). \nThe linear term equals M o(arte )2 + M, (v3; \u2014 v,). The second-degree \nterm is given by a.M/,K? (v2)\" + mo(v3 \u2014 1)? + (a,M,)Kyvo(v3 \u2014 04). \nThe coefficients a, , a2, MZ,, m2, etc., and the third-degree term are \ngiven in Appendix B.\n\nThe collector capacitance is a nonlinear function of collector-to-base \nvoltage (Veg) since the depletion layer width is a function of Vo, . \nThe exact functional relationship is determined by plotting the common- \nbase imaginary part of h.. as a function of collector-to-base voltage \n(Vez) as shown in Fig. 6.\u00b0 It is evident from the figure that C, follows \nthe 1/3 voltage law (Ref. 19; Equation 5-96);\n\nFig. 6\u2014 Collector capacitance nonlinearity \u2014 calculated and measured curves.\n\nThis nonlinearity is represented as a frequency (differentiation) and \nvoltage-dependent current generator as follows:\n\nThe above nonlinear current generators are incorporated in the linear \nequivalent circuit as shown in Fig. 2. The linear equivalent circuit \nparameters are obtained from the equivalent circuit characterization. \nThey can, for example, be computed from the h-parameters at different \nfrequencies. In general, the distortion is not a critical function of the \nlinear parameters. (Figs. 14 to 17).\n\nAll the nonlinear coefficients (K, , a: , m2 , etc.) are easily obtained \nfrom a simple computer program. The parameters to be specified along \nwith typical values for transistor type A-2486 are listed in Appendix C.\n\nThe Volterra series method is applied to the model to compute the \nsecond and the third harmonic distortion. The voltage at each node is \na nonlinear frequency-dependent function of the input voltage. Each \nnodal voltage is expressed by a Volterra series expansion of the generator \nvoltage; since the nonlinearities are small only three terms are con- \nsidered. The kernels at each node are determined from Kirchoff\u2019s current \nequations.\n\nThe Kirchoff\u2019s current law is applied at each node; the currents are \nnext expressed in terms of the generator voltage v, , the three nodal \nvoltages v, , ve , and v3 , and the known linear and nonlinear parameters. \nThe impedances are represented by their transforms and o denotes \nthat it operates on the voltage across it. The nodal equations are given \nbelow.\n\nSince each nodal voltage is to be expressed in terms of three Volterra \nkernels, there are nine unknown Volterra kernels to be determined from \nthe three equations. The problem of solving for nine unknowns from \nthree equations is resolved by noting that the polynomials x, x\u201d and \nz are linearly independent; hence, each degree term is separately \nand successively solved. The linear kernels are first determined; then \nthe second-degree kernels are determined in terms of the linear kernels; \nlastly, the third-degree kernels are evaluated in terms of the first- and \nsecond-degree kernels.\n\nLet A,(s), Bi(s), Ci(s) denote the transforms of the linear kernels \nat nodes one, two and three, respectively. From the nodal equations \n(18) to (20), the following vector matrix equation is derived.\n\nEquation (21) is solved by inverting matrix Pz(s) and post-multiply- \ning by the vector \n1 \nZ,(s)\n\nThe second-degree terms are equated next in (18) to (20). There are \ntwo types of second-degree terms; those arising from the unknown \nsecond-degree kernels [for example, (s; + s,)C,A2(s;, S2)] and those \narising from the known nonlinear coefficients and the known linear \nkernels [for example, K. T{_, B,(s;)]. The terms associated with the un- \nknown second-degree kernels are the same as were associated with the \nunknown linear kernels in (21), but at the harmonic frequency (s; + $2). \nThe following vector matrix equation is obtained for the second-degree\n\nwhere g. and fz represent the second harmonic contribution due to \nJo(V2 , V3 \u2014 \u00a5,) and y.(v3 \u2014 v2); hence,\n\nThe vector on the left side of (23) is known. Thus, the unknown kernels \nare determined by inverting the matrix P;(s, + s,) and post-multiplying \nby the vector on the left-hand side of (28). When s, = jw, , the inversion \nof the matrix and the post multiplication by the vector can be done \nnumerically.\n\nThe procedure for obtaining the third-degree kernels is almost the \nsame; the significant difference is that the vector on the left side not \nonly contains terms arising from the third-degree nonlinear parameters \nbut also includes second-degree coefficients which give rise to third- \ndegree terms by the interaction of the first- and the second-degree \nkernels. These interaction terms are denoted by Koz , Jos \u00bb Yo3 , respec- \ntively. For example, K3(B,) = K; II}_, Bi(s;), whereas K.3 = 2k.B,(s,) \nB.(s2, 83) which shows the interaction of the first- and the second- \ndegree kernels. The third-degree kernels are derived from the following \nequations:\n\nA computer program has been developed which calculates the kernels \nand the second and the third harmonic distortion. It uses existing pro- \ngrams to invert the matrix P,(s). The nonlinear coefficients are com- \nputed from the known and measured parameters. Computed and meas- \nured results at different currents are given in Fig. 7. The program has \nbeen extended to common-base and common-collector configurations.\n\nV. SIMPLIFIED DISTORTION EXPRESSIONS, THEIR PHYSICAL SIGNIFICANCE \nAND COMPARISON WITH EXPERIMENTAL RESULTS\n\nAnother advantage of the Volterra series method is that it permits \nderivation of closed-form expressions for second and third harmonic \ndistortion. These equations show the interaction between the various \nnonlinear parameters and the effect of frequency.\n\nThe model includes the base resistance (r,), the emitter resistance \n(r.,), the diffusion capacitance (C,), the load (R,) and the source im- \npedances Zg(s), and three nonlinearities, namely, exponential, ava- \nJanche, and hy, nonlinearities. In the computer program Cy, Ci. ,\n\nr., C,, m, and collector capacitance nonlinearity have been taken into \naccount. The expressions given below are for the common-emitter \nconfiguration.\n\nIn the third harmonic distortion term given below, the interaction \nterms due to the first- and the second-degree kernels have not been in- \ncluded mainly to reduce the complexity; in certain cases, they may be \nsignificant.\n\nThe interaction of different nonlinearities and their dependence on \nload impedance, source impedance, bias currents, bias voltage and \nfrequency is indeed somewhat complex. However, the closed form ex- \npressions derived above give a general qualitative picture which will \nbe discussed now.\n\nIt is important to know the effect of frequency on distortion. The \ndistortion depends not only on the frequencies of the fundamental tones \nbut also on the harmonic frequency of interest. As shown in Fig. 8,\n\nM, due to a + b product is better than M, of a \u2014 b product by 10 dB \nwith the two tones at 8.382 and 7.266 MHz. These measurements were \nmade with the transistor biased at 100 mA, 10V and with R, = 50Q \nand R, = 50. In curve (a) of Fig. 8, the fundamental tone was in- \ncreased from 2 MHz to 10.5 MHz and signals at 2f and 3f were measured. \nIt is seen that both M7, and M; improved with increase in frequency. \nA theoretical explanation on the basis of dominant terms (in this range \nof parameter values) in (28) and (29) is given below. In (28) as well as \nin (29) the terms in brackets are multiplied by a frequency-dependent \nterm\n\nIn this range of frequency (s = harmonic frequency), if K,(1 \u2014 a,) S \n| sC, | and if | (r, + R,)sC, | > 1 but K, > | (sC2) |, then the above \nterm reduces to K,/sC, which decreases with increase in frequency. \nHowever, the avalanche terms (M 2, M3, etc.) involve the terms sC, \n[in (28) and (29)] and S;C, in the numerator. Thus, if the avalanche \nterms are dominant, as at higher voltages, there should be no net con- \ntribution due to avalanche terms alone. The exponential terms (K.,/(K,)\u00b0 \nand K;/(K;)*] are multiplied by the factor\n\nThis term is independent of frequency if (sC.(r, + #,) + 1) > 1. \nThus the above discussion shows that distortion will improve with \nincrease in frequency at lower voltages and if | sC.(r, + R,) +1 |< 1. \nTo verify this statement, the voltage was increased to 20 volts and the \ninput resistance changed to 225. The plots of M, and M; with fre- \nquency, as measured, are given in curves labeled (b) in Fig. 8. It is \nseen that 17, and M; do not improve with increase in frequency. The \nsmall improvement can be attributed to the hz terms.\n\nIn general, increase in frequency increases distortion; this is especially \ntrue for the common base configuration. But as shown above, for certain \nranges of frequency and certain values of source impedance, distortion \ncan improve with frequency.\n\nThe load resistance is an external parameter which the circuit designer \ncan vary; hence, it is useful to know its effect on distortion. The second \nand the third harmonic terms are multiplied by 1/V R, and 1/R, ~\n\nterms, respectively; it shows that the distortion can be reduced by in- \ncreasing R,. However, the avalanche terms M@,, M., and I, are \nmultiplied by the R; term, so that increasing R, will increase the con- \ntribution from the avalanche terms. Thus, an increase in R, may in- \ncrease distortion or reduce it due to cancellation. (The contribution from \nthe collector capacitance terms also increases with increase in load (fz).) \nBecause of the above interaction, for a given set of parameters and \ninput frequencies and the harmonic frequency of interest there exists \nan optimum load RP, ; this, of course, can be determined using the com- \nputer program. In Fig. 9, the measured values of MM, and M; at different \nvalues of R; are plotted; in both cases increasing R, reduces distortion \nuntil the optimum value is reached and then distortion increases with \nincrease in Rf, .\n\nSource impedance is another important external parameter. The \nsource impedance affects the exponential nonlinearities K,/K{ in (28) \nand K;/K? in (29) by the factor\n\nAt low frequencies, this nonlinearity is reduced by the factor 1/[(1 \u2014 a) \n(R, + r,)K, + 1]. Thus, an increase in R, will reduce distortion from \nthis source. However, the contribution from other nonlinearities are \nincreased by | \n1+kK iE, a 1's) . \ni he, ad ep\n\nIt is seen from this expression that if K,(R, + 7,)(1 \u2014 a,) is greater \nthan 1, the other nonlinearities are not affected by the increase in R, . \nThus, if the exponential term is dominant, increasing #, reduces dis- \ntortion at low frequencies. At higher harmonic frequencies if \n| sC.(Z,(s) + 7r,) | > 1, the distortion terms are independent of the \nsource impedance since the [r, -+ Z,(s)] term in the numerator and \ndenominator cancel. This is well illustrated in the measured results of \nFig. 10. The second harmonic frequency being 0.7 MHz, | sC.(R, + 7) | \nis not much greater than one up to R, = 1000; hence, the second har- \nmonic distortion improves with increase in source resistance up to \n140Q. Further increase in ?, does not cause much change in distortion. \nThe third harmonic frequency 1s 17.3 MHz; hence, a change in R, does \nnot affect M7; appreciably. (| (sC.)-(R, + 7.) | > 1)\n\nIncrease in bias current usually reduces distortion due to the following \nreasons. The increase in emitter bias current reduces the exponential\n\nit becomes zero at Ig = Ig max/e, and then becomes negative, and \nincreases with further increase in J, . The coefficient a; decreases with \nbias current J,,. Thus, in general, an increase in bias current has the \neffect of reducing both the second and third harmonic distortion (Fig. 7) \n(at least until a. = 0).\n\nWhereas exponential and hp, terms are functions of bias current, the \navalanche and collector capacitance nonlinearities are affected by the \nbias voltage. The coefficient 7, increases with the voltage; but 17, and \nM, increase much more rapidly (Fig. 12). (Both the collector capacitance \nnonlinear coefficients y. , y3 decrease with the increase in bias voltage.) \nThe effects of change in bias voltage are especially pronounced at higher \nload resistance since avalanche (and collector capacitance) terms be- \ncome dominant. The third harmonic distortion decreases more with \nthe increase in voltage (Fig. 18) than the second harmonic distortion \ndoes.\n\nThe physical significance of the closed form expressions has been \nqualitatively discussed. Precise quantitative estimates can and have \nbeen obtained using the computer program. For example, the effect \nof varying linear parameters by fifty percent of their original values was \nstudied. The results show that the distortion does not critically depend \non the linear parameters (Figs. 14 to 17). The other transistor parameters \nsuch as Ig wax, Vezo, 1, etc., can also be varied.\n\nIt is often stated that in a multi-stage amplifier, the output stage \nalone determines the over-all distortion. Even though this statement \nis true to some extent, it is frequently found in practice that the effects\n\nof the previous stages cannot be ignored and sometimes the previous \nstage is dominant. This is especially true if both minimum noise figure \n(which requires lower bias current) and modulation requirements are \nto be met by a two stage amplifier. Two analysis tools based on Volterra \nseries are presented here which enable the study of such cascaded stages.\n\nThe first approach makes use of the cascaded formulae mentioned \nearlier; this method illustrates the cascade phenomenon and permits \nderivation of simple cascade rules.\n\nConsider two cascaded transistors (Fig. 1); let the output voltage \n(v2) of the first transistor be denoted by D(v,); the output voltage (v3) \nof the second stage by H(v.) and F(v,). The aim is to compute the kernels \nFi(s1), Fo(s, , 82), F'3(s; , 82, 83) knowing D and E. To calculate D(s;), \netc., it is necessary to know the load impedance of the first stage which\n\nis the input impedance of the second transistor. This can be computed; \nthus, for a given generator impedance and bias conditions, D(s,), \nD2(s1 , 82), D3(s; , 82 , 83) can be determined. H(s,), H2(s; , 82.) and \u00a33(s; , \nSo , 83) can be computed for a given load and bias conditions with R, = 0 \n(voltage v. is directly impressed across the second). Now expression \nv3 in terms of v, is given by\n\nIt is seen that F is related to H and D by the cascade formulae whose \nphysical significance is discussed below.\n\nwhich states that the overall gain in dB is the gain of the first stage \nin dB plus the gain of the last stage in dB.\n\nThe first term of the formula states that a given harmonic product \nfrom the first transistor D,(jw, + jw,) is amplified by the second transis- \ntor at the harmonic frequency E,(jw, - jw,). The second term shows \nthat the two fundamental tones are amplified by the first transistor \n[D1 (jw,)D,(+jw,)] and then the second transistor acts on these tones \nto produce distortion F,(jw, , jo).\n\nThe second term is the second harmonic distortion of the last stage. \nThe first term expresses the contribution from the first stage; it approxi-\n\nThis shows that if the gain of the last stage is high, the contribution \nfrom the first stage is small. Equation (85) is approximate in two re- \nspects; it neglects the frequency effects and the phase addition of the \ncontributions from the first and the second stage. In (85), the second \nstage gain in question is actually the ratio\n\nwhich involves the two fundamental and the harmonic frequencies. \nAs an example, a shaping network which was introduced increased the \ngain (18 dB) at the harmonic frequency (0.7) MHz) and decreased the \ngain at fundamental tones 15.2 MHz (8 dB) and 14.5 MHz (8 dB) \nwith the result the overall distortion was poorer by 34 dB.\n\nThe first term shows that the third harmonic product of the first stage \n[D3(si , S2 , S3)] is amplified by the last stage at the harmonic frequency \n[E,(s, + se + 83)]. The second term is the interaction term; it arises \nwhen the second-degree kernel of the last stage [E.(s, , s. + s3)] acts \non the sum of the fundamental [D,(s,)] and the second harmonic output \nof the first stage [D.(s2 , s3)]. The last term shows that the second stage \nthird-degree kernel [E3(s, , s2 , 83)] acts on the fundamental tones am- \nplified at the respective frequencies by the first stage [D,(s,)D,(s2)D, (ss) |.\n\nFrom (86), the overall third harmonic distortion is related to that \nof the individual transistors by\n\nThe first term is the contribution from the third harmonic term of \nthe first stage; it is given approximately by\n\nbeers harmonic | -- pew harmonic erage \nof the first stage in dBm of the second stage in dBm\n\nThe third term in (87) is the third harmonic distortion of the last stage \nin dBm.\n\nIt is seen that the effect of the first stage and the interaction term \ncan be reduced by increasing the gain of the last stage. Equation (389) \nillustrates that the second harmonic distortion of each stage should be \ngood. This may become a limitation if the first stage is biased at lower \ncurrents.\n\nIn the above simplified expressions [(88) and (39)] phase addition and \nfrequency effects have not been considered. In (88), 2 (gain in dB) \nactually represents\n\nIn (39) the second harmonic distortion is to be measured with two \ntones, one at the fundamental and the other at the harmonic frequency\n\nMoreover, the kernel must be made symmetrical by taking the average of \nthree possible combinations.\n\nThus, the simplified expressions (35), (88), and (39) are exact if the \ntransistor performance is not frequency dependent; in general, they\n\ncan be used to get a qualitative picture. Equations (34) and (87) are \nindeed exact and take into account the frequency dependence. The \ncomputer program is being extended to calculate (34) and (387).\n\nAn alternate approach to calculate the distortion of cascaded stages \nis to analyze the nonlinear equivalent circuit of cascaded transistors \nusing the nodal technique illustrated in Section IV. The nodal equations \nare derived first; next each nodal voltage is expressed in terms of the \nVolterra series of the input voltage; the resulting vector matrix equations \nare successively solved. Since the procedure is similar, the details are \nomitted.\n\nTwo common-collector stages were cascaded using this approach. \n(Fig. 18). The measured values at 120 mA, 10 V with 75 ohm source and \nload impedances were \u201487 dBm and \u2014112 dBm for the second and \nthe third harmonic distortion, respectively. The computed distortion \nvalues are \u201488.7 dBm for second and \u2014116.6 dBm for third harmonic \ndistortion. Thus, good agreement with experimental result 1s obtained.\n\nThe cascade formulas are simple, physically meaningful and yield \nrules of thumb to judge the effect of the first stages. The nodal approach \nis more complicated. However, the advantage of the nodal approach \nis that it is general and can be used for an amplifier. For example, a \ncascade of common-emitter and common-collector stages involves five \nnodes; if shunt feedback is used at the input and at the output, the \nsame program can be used to analyze this amplifier. (Cascade formulas \ndo not take feedback into account.) In general, the nodal approach can \nbe extended to study frequency-dependent nonlinear network with \nn nodes, if the nonlinearities are small.\n\nIn the initial design of L4 repeater a common-emitter\u2014common- \nemitter\u2014common-collector configuration\u201d\u00ae was used in the power am- \nplifier. The third harmonic modulation performance was not as good as \ndesired. This led first to the study of the output common-collector stage. \nAs shown in Fig. 19, the increase in source impedance increases the \ndistortion of the common-collector stage. Since the preceding common- \nemitter stage output impedance is high, the common-collector per- \nformance was not optimum. Secondly, the preceding common-emitter \nstage was studied because the gain of the common-collector stage is low. \n(see Section VI) As shown in Fig. 20, increase in load impedance beyond \noptimum R, degrades its performance radically. Since the common- \ncollector input impedance is high, the common-emitter stage perform- \nance was not optimum either. Thus, in the redesign work by Ken \nTantarelli, the common-collector output stage is not being used.\n\nAnother interesting application feature was the Improvement in \nmodulation performance of the common-emitter stage with increase \nin voltage. As shown in Fig. 8, it is a function of load impedance, and \nat about 1500, maximum improvement was obtained.\n\nNew coaxial systems are currently being studied to operate at higher \nfrequencies. Different configurations have been examined for the output \nstage. The model showed that common-collector and common-base \nperformance is poorer with an increase in frequency and thus the use \nof these stages as output stages was questioned (unless transistors with\n\nhigher f,\u2019s are available). Recently when a new, high-frequency modula- \ntion test set was built, experiments confirmed the prediction. The third \nharmonic coefficient (173) for a + b \u2014 ec product was 8 dB poorer at \n36.5 MHz (due to signals at 36.5 MHz, 40.1 MHz, and 43.1 MHz) \ncompared to the value at 17.3 MHz (due to signals at 14.5, 15.2, and \n16.6 MHz). The common-emitter configuration modulation performance \nsuffered only about one dB degradation.\n\nThis paper has presented a useful analysis tool for investigating the \nfrequency-dependent nonlinear behavior of transistors. A digital pro- \ngram for all the three configurations has been prepared. The results \nobtained compare favorably with experimental results. The closed-form \nexpressions yield a qualitative picture of distortion. The Volterra series \nproves useful in examining cascaded transistors; a few rules of thumb \nare derived and a general nodal analysis which can be extended to \ncascaded stages with feedback is developed. The practical applications \ncited show that the technique can be useful in the computer-aided \noptimal design of linear transistor feedback amplifiers.\n\nThe author acknowledges gratefully the cooperation he received from \nLee C. \u2018Thomas who is responsible for much of the initial development \nof the model; he has been particularly concerned about the cancellation \nmechanism which he observed in the analog computer simulation. The \nauthor wishes to thank Mr. Jack Huang, who first suggested the model, \nMiss J. A. Nicosia, who wrote the program, Mr. F. Kelcourse for many \nuseful discussions, and Mr. R. E. Maurer for reading the draft. The \nauthor would also like to thank Dr. F. H. Blecher for his encouragement \nand continued interest and guidance.\n\nIf the system is physically realizable, c.(\u00a2 \u2014 7,,\u00a2\u2014 72) = 0, for 7; > 1. \nHence, the limits can be extended to o.\n\nIntroducing dummy variables \u00a2, and t,, the two-dimensional trans- \nform is taken\n\nFor the second-degree case, consider two sinusoidal signals at fre- \nquencies f, and f,. The input x(7) equals, \n_ | exp (wat) + exp (\u2014 jw.) exp (Jw,7) + exp (\u2014jos7) \nAG Pe | sowecrerane= geek a= ae te eG \n(45) \nFrom (41)\n\nThis term occurs twice as does its complex conjugate. \nHence, the output due to the a + b term alone is\n\nThe 2m, term and its conjugate occur only once in (46); hence, it is \n6 db better. The response of the third harmonic kernel to three sinusoidal \ninputs is similarly treated.\n\nFor the system shown in Fig. 1, the cascade formula are given below. \nThe cascade relations can be symbolically written as\n\nF'4(s, , S2 , 83) = Ey(s, + 82. + 83)D3(s, , Se , 83) \n+ 2E.(s, , 82 + 83)Di(81)De(Se , 83) + E3(s: , 82 , 83) UJ Dy,(s;). (58) \nA physical interpretation of the formula for cascaded transistors is \ngiven in Section VI. The procedure for deriving the cascade relation is\n\nas follows: the output Z(t) of the last stage is expressed in terms of the \nVolterra series of its input. (Only two terms are considered)\n\n+f [Pet\u2014n,t\u2014 2) ued dr. (54) \nThe output of the first stage y(t) is related to its input by \nu(r) = [alr = a)2(e) do\n\nSubstituting (55) in (54), terms of the same degree are collected; as an \nexample, the first second-degree term equals\n\n2 \nfa ex(t = T) ll d,(r So Oy t \u2014 To) I] x(a;) do, . (56) \nt=1 \nTaking the two-dimensional transforms yields \n2 \nEy (8, + 82)Da(s, , 82) II X(s;). (57)\n\nA two-dimensional Taylor\u2019s series expansion of (58) is taken; 7, is \nexpressed by K(v.) and vez = vz \u2014 0, . Hence,\n\nThe coefficients m, , mz , and m, equal m; = Ig(M;/M,); 7 = 1, 2, 8, \nwhere I is the collector de bias current. \nThe hrz coefficients are given below:\n\nFrom (62) for g3(v2; v3 \u2014 v1), g3 18 obtained by replacing B,(s;) for \nv, and C,(s;) \u2014 A,(s;) for (v3 \u2014 v,); moreover, the kernel must also be \nsymmetrical. Since the procedure is the same as for g, it is omitted. \nThe interaction terms are given below:\n\nA2486 is an n-p-n silicon transistor with overlay type of construction. \nIts fr ranges from 800 to 1000 MHz. It is a power transistor with current \ncapability of 1 amp and can handle 2.2 watts of power.\n\n. Lotsch, H., Survey. of Nonlinear Distortions in Transistor Stages Including\n\n. Reynolds, J., Nonlinear Distortions and Their Cancellation in Transistors,\n\nMeyer, N. I., Nonlinear Distortion and Small Signal Parameters of Alloyed \nJunction T'ransistors, Danish Science Press Ltd., Copenhagen, 1960.\n\nThomas, L. C., Broadband linearization of Transistor Amplifiers, Presented at \nInternational Solid State Circuits Conference, February, 1967.\n\nNarayanan, S., Transistor Distortion Analysis Using Volterra Series Repre- \nsentation, (Oral presentation at International Conference on Communica- \ntions, June, 1967).\n\nVolterra, V., Theory of Functionals and of Integral and Integrodifferential \nEquations, Dover Publications, New York, 1959.\n\nWiener, N., Response of a Nonlinear Device to Noise, MIT, Radiation \nLaboratory, Cambridge, Mass., Report No. 129, (V-16), April, 1942.\n\nBose, A. G., A Theory of Nonlinear Systems, Technical Report 309, Research \nLaboratory of Electronics, MIT, May 15, 1956.\n\nWiener, N., Nonlinear Problems in Random Theory, MIT Press, Cambridge, \nMass., 1958. \nBrilliant, M. B., Theory of the Analysis of Nonlinear Systems, Technical \nReport 345, Research Laboratory of Electronics, MIT, March 3, 1958. \nGeorge, D. A., Continuous Nonlinear Systems, Technical Report 355, Re- \nsearch Lab of Electronics, MIT, July 22, 1959.\n\nBarrett, J. F., The Use of Functionals in the Analysis of Nonlinear Physical \nSystems, Statistical Advisory Unit, Report No. 1/57, Ministry of Supply, \nGreat Britain, 1957.\n\nBush, A. M., Some Techniques for the Synthesis of Nonlinear Systems, Sc-D \nThesis, Dept. of Electrical Engineering, MIT, Cambridge, Mass., May, 1965.\n\nNarayanan, \u00a7., Transform Methods for Special Nonlinear Systems, Ph.D. \nThesis, Carnegie Tech., Pittsburgh, Pa., May, 1965. .\n\nFlake, R. H., Volterra Series Representation of Time-Varying Nonlinear \nSystems, Proc. Second International Congress of IFAC on Automatic \nControl Based, Switzerland, Paper No. 408/1, 1963.\n\nKu, Y. H. and Wolf, A. A., Volterra-Wiener Functionals for the Analysis for \nNonlinear Systems, J. Franklin Inst., 281, No. 1, January, 1966, pp. 9-26.\n\nKelcourse, F. C. and Labbe, L. P., Transistor Feedback Amplifiers for 0.5 \nm/c-20 m/e Long Haul Coaxial Cable Transmission System, IEEE NEREM \nConference, 1964.\n\nThis paper considers the effectiveness of error-correcting codes for the \ntransmission of numerical data. In such a situation, errors in the nu- \nmerically most significant positions of a message are of greater con- \nsequence than are errors in the less significant positions. A measure \nof transmission fidelity based upon the average magnitude by which the \nnumbers delivered to the destination differ from the transmitted numbers \n1s developed and 1s referred to as the average numerical error (ANE). Codes \nare compared by comparing the ANE that results from their use.\n\nSignificant-bit codes are defined and the ANE resulting from their use \nis determined. For constant-symbol-rate transmission, the relative effect- \nweness of various coding schemes is analyzed when the error probability \nin the channel ts small, The ANE resulting from the use of certain specific \ncodes is numerically evaluated and compared.\n\nThe usual approach to coding is to ignore the actual meaning of the \ntransmitted symbols and to represent them in a purely statistical \nmanner. As a result, all message errors are assumed to be equally \ncostly and codes have been sought that simply reduce the probability \nthat a message is received in error.\n\nWhile this may.be appropriate for the transmission of some types \nof data, there are situations in which other criteria of goodness are \nof greater merit. If, for example, one is interested in the transmission \nof the temperature of a satellite, the probability that a particular \nobservation is transmitted incorrectly may have little direct relation \nto system performance whereas a measure of the average magnitude \nby which the received data differ from the data actually transmitted \ncould prove useful.\n\n* The material presented i in this paper is based upon the dissertation, Coding \nfor Numerical Data Transmission, submitted by the author to The Johns Hopkins\n\nUniversity in conformity with the requirements for the degree Doctor of \nPhilosophy.\n\nThis paper develops a criterion of transmission fidelity for numeri- \ncal data transmitted over a binary symmetric channel based upon the \naverage numerical error which occurs. Significant-bit codes are de- \nfined and the average numerical error resulting from their use is de- \ntermined for a binary symmetric channel with independent errors. For \nconstant-symbol-rate transmission, the relative effectiveness of various \ncoding schemes is analyzed when the probability that a symbol is \nreceived in error is small. In order to obtain a feeling for the utility \nof coding, the average numerical error resulting from certain specific \ncodes is numerically evaluated.\n\nThroughout this paper, the channel is taken to include all operations \nperformed upon the symbols during transmission. A binary symmetric \nchannel is defined to be a binary channel such that\n\n(z) the channel always gives one of the binary symbols at its output, \n(2) the probability that any particular sequence of errors occurs is \nindependent of the symbols transmitted.\n\nIn some sections, we shall consider a binary symmetric channel with \nindependent errors. This is a binary symmetric channel for which \nthe errors occur independently with probability p where 0 S p S 4 \nand p = 1 \u2014 4g.\n\nThe elements of the Galois field of two elements are denoted by 0 \nand 1. Let the symbol G denote component by component modulo 2 \naddition of vectors (or n-tuples) whose components are field elements. \nThe set of all such vectors forms a vector space I of dimension n over \nthe field of two elements. Because a field element can be viewed as a \nvector with one component, @ will also be used to denote the addition \nof field elements.\n\nA binary group code V is a subset of I which forms a group. Over \nthe field of two elements, any set of n-tuples that forms a group is \nindeed a vector space. Therefore, a binary group code V forms a sub- \nspace of I. The dimension of V is k.\n\nThe implementation of a binary group code can be viewed in the \nfollowing manner. The encoder receives & binary information symbols \n(called a message) from the source and determines from the message \n(n \u2014 k) binary parity check symbols (called an ending). The message \nand ending may be interleaved or transmitted sequentially to form a \nblock of length n (called a code vector). The decoder operates upon\n\nthe blocks of n binary symbols coming from the channel in an attempt \nto correct transmission errors and provides k binary symbols at its \noutput. The notation (7,k) is used to denote such a code.\n\nConsider the message (m,, \u2122Mz-1, *** , \u2122,). The code vector used \nto transmit this message will have m, , m,-1, -\u00b0* , m, 1n the k informa- \ntion positions. The (n \u2014 k) parity check positions that form the ending \nare denoted by \u00e9; , \u20ac2, ++ , \u20acn,-,- Lhe order in which the information \npositions and the parity check positions are arranged for transmission \nis arbitrary.\n\nLet H denote the parity check matrix for a binary group code. H \nis an (n \u2014 k) X n matrix whose entries are field elements. An n-tuple \nv is a code vector if and only if\n\nwhere H denotes the transpose of H. H can be written in a form such \nthat each column of H that corresponds to a parity check position in a \ncode vector is a distinct weight* one (n \u2014 k)-tuple. When this is done, \nlet C,(1 S | S k) denote the column in H that is in the position that \ncorresponds to position m, in a code vector.\n\nFor a binary symmetric channel, the order in which symbols are \ntransmitted can affect code performance. For the binary symmetric \nchannel with independent errors, the order in which symbols are trans- \nmitted does not affect performance. In the latter case, we can write \nH as\n\nA system for transmitting observations performed upon some physical \nprocess over a binary channel is shown in Fig. 1. So that the relation- \nship between the observed numbers and the code will be clear, a general \nformulation will be presented.\n\nIf each quantization step is of uniform size, the quantizer output \ncan be represented as A -++ Bz where A and B are constants and the \ninteger 7 indicates the quantization level. The \u2018\u2018source scale-to-binary \nconverter\u201d receives A + Biz from the quantizer and transmits 7 to \nthe encoder. The \u2018binary-to-source scale converter\u2019? receives some \ninteger 7 from the decoder and delivers A + Bj to the destination.\n\n* The weight of a vector v is the number of nonzero components in v and is denoted \nby wiv]. The distance between two vectors wu and v is wlu @ v]I.\n\nLet Pr {7 | 7} be the probability of receiving 7 at the decoder output \nwhen 7 served as the encoder input and let Pr {z} be the probability \nthat 7 is sent. The average numerical error (ANE) that occurs is\n\nIf all values of 7 are equally likely to be observed and if the range \nforzisO Si < 2\u00b0 \u2014 1, Pr {z} = 2. The range for 7 is thus0 Sj S \n2\" \u2014 1 and (3) becomes\n\nBr ok \nANE = & > yi \u00e9| Prt tj | a. \n7=( j= \nBecause B is a constant not dependent upon the particular coding \nscheme implemented, B may be set equal to 1 when comparing the \neffectiveness of different codes. Accordingly, we shall consider the \nexpression\n\nANE = laa eee. (4) \nFor a specified value of k, a given coding scheme is considered perferable \nto some other coding scheme if the ANE resulting from the implementa- \ntion of the given code is less than the ANE resulting from the alternative \ncode.\n\nThe code enters (4) through the terms Pr {j | 7}. Thus, for a binary \nsymmetric channel, the ANE will, in general, be dependent not only \nupon the error statistics of the channel but also upon the order in which \nthe symbols are transmitted.\n\nIt is possible to simplify (4) to an expression that involves terms of \nthe form Pr {7 | 0} exclusively. This reduces the number of terms by a \nfactor of 2* and demonstrates that knowledge of the error probabilities \nconditional upon zero being sent is sufficient to evaluate the ANE. \nHowever, it is necessary to develop some notation and to present two \nlemmas before proceeding to simplify (4). The proofs of the lemmas are \nomitted because the lemmas follow from the group property of the \ncode.\n\nWhen the integer 7 is to be sent, let us assume that the message \nultilized is the k-bit binary representation of 2 (which is denoted by \nB(z)) such that\n\nThe ending H; = (\u20ac,, \u20ac2, \u00b0** \u00bb \u20acx-x) required to encode B(z) 1s chosen \nso that the resulting code vector C'(z) satisfies (1).\n\nLemma 2: Let B(l) = Bit) B BY) as in Lemma 1. For fixed 1(0 S \nt < 2\u00b0 \u2014 1), as j successively takes on the values 0, 1, 2, --- , 2\u00b0 \u2014 1,1 \ntakes on each of the values in the range 0 S 1 S 2\" \u2014 1 once and only once.\n\nTheorem 1: Let all messages be equally likely to be transmitted and let \nthe channel be binary symmetric (but not necessarily with independent \nerrors). For these conditions, the average numerical error ts\n\nProof: By Lemmas 1 and 2, for each value of 7 and for a specified \nvalue of 1, there will be a unique integer 7, such that Pr {j, | 7} = Pr {/| 0} \nwhere B(l) = Bit) @ B(j,). From (4),\n\njx i) +H \u2014e | = 2-2\" \nBecause of the symmetries involved, \nQk\u20141 2k\u20141 \n2Dli-\u00e9tl= Dll + lH 7 | = 22\" \nThus, (6) becomes\n\nIn (5), notice that Pr {0 | 0} does not appear and that the terms Pr {7 | 0} \nare not weighted linearly in 7 but that the weighting coefficients go \nin steps as powers of 2 with several conditional probabilities having \nthe same weighting coefficient. Notice that the weighting coefficient \nfor Pr {z | 0} is 2\u2019-* where (j \u2014 1) is the largest power of 2 in z. All errors \nwith the same coefficient are of the same seriousness and a good code \nmust reduce these sets of probabilities rather than simply minimize the \nprobability that a few very large errors occur.\n\nBecause the set of messages B(z) (2\u00b0 S$ i S 2\u2019 \u2014 1) gives rise to \nthe set of conditional probabilities whose weighting coefficient in the \nANE expression is 2*~*, we shall call these messages the j-level messages\n\nand the corresponding conditional probabilities, Pr {2** | 0} through \nPr {27 \u2014 1 | 0}, the j-level conditional probabilities. The 0-level message \nis defined to be B(O) and the 0-level conditional probability to be \nPr {0 | 0}.\n\n(2) Component m; in each message is 1. \n(22) Components m,;(j + 1 S 7 S k) in each message are 0. \n(2722) Kivery possible (7 \u2014 1)-tuple occurs once and only once as com- \nponents m, through m,;_, of some j-level message.\n\nFor a perfect error-correcting code used with a binary symmetric \nchannel with independent errors, it is possible to compute the j-level \nconditional probabilities and thus the ANE from a knowledge of the \nweight distribution of the code vectors on each level (these weight \ndistributions have been referred to as level weight structures.)' The \nproblem of efficiently computing the level weight structures from knowl- \nedge of the parity check matrix has been discussed previously.\u2019\n\nIn order to permit the error-correcting capabilities of a code to \ncorrespond somewhat to the significance of the information positions, \nit is possible to formulate a type of code which uses a subcode to protect \nthe (k \u2014 k,) most significant positions of a message and simply transmits \nthe remaining symbols unprotected. The name significant-bit code \n(SB code) is used for this type of code. An SB code is specified by the \nparity check matrix H., and the ANE resulting from the use of an \nSB code is ANE sg, .\n\nThe code utilized to protect the (k \u2014 k)) most significant informa- \ntion positions will be named the base code. Because it is confined to \nthe (k \u2014 ko) most significant positions, we can abstract the base code \nand study it as a separate entity. Accordingly, the base code vectors \nare (n \u2014 ky)-tuples of which the first (k \u2014 k,) positions are the base \nmessages.\n\nAlthough the concept of SB codes is applicable to any binary sym- \nmetric channel, we shall assume independent errors in the following \nanalysis. Thus, from (2), the base code is specified by the base parity \ncheck matrix Hg, where\n\nIn this case, the code vector C(i) = B(z) | EH, where the symbol | \nindicates that C(z) can be partitioned into the k-tuple B(z) and the\n\n(n \u2014 k)-tuple LE; . Let B(z) be partitioned so that B(z) = B\u2019(2\u2019) | B\u2019(7\u2019\u2019) \nwhere B\u2019(2\u2019) denotes the (k \u2014 ky) most significant positions of B(:) \nand B\u2019'(2\u2019\u2019) denotes the k, least significant positions of B(z). Then\n\nThe range for 7\u2019 isO S 7\u2019 S 2\u00b0\u00b0** \u2014 1 and for?\u201d isO S 7\u201d S 2\" \u2014 1. \nLet Pr, {2\u2019 | j\u2019} denote the probability of receiving 7\u2019 when j\u2019 is sent \nusing the base code. By Theorem 1, the ANE for the base code (ANE;,) is\n\nk\u2014ko oi-] \nANU =: D2\" D3 Pig ao}. (9) \nj=1 i\u2019 =2Qin1 \nBecause the base code is used exclusively to protect the (k \u2014 kp) \nmost significant information positions, Hs, must have the form\n\nwhere 0 is used to represent an all-zero column of Hs, and where \nthe C/(1 <= 1 S k \u2014 k,) are the columns of H,. The coset leaders\u201d \nin the standard array\u201d for the SB code must be obtained from the \ncoset leaders in the standard array for the base code by expanding the \nbase coset leaders in length to n-tuples by inserting k, zeros in informa- \ntion positions 1 through k, of the expanded vectors. Because all vectors \nin column 7 of the standard array for the SB code will have B\u2019\u2019(7\u2019\u2019) \nin information positions 1 through fy ,\n\nWe shall now show that ANI.s, can be expressed in terms of the \nproperties of the base code.\n\nTheorem 2: Let the base code be defined as above. For a binary symmetric \nchannel with independent errors and when all messages are equally likely \nto be transmitted,\n\nI'rom Theorem 1, ANEs, = ANH\u2019 + ANE\u201d. \nLet us first analyze ANE\u2019. For 1 S 7 S k), the sum of the 7-level \nconditional probabilities is\n\nyi] \n\u00bb Pr {4 | 0} _ \u00bb poe Oe Lr; {0 | 0} \ni=Qin2 itt egint \nwhere we have used (10) and realized that 7\u2019 = 0 for all messages on \nthis level. Because every (7 \u2014 1)-tuple occurs as components m, through \nm;-, of some j-level message and m; = 1 in every j-level message,\n\n2ko\u2014j \nreggae vane Bleqgee - \n= a pr\u00bb (i ge w | (t\u2019\")} Pr, {oF 1 +. 5 | O}. \ni\u2019\u2019=0 \nAs 7\u201d runs through the range 0 S 7\u201d S 2\u00b0\u00b0 \u2014 1, each possible kp- \ntuple occurs once and only once. Therefore,\n\nBecause of the manner in which the sets were chosen, ANE\u201d can be \nexpanded as\n\nNotice that the situation k = k, can be included in this formulation \nif we define ANE, = 0 and Pr, {0|0} = 1 whenk = k, . Thus, uncoded \ntransmission can be regarded as an SB code in which k = ky.\n\nThe interpretation of (11) is interesting. The quantity >>\", 277-\"pq\" \nis the ANE that results from the uncoded transmission of ko-tuples. \nThus, ANI. gz is the ANE for uncoded transmission of ko-tuples weighted \nby Prz {0 | 0} plus 2\u201d\u00b0 times ANE, .\n\n(11) enables the computation of ANEgs,; from the properties of the \nbase code. Because the base code involves messages of length (k \u2014 ky), \nit is easier to analyze than the entire SB code.\n\nConsider two error-correcting codes which are denoted as V, and \nV.. Let V, be an (m,, &) code and V. be an (n,, k) code where n, \nmay or may not be equal to n,. Let \u00a2, denote the minimum weight \nof the n,-tuples that are not coset leaders in the standard array for \nV,. Similarly, let \u00ab, denote the minimum weight of the n.-tuples that \nare not coset leaders in the standard array for V,.\n\nwhere 7,;; is the number of n,-tuples of weight j in the column headed \nby C(z) in the standard array for V,. Thus, for V,, the average nu-\n\nThus, for p sufficiently small, if e, > e,, ANE, < ANE, and V, \nresults in less ANE than V,.\n\nThe minimum weight of the vectors that are not coset leaders in an \nSB code is 1. Thus, consider two SB codes denoted by V sp; and V gp \nwhere V sp; 1s an (n, , k) code and V spo 18 an (N2, k) code. V gp protects \nthe (k \u2014 ko,) most significant positions and V gg2 protects the (kK \u2014 ko.) \nmost significant positions of a message. By reasoning analogous to \nthat above, for p small, if kp, < ko2 and if the base codes used in V gp, \nand V spe correct all weight one errors, then V s,, results in less ANE \nthan V spe.\n\nWe thus have the following ranking of codes for p small. The ranking \n(in order of increasing effectiveness) assumes that the schemes are \ncompared for the same value of k.\n\n(it) An SB code protecting (k \u2014 ko) positions where k # ky. \n(277) An SB code protecting (k \u2014 ky + k\u2019) positions where k\u2019 > 0. \n(iv) An e-error-correcting code where e 2 1.\n\nTo obtain a feeling for the utility of coding for numerical data trans- \nmission over a binary symmetric channel with independent errors, the \nANE resulting from certain codes for k = 26 will be evaluated for\n\nconstant-symbol-rate transmission. Ref. 3 contains similar information \nfork = 1,4 and 11.\n\nLet ANEyc denote the ANID when no coding is used. Contrary to \nthe concept of code equivalence that is obtained under the assumption \nthat all errors are equally costly (i.e., when probability of message \nerror is used as the measure of code performance), the ordering of the \ncolumns of the parity check matrix can affect code performance. Thus, \nfor the (31, 26) perfect single error-correcting code (PSEC code), every \nordering of the columns of the parity check matrix could yield a distinct \nANE. Upper and lower bounds on the ANE for this code are obtained \nin Ref. 3 and are denoted herein as ANEys; and ANE;z , respectively.\n\nBy numerical computation, the ordering in (14) was found to result \nin as small an ANE as any other ordering tried. The number actually \ntried was by necessity a small fraction of all possible orderings of the \n26 columns. However, notice that C,. through C2, each have a one in \nthe same position thus assuring us that the number of weight three \ncode vectors on levels 12 through 26 will be the theoretical minimum for \nthis code (by Theorem 9 in Ref. 3). For values of p that are of primary \ninterest (less than 107\u00b0), this assures us that it is not possible to find \na different ordering that will result in a significantly better performance \n(although there are other orderings that in fact give equal performance). \nLet ANEp denote the ANE that results from the code specified in (14).\n\nAp = )111100001111000111100011107Z;)- (14) \nL1IDODLLIOOLLOOILOLIOOIIOILIIO!LI \nLOLOLODOLOLOLIOLIOLIIOLIOLOLIOII\n\nIf the columns of (14) are regarded as the 5-bit binary representations \nof integers, then the ordering from left to right corresponds to decreasing \ninteger value (with powers of two omitted because they appear in J;). \nSimilar ordering was observed to be preferable for the (15, 11) PSEC \ncode\u2019 and, by exhaustive search, actually found to be as good as any \nother ordering for the (7, 4) PSEC code\u2019.\n\nTable I compares ANE;; , ANEyg, and ANE,. For convenience \n(and so that the values given will agree with the data plotted in Figs. \n2, 8, and 4), the ANE has been normalized by dividing by 2\u00b0 \u2014 1 \n(i.e., the full-scale value).\n\nThe following SB codes are considered. For each, Hy and the nota- \ntion used for the resulting ANE in Figs. 2, 3, and 4 is given. Theorem \n2 permits the computation of the ANE for these codes from a knowl- \nedge of the base code.\n\nThe ANE is denoted as ANE,3,1) . \nBase Code 2: (5, 1) perfect double error-correcting code.\n\nBase Code 3: This base code uses independent (38, 1) PSEC codes to \nprotect the two most significant information positions. \n1 0 \n1 0 \nlas 01 i;\n\n0 | \nBecause the codes are used independently, the required conditional \nprobabilities for the base code can be readily calculated. The ANE \nis denoted as ANE\u00a2s3 31) ,\u00a23,1) -\n\nBase Code 5: This base code uses a (8, 1) PSEC code to protect the \nmost significant information position and a (7, 4) PSEC code to protect\n\nFigs. 2, 38, and 4 present ANEyo , ANEys; , ANE,: , ANE,p, and \nthe ANE of the SB codes considered. In each case, the ANE has been \nnormalized by dividing by 2\u00b0\u00b0 \u2014 1. For clarity, logarithmic scales are \nused as p decreases from 107\u00b0 until p becomes sufficiently small so \nthat the results for small p apply.\n\nThe following observations can be made for constant-symbol-rate \ntransmission.\n\n(2) Improvements in transmission fidelity are obtainable by the \nutilization of codes. It should be noted that no one code is the most \ndesirable for all p (0 < p < 4) and in some cases the codes that are \nbest for small p turn out to be less effective than uncoded transmission \nfor the larger values of p.\n\n(21) For k = 26, it can be shown that the probability that a message \nis received in error when the PSEC code is used is less (for 0 < p < 3) \nthan the probability that a message is received in error using any of \nthe SB codes considered. Thus, under the criterion of minimizing the \nprobability that a message is received in error, the PSEC code is pref- \nerable to any of the SB codes considered.\n\nHowever, when the ANE is used as a measure of code effectiveness \nfor numerical data transmission, we observe that the SB codes are \npreferable to the PSEC code for certain values of p. Thus, when com- \nparing codes, the ranking obtained using probability of message error \nas the performance index may not correspond to the ranking obtained \nusing ANE as an index. We can conclude that probability of message \nerror and ANE are not equivalent measures of code performance and\n\nthat, in some cases, the ANE can be reduced by using a code whose \nprobability of message error is not minimal.\n\n(71) For k = 26, consider the relative performance of the PSEC \ncode and the SB codes. When p is small, the PSEC code will be effective \nbecause it can correct all single errors (the only type that have much \nprobability of occurring) whereas a single error in certain positions \nof an SB code will result in a message error. For larger values of p, \nthere is an increasing chance that an error pattern will occur which the \nPSEC code cannot correct. The SB codes become effective in this situa-\n\ntion. If multiple errors occur during transmission such that the errors \noccurring in the (k \u2014 ko) most significant information positions and the \ncheck positions form an error pattern correctable by the base code, \nthis will be corrected leaving any errors in the ky least significant \ninformation positions uncorrected. Therefore, the most costly portion \nof a large number of error patterns can be corrected. As p increases, \nthe number of positions in the base code must decrease so that un- \ncorrectable error patterns in the positions covered by the base code \nhave a sufficiently small probability of occurrence so that the base code \ncan operate effectively. In other words, as p increases, more and more \nprotection must be provided for the significant bits so that the most \ncostly errors are prevented.\n\n(wv) For p small, the ANE from uncoded transmission is approxi- \nmately (2\u00b0 \u2014 1)p. For small p, the ANE as a fraction of full scale for \nuncoded transmission is thus very nearly independent of k.\n\n1. Buchner. M. M., Jr.. Computing the Spectrum of a Binary Group Code. \nBS.T.J.. 45, March, 1966. pp. 441-449.\n\n3. Buchner, M. M.. Jr.. Coding for Numerical Data Transmission, Ph.D. \nDissertation, The Johns Hopkins University, Baltimore, Maryland, 1965.\n\nVAcuav E. Brens8, A.B., 1950, Harvard College; M.A. and Ph.D., 1953, \nPrinceton University; Bell Telephone Laboratories, 1953\u2014. Mr. Benes \nhas been engaged in mathematical research on stochastic processes, \ntraffic theory, and servomechanisms. In 1959-60 he was visiting lec- \nturer in mathematics at Dartmouth College. He is the author of \nGeneral Stochastic Process in the Theory of Queues (Addison-Wesley, \n1963), and of Mathematical Theory of Connecting Networks and \nTelephone Traffic (Academic Press, 1965). Member, American Mathe- \nmatical Society, Association for Symbolic Logic, Institute of Mathe- \nmatical Statistics, SIAM, Mind Association, Phi Beta Kappa.\n\nMorcan M. Bucuner, Jr., B.E.S., 1961, Ph.D., 1965, The Johns \nHopkins University; Bell Telephone Laboratories, 1965\u2014. At Bell \nTelephone Laboratories, Mr. Buchner was engaged in a study of im- \npulse noise in an effort to understand its characteristics and its effects \nupon data communications. At present, he is on a military leave of \nabsence and is serving as a Lieutenant in the U. 8. Army Electronics \nCommand, Fort Monmouth, N. J. Member, IEEE, Tau Beta Pi, Sigma \nXi, Eta Kappa Nu.\n\nMyron 8. Guass, M.S. in Physics, University of Chicago, 1926; Bell \nTelephone Laboratories, 1926\u2014. Mr. Glass has been engaged in the \ndevelopment of electron devices and in applied magnetics and optics. \nHe is currently supervisor of a group in the Optical Device Depart- \nment engaged in the development of gas lasers and laser mirrors. \nMember, AAAS; senior member, IEEE.\n\nIra Jacoss, B.S. Physics, 1950, City College of New York; MS., \n1952, Ph.D., 1955, Purdue University; Bell Telephone Laboratories, \n1955\u2014. Mr. Jacobs has been engaged in studies of electromagnetic \nwave propagation in nonuniform and anisotropic media, radar cross- \nsection and antenna analyses, and in missile guidance and detection \nsystems. He is currently Head of the Military Communications Re-\n\nsearch Department in the Detection Systems Laboratory. His current \nactivities are largely in the field of communication theory. During \nthe summers of 1964 and 1966 he was a member of Institute for \nDefense Analyses study groups considering satellite multiple access \nand signal processing. He has served as Project Engineer on a study \nof weak-signal communication techniques and is presently Project \nKingineer on a deep-space communication study. Senior member, IEEE; \nmember, American Physical Society, American Association for the \nAdvancement of Science, Phi Beta Kappa, Sigma Xi, Sigma Pi Sigma.\n\nT. T. Kapota, B.S., 1953, Yokahama National University (Japan) ; \nM.S., 1956, Ph.D., 1960, University of California (Berkeley); Bell \nTelephone Laboratories, 1960\u2014. Mr. Kadota has been engaged in the \nstudy of noise theory with application to optimum detection theory. \nMember, Sigma Xi.\n\nSUNDARAM NARAYANAN, B. Tech., 1960, Indian Institute of Tech- \nnology, Kharagpur (India); M.S., 1963, Ph.D., 1965, Carnegie Insti- \ntute of Technology; Bell Telephone Laboratories, 1965\u2014. Mr. \nNarayanan is with the Coaxial Systems Studies group and is primarily \nconcerned with nonlinear distortion mechanism in transistors and in \ntransistor feedback amplifiers. Member, Sigma Xi.\n\nIF\u2019. N. H. Ropinson, B.A., Cambridge, 1946; M.A., Oxford, 1950; \nD. Phil., Oxford, 1955. At Oxford since 1950, where he teaches physics, \nMr. Robinson\u2019s research has mostly been in the fields of very low \ntemperature and nuclear orientation. As a visitor to Bell Telephone \nLaboratories in 1954-55, he worked on noise in electron beams. Since \nthen he has been a frequent visitor to the Laboratories and spent the \nyear 1965-66 there when he worked on non-linear optics.\n\nW. SHOCKLEY, B.Sc., 1932, California Institute of Technology; \nPh.D., 1936, Massachusetts Institute of Technology; Bell Telephone \nLaboratories, 1936-1955, 1965\u2014. Dr. Shockley is best known as the \ninventor of the junction transistor. For this and other contributions, \nto transistor physics, he received the 1956 Nobel Prize in Physics \njointly with his two former colleagues at Bell Telephone Laboratories, \nJohn Bardeen and Walter H. Brattain. During World War II, on \nleave of absence from Bell, he served as Director of Research for \nthe Navy\u2019s Anti-Submarine Warfare Operations Research Group and \nas expert consultant for the Office of the Secretary of War. He re- \nturned to Bell Laboratories after the war and became director of the\n\nsolid state physics research program. In 1953, he was named Director \nof Transistor Physics Research. During this period he made many \ncontributions to solid-state physics particularly in connection with \nthe transistor. In 1955 he left Bell Telephone Laboratories to join \nBeckman Instruments Ine. where he established the Shockley Semi- \nconductor Laboratory in Palo Alto, California, for research, develop- \nment and production of new transistor and other semiconductor \ndevices. In 1965 Dr. Shockley returned to Bell Telephone Labora- \ntories in the capacity of Executive Consultant. He presently holds \nthe position of Alexander M. Poniatoff Professor of Engineering \nScience at Stanford University. More than 70 United States Patents \nhave been granted for his inventions. Medal for Merit, Office of \nthe Secretary of War, 1946; Air Force Association Citation of \nHonor, 1951; Morris Liebmann Memorial Prize, IRE, 1952; Oliver \nE. Buckley Solid State Physics Prize, American Physical Society, \n1953; U.S. Army Certificate of Appreciation, 1953; Comstock Award, \nNational Academy of Sciences, 1954; Holley Medal, American So- \nciety of Mechanical Engineers, 1963; Wilhelm-Exner Medal, Oester- \nreichischer Gewerbeverein of Austria, 1963. Honorary Doctorates \nfrom the University of Pennsylvania, 1955; Rutgers University, 1956; \nand Gustavus Adolphus College, 1963. Consultant, Scientific Advisory \nPanel of U. 8. Army, Air Force Scientific Advisory Board; Fellow, \nIEEE, American Physical Society, American Academy of Arts and \nSciences; Member, President\u2019s Science Advisory Committee Panel on \nScientific and Technical Manpower, American Institute of Physics, \nSigma Xi, Tau Beta Pi.\n\nSmion M. 8zx, B.S., 1957, National Taiwan University, Taiwan, \nChina; M.S., 1960, University of Washington; Ph.D., 1968, Stanford \nUniversity; Bell Telephone Laboratories, 1968\u2014. Mr. Sze has been \nconcerned with the study of semiconductor device physics. At present \nhe is engaged in studies of metal-insulator-semiconductor devices and \ninterface states. Member, Sigma Xi, IEEE.\n\nWiuuiAM J. Tasor, B.S. (Chemistry), 1953, Rensselaer Polytechnic \nInstitute; A.M. (Physics), 1954, Ph.D. (Chemical Physics), 1957, \nHarvard University; U. 8. Army 1957-1959; Bell Telephone Labora- \ntories, 1959\u2014. His work at Bell Laboratories included research and \ndevelopment of microwave masers, the design of the maser for the \nTelstar\u00ae ground station, and investigation of light deflection tech- \nniques. He is currently involved in a study of domain wall motion \nin magnetic media.\n\nIn a recent brief\u2019 in the B.S.T.J., Zador presents, without proof, \nrealizability conditions for the input impedance of the lossless tapered \ntransmission line terminated in unit resistance. Upon a careful examina- \ntion of the brief, it appears that the conditions are not accurate. The \nfollowing analysis clarifies this point and, incidentally, provides alterna- \ntives to Zador\u2019s necessary conditions.\n\nConsider a nonuniform line (Fig. 1) with inductance per unit length \n\u00a3(x) and capacitance per unit length C(x) such that (to follow Zador)\n\nLet V(a,s) and I(z,s) be the voltage and current along the line with \npolarities as indicated in Fig. 1. The equations of the line are\n\ndV(z,s) \nae s\u00a3(x)I(x,s) \ndI(x,s) \nwe s(x) V(ax,s). \nEliminating J(z,s) and taking into account that \u00a37) = 1/C(x) we get\n\nrespectively. From the reference polarities of the voltages and currents \nin Tig. 1, we see that for a unit resistance termination at x = 0 we must\n\nhave \nV(0,s) = \u2014J(0,s). \nHence, if we impose the condition (following Zador) \ny(0,s) = V(O,s) = -2 \nthen for unit resistance termination we should have \ndV(0,s) _sV@O,s) as\n\n_ _s_ yllys) \n49) = oH y'(is) \nThus, the signs are wrong in Ref. 1. This is not the crucial error however. \nIn this paper, we will show that the difficulties in Zador\u2019s paper \narise from the following facts:\n\n(1) He does not consider the matched line. Unmatched lines tend \nto have almost periodic behavior for large real frequencies and hence\n\nthe network functions do not have limits at infinity. This point will \nbe made more precise in the sequel.\n\n(it) Multiplication of Z(jw) by exp (~\u20142jlw) in property (277) of the \nnecessity statement introduces periodic behavior at infinity even in the \nmatched case.\n\n(4722) Physical meaning has not been attached to the N; and D,. \nThese should obviously be identified with the well-known ABCD \nparameters to correct (27) of the necessity conditions.\n\nProperty (222) in the necessity statement does not appear to be true \nas stated. One can easily construct many counter examples.\n\nExample 1: The uniform line with (following Zador\u2019s notation) c(z) = 1 \nand length 7 = 1, terminated in a 1-ohm resistor. Obviously c(x) satisfies \nthe conditions stipulated by Zador, i.e., c(x) is positive and continuously \ndifferentiable in the interval 0 S x S 1. Clearly the driving point \nimpedance is\n\nClearly cos 2w does not have a limit fora \u2014 +o. \nConsider now a less trivial counter example.\n\nExample 2: The exponential line terminated in a unit resistance. With \nZador\u2019s notation c(z) = exp 27, and! = 1. In this case by solving Zador\u2019s \n(1) with the subsequent boundary conditions (appropriately corrected) \nwe find\n\nExample 3: Consider now the class of transmission lines which have \na, positive bounded and twice differentiable c(z) in the interval 0 S x S l. \nIt can be shown (see e.g., Ref. 2) that the ABCD parameters satisfy \nthe following asymptotic relations, for w large:*\n\nThese results follow from the classical theory of the asymptotic \nbehavior of the eigenfunctions of Sturm-Liouville problems.\u2019 The \nWKBJ method is a related subject. Schelkunoff has discussed these\n\n* The line is driven at the point z = |. The product of the inductance per unit \nlength and the capacitance per unit length is assumed to be unity.\n\nmatters in an elementary way in at least one of his textbooks (he does \nnot include the O(1/w) term).\n\nIf the line is terminated at \u00ab = O with a resistance Ry, we have for \nthe driving point impedance \nRoA(jw) + BUjw) | \nRoC (jw) nae o, (joa)\n\nSimilarly, fw) = Re exp (\u20142jlw)Z(jw) does not have a limit for \nw \u2014 +0, When R&,c(0) = 1, 1.e., when the line is \u201clocally matched\u201d\u2019 \nat \u00ab = 0, we have\n\nClearly, f(w) does not have the asymptotic behavior stipulated by \nZador; it does not even have a limit (because of the cos 2/w term). \nNote that the asymptotic formulas (10), (11), (12), and (13) are \nalso valid for a continuous positive c(x) which is piecewise twice dif- \nferentiable. This can be proven by partitioning the line at the dis- \ncontinuity points and finding the overall ABCD matrix by multiplying\n\nthe ABCD matrices of the sections of the line which now have a twice \ndifferentiable c(z).\n\nHence, property (277) of Zador\u2019s necessity statement could be replaced \nby the following: If (z) c(z) is a positive continuous and piecewise \ntwice differentiable function of the real variable x, (27) the line is term- \ninated in a unit resistance and c(0) = 1, then the following relation \nis valid for large w:\n\nvoltage reflection coefficient at x = / for the unit resistance terminated \nline, then\n\nwe can see, using Schelkunoff\u2019s results on wave propagation in stratified \nmedia,\u2019 that for w large\n\nTo generalize (following Schelkunoff\u2019) if c(0) = 1 and the first n \nderivatives of c(x) are continuous functions of z and vanish at the \nboundaries then for large w\n\nB.S.T.J. BRIEFS 1053 \nand therefore, \nais) = 4. + ota): (29) \nProperty (zz) in the necessity statement of Zador is also wrong.\n\nProof: The input impedance of the unit-resistance terminated line may \nbe written, in terms of the ABCD parameters, as follows:\n\nConsider a line with a twice differentiable c(z). In this case A(s), \nB(s), C(s), and D(s) are entire functions of order 1 and type I (see \nRef. 2), 1.e.,\n\nIn order to find Zador\u2019s representation with the N;, D; (\u00a2 = 1,2) \nfunctions we should be able to find an entire function \u00a2(s) # 0 such \nthat when we multiply both the numerator and denominator of Z(s) \nin (30) by this entire function, we get functions V;, D; (@ = 1,2) with \nthe properties stipulated by Zador.\n\nFrom (34), (85), (36), and (37) it follows that the functions (g(s) + \ny(\u2014s))/2 and (g(s) \u2014 e(\u2014s))/2 should be of type O in order that \nZador\u2019s N,; and D; be of type J. Consequently, the functions \u00a2(s) and \n\u00a2(\u2014s) themselves are of type 0. Therefore, it is impossible to find \nan \u00a2(s) such that o(s)v(\u2014s) = exp 2ls as Zador stipulates. So property \n(12) in Zador\u2019s necessity statement could be replaced by\n\nwhere k is a constant. Then NV, , D; (\u00a2 = 1,2) are proportional to the \nABCD parameters with proportionality factor k.\n\nFrom the above it follows that the sufficiency part as stated is in- \naccurate. It might be possible to alter the sufficiency conditions to \nmake them valid. In this case a proof must be given. The author has \ndone related work\u00ae on realizability conditions for nonuniform RC lines \nand is familiar with the difficulties involved in proving sufficiency \nconditions of this form.\n\nFinally, Zador\u2019s conjectures do not have an obvious physical in- \nterpretation and hence they should be justified.\n\n1. Zador, P. L., Realizability Conditions for the Impedance Function of the Lossless \nTapered Transmission Line, B.8S.T.J., 45, November, 1966, pp. 1667-1669.\n\n2. Protonotarios, E. N. and Wing, O., \u2018Analysis and Intrinsic Properties of the \nGeneral Nonuniform Transmission Line, To appear [HEE Trans. Microwave \nTheor. Tech. March, 1967.\n\n: mee E. L., Ordinary Differential Equations, Dover Publications, Inc., New York, \n1956.\n\n; Schelkunoff, S. A., Remarks Concerning Wave Propagation i in Stratified Media, \nCommun. Pure \u2018Appl. Math. June, 1951, pp. 117-128.\n\n. Protonotarios, E. N., On the analysis and Synthesis of the Nonuniform RC Dis- \ntributed Network, Doctoral Dissertation, Columbia University, May, 1966.", "title": "magazine :: Bell System Technical Journal :: BSTJ V46N05 196705", "trim_reasons": [], "year": 1967}
{"archive_ref": "bitsavers_BellSystemJV47N01196801_6097317", "canonical_url": "https://archive.org/details/bitsavers_BellSystemJV47N01196801_6097317", "char_count": 207905, "collection": "archive-org-bell-labs", "doc_id": 561, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc561", "record_count": 365, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bitsavers_BellSystemJV47N01196801_6097317", "split": "test", "text": "Unified Matrix Theory of Lumped and Distributed \nDirectional Couplers M.A. Murray-Lasso 39\n\nA Statistical Analysis of On-Off Patterns in Sixteen \nConversations P. T. Brady 73\n\nDielectric Loaded and Covered Rectangular Waveguide \nPhased Arrays V. Galindo and C. P. Wu 93\n\nSurface-Wave Effects on Dielectric Sheathed Phased \nArrays of Rectangular Waveguides C. P. Wu and V. Galindo 117\n\nADVISORY BOARD \nP, A. GORMAN, President, Western Electric Company \nJ.B. FISK, President, Bell Telephone Laboratories\n\nA. 8. ALSTON, Executive Vice President, \nAmerican Telephone and Telegraph Company\n\nThe usual design technique for waveguide band-rejection filters uses \nnarrow-band approximations and thus discrepancies generally exist be- \ntween the designed and measured response, particularly in the fairly wide \npassband. Nevertheless, this design technique has been used because of \nits sumplicity and because the filter configurations obtained are relatively \nsumple. Lately, a new design technique using transmission line synthesis \nbecame available which, theoretically, would yield the desired response. \nHowever, the physical realization results in a complicated configuration \nwhich leads to certain practical problems. This paper presents a modified \ntechnique which simplifies the structure without sacrificing performance. \nWith this modification the design procedure becomes very simple and many \nof the practical problems can be avoided. This paper gives precise design \ninformation and convenient design formulas. Furthermore, it shows that \nexcellent agreement between the designed and measured response can be \nachieved. |\n\nThe microwave waveguide band-rejection filter (BRF) now used in \nmany radio systems has many undesirable features. The designer finds \nthat the actual bandwidth is consistently narrower than the designed \nvalue, and that the filter has a unique passband VSWR behavior \nwhich becomes worse as the frequency goes farther away from the\n\nstopband. This becomes quite a severe problem in filters designed \nfor the high frequency or low frequency channels of the band. (For \nexample, in the 4 GHz band,* the return loss of a filter with the stop- \nband at 3710 MHz becomes progressively poorer as the frequency \napproaches 4190 MHz, and vice versa). In addition, the midband \nfrequency of the VSWR curve and of the corresponding delay curve \nis found to differ from the midband frequency of the insertion loss \ncurve at which the filter is tuned. The offset of a typical 3-cavity \nmaximally flat BREF at the 4 GHz band with 3 dB points at +17 MHz \ncan be as much as 2 MHz. Such an uncontrolled shift causes extreme \ndifficulties in the delay equalization of a radio system.\n\nThe present BRF design is based on the lumped-element low-pass \nprototype filter design technique: After making a proper frequency \ntransformation, the loaded Q of each cavity can be computed. The \ncavities are then separated by waveguide lengths which give an ef- \nfective spacing of an odd multiple of quarter wavelengths at the mid- \nband frequency, which is a standard technique to realize ladder filters \nin waveguide. It is desirable to keep the spacings as small as possible, \nbut, in order to avoid higher order mode coupling between cavities, it \nhas been determined that a three-quarter wavelength spacing is re- \nquired.\n\nIn a recent investigation, it was determined that neglecting the \nfrequency-dependent phase shift of the connecting lines results in both \nthe narrower bandwidth and the problem of high VSWR in the pass- \nband. A rigorous analysis was carried out for a three-cavity BRF. \nThe result shows that when the frequency dependence of the phase \nshift of the connecting lines is accounted for, an exact control of band- \nwidth is possible and the passband VSWR is improved; but the desired \ncharacteristic of the passband VSWR cannot be obtained as long as \nthe present form of construction is retained.\n\nVarious techniques for the exact synthesis of transmission line filters \nhave been available for many years.? However, they often are found \nto be not too practical for the design of waveguide filters because of \nthe large number of changes of characteristic impedances usually re- \nquired. There is already a sufficiently large number of discontinuities \nin a waveguide filter which are necessary to form the cavities. Addi- \ntional discontinuities with their associated fringing susceptances, such \nas would be needed to change the characteristic impedance, would \naggravate the practical realization difficulties. If these problems\n\ncould be overcome, the transmission line synthesis would offer the \nadvantage of being exact.\n\nThe possibility of constructing a waveguide BRF using transmis- \nsion line synthesis was mentioned by Schiffman and Matthaei.? Un- \nfortunately, the form they suggest involves three impedance changes \nin each connecting line and therefore is not practical. This paper pre- \nsents an improved form with only one impedance change in each \nconnecting line. The susceptance discontinuity present at each imped- \nance change is absorbed very naturally in the structure without caus- \ning any design difficulty. Design formulas for filters of up to five \ncavities are given in a very convenient form.\n\nThe problem of the midband frequency shift between the VSWR \nand the insertion loss was found to be the direct consequence of in- \naccurate correction for the connecting lines. The precise design proce- \ndure presented here guarantees the coincidence of the midband fre- \nquency of the VSWR and insertion loss characteristic.\n\nAn experimental model was built and tested, and the measured \nresult agrees very well with the theoretical prediction. With this de- \nsign technique, it is possible to achieve over 40 dB return loss across \nthe passband and to control the exact bandwidth and the midband \nfrequency. A pair of identical BRF were designed and in conjunction \nwith two hybrids, a constant resistance channel separation network \nwas built. Over 35 dB return loss across the band was observed; in \naddition, the delay distortion of the dropped channel was symmetrical \nwith respect to the midband frequency. Channel separation networks \nbuilt with BREs designed with the old techniques always had un- \nsymmetrical delay distortion for the dropped channel.\n\nII. BRF CONFIGURATIONS OBTAINED BY TRANSMISSION LINE FILTER \nSYNTHESIS\n\nThe easiest way to synthesize a transmission line filter is to use the \nknown low-pass prototype filters as done by Schiffman and Mat- \nthaei.? The reactive elements of the prototype circuit are replaced by \ntheir equivalent transmission line stubs (all of the same length J and \nof the same propagation constant @) using the frequency transforma- \ntion:\n\nwhere w is the normalized frequency of the prototype circuit and A is \na constant that determines the bandwidth of the filter. For a BRF, 1\n\ncan be chosen to be any odd multiple of a quarter wavelength of the \nmidband frequency. Schiffman and Matthaei chose 1 = dA,o/4. Then, \nthe shunt susceptances and series reactances of the prototype filter \ntransform as follows:\n\nIt is seen that the right side of (2) corresponds to the input suscep- \ntance of an open-circuited stub of characteristic admittance\n\nand that the right side of (3) corresponds to the input reactance of a \nshort-circuited stub of characteristic impedance\n\nAs shown in Fig. 1, a three-element low-pass prototype filter be- \ncomes a transmission line with three stubs using the transformation \n(1). These stubs are connected to the main transmission line at the \nsame point, which would be impossible in a physical realization. To \nsolve this problem, connecting lines between stubs have to be inserted \nby making use of Kuroda\u2019s identity. This identity allows the inter-\n\nFig. 1\u2014 The 3 element low pass prototype filter and its corresponding BRF \nwith transmission line stubs.\n\nchange of a stub and a connecting line (both of the same length /) as \nshown in Fig. 2. Adding a section of transmission line of appropriate \ncharacteristic impedance on both ends of the filter (Fig. 1) as shown \nin Fig. 3(a), the circuit retains its amplitude response. Applying \nKuroda\u2019s identity to the two shunt open-circuited stubs, one gets the \ndesired form in Fig. 3(b). With a proper replacement of the three \nseries stubs by rejection cavities (only an approximation), Fig. 3(b) \nforms three rejection cavities in cascade with quarter wavelength \nlines separation between the cavities.\n\nFor the most commonly used aperture-coupled rejection cavity, a \nquarter wavelength separation between the cavities would not be \nenough because of strong coupling resulting from higher order modes \nbetween the closely spaced discontinuities. It has been found, from \npast experience, that a three-quarter wavelength separation is neces- \nsary in most waveguide BRFs. For this reason, Schiffman and Mat- \nthaei treated the case of three-quarter wavelength separation between \ncavities. Following exactly the same technique as shown in Fig. 3, \none can add three pieces of quarter wavelength line on both sides of \nthe stubs, as seen in Fig. 4(a), without changing the amplitude re- \nsponse of the filter. Then, the final form of Fig. 4b is obtained by ap- \nplying Kuroda\u2019s identity three times on each side. The characteristic \nimpedances of the connecting line sections are tabulated for two and \nthree cavity BRF,\u00ae but no test result has been given in the same \nreference.\n\nTo realize the configuration of Fig. 4(b) in waveguide form, one \nwould, usually, change the height to control the characteristic imped- \nance such that the propagation constant @ remains unchanged. A \ntypical three cavity BRF would therefore assume the form shown in \nFig. 5. Theoretically, one may be able to compensate for the discon- \ntinuity susceptances at each step by proper adjustment of the line \nlength of each section. This, however, complicates the design pro- \ncedure unduly, and it is not sure how good the result may be. As \nmentioned before, the higher order mode interaction between closely \nspaced discontinuities may degrade the performance of the filter. \nSince a better configuration, in every respect, is being proposed, no \nattention was paid to the practical problems associated with this \nsuggested structure.\n\nThe only restriction which must be fulfilled in presently-available \ntransmission line synthesis is that all the line sections must have the \nsame length ! and the same propagation constant 8 so that the fre- \nquency transformation (1) can be applied. To construct a BRF, one\n\nFig. 4\u2014 Suggested 3 stub transmission line BRI to be realized in waveguide \nform.\n\nmay choose the length / to be any odd number of quarter wavelength \nof the midband frequency. For the waveguide BRF, it is convenient \nto choose 1 = 3A,g0/4. Then the shunt susceptance and series reactance \nof the filter become\n\nand the equivalence between prototype circuit and the transmission \nline stubs shown in Fig. 1 still holds except that the constant A is\n\nchanged due to the change of stub length J. The value of A may be \ncomputed from the given band edge attenuation specification; for ex- \nample, from the 3 dB point for maximally flat filters or from the \ncutoff frequency for equal ripple filters. Let w; be the edge frequency \nof the prototype filter and X,, be the corresponding waveguide wave- \nlength, then\n\nAfter obtaining the characteristic impedances of all the stubs, one \ncan apply Kuroda\u2019s identity to get the same form as Fig. 3(b), of \ncourse with 1 = 3A,9/4. This is almost the desired form for waveguide \nBRF except that the series stubs have to be replaced by cavities with \nthe proper Q values. There are many ways to find the equivalence be- \ntween stubs and cavities, all of which are approximations. Among \nthem, one method is found to be particularly convenient. The input \nimpedance of a short-circuited stub of 3A,0/4 length with characteris- \ntic impedance Zp 1s\n\nBy the well known partial fraction expansion of the tangent function, \none gets:\n\n7 4(8d0\\ < 1 \nZin = 940 \u2014 (2) pS SS ge (8) \nT g k=1 2 Ago \n(2k \u2014 1)\u00b0 \u2014 (8h) \nNg \nIn the vicinity of the resonance frequency (A,0/A, ~ 1), it is clear \nthat the k = 2 term dominates (8). Then:\n\nTaking this approximation, it is clear that the equivalent cavity must \npossess a loaded Q of\n\nThis approximation is found to be sufficiently accurate for frequencies \nclose to the resonance frequency as is the case for most waveguide \nfilters.\n\nMost of the advantages of this new configuration are obvious. As \nseen in Fig. 6, a three-to-one reduction in the number of steps is\n\nobtained, which makes the design procedure much simpler. In addi- \ntion, the spacing between discontinuities has become large, hence, one \nwould expect less practical problems in realization (because of higher \norder mode interaction). One rather subtle practical advantage of \nthis structure is that all the steps are located right under the center \nof the coupling apertures. For all inductive coupling apertures (the \ncircular hole is one of them), a capacitive compensating stud is \nalways required at the location of the hole, which in this structure \nhappens to coincide with the location of the steps. Because of the \ncapacitive step susceptance, the size of the compensating studs just \nhas to be made slightly smaller. No additional compensation is \nneeded for all the step susceptances. |\n\nA convenient table for filters with two to five cavities is presented \nin this section which gives the @ values for each cavity and the char- \nacteristic Impedance for each connecting line. The following symbols \nare used:\n\nn = number of cavities \nR, = normalized generator impedance \nFR, = normalized load impedance \nQ; = loaded Q value of zth cavity \ng; = normalized values of the low pass prototype filter \nZ;.:., = normalized characteristic impedance of the connecting line \nbetween the 7th and the (2 + 1)th cavity \nA = bandwidth constant defined in (6) \nn= 2\n\nTo construct connecting lines for a given characteristic impedance, \none may modify the heights 6; of waveguides according to\n\nInformation on cavity design has been available for a long time. \nA set of curves was plotted of loaded Q vs aperture size for different\n\nfrequencies, of cavity length vs aperture size for different frequencies, \nand of stud length vs aperture size for different frequencies. Such \nfrequencies were chosen that both the 4 GHz (8.7-4.2) and the 6 GHz \n(5.925-6.425) bands are covered.\n\nThe equivalent circuit of a properly compensated cavity of Fig. 7(a) \nis known (to the extent of the approximations used here) to be of \nthe form shown in Fig. 7(b) referred to the reference plane 7\u2019. Aside\n\nfrom the resonance circuit, an additional piece of transmission line \nof length / must be added at each side, in order to make the connecting \nline sections exactly 3\\,,/4 and this line element must be taken into \naccount accurately. It was found that the unpredictable midband \nfrequency shift between VSWR and insertion loss response, which is \none of the major problems of the previously designed BRF, is the \ndirect consequence of an inaccurate correction for this line element. \nA theoretical study showed that any inaccuracy in the length of the \nconnecting line may result in a shift in midband frequency of the \nVSWR response with respect to the insertion loss response.* One \ntypical 3-cavity 4 GHz BRF with Afzsg,3 = +17 MHz was found to \nhave a 2 MHz shift with a 35 mil error in each connecting line. \nBecause of its importance, an attempt was made to measure ac- \ncurately the values of J for various sizes of coupling aperture. The first \nset of data based on the symmetrical equivalent circuit (Fig. 7b) was \nvery disappointing since, for a simple measurement like this, the\n\n* Dissipation must be taken into account since this phenomenon does not exist \nin the lossless case.\n\nwidely spread measurement data was unexpected. It was found later \nthat the symmetry of the structure is destroyed by the tuning screw \non one side of the cavity (Fig. 7a); consequently, the symmetrical \nequivalent circuit 7(b) is no longer valid.\n\nTo make the equivalent circuit unsymmetrical, the circuit in Fig. 7(c), \nwhere the transmission lines on each side have different length, is \nintroduced. This does not solve the problem, however, because both \nthe length /, and 7, depend strongly on the amount of tuning added \nto the cavity. A simple plot of J, and 7, does not make sense unless \none could specify a fixed amount of tuning. Fortunately, there is a \nquantity, mainly 1, + 1,, which is very insensitive to the amount \nof tuning. As measured in the 4 GHz band, this quantity is accurate \nwithin a variation of 2 mils for a quite large tuning range. It is believed \nthat this would be true for the other frequency bands, too. Hence, \nplots of (J, + 1.)/2 vs hole diameter are presented in Tigs. 8 and 9 \nfor 4 and 6 GHz bands, respectively. For small tuning, it is true that \nlL wel, & (1, + 1,)/2. Furthermore, it was found that 1, , the length \non the tuning screw side, is always larger than /, , and that the dif- \nference, 1; \u2014 1,, may vary from zero to 20 mils depending on the \ntuning range for the sizes of apertures used in this study.\n\nTo verify the theoretical work discussed above and the available \ndesign data, a 3-cavity BRI was designed. Because of the approxima-\n\n4 Fig. 8 \u2014 Measured information on waveguide rejection cavities in the 4 GHz \nand,\n\nFig. 9\u2014 Measured information on waveguide rejection cavities in the 6 GHz \nband.\n\ntions involved in the equivalence between cavity and stub line, there \nmight still be deviations from the expected performance at frequencies \nwell removed from the center frequency. Therefore, the filter was \npurposely designed to operate in the highest frequency channel (4190 \nMHz) of the 4 GHz band. In this case, the passband extends almost \n500 MHz below the midband frequency of the filter. Any severe dis- \ncrepancy caused by the approximation should show up in this extreme \ncase. The filter was designed to have a maximally flat response with \nthe 3 dB points at +17 MHz, neglecting intrinsic losses. In order to \nconvey some idea of the \u2018\u2018exactness\u201d of the filter performance, the \nmeasured results are compared with the computed theoretical response. \nThis computed curve is the theoretical response of a transmission \nline filter of Fig. 3(b) (1 = 3),./4) with all line sections having an \nattenuation constant:\n\nAs seen in Figs. 10 and 11, the filter shows lower VSWR value in \nthe vicinity of resonant frequency which is attributed to slightly \nhigher intrinsic loss than the theoretical estimated value. However, \nover 40 dB return loss was obtained across the entire passband which \nis in very good agreement with the computed results.\n\nMeasured data were also taken from a channel-separating net- \nwork constructed with two identical BRFs and two hybrid junctions. \nThe channel separating network has over 35 dB return loss across the \nband. The delay distortion of the dropped channel was also measured, \nand the result shows that the delay distortion vs Af curve is symmet-\n\nFig. 10\u2014Computed and measured stop band performance of a typical 3 \ncavity maximally flat BRF.\n\nFig. 11\u2014 Computed and measured passband performance of a typical 3 cavity \nmaximally flat BRF\n\nFig. 12\u2014 Measured delay distortion of a channel separation network using \npresent design techniques as compared with that using previous techniques.\n\nrical with respect to the midband frequency. The measured data is \nshown in Fig. 12 where the delay distortion curve of the present exist- \ning channel separating network is plotted for comparison.\n\nThe author wishes to thank F. G. Joyal for carrying out all of the \nmeasurements on the correction line lengths and on the filter per- \nformance.\n\n1, Young, L., Matthaei, G. L., and Jones, E. M. T., Microwave Bandstop Filters \nwith Narrow Stop Band, IRE Trans. on Microwave Theory and Tech- \nniques, 10, November 1962, pp. 416-427.\n\n2. For example, see A. I. Grayzel, A Synthesis Procedure for Transmission Line \nee IRE Trans. on Circuit Theory, CT-5, September 1958, pp. \n172-181.\n\n3. Schiffman, B. M. and Matthaei, G. L., Exact Design of Bandstop Microwave \nFilters, IEEE Trans. on Microwave Theory and Techniques, 12, January \n1964, pp. 6-15.\n\nA new approach to speech synthesis by rule has been formulated and \nevaluated. A discrete set of symbols (phonemes and stress marks) 1s con- \nverted to a continuous acoustic waveform by a two-step transformation. \nThe first step involves conversion from phonemes to control signals capable \nof driving a terminal analog speech synthesizer. The second step 1s con- \nversion from control signals to the acoustic waveform.\n\nThis paper presents a design for the terminal analog synthesizer and \ndiscusses the new features of this device. It discusses in detail the method \nof converting from phonemes to control signals. It places primary emphasis \non determining the formant frequency control signals and the fundamental \nfrequency contour, and presents models for determining these contours \nfrom the input data. The paper includes an experimental evaluation of \nthe entire technique in terms of word intelligibility scores and consonant \nconfusion matrices.\n\nSpeech synthesis by rule is the method of converting from a discrete \nrepresentation of speech in linguistic units, that is, phonemes and \nstress marks, to a continuous acoustic waveform. Fig. 1 shows the \ntechnique for carrying out this transformation. The figure shows that \nthe discrete input is converted to continuous control signals by the \nsynthesis strategy. The synthesis strategy contains stored informa- \ntion about the phonemes and stored rules about the mutual effects of \nadjacent phonemes. The stored rules operate on the input sequence \nto produce the control signals for the synthesizer. The speech synthe- \nsizer converts the control signals to continuous speech. The synthesizer \nmay be a terminal analog, a dynamic analog of the vocal tract, or a \ncombination.\n\nDuring the past ten years there have been many attempts at syn- \nthesis by rule. The primary goal of these attempts has been to produce \nnatural sounding, intelligible speech. Of secondary importance has \nbeen the preservation, in some natural way, of the dynamics of speech \nproduction by embodying in the scheme the constraints imposed by \nthe human vocal tract.\n\nPrevious methods of synthesis by rule are generally classified as \neither articulatory or acoustic domain approaches. An articulatory \napproach uses physiological parameters such as tongue-tip position, \nand lip opening as the control signals for the synthesizer. The stored \ndata of the synthesis strategy are of the form of vocal tract configura- \ntions. An acoustic domain approach uses parameters such as formant \nvalues and fundamental frequency as control signals. The stored data \ninclude such information as target positions of formants and relative \namplitudes of phonemes.\n\nArticulatory domain approaches to synthesis by rule\u2019? have been \nmost successful in modelling the dynamics of the speech producing \nmechanism. Acoustic domain methods, such as the one presented here, \ncan impose the natural constraints of the vocal tract only indirectly, \nthat is, by rules which often lack a firm physiological basis. How- \never, acoustic domain approaches have enjoyed the most success in \nproducing intelligible, high quality speech,? * ** thus justifying and \nmotivating efforts along these lines. The technique for synthesis by \nrule, described in this paper, is an acoustic domain approach.\n\nThe next section gives a general description of the synthesizer. \nTerminal analog synthesizers of this type are common\u201d *\u00b0\u00ae and we \ndiscuss only the new features at any length.\n\nA terminal analog synthesizer models the speech-producing mecha- \nnism, which includes the vocal tract, excitation sources, and radia- \ntion impedance. The transfer function of the vocal tract can be re- \nduced to either a cascade of complex conjugate pole and zero pair \nnetworks, or a parallel addition of complex pole pair networks. The \ncascade representation was used because it reduced the complexity \nof the synthesis strategy by reducing the number of synthesizer con- \ntrol parameters.\n\nFig. 2 is a block diagram of the synthesizer used in this work. The \nsynthesizer was simulated on a computer at 20 kHz sampling fre- \nquency. There are two sources of excitation, a pitch impulse genera- \ntor, and a frication (noise) generator. To produce voiced speech \n(vowels, nasals, voiced stops, and voiced fricatives) the pitch im- \npulse generator output is gated by the switch to the upper arm of the \nsynthesizer. The nasal network is included in the upper arm only for \nnasal consonants. To produce whispered or aspirated speech, the \nfrication generator is gated by the switch to the upper arm of the \nsynthesizer.\n\nFo | VA F, F Fs \n(1-ANASAL) \nPITCH HIGHER \nIMPULSE SWITCH SHAPING FORMANT POLE ge & RADIATION \nNETWORK NETWORK CORRECTION NETWORK \nGENERATOR eislsate\n\nFRICATIVE \nFRICATION & oo) POLE AND SHAPING \nGENERATOR ZERO NETWORK \nNETWORK\n\nTo produce a voiceless fricative or the unvoiced component of \nvoiced fricatives, the frication generator excites the lower arm of the \nsynthesizer. For a voiceless fricative, the output of the voiced fricative \nexcitation network is constant. For a voiced fricative the output of \nthe voiced fricative excitation network modulates the frication gen- \nerator output. The details of this network are explained in Section 2.1 \nbecause this is an original design.\n\nThe higher pole correction network in the upper arm of the synthe- \nsizer compensates for the missing higher order poles.\u00ae *\u00b0 A new design \nfor this network, based on the properties of sampled data systems, has \nbeen formulated and is discussed in Gold and Rabiner\u2019s paper.**\n\nOne last feature of the synthesizer is provision for generating a \nvoice bar. (A voice bar is the quasi-periodic low frequency energy \nradiated from the region of the vocal cords during the closure inter- \nval of voiced stop consonants. During this interval the vocal cords \nare vibrating thus acting as the source of energy for the voice bar.) \nTo produce a voice bar, the middle arm of the synthesizer is used with \nthe switch gating the pitch impulse generator output to the shaping \nnetwork. The voice bar has energy only at low frequency similar to \nvoice bars of natural speech.\n\nThe outputs of the three arms of the synthesizer are added to \nproduce the speech. The synthesizer control signals are indicated in \nFig. 2 by arrows. These include four amplitude controls (avoicing, \nanasal, avb, afric); a derived amplitude control (1l-anasal); 14 pole- \nzero controls (both center frequency and bandwidth of F,, Pe, Fs, \nnpol, nzer, fpol, fzer); a switch control (va); and a fundamental \nfrequency control (Fo).\n\nThe network connecting the pitch impulse generator to the lower \narm of the synthesizer is used to provide the excitation for the un- \nvoiced component of voiced fricatives. Fig. 8 shows the relevant details \nof this network. (For clarity, certain components of the synthesizer \nhave been omitted from Fig. 2.) |\n\nThe ouput of the pulse generator is shaped to produce a suitable \npitch pulse. A complex conjugate pole pair resonator was used, but \nany suitably chosen network could have been used. The pitch pulses \nexcite a resonator tuned to the first formant of the fricative sound. \nA single resonance is the first order approximation to the transfer \nfunction of volume velocity (the signal of interest in Fig. 3) from the\n\nNOISE FRICATIVE \nGENERATOR NETWORK \nFig. 3 \u2014 Excitation network for voiced fricatives.\n\nglottis through the point of constriction of the vocal tract. A threshold \nlevel is subtracted from the output of the resonator and the result is \nhalf-wave rectified. These operations model the physical situation \nwhere turbulence is not produced until the volume velocity of the air- \nflow exceeds a threshold value. The output of the half-wave rectifier \nmodulates the output of a noise generator, preducing a pitch syn- \nchronous excitation for the unvoiced component of the fricative. The \nfinal unvoiced component is produced by exciting the fricative net- \nwork by this excitation. The voiced component is produced in the \nstandard manner, that is, by exciting the formant network by the \npitch pulses.\n\nSpectrograms of voiced fricatives produced by the above technique \nare quite similar to spectrograms from natural speech. Experimental \nevidence presented later shows that the synthetic fricatives are highly \nintelligible.\n\nSince formant contours are crucial to speech intelligibility (see \npages 220-234 of Ref. 9), the first step in transforming a discrete set \nof input symbols to the synthesizer control signals is to generate the \nformant contours. The method is explained in Sections 3.1, 3.2, and 3.3.\n\nOnce the formant contours are specified, the remaining control \nparameter contours are determined by delimiting certain characteristic \ntimes along the formant contours. At these times, motion of the other \nparameters is initiated or terminated. The techniques for generating \nthese contours are discussed in Section 3.4.\n\nPart IV treats the generation of a fundamental frequency con- \ntour. Table I shows international phonetic alphabet symbols and the \nletter equivalents used in the following sections.\n\nIn order to generate the formant contours from the phoneme input \nsequence, certain information must be supplied. Corresponding to\n\neach possible phoneme there must be data on formant target positions. \nThese data, along with other data, are included in a phoneme char- \nacterization table. Certain durational data are necessary, such as for \nstressed vowels. Finally a technique for generating formant transi- \ntions must be supplied.\n\nEach phoneme has a characterization independent of adjacent \nphonemes. The characterization includes formant information, source \ncharacteristics in the production of the phoneme, a description of \nwhether it is nasal or fricative, and a set of frequency regions sur- \nrounding the formant positions.\n\nThe formant information is a set of target positions for both center \nfrequency and bandwidth of formants one, two and three. The source \ncharacteristics describe the condition of the vocal cords during the \nproduction of the phoneme. If the vocal cords are vibrating the sound \nis voiced. The frequency regions of a phoneme represent the degree \nto which certain acoustic parameters must approximate the target \nvalues of these parameters in the context of connected speech. In an \narticulatory analog, the corresponding concept would be the extent \nto which a given vocal tract configuration must approximate the \ntarget configuration for the phoneme.\n\nThe frequency regions represent a compromise between choosing \na single characterization for a phoneme and considering it inviolate, \nand the realization that there are many acceptable characterizations \nfor a phoneme\u2014especially in the context of connected speech.\n\nTable II shows the phoneme characterizations we have used. The \nfirst three columns list the formant target positions of the phonemes. \nThe second three columns show the frequency regions of the phonemes. \n(The figures represent both + values.) The final three columns de- \nscribe nasality, fricative, and voicing characteristics of the phonemes. \nA + in any column indicates the presence of the feature and a \u2014 \nindicates its absence. The voicing condition of the voiced fricatives z \nand ZH is + indicating the two sources used to produce these sounds \non the synthesizer. The bandwidths of fF, and F; are held fixed at \n100 Hz and 120 Hz for all phonemes. The bandwidth of F, is 60 Hz \nexcept for nasals where it 1s 150 Hz. When a -+ appears in the nasality \nor fricative columns, a table look-up procedure is used to specify \npole-zero locations.*\n\n* All data referred to but not included in this paper are available in the au- \nthor\u2019s Ph.D. thesis which is available from the MIT library. (See Ref. 12.)\n\nVowel duration is specified only for stressed vowels. The durations \nof unstressed vowels are determined by the methods illustrated in Sec- \ntion 3.8. The duration of a stressed vowel is modified by its following \nphoneme. The longest vowels are those followed by voiced fricative \nconsonants; the shortest are followed by voiceless stop consonants.\u201d\n\nFor certain consonants maximum durations are specified. Conso- \nnant duration (as measured from human speech) is not a fixed quan- \ntity but is very dependent on context. (Four example, initial conso- \nnants are much longer than medial consonants.) The synthesis \nstrategy generates consonants whose duration is variable within cer- \ntain limits. Maximum durations are specified to prevent the consonant \nfrom being unnaturally long, hence objectionable.\n\nMaximum stop gap durations are specified for stops; aspiration \nduration, as a function of the succeeding phoneme, is specified for \nvoiceless stop consonants. Values of amplitude control signals are \nspecified for all phonemes. Rates of change of control signals are \nspecified for various phoneme classes (that is, vowels, nasals, frica- \ntives, and stops).\n\nThe technique for generating formant transitions (and hence for- \nmant contours) is a new one. We present it in detail because the entire \nsynthesis strategy is built around it.\n\nAs we stated, the motion of formants is one of the most significant \nfactors contributing to the intelligibility of speech. Smooth, continuous \nformant transitions are generally observed on spectrograms of real \nspeech. To match these characteristics, we used the solution to a \ncritically-damped second degree differential equation to describe the \ntransitions of formants. We chose a second degree equation because it \nprovided a good fit to data on formant transitions. We used a critically- \ndamped solution because it was completely specified from a single \ntime constant. Values of time constants were determined from examin- \ning formant transitions for real speech on spectrograms.\n\nThe input to the differential equation represents the formant target \nposition appropriate for the current phoneme. Since the current \nphoneme changes its value discretely, the input to the differential \nequation changes in a steplike manner. The formant motion, in re- \nsponse to this step input, is smooth and continuous. Thus motion \nfrom steady state value A1z to target value Af, beginning at time f = \n0 is of the form:\n\nIn general, motion between target positions does not proceed from a \nsteady state condition; that is, there are initial conditions. Motion to \na target whose formant value is Af from an initial formant position \nAv with an initial formant velocity Vi = dx/dt\\o_ is of the form:\n\nAt times when the input to the differential equation is changed dis- \ncretely, both the output value and slope are continuous. Thus the \nconcept of smooth, continuous formant transitions is realized in all \ncases.\n\nThe time constants of the differential equation are functions of the \nindividual formant and the pair of phonemes between which the \ntransition is being made. Hence, for each possible pair of phonemes, \nand each formant, a time constant is specified. Certain simplifying \napproximations reduce (by an order of magnitude) the number of \ntime constants that have to be specified.\n\nThe inputs to the differential equations change discretely in time in \na steplike manner. However, provision is made for delaying, to any \nformant, the steplike change in formant target position. Thus, in the \nmost general case, formants move independently of each other with \nunequal time constants of motion. This delay feature was found \nnecessary for only a few cases.\n\nThe phonemic goals change discretely in time. The decision of when \nto make the discrete changes, that is, when to initiate motion to new \nsets of formant targets, is based on the criterion that the formants \nmust first be within the phoneme frequency regions of the targets, and \nthen satisfy durational requirements of the phoneme, if there are any.\n\nFormants, in general, are in motion towards target values ap- \npropriate for the phonemes to be generated. Their motion is charac- \nterized by the solution to a differential equation. The time constant of \nmotion is a function of the phoneme from which motion began and \nthe phoneme which is being generated. Each formant moves with its \nown time constant and there is provision for delay in time of initiation \nof the motion of formants. When all formants are within the fre- \nquency regions of the target, a decision is made. If a stressed vowel \nis being generated, then a table look-up procedure determines the \ncorrect vowel duration and motion continues for the specified time. \nOnce a vowel of proper duration has been generated, motion towards \ntarget positions characteristic of the next phoneme is begun. If the \ncurrent phoneme is not a stressed vowel, motion towards the new \nphoneme targets is initiated as soon as all the formants are within \ntheir specified frequency regions.\n\nThe decision to start motion to new target values results in three \nseparate operations. First, new time constants for each formant are \ninserted into the respective difference equations. Second, the forcing \nfunctions (input) to the difference equations are changed in a step-\n\nlike manner indicating the changes in target positions. Finally, the \ninitial conditions of the difference equation are set to preserve con- \ntinuity of formant values and formant velocities. If the motion of \nany formant is to be delayed, the changes in the difference equation \nfor that formant are delayed appropriately.\n\nFig. 4 shows a typical cycle of events. Initially formants one and \ntwo (we shall neglect formant three in this example) are at target \npositions appropriate for phoneme 1. At time \u00a2, motion is initiated \nto phoneme 2. Formants one and two begin motion simultaneously \n(no delay is used here) with time constants 7}, and 77, respectively. \nTi. 18 much smaller than 77, so formant one moves more rapidly to \nits target value than formant two. Periodically the formant values \nare tested to see whether they are within the specified frequency regions \nof the targets. (The frequency regions are indicated by Al, A2 in \nFig. 4.) If they are not, the formants continue their motion, thus \nmoving closer to target. For the example in Fig. 4, formant one enters \nits frequency region prior to formant two. Until formant two enters \nits frequency region at time \u00a2, , formant one moves closer to its target \nposition. At time \u00a2, both formants are within the specified frequency \nregions and so a check is made on whether phoneme 2 is a stressed \nvowel or not. In this example phoneme 2 is not a stressed vowel, so \nmotion to phoneme 3 is initiated at t,. However, we now have the\n\ncase when the time of initiation of motion of formant one is delayed. \nHence at \u00a2. the target value and time constant for formant two is \nchanged, but formant one\u2019s target is unchanged. At time \u00a2{ the delay \nis terminated and formant one begins its motion.\n\nPhoneme 3 of Fig. 4 is a stressed vowel. So at time \u00a2;, when both \nformants are within the frequency regions for phoneme 3, motion to \ntargets for phoneme 3 continues for the specified vowel duration. \nAt time \u00a2 , following the vowel duration, motion begins toward targets \nfor phoneme 4. New time constants and targets are again inserted \nin the equations of motion. The process continues in this manner \nuntil all the input phonemes have been generated.\n\nThe motion of the remaining synthesizer control parameters (that \nis, nasal and fricative poles, and zeros and source amplitudes) is time- \nlocked to the formant motion. The source amplitudes (avoicing, \nanasal, afric, avb) begin to switch approximately one time constant \nafter the discrete phoneme goal is changed. The amplitudes change \nlinearly at predetermined rates. The nasal and fricative poles and \nzeros initiate motion at the time the phoneme goal is changed. The \nmotion is linear and the slopes are arranged so that the poles and \nzeros just reach their targets at the time the source amplitudes are \nswitched. The target positions are specified in a table.\n\nFor nonnasal sounds, the target positions of the nasal zero and pole \nare set to 1400 Hz. Thus the pole and zero will cancel each other in \nthese cases. Furthermore, for nasal sounds, the bandwidth of formant \none (nominally 60 Hz) is changed linearly to 150 Hz for the duration \nof the nasal. The bandwidth begins to change 50 msec before the \namplitudes switch and is linearly changed back to its nominal value \nin 50 msec after the nasal. For nonfricative sounds the fricative pole \nand zero target positions are set to 1500 Hz.\n\nOur model for generating fundamental frequency data is based on \nthe assumption that these data can be derived from data on laryngeal \ntension (LT) and subglottal pressure (Ps). A description of an ut- \nterance in terms of these variables is then used to produce the desired \nfundamental frequency data.\n\nThe model is based on that of P. Lieberman, in which the breath- \ngroup is defined as an underlying phonetic feature of American Eng-\n\nlish.1# The unmarked breath-group is characteristic of a simple, de- \n- Clarative sentence, whereas the marked breath-group characterizes a \nsimple interrogative sentence.\n\nThe feature breath-group is converted to a global description of an \nutterance in terms of Ps and LT. Fig. 5 shows the archetypal Ps con- \ntour, as suggested by Lieberman\u2019s data. Ps increases over the first \n300 msec of the utterance, and then remains constant until the last \n300 msec of the breath-group, at which point it decreases rapidly to \nzero. The LT contour for an unmarked breath-group is constant, \nwhereas for a marked breath-group it is characterized by a steady \nincrease over the last 175 msec of phonation. Fundamental frequency \nis linearly proportional to both Ps and LT. Since the archetypal Ps \ncontour falls at the end of a marked breath-group, the increase in LT \nmust compensate for the decrease in Ps to give fundamental frequency \na rising terminal contour. A slope of 0.6 Hz/msec, for the last 175 \nmsec of phonation, was assigned to the LT contour. This resulted in \na terminal rise of 60 Hz in fundamental frequency for a question.\n\nThe subglottal pressure contour is modified by both consonants \nand vowels. Two levels of stressed vowels are adopted. One level is \nreferred to as emphasis and only one vowel in a breath-group is em- \nphasized. The emphasized vowel provides the highest peak in the Ps \ncontour. All other stressed vowels are treated similarly. When a \nvowel is stressed, there is an increase in subglottal pressure for a pe-\n\nFig. 5\u2014 Archetypal subglottal pressure contour showing effects of vowel stress.\n\nriod of 500 msec, centered on the vowel. Fig. 5 shows an example of \nthis effect. The dashed curve shows the effects of placing stress on a \nvowel at the beginning of the breath-group. There is a rise in Ps \nearly in the breath-group and the increase is centered at ts, the mid- \npoint of the vowel steady state. If the stressed vowel had been the \nemphasized one, the only difference in Fig. 5 would be the amplitude \nof the increased Ps. It would have been 2.5 em H.O as compared to \n1.0 cm HO.\n\nThe effects of consonants on subglottal pressure have also been \nincluded in our scheme. Tor a voiceless consonant, subglottal pressure \nautomatically increases whereas subglottal pressure automatically de- \ncreased for voiced consonants. Thus the consonants introduce local \nperturbations to the Ps contour. The change in Ps for consonants \nis 1.0 cm H,0, and this change occurs over a period of 150 msec \ncentered on the consonant.\n\nAll vowels are unstressed except those followed by the symbol \nstrss. The symbol strss1 signifies the emphasized vowel. Word boun- \ndaries are signified by the symbol space and pauses by the symbol \npause. A question is signified by the symbol ques. The sentence boun- \ndary is indicated by the symbol end. The examples are:\n\n(2) This is an olive. \nTHEI strss S$ space 1 Z space AEN space A strss1 L1v end. \n(11) Why are you sad? \nw A Strss1 IY space AR space \u00a5 00 space \u00a7 AB strss D end. \n(272) We sang all day. \nW IY space Ss AE Strss1 NG space OW L space DETY Strss end.\n\nWhenever a word boundary (space in our code) occurs, consonants \non either side of the word boundary are affected. In this strategy an \ninitial consonant is lengthened by about 20 percent, whereas a final \nconsonant is shortened by a similar amount. A word boundary has no \neffect on phonemes which do not lie on either side of the word bound- \nary.\n\nIntelligibility tests were conducted to evaluate the scheme. To test \nthe rules in a limited environment, consonant intelligibility tests were \nrun. One test was intended to test perception of consonants in pre-\n\nstressed position. The schwa vowel uu always preceded the consonant \nand was used as a perceptual cue for stop consonants because it pro- \nvided a basis for perceiving the stop gap. The second test was intended \nto test perception of consonants in post-stressed position. The schwa \nvowel uH always followed the final consonant\u2014again providing a \nbasis for perceiving stop gap duration, bursts, and aspiration for \nstops.\n\nSixteen consonants were used: B, D, G, P, T, K, M, N, F, TH, S, SH, V, \nTHE, Z, and ZH. Five vowels (besides schwa) were used: Iy, AE, A, OW, \nand oo. For each test there were 80 possible stimuli. Twenty additional \nstimuli were used, ten initiating the test and ten concluding it, giving \na total of 100 stimuli per test. Only the middle 80 were used for eval- \nuation, and these were presented in random order.\n\nThree subjects were tested. Their results are summarized in the \ntwo confusion matrices shown in Table III. In prestressed position \n(uH-c-v), 73 percent were correct; in post-stressed position (v-c-UH), \n77 percent were correct. If r, rH and v, THE responses are pooled, as 1s \noften done, then the correct percentages increase to 79 in prestressed \nposition and 81 in post-stressed position. Ten prestressed consonants \nwere identified correctly more than 75 percent of the time: B, D, P, T, \nN, TH, 8, SH, Z, and zH. The post-stressed consonants identified cor- \nrectly more than 75 percent of the time were B, P, T, K, TH, S, SH, 2, \nand sH. The consonants which were identified incorrectly most often \nwere G, M, D, and kK.\n\nAn examination of the errors in the confusion matrices of Table III \nshows:\n\n(1) The voiced stop c was often confused with T and x; and D in \npost-stressed position was often confused with \u00aba. \n(72) The unvoiced stop K was often confused with T in prestressed \nposition. \n(22) The nasal Mm was often confused with B and v. \n(1v) The fricative pairs v, THE and F, TH were often confused.\n\nThese errors were the major confusions in the tests. The stop con- \nfusions primarily were caused by errors in frication burst positions. \nThe fricative pair errors were anticipated because of the acute acoustic \nsimilarities between these particular fricatives. The cause for the con- \nfusions between m and other phonemes is unknown. Further work re- \nmains to be done in this area.\n\nOne test contained simple declarative, interrogative and imperative \nsentences. A second test contained sentences chosen from a list of sen- \ntences used often in intelligibility tests.15\n\nThe sentences were presented to listeners who wrote down what \nthey heard. They were told to guess whenever in doubt. The sentences \nwere played a second time and the listeners were allowed to make \nchanges. The tests were scored on the number of words which were \ncorrectly identified (excluding only the and a as words). The results \nof the tests are as follows. For the test using simple sentences, eight \nlisteners had an average of 92 percent of the words correct after one \ntry, and 95 percent after the second try. For the test using the longer\n\nstandard sentences four listeners had an average of 83 percent of the \nwords correct after one try; and 86 percent after the second try.\n\nAs shown above, the percent intelligibility scores for sentences were \nsignificantly higher than for isolated syllables, primarily because of \nthe context of speech in a meaningful utterance. However, the longer \nthe utterance, the less intelligible 1t became. This is because rhythm \nand timing are much more important for a long sentence than for a \nshort, simple one.\n\nFig. 6a shows wideband spectrograms of the utterance \u201cLarry and \nBob are here.\u201d The spectrogram in the upper half of the figure is the\n\nFig. 6a\u2014 Wideband spectrograms of synthetic (top) and natural versions of \n\u201cLarry and Bob are here.\u201d\n\nsynthetic version. The lower spectrogram was made from the author\u2019s \nspeech. (The synthetic utterance was in no way modelled after or \nmodified by the natural utterance.) Fig. 6b shows narrowband spec- \ntrograms of both the synthetic and natural versions of this utterance.\n\nThis is a high degree of similarity between the spectrograms of the \nreal and synthetic speech. The durations of both the synthetic and \nnatural utterances are comparable. Fig. 6a shows that the variation \nof the formants for both versions is quite similar. Even the funda- \nmental frequency contours for these utterances are quite similar. As \nFig. 6b shows, both contours are peaked during the stressed vowels \nA in BoB and I in HERE. A careful examination of the narrowband \nspectrograms shows the decrease of fundamental frequency, for both \nutterances, during the initial and final B of Bos.\n\nThe stressed vowels of this utterance can easily be identified from \neither the long steady state duration of Fig. 6a or the peak in the \nfundamental frequency contour of Fig. 6b.\n\nThe results of the consonant intelligibility tests showed that most \nconsonants were reproduced accurately. The sentence intelligibility \ntests also produced good intelligibility scores, indicating a high degree \nof success for the major goal of this project.\n\nMany of the listeners made informal comments concerning the ma- \nchine-like quality of the speech, but no formal tests were run to meas- \nure the naturalness or quality of the synthetic speech. Current studies\n\nFig. 6b \u2014 Narrowband spectograms of synthetic (top) and natural versions of \n\u201cLarry and Bob are here.\u201d\n\nabout the characteristics of the source of voiced speech are expected \nto produce valuable information about the determinants of synthetic \nspeech quality.\n\nAmong the topics that will be considered for future work are the \neffects of stress and rhythm on timing of an utterance, the inclusion \nof more than one breath-group in an utterance, and studies of further \ncorrelates of word boundaries.\n\nAn acoustic domain approach to speech synthesis by rule has been \nformulated and programmed on a digital computer. Samples of speech \nhave been generated using our scheme and their intelligibility has \nbeen measured. The problem of automatically generating a funda- \nmental frequency contour by rule has also been investigated and one \npossible solution has been found.\n\n(iv) Natural fundamental frequency contours for both interroga- \ntive and declarative sentences.\n\n(v) A new synthesizer design having potential for high quality \nvoiced fricatives and provision for inclusion of a voice bar \nfor voiced stops.\n\n(vi) A new method of handling formant transitions (a differential \nequation approach) and transition durations (the matrices \nof time constants).\n\n(vii) Internal control of timing using system parameters with few \nexternal constraints.\n\n(viii) A new method of generating fundamental frequency data \nbased on a physiological theory involving simulated laryn- \ngeal tension and subglottal pressure information.\n\nI wish to acknowledge the invaluable advice and guidance of Pro- \nfessor Kenneth N. Stevens of the Massachusetts Institute of Tech- \nnology and Doctor James L. Flanagan of the Bell Telephone Labora- \ntories. I would also like to thank Doctors Harry Levitt and Peter \nBrady for valuable comments and criticisms of this manuscript.\n\nHenke, W., Dynamic Articulatory Model of Speech Production Using Com- \nputer Simulation, Ph.D. thesis, Massachusetts Institute of Technology, \n1966.\n\nRules, Speech Communication Seminar, Stockholm, August 29 to Septem- \nber 1, 1962, Paper F2\n\nGold, B. and Rabiner, L., Analysis of Digital and Analog Formant Synthe- \nsizers. Paper presented at 1967 Conference on Speech Communication and \nProcessing, Massachusetts Institute of Technology, and accepted for pub- \nlication in IEEE Trans. Audio and Elec., March 1968.\n\nRabiner, L., Speech Synthesis by Rule: An Acoustic Domain Approach, Ph.D. \nthesis, Massachusetts Institute of Technology, 1967.\n\nLieberman, P., Intonation, Perception and Language, Research Monograph \n38, MIT Press, Cambridge, Mass., 1967.\n\nUnified Matrix Theory of Lumped and \nDistributed Directional Couplers\n\nThis paper presents the theory of directional couplers using matrix \nformulation for lumped, distributed, or any linear tume-invariant black-box. \nThe starting point is the matrix theory of untform multiple coupled trans- \nmission lines in terms of which are brought in the concepts of characteristic \nampedance matrix, propagation matrix, reflection matrix, and scattering \nmatrix. Next, we use the results obtained with the theory of multiple coupled \ntransmission lines to derive expressions for the loading ampedance and \nvoltage ratios for distributed directional couplers. We do this using the \nspectral properties of the matrices. Then we generalize the concepis of \ncharacteristic impedance matrix and propagation matrix for a class of \nblack-boxes with the aid of the A, B, C, D transmission matrix. We give \nconditions for the loading impedance and expressions for the voltage \nratios, using the spectral theory of the transmission matrix. We discuss \nthe physical significance of the directional coupler effect at all frequencies \nin a vector-matrix framework and analyse in detail some lumped dtrec- \ntional couplers. Finally we discuss hybrid (lumped and distributed) \ndirectional couplers.\n\nThe directional coupler is an important device in many transmission \nsystems. The theory for the electrical design of certain types of dis- \ntributed-parameter directional couplers is well established through the \ncontributions of a number of researchers.*~\u2019\n\nThe purpose of this paper is to present the theory of transmission-line \nsymmetric directional couplers in matrix form and then extend that \ntheory to arbitrary lossless reciprocal circuits. There are many nota- \ntional and conceptual advantages in using a matrix formulation which \ngive further insight for the synthesis of different types of directional \ncouplers.\n\nThe starting point is to examine the coupled-line directional coupler \nas a particular application of multiple coupled transmission line theory \nand then to use the concepts of characteristic impedance matrix and \npropagation matrix to examine the whole problem, rather than one \nline at a time or one mode at a time. By using a matrix approach the \nfundamental properties of the modes become more evident and thus \nphysical intuition and mathematical reasoning blend to give a clearer \npicture of the situation. For instance, when the lines are no longer \nidentical it is the eigenvectors of the matrices which provide the in- \nformation of how to extend the mode concept.\n\nSeveral researchers\u2019 *\u2019 have analyzed the behavior of a set of multiple \ncoupled transmission lines. For reference in subsequent sections of this \npaper we present a brief account of this theory. For simplicity the \ndiscussion is restricted to two identical lines operating in the TEM \nmode. Fig. 1 shows schematically a differential section of two lines \nand a ground plane. The circuit of which Fig. 1 is a differential section \nobeys the following vector differential equations* in the steady state\n\nBy solving for I in (1) and substituting its value in (2), or solving \nfor V in (2) and substituting its value in (1) the following are obtained: \nav\n\n* The matrices V, I, Z, Y are all functions of frequency. For simplicity in the \nnotation this dependency is not explicitly indicated.\n\nwhere the 2-vectors.v, and v_ are arbitrary constants (dependent on \nfrequency) which depend on the boundary conditions. Their inter- \npretation is very similar to the one of single line theory, v. may be \ncalled the forward wave and v.. the reflected wave.\n\nEquations (8), (8) and (9) have the same form as the single trans- \nmission line equations. For this reason the matrices T, Y,, Z), are \ncalled propagation matrix, characteristic admittance matrix, and char- \nacteristic impedance matrix, respectively. Many properties of the \nsingle transmission line hold for the multiple case. In particular, if \nthe set of lines is terminated in a network whose open circuit impedance \nmatrix is equal to the characteristic impedance matrix of the set of \nlines, a vector of incident voltage waves traveling down the lines will \nexperience no reflection. The manner in which the voltages and currents\n\n* In evaluating WZY to calculate F the convention is made to associate with I \neigenvalues whose real part are positive and with \u2014I the ones with negative real \npart.\n\nin the lines interact is best understood by examining the matrix func- \ntions e~** and e**. (See Appendix.)\n\n2.2 feflection and Scattering Matrices \nIn single transmission line theory the voltage reflection coefficient \nI', at a discontinuity is defined as \nv. = pd. , (10)\n\nwhere v, is the incident voltage and v_ is the reflected voltage. The \nexpression for Ip in terms of the characteristic admittance Y, of the \nline and the input impedance Z, at the discontinuity is\n\na Vig Y \u2014 1 \n~ Z,Y,+1 \nThe concept of reflection coefficient can be generalized very simply\n\nfor the case of a multiple set of lines. Consider Fig. 2 depicting a pair \nof coupled lines of length 7 and matrices IT, and Y, terminated at\n\nx = Qin a device of open circuit impedance matrix Z,. According \nto (8) and (9) for z = 0 \nV0) =v+Vv_, (12) \nI(0) = Y,.(v. \u2014 v_). (13) \nThe box marked Z, obeys \nV(O) = Z_I(0). (14) \nSubstitution of V(0) and I(0) from (12) and (13) into (14) gives \nv, +tv_ = Z_Y,(vi \u2014 V-), (14b)\n\nFig. 2\u2014 Pair of coupled lines terminated in a circuit of impedance matrix Zz.\n\nDIRECTIONAL COUPLERS 43 \nThus, defining the reflection matrix Tp as a generalization of (10) \naccording to \nv. = Fev,, (16) \nit follows from (15) that \n| Ty = (Z:Y, + I) '(ZzY, \u2014 D), (17)\n\nwhich is the generalization to multiple lines of (11). \nThe scattering matrix \u00a7 is a reflection matrix in which the incident \nand reflected voltages are normalized.* It is defined by\n\nDefining the normalized load impedance matrix Z, according to \nZ, = YiZ1\u00a5i , (22) \nequation (21) may be rearranged as follows: \nw= Z@. +N\". \u2014 De... (23) \nComparison of (23) and (18) gives \nS=@Z,+1\"@, \u2014 0. (24)\n\nFrom either (17) or (24) it is clear that if the load has an open circuit \nimpedance matrix equal to Z, of the lines the reflected wave is zero \nsince both the reflection matrix Ip and the scattering matrix S vanish.\n\n(The superscript -++ denotes transposed conjugate of a matrix.) For a \nlossless device the incident power must be equal to the reflected power,\n\n* In the literature on the scattering matrix the variables are normalized with \nrespect to a diagonal matrix (usually a real matrix). Here the matrix may be complex \nand not necessarily diagonal. This derivation provides the physical interpretation \nfor a scattering matrix normalized with respect to a complex nondiagonal matrix.\n\nViewed from its terminal behavior the two coupled lines and ground \nof Fig. 3 can be studied as a four port which may be characterized by, \namong others, an impedance matrix \u00a2 or a transmission E matrix. \n(The E matrix is the extension to 2N ports of the concept of the A, B, \nC\u2019, D parameter matrix of two ports. See Ref. 11.)\n\n(The prime indicates the transpose matrix.) The E matrix, written in \npartitioned form, is\n\n\u2014 Bal _ | cosh T-l \nC|D Y, sinh T-l \nIf the two lines in Fig. 3 are identical then I is symmetric and hence \nI\u2019 = I so that in (29), A = D. Furthermore, for this case all the \nmatrices Z, , Yo , Y, A, B, C, D and their analytic functions commute \nand therefore may be treated unambiguously as ordinary numbers. \nFor instance (29) may be written\n\nThe analogy between the E matrix of (30) and the A, B, C, D parameter \nmatrix of a single transmission line considered as a two port is now \ncomplete.\n\nA symmetric matched directional coupler is a four port device whose \nscattering matrix is of the form of S of (81) when the ports are ad- \nequately numbered. 8 and 6 are, in general, frequency dependent. \nIt is well known in microwave circuits that, given any lossless reciprocal \n4-port, if all ports are matched, then the device is a directional coupler.\u201d \nBy properly numbering the ports it may be assumed that the directional \ncoupler has the following scattering matrix\n\nFor the case of two identical lossless coupled lines equally loaded \nat the four ports all that is necessary to obtain a directional coupler \nis to match one of the ports. This results from the great symmetry.\n\nFig. 4 shows two identical lines loaded at ports 2, 3, and 4 with \nequal resistances, #. To calculate the driving point impedance at \nport 1 the following procedure can be used. Consider Fig. 5. The \nequation\n\nFig. 4 \u2014 Two-terminal circuit formed by a pair of coupled lines above a ground \nplane loaded with resistances at three ports.\n\nis the matrix version of the well known formula for calculating the \ndriving point impedance at one port of a two-port when the other \nport is loaded. In (82) Z is the open circuit impedance matrix of the \ntwo-port at the left in Fig. 5, A, B, C, D are the matrices given by (29).\n\nOnce the Z of (82) is known the resistor R may be connected to \nport 2 as shown in Fig. 6 and the driving point impedance at port 1 \ncalculated by a second application of (82) although now, instead of \nmatrices, A, B, C, D are scalars. The matrices in (82) are all of the form\n\nFig. 5\u2014Intermediate circuit for calculating the driving point impedance at \nport 1 of the circuit in Fig. 4.\n\nFig. 6.\u2014Second intermediate circuit for calculating the driving point im- \npedance at port 1 of the circuit in Fig. 4.\n\nwhich hold for matrices (and hence for scalars considering them as \n1 X 1 matrices) may be used. The second application of (82) with \nthe aid of (88)-(41) gives\n\nFor port 1 to be matched z must equal R, hence, the condition for \ndirectional coupler effect is\n\nWhen the values given by (86) and (37) are substituted in (48), after \nsome algebra, the following equation results:\n\nIf a matched directional coupler at all frequencies is desired, (44) \nmust be satisfied at all frequencies. Some possible mathematical solu- \ntions are the following:\n\nEquation (45) or (46) and (48) give possible conditions for directional \ncoupler effect at all frequencies. If a narrow band directional coupler \nis desired one may match the coupler at the discrete frequencies which \nsatisfy (47). Since (47) is a transcendental equation it 1s not unreason- \nable to expect an infinity of roots. For instance, suppose (46) holds, \nand J is made so that\n\nThen the device will be a directional coupler at the frequencies that \nare roots of (49).\n\nSome explicit relationships for two identical lossless lines of inductance \nper unit length Z,, , mutual inductance per unit length L,. , capacitance \nper unit length of one line alone C, and capacitance between the two \nlines per unit length Cy, , are:\n\nwhich is the same as (55). Equation (54) implies that the lines have \nnegative mutual inductance, which is not achievable with parallel \nlines. This condition can, however, be satisfield with counter-wound \nlumped elements.\n\nThat is, for the frequencies given by (57), independent of the value \nof the loads (as long as they are all equal) the lines will be matched \nand will exhibit directional coupler effect.\n\nEquations (54) and (55) or alternatively (56) and (55) are not fre- \nquency-dependent. This means that the resulting circuit will be matched \nfor all frequencies. Therefore, the directional coupler effect will exist \nfor all frequencies, meaning that the coupling between uncoupled ports \nis zero at all frequencies. However, the coupling between coupled \nports is frequency-dependent. This dependency is derived as follows.\n\nConsidering the coupled transmission lines as the load to four un- \ncoupled lines, each of characteristic impedance FR as shown in Fig. 7, \nthe scattering matrix of the load is calculated according to (24) and \n(22) with\n\nand Z, given by \u00a2 of (28). Applying the methods of the appendix the \nmatrix \u00a2 may be expressed as follows:\n\nwhere R, , R; , R. , Ra are the spectral set of { and they are given by \n(165)\u2014(168) of the appendix and dz, , Ags , Age \u00bb Aza ATE the eigenvalues \nof \u00a2 and are given by\n\nFig. 7\u2014 Two coupled lines above a ground plane of impedance matrix given \nby equation (28) serve as load to four uncoupled lines of characteristic imped- \nance FR for calculation of scattering matrix of coupled lines.\n\nIf the values of Aya, Ara , Age \u00bb Ara aLe Substituted in (65)\u2014(68) and those \nare in turn substituted into (64) while the conditions indicated by \n(46) and (48) are imposed, the following result is obtained after some\n\nIf, instead of the condition of (46), the one of (45) is imposed plus (48) \nthen the following result is obtained\n\nwhere \n1 7 \nSis = Gshyl-+amhyl? cahyl+embyt? 9) \n; i \nSu = cosh y T+ sinhyl coshyl+sinhyl? = / \nwhere\n\nThe matrices of (69) and (75) with the aid of (70)\u2014(74) and (76)-\u2014(80) \ngive the frequency dependency of the coupling between the ports for \nthe two types of directional couplers derived (matched at all frequencies)\n\nTwo sets of conditions have been derived for obtaining the directional \ncoupler effect at all frequencies for transmission lines. Fig. 8 represents \na pair of lossless lines c, d which are coupled from x = \u2014/ to xz = 0. \nThe coupled lines are connected to four (uncoupled) lossless transmission \nlines a, b, e, f, each of characteristic impedance & and each terminated \nin &. The matrix Z, is the characteristic impedance matrix of the set \nof coupled lines (c and d in Fig. 8) and I its propagation matrix.\n\nConsider a case in which the eigenvalues of the matrix Z, of the \nlines c, d satisfy\n\nwhere I is the unit matrix. Assume that a pulse travelling down line \na occupies at time \u00a2, the position shown in Fig. 8. Lines a and b together \nmay be considered as a particular case of a set of multiple lines, in \nwhich both (Z,),, and (I),,\u2014the characteristic impedance and propaga- \ntion matrices of lines a, b\u2014are diagonal because lines a and b are \nuncoupled. Thus it is convenient to think of the pulse traveling down \nline a as a vector of pulses traveling down the two lines a, b; the second \ncomponent of the vector which corresponds to line b being zero. When \nthe vector of pulses arrives at the position \u00ab = \u2014l, indicated by \nin Fig. 8, the vector of pulses continue \u2018\u2018seeing\u2019\u201d\u2019 the same characteristic\n\nFig. 8 \u2014 Pair of coupled lossless lines c and d above a ground plane which has \nuncoupled terminated lines a, b, e, f of characteristic impedances R. This shows \nthe progress of a pulse to explain the directional coupler effect in physical terms \nfor a coupler with diagonal characteristic impedance matrix.\n\nimpedance matrix RI as before and hence no reflection of the vector \nis created at AZ.\n\nIn a vector formulation one speaks of reflections in a multidimen- \nsional sense. The voltage on line a may give rise to a reflected voltage \non line a [self reflection] or to a reflected voltage on line b [mutual \nreflection]. If all the lines are self matched and mutually matched \nthere will be no reflections whatever. Often a line might be self matched \nbut not mutually matched; then no self reflection will occur, but a \nmutual reflection will.\n\nAt time \u00a2, the second component of the vector pulse is no longer \nzero because the pulse on line ec which is coupled to line d induces a \npulse on line d as shown in Fig. 8. These component pulses will be\n\n* Although for simplicity in the explanations, parallel lines are assumed, this\n\nkind of coupler requires negative mutual inductances which are in general achieved \nwith counter-wound helices.\n\ndistorted because continually varying portions of them travel at \ndifferent speeds on the lines c and d.* When they reach x = 0 marked \nby N in Fig. 8, the vector again continues to see the same characteristic \nimpedance matrix and hence no reflections are caused at N. Finally \nthe pulses travel out on lines e and f and are dissipated at the resist- \nances &.\n\nFrom this it is clear that the two ports at Jf are uncoupled and also \nthe two ports at N are uncoupled. However a port at J/ is coupled to \nboth ports at N and vice versa. All the ports are self matched. This \nexplanation suggests a simple way of determining the conditions for a \nnonsymmetric coupler realized with transmission lines. Assume lines \nec and d are no longer identical but that the characteristic impedance \nmatrix of the set is still diagonal\n\nwhile the propagation matrix I is not. If the lines a and e have char- \nacteristic impedance (Z,),, and lines b and f have characteristic imped- \nance (Zo)o. and a, b, e, f are properly terminated, the coupled lines \nc, d will constitute a matched nonsymmetric directional coupler since \nthe discussion above holds for this case without modification.\n\nTig. 9 shows the same arrangement as Fig. 8. A pulse traveling \nto the right on line a is shown at \u00a2 = \u00a2, . The second component of the \nincident pulse corresponding to line b is zero since lines a and b are \nuncoupled. As the vector of pulses reaches the position M, the vector \nof voltage is reflected according to a reflection matrix because for \nthis case, the vector no longer sees the same characteristic impedance \nmatrix in the transition from lines a, 6 to lines c, d. However, because \nline a matches line \u00a2 there is no self reflection; only a mutual reflection \nappears on line b. Besides the reflected pulse, an identical transmitted \npulse appears on line d at time #, as indicated in Fig. 9. The appearance\n\n* The propagation matrix should not have equal eigenvalues, otherwise it will be\n\ndiagonal which, together with a diagonal characteristic impedance, implies un- \ncoupled lines.\n\na \\M \u20ac N e \n| \nt, ita t3)t3 \nw: VL. SULLLELLLLLA meV LEO ELEELELALLELAALAL LA ELL LLL LLL LET EY x 0) | 7 VALLE EELS Et \n_> <\u2014;\u2014 > <\u2014|\u2014> R \nt to lt oF (-\u2014) \n1 2 ve t3, ts \nfie ec ee es es a ee pe ee ee en etre ee A Ee \n| d ' f\n\nFig. 9\u2014Same structure as in Fig. 8 showing the progress of a pulse for a \ncoupler with scalar propagation matrix.\n\nAfter \u00a2, , as the two forward pulses travel along lines c and d they do \nso at the same speed, without distortion and do not interact with each \nother since the propagation matrix is diagonal. As the two pulses \narrive at point N they encounter a reflection matrix of opposite sign \nto the one they encountered at 17. This means that the pulse on line c \npasses through N undisturbed but creates on lines d and f reflected \nand,transmitted pulses identical to the ones created at M but of op- \nposite sign. Likewise, the incident pulse on line d goes right through NV \n(since each individual line is matched) creating transmitted and reflected \npulses on lines e and c, but being cancelled on line f by the transmitted \npulse created by the incident pulse on line c; thus, nothing comes out \nof line f.\n\nAt time \u00a2; right after the reflection at N the situation is depicted \nin Fig. 9. After t; the reflected pulses are traveling to the left at the \nsame speed undisturbed and undistorted on lines c and d. As they \narrive from the right at point M the pulse on line d goes out line b \nundisturbed but creating transmitted and reflected pulses on lines a \nand c. Likewise the pulse on line c goes out line a undisturbed creating \ntransmitted and reflected pulses on lines 6 and d, but being cancelled \non line a by the transmitted pulse created by the incident pulse on \nline d. This eliminates any delayed reflections on line a to the original \nincident pulse. The process continues in the same manner, the outgoing \npulses on lines a and f always being such that they cancel. This means \nthat the ports associated with lines a and f.are uncoupled, but the \nports of lines a, b and e are coupled.\n\nIt is clear that what is necessary for directional coupler effect on this \ntype of coupler is: all ports self matched, equal propagation velocities \nwithout attenuation or distortion. Hence it should be possible to \nrealize a nonsymmetrical directional coupler of this type whose propaga-\n\ntion matrix is a scalar matrix* by self matching its ports. This may \nbe useful for interconnecting lines a and b of different characteristic \nimpedances.\n\nBecause the reasoning above was made in the time domain with \npulses of arbitrary shape, the results hold for all frequencies. This \nis an example of the gain in insight owing to the vector-matrix for- \nmulation.\n\nLet us generalize several concepts introduced in Section 2.1. Con- \nsider a (2N + 1)-terminal network in which terminal 2N + 1 will \nbe grounded and ports from terminals 1 through 2N to ground will \nbe considered. Ports 1 to N will be considered input ports and Ports \nN + 1 to 2N output ports. Suppose the 2N-port is characterized \nby A, B, C, D N X N matrices. (Extensions to 2N-ports of the A, B, \nC\u2019, D parameters of a two-port). Assume the circuit is such that\n\nwhere I is the N X N unit matrix and A, B, C are N X N symmetric \nmatrices which commute.\n\nBy analogy with a multiple transmission line the characteristic \nimpedance matrix Z,) and the propagation matrix I are defined so \nthat they satisfy the following equations: :\n\nThe N X N matrices \u00a9 and Z, will also be symmetric and commute. \nThe matrix Z, is the open circuit impedance matrix of that network \nwhich, when connected to the output ports N + 1 through 2N of\n\nthe circuit whose A, B, C matrices are those of (86) and (87), will \nresult in an open circuit impedance matrix of Z, when the circuit \nis viewed at its ports 1 through N. This property is analogous to the \none of the characteristic Impedance matrices of a set of multiple \ncoupled transmission lines. The matrix TF, has virtually the same \nproperties of the matrix Tl of a set of coupled lines (although the \nsingle quantity J looses its significance as a length in case the 2N-port \nis a lumped circuit). For instance, if n identical 2N-ports are cascaded \nthe resulting 2N-port has a propagation matrix equal to nT.\n\nThe matrices I and Z, may be expressed in terms of the 2N XK 2N \nimpedance matrix Z of the 2N-port with the aid of (88) through (41)\n\nthe submatrices Z,, , Zi2 , Zo, , Zoo are N X N symmetric matrices \nand commute with each other. The characteristic impedance matrix \nmay also be expressed in terms of the so-called open and short impedance \nmatrices. If the N-vector V, and I, denote the voltages and currents \nat the N input ports and V, and I, denote the voltages and currents \nat the output ports, then if I, = 0, that is, the terminals on the output \nports are open then\n\nEquation (99) gives an experimental method of determining Z, if \nit is known that the network satisfies equations (81) and (82), and \nthe symmetry and commutativity conditions.\n\nIt is often convenient to analyze some lumped or distributed (or \ncombinations of lumped and distributed) systems as though they were \nmultiple transmission lines using such concepts as reflection matrix \nand incident voltage.\n\nConsider the connection shown in Fig. 10. Each network is an (2N-+1)- \nterminal network in which ports from each terminal to ground are \nmade. Ports 1 through N and 1\u2019 through N\u2019 are considered input ports. \nPorts N + 1 through 2N and (N + 1)\u2019 through (2N)\u2019 are considered \noutput ports. The voltage vector at the junction B whose components \nare the voltages of nodes N + 1, N + 2, --- , 2N to ground is denoted \nby Vz.\n\nThe vectors v, , V_, i, , i_, called incident voltage, reflected voltage, \nincident current, and reflected current at the junction B (assuming \nthe direction of propagation from left to right), are defined to satisfy\n\nFig. 10 \u2014 Connection of two 2N ports in cascade used to define the reflection \nmatrix.\n\nfrom the ports NV + 1, N + 2, --+ , 2N with network 1\u2019 disconnected. \nThe reflection matrix Vy,;,,- is defined according to\n\nVi = TV acarce-V\u2014 . (104) \nThe matrix T'y,;,,- satisfies the following relationship: \nTD arare \u2014 (Z/Z* + | Od A Ae \u201c= I), (105)\n\nwhere Z\u2019 is the N X N open-circuit impedance matrix of network \nM\u2019 as seen from ports 1\u2019, 2\u2019, --- , N\u2019 with network M disconnected. \nTis the N X WN unit matrix. The indices MM\u2019 on Yara, indicate the \ndirection of propagation from M to M\u2019. If the indices are reversed \nthe roles of Z\u2019 and Z are reversed, that is\n\nTaking advantage of the derivations done for transmission line \ndirectional couplers and the analogies introduced in Sections 5.1 and \n5.2, it is possible to write without further work, the equations of a \ndirectional coupler having the same mathematical symmetry of a \nmultiple transmission line directional coupler but which may have \nlumped components or combinations of lumped and distributed com- \nponents. Suppose a four-port is characterized by its E matrix whose \n2 X 2 submatrices A, B, C, D have the form of the matrix K of equa- \ntion (1) of the Appendix and satisfy equations (81) and (82). Without \nany further work it can be stated that if the load impedance z and the\n\nthe device will be a directional coupler for that frequency. \nThis result is deduced from equation (45). Likewise from equation \n(46) it may be deduced that if the four port is such that\n\nthen the device will also be a directional coupler for those frequncies \nfor which (110) and (111) are satisfied.\n\nIt is convenient at this point to exemplify with a simple lumped \ncircuit.\n\nConsider the four-port lumped circuit shown in Fig. 11. The E \nmatrix of the circuit of Fig. 11 may be calculated by cascading 3 sec- \ntions, the first and third containing only capacitors and the second \ncontaining the inductors and mutuals. By proceeding carefully much \nlabor can be saved using the spectral sets given in the Appendix. The \nresults are\n\nDIRECTIONAL COUPLERS 61 \nFrom (86) the hyperbolic sine and cosine of the propagation matrix is\n\nIf this condition is to be satisfied at all frequencies then \n(Li \u2014 Dy)(C + 2Cu)\u201d = in \u2014 Li2)C, (118) \n21a, + LC + 2Cy) = 2h \u2014 Ly)C. (119)\n\nBoth (118) and (119) are satisfied for the following choice: L,, = \u2014Ly., \nC = 0. Thus the circuit of Fig. 11 with C = 0 and Z,, = \u2014JZy,,. (per- \nfectly coupled counterwound inductors) is a directional coupler at all \nfrequencies, provided it is loaded at all ports with the impedance\n\nThe impedance z is frequency dependent. The voltage ratios will be \ngiven by (76) and (77),* that is\n\nIt is often convenient to express the equations of a directional coupler \nin terms of the A, B, C, D matrices directly instead of the Z,. and \nmatrices. Tor this purpose assume a lossless reciprocal four-port is \ncharacterized in terms of its A, B, C, D matrices which are of the form \nof the matrix K of equation (150) of the Appendix. Assume A = D. \nThe condition A\u201d \u2014 BC = I is automatically satisfied if the circuit \nis reciprocal. A, B, C, D commute, since they have the same eigen- \nvectors. Because all matrices commute they may be treated without \nambiguity as scalars. The open circuit impedance matrix is\n\nSuppose the ports are loaded with equal impedance z. The impedance \nmatrix \u00a2 normalized with respect to the matrix zI is\n\nwhere A\u201d, A\u2019, C*, C\u201d are the eigenvalues of A and C associated with \nthe sum and difference modes. (See Appendix.) The reflection matrix\n\n* Because for lumped elements F corresponds to rl in using the formulas derived \nfor distributed elements for circuits with lumped elements one should take 1 = 1.\n\nwhere R,, R,, R., Rz are the members of the spectral set of %, and \nwhich are given by equations (165) to (168) of the appendix.\n\nTo find the condition for self match at port 1 (which because of the \nsymmetry gives the condition of self match at any port) the eigenvalues \ngiven by equations (125) to (128) are substituted in (1380) and the \nupper left corner of 'z is equated to zero. After some algebra this yields\n\n(Apes) CA) sae yes 1 \nei CAP ea\") = A eee a0, 6 IB) \nEquation (131) 1s a quartic in z which may be rewritten \nz2(C*C\u2019) + 2(C*A7 + C7 A*) \n\u20142(A*B +B\u00b0A)-BB =0 (132)\n\nThe solutions of Equation (118) give the values of the impedances which \nwill match the four ports in terms of the eigenvalues of the matrices \nA, B, C. Although a quartic algebraic equation can be solved in terms \nof the coefficients, the solution is extremely cumbersome algebraically \nand it would be very difficult to see the effect of varying the quantities \nA*, A\u2019, B*, B\u2019, C\u2019, C\u2019. A sounder approach is probably to look at \nparticular simple cases. For instance if\n\nEq. (182) then reduces to \n\u20142(B*A\u2019) \u2014-B\u00b0B\u2019 = 0, \nwhose solution is \nz= \u2014-\u2014 > (136) \nObviously there are many possibilities. Some of the solutions are not\n\nis a root of equation (132). This fact can only be seen after a good deal \nof algebra, for this reason it is convenient to express (132) in different \nways so that different possibilities may be \u201cseen.\u2019\u201d? With this in mind \nequation (129) may be written\n\nUsing the spectral set of %, the following alternative expression for \nthe condition for the self-match of all ports is obtained\n\nIt is simpler (although not trivial) to verify that the conclusions asso- \nelated with equations (137) and (138) are true from (140) than from \n(132).\n\nThe reflected voltages caused by an incident voltage at port 1 may \nbe obtained from equation (130).\n\nVig CA EO a7. Oa Se \nEquations (123) through (148) all are good for any four-port, whether \nlumped, distributed, or made with combinations of lumped and dis- \ntributed elements. \n5.5 A Degenerate Situation:\n\nConsider the circuit of Fig. 12. Notice the structure is not physically \nsymmetrical. The A, B, C, D matrices of the circuit are\n\nAlthough in equation (144) A and D are apparently not equal, it turns \nout that Y and Z are orthogonal and therefore YZ = 0. Thus (144) reads\n\nPS ps : \nE 4 ~ LY\u00a5{I | a \nThe matrices A, B, C, D commute and satisfy equations (81) and \n(82). Thus, although the structure is not physically symmetrical, it \nis electrically symmetrical. When one attempts to use equation (87)\n\nto determine Z, one finds that the matrix C~* does not exist because \nC = Y is singular. Since Z, does not exist, equations (48), (70), and\n\nFig. 12\u2014 Lumped directional coupler which, when connected to a resistive load, \nexhibits directional coupler effect of all frequencies.\n\nThe load impedances are frequency invariant, which indicates that \nthe circuit may be matched with a constant at all frequencies and \nshould exhibit directional coupler effect at all frequencies when loaded \nwith a positive resistance of value ~WL,,/C,,. Using equations (141)- \n(143) the voltage ratios are found to be\n\nEquation (149) corroborates that the coupler exhibits directional coupler \neffect at all frequencies. This example illustrates the use of the direc- \ntional coupler equations in terms of the A, B, C, D matrices.\n\nThe formulation that has been developed allows the handling of \ncircuits with both lumped and distributed elements without any changes \nbecause the formulas are good for \u201cblack boxes.\u201d For example, for \nthe circuit shown in Fig. 13, the total E matrix is found by multiplying \nthe individual E matrices of the sections. The E matrix of section P \nor 7 is given by equation (145) while that of Q is given by equation \n(29). Once the total E matrix is known, it is partitioned into A, B, C, D \nmatrices and equation (132) applied to determine the proper z for \nterminating the coupler. When the coupler is thus terminated, equa- \ntions (141)\u2014(1438) yield the voltage ratios.\n\nFig. 13\u2014 Directional coupler configuration containing both lumped and dis- \ntributed elements.\n\nIn general, the algebra will get quite unmanageable and one will \nhave to resort to numerical calculations on a digital computer at \ndiscrete frequencies..\u00b0 The z may be found by numerically solving \nthe quartic equation (132) at a set of discrete frequencies and then \nrealizing it as a driving point impedance through successive approx- \nimations, or some similar procedure. The processes of normalizing the \nimpedance z and of making frequency transformations can be used \nvery effectively in the realization of directional couplers of this sort.\n\nAlthough strictly speaking all physical devices are distributed in \nspace and thus, in general, have transcendental transfer functions for \ncertain frequency regions, it might be possible to model the devices \naccurately enough with conventional ideal lumped elements, or more \ngenerally with elements having given frequency curves, which may \nbe given analytically or numerically. In this paper we give a matrix \ntheory for lumped and distributed circuits, keeping this fact in mind.\n\nBy using matrix formulation and treating the circuits as black boxes, \nit is possible to extend the classic theory of stripline directional couplers \nto more general circuits while still keeping many of the concepts (such \nas even and odd characteristic impedance) that have been found useful.\n\nThe paper makes evident the fact that the concepts of odd and even \nmode arise because of the special symmetry of the matrices and that \nthey correspond to their eigenvectors and eigenvalues. We indicate \nin the appendix that when such special symmetry is lost, the odd and \neven modes are also lost and it might be necessary to introduce one \nset of modes for the currents and another for the voltages. This fact \nis not simple to see without the matrix formulation.\n\neffect and gain considerable insight which is useful for realizing direc- \ntional couplers lacking double symmetry.\n\nWe have given general equations for analyzing and designing black- \nbox directional couplers in terms of the characteristic impedance and \npropagation matrices, and in terms of the transmission A, B, C, D \nmatrices. The latter may be necessary to analyze circuits whose char- \nacteristic impedance matrix (hence even or odd characteristic imped- \nances) does not exist but which have a scattering matrix.\n\nThe topic of the actual design of directional couplers with lumped \nor with lumped and distributed elements, and more specifically the \ndesign of multisection directional couplers using computer aids, is not \ntreated because it is the subject of a forthcoming paper.\n\nHere are some spectral properties of the principal matrices in the \npaper for ease of reference. *\n\nThe quantities associated with the vector U, are called the \u201ceven mode\u201d\u2019 \nor \u201csum mode.\u201d The quantities associated with the vector U, are \ncalled the \u2018\u2018odd mode\u201d\u2019 or \u2018\u2018difference mode.\u201d\n\nIn our main paper, the eigenvalues of the matrices Z, and I which \nare in the form of equation (1), are denoted by Z4 and y\u201d for the sum \nmode and Z;, and y for the difference mode.\n\nThis can be verified by a matrix multiplication similar to that of equa- \ntion (154).\n\nConcerning the so-called \u201c\u2018modes of propagation\u201d of a set of two \ncoupled lines, when the I matrix of the two lines has the double sym- \nmetry exhibited by the matrix K of equation (1), the eigenvectors \nof the matrix are those given by equation (4). Since the matrix is \nsymmetrical, the eigenvectors of the transposed matrix I\u2019 are the \nsame; therefore, one may speak of the \u2018\u2018sum mode voltages and cur- \nrents\u201d and \u2018\u2018difference mode voltages and currents.\u2019\u201d\u2019 However, it might \nhappen that I does not have the form of K in equation (1). Then the \nconcept of sum and difference modes disappear because the eigenvectors \nthat will result will not be quite so simple. If he wishes, one may then \nspeak of \u2018\u201c\u2018first mode\u201d and \u2018\u2018second mode,\u201d associating each mode with \neach eigenvector.\n\ncoincide. However, if I is not symmetrical then the eigenvectors of \nIr\u2019 will not be the same as those of Fr. It will be necessary to speak of \n\u201cfirst voltage mode,\u201d \u201csecond voltage mode,\u201d \u201cfirst current mode,\u201d \nand \u201csecond current mode\u201d because the voltage modes will differ from \nthe current modes if I is not symmetrical. The eigenvalues of a matrix \nand its transpose are always the same, hence no distinction is necessary \nfor the propagation constants of the voltage and current modes.\n\nFirestone, W. L., Analysis of Transmission Line Directional Couplers, Proc. \nIRE, 42, October 1954, pp. 1529-15388.\n\n. Fel\u2019dshtein, A. L., Synthesis of Stepped Directional Couplers, Radiotekhnika\n\ni Elektronika, 6, No. 2, February 1961, pp. 234-240; English translation, \nPergamon Press, New York, 1961, pp. 75-85.\n\nand Transmission-Line Stepped-Impedance Filters, Proc. IEEE, 110, Feb- \nruary 1963, pp. 275-281.\n\nPipes, L. A., Matrix Theory of Multiconductor Transmission Lines, Phil. \nMag., Series 7, 24, July 1937, pp. 97-113.\n\nHuelsman, L. P., Circuits, Matrices and Linear Vector Spaces, McGraw-Hill \nBook Company, Inc., New York, 1963.\n\nMontgomery, C. G., Dicke, R. H., and Purcell, E. M., Principles of Micro- \nwave Circuits, McGraw-Hill Book Company, Inc., New York, 1948.\n\nMurray-Lasso, M. A., A Digital Computer Simulation of a Class of Lumped \nand/or Distributed Four-Ports, Proc. SHARE-A.C.M. Design Automation \nWorkshop, Los Angeles, Calif., June 19-22, 1967. .\n\nFrazer, R. A., Duncan, W. J., and A. R. Collar, Elementary Matrices, Cam- \nbridge University Press, Cambridge, 1963.\n\nThis ts a summary of data from an extensive analysts of on-off speech \npatterns in 16 experimental telephone conversations. The on-off patterns \nare determined by a fixed threshold speech detector having certain rules \nfor rejecting notse and for filling in short gaps (for example, from stop \nconsonants). Distributions are obtained for ten events, including talk- \nspurts, pauses, double talking (stmultaneous speech from both parties), \nmutual silence, etc. Particular emphasis ts placed on events surrounding \ninterruptions. The entire analysis 1s performed for three speech detector \nthresholds, since most of the data are strongly influenced by choice of \nthreshold. Observations are made about the influence of threshold on the \ndata, properties of speech invariant with choice of threshold, and differences \nbetween male and female speech patterns.\n\nA statistical analysis of the on-off speech patterns of 16 recorded \nconversations has been obtained by a computer program written by \nMrs. N. W. Shrimpton in 1968, and recently modified by the author. \nThe data can serve the following purposes.\n\n(i) They can illustrate the effect of variation of threshold setting \non the resulting speech data. This problem has been plaguing virtually \nall researchers who have attempted to arrive at the \u201cbasic\u201d talkspurt- \npause patterns, that is, patterns which represent the subjective on-off \nbehavior, either as intended by the speaker or as perceived by the \nlistener. (There is, of course, no certainty that such on-off classifica- \ntion actually occurs during normal talking and listening.)\n\n(7) They can guide the design of voice operated devices, such as \nconventional echo suppressors? or an adaptive transversal filter echo \ncanceller,? both of which have critical timing problems in the inter- \nvals surrounding interruptions.\n\n(12) They can provide material for building stochastic models of \nspeech patterns in conversations. Several studies have already used \nbasic models (such as Markov processes) to approximate talkspurt \nand pause durations.*: * These models could be useful in predicting \nconversational behavior over special circuits, such as those containing \ntransmission delay.\n\nApplicability of the data to the above-mentioned purposes is influ- \nenced by the source and nature of the specch material and by the \nspeech detector used to obtain on-off patterns. Section IT is a descrip- \ntion of the speech material, and Section III contains a description \nof the speech detector. This detector tries to yield patterns as close \nas possible to the original waveform, while making certain correc- \ntions to make the pattern representative of perceived speech patterns. \nThese corrections, requiring two arbitrary parameters, include rejec- \ntion of impulse noise operation and bridging of gaps caused by stop \nconsonants. The third parameter, threshold, has such an effect on the \ndata that the analysis is performed for a range of thresholds. In other \nrespects however, the detector preserves fine details of timing of \nevents; for example, the attack and release times are less than 5 msec.\n\nCharacterization of speech for speech detectors in the telephone \nsystem is a different problem from characterization of speech for \nmodeling conversational speech patterns. The data of the present \nstudy are not intended to provide a basis for characterizing speech \nfor telephone system speech detectors. Within the constraints of the \ncorrections described in the preceding paragraph, however, the data \ncan be extended to predict the behavior of certain speech detectors \nas explained in Section 5.2.\n\nOf the 16 conversations, eight, obtained from four male pairs and \nfour female pairs, lasted about 7 minutes each and were documented \nin a previous paper.> The remaining eight, also four male and four \nfemale pairs, lasted about 10 minutes each. The subjects talked over \na 4-wire circuit such as illustrated in Fig. 1. The losses were typical \nof a long distance call, and there were no degrading factors such as \nnoise, echo, or delay. The voices were recorded at the zero transmis- \nsion level points (0 TLP), determined to be 6 dB \u201caway from\u201d the \ntransmitters. The 0 TLP is an arbitrary reference level used to es- \ntablish relative levels in a telephone circuit.)\n\nThe members of each pair were close friends; we have found that \nconversation between strangers can be restrained and halting. Their \ninstructions were as follows.\n\n\u201cYour task in this experiment will be to converse with each other \nfor approximately 10 minutes. You may talk about anything you \nwish, but keep in mind that you will be recorded. The recording will \nbe kept private and will be used for computer analysis of speech. \nWe ask that you both talk frequently; if only one person talks the \nconversation will be of almost no value to us.\u201d\n\nThis method seemed to produce natural conversational speech which \nwas not restrained by the subjects\u2019 knowledge that they were being \nrecorded.\n\n(1) The experimental calls are recorded, and can be studied for \ncontextual material, etc.\n\n(w) The subjects are indeed conversing, rather than momentarily \nsetting the phone down, or even switching off to other persons. In \nshort, in the experimental calls, the subjects and tasks are known.\n\nof a call with active interchange, as our experimental calls generally \nhad. This is especially true in echo suppressor? and speakerphone\u2018 \nstudies.\n\nThe technique of obtaining on-off speech patterns, although already \ndocumented,\u201d is summarized as follows. A flip-flop is set any time \nspeech (full-wave rectified and unfiltered) from speaker A crosses a \nthreshold. This flip-flop is examined and cleared every 5 milliseconds, \nwith the output being a 1 if the threshold was crossed, 0 otherwise. \nThe resulting string of 1s (spurts) and Os (gaps) is examined for short \nspurts; all spurts <15 msec* are erased. After this is done, all gaps \n<200 msec are filled in to account for momentary interruptions, such \nas those due to stop consonants. The resulting on-off pattern consists, \nby definition used here, of talkspurts and pauses. An identical procedure \nis used for speaker B.\n\nThree thresholds have been chosen: \u201445 dBm0Ot (most sensitive), \n\u201440, and \u201435. These values seemed to bracket the range between \nexcessive noise operation and insufficient speech operation. The average \npeak level (apl)f for all 32 speakers was \u201418.9 dBm re 0 TLP, 26.1 dB \nabove the most sensitive threshold. If one prefers VUs, a previous \nstudy\u2019 showed that VUs obtained by Miss K. L. McAdoo (an ex- \nperienced VU meter reader) are roughly 6 dB below the apls, hence, \nthe average VU for that observer would have been near \u201425 dBm.\n\nAlthough all means reported here are exact (in that they equal the \ntotal time in an event divided by the number of event occurrences) \nthe medians are not exact because the measuring intervals are arbitrarily \ncategorized. For example, the median talkspurt at the \u201445 dBm \nthreshold is somewhere between 750 and 800 msec, and is reported \nat 775 msec, the interval midpoint. The measuring intervals are roughly \nproportional to the lengths of events; the intervals are as short as \n10 msec for events $200 msec and as long as 1 second for events \n> 6 seconds.\n\n* This was originally 10 msec, but 15 msec seems to be required for good im- \npulse noise rejection.\n\n< Average peak level is a measure of speech level based on the average log \nrectified speech voltage.\u2019\n\n(z) Percent of time each person talked, averaged over 32 persons, \nobtained for each person by dividing his total speech time by the \nlength of his conversation.\n\nTable IL shows two measures made on the entire sample of 137.4 \nminutes of conversation: the percent of time in double talking, and \nthe percent of time in mutual silence. Notice that mutual silence is \nthe complement of the event that one or both speakers are talking.\n\nTen events were defined and measured. Figs. 2 through 11 are cumu- \nlative distribution plots of the events. The arrows show which event \nis being measured. For example, in Fig. 2, which shows the talkspurt \ncumulative distribution, there are three events illustrated and indi- \ncated by the arrows.\n\n(v) Alternation silence\u2014the period of mutual silence between the \nend of one speaker\u2019s talkspurt and the beginning of the other\u2019s. Event \n5 is a subset of 4. If a speaker alternation results from an interrup- \ntion so that there 1s no mutual silence period, then an alternation \nsilence has not occurred. (There are no negative alternation silences.)\n\n(vz) Pause in isolation\u2014a pause in which the other speaker is silent \nthroughout the pause. Event 6 is a subset of both 2 and 4.\n\n(viz) Solitary talkspurt\u2014a talkspurt which occurs entirely within \nthe other speaker\u2019s silence. Event 7 is a subset of 1.\n\n(vii) Interruption\u2014if A interrupts B, the time at which A\u2019s talk- \nspurt begins determines the start of an interruption. The interruption \nterminates at the end of A\u2019s talkspurt, unless B stops and then inter- \nrupts A, in which case A\u2019s interruption terminates upon B\u2019s counter \ninterruption.\n\n10 20 40 60 100 200 400 600 | 2 4 6 6810 20 40 60 \n\u2014\u2014\u2014 MILLISECONDS SECONDS \nSPEAKER \nA DOOM EXQy SON MOM wW OOOO \nSPEAKER [eB 70 A ATO Bke--->| B To Ales \nB\n\nFig. 6\u2014 Alternation silences for 32 subjects. A to B and B to A have been \ncombined.\n\nFig. 7 \u2014 Pauses in isolation for 32 subjects. Events from A and B have been \ncombined.\n\nFig. 8\u2014 Solitary talkspurts for 32 speakers. Events from A and B have been \ncombined.\n\nSPEECH ROSES QQC.\u00aeMeuuw MSG RSP Mowm A837 \nB KB kB KB>  k--B-=ft TA kB \nSPEECH QO MON DOH REX \u2122|QAQQ4yg Ry\n\nSPEECH MS S| SBS SX \u201cngn SSS SSn1w _ ESXV\u201c\u201coyr \n4 AA =| KA KAA as SA Kye \nSPEECH Sy SQ99y Ss\u201c MEXSSAA\u2019WN SXs9ou\n\nof 6\u2019s talkspurt is entered here, unless A pauses and then again inter- \nrupts the same B talkspurt. The first \u201cspeech after interruption\u201d \nwould terminate upon A\u2019s reinterruption, and a second speech after \ninterruption would begin.\n\n(x) Speech before interruption\u2014if A interrupts B, B\u2019s speech in- \nterval up to the interruption is entered here. If A then pauses at time \nt; and reinterrupts at time ts, (assuming B continues talking), a new \nB speech before interruption (f2-\u00a2,) is entered. If A continues talking \nand B pauses and then counter interrupts, the length of B\u2019s pause is \nentered as A\u2019s speech before interruption.\n\nTable III lists the mean, median, and number of events for talk- \nspurts, pauses, and doubletalks of the entire 137.4-minute conversa- \ntion sample. Notice that the talkspurts and pauses represent 274.8 \nminutes of speech, since the A and B speech samples can be sepa- \nrated and placed end to end.\n\nTable IV lists the means of the averages of the categorized events \nper person (or in some cases per conversation). For example, the\n\nmean talkspurt length of 1.866 second for a \u201445 dBm threshold is \nthe average of 32 numbers, each in turn being the average talkspurt \nlength for a particular speaker. The o reported* is the standard devia- \ntion of the 32 (or 16) averages among speakers (or conversations).\n\n} Values of n were obtained by dividing the total number of events for \u201440 dBm \nthreshold by k. These numbers thus give a rough idea of the frequency of these events \nper person (or conversation). These same values also apply to Table V.\n\nTable V lists the means of the medians of the categorized events \nper person (or conversation). For example, the \u201caverage of median\u2019\u2019 \ntalkspurt length of 0.788 second for a \u201445 dBm threshold is the \naverage of 32 talkspurt medians, with 0.229 second as the standard \ndeviation of the 32 medians among speakers.\n\nTable VI lists the means of the averages of the events per person \nor conversation, for the men and women separately. (Data on aver- \nages of the medians are available from the author.)\n\nAs we mentioned, some researchers have postulated first-order \nMarkov processes to model speech patterns. A conversation, at any \ninstant, can exist in one of four states depending on who is talking:\n\nneither, A, B, or both. If the conversation is in state 21 (1 = 1,2,3,4) at \nsome time \u00a2#, it may be of interest to know the probability of being \nin state 7 (7 = 1,2,3,4) at \u00a2 + At, possibly to establish a crude Markov- \nian model for distributions of times in each state. Notice, however, \nthat the simplest Markovian model will predict that each state will \nhave an exponential distribution, which is a hypothesis not generally \nsupported by the data. (For example, mutual silence is the event rep- \nresenting the first state, and a glance at Fig. 5 shows that this dis- \ntribution is strongly colored by the 200 msec fill-in time.)\n\nThis paper is primarily a collection of data, and is not intended \nto pursue the problem of modeling conversational behavior. We shall \ntherefore simply list the transition matrix for the \u201440 dBm threshold*\n\n* Transition probabilities for the other thresholds may be obtained from the \nauthor.\n\nThreshold \nEvent (dBm) \n\u201445 \nTalkspurt \u2014A0 \n\u201435 \n\u201445 \nPause \u201440 \n\u201435 \n\u201445 \nDouble talk \u201440 \n\u201435 \nMutual \u201445 \nsilence \u201440 \n\u201435 \nAlternation \u201445 \nsilence \u201440 \n\u201435 \nPause in \u201445 \nisolation \u201440 \n\u201435 \nSolitary \u201445 \ntalkspurt \u201440 \n\u201435 \n\u201445 \nInterruption \u2014 40 \n\u201435 \nSpeech after \u201445 \ninterruption \u201440 \n\u201435 \nSpeech before \u201445 \ninterruption \u201440 \n\u201435\n\n* Compares events at a common threshold. \n+ Compares males at \u201440 dBm with females at \u201445 dBm threshold; similar for \nlast column. All significance levels from #-test.\n\nin Table VII. The table indicates transition probabilities for 5 msec \ntime steps. For example, if both are talking, the probability 1s 0.98095 \nthat they will still both be talking 5 msec later. The conversation will, \ntherefore, leave the state with p = 1.0 \u2014 0.98095 = 0.01905. If a \nPoisson termination process* is assumed for terminating the event, \nthe conversation would leave the state in 1 msec with p = 0.01905/5.\n\nEvents of one subject that do not involve interaction with talk- \nspurts of his partner include pauses in isolation and solitary talk- \nspurts. Data on behavior during double talking should be contrasted\n\nwith data on these \u201cisolated\u201d events rather than, for example, the \ndistribution of all talkspurts, since this distribution includes events \nduring double talking.\n\nThe data are notably influenced by threshold changes. The author \ndoes not believe it is possible, from results reported here, to establish \na single \u201ccorrect\u201d threshold. It is possible, however, to draw certain \nconclusions which are threshold independent (see (vz) and (vit) below, \nand Section 5.2.)\n\n5.1 The Data \nWe know from the data that: \n(2) As the threshold 1s raised (speech detector made less sensitive),\n\nevents which measure periods of talking tend to decrease in length, \nsince the longer events tend to be broken up into short ones. These\n\n* A good discussion of Poisson and Markovian processes may be found in Ref- \nerence 8.\n\nevents include talkspurts, double talks, solitary talkspurts, interrup- \ntions, and speech before and after interruption.\n\n(it) As the threshold is raised, events which measure periods of \nsilence tend to increase in length. These events include pauses, mu- \ntual silences, alternation silences, and pauses in isolation.\n\nThere are some individual speaker exceptions to these observations. \nFor example, male No. 14 talkspurt averages are 1.683, 1.620, and \n1.759 seconds for \u201445, \u201440, and \u201435 dBm thresholds, respectively. \nTwo other male speakers exhibit such a reversal for talkspurts. In \ngeneral, however, conclusions drawn from the gross data are true of \nmost speakers or conversations.\n\n(2) The distribution functions of events resulting from periods of \ntalking seem in general more strongly affected by threshold shifts \nthan those resulting from silences. Compare, for example, talkspurts \nvs pauses, or solitary talkspurts vs pauses in isolation. The mutual \nsilence distribution, however, seems strongly influenced by threshold \nchanges.\n\n(iv) For all events, the number of times they occur (7) is notably \ninfluenced by the threshold. This is particularly true of pauses in \nisolation, whose distribution remains virtually unaffected while n \nchanges from 1890 to 3322 for a 10 dB threshold shift.\n\n(v) As the threshold is raised, the number of talkspurts tends to \nincrease. This trend will obviously be reversed if the threshold be- \ncomes so high that only a few spurts of energy clear it. But for low \nthresholds, as threshold is raised, long talkspurts are apparently being \nbroken up into shorter segments at a faster rate than that of low \nlevel talkspurts being left below the threshold.\n\n(vt) For any particular threshold, the cumulative distributions of \nspeech before and after interruption are practically identical, as seen \nfrom a comparison of Figs. 10 and 11.\n\n(vit) Interruptions tend to be much shorter than solitary talk- \nspurts, as would be expected because the interrupter might merely \nbe trying to get attention rather than make a statement. Also, some \ninterruptions are really not deliberate interruptions but rather ac- \nknowledgments, such as \u201cuh huh\u201d and \u201cum.\u201d This effect may be seen, \nfor example, at the \u201440 dBm threshold for which 17 percent of the \ninterruptions are less than 100 msec long, while only 9.5 percent of \nsolitary talkspurts are less than 100 msec.\n\n(vu) Many speech detectors operate with a hangover, rather than \na fill-in, to bridge short gaps. By shifting the talkspurt distribution \n200 msec to the right and the pause distribution 200 msec left, one\n\ncan determine the distributions for these events which would have \nresulted if a 200-msec hangover were used instead of fill-in. However, \nthe \u201cinteraction event\u201d (double talking, etc.) distributions will be \nchanged in a manner which cannot be determined from our present \ndata.\n\nTable VI shows that when male and female speech is compared at \nthe same threshold, four events show a statistically significant dif- \nference:* talkspurt, pause in isolation, solitary talkspurt, and speech \nbefore interruption. With the exception of pause in isolation, which \nis significant only at \u201445 dBm threshold, these are events resulting \nfrom talking rather than silence.\n\nSome of the apparent difference in male and female speech may \nresult from a difference in average levels. The average speech level \nfor the females was 5.94 dB below the average male speech level \n(measured in apl). When male speech at \u201440 dBm threshold is com- \npared with female speech at \u201445 dBm, and when male speech at \u201435 \ndBm is compared with female speech at \u201440 dBm, thus roughly \ncompensating for the average 6 dB level difference, the significant \ndifferences previously observed tend to disappear. New events\u2014pause \nin isolation, and possibly mutual silence and speech before interrup- \ntion\u2014become significant. It thus appears not possible to completely \neradicate differences in male and female speech with a simple level \nadjustment, although a level difference does account for differences \nobserved in certain events.\n\nThese conclusions are of particular interest in view of a recent \nstudy by Krauss and Bricker,?> who made measurements of verbal \ninteraction (measured from transcripts of the conversations) when \npairs of men and pairs of women talked over a circuit containing \nvoice-operated, fixed threshold devices. The verbal behavior of the \ntwo sexes was significantly different in certain tasks. One wonders \nif the devices operated differently on the male and female speech, \nas they did in the present study. This could be a contributing factor \nin bringing about the behavioral difference reported by Krauss and \nBricker.\n\nWe hope that the publication of these data will encourage other \nresearchers to make further observations leading toward a general\n\nmodel of the speech patterns occurring in conversations. We also hope \nthat by emphasizing the events surrounding double talking and other \nspeaker interaction, it may be possible to draw conclusions regarding \ndifficulties in conversing on certain circuits that have voice-operated \ndevices.\n\nI am especially grateful to Mrs. Lynn Evans, who cheerfully spent \nconsiderable time making hand tallies and desk-calculator analyses \nof the computer printouts, and to C. J. Gspann, who set up the inter- \nface between the speech detector and computer.\n\n1. Brady, P. T. and Helder, G. K., Echo Suppressor Design in Telephone Com- \nmunications, B.S.TWJ., 12, November 1963, pp. 2893-2917.\n\n3. Jafie, J.. Cassotta, L., and Feldstein, S., Markovian Model of Time Patterns \nof Speech, Science, 1d, May 15, 1964, pp. 884-886.\n\n4. Brady, P. T., Queueing and Interference among Messages in a Communica- \ntion System with Transmission Delay, Ph.D. Thesis, NYU Department of \nElectrical Engineering, June 1966.\n\n5. Brady, P. T., A Technique for Investigating On-Off Patterns of Speech, \nBS.T.J., 44, January 1965, pp. 1-22. |\n\n6. Clemency, W. F. and Goodale, W. D., Jr., Functional Design of a Voice- \nSwitched Speakerphone, B.S.T.J., 40, May 1961, pp. 649-668.\n\n7. Brady, P. T., A Statistical Basis for Objective Measurement of Speech \nree. \"BST J. , 44, September 1965, pp. 1453-1486.\n\nDelay on the Efficiency of Verbal Communication, J. Acoust. Soc. Amer., \n41 No. 2, September 1966, pp. 286-292.\n\nThis study examines the effects of loading or covering a phased array \nwith dielectric materials. It studies in detail the effect of dielectric geometry, \ndielectric constant, and sheath thickness on the wide angle array performance \nof an array of rectangular waveguides in the H and quasi-E plane modes \nof scan. We obtain numerical solutions of the integral equations describing \nan array covered with thick dielectric material. The resulis show that we \ncan obtain a match over a wide scan angle for the array by appropriate \nuse of dielectric geometries, and we discuss the advantages and disadvantages \nof several geometries.\n\nThe advent of swift aircraft, missile warfare, and the need for \nmodern radar to accomplish multifunction detection has given im- \npetus for a considerable amount of research into phased-array anten- \nnas. Such arrays consist of a large group of small radiators in a grid, \nfrequently a rectangular grid, and, most important, correlated in phase \nand amplitude. The radiated beam can be steered by an electronically- \nvariable linear taper of the phase correlation among elements. (See \nFig. 1).\n\nConsiderable knowledge of the behavior and problems of such ar- \nrays has been obtained in recent years by experimental and theoretical \nstudy of phased linear and parallel plate arrays. For example, it is \nwell known that the coupling coefficients between any single excited \nelement in the array and any terminated inactive element is uniquely \ndetermined by the inverse Fourier series transform of the reflection \ncoefficient as a function of scan angle determined when all elements \nare excited.1 Hence, by studying the array behavior for all possible \nlinear tapers of phase, we can determine the behavior of the array \nfor any phase or amplitude distribution among the elements.\n\nFig. 1\u2014JInfinite array geometry. Top: H-plane scanning in X-Z plane, \nBottom: Quasi-E plane scanning in Y-Z plane.\n\nOf fundamental importance in designing such arrays is a knowledge, \nand control, of the mutual coupling (the coupling coefficients) be- \ntween elements in an array. For arrays which scan over wide angles, \nthis coupling very seriously affects the array, so substantial effort has \nbeen made to understand mutual coupling. Because the arrays of in- \nterest are very large and consist of very many elements, theoretical \nstudies have generally assumed the arrays to be infinite in extent. \nThe usefullness of this approximation for elements located near the \ncenter of a large array has been verified, and in fact, the approxima- \ntion is frequently valid to within several elements from the edge.\n\nBecause the arrays are generally very large, the coupling between \ngreatly separated elements, the asymptotic coupling, is also of interest \nand has been theoretically studied.\u201d ? In general, planar phased ar- \nrays with terminated elements behave like lossy surfaces and have \nan asymptotic 1/r? decay of coupling between elements separated by \nthe distance r along the array surface.\n\nAmong other fundamental and interesting early developments 1s \nthat the transmission coefficient of a phased array with all elements \nexcited, as a function of scan angle, is directly related and proportional \nto the radiation pattern of a single-excited element in the array as a \nfunction of far-field observation angle.t As we mentioned, the two \nphysical situations (all the elements excited with a linear phase taper \nand only one element excited) are uniquely related. Hence, an ana- \nlytical formulation for only one or the other situation is necessary. \nAlthough most work has concentrated on the linear phase taper case, \nwith the consequent application of periodicity conditions and Floquet\u2019s \ntheorem,! *\u00ae some work has proceeded by directly attacking the \ncase with only a single element excited.\u201d\n\nOne of the most advantageous approaches to phased array problems \nhas been through the use of high-speed computers and numerical solu- \ntions of the appropriate integral equations.\u2019 We use this approach \nin this study, which attempts to discover some problems and solu- \ntions associated with covering phased arrays with radomes. The \nprincipal problem is, of course, to maintain a good impedance match \nto the array over a wide scan angle when the phased array radome is \nincluded in the design. Now many antennas have radomes covering \ntheir moving mechanical parts and their interior electrical components. \nA planar phased array can be so protected by covering the array with \na dielectric sheath or by loading it with a dielectric material. Hence \nour study concentrates on this type of cover. |\n\nWhereas ordinary radomes usually are designed to have the least \neffect on the antennas they cover, phased array covers often can be \nmade to very substantially improve the wide angle scan performance \nof the array. In fact, Magill and Wheeler recently have shown that a \ndielectric sheath cover can greatly improve the wide angle match of \nthe array.2 However, their analysis was a transmission line analysis \nin the sense that it did not take into account the interaction of the \nevanescent modes, generated at the array interface, with the dielectric \nsheath.\n\nMore recently, Lee\u00ae has made an analysis restricted to an array of \nthin-walled parallel plates, wherein the interaction of a limited num- \nber of evanescent modes with the dielectric sheath is taken into ac- \ncount. His results bear out the possibility of improving the array \nmatch with a dielectric sheath.\n\nBy using a somewhat different and more powerful analytical ap- \nproach, wherein the integral equations describing the array with a\n\ndielectric covering are solved by an accurate numerical technique \n(basically Galerkin\u2019s method*\u2019), we may obtain a solution with few \nrestrictions. The interaction of virtually all the modes at the array \ninterface with one or more dielectric sheaths is accounted for with \nvery little more difficulty than that entailed in the solution for the \nuncovered array. We made an extensive study of the rectangular ar- \nray shown in Fig. 1 by this method for two modes of scan, the quasi-ld \nand H planes of scan.\u2019 We use the term \u2018\u2018quasi-E\u201d\u2019 because the ad- \njacent columns of elements in the bottom part of Fig. 1 are out of \nphase by 180\u00b0.* We assume that the waveguides in each case are \nexcited in the dominant mode and we compute the parameter of par- \nticular interest, the reflection coefficient R of this mode, from the \naperture field determined by the integral equations.\n\nWe divided the complete study into two parts. The first, considered \nin this paper, analyzes the effects of loading the waveguide with di- \nelectric or covering the array with a very thick sheath where only one \ngrating lobe, at most, is present in the sheath. Generally speaking, a \nthick sheath is used with only lower dielectric constant materials be- \ncause with higher dielectric constants in a thick sheath a great and \nvery frequency-sensitive mismatch arises. Furthermore, the presence \nof two grating lobes in the sheath gives rise to surface wave phe- \nnomenon. This is the subject of the second part of our study, which \nwe have relegated to another part.\u2018 In that paper we plan to deal \nwith thinner sheaths, multiple sheaths, and some anomolous surface \nwave effects that we have observed in arrays with dielectric covers.\n\nFig. 2 shows the three dielectric geometries that we analyzed. In \neach case we assume a moderate fixed waveguide wall thickness\u2019 to \nexist in the plane of scan only. The top figure illustrates the \u201cloaded\u201d \narray with a symmetrical iris in the aperture (quasi-/ scan only, Fig. 1). \nFig. 2 also shows two other thick sheath covers. By \u2018\u201c\u2018thick\u2019\u2019 we mean \nthat there is very little interaction between the evanescent modes \ngenerated at the aperture plane (\u00a2 = 0) and the second dielectric \nboundary removed from the array interface (\u00a2 = -+d,). We may test \nthe validity of this assumption by estimating the relative amplitudes \nof the evanescent to propagating modes at z = +d, when the second \nboundary is not present. We made such a validity check with most\n\n* The waveguides are excited in this manner to reduce to a more easily \nnumerically tractable one-dimensional integral equation the two-dimensional \nintegral equations which result in the usual E-plane scan.\n\nFig. 2\u2014 Dielectric sheath geometry. Top: Dielectric loading. Middle: Sheath \ninside guides. Bottom: Sheath outside guides.\n\nof the thick sheath results. Notice that with this assumption the input \nimpedance or reflection coefficient is a periodic function of the distance \nae \nActually the integral equations that are solved require only a slight \nmodification to go from a \u201cthick\u201d? approximation to a \u201ctotal\u201d account- \ning of the interaction of higher order modes with the dielectric interface \nremoved from the aperture by d, . The actual equations solved for the \ndielectric loaded case take the forms:\u201d\n\n2\u00a5oe,.(y) = / - p> Vcr Deng 2. ARON CON ACD A \n= a 1 \nfor the unknown tangential electric field in the aperture in the quasi-E \nplane scan case, and\n\nfor the unknown magnetic field over an entire cell (6 X d, see Fig. 1). \nThe \u00a2,(x) and e,,(y) are appropriate interior orthonormal waveguide\n\nThe wn(%) or un(y) are appropriate exterior orthonormal modes \npertinent to the periodic structure and are obtained from an applica- \ntion of the Floquet theorem:\n\nBy the laws of transmission of a plane wave through a plane dielectric \nboundary (Snell\u2019s law), the quantity (kb sin @) or (kd sin 6) is unchanged \nby the presence of the dielectric. Hence the interior and exterior modes \nare independent of the dielectric constant. By using the h S c limits \nin (1) we also allow for the presence of a thin metallic iris directly at \nthe aperture plane (see Fig. 1).\n\nThe incident electric field in (1) is given by e,,(y) exp (\u2014 7852) and \nthe incident magnetic field in (2) by \u00a2,(x) exp (\u2014j8%z), where the \ninterior modal propagation constants are given by\n\nNow the coefficients of the interior and exterior dyads in (1) and \n(2), the Y,, YZ, Z,, and Z/, take a form that is dependent on the \ndielectric sheath geometry. For the dielectric loaded case they become \nsimply the modal admittances and impedances:\n\nFor the dielectric sheath cases, with one or more sheaths, the Y,,, \nY/,, Z,, and Z/% become the modal admittances or impedances ap- \npropriately referred to the aperture plane (2 = 0). These modal admit- \ntances or impedances are obtainable by the usual transmission line equa- \ntions. For example, suppose a single dielectric sheath inside the guides \nis considered (middle of Fig. 2) in the quasi-E plane scan case. Define\n\n, = 6. in the empty portion of the guide. (9) \nY,, = admittance in the dielectric region. \nY, = admittance in the air region. \nThen the coefficients Y,, become\n\nwhile the exterior Y/ coefficients remain unchanged (unless an exterior \nsheath is simultaneously included). The free term on the left becomes, \nif we postulate the same incident field as earlier,\n\nSolving equations (1) and (2) by the Galerkin\u2019\u00ae (or Ritz) method \nmeans that (1) and (2) are approximated in an NV-dimensional subspace* \nof the complete Hilbert space.\u2019 One way of testing the accuracy of this \napproach is to choose two very dissimilar subspaces for approximation\n\n* Approximation in an N-dimensional subspace leads to a set of N linear \nequations to be solved by well-known matrix inversion methods.\n\nand then compare the attained results. One subspace choice was that \nspanned by a set of N equally-spaced pulses\u2019 (that is, we sample the \nfield at N points along the z or y axis). In this case, R is determined by \naveraging the coefficients of the pulses. Other bases used were the first \nN modes, e,,(y) and y,,(z). In the case of e,, , R is determined directly \nfrom only the coefficient of the e,,(y) term.\n\nWe made a number of additional checks on the solutions. We observed \nthe convergence of the solutions with increasing NV, and we verified the \nconservation of energy between incident, transmitted, and reflected \nwaves. For certain angles of incidence we compared the results with \nthose obtained previously by Marcuvitz and Lewin.\u2019*\"*\u201d We also checked \nsome of the results against values for R obtained experimentally. \n(See Ref. 5 for example.) We checked the thin dielectric sheath results \n(using the exact formula (10) for example) against thick dielectric \nsheath results (using the approximate formula (8) and subsequent ap- \nplication of the transmission line equations to the dominant mode).\n\nIn the thick sheath numerical results which follow, we restrict the \nresults to include only the cases wherein at most a single propagating \nmode exists in any region where relative \u00a2 is greater than one.\n\nWe first consider the dielectric-loaded array (top of Fig. 2). In \nreality this may be viewed as an infinitely thick sheath with only one \ndielectric boundary interacting with the array interface (2 = 0). The \nphase of #, the reflection coefficient, and the amplitude of R are \nplotted as a function of scan angle (kb sin @ for the H-plane, and kd \nsin 6 for the quasi-E plane), with e as a parameter. A moderate, but \nfixed, guide wall thickness in the plane of scan is assumed in all the \ndata.\n\nFig. 3 gives some typical results for the H-plane scan direction \nwith the waveguides loaded with e = 0.9 to \u00ab = 3.0. The change in R \nwith \u00ab between curves is smooth. Between \u00ab = 0.9 and e\u00ab = 1.1, how- \never, the change in | R | is great. This may be attributed to the fact \nthat cutoff of the dominant waveguide mode occurs at \u00ab = 0.872.\n\nWe notice that for e ~ 1.3, the angular response is nearly flat, both \nin amplitude and phase. In fact, for all wavelengths examined there \nappears to be at least one value of \u00ab for which a nearly flat angular \nresponse for #& is obtained. It should be noted that even if the magni-\n\ntude of & were large, the flatness of the response in both amplitude \nand phase permits matching the array for all angles in the region of \nflat response, at least at one frequency.\n\nThe discontinuity in slope of these curves at kb sin 6 = 2r(1\u2014)A/b) \ncoincides with the onset of a grating lobe at that angle. Notice that \nthe singularity in the derivative of the modulus of R( | R | ) lies on the \nright side of grating lobe incipience, but on the left side for the phase \nof R. This is the same as that found for thin walls\u2019 * and it is plausible \nthat this will lead to the same asymptotic coupling.\n\nThe table in Fig. 3 shows the self-reflection coefficient Co and the \ncoupling to the adjacent element C; when a single waveguide element \nis excited and the others merely terminated with a perfect match. \nThe adjacent element coupling is found to be an order of magnitude \nsmaller for the H-plane than for the quasi-E plane. For e = 3.0, some \nhigher-order coupling coefficients for the H-plane case of Fig. 3 are:\n\nFig. 3 also shows the transmission phase and amplitude curves for \n\u00ab = 3.0. These curves are in fact the far field patterns when a single \nwaveguide element is excited? (except that the maximum value for 6 \nis somewhat less than 90\u00b0 since 6/A > 4). The very flat phase curve \nis an indication that the phase center for the singly excited element \nlies in the aperture plane.\n\nFor a wavelength further removed from the waveguide cutoff length, \nthe variation of R between curves of constant e is considerably re- \nduced. Fig. 4 gives some typical results for b/A = 0.400. (The dielec- \ntric loading here permits an element spacing of less than 2/2.)\n\nFig. 5 gives typical quasi-E plane results, where R(6), | Co |, and \n| C1 | are shown as a function of \u00ab from e = 0.8 toe = 1.6. This range \nof e\u00ab generally depicted all the important characteristics observed. \nFurthermore, a slightly greater value of \u00ab than e = 1.6 would cause the \nwaveguides to multimode (more than one mode propagates). Notice\n\nFig. 4\u2014 Dielectric loading inside waveguides for H-plane scanning (element \nspacing < A/2).a = 0.937b = 0.3748), b = 0.400X.\n\nFig. 5\u2014 Dielectric loading inside waveguides, for quasi-E plane scanning. \na=b=d= 0.5714), h = c = 0.937d = 0.5354).\n\nin the table of coupling coefficients that the adjacent element coupling \nmagnitudes are an order of magnitude greater than those for the H- \nplane scan case, independent of dielectric constant. This is also true \nfor higher coupling coefficients for the case depicted in Fig. 5, as shown \nbelow:\n\nThis behavior is attributable to both element spacing and polarization. \n(By making b < }/2, and with an appropriate dielectric loading, \nH-plane results very similar to these quasi-E plane results are obtainable. \nFor example; see Fig. 4. The \\ we refer to here is that which is appro- \npriate at the aperture for z > 0.)\n\nThe curves of | & | in Fig. 5 show that total reflection occurs beyond \na critical angle (~100\u00b0). This occurs because the element spacing d \nis less than \\/[4 \u2014 (d/b)?]?. Again, the infinite slopes for | R | and for \nthe phase of R, which occur at this critical angle, exhibit the same \nbehavior found in the thin wall analysis\u2019\u2019\u201d when e = 1. Hence the same \nasymptotic behavior of coupling, exp (\u2014jkr)/r?, may be expected for \nthick walled dielectric loaded arrays. (Here r is the distance between \nthe excited and coupled element.)\n\nA point of special interest in connection with these curves is the ap- \npearance of a resonance that occurs near the critical angle. A sharp \ndip in | & | occurs precisely at the same angle for which the slope of \nthe phase of R curves has a maximum. Although the sharpness of the \nresonance increases gradually with e\u00a2, it is interesting that there is no \nresonance for e < 1.0.\n\nFig. 6 illustrates the transmission phase and amplitude, or equiva- \nlently, the far field pattern of a singly excited element.\n\nIn Fig. 7 we illustrate the effect of a capacitive iris loading (that \nresults when h < c) together with dielectric loading. In this case we\n\nTig. 6 \u2014 Transmission coefficient for dielectric loading inside waveguides, with \nquasi-E plane scanning. a = b = d = 0.57144, h = c = 0.937d = 0.53542.\n\nFig. 7\u2014Dielectric loaded waveguides and iris loaded apertures, for E-plane \nscans, a=b= ,d = 0.400A, h = 0.6d = 0.240 and h = 0.937d = 0.3748), \n= 0.937d (h = 0. 937d => fully open aperture; h < 0.937d => iris at aperture).\n\nhave let a = 6 > o so that a true E-plane scan in a parallel plate ar- \nray is considered. The effect of the iris tends to flatten the phase and \namplitude responses, particularly the phase response, at the expense \nof a somewhat greater average | Ff |. The larger average | R | can, how- \never, be uniformly reduced for all scan angles in a region of flat R(@) \nresponse.\n\nThe solutions of the integral equations (1) and (2) are actually \ncomplete solutions of the boundary value problem. The fields as well \nas the scattering matrix are determined. The variation of E, in the\n\naperture as a function of scan angle (kd sin @) is sharper when the \nwaveguides are loaded. Particularly interesting is the field change near \nthe critical angle kd sin 6 = d/[4 \u2014 (/b) *}?, which has the value 99.57\u00b0 \nfor the results shown in Fig. 8. Since the relevant eigenvalue equation \n(See p. 157 of Ref. 4.) for [#,(6) + H,(\u20146)] has a Hermitian kernel* \nfor |kd sin @| > 99.57\u00b0, the phase of [#,(6) + E,(\u20146)] should be \nconstant in this region. By observing the phase of the approximate field\n\n*Since no power is radiated for | kd sin 6 | > X / [4 \u2014 (A/b)2]4 when the \nphasing is directed in the +6 and \u2014\u00e9 directions simultaneously, the phased \narray behaves lke a closed system, a cavity, and it is easy to show that a \nHermitian kernel results. It is well known, then, that the eigenfunctions (the \nfield solution here) of such a kernel have no varying phase.\n\nsolution at angles greater than the critical angle, we may obtain some \nevaluation of the errors in this solution. For most angles the errors are \nsmall. However, it is evident that errors in phase do occur for @ near \nthe critical angle and also for y near the singularity obtained for | E, |. \nNevertheless, since the computation of R is an averaged quantity over \nthe range of y, these errors do not greatly affect the values obtained \nfor R.\n\nWhen we add another dielectric boundary inside the waveguides, \nthat is, when we place a dielectric sheath inside the guides, the results \nobtained for R are substantially different. This is true despite the fact \nthat we will consider only thick sheaths, in the sense described earlier. \nThe large change in R(@) behavior occurs because, even when a thick \nsheath is assumed, the second dielectric boundary is accounted for by a \nbilinear transmission line transformation which changes the input varia- \ntion of R with 6. (A linear transformation would leave R(6) functionally \nunchanged.)\n\nIn the following we will keep the dielectric constant fixed and plot \nR(@) versus kb sin @ with d,, the sheath thickness, as a parameter. \nThe phase of # will be referred to the aperture plane, z = 0, although \nR is the reflection coefficient for, or into, the region zg < \u2014d,. The \nchoice of \u00a2 in any given figure was made so that the illustrated results \nwere typical of a wider range of \u00ab. With the thick sheath approximation, \nthe results will repeat every half guide wavelength, so that d, is varied \nover only one half wavelength. The minimum d, for which the thick \napproximation is valid is determined by the relative decay of the first \nevanescent mode in the distance d, . This decay factor, df, is presented \nwith each curve. It is found, generally, that df \u2018<= 0.1 is sufficient for \nthe thick sheath approximation to be valid. This result is usually \nsatisfied for some d, = \\,/2. Of course, by adding a sufficient number \nof multiples of a half guide wavelength to d, , the results must become \nvalid to any accuracy desired.\n\nIn Fig. 9 we have illustrated some typical results with e = 2.0. For \nany given e we have found that there exists a thickness, d,, for which \nboth the amplitude and phase of the reflection coefficient is flat over \nthe generally useful region of scan angle (region in which only one \nlobe radiates). However, as the dielectric constant 1s increased, the\n\nfrequency sensitivity of the angular response is increased. This is \nmanifested, basically, by an increased spread between the curves \nshown. Furthermore, since the nearly flat curve also is found to have \nthe maximum amplitude of R (d, = 0.618b in Fig. 9), the necessity of \nmatching out a larger | R |, when \u00ab is greater, further aggravates the \nfrequency sensitivity problem that occurs with increasing e\u00ab. The re- \nsults in Fig. 10, when compared with those in Fig. 9, show the effect \nof increasing e.\n\nQualitatively similar results are obtained in the quasi-E plane scan \ncase as illustrated in Figs. 11 and 12. The increasing dielectric con- \nstant, illustrated by comparing the results in Figs. 11 and 12 (e = 1.2 \nand \u00ab = 1.6, respectively), causes a greater spread between curves and, \nconsequently, a greater frequency sensitivity.\n\nIn addition, we again notice a sharp resonance here (increasingly \nsharper with greater \u00ab), just as we noticed when only a single dielectric \nboundary was present. The primary difference between these results \nand those for a single dielectric boundary is that we may keep the \ndielectric constant fixed here and vary the thickness, d,, to obtain a \nflat response. This is done, however, with the cost of increased fre- \nquency sensitivity.\n\nWe notice that very similar results are obtained in the quasi-E \nplane scan independent of whether the second dielectric boundary is \nplaced inside or outside the waveguides; that is, independent of \nwhether there is a dielectric sheath inside or outside the waveguides. \nFig. 13 illustrates a typical result. In this figure the value of \u00ab is \nheld fixed while the sheath thickness (see the bottom figure of Fig. 2) \nis varied from curve to curve.\n\nboundary, at 2 = +d,, and the array face is accounted for, with all \nthe evanescent modes, then the results shown by the circles and tri- \nangles are obtained. The results agree very well with those results \nin which higher order mode interaction with the second dielectric \nboundary is neglected, that is, the sheath satisfies the earlier specified \nthickness criteria.\n\nFig. 14 shows a useful way of estimating what value of d, will be \nproperly \u201cthick\u201d. Notice first that the rate of decay of higher order \nmodes away from the array face is a function of scan angle in the \nexterior sheath case. Hence, a single value cannot be used as a decay \nfactor for all 6. In Fig. 14, however, the actual ratios of the first and \nsecond evanescent mode amplitudes to the propagating modes is\n\nFig. 11\u2014 Dielectric sheath inside waveguides, for quasi-E plane scanning. \ne-\u201412,\u00a2,=\u201410,@a=-b=d = 0.5714, \u00a2c = h = 0.987d = 0.5354).\n\nFig. 12\u2014Dielectric sheath inside waveguides, for quasi-E plane scanning. \ne=16,\u00a2 = 10,a = b= d = 0.5714), c = h = 0.987d = 0.5354).\n\nplotted versus kd sin 6. Again we see that if RD1 and RD2 (see Fig. \n14 for definitions) are less than about 0.1, then the parientar d, is \nthick.\n\nA \u201cthick\u201d d, means that, for a given scan angle, the results for R (6) \nwill repeat periodically so that\n\nfor alln > 0. The question remains whether \\,, = 27/82\" (see equation \n(7)) varies rapidly with kd sin 6. This may be answered by examining the \ngrating lobe structures in the dielectric sheath as compared with that \nin free space, as shown in the inset in Fig. 13. This structure is obtained\n\nFig. 13\u2014 Dielectric sheath outside waveguides, for quasi-E plane scanning. \ne\u2014 1818,a =b=d = 0.5596), c = h = 0.9387d = 0.52432.\n\n(Actually (12) defines the intersections shown in the Fig. 13 inset. \nThe total grating lobe structure requires setting the two-dimensional \nz-directed propagation constants to zero.) Now the z-directed wave- \nlength in the sheath is given by\n\nwhich is a slowly-varying function of 6 in that part of the grating lobe \nstructure bounded by the dashed lines in the Fig. 13 inset. Only this \nregion is useful, because the propagating wave in the dielectric is \ntotally reflected at (kd sin 0)\u201d = k* \u2014 (x/b)\u2019(y,> 90.6\u00b0). Hence the \npole in \\,, doesn\u2019t affect the variation of \\,,, and A,. varies very little \nwith 6, from y, = 0 to \u00a5? = (k* \u2014 (x/b)\u2019), providing that\n\nSE = 3, (13) \nve- &) \n2b \nWe should mention that the quasi-E plane results for the sheath \ninside and outside the waveguides are similar, primarily because of \nthe element spacing that causes total reflection to occur at the defined \ncritical angle. When the element spacing is changed, very markedly \ndifferent results can be obtained. The results depicted in Fig. 15 for \nthe H-plane are typical of this. \nIn examining the grating lobe diagram in the inset of Fig. 15 we \nnotice that, in the shaded region, two waves propagate in the dielectric\n\nsheath whereas only one wave propagates in free space. This is a \npotentially very useful operating region for the phased array because\n\nRD = (AMPLITUDE OF Ist EVANESCENT MODE AT Z=ds \nne (: AMPLITUDE OF PROPAGATING MODE\n\na dg\u2014=0224d \n0.1 \n0.447 d \n0.671 d \nO beeas 1.006 d \n0.3 \nAe AMPLITUDE OF 2ND EVANESCENT MODE AT zat) \nGi. =( AMPLITUDE OF PROPAGATING MODE \nra) \naa \n0.! \n0\n\nFig. 14\u2014 Validity check of transmission line approximations, for quasi-E \nplane scanning with a dielectric sheath outside waveguides. e = 1.818, a = b = \nd = 0.5596, c = h = 0.937d = 0.5243.\n\n\u201d) \nuJ \no: \n\u00a9) GRATING \nwl STRUCTURE \nIN FREE SPACE \n= GRATING | \nac STRUCTURE \nIN SHEATH \nWw \noO \nWw \nY) \n< \nL \nao.\n\nonly one beam will be radiated. However, we find that interference \nbetween the two propagating waves in the dielectric causes some very \ninteresting anomolous results associated with what might be described \nas unattenuated surface waves. This effect 1s markedly different from \nthat which may occur for a sheath inside the guides (although as a \nfunction of frequency or sheath thickness, as opposed to scan angle, \nsimilar results may occur). Further discussion on this subject is \ndeferred to another paper.\n\nThe results in Fig. 15 for the | & | and the phase of R are shown in \nthe scan angle region in which only one wave, at most, propagates in \nthe sheath. In this angular region the results are also very different \nthan those for the interior sheath in that the response is comparatively \nflat for a wide range of sheath thicknesses.\n\nIn summary we may state that dielectric loading or covering has a \nvery substantial effect on the array performance to the extent that an\n\narray and dielectric cover should be designed as an integral unit in- \nstead of being designed for minimal effect or match correction. Never- \ntheless, the additional parameters available in the design of an array \nwith a dielectric cover can be used to match the array over a wide \nangular and frequency region.\n\nNo particular method of loading or covering appears to be uni- \nversally superior except that when only one dielectric boundary is \npresent (Fig. 2, top ), the wide angle match appears to be less fre- \nquency sensitive than when a sheath or two boundaries are present. \nFurthermore, with thick sheaths, the maximum permissible value of \u00ab \nis small before serious matching problems occur (a large | R | together \nwith a flat R[6]). Finally, by placing the sheath inside the wave- \nguides instead of outside, certain anomolous reflection phenomena, \nassociated with the exterior sheath grating lobe structure and sur- \nface waves, can be avoided.\"\n\nAlthough we did not give analytic proof, the amenonl results do \nindicate that the asymptotic behavior of the coupling coefficients? is \nnot altered by the presence of the dielectric materials.\n\n1. Wu, C. P. and Galindo, V., Properties of a Phased Array of Rectangular \nWaveguides with Thin Walls, IEEE Trans. on Antennas and Propagation, \nAP-14, No. 2, 1966, pp. 163-172.\n\n2. Galindo, V. and Wu, C. P., Asymptotic Behavior of the Coupling Coef- \nficients for an Infinite Array of Thin-Walled Rectangular Waveguides, \nIEEE Trans. on Antennas and Propagation, AP-14, No. 2, March 1966, \npp. 248-9.\n\n3. Galindo, V. and Wu, C. P., On the Asymptotic Decay of Coupling for Infinite \nPhased Arrays, paper presented at International Union of Radio Science, \nWashington, D. C., April 1966. (To be published.)\n\n4, Galindo, V. and Wu, C. P., The Relation Between the Far-Zone Pattern \nof the Singly Excited Element and the Transmission Coefficient of the \nPrincipal Lobe in an Infinite Array, IEEE Trans. on Antennas and \nPropagation, AP-14, No. 3, March 1966, pp. 397-398.\n\n5. Galindo, V. and Wu, C. P., Numerical Solutions for an Infinite Phased \nArray of Rectangular Waveguides with Thick Walls, IEEE Trans. on \nAntennas and Propagation, AP-14, 1966, pp. 149-158.\n\n6. Galindo, V. and Wu, C. P., Integral Equations and Variational Expressions \nfor Arbitrary Scanning of Regular Infinite Arrays, IKEE Trans. on An- \ntennas and Propagation, AP-14, No. 3, May 1966, pp. 392-394.\n\n7. Galindo, V. and Wu, C. P., A Variation Expression for the Dominant Mode \nCoupling Coefficients Between The Elements in an Infinite Array, IEEE \nTrans. on Antennas and Propagation, AP-14, No. 5, September 1966, pp. \n637-639.\n\n8. Magill, E. G. and Wheeler, H. A., Wide Angle Impedance Matching of a \nPlanar Array Antenna by a Dielectric Sheet, IEEE Trans. on Antennas \nand Propagation, AP-14, 1966, pp. 49-53.\n\n9. Lee, S. W., Impedance Matching of an Infinite Phased Array by Dielectric \nSheets, IEEE Elec. Letters, 2, No. 10, October 1966, pp. 366-368.\n\n10. Kantorovich, L. V. and Krylov, .V. I., Approximate Methods of Higher \nAnalysts, Interscience Publishers, New York, 1958.\n\n11. Wu, C. P. and Galindo, V., Surface-Wave Effects on Dielectric Sheathed \nPhased Arrays of Rectangular Wave-Guides, B.S.T.J. 47, No. 1, January \n1968, pp. 117-142 (next article in this issue).\n\n12. Waveguide Handbook, N. Marcuvitz, ed., MIT Radiation Lab. Series, 10, \nMcGraw-Hill Book Company, Inc., New York, 1951.\n\nSurface-Wave Effects on Dielectric Sheathed \nPhased Arrays of Rectangular Waveguides\n\nA further study of the effects of dielectric slabs on the radiation char- \nacteristics of an infinite array of rectangular waveguides has been carried \nout. It is found that, in addition to causing substantial and sometimes \nbeneficial changes in the array performance, the presence of dielectric slabs \ncan give rise to sharp resonant peaks in the reflection coefficient at certain \nscan angles. T\u2019'he occurrence of such resonant peaks at which total reflection \noccurs 18 contingent upon the presence of space harmonics which have \nsurface-wavelike field distribution. Extensive data for both the H and E \nplanes of scan, when the array 1s covered with a single slab, have been obtained \nand are presented here. This paper discusses the influence of the dielectric \nconstant, slab thickness, and waveguide wall thickness on the resonant \npeak location, and points out the relationship between the resonance phe- \nnomenon and the surface wave propagation over a corrugated surface. It \nalso presents some further results for the thin sheath and their extension \nto multiple sheaths.\n\nWe presented the radiation properties of a dielectric-loaded rectan- \ngular waveguide array in some detail in a previous paper? for two \nplanes of scanning: the H plane and a quasi-E plane. (See Fig. 1.) \nWe discussed the effects of a thick dielectric sheath placed inside or \noutside the array aperture. We concluded that the presence of a \ndielectric material can cause a substantial change in the array per- \nformance, so that dielectric material should be considered an integral \npart of the array when it is designed. Moreover, we demonstrated \nthat the effects of a dielectric may be used to improve the match \nperformance of an array, perhaps at the cost of a larger frequency \nsensitivity, by a judicious choice of the added physical parameters.\n\non an extensive further analysis. We completely removed the restric- \ntions placed on the parameters in the previous work in order to analyze \nthe full effects of a dielectric medium. In particular, since dielectric. \nslabs can support surface waves, we direct special attention toward \ninvestigating the possibility of anomalous array behavior when dielec- \ntric slabs are used to cover the array. Indeed, a resonance phenomenon \nis found to exist due to the presence of trapped- (or surface-) wave \ntype space harmonics at the air-dielectric interface. One important \neffect of such space harmonics is to cause the appearance of sharp: \nresonant peaks in the reflection coefficient at certain scan angles. The \npeaks attain, for all practical purposes, values of unity. We present \nresults for a dielectric slab of arbitrary thickness placed over an array, \nthereby, removing the previous restriction to thick slabs. We also dis- \ncuss radiation through a stratified medium.\n\nThe approach we use in this work, as in previous ones,?? is based \non an integral equation having either the aperture electric or the \naperture magnetic field as the unknown function. One of the advan- \ntages of such an approach is that the integral equation may be easily \nand rigorously derived without any limitations from the physical \nparameters of the problems, and the approach is readily adaptable \nto a more general class of problems such as, for example, an array \ncovered by a stratified dielectric medium. Although the basic proce- \ndures for formulating the integral equation have been discussed else-\n\nwhere,\u201d 74 in order to facilitate a further discussion of the formula- \ntion and to make this paper more or less self-contained, we present \na brief derivation here and point out the modifications necessary \nfor an extension to a more general situation of stratified media.\n\nThe procedure in the integral equation method is first to expand \nthe fields into the appropriate normal modes in the various regions \nand then to match the boundary conditions across the interfaces. \nConsider, for example, an infinite array of parallel plates, covered \nwith a single dielectric slab and scanned in the H plane as shown \nin Fig. 2. It is convenient in this case to divide the space into three \nregions, the region inside the waveguides, the region inside the dielec- \ntric slab, and the free space region. However, as will become evident \nshortly, it turns out that the geometry of the problem is such that \nit suffices to partition the space into two regions, inside and outside \nthe waveguides.\n\nThe orthonormal modal functions and the modal impedances per- \ntinent to the waveguide region are the usual ones given by\n\nare the z-directed propagation constants for a waveguide filled with \na dielectric of dielectric constant e;.\n\nBy using the Floquet theorem, it can be easily shown that the \ndielectric slab and free space regions have identical modal functions, \nowing to the requirement that the tangential fields must be continuous \nat all points across the dielectric-free space interface. The normalized \nmodal functions take the usual form:\n\nYm(z) = Jie penny, m = 0, \u00a31, \nwhere JT, is the phase shift per unit length. The modal impedances\n\nfor these two regions are different, however. They are given respec- \ntively by\n\nWhen the fields in the various regions are expanded into the nor- \nmal modes under the situation in which the waveguides are excited \nin the fundamental mode of unit amplitude, we have\n\nA time convention of exp (\u2014jwt) is assumed and suppressed for brevity \nthroughout this paper. The J\u2019s are the unknown modal coefficients. \nWaves of all modes travelling in both the positive and negative z \ndirections are included in the fields for 0 S z S d,, because this re- \ngion is situated between two interfaces. Likewise, the fields for z S$ 0 \ncontain waves travelling in the negative z direction due to the scat- \ntering at the array aperture. \nWe find, by applying the boundary conditions at z = 0, that\n\nHence, on account of the orthonormality among the modal functions, \nit follows that\n\nBy observing that the right and left sides of (4) are the field expan- \nsions with respect to the same set of modal functions, one may 1m- \nmediately write\n\n[ieitmts + Toe\" 784 = I\u2019 \nAe aaa aes ad Ee \nEquations (5) show that the air-dielectric interface at 2 = d, is a \nsimple one in that the m\u2122 order mode in the dielectric region couples\n\nonly into the same order mode in the free space region and vice versa. \nThe implication of this result is that the fields at the air-dielectric \ninterface are completely determined when the fields at the array aper- \nture are obtained. Therefore, it is necessary only to solve for the \naperture field alone. With this in mind, we may then make use of \n(2), (8) and (5) to derive an integral equation having only the aper- \nture magnetic field as the unknown function. Thus,\n\nIt is clear from expression (7) that the equivalent impedances Z,/ \nfor the m\u201c order modes are the familiar input impedance of a trans- \nmission line, which has a characteristic impedance Z? , propagation \nconstant 6,,, and length d,, and is terminated in a load impedance \nZ,. Notice also that only quantities pertaining to the m\u2018 space \nharmonic appear in (7). All these facts suggest that the space exterior \nto the waveguides may be treated as a single region from the onset, \nas observed earlier, provided that the effects of the dielectric slab are \ntaken into account through the use of appropriate modal impedances. \nMoreover, the integral equation given by (6) is readily extended to a \nmore general situation in which the array is covered or loaded with \nstratified dielectric media (or both covered and loaded). Only the \nmodal impedances need to be modified for this purpose.\n\nAs an example, suppose we wish to study the properties of an array \ncovered by a stratified medium as Fig. 3 shows. Equation (6) still is a \nvalid integral equation to use. In this case, the Z, are the usual wave- \nguide modal impedances, whereas the Z,\u2019 are the modal impedances \nas seen at the array aperture of the N layer stratified dielectric medium, \nwhich may be calculated by standard techniques.\u201d\n\nThe integral equation appropriate to the quasi-E plane of scan may \nbe derived in a similar manner. In a quasi-E plane of scan, the scanning \ntakes place in the plane of the electric field while there is a sinusoidal \nfield variation in the direction normal to the plane of scanning.*\n\n*Such a mode of scanning results from a planar array of rectangular wave- \nguides, which is scanned in the E plane direction with a fixed scan angle of 180\u00b0\n\napplied in the H plane direction. This special scan case is considered for the sake \nof simplifying the problem. See Ref. 4.\n\nThe exterior modal functions for the quasi-E plane scan are the same \nas those for H plane scan, according to the Floquet theorem, but the \nwave modes which the waveguides support are different. The orthonor- \nmal modal functions for this case are\n\nHere we have used c for the internal waveguide width and d for the \nelement spacing. A sinusoidal variation of sin [ (7/b)x], which applies \nto all tangential field components, is omitted for brevity.\n\nThe integral equation with the aperture electric field as the un- \nknown is given by\n\nare the interior modal admittances and the Y/ are the exterior modal \nadmittances with the presence of dielectric slab(s) appropriately taken \ninto account.\n\nThe method used in solving the integral equations is basically that \nof Galerkin,\u00ae * the method of moments. Briefly, in this method, the\n\nfirst step is to expand the unknown function as a linear combination \nof NV linearly independent functions. Substitution of the representa- \ntion into the original integral equation leads to an approximate equa- \ntion. The difference between the left and right sides of the approxi- \nmate equation is then required to be orthogonal to the set of functions \nindividually, thus yielding a set of N equations in N unknowns. The \nresulting set of equations may then be inverted with the help of an \nelectronic computer.\n\nThere are different sets of functions one may choose to use in this \napproach. Although the choice, aside from consideration of computer \ntime and convenience, seems to be largely a matter of personal pref- \nerence, it is desirable to incorporate as much prior knowledge about \nthe problem as possible. We have used the set of first N modal func- \ntions to obtain most of the results reported here. This step is tanta- \nmount to assuming the higher order modal coefficients to be zero. For \nexample, to solve (6), we set\n\nThe expression (9) is then required to be orthogonal to the set of func- \ntions y\u00a5(z), 1 = 0, +1, --- , 4M, individually, thus yielding\n\nAL ro) \nDe , LC anC#, + Zh, 5h = 27,C%, 1=0,41,---,4M. \nm=-\u2014AL n=1 \n(10) \nThe set of equations (10) may be solved by a standard technique. The \nreflection coefficient R is then obtainable from\n\nSince it appears that there is no convenient scheme for estimating \nthe error in the type of problem being considered here, the following \nprocedures have been used to ascertain the accuracy of the results. \nThey include:\n\n(t) Observing the \u201cconvergence\u201d of the solution by increasing JN, \nthe number of functions used to represent the unknown aper- \nture field.\n\n(iz) Using different sets of functions, such as the set of piecewise \nconstant or pulse functions for the approximation.\n\n(271) Comparing the results with known solutions obtained by dif- \nferent methods where applicable.* * &\n\n(iv) Applying the variational principle to check the adequacy of \nthe algorithm used in the numerical procedure.\n\nIn the previous report,! we placed the emphasis on the results for \nthe situations where the waveguides are either completely filled with \na dielectric material or loaded with a dielectric slab. We also dis- \ncussed some preliminary results for covering the array with a di- \nelectric sheath. We obtained that data for a range of parameters such \nthat, at most, one propagating mode could exist inside the dielectric \nregion, and only relatively thick slabs were considered. Such a choice \nof the parameters was necessary in order that an approximation based \non the transmission line theory could be applied.\n\nThe results presented here are concerned largely with the effects of \ndielectric sheaths covering the array. In particular, we concentrate \non the situation in which more than one wave can propagate inside \nthe dielectric region so that there will be trapped- (or surface-) wave- \nlike space harmonics at the air-dielectric interface. Although in prin- \nciple it is still possible in such cases to apply a generalized transmis- \nsion line approximation, this approach might not be very convenient \nin practice. The modification required for generalization depends on \nthe number of modes which can propagate inside the dielectric, and \nthis number is, in turn, dependent on the dielectric constant being \nused. Moreover, the minimum slab thickness necessary for a valid \napproximate calculation might sometimes become so large that it \nwould exclude a useful range of practical interest. Therefore, it is \ndesirable to proceed with the solution of the appropriate integral \nequations without introducing any intermediate steps. Thus, the ef- \nfects of the air-dielectric interface at 2 = d, on all the modes gen- \nerated at the array aperture may be fully taken into account. By \ndoing so, we are also able to obtain data for comparison with those \ncalculated by using the transmission line theory, and thus gain a\n\ngeneral feeling for the accuracy of this type of approximation. Some \nresults of such a comparison are presented in Ref. 1.\n\nThe effect of an ordinary dielectric (\u00ab > 1) on wave propagation is \nto slow down the phase velocity, or equivalently to shorten the wave- \nlength. Hence, when a dielectric slab 1s placed over a phased array, \ntwo apparent element spacings (in terms of the wavelength) have to \nbe considered: one inside the dielectric medium and the other in the \nfree space; the former always greater than the latter. A difference in \nthe element spacings as seen in the two regions results in different scan \nangles for the appearance or disappearnce of grating lobes in the re- \nspective regions. Consequently, when the fields are expanded into normal \nmodes according to the Floquet theorem, there will be a range of scan \nangles over which the number of propagating modes inside the dielectric \nis larger than that in free space. A mode is said to be propagating when \nthe corresponding z-directed propagation constant is real. In a linear \narray, this propagation constant is given by\n\nwith an appropriate e\u00ab for each region. It is easy to show that for m < \nb/d. < (m + 1/2), where m is an integer and X, is the wavelength of a \nplane wave in a medium with dielectric constant \u00a2\u00ab, the number of \npropagating modes will change from (2m-+1) forO S$ T,b < 2r(b/d,.\u2014m) \nto (2m) for 2r(b/A. \u2014 m) S T,b S +. On the other hand, when (m + 1/2) \n< b/d. < (m + 1), the number of propagating modes will increase by \none from (2m + 1) to (2m + 2), when the scan angle passes from \n0< T,b < 2x(m + 1 \u2014 DS.) to 2r(m +1 \u2014 B/D.) < T,b < o.\n\nThe wave modes which are propagating inside the dielectric and \nare evanescent in free space have the same field distribution as that \nof a surface wave. Such wave modes have profound effects on the \nradiation characteristics of a phased array as we will see in the ex- \namples. It is important to emphasize that a single wave mode of this \ntype alone is not sufficient to satisfy the boundary conditions. In \nother words, all the modes are required to constitute a correct solu-\n\n3.2 H-Plane Scan Results \nTigs. 4 and 5 give the reflection coefficients as a function of scan \nfor an infinite array of rectangular waveguides covered with a single\n\nFig. 4\u2014 Reflection coefficient R vs scan angle T.b for H plane scan with \na/\\ = b/d = 0.5714, and e = 3.0625.\n\ndielectric sheath and scanned in the H plane. The results are obtained \nfor the following parameters: b/\\ = a/\\ = 0.5714, e = 3.0625 with the \nthickness of the dielectric slab d, varied over one i, at an increment of \n\\./8, where \\, is the wavelength of a plane wave in the dielectric medium. \nNotice that the element spacing is measured in terms of the free space \nwavelength \\, for b/d is the quantity which determines the number \nof radiated beams at a certain scan angle. With the given arrangement, \nthe element spacing is such that the array radiates one beam for 0 S$ \nT,b5 S 2r(1 \u2014 b/d) = 154\u00b0, and two beams when 27(1 \u2014 b/d) S T,b S\n\n180\u00b0. The dielectric constant of the slab makes the apparent element \nspacing in the medium to be b/A, = 1. Thus, there are always two \npropagating modes inside the dielectric slab for all scan angles. As a \nresult, we have in 0 S T,b S$ 2r(1 \u2014 b/X) a mode which exhibits a \nsurface-wavelike behavior by being propagated inside the dielectric \nand evanescent in free space. The effect of such a mode is to cause \nthe appearance of sharp resonant peaks in the reflection coefficient at \ncertain scan angles as is evident from the graphs.\n\nBased on the results presented here and some further calculations \nat different wavelengths, we may make the following general observa- \ntions:\n\nFig. 5\u2014 Reflection coefficient R vs scan angle 7.b for H plane scan with \na/\\ = b/A = 0.5714, and e = 3.0625.\n\n(1) When the thickness of the dielectric slab is relatively small, no \nresonant peak occurs.\n\n(72) When the thickness is increased beyond a critical value, \nusually in the neighborhood of 3A,/16, a resonant peak starts \nto appear at a scan angle close to the value 27(1\u2014b/A), and \nthe peak is usually preceded by a dip.\n\n(17) Increasing the slab thickness causes the peak to move toward \nthe broadside direction and the peak becomes sharper.\n\n(iv) A further increase in the slab thickness makes more than one \npeak appear.\n\nThe dielectric constant used in obtaining the results for Figs. 4 and \n5 was chosen to have a value e = (A/b)*. This creates a situation where \none space harmonic possesses a surface-wavelike field distribution in \n0 < T,b S$ 27r(1 \u2014 D/)). It is possible for a resonant peak to appear \nat any scan angle within this range. If a smaller dielectric constant \nhad been used, the range of scanning over which a resonant peak might \nappear would be reduced accordingly. On the other hand, a larger value \nof dielectric constant would give rise to more than one surface-wavelike \nspace harmonic, and it is possible for resonant peaks to appear with \ndielectric slab of thickness even smaller than 3),/16.\n\nRecall that when we use a dielectric slab of small dielectric constant \nand large slab thickness, the transmission line approximation may be \neffectively applied to yield useful results.t The reflection coefficients \ncalculated under such conditions are periodic functions of the slab \nthickness; hence, the calculations only need be performed over a \nperiod, that is, a half guided wavelength. If the dielectric constant is \nlarge so as to permit more than one propagating mode inside the di- \nelectric, however, the periodic property 1s no longer present, and \nseparate calculations have to be carried out for different slab thick- \nnesses,\n\nFig. 6 is typical of the transmission coefficients calculated for b/A \n= a/r = 0.5714, \u00ab = 3.0625 and d; = ,/2. These transmission coef- \nficients are referred to the air-dielectric interface z = d,. The figure \nshows both the transmission coefficients for the zero order mode \n(corresponding to the main beam) and the first space harmonic (cor- \nresponding to a grating lobe). These coefficients are defined so that \nthe sum of the reflection coefficient squared and the square of the \ntransmission coefficient(s) equals one. Thus, if 7\u20199 and 7\u2019; denote the\n\nFig. 6 \u2014 Transmission coefficients 7\u2019 vs scan angle T.b for H plane scan with \na/\\ = b/d\u2019 = 0.5714, e = 3.0625, d, = 0.5r..\n\ntransmission coefficients of the main beam and the grating lobe re- \nspectively, then\n\nClearly, 7\u2019, has significance as a transmission coefficient only in the \nscan range 27(1 \u2014 D/A) S$ T,b So.\n\nThe graph of | 7 | shows a sharp dip at the scan angle of 7, = 70\u00b0, \nat which the reflection coefficient attains its peak value of one. Notice \nthat the phase of 7\u2019) exhibits a discontinuity of 180\u00b0 at this scan angle. \nThis is so because both the real and imaginary parts of 7) go through \nzero and then change sign as the beam is scanned past this scan angle. \nThis appears to indicate that | 7) | does go to zero rather than approach \nzero, or equivalently, that | R | actually attains the value one. The \nsignificance of the difference between | R | approaching one and | R | \nactually attaining one lies in whether the match of an array can be \nimproved by network compensation. Moreover, there is an intimate \nconnection between the fact that | R | = 1 and the surface wave propaga- \ntion along a plane corrugated structure. We discuss this in detail in \nSection IV).\n\nAnother point which might be of interest is that 7 is the radiation \npattern in the angular range 0 S @ S sin\u2122*(\\/2b), when a single ele- \nment is excited with the rest of the elements terminated in the char- \nacteristic impedances.\u201d The radiation pattern for the remaining angular \nrange sin-'(\\/2b) < 6 < 2/2 may be obtained from the curve of 7; \nin the scan range 27(1 \u2014 b/d) S T,b S 7, reflected with respect to \nthe T.b = a7 axis.\n\nThe incident wave in the E plane scan is a TEM wave with the \nelectric field polarized in the direction normal to the waveguide walls. \nFigs. 7 and 8 give the results for both the amplitudes and phases of \nthe reflection coefficients as a function of scan. The set of parameters \nused for obtaining these results is: d/A = 0.5714, c/d = 0.85. The ar- \nray is covered with a single layer of dielectric material flush with the \narray aperture. A dielectric constant of e = 3.0625 is chosen for the \nsame reason as for the H plane, namely to have one surface wavelike \nspace harmonic present over as wide a scan angle as possible. Rela- \ntively thick waveguide walls are used in this case in order that the \nwaveguides support only the dominant ITEM mode, because the ele- \nment spacing is larger than 4/2.\n\nFrom the numerical data we may observe that, in the absence of \na dielectric material, the reflection coefficient is relatively flat over a \nlarge range of scan angles, except in the vicinity of the grating lobe \nformation angle (in this case T,d = 154\u00b0), at which we find a sharp\n\nFig. 7\u2014 Reflection coefficient R vs scan angle Tyd for E plane scan with \nd/x = 0.5714, c/d = 0.85 and e = 3.0625.\n\nFig. 8\u2014 Reflection coefficient R vs scan angle T,d for E plane scan with \nd/\\ = 0.5714, c/d = 0.85 and e = 3.0625\n\npeak. This is in marked contrast with the H plane scan case (see \nFig. 4). The peak value does not reach one, however. When a dielec- \ntric sheath of thickness A./8 is used to cover the array, the effect is \nto raise the level of reflection almost uniformly over all the scan \nangles. In particular, the peak which occurs at the grating lobe forma- \ntion angle is seen to reach a value of unity. Although not presented \nhere, the results for thinner dielectric sheaths show similar tendencies \nwith the exception that the peak values are not necessarily unity. \nWhile the position for the appearance of the resonant peaks remains \npractically unchanged when the dielectric slab is relatively thin, it \ndoes start to shift toward the broadside direction as the thickness is \nincreased beyond a critical value of about \\,/4. A further thickening \nof the dielectric sheath eventually leads to the emergence of multiple \nresonant peaks.\n\nA quasi-E plane scan is a scan in which the fields have a sinusoidal \nvariation in the direction perpendicular to the plane of scanning.\u2018 \nThe reason for considering such a scan condition is that it enables us \nto simplify a vector three dimensional problem for a planar array to a \nscalar two dimensional one.\n\nThe z-directed propagation constants of the exterior modes for this \ncase are given by\n\nFrom this expression, it is easy to show that when m < dVe \u2014 (d/2b)\"/d \n< (m + 1/2), there will be (2m + 1) modes propagating for\n\nthe number of propagating modes will increase from (2m + 1) to \n(2m + 2) as the scan angle is steered through the angle of transition \nTd = 2r[m +1 \u2014 dWVe \u2014 (d/20)\u2019/XI.\n\nWith the parameters of the array chosen to be a/\\ = b/d = 0.5714, \nc/d = 0.937, the array radiates one beaminO S 7T,d S T,,d = (2rd/))- \n[1 \u2014 (\\/2a)]?, and no beam in 7,,d S T,d S wz. Fig. 9 shows some\n\nFig. 9 \u2014 Reflection coefficient R vs scan angle T,d for quasi-E plane scan with \nb/\\ = d/x = 0.5714, c/d = 0.937, \u00ab = 3.825.\n\nresults of the reflection coefficient for this array when it 1s covered by a \ndielectric sheath with dielectric constant 3.8385. The choice of this \nrelatively large value of dielectric constant is again dictated by the \ninterest in making as large a range of scan angle as possible for a mode \nin the region outside the waveguides to have surface wavelike behavior.\n\nFrom the results for both the H plane and \u00a3 plane scans as we already \ndiscussed, it might be anticipated that a resonant peak would be \nencountered with a relatively thin dielectric slab. The actual calculation, \nhowever, shows that this is not so. In fact, the reflection coefficient \nstarts to show sharp peaks only when the slab thickness exceeds 3),/4, \nwhere\n\nMoreover, even when a resonant peak is present, the reflection coeffi- \ncient as a function of scan is usually quite flat except near the scan \nangle at which the resonant peak appears.\n\nSo far, we have presented results for the situation where the element \nspacing and the dielectric constant were kept constant while the thick- \nness of the dielectric sheath was varied. We studied the effects of the \nsheath thickness on the location of the resonant peak in some detail \nunder these conditions.\n\nNow let us look at some data showing the effect of the waveguide \nwall thickness. Fig. 10 shows both the amplitude and phase of the \nreflection coefficient as functions of the scan angle for an array which \nis scanned in the E plane and is covered by a single dielectric sheath. \nWe obtained the results with fixed values for the element spacing \nb/X = 0.5714, a dielectric constant of the sheath e = 3.0625, and its \nthickness d,; = 0.5 A,. However, the waveguide wall thickness (or \nequivalently, the size of the radiating aperture) is varied over a \nwide range. It is evident from the graphs in Fig. 10 that the size of \nthe radiating aperture has substantial effects on the reflection coef- \nficient. The scan angle for the appearance of the resonant peak is \nshifted toward the broadside direction as the aperture decreases. \nSimilar shift in the scan angle for the resonant peak has also been \nobserved in the case of the H plane scan. Moreover, loading the wave- \nguides with a dielectric material also causes similar effects. It is ap- \nparent that the resonance phenomenon is strongly dependent on the \naperture impedance which is a function of the array parameters.\n\nThe results presented earlier indicate that when a dielectric slab \nthicker than ),/4 is used to cover an array, an important effect is \nmanifested by the appearance of sharp resonant peaks in the reflec- \ntion coefficient. Such resonances may be avoided if thin dielectric\n\nsheaths are used. A thin dielectric sheath covered array for H plane \nscanning has these dimensions: a/A = b/d = 0.5714, and the dielectric \nslab has a fixed thickness ds = A/16. We obtained the results for \ndielectric constants ranging from 1.25 to 3. As the graphs in Fig. 11 \nshow, even though it does not cause \u2018resonance,\u2019 a thin dielectric \nslab does have a considerable effect on the array match characteristics. \nThis is true even when the dielectric constant is close to one. In fact, \nin the example shown here for e = 1.25, the reflection coefficient seems \nto have a larger variation than that of an uncovered array. How- \never, by varying the dielectric constant, and perhaps also the thick- \nness of the slab (as long as the slab is thin), it 1s possible to obtain a \nsuitable combination of these two parameters and thus achieve a flat \nresponse in both amplitude and phase of the reflection coefficient at a \nsingle frequency. Such a possibility has been demonstrated previously.* \nThe significance of this possibility is that a thin dielectric slab may be \nused as an effective means for the wide angle match of an array over \na narrow frequency band.\n\nWe have shown the effects of a dielectric slab on the radiation char- \nacteristics of a phased array. One outstanding feature of the substan- \ntial changes which the dielectric slab brings to array performance is \nthe appearance of resonance peaks in the reflection coefficient of the \narray at certain scan angles. This phenomenon is caused by surface- \nwavelike space harmonics at the interface between the dielectric and \nfree space.\n\nWhen the fields in the free space region are expanded into a gen- \neralized Fourier series according to the Floquet theorem, each space \nharmonic may be viewed as homogeneous or inhomogeneous plane \nwaves, depending on whether it 1s propagating or evanescent. In the \nabsence of a secondary boundary at z = d, introduced by the dielectric \nslab, the plane waves generated at the array aperture either propagate \naway from it if they are homogeneous, or decay away if inhomo- \ngeneous.\n\nWhen an air-dielectric interface is introduced at 2 = d,, it causes \neach space harmonic generated at the aperture to be reflected and \nrefracted upon encountering the interface as it travels away from \nthe aperture. Naturally, only wave modes of the propagating type \nare significantly affected, because those of the evanescent type are \nusually rapidly attenuated away from. the aperture. Furthermore,\n\nFig. 11\u2014 Reflection coefficient R vs scan angle T.b for an array covered with \na thin dielectric slab (H plane scan, a/\\ = b/A = 0.5714, d, = 2/16).\n\nmodes which exhibit a surface-wavelike field distribution suffer total \nreflection at the interface. The energy reflected from the interface is \nreturned to the aperture as a wave incident to the array from the ex- \nterior region, and is scattered there.\n\nIn the presence of a dielectric sheath, there usually exists a range \nof scan angles over which there are two or more propagating modes \ninside the dielectric while only one propagating mode is present in \nthe free space region. Under such circumstances, the two propagating\n\nmodes inside the dielectric interact as they are multiply-scattered at \nthe array aperture. The degree of interaction depends on the various \narray parameters as well as the scan angle. It 1s then possible that, \ngiven a suitable set of parameters, the multiple scatterings between \nthe two interfaces might lead to a situation in which a large reflection \nis generated at certain scan angles. The numerical examples in Sec- \ntion IIT indicate that the reflections can be so high as to reach unity \nfor all practical purposes. In fact, it has been inferred from the \nnumerical results for the transmission coefficient that the reflection \ncan indeed reach the value one exactly.\n\nThe fact that the modulus of the reflection coefficient can attain \nexactly the value of unity assumes some special significance. For \nwhen this happens, inside the waveguides at a distance away from \nthe array aperture such that all the evanescent modes are sufficiently \nattenuated, the incident and reflected waves combine to form a pure \nstanding wave. This implies that the tangential electric and mag- \nnetic fields alternatively go through zero at one half guide wavelength \nintervals. Specifically, by writing R = e/, the resultant tangential \nelectric field under these conditions may be expressed as\n\nHence, for 2 = \u2014L, such that (a,;L,+9/2) = na, or equivalently Ln \n= (nr \u2014 \u00a2/2)/a,, the tangential electric field vanishes. As a conse- \nquence, electric conductors may be introduced at these positions with- \nout disturbing the field distributions of the entire system. When this \nis done, however, the array is then transformed into a corrugated \nstructure which is completely isolated from the source region. The \nsolution for the aperture fields obtained under these situations may \nthen be regarded as the solution of a surface wave which propagates \nalong a corrugated structure.\n\nThere are two aspects of this conclusion which deserve further \ncomments. A corrugated surface has long been known as a structure \nwhich is capable of supporting surface waves. The characteristics of \nsuch a system are that it supports only the TM type surface wave, \nrelative to the direction of propagation, in the absence of a dielectric \nmaterial and that the period of corrugation be less than a half free \nspace wavelength in order for the wave to propagate. Moreover, there \nis some requirement in regard to the depth of the corrugation for the \nsurface to be properly reactive.\n\nThe introduction of a dielectric sheath above the array aperture or \ncorrugated surface produces two significant effects. One is to enable \na surface wave to propagate in the TE mode, corresponding to the \ncase of H plane scan, as well as in the TM mode. The other is that the \nsurface wave propagation is possible even though the period of cor- \nrugation is larger than a half free-space wavelength. It 1s important \nto observe that when the corrugation period is larger than a half wave- \nlength, the zero\u00ae space harmonic of the fields outside the corrugation \nis a propagating mode. Therefore, it is necessary for the modal coef- \nficient of this mode to be identically zero in order for the fields to \nconform to that of a surface wave. The results presented in this calcu- \nlation indicate that indeed this is the case.\n\nThe scan angles at which resonant peaks appear may be determined \naccurately by the transverse resonant method when the radiating \naperture is small. This situation applies readily to the case of the E \nplan scan. In this method, the sum of impedances looking towards the \npositive and negative z directions at some reference plane is set equal \nto zero, thus yielding a characteristic equation. It is convenient to use \nthe array aperture as the reference plane. When the radiation aperture \nis small in comparison to the size of a periodic cell, the \u201caverage\u201d im- \npedance looking toward the array side is almost zero. The impedance \nlooking away from the aperture may be approximated by that of the \nlowest order mode, in this case the (\u20141)st space harmonics. The im- \npedance of this mode is given by\n\nBy setting this expression to zero and using the appropriate modal \nimpedances, we find\n\nWhen this equation is used to determine the scan angle for the ap- \npearance of the resonant peak, it is found that excellent agreement \nwith actual calculation is obtainable for aperture size up to 10 per- \ncent of the size of the periodic cell. For larger aperture sizes, the \nagreement gradually becomes poorer. This is because the impedance \nlooking toward the waveguide side no longer remains negligibly small. \nIt appears at this time that an accurate determination of the scan \nangle for resonant peaks under such conditions is best carried out by\n\nsolving the boundary value problem directly. Fortunately, this is a \nrelatively easy task nowadays with the help of high speed electronic \ncomputers.\n\nThe numerical results obtained so far have revealed that the occur- \nrence of sharp resonant peaks is associated with rather thick dielec- \ntric sheaths, and such resonance may be avoided by using thin sheaths. \nThus, dielectric covering of an array is still a useful tool for the pro- \ntection of the array from its environment. More importantly, the scan \n(or incident angle) dependent reflectivity of dielectric slabs may be \nutilized to advantage in improving the match performance of an array. \nThe feasibility of this desirable feature has been demonstrated by \nextensive data presented herein and in [1]. It has also been suggested \nby other workers.*: 7\u00b0 Although, based on the calculated results, the \nimprovement in array match by a thin dielectric sheath is obtainable \nat the expense of a higher frequency sensitivity, it seems possible to \nachieve a broadband compensation by using multiple thin sheaths. \nFurther work is being carried out along this line, and the results will \nbe reported at a later date.\n\n1. Galindo, V. and Wu, C. P., Dielectric Loaded and Covered Rectangular \nWaveguide Phased Arrays, presented at International Symposium on An- \ntennas and Propagation, Palo Alto, California, December 1966.\n\n2. Galindo, V. and Wu, C. P., Numerical Solutions for an Infinite Phased Array \nof Rectangular Waveguides with Thick Walls, IEEE Trans. Antennas and \nPropagation, AP-14, March 1966, pp. 149-158.\n\n. Wu, C. P. and Galindo, V., Properties of a Phased Array of Rectangular \nWaveguides with Thin Walls, IEEE Trans. Antennas and Propagation, \nAP-14, March 1966, pp. 163-172.\n\n. Hildebrand, F. B., Methods of Applied Mathematics, Prentice-Hall, Inc., \nNew York, 1954, pp. 451-452.\n\n. Kantorovitch, L. V. and Krylov, V. I., Approximate Methods of Higher \nAnalysis, Interscience Publishers, New York, 1958.\n\n. Waveguide Handbook, N. Marcuvitz, ed., Radiation Laboratory, Series 10, \nMcGraw-Hill Book Co., New York, 1950.\n\n. Lee, S. W., Impedance Matching of an Infinite Phased Array by Dielectric \nSheets, Electronics Letters, 2, October 1966, pp. 366-368.\n\n. Galindo, V. and Wu, C. P., The Relation Between the Far-Zone Pattern \nof the Singly Excited Element and the Transmission Coefficient of the \nPrincipal Lobe in an Infinite Array, IEEE Trans. Antennas and Propaga- \ntion, AP-14, May 1966, pp. 397-398.\n\n10. Magill, E. G. and Wheeler, H. A:, Wide Angle Impedance Matching of a \nPlanar Array Antenna by a Dielectric Sheet, THEE Trans. Antennas and \nPropagation, AP-14, January 1966, pp. 49-53.\n\n11. Collin, R. E., Field Theory of Guided Waves, McGraw-Hill Book Co., New \nYork, 1960, Chapter 8.\n\nA systematic approach to the analysts and construction of channel \ncodes for digital baseband transmission is presented. The structure of the \ncodes ts dominated by the set of requirements imposed by channel charac- \nteristics and system operation. These requirements may be translated into \nsymbol sequence properties which, in turn, specify a set of permissible \nsequence states. State-dependent coding of both fixed and variable length \n1s a direct result. Properties of such codes are discussed and two examples \nare presented,\n\nBinary information to be transmitted over a baseband digital system \nmust typically be encoded into a sequence of symbols suitable for \ntransmission through the channel. The structure of such codes is \ndominated by the set of requirements imposed by considerations \nsuch as channel characteristics and system operation. Several codes \ndesigned for baseband transmission have been discussed in the litera- \nture, 23 but the analysis of such coding has received little attention. \nThis paper presents a systematic approach to the description and \nconstruction of codes for such application. The symbol sequence to \nbe generated by the code is viewed as the output of a sequential ma- \nchine with a set of permissible states derived from the imposed re- \nquirements. State-dependent codes, both of fixed and of variable \nlength, are a direct result. Various properties of such codes are dis- \ncussed, and two examples are presented.\n\nAn advantage of using a sequence-state point of view is the ease \nwith which it may be adapted to computer analyses, which are feasi- \nble due to the typically short word length of such codes (usually less\n\nthan eight symbols), and useful because normal design procedure \noften entails the construction of a number of codes, derived from \ndifferent sequence requirements, in order to compare properties.\n\nA block diagram of a baseband digital transmission system is shown \nin Fig. 1. Binary signals are encoded into a sequence of pulses (sym- \nbols) which are transmitted over a channel with regularly spaced re- \nconstructive repeaters.* Typically, a number of factors strongly \ninfluence the choice of a channel code. Examples are:\n\n(2) The symbol sequence is often required to furnish timing infor- \nmation to the repeaters.\n\n(72) The channel may impose restrictions on the spectral shape of \nthe sequence. Most such systems, for example, have a transmission \nnull at de. This implies that long strings of SyvOr of one polarity \ncannot be tolerated.\n\n(122) Provisions for monitoring the channel error rate during normal \noperation may be necessary.\n\n(tv) Pulse sequence requirements must be satisfied independently \nof the source statistics.\n\nRequirements such as the above may usually be translated into \nlimitations on the length of strings of like symbols, specifications of \nsets of allowable transitions between symbols (such as when a + 2 \npulse must be followed by a \u2014 2 pulse to avoid timing jitter due to \nasymmetries), and bounds on the variation of the running digital \nsum. The latter is defined as\n\nT \n\u00a3(T) = D Qi y \nt=T'o \nwhere {a;} are the weights assigned to the channel symbols and Ty \nis an arbitrary but fixed time.\n\nThe objective of the code designer is to specify a code which meets \nthe imposed conditions while obtaining maximum information ca- \npacity (average number of bits per channel symbol) or alternatively \nto optimize sequence properties.\n\nPULSE STREAM \nBINARY SUITABLE FOR BINARY \nwalks CopER LLRANSMISSION) REPEATER REPEATER DECODER\n\nA code may be defined as a mapping from the binary input infor- \nmation to the ensemble of symbol sequences which satisfy the im- \nposed criteria. The allowable symbol sequences can, in turn, be speci- \nfied by a set of permissible states {s;} for each point, describing such \nquantities as the present value of the running digital sum, the pre- \nceding number of symbols of a given type, etc. In addition to the {s;}, \nit is convenient to define a set of allowable words {w,;}, which take the \nsequence through a succession of allowable states, one for each sym- \nbol in the word. A necessary condition for w; to be permissible is that \nthere exists at least one state s, such that the states resulting from the \nuse of w,; at s, are all allowable. In addition, the words to be used in \nthe code may be restricted to a subset of {w;} which satisfies require- \nments not specified by a particular choice of states, such as various \nerror-correcting properties. The latter restriction will not be treated \nin this paper.\n\nIt is assumed in this paper that each code word carries an integral \nnumber of information bits, and that the ratio of bits per symbol re- \nmains constant over each word. The latter assumption is a sufficient \ncondition for synchronous transmission.\n\nThe simplest codes of the above type are those of fixed length, \nwhich may be defined as mappings from the set of input binary words \nof length logel to the set of allowable words {w,} of length N. These \nmappings are generally state-dependent. That is, the choice of a code \nword may be a function of the state occupied by the sequence.\n\nDefimtion: W(s;) is the set of words (alphabet) of length N which \ntake the symbol sequence from state s; through a succession of allow- \nable states.\n\nThe fact that each word w; must carry a = logeL information bits \nimplies that words used in a code must terminate in states for which \nW (s) contains a minimum of 22 words.\n\nDefintion: The principal states {S,} of the channel sequence are \nthose states for which W{S,,} contains at least 2\u00a2 words each of which \ntakes the sequence to another principal state.\n\nficient condition for the existence of a code. It is possible to implement \ncomputer search routines for finding such states. The appendix con- \ntains a description of a search method which was used as an aid on \nconstructing the two codes in Section 5.2.\n\nA given code actually occupies a subset {c,} of the principal states \nat the end of code words. Let W*(c,) C W(c,) be the words actually \nused for coding when the sequence occupies the state o, .\n\nThe above relation partitions the set of terminal states into alphabet \nclasses Rz, g = 1,2,...Q. A set of words W*(R,) is associated with \neach such class. Each W*(R,) contain 2\u201c words which are used when \nthe sequence occupies one of the terminal states in the class Ry (1e., \neach word in W*(R,), g = 1, 2,... Q is associated with one binary \ninput word b,). It is advantageous to minimize the number of alphabet \nclasses in order to simplify the coding and decoding circuits.\n\nThe coder tracks the state of the symbol sequence, and at the end \nof each word, codes the next binary word 6, into a word chosen from \nthe alphabet corresponding to the state occupied by the output se- \nquence. For example, the Paired Selected Ternary code book con- \nsists of two alphabets, which contain words of zero or positive weight \nand zero or negative weight, respectively. The coder switches to a \ndifferent alphabet after use of a word with non-zero weight.\n\nThe class of codes defined in Section 2.1 may be uniquely decoded \nprovided the state of the symbol sequence is known. Frequently, how- \never, it is not feasible for the decoder to distinguish which member \nof a given subset of states the sequence occupies at the end of a code \nword. For example, the states may be functions of the instantaneous \nsymbol sequence sum. Here a single error in detection may result in \nan unbounded string of decoding errors. In addition, implementation \nof state-dependent decoding requires circuits to track the sequence \nstate. Thus, state-independent decoding may often be necessary.\n\nDecoding independently of the state is possible if and only if one \nbinary word 6, corresponds to each code word w; \u00ab {w,;}. That is, the \nstate-dependent mapping of {b.} into {w;} must have a unique in- \nverse. If decoding is to be independent of the state within a given \nsubset of the terminal states, the above mapping must have a unique \nInverse within this subset.\n\nGiven a group of terminal states {o,}, q = 1, 2,...Q, and their \nassociated alphabets W*(o,) [each containing 2\u201c words], it is desirable \nto determine whether it is possible to code so that decoding is in- \ndependent of g. The following are necessary and sufficient conditions \nfor the above:\n\nCondition 1: A sufficient, but not necessary, condition for the existence \nof an assignment of binary words to the members of W*(c,) q = 1, 2, \u00b0\u00b0: Q \nso that decoding is independent of t is that for any integers u, v such that \nIlsu<vsQ\n\nProof: Suppose that a binary word b, has been assigned to w, in the \nalphabet W*(c,). Then it is possible to give w, the same binary assign- \nment in W*(o,41) \u00ab++ W*(o,). The lack of necessity of the above condi- \ntion may be demonstrated by a renumbering of the alphabets.\n\nCondition 2: A necessary and sufficient condition for the possibility of \nassigning binary words to the members of W*(o9), 6 = 1, 2 ++: \u00a9 so \nthat decoding 1s independent of 61s that for! SuSv Sx S Oand for each\n\nbe equivalent (in the sense that they are assigned the same binary word). \nThe necessity of the condition follows from the fact that if\n\nthen some other word w, \u00ab W*(c,) must be assigned the binary word \nassigned to w, in the uth alphabet. But if w,,w,eW*(c,),1 Sms Tf, \nstate-independent decoding is precluded.\n\nIt is obvious from the above conditions that state-independent \ndecoding is always possible when there are only two alphabets (as in \nPaired Selected Ternary\u2019). An example of a group of alphabets which \ndo not admit state-independent decoding is shown in Table I.\n\nThe received symbol sequence must be correctly partitioned into \nblocks of words length N before decoding. This may be done by ob-\n\nserving whether the received words coincide with the words utilized \nin a code, by tracking the sequence state to determine whether ends \nof the received words coincide with sequence terminal states, or by \nlooking for the presence of nonallowable sequences of permissible \nwords.\n\nAttempts to increase fixed length state-dependent code efficiency \nwith constraints on pulse sequence properties result in increased \nword length, and thus in rapidly mounting coder and decoder com- \nplexity, reframe times and sensitivity to detection errors. Hence, \nvariable length codes, which may combine the advantages of short \nand long word lengths, are frequently advantageous.\n\nConsider a sequence of symbols generated by a fixed length code \nof word length N. The sequence by definition occupies a principal \nstate at the end of each word. Often, however, a principal state will \nbe entered in the middle of a word. This offers the possibility of using \nthe word fragment to convey information, and of starting any of at \nleast 2* words from this point (by definition of a principal state). The \nresult, subject to a number of restrictions outlined below, is a variable \nlength code.\n\ncompared to the 256 required for a single alphabet of a fixed length \ncode of equal efficiency and sequence properties.\n\nThe structure of variable length codes required to satisfy sets of \nsequence requirements is quite similar to that of fixed length codes. \nA number of special features, however, arise from the presence of \nwords of different lengths.\n\nThe requirement of synchronous transmission, coupled with the \nasumption that each word carries an integer number of information \nbits, implies that the word lengths in a given code are integer multiples \nof a basic word length N, where N is the smallest integer for which \nthe bit per symbol ratio a/N is that of two integers (i.e., if the bit per \nsymbol ratio is 1.5, the basic word length is 2). The proof of this state- \nment follows from the fact that the word length K,;N must be such \nthat the number of bits per word Kja is an integer [this paper covers \nthe transmission of binary data]. Suppose that K; = n + b, where n \nis an integer and 0 < b < 1. Then ba and DN must also be integers, \nwhich violates the definition of N.\n\nThe set of allowable words may be written as {wy},7 = 1, 2,..., \nM, where wi; is the 7th word of length 1.\n\nDefintion: W;(s;) is the set of words of length 2:N,1 = 1,2,...,M, \nwhich may be used in the permissible state s; without violating the \nsequence requirements.\n\nA set of principal states may be defined for a symbol sequence as- \nsociated with a variable length code in a manner analogous to that \nof Section IT.\n\nDefimtion: The principal states {S,} are those allowable states for \nwhich W(S,) contains sufficient words terminating in principal states \nto maintain an information rate of a bits per N symbols.\n\nThe number of words of length 1V,27 = 1,2... M required to main- \ntain a bit rate of a bits per N symbols is given by\n\nwhere L,[W(s,)] is the number of words in W,,(s;). Equation (1) clearly \nindicates the advantage of using as many short words as possible.\n\nThe terminal states {c,,} are as defined in Section II. Similarly, \nW*(o,,) and W*(c,,) are those members of W;(c,,) and W(c,,), respec- \ntively, which are actually used in the code; two terminal states o,, \nand o, are in the same alphabet class if and only if W*(c,,) = W*(o,).\n\nfor all w,;; , wi, \u00a9 W*(c.) and w;; ~ wi, . P;(w;,) denotes the first \u00abNV \nsymbols of the word w,, . The above condition follows from the fact \nthat w;; \u00ab W(S.) implies that w;; , when used in the state S, , takes \nthe sequence to another principal state.\n\nState-independent decoding of a variable-length code is desirable \nfor the same reasons as for one of fixed length. The criterion for such \ndecoding is the same as for a fixed length code \u2014 the state-dependent \nmapping from the binary words {b,} to the words to be transmitted in \nthe channel must have a unique inverse. Two considerations which \naffect the complexity of this mapping are the presence of words of \nlength 7V,2 = 1, 2,..., M and the existence of words which consist \nof concatenations of two or more shorter words. Such word combina- \ntions will be called concatenated words. It follows from Section 3.2 \nthat a concatenated word and its prefix cannot occupy the same \nalphabet. From Section 3.2, it can be concluded that a concatenated \nword of length 2N is a legitimate word in a given alphabet if no \nterminal state is reached with the KNth symbol for0 < K<1A \nrequirement for state-independent decoding is that the binary words \nassigned to a concatenated word be concatenations of the binary \nwords assigned to the words of which the concatenated word is com- \nposed. Section 5.2 contains an example of a variable-length code \nwhich satisfies these criteria.\n\nThe process of framing (at the decoder) a symbol sequence gen- \nerated by a variable-length code may be divided into two parts: \npartitioning the sequence into blocks of length N (the basic word \nlength), and properly grouping these blocks into words of length iN, \na=1,2, ...,M. Framing may be performed in much the same man- \nner as for a fixed-length code. A misframe condition may be detected \nby noting the presence of nonpermissible words, word sequences which\n\ndo not occur in the code, and/or the occupation of a nonterminal state \nat the end of a word.\n\nIt is often possible to substantially simplify the framing process by \nensuring that secondary misframing (the condition that results from \nimproper grouping of blocks of length N into words) is self-terminat- \ning. This restricts the need for framing to blocks of length N (the \nbasic word length), and frequently results in superior framing per- \nformance compared to a fixed length code of equal efficiency and \nsequence properties.\n\nConsider a state-independent decoder which stores all word prefixes \nof length iN,2 = 1,2,. .. , M\u20141, and which operates in the following \nmanner: If the first fN symbols that are received form a prefix of a \nlegitimate word, the decoder considers the first (f+1) N symbols. \nOtherwise the first fN symbols are decoded into an arbitrary binary \nword of length fa. The concatenated words are treated as strings of the \nshorter words of which they are composed.\n\nA sufficient condition for secondary misframing to be self-termi- \nnating in the above decoder is that (excluding concatenated words)\n\nfor all r, k, 7, m, e such that e > r andi > r. E,[wz;] denotes the last \nrN symbols of the word w,;. This ensures that the secondary mis- \nframing condition terminates with the ending of the word wx.\n\nIt is often necessary to monitor the level of channel performance \nwithout interrupting service. Codes with bounded sequence sums \nreadily lend themselves to this objective, especially in the low noise \nenvironment typical of telephone communications systems. This is \nbecause an error is quite likely to eventually cause the sequence sum \nto exceed the imposed bounds. Alternatively, the behavior of the \ndigital sequence at the bound can be observed. For example, not more \nthan one consecutive zero (null pulse) might be permitted at extreme \nsequence sum values, while strings of length N > 1 may occur at \nintermediate sums. Here, the detection of a string of zeros at an ex- \ntreme sequence sum value indicates that an error has occurred in \ndetection. An advantage of these methods' is that error monitoring \ncan be performed without framing.\n\nStandard methods of error monitoring (or error detection) can also \nbe incorporated in a code of this form. A possibility is the require-\n\nment that the words have even digital sums, or more generally that \nthe terminal states {o;} be separated by a minimum \u201cdistance.\u201d\n\n(zv) Terminal states and code words are chosen for (if necessary) \nstate-independent decoding, to optimize spectral properties, error \nmonitoring and framing statistics. This usually requires much tedious \ncomputation.\n\nIt may occur that a given set of sequence requirements are such \nthat a set of principal states does not exist for codes of word length \nN, but that a length of MN yields an acceptable code. In this case, a \nvariable length code of basic word length N with maximums length \nMN might be a possibility. It is advantageous to use as many words \nof length N as possible, and to utilize the longer lengths to achieve a \nsufficient number of words inn each terminal state.\n\nThe following two ternary codes, one of fixed length and one of \nvariable length, are examples of results of the above procedures. \nThese codes were designed for possible use in a future high-speed \nbaseband digital communication system.\u2019 The symbols to be trans- \nmitted on the line consist of positive, negative and null pulses. The \nmain objectives in the design of these codes was to restrict the varia- \ntions in the running digital sum of the pulse stream (with positive, \nnegative, and null pulses assigned values of +1, \u20141, and O respec- \ntively) to avoid de buildup in the channel, and to have as many transi- \ntions between symbols as possible in order to provide energy for \nself-timing in the repeaters.*\n\nThe M848 selected ternary code (Table II) is an example of a \nfixed length, three alphabet code of word length three. The digital \nSequence sum varies over six levels, which are arbitrarily labeled \nzero to five. These six levels are the allowable states. Terminal states \noccur only at values of one to four inclusive. There are three alphabet \nclasses. R, and Rg correspond to terminal states with sequence sums \nof one and four respectively, while R. corresponds to a sum of either \ntwo or three. The choice of M843 words is such that strings of zeros \nand symbols of like sign are limited to lengths of four and five re- \nspectively.\n\nCoding is performed as a function of the terminal state. For ex- \nample, if 00110100 is to be coded when the value of the sequence sum \nis +1, the first four bits are coded as O\u2014+ and the second as 0+-+. \nIt can easily be seen from Table II that decoding is independent of \nthe state.\n\nThe VL43 code [Table III] is an example of a three alphabet \nselected ternary code of variable length. The basic word length is \nthree and the maximum length six. The bit per symbol ratio is 4/8, \nas in MS48. Secondary misframing is self-terminating and decoding is \nstate-independent. The variable length feature permits a decrease \nin the variation of the running digital sum from six to five levels, and \na significant increase in density of transitions. There are three \nterminal states (01, 02, 03) each with its own alphabet. W*(c2) con- \ntains only words of length three. W*(c,) and W*(o3) each contain 16 \nwords of length six (each carrying eight bits) and 15 words of length \nthree. Both W*(o,) and W*(o3) contain concatenated words.\n\nCoding is performed in much the same manner as for MS43. For \nexample, if 110000010000 is to be coded when the state is o2, the first \nfour bits are coded as \u2014\u2014+, after which the state is o,. The next ian \nbits are coded into a six symbol word, -+\u2014+\u2014+.\n\nThis paper has presented a description of a number of properties of \nmultilevel coding for synchronous baseband transmission. The domi- \nnant feature of such coding is the set of pulse sequence requirements \nimposed by channel characteristics and system operation. It was \nshown that multialphabet and variable length codes are a natural \nconsequence of attempting to attain high efficiency with codes of this\n\nThe author is indebted to Miss N. K. Shellenberger for program- \nming support, and to J. M. Sipress for many valuable discussions.\n\n(\u00ab) The maximum number of different levels which can be assumed \nby the running digital sum (with positive, zero and negative pulses \nassigned values of +1, 0, and \u20141, respectively) is 2Q + 1.\n\nThe first step is a search for the set {w} of words of length N which \nsatisfy the above conditions. \n+++ Qy; , with a;; = \u20141, 0 or +1. It is then required\n\nNext, a set of allowable states s(t,u,v) is defined, where \u00a2 denotes \nthe value of the running digital sum of the pulse sequence, w the \npreceding number of zeroes and v > 0 (v < OQ) the number of preced- \ning positive (negative) pulses. It is required that\n\nThe words {w}, when used in a particular state s(t,u,v) take the \nsequence through N states. A word wy; is permissible in s(\u00e9,u,v) if \nand only if the N resulting states are allowable. Each word in {w} \nis tested to determine the states in which it is permissible. The re- \nsult is that a set of words {w}:4,\u00bb is associated with each state s(t,u,v).\n\nWhen the above procedure is completed, the program searches for \nthe principal states. States which contain less than 2* words are dis- \ncarded. Then, for each remaining state s(t,u,v), those words in \n{whtu,o are eliminated which terminate in a state which has been \ndiscarded. If {w}z.4,, contains fewer than 2\u00a2 words after this opera- \ntion, s(t,u,v) is eliminated. This procedure is continued until either \nno states remain or the routine runs through a complete cycle of \nstates without eliminating any words or states. In the latter case, the \nremaining states are the principal states, {o,}, and the remaining \nwords are {W (aq) }.\n\n1. Sipress, J. M., A New Class of Selected Ternary Pulse Transmission Plans \nfor Digital Transmission Lines, IEEE Trans. Commun. Tech., COM-13, \nNo. 3, September, 1965, pp. 366-372.\n\n2. Davis, C. G., An Experimental Pulse Code Modulation System for Short- \nHaul Trunks, B.8.T.J., 41, January, 1962, pp. 1-24.\n\n4, Mayo, J. 8., A Bipolar Repeater for Pulse Code Modulation Signals, B.S.T_J., \n41, January, 1962, pp. 25-97.\n\n5. Dorros, I., Sipress, J. M., and Waldhauer, F. D., An Experimental 224 Mb/s \nDigital Repeatered Line, B.S.TJ., 45, No. 7, September, 1966.\n\nPaut T. Brapy, B.E.E., 1958, Rensselaer Polytechnic Institute; \nM.S.E.E., 1960, Massachusetts Institute of Technology; Ph.D., 1966, \nNew York University; Bell Telephone Laboratories, 1961\u2014. He has \nworked in statistical modeling of on-off speech patterns and of \nspeech level distributions, and in studying two-way transmission \nof speech and data on circuits containing transmission delay.\n\nP. A. Franaszek, Se.B., 1962, Brown University; M.A., 1964, Ph.D., \n1965, Princeton University. Bell Telephone Laboratories, 1965\u2014. Mr. \nFranaszek has been concerned with the analysis of digital data trans- \nmission problems. Member, Sigma Xi, Tau Beta Pi, AAAS.\n\nVictor GALINnbo, B.S. (E.E.), 1954, New York University; M.S. \n(E.E.), 1962, and Ph.D. (E.E.), 1964, University of California, \nBerkeley; Hughes Aircraft Company, 1954-1957, 1958-1960; M.I.T. \nLincoln Laboratory, 1957-1958; Bell Telephone Laboratories, 1964\u2014. \nMr. Galindo has been engaged in applying electromagnetic theory \nto studies of microwave transmission devices, antennas, and phased \narrays. Member, Eta Kappa Nu, Tau Beta Pi, IEEE.\n\nMarco A. Murray-Lasso, M.S., E.E., 1962, and Sc.D., E.E., 1965, \nboth from Massachusetts Institute of Technology; Bell Telephone \nLaboratories, 1965\u2014. Dr. Murray-Lasso has done theoretical research \nin communication system components, computer studies involving \npredicting the behavior of directional couplers, and design work on \ncontrol and data reduction systems. Member, IEEE, Sigma Xi, As- \nsociation of University Professors of Mexico, and Association of \nMechanical Electrical Engineers of Mexico.\n\nLAWRENCE R. Rasiner, \u00a7.B. and \u00a7.M., 1964, and Ph.D., in elec- \ntrical engineering, 1967, all from the Massachusetts Institute of Tech- \nnology. From 1962 through 1964 he participated in the cooperative \nplan in electrical engineering at Bell Telephone Laboratories, Whip- \npany and Murray Hill, New Jersey. He worked on digital circuitry, \nmilitary communications problems, and problems in binaural hearing. \nHe joined the staff of Bell Laboratories in 1967 and has been engaged\n\nin speech communications research. Member, Eta Kappa Nu, Sigma \nXi, Tau Beta Pi, Acoustical Society of America, and IEEE.\n\nHan-cuiu Wane, B.8.H.E., 1955, Cheng-kung University, Taiwan, \nChina; M.S.E.E., 1960, University of Notre Dame; Ph.D., 1965, Poly- \ntechnic Institute of Brooklyn; Chinese Government Radio Adm., \n1956-1958; Polytechnic Institute of Brooklyn, 1964-1965; Bell Tele- \nphone Laboratories, 1965\u2014. He is currently engaged in developing \nmicrowave components for radio relay systems.\n\nC. P. Wu, B.S., 1956, National Taiwan University; M.S., 1959, \nand Ph.D., 1962, Ohio State University; Bell Telephone Laboratories, \n1962\u2014. Mr. Wu was an assistant instructor at the National Taiwan \nUniversity, 1956-57. He has done research in electromagnetic radia- \ntion in anisotropic media. His present work includes phased array \nantennas and electromagnetic scattering. Member, IEEE, Sigma Xi.", "title": "magazine :: Bell System Technical Journal :: BSTJ V47N01 196801", "trim_reasons": [], "year": 1968}
{"archive_ref": "bitsavers_BellSystemJV53N10197412_17899315", "canonical_url": "https://archive.org/details/bitsavers_BellSystemJV53N10197412_17899315", "char_count": 593142, "collection": "archive-org-bell-labs", "doc_id": 625, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc625", "record_count": 712, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bitsavers_BellSystemJV53N10197412_17899315", "split": "test", "text": "Line-Power Feed \nLine-Protection Switching \nCentralized Transmission Surveillance\n\nJ. B. FRY, Art and Production Editor \nF. J. SCHWETUE, Circulation \nW. G. SCHEERER, Coordinating Editor of L5 System Articles\n\nTHE BELL SYSTEM TECHNICAL JOURNAL is published ten times a year by \nthe American Telephone and Telegraph Company, J. D. deButts, Chairman \nand Chief Executive Officer, R. D. Lilley, President, J. J. Scanlon, Executive \nVice President and Chief Financial Officer, F. A. Hutson, Jr., Secretary. Checks \nfor subscriptions should be made payable to American Telephone and Telegraph \nCompany and should be addressed to the Treasury Department, Room 1038, 195 \nBroadway, New York, N. Y. 10007. Subscriptions $15.00 per year; single copies \n$1.75 each. Foreign postage $1.00 per year; 15 cents per copy. Printed in U.S.A.\n\nCoaxial-carrier transmission systems constitute a significant portion \nof the Bell System long-haul transmission facilities. These systems \nhave been developed over several decades to provide the basis of a \nhigh-quality, high-capacity, long-distance communications network.\n\nAfter extensive exploratory work on wideband amplifiers and coaxial \ncable, the feasibility of a coaxial-carrier system was demonstrated in \n1936 between New York and Philadelphia. The success of this trial \nwas followed by development of the first Bell System coaxial-carrier \ntransmission system, the L1. Placed in service in 1941, the L1 system \nwas initially capable of carrying 480 four-kHz two-way message chan- \nnels per pair of 0.270-inch-diameter coaxial cables, with a repeater \nspacing of 5.5 miles. Soon, 0.375-inch-diameter cable became standard \nand, with system improvements, the vacuum-tube-operated L1 system \nwas capable of carrying 600 circuits per coaxial pair with 8-mile \nrepeater spacing. Its capacity was later increased to 720 circuits.\n\nThe major expense of the coaxial system has been in the outside \nplant area: cable, cable placement, right-of-way, and buildings, \nincluding aboveground or underground structures for housing re- \npeaters. Once this portion of the system is established, development \nof electronic equipment to provide maximum utilization of the cable \nbecomes economically attractive. Each successive generation of \nrepeaters achieved wider transmission bandwidth and, hence, larger \nchannel capacity through use of shorter repeater spacing, new tech- \nnology, and more advanced system concepts. Use of cables with more\n\nTelephone Typical \u2018 \nCircuits pone PEPIN \nFirst Units Repeater Honor: Other Key \nSystem Services. Pee poi aere Technology Repeater Types Equalization System Features \nPer Sheath | Pray \nCoax Per (Miles) \nPai Cable* \nair \nLi 1941 720t 720 4 8 Vacuum tube Manually and automatically | Adjustable static \n2,160 8 adjusted regulating repea- and dynamic \nters; equalizing repeaters \u201cbump\u2019\u2019 shapes \nL3 1953 1,860 5,580 8 4 Vacuum tube, Buried-thermistor and line- Manual \u201c\u2018cosine\u201d\u2019 Hardened con- \n,300 12 statistically con- pilot-controlled regulating shapes; dynamic figuration \ntrolled key repeaters; equalizing broad shapes \ncomponents repeaters \nL4 1967 3,600 32,400 20 2 Discrete transistor, | Fixed basic repeaters; regu- Static \u2018\u2018bump\u201d Noise objective more \nprinted wiring lating repeaters controlled shapes stringent by 4 dB \nboard by both a buried thermis- \ntor and line pilot; \nequalizing repeaters \nL5 1974 | 10,800t | 108,000{ 22 1 Discrete transistor, Static \u2018\u2018bump\u201d\u2019 Phase-shaping net-\n\nt Originally 480 circuits per coaxial pair, but widely used at 600-circuit capacity.\n\nt$ Extensions to L5 are being developed to provide 13,200 telephone circuits per coaxial pair, or 132,000 circuits per 22-tube cable.\n\nworks to control \nthird-order modu- \nlation addition; \nsame noise objec- \ntives as in L4\n\ncoaxial units per sheath increased route capacity and further reduced \nper-channel-mile costs.\n\nFrom the outline of the evolution of coaxial-carrier systems in Table \nI, we can see that, in 33 years, the channel capacity of repeatered \ncoaxial line has increased by a factor of 22.5\u2014from 480 to 10,800 \nchannels, and further increases are anticipated. In the same period, \nimprovements in cable technology allowed the manufacture of cable \nwith 5.5 times more coaxial units in the cable sheath\u2014from 4 tubes to \n22 tubes\u2014resulting in a 10-fold increase in signal-carrying capacity, not \nincluding the two units reserved for service protection. The total \nimpact, then, was a 225-fold increase in route capacity. During the \nsame 33 years, Bell System circuit miles increased 485-fold.\n\nThis issue of The Bell System Technical Journal describes in detail \nthe latest in the line of coaxial systems\u2014the L5 Coaxial-Carrier \nTransmission System. The articles include descriptions of an advanced \nsystems approach and sophisticated repeatered-line and equalization \ndesigns. Also included are the novel concepts in repeatered-line power- \ning, line-protection switching, equipment-performance surveillance, \ncentralized maintenance, and carrier reference-frequency generation. \nOther articles describe the new multiplex and signal-administration \nequipment and the important role in the success of L5 played by \ninnovations in physical design and thin-film techniques and by the \nuse of ultralinear semiconductor devices. The many computational \naids and measurement facilities that were effectively used in the \ndevelopment of the system are also discussed.\n\nThe initial L5 system\u2014815 miles of cable, 14 stations, 850 manholes, \n3400 manhole repeaters, and over 250 bays of transmission equip- \nment\u2014went into service on January 38, 1974, fulfilling a schedule \ndeveloped six years earlier. This on-time completion of such a massive \nsystem required the dedicated effort of many individuals in the Bell \nSystem companies. Bell Laboratories people conceived and developed \nthe system and its components; Western Electric people were re- \nsponsible for manufacture of cable and electronic equipment and \nfor installation of main-station equipment; and AT&T and Long \nLines people were actively involved in system planning, coordination, \nroute selection, cable placement, installation of line equipment, and \noperational testing. It is to this skilled Bell System team that this \nissue is dedicated.\n\nCopyright \u00a9 1974 American Telephone and Telegraph Company \nTue Bet System TECHNICAL JOURNAL \nVol. 53, No. 10, December 1974 \nPrinted in U.S.A.\n\nThe LS Coaxial-Carrier Transmission System is a new solid-state \nbroadband system designed to transmit 10,800 long-haul message channels \non standard 32-inch (9.5-mm) coaxial cable. Alternatively, high-speed \ndigital signals may be transmitted in selected mastergroups. This paper \ncontains an overall system description and describes in detail system design \nconsiderations such as the transmission plan, frequency allocation, and \n4000-mile notse performance. Also considered are the significant proper- \nties of the coaxial medium, the ways they affect the hierarchy of equalizers \ntraditionally found in long-haul coaxial systems, and the manner in \nwhich the several levels of equalization relate to each other and to the \nfundamental transmission phenomena within the system.\n\nDuring the last 30 years, long-haul transmission facilities have been \nrapidly expanded to meet the increasing need for both voice and data \ntransmission, and it is expected that rapid growth will continue in the \n1970\u2019s although possibly at a slower rate than the 15 percent per \nyear which typified the late 1960\u2019s. Thus, during this period there \nhas continued to be a need for new systems providing increased \nmessage capacity at lower cost to supplement the systems already \navailable in the long-haul plant. As can be seen in Table I, the L5 \nCoaxial-Carrier Transmission System is the latest in a family of coaxial \ntransmission systems developed to date.*\n\nIt is evident that the newer coaxial systems have achieved increased \ncapacity because of better utilization of the transmission medium (i.e.,\n\n*The sheath capacities listed in Table I are the ones most commonly associated \nwith each of the systems. The initial L3 application, in 1953, was actually on Coax-8 \n(i.e., a cable which includes eight coaxials) with a total sheath capacity of 5580 \ncircuits on three working pairs, with one pair reserved for protection.\n\nLi L3 L4 L5 \nFirst service 1941 19538 1967 1974 \nTechnology Tubes Tubes Transistors | Transistors \nand HIC\u2019s \nRepeater spacing (miles) 8 4 2 1 \nChannels per pair 600 1,860 3,600 10,800 \nTotal channels/sheath 1,800 9,300 32,400 108,000 \nNoise objective (dBrnc0) 44 44 40 40\n\nwider bandwidth) and because of an increase in the number of coaxials \nincorporated within the sheath. An important characteristic of such \nlong-haul systems is the dominant effect which the installed cost of \nthe transmission medium has on the overall cost. By providing triple \nthe message capacity on L5 as compared to L4, for example, the cost \nof a message-circuit mile on L5 is substantially less than that of a \nmessage-circuit mile on L4, even though the number of repeater \nstations is doubled and the electronics required is of wider bandwidth \nand of increased sophistication. In addition, the older L-carrier systems \nuse the same transmission medium, and it is possible to convert many \nexisting L1, L3, or L4 routes to L5 by adding the required number of \nnew repeater sites to the existing sites of the earlier system. In this \nmanner, increased message capacity can be obtained at an incremental \ncost that is even more attractive than on a new installation.\n\nAs noted in Table I, the L5 coaxial system provides 10,800 toll-grade \nquality message channels (on a pair of 0.375-inch disc-insulated coaxial \ncables). In a Coax-22 cable configuration, the resultant total channel \ncapacity is 108,000 channels per cable sheath when a 1:10 protection \nstrategy is employed. The increase in the bandwidth of the system \nrequired for this capacity is made possible in part by setting the nominal \nrepeater spacing at 1 mile and in part by several significant techno- \nlogical advances in such areas as repeater, network, and device \ndesign.!-3\n\nThe Ld system includes fixed gain \u2018\u201c\u2018basic\u2019\u2019 repeaters at 1-mile \nnominal intervals, adaptive \u2018\u2018regulating\u201d\u2019 repeaters at 7-mile maximum \nintervals, adjustable gain \u2018\u2018equalizing\u201d repeaters at 38-mile maximum \nintervals, and power-feed main station repeaters at 75-mile maximum \nintervals.* Line protection switching is provided on a 1-for-10 basis at \nmaximum intervals of 150 miles.\u00ae\n\nThe adjustable equalization of the L5 repeatered line is partially \nmanual and partially automatic. Manually adjusted equalizers \ncompensate for the transmission deviations, which are essentially \ntime-invariant. These equalizers are adjusted to minimize the mean- \nsquared error of the broadband channel upon initial installation, and \nany subsequent readjustments are normally done out of service. The \nuse of out-of-service, mean-squared-error adjustment procedures \npermits a substantial improvement in the transmission quality for a \nparticular array of adjustable equalizers.\n\nAutomatic equalizers compensate for the time-variant transmission \ndeviations. These equalizers are located in each regulating, equalizing, \nand main station repeater and fall into two categories. The automatic \nequalizers for the cable-temperature effect are located in each of these \nrepeaters and utilize both open-loop regulating circuits, which detect \nthe resistance of a locally buried thermistor, and closed-loop regulating \ncircuits, which detect the level of a full-time line pilot. The automatic \nequalizers for the repeater-temperature effect and the regulating \nrepeater tracking error are located only in the receiving main stations \nand utilize closed-loop regulating circuits equipped with \u2018\u2018memory,\u201d\u2019 \nwhich detect the levels of four full-time line pilots.\n\nProvisions are included at appropriate main stations for adding, \ndropping, or branching any jumbogroup (3600 channels) at line \nfrequency. \u2018Dropping\u2019 involves connection to the jumbogroup \nmultiplex (Jmx), a new frequency-division multiplex (rpm) terminal \ndesigned for L5, which translates up to three basic jumbogroups \n(0.564 to 17.548 MHz) to appropriate line frequency assignments \nbetween 3.124 and 60.556 MHz.\u2019 \u201c\u2018Branching\u201d\u2019 involves connection, \nat line frequency, to or from another L5 facility. \u2018\u2018Adding\u201d\u2019 involves \nthe addition to an L5 line of circuits originating from the mmx. The \nmastergroups of a basic jumbogroup are administered much like the \nmastergroups of an L4 signal. The basic jumbogroup can consist of \nany available combination of analog and digital mastergroup signals. \nWhen frequency assignments permit, direct connection with other \nfacilities is made possible by connecting all mastergroups to the Jmx \nvia a basic jumbogroup trunk bay (BsetT). This trunk bay permits \nmastergroups from several different sources, such as mastergroup \nmultiplex (amx-2), L-carrier mastergroup digital (LMp) terminals, \nand radio systems, to be combined into a basic jumbogroup.\u00ae The \ncarrier generation within the ymx requires reference frequencies of an \naccuracy heretofore unavailable in the Bell System and consequently\n\na jumbogroup frequency supply (sFs) is included at all smx locations \nfor this purpose.\u00ae\n\nAt selected main stations, an L5 transmission surveillance center \n(tsc) will be utilized to centralize and partially automate the surveil- \nlance and fault location of several hundred miles of repeatered line \nand associated main station and multiplex equipment.\u2019 In the initial \nL5 installation between Lillyville, Pa., and Hillsboro, Mo., a single \nTsc serves these functions for 664 miles of \u2018\u201c\u2018backbone\u201d\u2019 route Coax-22, \nand 151 miles of \u2018\u2018sideleg\u2019\u2019 Coax-12.\n\nAll long-haul coaxial systems share the common objective of toll- \ngrade signal-to-noise quality in a 4000-mile circuit. In the L5 system, \nthis involves equalization of over 100,000 dB of cable loss at the upper- \nmost message frequencies and almost 132,000 dB at the uppermost \n\u201ccontrolled\u201d frequencies near 70 MHz. Equalization of this loss is \naccomplished with a combination of wideband circuits, all of which \nmust be low in noise, highly linear, and extremely stable. It is in the \nachievement of the 4000-mile transmission objective that the familiar \ncoaxial system hierarchy of increasingly complex repeaters evolves. \nAmong the general features of the system, the frequency allocation \nwill be considered first. Although many different factors must usually \nbe considered during the specification of a frequency allocation, only \na few can be reviewed here.\n\nThe controlling consideration in establishing the frequency allocation \nis the overall bandwidth involved\u2014both in terms of hertz of bandwidth \nand in terms of the ratio of uppermost to lowermost frequencies to be \nused\u2014which appears to be consistent with the state of the art at the \ntime, or within the evolutionary lifetime of the system development. \nAn important and relevant experience of previous coaxial-systems \ndevelopment has shown the difficulty and expense of simultaneously \ncontrolling the transmission characteristic at the highest and lowest \nfrequencies of a spectrum that spans a frequency range of as much as \nfive to six octaves. Realizable transformers, inductors, and capacitors \nare in most cases far from ideal components, and parasitic effects be- \ncome especially troublesome when the frequency range is more than a \nfew octaves. As the system bandwidth approaches five octaves or so, \nparasitic effects frequently become dominant. Thus, for a given band-\n\nwidth, there may be significant advantages to raising, within reasonable \nlimits, the lowest frequency requiring careful control.\n\nCompatibility with the existing rpm hierarchy is another relevant \nfactor and, in the case at hand, the basic building block was taken to be \nthe six-mastergroup assembly formed in accordance with the L4 \ntransmission plan. The six mastergroups are arranged between 0.564 \nand 17.548 M\u00a5Uz and are designated the \u2018\u201c\u2018basic jumbogroup.\u201d\n\nA basic system objective was the transmission of at least three \ntimes the L4 message capacity of a single jumbogroup. The assignment \nof the jumbogroup line frequencies and the spacing between the jumbo- \ngroups depends mainly on the bandwidth considerations already \ndiscussed and the types of signals and signal administration procedures \nplanned for the system. In L35, the capability to select any jumbogroup \nfrom the line spectrum at line frequency (\u2018branching and dropping\u2019) \nwas a basic system objective. Questions relating to the design of \nappropriate bandpass and band-stop filters substantially affect the \nguard spacing required between jumbogroups. In addition, at the time \nthat it was necessary to make the frequency allocation decisions, the \nability to displace an entire analog jumbogroup with a high-speed \ndigital signal was also a basic objective.\n\nThe frequency allocation of the L5 system is shown in Fig. 1, with \nsix mastergroups, each containing 600 message channels, making up \nthe basic jumbogroup. In some cases, a mastergroup in the L5 line \nspectrum will transmit a 13.29-megabits-per-second digital signal by \nmeans of the L-carrier mastergroup digital (LMp) terminal,\" a digital \nrepeater spaced at maximum intervals of 300 miles. These digital \nchannels will be used to carry a variety of digital signals including \nwideband data, pps, and digitized Picturephone\u00ae signals.\n\nThe regions around 1.59 and 68.65 MHz are used to transmit fault- \nlocating signals which in turn are used to remotely locate repeater \nfaults from a centralized location equipped with an L5 transmission \nsurveillance center.\n\nThe region around 68.78 MHz is used to transmit signals required \nby the line-protection-switching system (Lpss-3) for interstation com- \nmunication, and control and verification of Lpss operation.\n\nPilots are transmitted below, between, and above the jumbogroups \nto control the L5 automatic equalizers. The automatic equalizers for the \ncable-temperature effect, located in the regulating repeaters, are \ncontrolled by the incoming level of the 42.880-MHz pilot (which is \noften referred to as the \u2018\u2018temperature pilot\u2019\u2019). The automatic equalizer\n\nIN MHz 70.144(C) FL: FAULT-LOCATING SIGNAL \n42.496(C) E: EQUALIZING PILOT \nSW: LPSS TELEMETRY \n42.496(C) \n45.056(C) \n91.648(C) \n40.448(C) \nLLeo Oo \u00b0 SsSsom \nao + 8 aSBR Ss \neminent Se ; \u00a9 oo OO \n-- N Qa g oon oo \nI (goeood|Geoooo (feoooo I I \nS 88 8 5 : \n: & 8 8 8 8\n\nFig. 1\u2014L5 frequency allocation. The basic jumbogroup shown is equivalent to the \nL4 line spectrum. The carrier frequencies indicated are those used to translate the \nbasic jumbogroup to the L5 line frequency assignment.\n\nfor the repeater-temperature effect and the regulating repeater tracking \nerror, located in receiving main stations and designated the E3 equal- \nizer, is controlled by the incoming levels of all four pilots.\n\nAfter the establishment of at least an approximate frequency \nallocation, the central concept in the design of the coaxial system \nbecomes the equalization of the 4000-mile cable loss. Furthermore, \nit must be done in such a way that the residual transmission deviations \nare less than 0.01 percent of the top channel cable loss, while accum- \nulated noise from all sources, including associated multiplex, must sat- \nisfy the long-haul noise objective of 40 dBrncO. The general features \nof the system layout selected to meet these goals are described next.\n\nThe most general features of a repeatered-line layout in long-haul- \ncoaxial systems depend mainly on power-feeding techniques, equaliza- \ntion strategy, reliability requirements, and convertibility of older sys- \ntems. The equipment providing power for the remote repeaters, the fine- \ngrained equalization of transmission, and the automatic line-protection\n\nswitching will generally be of such size and sophistication as to require \na station environment as opposed to the manholes that are suitable \nfor the simpler repeaters. The cost of these main stations is many times \nthat of a manhole, and it is essential that the number of such stations \nbe minimized for system costs to be competitive with alternative ways \nof providing the circuits.\n\nThe power-feed configuration maximizing the length of the power \nspan for a given voltage is the double-ended connection shown in \nFig. 2.\n\nThe L5 line power supplies provide a nominal 0.91 A de via the \ncenter conductors of the coaxial cables for powering up to 75 miles \nof remote line repeaters. This is a convenient submultiple of the L4 \npower span length and requires a maximum line voltage of only \n+1150 V, with the result that physical designs of high reliability \nshould be readily achievable, and corona \u2018\u2018popping\u201d\u2019 in the installed \nsystem should be minimal.\n\nThe basic layout is affected by the equalization strategy mainly by \nthe effect the latter has on repeater signal loading and the extent of \nsignal level misalignment that must be accepted by the line repeaters. \nThe L5 strategy in most cases uses pre- and postequalization in equal \nparts and, as a result, repeater load requirements can be reduced for a \ngiven transmission deviation from nominal or, for a fixed repeater \nload capacity, the permissible distance between equalizers can be \nincreased, since the magnitude of the transmission deviation is usually\n\n= GRD PROT \n(OPERATES WHEN \nIMBALANCE IN TWO LINES \nEXCEEDS PRESET THRESHOLD)\n\nFig. 2\u2014Simplified diagram of line power feed. Plus or minus voltage is developed \nby dc-de converters and combined with message in power separation filters prior to \napplication to coaxial line.\n\nproportional to distance. Finally, when the misalignment is contained \nto within about +5 dB, it can be shown that this approach achieves \noptimum signal-to-noise ratio in both second- and third-order inter- \nmodulation-limited systems. Figure 3 shows the signal-to-thermal \nnoise penalties associated with single-ended equalization, at either \nthe receiver or the transmitter only, and with equal parts of pre- and \npostequalization.\n\nFinally, a relatively simple adjustable equalizer is included in the \nmidspan manhole repeater so that considerations of equalization and \nload capacity do not entirely determine the main station interval. \nThe midspan equalizer provides coarse adjustment of signal levels as \nthey traverse the power feed span, and not the fine-grain compensation \nof transmission deviations, which is accomplished only at the main \nstations.\n\nPre- and postequalization is applied throughout the L5 repeatered \nline except for the automatic equalization of the repeater-temperature \neffect, the controlling considerations in this case being simplification \nof system operation and reduction of system complexity. The repeater- \ntemperature effect is equalized only in the receiving portion of main \nstation repeaters, which are up to 75 miles apart.\n\nReliability affects system layout, since the reliability of the system \nelectronics determines the length of the protection-switching span and\n\nHO\u201d 38 6. \u00ab4 22 0 2 4 6 8 10 \nMISALIGNMENT 1N DECIBELS \nFig. 3\u2014Signal-to-thermal noise penalties resulting from signal misalignment. For \nmisalignment less than about +5 dB, pre- and postequalization in equal parts \n(curve 2) also results in negligible penalties in second- and third-order intermodula- \ntion noise (not shown).\n\nSWITCHING REGULATING MIDSPAN SWITCHING \nPOWER-FEED REPEATER (EQUALIZING) POWER-FEED POWER-FEED \nMAIN STATION \\ REPEATER MAIN STATION MAIN STATION\n\nBASIC \\ \nREPEATER __\u00a5 \nsee RECEIVING | tons RECEIVING \naan REPEATER | fora NeR REPEATER\n\nthe number of lines required as standby for any given availability \nobjective to be achieved. The predicted reliability of the L5 line \nrepeaters permits up to two power-feed spans to be traversed before \nautomatic line-protection switching is required, and then on a 1-for-10 \nbasis. In the context of the repeatered line, the basic building block \nof the system is the switching span and, in the absence of signal \nadministration needs (dropping, branching, etc.), such spans as are \nshown in Fig. 4 are connected in tandem until the distance between \nterminal stations is bridged.\n\nIn addition to the provision of repeatered-line power, fine-grain \nequalization, and (sometimes) line-protection switching, other im- \nportant functions are carried out at the different types of L5 main \nstations, and, depending on the particular features required of main \nstations, they fall into one of three categories. In order of decreasing \ncomplexity they are:\n\n(\u00a2) Terminal station/terminal main station (Ts/TMs). \n(21) Switching power feed main station (SPFMS). \n(i272) Power feed main station (PFMs).\n\nThe functions performed at these stations which are directly related \nto the repeatered line are alike. The differences among them are \nin the signal and pilot administration (line connecting) functions, \nthe line-protection switching, the transmission-surveillance capabilities, \nand the multiplexing and related signal-processing functions.* Some \nmain station and repeater functions are summarized in Fig. 4, and \nsome possible terminal interconnections for jumbogroup administration \nare shown in Fig. 5.\n\nA basic objective in the design of the L5 system was to achieve a \nsystem requiring a minimum of day-to-day routine involvement of \ncraftspeople. The L5 main stations, as well as the repeatered line, will \ngenerally be operated without personnel present. This has been made \npossible in part by developing equalization procedures that should \nrequire activity only rarely, and in part through the implementation \nof centralized alarm, surveillance, and fault-locating techniques. The \nlayout of the initial L5 commercial installation between Lillyville, Pa., \nand Hillsboro, Mo., is shown in Fig. 6 and, for purposes of the follow- \ning discussion of system organization and administration, is treated as \ntypical.\n\nInstallation of the L5 repeatered line is accomplished by telephone \ncompany personnel and is controlled and supervised at \u2018\u2018maintenance \ncenters.\u201d\u2019 Six such maintenance centers are shown in Fig. 6; the four \nat Danville, Xenia, Morgantown, and Noble have responsibility for \nall but the 30 repeater sections near Lillyville and the 14 repeater \nsections near Hillsboro. Those four stations are consequently each \nresponsible, on the average, for the installation and maintenance of \nnearly 200 miles of repeatered line which, when fully equipped, will \ninclude approximately 4000 line repeaters.\n\nUpon being received from the factory, every L5 repeater is carefully \ntested at one maintenance center before installation along the re- \npeatered line. The tests made involve primarily transmission measure- \nments that are executed, on site, by a computer-operated transmission \nmeasuring set and associated test stand. A hard copy of the test \nresults is produced during the process and, in the case of an acceptable \nrepeater, is retained in the maintenance center as part of the permanent \nhistory file on that repeater. Regulating repeaters and fault-locating \noscillators are among system components requiring on-site adjustment\n\nRECEIVING TRANSMITTING \n' REPEATER i = REPEATER | \n7 \n| \nCOAXIAL | / paranoid TRANSMITTING : | COAXIAL \nLine | |P BRANCH Plot LINE \nS EV/EZ/E3 | CIRCUIT circuit [23 heal E1/E2 S \nI |e S| \u201c(e2) an \n| RECEIVING TRANSMITTING \nEQUALIZER i) = EQUALIZER \nTO OTHER FROM OTHER \nL5 SYSTEM L5 SYSTEM\n\nFig. 5\u2014Jumbogroup administration. In the hypothetical case shown, sal is through-connected at line frequency and JsG2 is \nbranched on to another L5 system also at line frequency. Ja3 is dropped via JMx and the Byer to basic jumbogroup for interconnec- \ntion with mmx-2, L4, a radio system, or a combination thereof.\n\nBEAUCOUP KENTUCKY \nHILLSBORO MAIN ROUTE \u2014 664 MILES \nSIDELEGS | \u2014 151 MILES \nTOTAL \u2014 815 MILES \nMISSOURI (a)\n\n(2) HILLSBORO, MO. \n[>] BEAUCOUP, ILL. \n() NOBLE, ILL. \n[>]  SHELBURN, IND. \n(2) MORGANTOWN, IND \n[>]  CONNERSVILLE, \n? \n(2) S XENIA, OHIO \nw \n[>] HILLIARD, OHIO \n(>) DANVILLE, OHIO \n[3]  WAYNESBURG, \n(2) LILLYVILLE, PA. \nGy CINCINNATI, OHIO \nGY? covumaus, ox10 \nor DAYTON, OHIO \n<q WILLIAMSTOWN, KY\n\n(b) \nFig. 6\u2014(a) Geographical layout of initial L5 commercial installation between\n\nLillyville, Pa., and Hillsboro, Mo. (b) Administrative layout of Lillyville-Hillsboro \nroute.\n\nin accordance with the particular repeater station to which they are \nassigned, and those adjustments are normally made utilizing the same \ntest setups.\n\nInstallation of the L5 main station equipment is usually done by \nWestern Electric installation personnel and, prior to turnover of the \nequipment to the telephone company, a variety of transmission, noise, \nand operational tests is usually necessary. In most cases, these tests \nutilize the same kind of test equipment required by the telephone \ncompany to install, maintain, and operate the system.\n\nAfter the testing and installation of the repeaters and fault-locating \noscillators have been completed between a pair of power-feeding points \n(such as Lillyville and Waynesburg in Fig. 6), and after the main \nstation equipment has been installed and tested, power is applied by \noperating the line-power-feed dc-dc converters at both ends of the line. \nThe converters at each end are adjusted to share equally the load of \nthe intervening repeatered line (except in the case of very short power \nspans, which can be powered by converters at one end only). With \npower applied to the line, a detailed step-by-step process, known as \nthe L5 line acceptance procedure and appropriately documented, is \nexecuted, which permits the systematic identification and elimination \nof any system faults that may have escaped detection during the \nearlier testing of the individual repeaters and cable sections. In \naddition, such engineering or operational errors as incorrect line- \nbuild-out specification, incorrect fault-locating oscillator adjustment, \nor improper pre-regulator settings can, with a combination of engineer- \ning and on-site analysis, readily be detected. The line acceptance \nprocedure flows from a detection of the gross faults that may have \nexisted, through a regulator-by-regulator pilot level measurement \nsequence, to the measurement, evaluation, and equalization of the \nchannel between main stations, and ultimately to the noise loading \nof the tandem power-feed spans making up a protection-switching \nspan (such as between Lillyville and Danville in Fig. 6). Upon com- \npletion of this procedure, the repeatered line is ready for the next \nlevel of system testing, which would consist of such procedures as \nexercise and validation of the protection-switching system, including \ntandem operation (such as from Lillyville-to-Danville-to-Xenia in \nFig. 6) and multiplex-to-multiplex measurements and evaluation \n(such as between Lillyville and Xenia).\n\nA wide variety of status and alarm indications has been made \navailable in the L5 system design to provide for complete and ready\n\nremote verification of system health as well as diagnosis of system \ntroubles when they occur. Status indications are provided for such \noperating parameters as the dynamic equalizer range setting and the \ncondition of the protection-switching system, while alarm indications \nare developed based on the power level of any line or jmx pilot, the \ncondition of any fuse, and a variety of other parameters. These alarm \nand status indications are transferred to a central alarm station which, \nin Fig. 6, is located at Williamstown, Ky., off the L5 route. This station \nis part of the E2 Status Reporting and Control System and, in the \nfollowing discussion, references to E2 will be to this system, not to the \nE2 equalizer that is an integral part of the L5 system. Transmission \nbetween the Ld stations and the E2 central station takes place over a \ndata network which is a four-wire private line, meeting requirements \nsimilar to those of a standard voice-band data service. (In the case of \nthe initial L5 installation, the data facility is located internal to the co- \naxial cable, utilizing two of the loaded 19-gauge pairs, as far as \nCincinnati, and utilizes an external private line circuit between \nCincinnati and Williamstown.) Each L5 main station is equipped with \nat least one E2 \u201cremote unit,\u2019\u201d\u2019 which provides the interface between \nthe L5 system alarms and the E2 data facility. The larger Ld stations, \nsuch as Xenia and Morgantown on the initial route, require two E2 \nremote units, because of the quantity of equipment to be located there. \nThe total load on the Williamstown E2 central is 16 remote units, the \nupper limit for a \u2018manual\u2019 E2 central on a single data network.\n\nIn its normal operating mode, the E2 central sequentially interro- \ngates (alarm polls) each remote unit for its alarm poll response. When \nall remote units have been polled, the process is successively repeated. \nA manual operation at the central or a demand by the L5 transmission \nsurveillance system (Tss) for use of the data facility will interrupt this \notherwise automatic and repetitive alarm polling. The manual opera- \ntions at the central include calling up a particular station display for a \ndetailed alarm report as well as controlling such L5 system operations \nas effecting a line switch or operating the line-feed converters.\n\nsurveillance center (Tsc), located at Xenia. Equipped with a mini- \ncomputer central processor, and loaded with appropriate data-base, \noperating-system, and applications programs, that Tsc controls the \nfault-location and surveillance processes for all the repeaters and main \nstations in the initial system. (The total mileage represented by \nFig. 6 is approximately 815 miles. When fully equipped, the equipment \n\u201cdomain\u201d of the tsc at Xenia will include over 17,000 line repeaters \nand more than 250 transmit-receive bays, as well as associated multi- \nplex.) Virtually all the periodic maintenance measurements planned \nfor the system will be carried out at the Tsc and, by supplying a full \nduplex data circuit between Xenia and Williamstown, operation of \nthe tsc can be transferred to Williamstown during Xenia\u2019s unmanned \nintervals.\n\nThe transmission plan of an analog coaxial system is determined by \nadjusting the bandwidth, repeater spacing, and signal transmission \nlevels so that prescribed channel objectives are realized over the life \nof the system. To transmit over a particular bandwidth on a specified \ncable, the repeater spacing is determined by the available gain, noise \nfigure, load-carrying capacity and linearity of the repeater, and the \nmargins required to allow for cable and repeater variations resulting \nfrom aging, temperature changes, and manufacturing limits.\n\nTable IT illustrates how the basic repeater noise, load capacity, and \nlinearity design objectives became progressively more stringent as \ncoaxial-system design proceeded from L3 through L5. In comparing\n\nSystem Data L3 L4 L5 \nTop frequency (MHz) 8.3 17.5 60.6 \nRepeater performance, referred \nto top channel: \nNoise figure (dB) 1] 6.5 5.5 \nLoad capacity (dBm) 16 21 24 \nNonlinearity \nSecond order (dB) \u201461 \u201470 \u201470 \nThird order (dB) \u2014 96 \u2014 100 \u2014110* \nInsertion gain (dB) 44 33 31 \nTransmission level (dB) \u2014lil \u201414,2 \u201413.6\n\nthe repeater design parameters of the table, it is important to note \nthat the required noise, load capacity, and linearity improvements \nfrom L3 through L5 are achieved at progressively higher top channel \nfrequencies. Table II shows, for example, that operating at one-half the \nL4 repeater spacing and more than three times the L4 top channel fre- \nquency, the L5 basic repeater requires a 1-dB better noise figure, 3-dB \nmore load capacity, and 10-dB better third-order nonlinearity than \nthe L4 basic repeater, with all parameters referred to their respective \ntop channel frequencies. Top channel insertion gain is slightly lower in \nthe L5 repeater despite the higher top transmitted frequency because \nof the 2:1 reduction in repeater spacing. \u00b0\n\nThe L5 system signal-to-noise design objective is based on meeting a \n4000-mile per channel noise objective of 40 dBrncO0. The 40-dBrnc0 \ntotal is allocated as 39.4 dBrncO to the high-frequency line and 31.5 \ndBrnc0 to the terminals (Fig. 7 shows how the 31.5 dBrnc0 terminal \nallowance relates to a possible multiplex deployment). In the signal- \nto-noise design of the system, high-frequency line transmission levels \nare shaped with frequency to realize noise performance in every \nchannel as nearly equal as possible. In L5, second-order modulation\n\nJMX JMX JMX JMX JMX \nJUMBOGROUP \u2014\u2014 >\u00bb \n(3600 CHANNELS) \nMMX-2 MMX-2 MMX-2 MMX-2 MMX-2 \nMASTERGROUP- \u2014-> \\A \n(600 CHANNELS) t \nLMX-3(SG) LMX-3(SG) LMX-3(S6) \u00a7 EXPOSURES LMX-3(SG) \nSUPERGROUP. \u2014\u2014-\u00bb> \\A \n(60 CHANNELS) t \nLMX-3(GRP) LMX-3(GRP) 3 EXPOSURES LMX-3(G RP) \nGROUP- \u2014\u2014\u2014\u2014 > Ss \n(12 CHANNELS) | \nAG 3 EXPOSURES AG \nORIGINATE TERMINATE\n\nFig. 7\u2014Possible multiplex terminal deployment in a long-haul interconnection. \nThis example includes 11 intermediate multiplex operations between the originating \nand terminating stations, which is well above the number typically encountered.\n\nnoise is the controlling modulation contributor because of the use of \nthe phase-shaping networks discussed below, which serve to reduce the \nthird-order intermodulation distortion to a relatively insignificant \nlevel. In L5, application of signal level shaping techniques results in \nthe transmission levels shown in Fig. 8a and in the predicted zero \nlevel noise shown in Fig. 8b. The curve labeled \u2018\u2018Total\u2019\u2019 in Fig. 8b \nis the predicted noise under nominal conditions. The uppermost curve, \nlabeled \u2018\u2018Total (with allowances),\u2019\u2019 is the predicted noise performance \nincluding penalties for misalignment, misequalization, and the like. \nAs can be seen, the average noise over all channels is predicted to \nbe 39.7 dBrncO under nominal conditions. The use of frequency \nfrogging of the jumbogroups and mastergroups will tend to cause \nthe typical 4000-mile L5 channel to approach that average value in \nnoise performance. In addition, measurements on the L6 field trial \nand first commercial installations indicate close agreement with \n4000-mile predicted noise proportioned to the appropriate test system \nlength. A controlling factor in the determination of the allowable \nsecond- and third-order modulation coefficients listed in Table II is \nthe use of repeatered line phase-shaping networks in each Ld regulating \nrepeater. Such networks are used to produce a significantly less linear \nrepeatered line phase-frequency characteristic, resulting in reduced \ncorrelation of third-order (A + B \u2014 C)-type product addition from \nrepeater to repeater. In the absence of phase-shaping, such product \naddition would occur on a nearly in-phase or voltage-addition basis, \nwith the repeater-to-repeater terms approaching 20 log N, where N \nis the number of tandem line repeaters.*\n\nIn the presence of the phase-shaping networks, such as those used in \nL5, substantially lower correlation of (A + B \u2014 C) product addition \nresults. For the LS 6-mile regulating-section phase-frequency charac- \nteristic shown in the upper curve of Fig. 9, the repeater-to-repeater \naddition term within a frogging link is approximately 15 log N, cor- \nresponding to a per-frogging-link (A + B \u2014 C)-type product reduction \nof about 15 dB relative to the \u201c\u2018no-phase-shaping\u201d values. Thus, in Ld, \nthe use of phase shaping reduces third-order intermodulation noise to \nrelatively insignificant values, and the system is second-order modula- \ntion and thermal-noise limited.\n\n* Product addition according to a particular law occurs over \u201c\u2018frogging\u201d\u2019 link inter- \nvals of up to 800 miles. Because of the randomizing effect on product and source \nfrequencies produced by frogging, random or power addition is assumed to occur \nfrom frogging link to frogging link for all modulation product types.\n\nFig. 8\u2014(a) Repeater output transmission levels. The discontinuities in the TL shown result from using a stepped approximation to \nthe optimum (smooth) preemphasis curve, derived by transmitting Ja2 1.7 dB hotter than jel and ja3 1.8 dB hotter than 3q@2 at \nthe ymx output. (b) Predicted zero level noise for 4000-mile L5 system. The prediction is based on extensive individual repeater \nand transmit-receive bay measurements, and has been confirmed by noise loading measurements made on lengths of the initial \nsystem up to 750 miles long.\n\nFig. 9\u2014Relative phase (linear component subtracted) of L5 regulating section\u2014 \nwith and without phase-shaping networks in regulating repeater. The parameter } \nis a measure of the \u201c\u2018stiffness\u2019\u2019 of the \u00a2-f characteristic of the networks, and f, is \nthe frequency at which the phase shift of each of the four sections is 180 degrees.\n\nThe typical L5 cable (new installation) will be a Coax-22 and will \ncontain 22 disc-insulated coaxials and 42 polyethylene-insulated\n\nconductor pairs. (A cutaway view of a Coax-20 is shown in Fig. 10.) \nThe paired conductors in these cables are used for support systems \nsuch as order wire, fault location, and air-pressure telemetry.\n\nThe loss characteristic of the cable over the band of interest in L5 \ncan be approximated by\n\nThe first term of the right-hand side is a result of the resistance loss \nof the copper conductors and, at L5 frequencies, owing to skin effect, \nthis loss is proportional to the square root of frequency. The second \nterm reflects the dissipation in the polyethylene disc spacers that \nmaintain the relative positions of the inner and outer conductors, and \nit increases linearly with frequency. (This term is often called the \n\u201cpower factor\u201d term.) The third term reflects the temperature depend- \nence of the copper resistivity, and is also proportional to the square \nroot of frequency.\n\nAlthough the cable loss representation in eq. (1) is perfectly adequate \nover the transmission band of Ld and its predecessors, distortions in the \ncable loss will be encountered as system bandwidths are extended \nfurther, which in some cases may limit those bandwidths. For example, \nstructural return loss (SRL) spikes are encountered, which result from \nthe cable-stranding operation during manufacture. The strander \nperiodically deforms the coaxials as they precess about the cable \ncore, and, for a fixed stranding lay, the sru spikes are coherent and \nresult in relatively narrow-band transmission deviations from the \nbehavior predicted by eq. (1). The current lay length for the outer \ncoaxials in Coax-20 (cables having 20 coaxial tubes within the sheath) \nand Coax-22 is 36 inches +3 inch and results in sri spikes at 157 + 2.5 \nMHz. Other distortions are caused by seam interaction or \u2018back \ntwist,\u2019\u2019 which is also a product of the stranding operation. These occur \nin the outer coaxials of the Coax-20 and Coax-22 around 170 MHz. \nThe inner units of Coax-20 and -22 (8 in each case) and all the units of \nCoax-12, since they have shorter lay lengths, exhibit these effects at \nhigher frequencies than those noted for the 36-inch lay length. For the \npurpose of designing the hierarchy of repeaters and equalizers that \nmake up the repeatered line, the loss described by eq. (1) becomes the \nbasic objective for repeater gain per mile of system, as discussed next.\n\nFig. 10\u2014Coax-20 of serrated seam construction. This cable is the backbone of \nnearly all existing L4 routes, and includes 52 paper-insulated wire pairs in addition \nto the 20 coaxials. New L5 construction will normally utilize Coax-22, specifically \ndeveloped for L5.\n\nFigure 11 shows that the loss of 1 mile of 2-inch coaxial cable reaches \nabout 33 dB at 70 MHz. It accumulates to over 130,000 dB in 4000 \nmiles, and is compensated by a variety of equalizers located throughout \nthe system. When the transmission characteristic or effect being \nequalized is both predictable and time-invariant, fixed equalizers are \nappropriate, but when the effect to be equalized is either unpredictable \n(in a given installation) or time-invariant, some form of adjustable \nequalization is necessary.\n\n4.2.1.1 Nominal cable loss. The principal loss in the system that is \nboth time-invariant and predictable is the nominal cable loss, which is \ncompensated by the fixed-gain basic line repeaters. If the 4000-mile \nsystem transmission objective were to be met under nominal conditions \nby adequately matching the gain of the basic line repeaters to the loss \nof the adjacent cable section, a match would be required of approxi- \nmately 0.0008 dB at the uppermost L5 frequencies, which is neither \nrealistic with respect to design complexity nor consistent with measur- \ning and control capabilities of present-day manufacturing processes. \nThe principal transmission components of the basic repeater are a\n\n4.2.1.2 Deviations from nominal repeater spacing. When the repeaters can- \nnot be installed at the nominal 1-mile intervals, the resultant spacing \ndeviations are compensated for by the abovementioned LEO net- \nworks which simulate the loss of a length of the coaxial cable, and \nare available in 0.1-mile increments from 0.1 to 0.5 mile. Because \nof obstacles, access problems, or other geographical limitations, it \nis usually impossible to achieve exactly 1-mile spacing between ad- \njacent repeater sites. In such cases, the line repeaters are assigned \none of the available LBo\u2019s to increase the loss of shorter cable sections \nto that of 1 mile of cable. For unusual obstacles such as river crossings, \nfor example, a limited number of long repeater spacings may be \npermitted by requiring the repeater sections on both sides of the long \nsection to be shorter than nominal. The administrative rules for all \nthese cases are described in detail in engineering documentation \navailable to the telephone company engineer at the time of route \nlayout and route planning.\n\n4.2.1.3 Average design error. In previous coaxial systems, such as L3 \nand L4, the equalization of the average difference between line repeater \ngain and cable loss (i.e., the average \u201cdesign error\u2019) has been attempted \nwith families of fixed equalizers called \u201cdesign deviation equalizers\u201d\u2019 \nor simply \u2018\u2018deviation equalizers.\u2019\u2019 The functions of such equalizers are, \nfirst, to contain the transmitted signal levels at close to nominal \nvalues and, second, to enhance, in conjunction with all the other \nequalizers in the system, the ultimate transmission response of the \nsystem. While the first of these can often be guaranteed with relatively \nhigh confidence, the enhancement of the equalized transmission \nresponse is more difficult to achieve as a product of such equalization \nstrategy.1 This section discusses some reasons for this and some \nconsiderations involved in seeking an optimum overall set of equalizers \n(fixed plus adjustable).\n\nAs mentioned before, the initial design objective for a deviation \nequalizer is the complement of the average design error of the line \nrepeaters. As used here, this error is the difference, on the average, \nbetween the gain of the line repeaters and the loss of the associated \nnominal length of coaxial cable. Ideally, the basis for computing the \naverage difference would be the universe of all manufactured line\n\nrepeaters and cable. In reality, the basis must be the largest sample \nof each that is available before final manufacturing commitments \nmust be made on the equalizers.\n\nEven if the optimum information were available when desired, \nnetwork realization limitations leave a residual error requiring further \nequalization, and when the deviation equalizers consist of a family of \nseveral different designs (as in L4, for example) there are several \ndifferent realization errors. Naturally, the specific errors for different \nphysical models of the \u201csame\u201d network also tend to be different \n(because of what is usually called the manufacturing deviation). \nFinally, even if the realizations were ideal, the average design error \ncharacteristic of the repeaters used as a basis for the deviation equalizer \ndesign will not always be characteristic of the set of repeaters in any \nparticular line section.\n\nThe collective impact of these factors is a tendency to introduce \n\u201cripples\u201d into the frequency response of the channel. Even though \ncharacterized by smaller amplitude deviations than the unrippled \ndeviation, the rippled frequency response can often be more difficult to \nequalize. Of equal importance, the variations from line to line in the \nrippled characteristic resulting from the above factors will often be \nmore difficult to track with a selected set of adjustable equalizers than \nwould the line-to-line differences in the unrippled deviation (i.e., \nthat existing before the application of the deviation equalizers).\n\nTo the extent that these factors contribute significantly to the \nultimate residual error of an equalized channel, the transmission \nresponse of that channel would apparently benefit from minimizing \nthe number of different types in the family of networks, and/or \nminimizing the number of locations at which the deviation equalizers \nare inserted. The straightforward application of these guidelines leads \nto a single deviation equalizer design inserted as infrequently (along \nthe repeatered line) as possible, while the most extreme interpretation \nwould lead to no deviation equalizers as such, a condition that probably \ncould be achieved only with the addition of relatively costly adjustable \nequalizer stations, since the signal levels must still be contained \nwithin the repeater load capacity as the signal traverses the repeatered \nline. *\n\nIt is possible that the latter might be a desirable solution for a single \nequalizing section, which in L5 will be up to two power-feed spans.\n\n*The cost of a fully equipped L5 manhole equalizing station in a medium-cost \nzone is about twice that of a fully equipped regulating repeater station, which would \nnormally be used otherwise.\n\nHowever, when there is no attempt whatever (such as with deviation \nequalizers) to reduce the average misalignment over all equalizing \nsections to a value nearer zero, then all the adjustable equalizers in a \nlong circuit would tend to be set to the same side of their nominal \nsetting. As a result, there would be no tendency whatever to cancel \n(or randomize) the errors and transmission distortions originating in \nthe equalizers themselves and in their inherent equalizing limitations, \nand these would accumulate systematically in a long circuit, quite \npossibly becoming the major source of residual overall transmission \ndeviation. Short of this extreme, a solution based on a single network \ndesign and application in a minimum number of locations is attractive \nfrom several viewpoints, and such a solution has been incorporated \ninto the L5 design.\n\nThe L5 deviation equalizer design provides one-third the total \n\u201cdesign error\u2019 equalization required by a 66-mile repeatered line \n(approximate average power span length) in single networks located \nin the transmitting and receiving main stations and at the midspan \nequalizing station. The network has been designated the \u201822X\u201d \ndeviation equalizer, since the design objective is approximately the \ndesign error of 22 miles of L5 repeatered line. The particular method \nused to realize the deviation equalizer design is described in detail in \nRef. 1. As described there, the L5 deviation equalizer, while in principle \ncorrecting the average design error, is realized in the final design stages \nusing computer optimization routines that interact with the available \nadjustable equalizers. Figure 12 shows the signal misalignment as a \nfunction of distance, per 3 dB of 66-mile correction, employing the \n22X approach and assuming uniform accumulation of the misalign- \nment. As far as signal-to-noise penalty is concerned, it can be seen that \nthis arrangement is equivalent to pre- and postequalization, in equal\n\nFig. 12\u2014Equalization of the average design error. The relative signal level is \nshown for a case involving 3 dB of misalignment in 66 miles; three identical deviation \nequalizers are placed at the transmitter and the receiver, and midway between.\n\nX in miles \nTRMTG RPTR MIDSPAN RPTR RCVG RPTR \n55 to 75 Yes Yes Yes \n33 to 55 Yes No* Yes \n11 to 33 No * Yes \nLess than 11 No * No\n\nparts, of 44-mile sections having +1-dB signal excursion in tandem \nwith 22-mile sections having +0.5-dB signal excursion, each section \nincurring signal-to-noise penalties very near zero (see Fig. 3).\n\nNaturally, administrative rules are required to assist the telephone \ncompany engineer in the determination of his 22X equalizer needs. \nWith \u2018\u201c\u2018quantum steps\u201d of 22 miles of fixed equalization, it is possible to \nspecify the fixed equalization to within +22/2 miles of the ideal value, \nand the application rules specified for L5 are shown in Table III. Note \nthat execution of the tabulated rules should approach compensation, on \nthe average, of the design error in a long (multispan) system that oper- \nates at a uniform mean temperature. In so doing, the rules should result \nin slight over- or undercompensation of the individual equalizing \nsections which will be of varying lengths. This effect, along with the \nsignificant variations in mean cable temperature that normally will \nbe characteristic of long systems, should result in a tendency to \nrandomize the adjustable equalizer settings, thereby mitigating the \nsystematic accumulation of the errors and distortions originating in \nthe adjustable equalizers themselves, and in their inherent equalizing \nlimitations.\n\n4.2.2 Compensation of transmission effects requiring adjustable equalizers\n\nAs discussed above, those transmission effects that are ezther time- \nvariant or unpredictable (or both) require some form of adjustable \nequalization. Adjustable equalizers can be automatically or manually \nadjusted and the process can be carried out either in- or out-of-service, \nas particular circumstances and considerations warrant. In the case \nof L5, each of these possible types of equalizer and equalization \nprocesses is utilized among the various members of the equalization \nhierarchy.\n\n4.2.2.1 Cable-temperature effect. The largest time-variant transmission \ndeviation in a coaxial cable system, normally orders of magnitude \ngreater than any other, is the cable-temperature effect, shown in \nFig. 13 for 1 mile of 3-inch cable and for a 20\u00b0F change in cable tem- \nperature. Extrapolated to 4000 miles, the temperature dependence \ndepicted here corresponds to about +3000 dB of loss variation at the \nuppermost L5 frequencies, uniformly distributed over that distance. \nIn LS, this is corrected at intervals not greater than 7 miles in \u201c\u2018regu- \nlating repeaters\u201d? which, in addition to compensating for the nominal \ncable loss, provide automatic equalization of the cable-temperature \neffect. The regulating repeater compensates for the changing cable \nloss within a \u201c\u2018regulating section\u201d\u2019 which, in typical installations, has \nan average length of approximately 6 miles.\n\nIn addition to the components found in a basic repeater, the major \ntransmission components of the regulating repeater are a pair of \nadjustable equalizers having the adjustable loss frequency capability \nshown in Fig. 13 for the cable-temperature effect. The first equalizer \noperates in a closed-loop feedback mode and postequalizes for a portion \nof the preceding regulating section. The second equalizer operates in \nan open-loop mode by sensing the resistance of a thermistor buried in \nthe ground near the cable, thereby pre-equalizing for a portion of the \ncable-temperature effect of the ensuing regulating section. These two \nequalizers restrain the maximum level deviations resulting from the\n\nFig. 13\u2014Effect of changing earth temperature on 1-mile cable loss. The indicated \ntemperature range is only rarely exceeded within the continental U.S. at the cable \nburial depth of 4 feet.\n\ncable-temperature effect to about 2.5 dB within a regulating section \nthrough the use of the pre- and postequalization method. When the \nsystem is properly aligned, the pre-equalizer in one regulating repeater \nand the postequalizer in the subsequent regulating repeater each \ncompensate for one-half the cable loss variation of the intervening \ncable section. The remaining components and a more detailed descrip- \ntion of the regulating repeater operation are included in Ref. 1.\n\n4.2.2.2 Repeater-temperature effect. The other major time-variant trans- \nmission deviation is the repeater-temperature effect. This results \nfrom both the regulating repeater tracking error associated with \nequalizing the cable-temperature effect (a maximum of 0.10 dB per \nregulating section) and the direct effect of changing temperature on \nthe transmission responses of the various line repeaters (less than 0.02 \ndB per repeater section). Suitable allowances had to be made in the \nrepeater signal loading requirement to permit this effect to be equalized \nat the receiving end only of each power-feed span (up to 75 miles long). \nThis is accomplished in the \u2018\u2018dynamic equalizer,\u201d a cause-associated, \nadaptive equalizer with memory,* the important characteristics and \noperational features of which are described in Ref. 1.\n\n4.2.2.3 Statistical manufacturing deviation. The largest time-invariant \ntransmission deviation requiring adjustable equalization is the sta- \ntistical manufacturing deviation that results from the (usually small) \ndifferences in electronic component values within their specified \ntolerance ranges. Thus, the exact frequency shape of this deviation is \nnot predictable for any given collection of repeaters in a particular \ninstallation, although it can be predicted to fall within certain over- \nall limits. In L5, this deviation is compensated by equalizers El \nand E2 which are manually adjustable because the deviation is time- \ninvariant once identified. These equalizers consist of 28 adjustable \nBode-type equalizers (\u201cbump\u201d equalizers)\u201418 in the E2 and 10 in the \nEl. Both E2 and El equalizers are located in the transmitting and \nreceiving main stations, while only the El equalizer, mechanically \nadapted for manhole use, is utilized at the midspan equalizing re- \npeater. The deployment chosen for the equalizing stations is the result \nof seeking a reasonable balance between the amount of misalignment \nthat would have to be tolerated by the line repeaters, with resultant \nsignal-to-noise penalties and effects on repeater power requirements,\n\n*The memory in the dynamic equalizer permits the transmission response of the \nequalizer to remain unchanged if any equalizing pilots are lost.\n\nand the proliferation of the relatively complex (and relatively costly) \nequalizing repeaters that would otherwise be required to keep the mis- \nalignment and associated signal-to-noise penalties small.\n\nTo estimate the static misalignments to be encountered, it is neces- \nsary to know some transmission properties of the repeaters which, at \nthe outset of a system design, either are in the form of repeater require- \nments or, at times, are based on previous experiences with similar \ntransmission components. In the case of the L5 design, a requirement \nwas imposed on the line repeaters that (z) the average gain of the \nrepeaters match the nominal cable loss within +0.1 dB and (772) the \nspread among the gains of all manufactured repeaters would be less \nthan +0.1 dB with respect to that average. This, in combination with \nthe deviation equalizer strategy described in the preceding section, \nmade it possible to specify and design a single midspan (between \npower-feed stations) adjustable equalizer having both adequate range \nand adequate margin (against the possibility, for example, that the \nrepeater transmission requirements turned out to be unrealistic, or \ntoo costly to achieve, and might thus have been relaxed somewhat). \nAs mentioned earlier, this also resulted in the spacing of the relatively \nexpensive power-feed stations being limited by line power considera- \ntions, rather than by equalization and repeater load capacity \nconsiderations.\n\nThe selection of the type of equalizer terms, or \u2018\u2018shapes,\u2019\u2019 to be used \nand the number of terms in the equalizers, as well as the deployment \nof the shapes in the frequency domain, was based in part on computer \nsimulations of the L5 channel as it was estimated to be, and in part \non engineering judgments that evolved from experiences with earlier \nsystems such as L3 and L4. For example, it was judged possible to meet \nthe line repeater-cable loss matching requirement with about four or \nfive peaks in the matching error as a function of frequency. (Iive was \nthe actual number.) It was further reasoned that (7) the action of the \ndeviation equalizers, intended to complement the line repeater error \nfunction, would tend to double that number of peaks in the resultant \n\u201cfixed equalized\u201d characteristic; (77) the midspan equalizer, while as \nsimple as possible, should have a number of terms equivalent to the \nnumber of deviation peaks expected in the characteristic for which it \nwas to compensate (8 to 10); (iz) the action of the midspan equalizer \nwould tend to increase the number of deviation peaks in the resultant \n\u201cceoarse-equalized\u201d\u2019 characteristic by about the number of terms in \nthe equalizer (presumably, 8 to 10). This leads to a main station \nequalizer being required to deal with a transmission characteristic\n\nhaving potentially 16 to 20 peaks, whereupon the equalizer would \ncorrespondingly be required to have a similar number of terms. This \ntype of reasoning provided starting conditions for various computer- \naided optimization routines that ultimately resulted in the 10-term \nE1 equalizer and the 18-term E2 equalizer. Similar computer simula- \ntions were utilized to evaluate the effectiveness of such different types \nof equalizer shapes as the Bode-equalizer-derived \u2018\u201c\u2018bump\u201d\u2019 shapes (as \nused in L4) and cosine equalizers (as used in L3). In each simulation \nundertaken, a given number of terms could be deployed more effectively \nwith \u201cbumps\u201d than with alternatives, consequently leading to the \nselection of the \u201cbump\u201d\u2019 equalizers for L5. While the initial analyses \nand selection of equalizer characteristics had to be based on expected \nL5 channel performance, the final deployment of \u2018\u2018bump\u201d\u2019 shapes, \nas manufactured for the initial commercial application, was based on \nactual L5 repeatered line realizations in the L5 field trial. (The field \ntrial took place between Cedarbrook and Netcong, N. J., from mid-1970 \nto mid-1972, with each of four coaxial lines equipped and evaluated \nover a distance of approximately 100 miles. The trial resulted in a \nvariety of important system improvements. )\n\nThe manually adjustable equalizers are intended to be set upon \ninitial installation and only rarely thereafter\u2014for example, when a \nsubstantial increase in the amount of equipment operating in the \nmanholes results in significant changes in the annual mean operating \ntemperature of the repeaters. Readjustments of the El and E2 equal- \nizers under such conditions will restore the mean-annual setting of the \ndynamic equalizers (E3) to midrange, will improve the overall trans- \nmission response (since the 28-term E1/E2 combination is more effec- \ntive at equalizing the temperature offset effects than the four-term \ndynamic equalizer E3), and will improve signal-to-noise performance \nby reducing the maximum misalignment. Such readjustment would be \naccomplished out-of-service and would involve determination of the \nchannel misalignment at.a set of frequencies (three per adjustable \nterm) selected to permit optimum channel equalization in a mean- \nsquared-error sense. The misalignment of the channel is measured \nusing programmable transmission measuring equipment under the \ncontrol of what amounts to a special-purpose computer designated the \nEqualizer Adjustment Unit. The detailed implementation of equalizer \nadjustment is described in Ref. 1.\n\n4.2.2.4 Other factors affecting system transmission. For completeness, two \nother factors that result in time-variant transmission effects are\n\nmentioned here. One of these is the aging of the elements making up \nthe system. Aging effects usually are much smaller and much slower \nthan any effects already discussed. In a solid-state system designed \nwith careful attention to component selection and component stresses, \nthere is no aging mechanism comparable to that which was charac- \nteristic of the electron tube systems L1 and L3. In L5, aging effects are \nexpected to be small enough to be insignificant with respect to trans- \nmission quality or with respect to such operational aspects as equalizer \nadjustment intervals.\n\nThe second of these effects is that associated with maintenance \nwhich, along with the aging effect, is normally one of the smallest \ntime-dependent sources of change in the transmission response of a \nsystem. Over long periods of time there will inevitably be occasional \nrepeater failures with subsequent replacement of these repeaters by \nspares not having the same transmission response as the failed repeater. \n(Present estimates of repeater reliability suggest a mean-time-to- \nfailure in a maximum-length power span of 75 miles because of repeater \nfailure at between one and two years. Individual failures are, of course, \nprotected by the tess.) When they become large enough to be detected, \nboth aging and maintenance effects would normally be compensated \nby one of the occasional readjustments of state equalizers El and \nF2.\n\nWhen the characteristics of the L5 repeatered line are examined, \nit is evident that it shares many features with previous systems, and \nthus in some ways is a hybrid offspring of the L3 and L4 predecessors. \nFor example, while the adjustable equalizers El and E2 consist of \nbump shapes like L4, the adjustments are made out-of-service and \nin such a way as to minimize the residual mean-squared error of the \nequalized channel, like L3. The cable-temperature effect is compen- \nsated by using both open- and closed-loop regulators which pre- and \npostequalize the changes in cable loss in equal parts, like L4. But \nthe 22X deviation equalizer strategy of L5 is much more akin to \nthe L3 approach. (In L4 there are five different designs, one of which \nis installed in every regulating repeater.) While all three systems \nutilize line-build-out netowrks to provide some spacing flexibility, \nthey have fault-locating systems that are quite different. And only \nthe L5 system supplements the basic fault-location system with \na centralized, computer-operated transmission surveillance system\n\ncapable of measuring and monitoring a wide variety of system oper- \nating parameters.\n\nAt the time of this writing, the development of the L5 system in its \ninitial configuration had been completed with all important noise and \ntransmission objectives having been met. Continuing efforts in the \nsoftware offerings of the transmission surveillance system are expected \nto increase the already considerable capabilities for automatic and \nsemiautomatic system maintainance, fault location, and trouble di- \nagnosis. A modest level of continuing effort involving the optimiza- \ntion of deviation equalizer designs and modification of equalizer \ndeployment should further improve the system transmission response \nand reduce system noise.\n\nEffort is also under way to replace the existing mastergroup multi- \nplex equipment (mmx-2c) with individual mastergroup translators, \nresulting in reduced noise contribution from that level of multiplex \nat lower cost. The possibility of alternative protection strategies for \nthe jumbogroup level of multiplex are also being examined. In addition, \nthe basic frequency allocation described in this paper is being reviewed \nwith an eye toward possible spectrum reorganizations that would \npermit an increase in L5 capacity to 22 mastergroups using a new \nfamily of mastergroups and multimastergroup multiplex equipment.\n\nFinally, questions continue to be raised with respect to future \ncoaxial systems\u2014whether or what or when they may be. Although \nadvanced coaxial systems having substantially greater capacity than \nL5 seem clearly to be within the technological capabilities of the \nrelatively near future, the need and economic desirability of such \nsystems has not yet been established. Whether or not L5 becomes the \nlast of the major long-haul coaxial developments will be determined \nin the years ahead primarily by considerations of circuit growth \npatterns and economic conditions and by the relative viability and \navailability of alternative instrumentalities such as possible new micro- \nwave radio, millimeter waveguide, and optical fiber systems.\n\n1. E. H. Angell, Y. S. Cho, K. P. Kretsch, and M. M. Luniewicz, \u2018\u201c\u201cL5 System: \nRepeatered Line,\u201d B.S.T.J., this issue, pp. 1935-1985.\n\n2. F. A. D\u2019Altroy, R. M. Jacobs, J. M. Nacci, and E. J. Panner, \u2018\u201cL5 System: \nUltralinear Transistors,\u201d B.S.T.J., this issue, pp. 2195-2202.\n\n3. J. L. Garrison, A. Olsen, Jr., and T. H. Simmonds, Jr., \u201c\u201cL5 System: Transmission \nNetworks and Magnetic Components,\u201d B.S.T.J., this issue, pp. 2203-2248.\n\n5. J. H. Green and R. W. Sanders, \u2018\u201cL5 System: Line-Protection Switching,\u201d \nB.S.T.J., this issue, pp. 2011-2034.\n\n6. R. K. Bates and D. J. Zorn, \u201cL5 System: Signal Administration and Inter- \nconnection,\u201d B.S.T.J., this issue, pp. 2129-2145.\n\n8. J. F. Barry, 8. Narayanan, and J. F. Oberst, \u2018\u201cL5 System: Jumbogroup Frequency \nSupply,\u201d B.S.T.J., this issue, pp. 2109-2127.\n\n9. J. L. Thomas, R. E. Anderson, and P. J. Baun, \u2018\u201cL5 System: Centralized Trans- \nmission Surveillance,\u201d B.S.T.J., this issue, pp. 2035-2064.\n\n10. J. F. Gunn, J. S. Ronne, and D. C. Weller, \u2018\u2018The Pzcturephone\u00ae System: Master- \ngroup Digital Transmission on Modern Coaxial Systems,\u201d B.S.T.J., 50, No. 2 \n(February 1971), pp. 501-520.\n\n11. Bell Laboratories Staff, Transmission Systems for Communication, Fourth Edition, \nrevised, pp. 331-333.\n\nThe Ld repeatered line is presented from the viewpoint of a distributed \nequalization process. Reliable transmission of 10,800 circuits over 4000 \nmiles of coaxial cable with minimum noise is the sole objective of this \nprocess. The strategy 1s to provide equalization in cause-associated incre- \nments that place specrfic bounds on signal-level excursions, thereby in- \nsuring an ultralinear, low-noise predictable transmission medium. The \nindividual causes of signal misalignment, both static and dynamic, are \nexamined and the realization of the strategy, which forms a hierarchy of \nequalizers, 1s described.\n\nThe basic function of the Ld repeatered line is to provide a lossless \ntransmission facility between any two L5 terminal stations, which \nmay be located as far as 4000 miles (6400 kilometers) apart, with \nminimum noise penalty. The objective is to establish and maintain \nthe insertion loss of the 4000-mile line over the message band to be \nwithin +4 dB and the noise to be less than 39.4 dBrnc0 in any of the \n10,800 channels transmitted.\n\nThe transmission medium for the L5 line is the standard Bell System \n0.375-inch disc-insulated coaxial cable. This cable has a loss* that can \nbe conveniently expressed as follows:\n\n* A 2.1-percent factor has been included in this loss to account for miscellaneous \nfactors such as stranding.\n\nThere are two frequency bands of interest. The first is the message \nband from 3.1 to 60.6 MHz over which the 4000-mile objectives must \nbe met. To meet these long-haul objectives, however, and to accommo- \ndate other system functions, a wider band from 1.6 to 66 MHz is \ncontrolled. In eq. (1), the loss associated with the first bracketed term \nis a function of frequency only. This is called the static loss. The second \nbracketed term is not only a function of frequency, but also of cable \ntemperature and, hence, of time. This component of cable loss is called \nthe dynamic cable loss.\n\nThe strategy to equalize both static and dynamic losses is to provide \na cause-associated equalization hierarchy\u2014that is, the various equip- \nment making up the L5 repeatered line each compensate for a particular \ncause of line misalignment. This relationship is shown in Fig. 1.\n\nThis paper discusses the misalignment causes and the associated \nequipment design and realization.\n\nStatic equalization of the cable is provided in three levels. The first \nis provided by the basic repeater. With the repeaters spaced at 1-mile\n\nintervals, the gain required to match the static component of the cable \nattenuation given in (1) varies from 4.97 dB at 1.6 MHz to 32.04 dB \nat 66 MHz.\n\nThe second level is provided by the deviation equalizer. It is neither \neconomically nor technically feasible to design the basic repeater \nwith a gain response that exactly matches the cable loss function. \nThe realization of the repeater design results in a slight mismatch \nbetween the average basic repeater gain and the average cable loss. \nThis mismatch is referred to as the average design error. An objective \nwas established to hold this error to less than +0.1 dB over the \nfrequency band in a 1-mile section or to less than +7.5 dB in a 75-mile \npower-feed section. Compensation for this misalignment is provided \nby the deviation equalizer located at the transmitting and receiving \nends of a power-feed section and in midspan.\n\nEven if it were possible to exactly match the nominal attenuation \nof the cable by the gain of the basic repeater, there would still be \nappreciable misalignment in line sections involving many repeaters. \nThis is due to the statistical manufacturing deviations in repeaters \nand cables, and is impossible to predict for a given line section. The \nresulting misalignment from both causes is referred to as static line \nmisalignment or static deviation. Equalization of this static deviation \nforms the third level of static equalization in the L5 system, and is \naccomplished with adjustable equalizers referred to as El and E2 \nequalizers.\n\nSpecific causes for the statistical manufacturing deviations are \nknown. Repeater gain deviations are the result of component and \nassembly tolerances, while cable loss deviations are the result of \nvariations in the copper conductivity, cable geometry, and dielectric \ndisc conductivity. Copper conductivity and cable geometry influence \nthe A parameter of the cable loss equation and typical variations (one \nsigma) resulting from this effect are 6 dB at 66 MHz over 75 miles. \nDielectric disc conductivity, on the other hand, affects the B parameter \nof the cable loss equation and amounts to a 1.5-dB change for a one- \nsigma variation at 66 MHz over 75 miles. However, changes in disc \nmaterial have occurred since the initial cable production some 25 \nyears ago. Measurements made on a number of these earlier cables \nindicate a variation in cable loss of about 23 dB at 66 MHz per 75 \nmiles. Since, in addition to new installations, L5 is also intended as a \nretrofit system for some of these earlier cable applications, this varia- \ntion becomes important.\n\nEquation (1) shows that the cable attenuation varies as the tempera- \nture of the cable changes. In the L5 system, the cable is buried four \nfeet beneath the ground surface, where the maximum temperature \ndeviation expected in the United States is +20\u00b0F. This results in a \n52-dB change in cable attenuation at 66 MHz for 75 miles. Such \nchange in attenuation is compensated for by repeaters that are called \nregulating repeaters. These repeaters automatically compensate for \nchanges in the cable attenuation by sensing the temperature of the \nearth at cable depth and by detecting the level of a pilot tone which is \ninserted into the message signal at L5 main stations.\n\nThis is not the only component of dynamic misalignment that must \nbe equalized, however. Since it is not possible to match exactly the \ndynamic change of the cable over the entire frequency spectrum, there \nwill be a slight misalignment or slight difference between the change \nin gain of the regulating repeater and the change in loss of the cable, \nreferred to as the tracking error. In addition, slight differences occur \nbetween the change in gain of the basic and regulating repeaters \nthemselves as the temperature of their environment changes. These \ntwo effects, taken together, are referred to as the residual dynamic \ntransmission deviation. While over any short-line section this deviation \nis very small, the accumulated effects over a 75-mile section are \ntypically +3 dB at 66 MHz. This effect is automatically equalized \nby an E3 equalizer. The time-varying transmission deviations of an \nLd line are detected by four pilot tones spaced across the line frequency \nspectrum. Four networks in the E3 equalizer automatically respond \nto the pilot levels to correct the residual dynamic transmission \ndeviation.\n\nThe function of the basic repeater is to provide a fixed gain to \ncompensate for the attenuation of 1 mile of coaxial cable. Its charac- \nteristics have the dominant effect on the overall system performance. \nThus, the requirements it must meet are stringent to assure that the \noverall 4000-mile system objectives are met. These requirements cover \nnumerous areas such as specification of gain match to cable loss, \nintermodulation distortion, power-handling capacity, noise figure,\n\nreturn loss, and temperature coefficient. The simultaneous achievement \nof all these specifications is a major accomplishment in the design of \nthe L5 system.\n\nTwo of the most important design requirements are (2) that the \naverage gain of the basic repeater must match the loss (at 55\u00b0F) of one \nmile of coaxial cable over the message band to within 0.10 dB and \n(27) that the two-sigma limit on the distribution of repeater gains must \nnot differ from the average characteristic by more than 0.10 dB. These \nrequirements necessitate an amplifier design with an overall square- \nroot-of-frequency gain shape of 6.91 dB at 3.1 MHz to 30.69 dB at \n60.6 MHz.\n\nThe second- and third-order interchannel modulation distortion \n(hereafter referred to as intermodulation noise) of the repeater must \nbe extremely small. The second-order intermodulation coefficient \nobjective varies from \u2014105 dB at the low end of the spectrum to \u201470 \ndB at the high end. The third-order intermodulation coefficient \nobjective varies from \u2014128 dB at the low end to \u2014110 dB at the high \nend. These objectives are derived from calculations involving the \nrepeater noise figure, output spectral density of the signal, and the \nassumed law of addition, from repeater to repeater, for intermodulation \nnoise.\n\nTo avoid a rise in system noise because of peak busy-hour message \nload, the basic repeater must be capable of sustaining a load of +24 \ndBm before overloading. For the L5 system, the conservative definition \nof overload is used as the point where the modulation coefficient \ndegrades by 0.5 dB. An ultralinear high-power amplifier is obviously \nrequired as the result of the overload and intermodulation objectives.\n\nNoise figure of the repeater obviously affects system signal-to-noise \nperformance and therefore should be kept to a minimum. The noise \nfigure objective is 8.5 dB at the low frequencies and slowly decreases \nto 5.5 dB at the high frequencies. Contrary to overload and inter- \nmodulation, this requirement calls for a low thermal noise amplifier. \nAdditional objectives that influenced the amplifier design were surge \nprotection, return loss, and temperature coefficient.\n\nTo summarize, the basic repeater objectives dictate a shaped-gain, \nlow-noise, high-power ultralinear repeater. One method of achieving \nthe shaped-gain requirement would be to design a flat-gain amplifier \nand a separate network for the shaped loss. The shaped loss would be \nequal to the maximum gain required minus the minimum gain required\n\n(for signal transmission). This amount of loss (25.13 dB), if placed at \nthe input, would lead to an excessively high noise figure at low fre- \nquencies; if placed at the output of the amplifier, this loss would \nexcessively penalize intermodulation distortion and overload. On the \nother hand, these penalties may be avoided by designing the amplifier \nwith a square-root-of-frequency shape (Vf) network in the feedback \npath, thus creating a shaped-gain amplifier. As a result, additional \nnegative feedback is available; it improves the intermodulation dis- \ntortion and gain deviations resulting from u-path variations and \nimproves return loss. However, the shaped-gain feedback amplifier \nmust be properly compensated to avoid high-frequency stability \nproblems.\n\nThe realization of the basic repeater includes two feedback amplifiers \nand a number of passive networks shown in block diagram form in \nFig. 2. Both amplifiers have shaped loss networks in their feedback \npaths to accomplish the Vf gain required. Details of the amplifiers \nand other networks are given in the sections that follow.\n\nThe twin jacks, shown at the repeater input and output, are located \nbehind the mounting surface for the repeater in the manhole. Their \npurpose is to provide connection from the coaxial line to the repeater \nwith an additional port exposed for outside connection. The outside \nconnection serves two purposes. First, it allows injection of fault- \nlocating tones through bridging pads to the repeater input and output. \nSecond, it allows for \u2018power patching\u201d of repeaters. This is done by \nremoving the bridging pads and replacing them with a coaxial cable \npatch. The repeater may then be removed and replaced without \ninterrupting the coaxial line de power. This simplifies and expedites \nrepeater replacement.\n\nThe bridging pad has an impedance of about 2000 ohms facing the \ntwin jack. Since it connects with a coaxial plug, there is also a parasitic \ncapacitance to ground. To maintain the 75-ohm impedance required \nat the repeater input and output, an inductor was designed into the \ntwin jack. This is shown schematically in Fig. 3. By choosing Z such \nthat VL/C = 75 ohms, a lumped section of 75-ohm coaxial cable was \napproximated, and the impedances were maintained. The shunting \neffect of the 2000-ohm resistor is compensated for by the impedance \nof the low-frequency networks that are discussed in Section 2.33.\n\nCOAXIAL CABLE USED IN POWER PATCHING \nIe ee ee ee ee ee we ee we ee we ww os we ae ae ~~\n\nThe effect of the twin jack and bridging pad are included in the \nperformance specifications of the repeater.\n\nThe earth-ground filters, shown at the repeater input and output, \nconsist of sections of high-voltage capacitors and ferrite beads which \nform a filter. They provide transmission of signal and power through \nthe outer wall of the repeater (which must be kept at earth ground \nfor safety reasons) to an inner \u201cfloating ground.\u201d Signals developed \nacross the high-voltage output capacitors are then sufficiently attenu- \nated by the filter so as not to couple into the input high-voltage capaci- \ntor, and vice versa. Failure to provide this filtering action would result\n\nin signals coupling from output to input, which produce unequalizable \nripples across the frequency spectrum of the repeater. The earth- \nground filters are the only high-voltage components in the repeater ; \nall other components operate with respect to the internal \u2018\u2018floating \nground.\u201d\n\nThe two low-frequency networks (see Fig. 2) differ slightly in internal \ncomponent choice, but perform essentially the same functions. The \noriginal purpose of the low-frequency networks was to provide some \nloss at the very low frequencies of the L5 spectrum to compensate for \nexcess gain of the amplifiers. However, during the evolution of the \nrepeater design, other functions were added to the low-frequency \nnetworks. A simplified schematic of a low-frequency network is shown \nin Fig. 4. The network, comprised of Z1, Z2, and ZO, provides the \nlow-frequency loss shaping previously mentioned. Resistor R1 provides \nboth an impedance match to 75 ohms (in conjunction with the 2000 \nohms in the bridging pad) and a current limiting function for surge \nprotection. It has a small flat loss effect on the repeater response, \nwhich is compensated for by the main amplifiers.\n\nG1 is a gas tube surge protector having a de voltage breakdown of \nabout 90 volts and, under normal circuit conditions, introduces only \n1.5 pF of shunt capacitance. Inductor L1 provides for the separation \nof the dc line current from the L5 message signal. It maintains a high \nreactance over the frequency band to keep signal losses to a minimum.\n\nThe line-build-out network (LBO) is a passive network that provides \nloss equivalent to a given length of coaxial cable. It is used to reduce \nthe repeater gain from the equivalent of 1.0 mile of cable down to \n0.5 mile in steps of 0.1 mile. This allows flexibility for those cases in \nwhich the sections of cable are shorter than 1.0 mile (usually for easy \naccess to the repeater location or because of a physical obstruction in \nthe location of the manhole). Since the line-build-out network is an \nintegral part of the basic repeater, a separate repeater identification \n(code) must be associated with each of the six line-build-out networks. \nThis is in contrast to earlier L-carrier systems in which the line-build- \nout network was provided to the field as a separate network. The \nadvantage of this approach is that it allows for precise control of the \ngain interaction between the line-build-out network and the input \nimpedance of the power amplifier.\n\nTo achieve the required gain, noise figure, and distortion require- \nments of the basic repeater, new ultralinear wideband transistors were \nrequired. The characteristics of these transistors and other transistor \ndesign considerations are covered in a companion article.?\n\nAs previously mentioned, each amplifier uses negative feedback. \nGood stability of the amplifier requires minimization of the length of \nthe w8 path, which was achieved by employing thin-film technology \nwith appliqued components, yielding a tri-level circuit realization: \n(see Fig. 5). The three layers thus formed are comprised of (7) thin-film\n\nresistors and metal system interconnections, (77) discrete leaded \ncomponents (mostly inductors, transformers, and transistors) and \n(217) leadless ceramic chip components (capacitors and thick-film \nresistors). This tri-level circuit minimizes the w8 path length and \nincreases the available feedback. In addition, since thin-film resistors \nand ceramic chip capacitors have no external leads, parasitic effects \nare minimized, thus improving the circuit performance at the high \nfrequencies required in the L5 system.\n\nThe feedback networks of both amplifiers required precision com- \nponents to achieve the required gain characteristic. An initial tolerance \nof +0.1 percent is maintained with the thin-film-shaping resistors, \nand selection limits of +0.03 dB flatness across the L5 spectrum are \nrequired on the transformers.\n\nOther details of the physical realization of the repeater are discussed \nin more detail in a companion article.\u2019\n\n2.3.5.1 Preamplifier. The preamplifier is a negative feedback, hybrid- \ninput, hybrid-output amplifier depicted in Fig. 6. This circuit configura- \ntion was chosen, after careful computer simulation of various alter- \nnatives, for its low-noise figure while still maintaining good inter- \nmodulation performance and stability. The input and output hybrid \ntransformers have impedance ratios of 75:65 + 28 ohms. This ratio \nwas chosen to minimize the amplifier noise figure, while still maintain- \ning physical realizability and reproducibility of the hybrid. Decreasing \nthe \u201c\u2018through loss\u2019\u201d\u2019 of the hybrid to decrease the noise figure results \nin increasing the loss through the uf ports of the hybrids, which then \nresults in an increased minimum gain of the amplifier. It is this mini- \nmum gain, resulting from the hybrid effect, that is compensated for \nby the low-frequency networks.\n\nThe input stages were chosen as common emitter, biased at 30 mA \nand 5 volts, to achieve a low noise figure and to provide sufficient \ngain to minimize the effect of the noise figure of the output stages on \nthe overall amplifier noise figure. The output stages are biased at 110 \nmA and 12 volts for optimum intermodulation performance. Both input \nand output stages consist of two transistors connected in parallel to \nreduce the intermodulation noise as follows. The input signal current \ndivides equally, and each transistor then carries one-half the total \ncurrent. The controlling third-order nonlinearity is primarily current- \ndependent and, as a result, the output distortion of each transistor is \nreduced to one-eighth [($)*]. When combined from both transistors, \nthe output distortion is one-fourth that of a single stage, or a 12-dB \nimprovement. Similarly, for second-order distortion, a 6-dB improve- \nment is achieved. To achieve these improvements, the transistors are \npaired by matching their current gain.\n\nTwo adjustable elements are in the preamplifier\u2014a capacitor and a \npotentiometer. Each has a very limited range and is used for factory \nadjustment of the gain response, particularly at high frequencies. \nComputer sensitivity runs were used to choose the variable elements for \nmatching the normal amplifier gain shape variation. The capacitor \naffects the balance between loop and local feedback and hence in- \nfluences the closed loop gain through the yf effect, which varies the \nrepeater gain only at the very high frequencies. The potentiometer \nis in the bridged-T 6-network. Changes in its resistance directly vary \nthe 6 loss and, hence, the amplifier gain. This adjustment changes a \nbroad, high-frequency shape and is made after the total repeater has \nbeen assembled, to set the gain at 42.880 MHz (the temperature pilot) \nto within +0.01 dB of the nominal value. This precise gain adjustment\n\nminimizes gain error introduced by the basic repeater at the tempera- \nture pilot frequency and therefore reduces the range requirements of \nthe regulating repeater.\n\n2.3.5.2 Power amplifier. The most difficult objective to meet in the \nbasic repeater was the intermodulation distortion. After an extensive \nprogram of computer modeling of transistor and circuit nonlinearities\u00ae \nand laboratory evaluation, the circuit topology shown in Fig. 7 was \nchosen for the power amplifier. It has shunt feedback at the input \nand hybrid feedback at the output. The \u00bb path is a common emitter- \ntransformer-common-base arrangement. The interstage transformer \nprovides current gain to minimize the effect of distortion in the \ncommon emitter stage and also provides a more nearly optimum set of \ninterface impedances between the stages. An autotransformer was \nchosen as the output hybrid to obtain maximum bandwidth and min- \nimum phase shift.\n\nAs in the case of the preamplifier, all transistors are paralleled for \nimprovement in distortion. The bias point of 15 volts. and 110 mA \nwas chosen to minimize the third-order distortion.\n\nComputer simulation revealed that the closed-loop gain at high \nfrequency was sensitive to the capacitance at the common base output \nnode. This node capacitance varies as a result of variation in transistor \ncapacitance, surge diode capacitance, choke winding capacitance, and \ncircuit parasitic capacitance. Because these capacitances cannot be \ncontrolled to the desired tolerance, an adjustable capacitor was added.\n\nTo meet the distortion requirements, the loop feedback was maxi- \nmized. To maintain stability margins at the maximized feedback, an \ninput hybrid was not used. The input network serves to form part of the \noverall amplifier gain shaping, but at the expense of presenting a \nnonconstant impedance to the preamplifier (or LBo, if used). This \nimpedance mismatch creates a gain interaction when line-build-out \nnetworks are placed between the preamplifier and the power amplifier. \nTo offset this interaction, power amplifiers intended for use in repeaters \nwith line-build-out networks are adjusted to a different nominal gain.\n\nBuried cable systems, although shielded, are still subjected to a \nnumber of transients induced from within and without the system. \nThese include (2) high-voltage line turnup,\u00ae (22) power patching and \nsubsequent insertion of an uncharged repeater into the high-voltage \nline (227) lightning, and (zv) 60-Hz induction. To protect the repeater\n\nOne performance feature that must be determined very early in the \ndevelopment of a coaxial system is the match of the gain of the line \nrepeaters to the loss of the cable. This information 1s necessary for the \ndesign of the deviation equalizer, which is described in the next \nsection. Although prototype information was available on the degree \nof match, the results of actual production runs with final component \ntypes are required to establish a sufficiently accurate frequency \nresponse of the repeaters. These data were obtained with the coopera- \ntion of the manufacturer, Western Electric, from measurements of the \nearly production repeaters by Bell Laboratories personnel. The details \nof the accuracy and environment of these measurements are described \nin a companion paper.\u2019 The results of these measurements, when \ncompared to the installed line measurements, established a high degree \nof confidence in the reproducibility of the design and in the validity \nof the parametric equation describing the cable loss. The results also \nprovided repeater temperature coefficients which were required for \nthe design of the network shapes in the E3 dynamic equalizer.\n\nThe following illustrations demonstrate some performance features \nof the basic repeater. Figure 8 illustrates the match to cable loss and Fig.\n\nFig. 8\u2014Average deviation of basic repeater gain from the average loss of 1 mile of \ncoax cable.\n\n9 represents typical input and output return loss. Overload and temp- \nerature coefficients are shown in Figs. 10 and 11, respectively. The \nnoise figure of a basic repeater is plotted in Fig. 12. Second- and \nthird-order modulation are depicted in Fig. 13.\n\nIt has been pointed out that exact equalization of the static attenua- \ntion component of the cable loss is not feasible. The design trans- \nmission objective of the basic repeater is to match the nominal cable \nattenuation characteristic to within +0.1 dB. The difference between\n\nthe gain of the average repeater and the nominal loss of the cable is \ncalled the average design error and is obviously predictable. Equali- \nzation of the average design error is accomplished by providing fixed \nequalizers in the transmitting and receiving repeater in the main \nstations and in the equalizing repeater located at approximately the \nmidspan location between main stations. In previous L-carrier systems, \nthe loss characteristic of the deviation equalizer was determined as\n\nDetermine the average design error as a function of frequency . \nfor each type of line repeater.\n\nDetermine the cumulative average design error for the average \nexpected line section length.\n\nModify the cumulative design error to compensate for any \ndesign error at the temperature pilot frequency by adding (or \nsubtracting) the function KV f/fr, where K is the cumulative \ndesign error at the temperature pilot frequency, fr. This is \nnecessary because any gain offset at the pilot frequency will \nbe operated on by the regulating repeaters as if it were caused \nby a line temperature change.\n\nApportion the resultant characteristic, which is the desired \ntotal insertion loss characteristic (inverted in sign, of course) \nof all line deviation equalizers, equally to all equalizers. \nRealize a network to match the desired characteristic according \nto some error criterion (usually a minimum mean-squared \ncriterion).\n\nIn any actual line section, a residual gain or loss characteristic will \nstill remain, since the line will not contain all average repeaters and \ncable, and will be of a different than average length. It is difficult to \ndetermine beforehand whether this residual characteristic is equalizable\n\nwithin the capabilities of the variable equalizers provided in the \nsystem.\n\nIn the L5 system, a different design approach was taken. Rather \nthan the deviation equalizer being designed to match the average \ndesign error, it was designed by a computer program so that the \ndifference between the deviation equalizer and the average design \nerror is optimally equalizable by the L5 adjustable equalizers.\n\nThe results can be seen in Fig. 14. Curve (a) shows the average design \nerror of the average L5 power-feed section consisting of 60 basic and \n10 regulating repeaters. Curve (b) shows the difference between the \naverage design error and the deviation equalizers. Curve (c) shows the \nequalized characteristic predicted by the computer program using \nmathematical descriptions of the equalizers.\n\nThis approach has another advantage in terms of the system \ndevelopment time. Repeater development and variable equalizer \ndevelopment both require long intervals. Using the former approach, \noptimum variable equalizers cannot be developed concurrently with \nthe repeaters. In the new approach, the deviation equalizer is used to \noptimally \u201cmarry\u201d the repeater design error and the equalizers. The \ndevelopment time of the fixed deviation equalizer is very short com- \npared to the repeaters and variable equalizers.\n\nAs indicated in the previous section, gain deviations of the line \nrepeaters resulting from manufacturing differences from repeater to \nrepeater and differences in cable loss from section to section, taken \ntogether, result in a static misalignment of a line section. Since there \nare a large number of contributors to this static misalignment, mis- \nalignment of any line section as a function of frequency is extremely \ndifficult, if not impossible, to predict a priort. An equalizer designed \nto compensate for this misalignment must therefore be able to accom- \nmodate a wide range of shapes. This implies an adjustable equalizer \nwith considerable flexibility.\n\nOn the other hand, the static misalignment of any line section will \nbe a very slowly varying function of time. These considerations led \nto the design of manually adjusted equalizers that are adjusted upon \ninitial installation and, at very infrequent intervals thereafter, gov- \nerned by such factors as route growth and line equipment replacement.\n\nIn-service versus out-of-service adjustment is another important \nconsideration affecting adjustable equalizer design. While in-service \nequalization is preferable from an operational standpoint, the con- \ncomitant constraints imposed on the location of equalizer adjustment \ntones in the frequency spectrum make the achievement of an optimum \nequalizer setting difficult. Since the equalizers will be adjusted in- \nfrequently, however, out-of-service adjustment imposed no system \npenalty and, in fact, is considered necessary to achieve the objective \nof +0.4 dB in a switching section.\n\nIn wideband coaxial systems, two types of equalizers are commonly \nused for amplitude equalization\u2014transversal equalizers\u2019 and Bode\n\nequalizers.\u2019 After extensive studies conducted early in the L5 develop- \nment, Bode equalizers were selected primarily because the algorithm \nfor adjusting the equalizer setting to match the shape of the line is \nsimpler and converges more rapidly to the optimum settings than that \nfor transversal equalizers. This is an important consideration. Achiev- \ning the 4000-mile equalization objective depends to a large degree on \nthe ability of the telephone company craftsperson to achieve the \noptimum equalizer settings within the individual line sections. There- \nfore, it is a distinct advantage to have a simple algorithm. Furthermore, \na simple algorithm is much more amenable to mechanization. Mecha- \nnization, in turn, insures the achievement of optimum equalization. \nThis is discussed in detail later.\n\nIn the L5 system, two manually adjustable equalizers, the El and \nE2 equalizers, are provided to compensate for the residual static \nmisalignment of the Ld line sections.\n\nThe El equalizer contains ten broad shapes spaced across the L5 \nfrequency band from 1.6 to 66 MHz plus one very narrow shape \ncentered at the L5 temperature pilot location. They are shown in \nFig. 15. In each L5 section are three El equalizers, one in the trans- \nmitting main station, one in the receiving main station, and a third \nat the equalizer manhole. From a functional standpoint, the three E1 \nequalizers are identical. The manhole E1 equalizer differs from the \nothers in that it is packaged to fit in a manhole apparatus case and \nalso in that it derives its power from the de current on the center \nconductor of the coaxial cable. The primary function of the El equal- \nizer is to provide relatively coarse equalization of the L5 line and to \ncontain the message signal within relatively narrow levels as it tra- \nverses the line so that excessive noise penalties will not accrue.\n\nThe nominal insertion gain of the El equalizer\u2014that is, the gain \nof the equalizer when all bumps are in their reference state\u2014is 0 dB. \nThe gain or loss around zero of each bump can be set independently \nof any other bump by means of a manual control on the front panel \nof the equalizer. Note that, with the exception of the pilot bump, \neach bump shape has a zero crossover close to the center frequency \nof the bump immediately preceding it. Taken collectively, we refer \nto this property of the El bumps as being one-way orthogonal with \nrespect to their center frequency. The significance of this property \nis that, if the required gain setting of bump 7 is determined by a \nmeasurement of the line with only bumps 7 \u2014 j,7 = 1, 2, ---,; 7-1\n\npreviously set, then the required gain will not be altered by the \nsubsequent setting of bumps 7+ j, 7 = 1, 2, ---, N \u20147 (where N \nis the total number of bumps in the equalizer). It is this property that \nallows rapid convergence of the equalizing algorithm.\n\nThe pilot bump has a very specific purpose. The transmission level \nlayout of the L5 line is such that the absolute power level of the tem- \nperature pilot at the input to the El equalizer is always the correct \nlevel. If the E1 equalizer were to change the pilot level (by insertion \nof gain or loss by bump number 7, shown in Fig. 15, for instance), \nthen the regulating repeaters in the line following the El equalizer \nwould act to correct the change with Vf shape. In general, this would \nprevent the equalization from converging and the line could not be \nequalized. The purpose of the pilot bump, then, is to maintain the \npilot leaving the E1 equalizer at its correct transmission level regardless \nof any other gain setting within the equalizer. The pilot equalizer is \nnarrow enough so that it does not affect the equalization of the message \nband surrounding the pilot frequency.\n\nThe E1 equalizer is realized by a series of four amplifiers alternately \ninterconnected with four passive networks, as depicted in Fig. 16. Two \nnetworks are series double-bump Bode networks and the third is a \ntriple-bump Bode network. The theory of Bode network design and \nthe realization of the networks are discussed in a companion paper.\u00ae \nThe amplifiers in the E1 equalizer serve several purposes. First, the \ngain of the amplifier makes up for the flat loss of the networks so that \nthe overall insertion loss of the equalizer is 0 dB. Second, they provide \nisolation between the Bode networks so that the return loss interaction \ninherent in these networks is minimized. Finally, each amplifier \nprovides one bump shape by the inclusion of a Bode network in its \nfeedback path. A typical amplifier schematic is shown in Fig. 17.\n\nThe most difficult design problem in the El (and E2) equalizers \nwas the achievement of the overall frequency response. The gain or \nloss of each amplifier or network comprising the equalizers is adjusted \nduring its manufacture to be within +0.05 dB of its respective nominal \nvalue. The requirement on the overall equalizer is that, in its reference \nstate, the gain of the equalizer should be within +0.1 dB of zero when \nall the Bode networks are in their flat condition. This has been achieved \nby providing a trimming network adjusted during the assembly of \nthe El equalizer to trim the characteristic of the El to within the \ndesired requirements. Because the network and amplifier gains are \ntightly controlled, the trim equalizer is a simple network containing \nonly a flat and a slope term.\n\nAMPLI- EQUALIZER 2. AMPLI- EQUALIZER 3 AMPLI- EQUALIZER 1) = AMPLI- \nFIER 4 FIER 3 FIER 2 FIER 1\n\nThe E2 equalizer is similar in design to the El equalizer. It contains \n18 narrow shapes spaced across the 1.6- to 66-MHz spectrum. E2 \nequalizers are located in main stations at both the transmit and receive \nends of a power-feed section. The location of the E2 bumps are indi- \ncated in Fig. 18. The primary function of the E2 equalizer is to provide \nthe fine-grained equalization of the L5 signal so that, coupled with the \nE1 equalizer, the overall equalized response of a switching section is \nwithin +0.4 dB of zero from 1.6 to 66 MHz.\n\nThe E2 equalizer contains seven amplifiers alternately intercon- \nnected with seven networks. Six networks are series double-bump \nBode networks. The seventh network is a trim equalizer similar to \nthe one in the El equalizer. Six amplifiers contain Bode networks in \ntheir feedback paths, while the seventh is a flat-gain amplifier. Salient \nfeatures of the El and E2 equalizers are shown in Table I.\n\nFrequency Response (MHz) 1.6 to 66 1.6 to 66 \nNominal Gain (dB) 0.0+0.1 0.0 + 0.1 \nInput Return Loss (dB) >30 >30 \nOutput Return Loss (dB) >30 >30 \nNumber of Shapes 10 Broadband 18\n\n+1.0 Pilot Adjustment \nNumber of Amplifiers 4 7 \nNumber of Networks 4 7 \nFull Load Output Power (Single +6.5 +4.5\n\nThe 4000-mile equalization objective for the L5 system is a very \ndifficult objective to meet. Therefore, great attention has been given \nto the selection of the Bode network shapes used in the El and E2 \nequalizers and to the determination of the best equalization algorithm \nso that optimal equalization is achieved from section to section.\n\nFrom a transmission viewpoint, we can consider the El and E2 \nequalizers as a single equalizer with 28 Bode networks whose frequency \nresponse, HQL(f), is given by\n\nwhere f indicates the frequency and g,B;(f) represent the gain and \nfrequency response of the kth Bode network, respectively.\n\nIf M(f) represents the misalignment to be equalized, the residual \nerror, H(f), will be, after equalization,\n\nThe purpose of the equalization algorithm is to minimize a certain \nmeasure of E(f) in eq. (3) over the frequency range of interest. The \nideal condition would be to make E(f) equal to 0 over the entire \nfrequency range by proper selection of B,(f) and gx. In reality, how- \never, this is not possible. The L5 equalization strategy is to minimize \nthe mean-squared-error (MsE) function of E(f), which is defined as\n\nThe optimum set of gains, g;, which results in the minimum MsE \ncan be found if the gradient, G;, of the msE with respect to each gain, \ngz, becomes zero.?\u00b0! That is,\n\nThe MsE algorithm given in Ref. 10 solves eq. (7) with G = 0 by \nthe steepest descent method. This may be readily implemented in an \nautomatic equalizer control circuit, but not in the manually adjusted \nequalizers. The steepest descent algorithm requires simultaneous \nadjustment of all the gain settings. An equalizer adjustment algorithm \nbased on the Gauss-Seidel iterative method discussed in Ref. 11 is \nsuitable and hence adapted for the manual El and E2 equalizer \nadjustment. This algorithm calls for the following equalizer gain \nadjustment procedure:\n\n(z) Adjust the first gain, g1, until its corresponding gradient Gi = 0. \n(72) Repeat for the second gain, gz, through the last one, g2s, until \ncorresponding gradients become zero, thus completing one \niteration. \n(177) Repeat the above steps.\n\nSo far, it has been assumed that the gradient G, can be continuously \ncalculated during the adjustment of the kth Bode network gain, gy.\n\nOne approach is to compute the gradient by eq. (5). The gradient is \nobtained by cross-correlating the equalizer shapes B:(f) with error \nE(f). This procedure requires continuous error information over the \nfrequency range of interest and results in a rather complex hardware \nrealization.\n\nIt is shown in Ref. 10, however, that the gradient of the msE with \nrespect to a particular network gain can be very closely approximated \nby the following simple relationship :\n\nwhere E(f) is the channel residual error defined in eq. (3), fz2 is the \ncenter frequency of the Bode network B;(f), and fii and f,3 are lower \nand upper side frequencies of B,(f) such that\n\nBi(fis) = Bu(fis) = 2Be( fre). \nIf Bi(fi2) is normalized and its value is 1, eq. (9) becomes simply \nGi, = 2E (fia) + E (fiz) + Eas). (10)\n\nIn other words, by sampling the channel error at three frequency \npoints and by properly weighting these errors, we may approximate \nthe necessary gradient information. This approach is called the \nsimplified msE algorithm.\n\nIn Ref. 11, it is shown that a further approximation has resulted \nin a very simple way of computing the gradient:\n\nThat is, the gradient is approximated by sampling the channel error \nat a single frequency (usually the center frequency of the Bode net- \nwork). An equalizer adjustment method with the gradient obtained \nas in eq. (11) has been referred to as the zero forcing (zr) algorithm, \nand has been used, for example, in the L4 line equalization. When \nthe set of shapes B;, are near-orthogonal, the zF algorithm results in a \nrapid convergence to the optimum gain settings.\n\nExtensive computer simulation and field testing have been conducted \nto compare the effectiveness of the three adjustment algorithms in \nwhich the gradients have been calculated according to eqs. (5), (10), \nand (11). The results are shown in Fig. 19, which indicates that there \nis considerable improvement in the simplified Ms& algorithm over the \nzF algorithm. However, little improvement is achieved when the \ngradients are calculated according to the more complex relationship\n\nin (5). In the L5 equalization plan, the simplified sr algorithm has \nbeen implemented with the aid of a unit called the equalizer adjust- \nment unit (BAU).\n\nIn the equalization process, the BAv is used in conjunction with the \n90-type transmission measuring set composed of the 90G oscillator, \n90H detector, and 90F digital control unit,!2 which is located in every \nL5 main station.\n\nFor equalization, two separate EAU\u2019s and transmission sets are \nrequired, one at the transmitting station and one at the receiving \nstation (see Fig. 20). During equalization, each EAv is set by its \noperator for a particular Bode network, say B;,, and the Eau outputs \nthe appropriate control information to the 90 set for that network. \nAt the transmitting station, the command portion of the EAU instructs \nthe 90G to generate specified levels and proper frequencies corre- \nsponding to the network to be adjusted. In the zF mode, the EAU \ncauses the 90G to generate only frequency fi2, which corresponds to\n\nMANUAL CONTROL MANUAL \u2018CONTROL MANUAL CONTROL AUTOMATIC \n*eeMN0 11 ss 428 | Wide. 0 Wtlee*NO 1W4)---$28 CONTROL \nrf 2H fd a fy (:*\u201d 5 a i \nALIZER] [EQUALIZER EQUALIZER y) EQUALIZER] |EQUALIZER] JEQUALIZER \n\u2018 / TE \nx a E(fx1) \nREPEATERED LS5 LINE Elfica) \n1 4 Elf3) \n! \nq 1 \noS 7 --\u2014-t---. \n90H | 90G 90H \n| DETECTOR | | OSCILLATOR | DETECTOR \noe oo ed Le a ed) \nq Tupve) (IDLE) . \n' 1 ' \nt 1 t \n1 t \nt | \n90F __J to 90F \nDCU DCU \nEAU EAU.\n\nthe center frequency of B,. If the MsE mode is selected at the Eau, \nthe 90G will generate sequentially frequencies fi, fizz, and fi3 as \ndefined in eq. (10). These frequencies are then transmitted on the \nLd line.\n\nAt the receiving location, the EAv initially establishes synchroniza- \ntion between the transmitting 90G and the receiving 90H. The 90H \nthen receives these signals, measures the received level, and transmits \nthe resulting level information to the arithmetic portion of the Eau. \nThere it is processed and the resultant gradient information is displayed \nto the operator in numeric form. The displayed gradient is calculated \nby either eq. (9) or (11), depending on whether the mss or zF mode, \nrespectively, has been selected by a front panel switch located on \nthe BAU.\n\nIn the equalization procedure, since the gradient information is \ndisplayed only at the receiving end of the line, the adjustment is under \nthe control of the operator at the receiving station. The receiving \noperator relays the correct gain setting for each equalizer bump to \nthe operator at the transmitting station and to the craftsperson at the \nmanhole equalizing repeater location. To achieve the proper pre- and \npostequalization during the adjustment of the E1 equalizer, one-half \nthe error is compensated for by the equalizer in the manhole and one- \nquarter of the error is compensated for in both the transmitting and \nrecelving main-station equalizers. In the case of the E2 equalizer, \none-half the correction is inserted at both the transmitting and re- \ncelving main stations. The ultimate objective is to adjust all the equal- \nizer shapes so that the resultant gradient, as displayed on the Bau, is \nzero for all shapes.\n\nIn the equalization procedure, the zF algorithm is first used because \nof its rapid convergence to the near-optimum gain settings. The MsE \nalgorithm is then used to \u2018\u201c\u2018fine tune\u201d the equalization. The iterative \nprocedure is as follows:\n\n(22) Select El equalizer, shape B,. Adjust the equalizers (one-half \nthe correction in the manhole equalizer, one-quarter in both \ntransmitting and receiving main stations) until the gradient \nof the error is less than 0.05 dB.\n\n(tv) Select E2 equalizer, shape Bi. Adjust the equalizers (one-half \ncorrection in both transmitting and receiving main stations) \nuntil the gradient of the error is less than 0.05 dB.\n\n(v) Repeat (iv) for E2 shapes By through Bos. \n(vt) Select the msE mode of gradient calculation. \n(vit) Iterate steps (77) through (v) until the gradient is within +0.05 \ndB of zero for all shapes.\n\nIn practice, convergence of the iteration is quite rapid so that usually \ntwo or three iterations are sufficient.\n\nIn both the L4 and L5 systems, the coaxial cable is buried at a \ndepth of four feet. The change in cable loss resulting from a +20\u00b0F \ntemperature change, for power-feed spans approaching 75 miles in \nlength, is +52 dB at 66 MHz. This is the magnitude of the equalizing \nfunction of the regulating repeaters at the high end of the L5 frequency \nspectrum.\n\nThe strategy in providing compensation for this seasonal dynamic \nchange in loss is focused at minimizing the signal-to-noise impairment \nresulting from misalignment caused by the temperature effect. In \nthe L5 system, the distance between regulating repeaters is a maximum \nof seven miles. With this bound established, the misalignment resulting \nfrom cable temperature changes is reduced from +52 dB to +4.8 dB.\n\nFig. 21\u2014Regulating section\u201466-MHz misalignment at the regulating repeater \noutput and input owing to cable temperature variation.\n\nWhen both pre- and post-correction are provided, the misalignment \nis further reduced to +2.4 dB as shown in Fig. 21.\n\nA second effect that the regulating repeater must compensate for \nis not directly related to temperature, but more to distance. Since the \nbasic repeater is a fixed gain amplifier designed to equalize the loss \nof one mile of cable at 55\u00b0F, adjustments are made in the nominal \nspacing of the repeater manholes when cable is placed in areas having \nother than a mean annual temperature of 55\u00b0. In warmer areas, the \nmanholes are placed at intervals shorter than 1 mile; in colder areas, \nplacement intervals are greater than 1 mile. A 5-degree error in mean \nannual temperature estimation for system design results in 2158 feet \nof cumulative error in manhole placement over 75 miles. This amounts \nto 18.1 dB of misalignment at 66 MHz or, as in the case of the L5 \nsystem, 1.2 dB in a 7-mile regulating section. The loss relationship as \ndefined in (1) is predominantly Vf.\n\nThere is a third source of misalignment. One inevitable situation \nthat plagues those who engineer cable systems is the inability to place \na manhole at the locations dictated by the nominal spacing. Right-of- \nway procurement, convenient access, terrain considerations, and \ndensely populated and urban areas all contribute to a deviation from \nnominal spacing. The accumulation of these deviations within a \nregulating section results in additional misalignment that is accommo- \ndated by the line-build-out networks in 0.1-mile increments, and the \nremainder is equalized by the closed-loop regulating action of the \npostregulator described in the next section. Therefore, equalization of \ncumulative spacing deviations of +0.05 mile, equivalent to +1.6 dB \nat 66 MHz, is another objective of the regulating repeater.\n\nOther sources of misalignment, such as the deviation in basic \nrepeater gain from the nominal gain, the uncertainty in placed cable \nlengths, and others, are estimated to be +0.4 dB. Table II lists the \nmisalignments and establishes the total range required of both the \npost- and preregulator circuits at 66 MHz for 7 miles.\n\nTable Il\u2014 Misalignments allocated to the regulating repeater \nCause Magnitude \nChange in cable loss caused by temperature variation (+20\u00b0F) +4.8 dB \nCumulative length error (+0.038 mi) +1.2 dB \nCumulative spacing deviation (\u00a30.05 mi) +1.6 dB \nOther +0.4 dB \nTotal +8.0 dB\n\nThe preregulator automatically accommodates for +2.4 dB of the \ncable loss variation resulting from temperature changes, and the post- \nregulator must accommodate the remaining -+-5.6 dB. This effect is \nillustrated in Fig. 22.\n\nThis suggests a range requirement of 11.2 dB for the postregulator. \nHowever, because of circuit realization limitations, only 8.7 dB was \nattainable in either a post- or preregulator network while still meeting \na broadband objective of less than +0.015 dB per dB of tracking \nerror. It was found that line acceptance procedures overcome this \nlimitation if the uncertainties and deviations accumulate in such a \nway that the range of the regulating repeater is exceeded. These \nprocedures locate the excess variations which are then compensated \nfor with either reassignment of line-build-out networks or readjustment \nof preregulators.\n\n(7) To equalize the dynamic temperature-dependent term of the \ncable equation in a pre- and postequalization strategy.\n\n(it) To accommodate automatically the uncertainties and cumu- \nlative spacing deviation in a postequalization strategy.\n\nThis total function is aimed at minimizing the signal-to-noise \nimpairment by maintaining the distributed misalignment at small \nmagnitudes.\n\nFig. 22\u2014Regulating section\u201466-MHz misalignment at the regulating repeater \noutput and input resulting from cable temperature variation and other sources of \nmisalignment.\n\nThe regulating repeater consists of those circuits that comprise a \nbasic repeater plus additional circuits to perform the pre- and post- \nregulating functions. These additional circuits are located between \nthe preamplifier and power amplifier as shown in Fig. 23.\n\nIt has been established that the misalignments allocated to the \nregulating repeater are predominantly losses or gains that vary with \nfrequency in a Vf manner because of temperature and length. There- \nfore, the equalization networks must also vary as Vf. The misalignment \nallocated to the preregulator is one-half the variation resulting from \ntemperature. Control of the preregulator is established by sensing earth \ntemperature in the vicinity of the manhole. To establish proper track- \ning of the cable temperature, the sensor must be placed in an earth \nenvironment similar to that in which the cable is buried. The depth \nof the sensor is particularly important. Errors in depth significantly \naffect the amplitude and delay of earth temperature variations with \nrespect to the cable.\n\nControl of the postregulator is established by sensing the level of a \nsingle frequency, designated the temperature pilot, which is applied \nat the transmitting end of the cable. This method of control has the \nadvantage of automatically correcting for all misalignments of this \npilot. Therefore, the remaining half of the temperature effect on cable, \ncumulative spacing deviations, and uncertainties in cable length and \ndesign temperature are equalized in the postregulator. However, pilot \ncontrol may be troublesome if the losses incurred are not Vf, since \nthe correct equalization shape will not be provided. In the L5 system, \nthis difficulty is avoided by the proper selection of the pilot frequency \nand the attention given to the basic repeaters and to the fixed and \nadjustable equalizers at this frequency. Although maximum sensi- \ntivity to cable temperatures occur at the high end of the L5 spectrum, \nthe pilot was chosen to be at 42.880 MHz. The reasons for this choice \nare\n\n(i) The temperature coefficient of the basic repeater (not vf) \ncould be made to approach 0 at this frequency, but is a maxi- \nmum at higher frequencies.\n\n(771) Return loss is better controlled and therefore interactive \neffects are minimized.\n\n(ziz) This frequency is in the guard band between the second and \nthird jumbogroups.\n\na ee eee ee ZERO LOSS AT NOMINAL \nT TEMPERATURE AND MANHOLE SPACING \u2014\u2014\u2014\u2014\u2014\u2014\u2014\u2014\u2014\u2014~\u2014\u2014\u2014 7 \nTEST PORT \n| PHASE-SHAPING \nDEVIATION BUFFERED BUFFERED \nEQUALIZER YF NETWORK Yf NETWORK\n\nFROM \nINPUT OUTPUT \nDC GAIN WITH @EROM \nTHERMISTOR \nCOMPENSATING ihe TEMPERATURE \nZERO AND SENSOR \nDRIVER NONLINEAR \nMAPPING CIRCUITS \nDRIVER \nrn AND DRIVE \nBAND AMPLIFIER\u2014 \nPILOT DETECTOR\n\nFig. 23\u2014Additional circuits that perform the pre- and postregulating functions.\n\nEach pre- or postregulator contains identical Bode equalizers (vf \nnetworks?) with loss adjustment controlled by varying a single resistive \nelement. This element is an indirectly heated thermistor with the \nheater winding electrically isolated from the thermally controlled \nresistor.\n\nGain is provided in each regulator to compensate for the loss in the \nBode equalizer and the associated networks by single-stage thin-film \nhybrid integrated-circuit amplifiers. Two amplifiers and the Bode \nequalizer are packaged to form the Vf network. The amplifiers also \nprovide the proper terminating impedances for both ports of the \nequalizer to achieve an accuracy that is better than 1 percent in \nmatching the loss variation which is Vf. The type of transistor used in \nthe basic repeater circuits is also used in the single-stage amplifier to \nachieve the noise figure and linearity requirements.\n\nOne critical aspect in the design of the preregulator control circuitry \nis to accurately map the nonlinear resistance-versus-temperature \nfunction of the buried ground temperature sensor into a linear loss \n(in dB) function. In addition, provision had to be made to control the \nsensitivity of the function since regulating sections can vary signifi- \ncantly in length, and to adjust for the mean annual temperatures since \nthe same sensor is placed in all climatic areas. The response of the \ntemperature sensor is shown in Fig. 24. The range expected is +20\u00b0F\n\nThe nonlinear mapping was accomplished through the use of \noperational amplifiers. With the use of negative feedback and scaling, \nthe control circuit provides a temperature tracking function that is \naccurate to better than 2\u00b0F. The mean annual adjustment is made \naccessible so that readjustments may be made safely in the manhole \nwhen the repeater is installed and powered, should line acceptance \nprocedures indicate that the postregulator range is being exceeded.\n\nIn the postregulator, the temperature pilot is selected from the \nmessage band with a crystal filter, amplified, and converted to a direct \ncurrent in a peak detector. Comparison of this direct current signal to a \nreference results in an error signal that is amplified and applied to the \nheater of the thermistor in the Bode equalizing network. The heater \nsets the thermistor resistance in the Bode network and the pilot level \nis maintained constant at the output for changes in level at the input.\n\nThe goal in this control circuitry is to achieve a reasonable response \ntime while maintaining a well-damped, nonenhanced system of many \npostregulators in tandem.\n\nA system of tandem regulators may be modeled as shown in Fig. 25. \nIn this figure,\n\nG(s) = the control loop transfer function of a single regulator, where \nsis a Laplace transformed variable.\n\nThe normal expected response of this system is to have H,; vary \nseasonally while tracking each cable section loss change that results \nfrom the cable temperature change. This variation is extremely slow \n(one cycle per year) and, since G(s) is large for such a slow variation, \nthe envelope gain of the pilot is nearly zero while the message envelope \nis properly tracking (A = Vjfm/fp) the pilot envelope to maintain the \nlevels constant at the output of each regulator. Therefore, each \nregulator compensates for its preceding section of cable.\n\nFor very fast variations in the pilot envelope, the envelope gain to \nthe pilot is unity, and the message envelope gain is nearly zero. \nMessage channels, therefore, are not affected by the faster per- \nturbations.\n\nWhen G(s) has a single pole, no gain enhancement (\u00a3,./E\u00bb; > 1) \nwill occur, and minimum overshoot to transients will result. When \nG(s) has more than a single pole, gain enhancement will occur where \nG(s) approaches unity, and the system will not be critically damped \nto transients.\n\nIn the postregulator, the loop response contains several poles. These \npoles are a result of the indirectly heated thermistor. A plot of the \nfrequency response of the loop gain is shown in Fig. 26.\n\nTo restrict the gain enhancement to a tolerable magnitude, the \nresponse is compensated for by using an operational amplifier to \nprovide a pole cancelling zero in the vicinity of 2 Hz. This results in \nan enhancement response as shown in Figs. 27 and 28 where computed \nand measured data are compared and in a transient response as shown \nin Fig. 29.\n\nOne final aspect of the circuitry pertains to a transfer arrangement \nin the event of pilot failure. The loss of the temperature pilot initiates \na switch to the spare line. This pilot loss could also cause the envelope- \ncontrolled regulators to go to an extreme gain condition. To prevent \nthis high gain from occurring, heater control of the postregulator is \ntransferred from envelope control to preregulator control. Under these \ncircumstances, those misalignments allocated to the postregulator \nare no longer equalized. However, in an L5 switching section (150\n\nFig. 27\u2014Computed pilot envelope response for 80 regulators in tandem with and \nwithout thermistor pole cancellation. Measured data points are compared.\n\nFig. 28\u2014Computed channel envelope response for 80 regulators in tandem with and \nwithout thermistor pole cancellation.\n\nFig. 29\u2014Transient response of 80 regulators in tandem. Upper trace is the input \nexcitation of 0.8 dB.\n\nmiles), the misalignments owing to cumulative spacing deviation and \nuncertainties tend to be both positive and negative, and hence there \nis some cancellation. The net result is that gain is maintained near \nnormal. Also, regulator response time is fast so recovery from typical \nno-pilot misalignments of 6 dB is complete in less than 5 seconds for a \nswitching section containing 25 regulators.\n\nThe networks in the transmission path of the regulating repeater \ncontribute to the noise and linear performance of the repeater. These \nparameters are listed and compared to the same parameters of the \nbasic repeater in Table III.\n\nThe final level of equalization in the L5 system is performed by an \nautomatic pilot controlled equalizer which equalizes for a number of \ndynamic effects. The primary effect is that associated with the small \nbut significant change in the gain of the L5 line repeaters because of \nseasonal temperature changes. Repeaters located in manholes are \nsubject to ambient temperature changes roughly equivalent to the \n+20\u00b0F temperature change of the coaxial cable itself. Line repeaters \nhaving an average temperature coefficient of less than 0.002 dB per \ndegree Fahrenheit at 66 MHz introduce less than 3-dB change in line \ngain over a 75-mile power-feed section. The variation of this deviation \nwith frequency is predominantly non-Vf and is therefore not compen- \nsated for by the regulating repeaters.\n\nIn addition, as discussed in the previous section, a tracking error is \nassociated with the inability of the regulating repeaters to accurately \nmatch the change in loss of the cable as a function of frequency and \ntemperature. Over a 75-mile power-feed section, deviations of up to \n1.6 dB at 66 MHz can be expected.\n\nWhile seasonal temperature variations have the main effect on \nline gain change, there are other second-order, longer term effects \nsuch as component drift, aging, and increases in the manhole ambient \ntemperature as the number of equipped coaxial lines in a route is \nincreased. All these taken together represent the total dynamic \ndeviation of the Ld line and are equalized by the E3 dynamic equalizer.\n\nThe realization of the E3 equalizer has been a function of a number \nof system considerations. In previous sections, it was pointed out that \nthe static deviation, being a result of statistical manufacturing devia- \ntions of many components, was impossible to predict a priori. Thus \nthe El and E2 equalizers are very flexible and have many degrees of \nfreedom to match a wide variety of shapes.\n\nOn the other hand, we expect the dynamic deviation to be much \nbetter behaved. In previous L-carrier systems, the gain of the line \nrepeater as a function of the repeater temperature was a broad function \nof frequency. Predictions based upon gain sensitivity analysis indicated \na similar behavior for the L5 repeatered line. We would expect then \nthat an equalizer designed to compensate for this effect would also be \na broad function of frequency with correspondingly fewer degrees of \nfreedom. Furthermore, this function could be measured a priori by \nmeasuring the gain deviation of a sample of repeaters for various \nambient temperatures.\n\nWhile the dynamic misalignment of the L5 line is a slowly varying \nfunction of frequency, the overall equalization objective of the L5 \nline dictates that automatic equalization be employed rather than \nmanual equalization, as in the case of El and E2 equalizers. While \nthere are many ways we could conceive of providing automatic \nequalization, the use of pilot tones in the composite message signal \nand independent of the message signal itself allows a relatively simple \nand reliable method of providing such equalization. Therefore, pilot \nfrequencies have been included above and below the message band as \nwell as between jumbogroups to provide for dynamic equalization. \nThe pilot between jumbogroup 2 and 3 is the same pilot used for \ntemperature regulation.\n\nAll these considerations have led to the E3 dynamic equalizer with \nfour degrees of freedom, each degree of freedom being automatically \ncontrolled by a line pilot.\n\nUnlike all other equalization in the L5 system, the E3 dynamic \nequalization is provided only on a postequalization basis. There are two \nreasons for this: In the first place, the expected misalignment over a 75- \nmile power feed span is about 3 dB at 66 MHz. The noise penalty as- \nsociated with postequalization only, under these conditions, is on the \norder of 1 dB and has been included in the system margins. In addition, \nany scheme to provide both pre- and postequalization on a dynamic \nbasis for all four pilots would be considerably more expensive and is \nnot economically justifiable. Thus, the E3 dynamic equalizer is located \nin main stations at the receive end of each Ld power-feed section.\n\nFigure 30 is a block diagram of the E3 dynamic equalizer. Basically, \nit consists of four high-frequency networks through which the L5 line \nsignal passes. Each network has a variable gain shape such that the \ncombined frequency characteristic of all four networks closely matches \nthe expected residual dynamic misalignment of the L5 line. Three of \nthese shapes are Bode bumps, while the fourth shape is a flat function \nof frequency (Fig. 31).\n\nIn Fig. 30, the first network is a series Bode network imbedded \nbetween two flat-gain amplifiers. The amplifiers provide an overall \ngain of 6.55 dB and provide good input and output return loss. Im-\n\nbedded in the Bode network is a thermistor whose resistance is con- \ntrolled by the 42.880-MHz temperature pilot at the E3 equalizer \noutput through the control circuit feedback path. This resistance in \nturn controls the gain of the Bode network such that the level of the \n42.880-MHz pilot at the output of the equalizer is held at a fixed \nreference level. .\n\nThe second network is a broadband variolosser that provides a flat \ngain or loss around a nominal loss of 12 dB. The amount of gain or \nloss is controlled by a thermistor in the variolosser whose resistance is \ncontrolled by the level of the 20.992-MHz pilot at the output of the \nequalizer.\n\nThe third network is a series Bode network similar to the first \nnetwork. In this network, the gain of the Bode network is determined \nby the level of the 66.048-MHz pilot at the output of the equalizer.\n\nThe fourth network is a broadband feedback amplifier of nominal \n14.0-dB gain with a Bode network in its feedback path. This network \nalso contains a thermistor whose resistance is controlled by the level \nof the 2.976-MHz pilot at the equalizer output.\n\nThe E28 equalizer control circuits contain a pilot pick-off circuit \nwhere the four E8 line pilots are separated from the message signal \nand converted to four de voltage levels, each proportional to the \nabsolute power level of the respective pilot. The remainder of the \ncontrol circuits are digital circuits. The heart of each digital circuit \nis a binary up-down counter coupled with a digital-to-analog converter \nthat acts as a digital integrator in the feedback loop. The system \nworks in the following fashion: If the pilot level at the output of the \nequalizer deviates more than about 0.05 dB from nominal, it is detected \nby a change in the de pilot voltage. Under this condition, the up-down \ncounter is allowed to count. This count is converted by the digital-to- \nanalog converter to a de current, which is then applied to the thermistor \nin the corresponding network. This causes the network gain to change \nin a direction to restore the output pilot level to its correct value. \nWhen the pilot returns to within +0.05 dB of nominal, the counter is \nstopped and the gain of the network is held at that level.\n\nThere are several advantages in using digital circuits in the feedback \nloop when space and power consumption are not critical.\n\n(2) With a digital integrator within the control loop, an equiv- \nalently higher loop gain can be obtained than with an analog\n\nintegrator. The steady-state accuracy of the control loop is \ndetermined primarily by the size of the quantizing steps, the \ntracking between the digital-to-analog converter, and the \nnetwork control characteristic (input voltage versus network \ngain in dB).\n\n(it) If an abrupt change in pilot level or loss of pilot occurs because \nof an abnormal line condition, the digital memory holds the \nequalizing network control fixed.\n\n(tit) The midrange gain of each equalizing network is precisely \ndefined by the state of the up-down binary counter so that the \nnetwork may be set at midrange with a local key or by remote \ncommand.\n\n(wv) The control loop dynamic response can be controlled by the \nrate at which pulses are clocked into the up-down counter.\n\n(v) The binary counter state can be monitored to indicate equalizer \ngain and provide for control and alarm functions such as \nlocking the gain at any given state and giving a warning of \nend of equalizing range.\n\nIf M(f, t) represents the time-varying misalignment to be equalized, \nthe residual error after equalization will be similar to eq. (3) and can \nbe expressed by\n\nwhere f and \u00a2 indicate the frequency and time, g, and B; are the gain \nand input-output relationship of the kth network, respectively, and \nthere are four adjustable networks in the equalizer.\n\nThe E83 equalizer is adjusted continuously in-service by sampling \nchannel misalignments at four pilot frequencies. In this way, four \npilot signals generate the four equalized channel errors. The network \ngains are adjusted until the four errors become zero.\n\nThe block diagram shown in Fig. 30 can be represented by the \nfunctional block diagram shown in Fig. 32. Figure 32 indicates that \nthe E8 equalizer is a multivariable system whose input and output \nrelationship can be expressed by the following equation.\n\nReferring to Fig. 32, the gain g is obtained in the feedback loop and \nexpressed by\n\nwhere G isa 4 X 4 control matrix, and W isa 4 X 4 weighting matrix. \nG is an operator and determines the dynamic behavior of the four \nfeedback control loops and is expressed approximately by\n\nwhere K and T are the constants andI isa 4 X 4 unity matrix. From \neqs. (13) and (14), the input-output relationship of the equalizer \nbecomes\n\nAll the main and power-feed stations along the L5 coaxial line are \nequipped with E8 equalizers. Hence, any disturbance leaving one \nequalizer will be propagated to the following equalizers whenever \nthose equalizers are controlled by the same pilot signals. If pilot \nsignals are blocked and reinserted every Nth station, then up to N \nequalizers are effectively connected in tandem, and when the individual \nfeedback loop is designed, it is necessary to consider the effects on the \nsubsequent N \u2014 1 equalizers.\n\nThe dynamic behavior of the E3 equalizer is mainly determined by \nB and G in egs. (13) and (14), respectively. As shown in eq. (13), the \nmatrix B is determined by the network shapes in E3 and pilot fre- \nquencies. When B and G are finally designed, a further improvement \nmay be obtained by a suitable choice of the weighting matrix W shown \nin eq. (14).\n\nIn reality, the control loop transfer function, G, includes nonlinear \nelements (e.g., thermistor), and the frequency domain approach to the \nanalysis and synthesis of the control loop to satisfy the transient \nbehavior becomes less accurate. A digital computer simulation in the \ntime domain was developed to predict the behavior of the N-tandem- \nconnected equalizers when the transfer function of the single control \nloop, G, is known. The computer results were used to modify the \ncontrol loop transfer function to satisfy the system requirements of \nthe N-tandem-connected equalizers.\n\nFigure 33 shows a transient response of four E3 equalizers connected \nin tandem when input pilots are step-disturbed by 2 dB. Figure 34 is\n\nsimilar to Fig. 33, but includes the transient effects of 40 regulating \nrepeaters within four power-feed sections (about 250 miles). These \nresults were measured in the first L5 installation.\n\nThe concept of a hierarchy of repeaters in which the basic repeater \nis the fundamental building block was originated in the L4 system. \nExperience with the L4 system has proven this concept to be sound, \nand the L5 repeaters are similarly based. Design emphasis was focused \non the realization of an ultralinear low-noise repeater with a frequency \nresponse that is consistently reproducible in manufacture. This\n\nFig. 36\u2014Measured line noise (noise loading) extrapolated to 4000 miles (measured \nlength: 750 miles).\n\n1. G. H. Duvall and L. M. Rackson, \u2018L4 System: Coaxial Cable and Apparatus,\u201d\u2019 \nB.S.T.J., 48, No. 4 (April 1969), pp. 1065-1093.\n\n2. F. A. D\u2019Altroy, R. M. Jacobs, J. M. Nacci, and E. J. Panner, \u201cL5 System: \nUltralinear Transistors,\u201d B.S.T.J., this issue, pp. 2195-2202.\n\n4. J. L. Garrison, L. P. Labbe, and C. C. Rock, \u201cL4 System: Basic and Regulating \nRepeaters,\u201d B.S.T.J., 48, No. 4 (April 1969), pp. 841-889.\n\n. R. M.-M. Chen, C. F. Hempstead, Y. L. Kuo, M. L. Liou, R. P. Snicer, and \nE. D. Walsh, \u201c5 System: Role of Computing and Precision Measurements,\u201d \nBS.T.J., this issue, pp. 2249-2267.\n\nCopyright \u00a9 1974 American Telephone and Telegraph Company \nTue Brevi System TECHNICAL JOURNAL \nVol. 58, No. 10, December 1974 \nPrinted in U.S.A.\n\nThe repeaters in a pair of LS coaxial-carrier transmission lines are \npowered in a series loop with two dc-to-dc converters at each end of a power \nspan. The loop ts normally grounded at one end with the converter voltages \nbalanced to minimize the voltage to ground at the floating end. As with the \nL4 system, automatic grounding of the floating point is provided to limit \nthe line voltage to ground and to enable the two lines to be powered in- \ndependently during turnup and under trouble conditions. A new dc-to-dc \nconverter powered by a 140-V battery has been developed. It employs a \nsingle power stage to deliver 910 mA at output levels up to 1150 V. A \ndc-to-dc converter was also developed for operation on 24-V battery power \nfor use at stations where installation of a new 140-V battery plant is not \nwarranted. The dc-to-de converters employ hybrid, thin-film, integrated \ncircutts to achieve the needed level of precision without dependence on field \nadjustments. They also employ many new automatic control features that \nsimplify operation of the four converters in a series loop and permit \nturnup by remote control via the E2 status reporting and control system. \nAnother new feature is a spare converter with reversible polarity and a \ndecentralized high-voltage patching arrangement.\n\nThe repeaters and equalizers in the Ld coaxial-carrier transmission \nsystem are spaced at approximately one-mile intervals along the cable \nsystem. Dc-to-de converters operating on station battery power and \nspaced up to 75 miles apart are employed to furnish a constant current \nof 910 mA at voltages up to 1150 V dc over the coaxial line to the \nseries-connected repeaters and equalizers. This article describes the \nline-power-feed system and equipment with emphasis on the features \nthat simplify (or automate) system operation and that minimize the\n\neffect of foreign voltages which may be injected into the system \nbetween stations.\n\nThe L5 repeatered lines, like those of the L4 system, are powered \nin pairs using a series loop including two dc-to-de converters at each \nend of a power-feed section (see Fig. 1).! For cables shorter than 37.5 \nmiles, the pair of converters at one end is omitted. The system is \npermanently grounded at one station and the output voltages of the\n\nREPEATERS AND FOR LESS \nEQUALIZERS THAN \n37.5 MILES \n7 \n1 ge \n0 TO 1150 \nVdc \ndc INPUT \n(STATION \nBATTERY) de INPUT\n\nSEPAR\u2014 > _ : > SEPAR~ \nTRANSMISSION SLTER FEA TRANSMISSION \nSIGNAL SIGNAL \nOUTPUT + 910 mA de | INPUT\n\nfour series-connected converters are balanced under normal conditions \nsuch that the voltage is near zero between the floating and local-ground \npoints at the distant station. Therefore, the highest voltage to ground \nis 1150 V and no path exists for the flow of direct current through \nearth-metal interfaces between stations. This avoids electrolytic \n\u2018corrosion at such interfaces under normal steady-state conditions.\n\nUnder certain abnormal conditions, grounding of the floating point \nis desired. If, for example, a foreign longitudinal voltage appears \nbetween grounded points in the system, grounding of the floating- \nground point may be needed to limit the peak voltage appearing \nbetween the center conductors and outer conductors of the coaxial \ncables. This arises from the tendency for such voltages to be super- \nimposed on the output voltage of the converters connected to the \nfloating ground point. A second abnormal condition which warrants \ngrounding of the floating point is the appearance of a cable fault or a \nconverter failure that would reduce the current in the power-feed loop \nbelow an adequate level. By detecting such a fault and automatically \ngrounding the floating point, normal current (and transmission) can be \nmaintained in one of the coaxial lines.\n\nThe principal electrical performance requirements that influenced \nthe design of the line-power-feed system are summarized in Table I. \nThe allowance for de earth potentials as high as 15 V per mile is \nlarger than that used in designing previous systems of this type. \nSuch earth potentials are caused by geomagnetic storms that ac- \ncompany solar flares. The effects are largest on east-west routes at \nhigh latitudes in areas with high soil resistivity. Large de earth po-\n\nTable |\u2014 System requirements \nNominal line-current 910 mA \nLoop voltage drop 120 to 4600 V \nLine-current variation (normal +40 mA for combined variations of \nlimit) +100 V in loop drop, +2 V per\n\nmile de earth potential, 10-50\u00b0C \nambient temperature, 120-152 V \nconverter dc input voltage.\n\nLine-current variation (during +110 mA for earth potentials up to \nunusual geomagnetic disturbances) +12 V/mile.\n\nSystem-withstand capability (no DC earth potential up to +15 \nprotective shutdowns or equip- V/mile. 60-Hz induction up to \nment damage) 1500 V rms for 0.1 s.\n\ntentials occur infrequently and persist only for a few minutes in a \ngiven locality. The design objectives of avoiding major transmission \nimpairment up to 12 V per mile, and avoiding protective power shut- \ndowns up to 15 V per mile are considered to be quite conservative, \ninasmuch as earth potentials approaching these levels are expected to \noccur only rarely and in limited areas of the United States.\n\nHigh levels of longitudinal 60-Hz induction can be caused by large \nunbalanced currents under fault conditions on major commercial \npower lines which parallel the cable route. The \u201csystem-withstand\u201d \ncapability of 1500 V is considered adequate for most applications. \nWhere routes have unusual exposures to high levels of 60-Hz induction \n(i.e., more than 1500 V), additional measures, such as supplementary \ncable shielding, must be considered.\n\nA major factor affecting converter design was the desire for sim- \nplicity of system operation. In particular, it was desired to be able to \nturn up converters at unattended stations by remote control. This \nimplied a need to design the converters and protective ground circuits \nto perform several functions automatically. These functions include:\n\n(2) Soft start\u2014To increase line current gradually over a period \nof about 15 seconds. \n(71) Converter voltage limiting\u2014To avoid an excessive output \nvoltage from the first converter turned on. \n(2) Automatic grounding\u2014To ground the floating point during \nturnup. \n(iv) Automatic ungrounding\u2014To unground the floating point \nupon completion of turnup. \n(v) Converter current limiting\u2014To avoid producing excessive \ncurrent into a short circuit. \n(vt) High-impedance shutdown\u2014To prevent turnup into an open \ncircuit (avoids unnecessary personnel hazards).\n\nTo provide for faster power restoral after a protective shutdown \ncaused by an unforseeable momentary transient, the converter would \nbe designed to automatically make a single attempt to restart. To \nmaintain the converter and to provide for fast restoral of power in \nthe event of converter failure, a manually patchable spare converter \nand test-load facility would be available at each power-feed station.\n\nSimple in-service pushbutton tests would be employed to verify \nthat protective converter shutdown and grounding circuits are func-\n\ntioning. The need for adjustments to the system in the field would be \nminimized.\n\nIn summary, it was thought that the goal of high system reliability \nwould be best served by accepting moderate increases in the complexity \nof the power-feed equipment to simplify its use in the field and permit \nremote control.\n\nThe basic element in the converter is the power conversion section \nwhich changes the de input voltage available from the station batteries \nto a higher voltage that is controllable and is isolated from the input. \nPower converters for the L5 system were designed to operate on 24-V \nand 140-V battery plants. The power conversion technique for 24-V \ninput is similar to that developed for the L4 system,! and employs up \nto five pulse-width-modulated power stages. Each power stage employs \npower transistors switching at 20 kHz and delivers approximately \n200 W at 230 V and 910 mA.\n\nA new type of transistor-converter circuit?-* was developed for use \nwith the 140-V battery plants that will be used increasingly in the \nBell System.> The new converter circuit employs transistors rated at \n200 V and 12 A and delivers approximately 1 kW from a single unit. \nThe new circuit, illustrated in Fig. 2, separates the pulse-width-control \nfunction from the power-inversion function by employing a switching \nregulator (Q1, D1) operating at 40 kHz on the input side of a bridge- \ninverter circuit (Q2, Q3, Q4, Q5) that operates at 20 kHz. A principal \nfeature of this circuit is the presence of the inductor which averages \nthe voltage pulses at the input to the inverter rather than placing the \ninductor in the conventional position at the output of the rectifier \nbridge. This provides a number of advantages:\n\n(z) Eliminates voltage and current surges that ordinarily \nappear in the transistors and diodes as a result of rectifier-diode \nrecovery and/or transformer saturation.\n\n(tv) Permits effective utilization of techniques to reduce switching \nloss.\n\nBy using nonlinear networks to improve the switching locus of the \ntransistor, the switching losses were kept quite low even at a switching\n\nfrequency of 40 kHz using transistors with 0.75-us rise and fall times. \nThese techniques are illustrated for the switching regulator section \nof the converter in Fig. 3. In Fig. 3a, the switching transistor operates \ninto a highly inductive load (L1) with the inductor voltage clamped \nby fly-back diode D1. In the absence of the nonlinear loci-control \nnetworks, A and B, the adverse switching loci shown in Figs. 3b and 3c \nare obtained at turn-on and turn-off, respectively, of transistor Ql. \nThe peak and average switching losses in transistor Q1 are 1500 W \nand 45 W, respectively. Inclusion of networks A and B yields the im- \nprovements in switching loci shown in Figs. 3b and 3c. This reduces \nboth the peak and average transistor switching losses by a factor of 10. \nThe total switching losses are reduced by a factor of 3, taking into \naccount the losses in the added resistor, Rl. The achieved conversion \nefficiency of 90 percent (96 percent for the switching regulator alone) \nis considered quite high at an operating frequency of 40 kHz.\n\nThe desired output voltage-current characteristic for a converter \nintended to feed a line of maximum length is illustrated in Fig. 4. There \nare three modes of regulation corresponding to the three linear segments \nof the voltage-current characteristic.\n\nNormal operation of the converter is in the middle section called \nthe impedance mode. The slope resistance of this section, 3370 ohms,\n\n/ NETWORKS SWITCHING \nLOCUS WITHOUT \nNETWORKS \n_ IMPROVED ee \n/ SWITCHING ing \n; LOCUS\n\nFig. 3\u2014(a) Switching regulator with locus-control networks. (b) Q1 switching \nlocus at turn-on. (c) Q1 switching locus at turn-off.\n\nis chosen to limit the variations in cable current to +110 mA in the \npresence of de earth potentials up to 12 V per mile. The \u2018\u2018level\u201d\u2019 of the \nimpedance segment is adjustable in the field to accommodate cables \nof various lengths and to provide means for balancing the voltages \nproduced by the four converters in the series loop. None of the other \noutput characteristics, alarm, or protective shutdown levels require \nadjustment in the field. Instead, the necessary precision (approxi- \nmately +2 percent) is built into the equipment using precise integrated \nvoltage regulators, thin-film tantalum resistors, and integrated opera- \ntional amplifiers.\n\nAt current levels above 1000 mA, the regulating mode automatically \nchanges to the current-limiting segment, where the slope resistance \nincreases to 24,500 ohms. This limits the output current to less than \n1050 mA for any short-circuit fault on the line. Normal operation in a\n\nOUTPUT CURRENT NORMAL \nCURRENT REGULATING MODE - OPERATING POINT \nt (24,500 OHMS) \u201c4 (MAXIMUM \n0 N 1000mA 7 i eang\n\nmode with such a flat (i.e., high-impedance) slope would be undesirable \nbecause of the difficulty of maintaining good voltage balancing be- \ntween the four converters. At output-voltage levels above 1520 V, \nthe regulating mode automatically changes to the voltage-limiting \nsegment, where the slope resistance decreases to 43 ohms. This limits \nthe converter output voltage to 1550 V while power is being turned up. \nWithout this feature, the first converter turned on in the series loop \nwould deliver higher output voltage in attempting to supply power \nto the complete loop. When the converter is turned on, the output \nvoltage of the converter is increased slowly in the open-loop control \nmode over a period of 15 seconds to minimize the voltage and current \ntransients applied to the system and to enable the feedback regulator \nto take control smoothly and restrict the output voltage and current \nlevels not to exceed the characteristic shown in Fig. 4.\n\nThe circuit implementation of the triple-mode regulator is indicated \nin Fig. 5. The output voltage and current of the converters are sensed \nusing saturable reactors to achieve dc isolation. A negative de voltage \nproportional to converter output voltage is presented at point a and a \nnegative dc voltage proportional to converter output current is pre- \nsented at point b. These two voltages are compared to a reference \nvoltage in three different ways at the inputs to each of three \u201c\u2018current- \nsumming\u201d operational amplifiers. The outputs of the three operational \namplifiers are combined via diodes D1, D2, and D3 to control the\n\noutput of a 40-kHz pulse-width modulator which, in turn, controls the \noutput voltage and current of the dc-to-de converter. The three diodes \nassure that only one operational amplifier (the one with the highest \nlevel) controls the converter. In the current-limiting mode, operational \namplifier No. 2 is controlling. Note that the output-current level is \ncompared to a reference level at current summing point c. In the \nvoltage-limiting mode, operational amplifier No. 1 is controlling with \nthe converter output voltage compared to reference at current-summing \npoint d. In the impedance mode, operational amplifier No. 3 is in \ncontrol. In this case, a linear combination of the output voltage and \ncurrent signals is compared to reference at summing point e. The slope \nof the converter output characteristic in the impedance mode is \ndetermined by the ratio of the combining resistors R1 and R2.\n\nVOLTAGE CURRENT \nSENSING SENSING \nSATURABLE SATURABLE \nREACTOR REACTOR \n=\u2014_\u2014 \\ 7 \nTO CABLE \nR \nE \u2014 \n140 V Cc i \nco) \nT \nTO GROUND \nVOLTAGE ip & io\n\nThe feedback regulator circuit and the ramp generator (used for \nsoft starting) were realized as a hybrid integrated circuit using beam- \nlead operational amplifiers, thin-film tanalum resistors, and ceramic- \nchip capacitors. The resistors in the current-summing networks are \nmatched to 0.25 percent using an anodizing process.\n\nGeneral alarm and protective features were included in the converters \nas follows:\n\nLow-current alarm (\u20145%) High-current shutdown (+32%) \nHigh-current alarm (+7%) High-voltage shutdown (1750 V)\n\nA circuit is also employed to detect line-circuit fluctuations resulting \nfrom an arcing fault in the repeatered line and to activate converter \nshutdown. As a supplementary measure to protect personnel from \nhigh-voltage hazards, a circuit is provided which prevents turn-up \nof the converter if the coaxial line has an open circuit. These functions \nare realized with three additional hybrid integrated circuits. The \nprecision of these circuits (+2 percent end of life) is such that no field \nadjustments are needed. Means are provided for simple (pushbutton) \n20, no-go in-service tests of these protection and alarm functions.\n\nThe function of the ground-protector circuit shown in Fig. 1 is to \nautomatically ground or unground the system\u2019s floating-ground point \nunder appropriate conditions. When the converters are turned down, \nthe floating point is grounded. The ground is automatically opened \nafter two conditions are satisfied: (7) the sum of the magnitudes of the \ntwo line currents exceeds 1500 mA and (77) the difference between \nthe magnitude of the two line currents is less than 34 mA. This assures \nthat all four converters have been energized and that the opening of \nthe ground connection will not produce a voltage at the floating \nground greater than 129 V.\n\nThe floating point is automatically grounded after a short delay if \nthe voltage exceeds 250 V to ground or instantaneously if the voltage \nexceeds 800 V +15 percent to ground. Possible causes of voltages \nsufficient to actuate the automatic grounding circuit are a converter \nfailure, a line fault, the appearance of an abnormal de earth potential \nbetween stations, a nearby lightning strike, or abnormal 60-Hz \ninduction. As in the case of other automatic features, a pushbutton,\n\nin-service test is provided to verify the proper functioning of these \nfeatures.\n\nVil. SYSTEM RESPONSE TO DC EARTH POTENTIALS BETWEEN \nPOWER-FEED STATIONS\n\nAs noted earlier in this paper, abnormal solar-flare activity can \ncause the appearance of earth potentials up to 12 V per mile. Such an \nearth potential appears as a voltage between station grounds as shown \nin Fig. 6. The effect of earth potential will be discussed for the case of a \nmaximum-length system where the effect is maximum. Until the earth \npotential reaches 3.33 V per mile (250 V for 75 miles), the earth \npotential appears at the floating ground point. The line currents are \nnot affected since the system is grounded at only one point. As the \nearth potential increases further, the threshold for the automatic \ngrounding circuit is exceeded and the system becomes grounded at \nboth stations. The earth potential now appears as an aiding voltage \nin one loop (cable plus two converters) and as a bucking voltage in \nthe other loop (see Fig. 6). Thus, in a pair of cables, the cable current \nincreases in one cable and decreases in the other.\n\nUpon taking into account the nonlinear output characteristic of \nthe converters and the major nonlinearities in the repeaters, the \noverall effects of de earth potentials are shown in Fig. 7. The hysteresis \neffect is the result of the margins purposely designed into the automatic\n\ngrounding and ungrounding circuits to avoid unstable operation. The \ncurrent variations in the cables are less than 110 mA for earth po- \ntentials less than 12 V per mile (900 V in a 75-mile power-feed section).\n\nThe appearance of a burst of 1500-V, longitudinal, 60-Hz induction \nnear a power-feed station will cause a peak line voltage to ground of \n2300 V and a peak line current of 2.2 A. If the 1500-V, 60-Hz induction \nis closer to the center of a power-feed section, the peak voltage will \nbe less, but the peak current can increase to 3.5 A.\n\nThe action of the diodes in the converter output rectifiers is such \nthat the effect on the converter is similar to the application of an \nalternating series of open and short circuits. Fast current-limiting \ncircuits are included in the converter to reduce the peak currents in \nthe switching transistors. A delay is included in the high-current \nshutdown circuits to avoid an unwanted power shutdown during such \nbursts of 60-Hz induction, which are expected to be less than 100 ms \nin duration.\n\nIX. HIGH-VOLTAGE PATCHING ARRANGEMENT FOR SPARE CONVERTER \nAND TEST LOAD\n\nA spare converter is provided for use in quickly restoring power to a \nline in the event of failure of a regular line-feed converter. A manual\n\nhigh-voltage patching arrangement is provided to permit connecting \nthe spare converter to substitute for the original converter for any \ncoaxial line and, at the same time, connecting the original converter \nto the test load contained within the spare converter. To achieve this, \nthe output of the spare converter and the test load are multipled to \njacks in a protected area at the top of each converter, as shown in \nsimplified form in Fig. 8. There are two types of plugs shown in Fig. 8, \nPl and P2. The plugs are shown in their normal positions in Fig. 8. \nNote that there is a Pl type plug for each converter normally associ- \nated with a particular coaxial line, but only one P2 type plug which is \nnormally associated with the spare converter. In the normal state, \neach individual converter feeds a particular line and the spare con- \nverter 1s connected to its internal test load. Under this condition, the \nspare converter may be exercised into the test load to assure its \noperability should it be needed to substitute for an individual con- \nverter. The spare converter also provides a means for checking opera- \ntion of the spare converter plug-in units at each station.\n\nWhen an individual converter fails, the Pl plug from the failed \nconverter is manually interchanged with the P2 plug from the spare \nconverter. As can be seen from Fig. 8, the failed converter will connect \nto the test load and the spare converter to the coaxial line. After the \nfailed converter has been repaired and tested, power is turned off on \nthe line and the patch plugs are restored to their normal positions.\n\nThe patching is somewhat more complex inasmuch as there are both \npositive and negative converters for the individual lines and the \nconverter output polarity of the spare converter is reversible. The\n\nREGULAR LINE FEED CONVERTERS SPARE CONVERTER \nFEEDING INDIVIDUAL CABLES AND TEST LOAD \n(USUALLY ONE PER STATION)\n\nP1 plugs from positive converters are slightly different from P1 plugs \nfrom negative converters and this difference is used to control the \noutput polarity of the spare converter when it is used to power a \ncoaxial line. There are safety interlock features built into the patch \nplugs to assure that converters cannot be energized unless the plugs \nare in valid positions and the covers over the patch panels are closed. \nMajor design objectives for the patching arrangement were to keep it \nsimple and safe for use by the telephone craftsperson in the field and to \nfacilitate the addition of converters without requiring a power shut- \ndown on all cables at a station.\n\nThe physical design of the line-power-feed equipment for the L5 \nsystem was based on the following factors:\n\n(t) Two power sources (converters) and their associated ground \npanel are always provided as a group to power a pair of re- \npeatered lines (see Fig. 1), and thus can be manufactured and \nsubsequently installed in the field as a single entity.\n\n(it) The two paired power sources (converters) have the same \noutput voltage magnitudes but different polarities.\n\n(i2t) The required line-feed voltages can be predetermined and the \npower-feed equipment can be ordered from the factory with \nthe proper output-voltage capability.\n\nThese factors resulted in a physical design which integrates two power \nsources (converters) and a ground panel into a single structure (see \nFig. 9). While the structure has well-partitioned volumes for each \nfunction\u2014converters and ground panel\u2014they are not distinct, separ- \nable, physical entities.\n\nThe physical design of the line feed has been influenced, in detail, by \nother factors. For instance, the L5 line-feed equipment had to be \ncompatible with L4 line-feed equipment in some respects to facilitate \nstation rearrangements when L4 lines were converted to L5. The \nheight and width had to be the same as in L4 or some rational sub- \nmultiple thereof. The depth of the L5 equipment was designed to be \napproximately one-half the depth of the L4 power-feed equipment.\n\nTo facilitate service restoral and maintenance, replaceable plug-in \nmodules and a spare converter were included in the design. Personnel \nprotection was required because of the high-voltage potentials present.\n\nOther design factors were the usual requirements for (2) means for \nheat removal, (27) convenient cabling design, (777) human engineering \nfeatures, and (zv) esthetic considerations.\n\nThe line-power-feed equipment cabinet shown in Fig. 9 contains \ntwo converters and a ground panel. High-voltage patching compart- \nments and a common area for battery input and alarm connections are \nalso provided. Figure 9 shows the 24-V-battery input design, which is \n27 inches wide, 133 inches deep, and 7 feet high. The 140-V-battery \ninput version, designed for stations with the new 140-V de distribution \nsystem shown in Fig. 10, is narrower. The same partitioning pattern \nhas been followed in the 140-V design. Three 140-V cabinets placed \nside by side are equivalent to two 24-V-input designs and are thus \ncompatible with floor plans.\n\nNormal operating controls are mounted on the front panel. No \nother facilities for adjustments are provided within the equipment. \nTrouble indicators are a lamp at the top of the cabinet, and a lamp for \neach converter.\n\nFigures 11 and 12 show the regular line-feed cabinets with their \ndoors open to illustrate the plug-in modules. The lower doors are inter- \nlocked to deactivate the equipment (one converter at a time) since \nthe high voltages are generated in this area. Access doors to other \nhigh-voltage areas, including the patching area, are also interlocked \nand protective covers are furnished.\n\nThe cabling plan is designed to minimize cable installation within \nthe cabinet. Field-connected cable terminates at the top of the cabinet. \nThe high-voltage outputs and the connections to the spare converter \nare protected by flexible conduit and enclosed metal ducts.\n\nReplaceable plug-in modules, which contain the critical electronic \ncircuits, are used to facilitate rapid field repair and manufacture. The \nmodules for a 24-V input converter are shown in Fig. 13. The active \nelectronic circuits for each converter, except for the power stages and \noscillator, are contained on five printed-wire boards which provide the \nplant automatic features described earlier. Each board consists of a \ndouble-sided epoxy-glass board assembly with a standard plastic face\n\nplate. Connection to the unit is through printed-circuit contact fingers \nwhich plug into connectors mounted on the housing for the five \nboards. Each converter has an identical complement of five boards. \nAlthough the boards were initially designed for application in the \n24-V-input version, they were applied without change in the 140-V- \ninput version to simplify manufacture and repair. Other printed-wire-\n\nboard assemblies are component parts of the power stages, oscillators, \nand ground panels.\n\nThe feedback regulation, alarm, and protective features incorporated \nin the converters utilize four specially developed hybrid integrated\n\ncircuits shown in Fig. 14. The hybrid circuits include thin-film tantalum \nresistors with precise ratios (0.25 percent initial) to provide long-term \nstability and eliminate the need for adjustments over the anticipated \nlife of the equipment. The resistors, positioned in close proximity on a \nceramic substrate, are aged prior to final trim anodization to achieve\n\nthe required accuracy and stability. Additional advantages in circuit \nperformance are realized by including associated components on the \nsame substrate, which assures minimum-length paths for critical \ninterconnections. These additional components include beam-leaded \noperational-amplifier chips, ceramic-chip capacitors, and beam-leaded \ndiodes. Stitch crossovers are also used for interconnections on the \nsubstrate. Solder reflow is used to attach the capacitors and thermo- \ncompression bonding is used for the operational amplifiers and diodes\n\nFig. 15\u2014Unencapsulated development model of feedback regulator with hybrid \nintegrated circuitry.\n\nas well as the substrate lead frame. An early development model of \nthe feedback regulator hybrid, unencapsulated, is shown in Fig. 15. \nThe hybrid circuit is mounted to the printed-wire board by reflow \nsoldering the lead frame to presoldered terminal areas on the boards. \nThe use of hybrid integrated circuitry eliminates a significant number \nof soldered connections, which would have been required in a discrete \nrealization and thus should enhance reliability.\n\nEach converter (line and spare) has an associated high-voltage \npatch compartment containing the patch plug and mating receptacle. \nAccess to the patch plug is gained by opening an interlocked hinged\n\npanel, as shown in Fig. 16. A small window in the line-feed converter \npanel is provided for visual observation of the patch status, i.e., to \ndetermine whether the normal plug or the spare plug is in place in the \nline-feed converter. A status lamp is provided at each line-feed con-\n\ni \nTititiiitiit414++ad : \nSUCRCeseseezeesee E \n+ 43-3-$.4-4-4-4-4-4-4-4-5-4-4-4 2 q \nPett ith tt dd 1444-444 4 \nBRAVES RKR ES RROS \nPt4335-$+-44-4-1-4-4-4-4-5-0 q \n= Mt \nSRAATE CCE SAAR eee i \nPESRCCSL ESSERE \nRSRRRRERRCERR ERAS ee \nbbb bededd-b-b-5-544-4-4-44.2 zt ie \nt Lhd db$-+4-454-4444-11 ; ae: \nBOCRPSSL ESCH SSHEH \u00bb Shes \n2 Rb td bdbcb4-5-4-4-4.44-4.1 BBS \nig b db d-d-t.h-$eh4-4-4-4.4.4-4-40 \nSHEEHGniness =| gs \neee ee a poteratiti titi ty ee : \nRASHAT SORT FEO eEEE SUVKARAROTA RARER 3 \u00a5: \neet ttt ttt e4 SERSSR ESSEC ES Ana oe \nRosaseasaaanssace e ptittttttt ttre ry i z \n:\n\nOye SERRE RE ye \noo $SSENe SaneNes sane \npeetitiii tim tit ttt \nbps TTT TTT \nPA an i RRR X \u00bb\n\nverter and indicates whether the spare converter has been placed in \nservice as a substitute for a regular line feed or is available for \nassignment.\n\nThe spare converter services a number of line-feed converters, the \nquantity being a function of the acceptable system repair time. It \nshould be noted here that the transmission-system-protection switching \nfacility maintains service in the event of a converter failure. Arrange- \nments are available to provide a multiplicity of spare converter systems \nfor regular line feed within a single office if system-reliability require- \nments warrant.\n\nSpare converters are available for \u201424-V and 140-V input voltages. \nThe cabinet sizes are identical to those of the regular line-feed \nconverters.\n\nThe precision obtained through the use of hybrid-integrated-circuit \ntechnology permitted the inclusion of many automatic control features \nin the power-supply system for the L5 repeaters. It is expected that \nthe resulting simplification of system operation and the minimal \nnumber of adjustments required in the field will contribute significantly \nto overall system reliability.\n\nA. P. Walsak was responsible for physical design, and circuit de- \nsign responsibilities were shared by R. Ostapiak, P. P. Untamo, R. E. \nSchroeder, and J. R. Meszar.\n\n2. E. T. Calkin and B. H. Hamilton, \u2018\u2018Circuit Techniques for Improving the Switch- \ning Loci of Transistor Switches in Switching Regulators,\u2019 IEEE Conf. Record \nof 1972 Seventh Annual Meeting of IEEE Industry Applications Society, \npp. 477-484.\n\n. Ibid., \u201cA Conceptionally New Approach for Regulated de to de Converters \nEmploying Transistor Switches and Pulse Width Modulation,\u201d pp. 485-494.\n\nCopyright \u00a9 1974 American Telephone and Telegraph Company \nTHE Bey System TECHNICAL JOURNAL \nVol. 53, No. 10, December 1974 \nPrinted in U.S.A.\n\nA new system has been designed to provide protection switching capa- \nbility for the LS repeatered line. This line-protection switching system, \ncalled tpss-3, provides one standby line for protection of as many as 10 \nservice carrying lines. Switching equipment ts located in terminal, ter- \nminal main, and switching power-feed stations. Economic and reliability \nconsiderations allow switching spans of up to 150 miles. Either tempera- \nture pilot deviations or excessive total signal energy initiates automatic \nswitching. Manual control of all switching functions 1s provided locally \nand via telemetry.\n\nA dedicated pcm signaling system maintains interstation switching \nsystem coordination. Identical information ts transmitted in complemen- \ntary form over two adjacent L5 channels. Parity and complementary \nchecking as well as automatic retransmission of failed codes enhances \nerror immunity.\n\nThe need for new switching capabilities and the desire for a modern \nswitching facility for the L5 Coaxial-Carrier Transmission System \nhave led to the development of a new line-protection switching system \ncalled tpss-3. Several objectives were paramount at the beginning of \nthe new development. One was to design an interstation signaling \nsystem that is relatively immune to both line noise and hits (short \ninterruptions of transmission), since a signaling error can cause a \nservice failure because of improper switch activation. Another objec- \ntive was to provide a design that requires a minimum of effort for the \naddition of new lines.\n\nA newly installed coaxial-carrier system usually has only one or \ntwo of the ultimate regular lines equipped and connected. Additional \nlines are equipped and connected to the system as traffic growth \nrequires. Thus, the process of adding lines to a working system is \nimportant. The upss-3 bay is factory-wired and tested for its full \ncapacity of 10 regular lines and one standby line. Equipping switching \ncapability for newly added regular lines on an L5 system requires only \nreplacing dummy plug-in modules with active modules in the Less-3 \nbay and adding a few control leads to the new line bays. No new intra- \nbay wiring changes or connections are involved.\n\nA new series of coaxial switches has been developed using diodes as \nthe switching elements. These switches were designed specifically to \ncomply with the L5 system bandwidth and modulation performance \nstandards.\n\nA reliability analysis of estimated failure rates, mean time to failure, \nand service outage was made for the Ld coaxial system. This included \nestimates relating to the L5 line, main-station, and switching equip- \nment. These studies were then projected to estimate average outage \ntimes for various types of switching sections through the use of sys- \ntem models. In addition, the probability of cable damage has been \nanalyzed and is included in the overall system outage predictions. An \noutage is a service loss, ie., a failure not remedied by protection \nswitching.\n\nThe results of this study show that the use of average switching \nintervals of 120 miles, total system length of 4000 miles, and one pro- \ntection coaxial line for 10 regular lines will result in adequately small \noutage times. With this arrangement, service outage time because of \nindividual line failures is significantly less than outage time caused by \nmassive failures such as a man-made fracture of the buried cable. \nModification of switching section length and spare-to-regular ratio will \nnot reduce the latter outage time, which is controlling.\n\nA broadband transmission system such as L5 is composed of two \nbasic elements: the office or main station equipment and the repeatered \nline equipment. The main station equipment combines and separates \nthe system message content and performs functions such as power \nfeed, equalization, line switching, and fault location. The line equip-\n\nment provides the transmission equipment required to connect main \nstations together. The line-protection switching system provides the \nbackup protection to guarantee transmission between main stations \nwhen one of the regular lines fails. This is accomplished in L5 by \nutilizing the LPss-3 equipment to control and cross-connect the main- \nstation line-connecting equipment to the standby lines provided in the \ntransmission medium. (In the coaxial cable, one unit coaxial is pro- \nvided for standby or protection use in each direction of transmission, \nsince L5 repeaters are unidirectional.)\n\nFigure 1 is a diagram of an L5 line-switching section. It has been \nsimplified to show only one regular line and the standby line for each \ndirection of transmission. The protection equipment is located within \nthe main stations at each end of the switching section (up to a maxi- \nmum of 150 miles apart). Within the station, the protection equipment \nis divided into two basic locations. The line failure detectors and \ncoaxial switches are located in the individual line bays, while the \ncommon switching equipment is in the Lpss-38 bay. The inputs to the \ndetector circuits and the access of the interstation signaling commands \nare derived from the line-connecting circuits.\n\nThe detector circuits are located in the switch initiator, Fig. 2. \nThese include both transmitting and receiving detectors. The receiving \ndetectors monitor the 42.880-MHz line pilot and the total system \naverage power. If the pilot deviates more than 5 dB from nominal for \nmore than 2.5 ms, a switch request is initiated (whether a switch takes \nplace depends on many conditions and is discussed in detail later). If \nthe average power exceeds its threshold, a switch request is issued, \nfollowed 50 ms later by a termination request (as before, the results of \nthese requests depend on other conditions and are covered in detail \nlater).\n\nTransmitting detectors monitor the 42.880-MHz pilot. Each time \nthat pilot exceeds the +5-dB limit, a 50-ms signal is generated in the \nLPss-3 transmitting switch-control circuits, preventing a line switch \nfrom taking place during that interval. The assumption is that the \nfailure has occurred in a previous switching section (the probability of \nsuch is very high) and that section\u2019s protection-switching facility will \nremove the apparent problem. If both transmit and receive detectors \nin a switching section persist in the failed state beyond a reasonable \nswitching interval, a switch is completed in that section to obtain the \nbenefit of the independent standby line temperature pilot source.\n\nAA AA REGULAR AA AA \nee \nH HYBRID TRANSFORMER MCS MESSAGE CUTOFF SWITCH \nS| SWITCH INITIATOR LTS LINE TERMINATE SWITCH \nTLS TRANSMIT LINE SWITCH DIR DIRECTOR SWITCH\n\nTwo basic types of coaxial switches are used in the protection of L5. \nOne is the director switch, a 1 X 11 solid-state switch without any \ncontrol of its own; it is a slave to the switches with which it operates. \nOne director is located at each end of the standby line (Fig. 1). Its \ncommon port is connected to the line equipment, and its 11 other ports \nconnect to the line-connecting circuits via line switches. With no pro- \ntection switch in force, the standby line-connecting circuits are con- \nnected to the standby line. With a protection switch in force, the \ndirectors route the message from the transmitting line-connecting \ncircuit for the failed line, through the standby line, and on to the \nreceiving line-connecting circuit for the regular line.\n\nThe remaining line switches are combinations of the basic solid- \nstate T structure. This structure uses two series elements and a shunt \nelement. In the pass condition, the series elements provide low loss and \nlow distortion, while the shunt element has high loss. The stop condi- \ntion is just the opposite, with the series elements providing high inser- \ntion loss to the signal and the shunt element acting as a short to ground \nfor any signal passing the first series element. The line switches are \ncomposed of from two to four T\u2019s. In Fig. 1, the T\u2019s have been simpli- \nfied to show only their normally released state. (The straight line across \nthe path indicates the pass condition of a T, while the X indicates the \nstop condition.) The transmitting and receiving line switches connect\n\nthe line-connecting circuits to the standby line via the director switches. \nAs pointed out previously, the director switches are slaves to the line \nswitches. In fact, the solid-state elements of the director switches are \nthe final elements of the T\u2019s of the line switch to which they connect. \nControl of the director is by a de current from the line switch to the \ndirector over the center conductor of the coaxial cable carrying the \nmessage between them. Two other switches are used in protection \nswitching, one located at each end of the switching section. The trans- \nmitting end of each line, both regular and standby, contains a message \ncutoff switch. The function of this switch is to remove the message \nload from the line, under controlled conditions, to permit line measure- \nment for special purposes\u2014equalization and, if necessary under high \nnoise conditions, fault location. The line-terminating switch is located \nat the receiving end of the section, immediately following the receiving \nline switch. Its function (detailed in Section 4.9) is to prevent the \npropagation of system overload conditions.\n\nHigh noise immunity and low power dissipation are more important \ncharacteristics than speed of operation in Lpss-3 circuitry. With this \nin mind, saturated logic integrated circuits have been used in Lpss-3 \ndesigns. Timing circuits and NAND logic functions are provided by a \nfamily of diode-transistor logic (pTL) circuits. Resistor-transistor logic \n(RTL) circuits are used in the signaling system to provide Nor and \nEXCLUSIVE OR logic in addition to clocked flip-flops for parallel-to-serial \nand serial-to-parallel conversion. Set-reset flip-flops from both families \nare used as memory elements throughout the system.\n\nThere are two varieties of L5 offices from an Lrss-3 point of view: \nswitching power-feed stations and terminal or terminal main stations. \nThe physical arrangement of equipment in the Lpss-3 bay is the same \nfor either type station, but the manner in which Lpss-3 interfaces with \nthe line equipment differs. The Lpss-3 bay is mounted in the same \naisle as its associated line bays. The reasons for choosing this approach \nare twofold: it was possible to design one universal arrangement that \nfits all applications, and the maintenance operation is simplified by \nalways keeping functionally associated equipment in the same aisle \nalignment. In the terminal and terminal main stations, the Lrss-3 \nequipment is associated with a line section. In the switching power-feed \nstation, the Lrss-8 equipment is associated with one direction of\n\ntransmission in two line sections because the line bays are through- \ntransmission units. No message administration is performed at these \nstations, so the optimum line arrangement is to have the receiving and \ntransmitting circuits all in the same bay.\n\nProtection switching operations require coordinated actions at \ndiverse physical locations. Transfer of service from a regular to a \nstandby line, for example, requires properly sequenced operation of \nline switches separated by up to 150 miles. This interstation coordina- \ntion is maintained by the Less-3 signaling system, a self-contained sub- \nsystem that generates and detects digitally encoded carrier signals. \nThe L5 system itself is the medium for transmittal of these signals; \nthe associated frequency band is blocked at each switching main \nstation to eliminate signaling interference among switching sections.\n\nThirty-nine digital code words are in active use within the system; \neach has a specific purpose and consists of seven bits (e.g., 1100101). \nThe digital rate is 2 kilobits per second. When signaling is not taking \nplace, an alternating 1-0 pattern is transmitted to allow immediate \nalarming whenever interstation signaling continuity is lost. The seven- \nbit code word is preceded by two successive 1\u2019s to mark the beginning \nof the code word.\n\nThe transmitter generates an FsK signal with a 1 bit corresponding \nto energy at 68.76 MHz and a 0 bit corresponding to energy at 68.78 \nMHz. The receiver separates and independently detects the two \nchannels. With no transmission or detection errors, the two channel \noutputs are complementary.\n\nError-free detection is of prime importance. With this in mind, the \n39 code words have been chosen to have even parity to allow detection \nof errors resulting from the permutation of a single bit. Additional error \ndetection is made possible through complementarity checking of the \ntwo channels for each bit of the code word. A transmission medium \ndisturbance must simultaneously permute a 0 to a 1 in one channel \nand a 1 to a 0 in the other channel in at least two bit positions to cause \nan erroneous receiver output.\n\nError detection at the receiver causes the incoming word to be re- \njected; no attempt is made at error correction. To prevent switching\n\nsystem lock-up under this condition, the transmitter automatically \nretransmits the code word every 15 ms until evidence is received that \nthe command has been properly decoded at the remote location. The \nsequence is as follows: An originating command is transmitted to the \nremote station to cause the change of state of a line switch. With \nthis action completed, the remote signaling transmitter generates an \nanswering command that is routed back to the originating location. \nThe proper decoding of the answering command stops the retrans- \nmission of the originating command. At the remote location, cessation \nof the incoming originating command is taken as evidence that the \ndecoding of the answering command has been successful, and the \nretransmission of the answering command is stopped. Every signaling \noperation consists of a round-trip operation as described above to \nverify the completion of each originating command.\n\nSpecial provision has been made to allow in-service exercising of \nthe signaling system to verify proper operation. Local or remote (via \ntelemetry) manual action causes the transmitter to generate a TEST \ncommand, which is decoded at the far end of the switching section. \nA TEST RECEIVED command is then returned, completing the round \ntrip and causing the signaling to return to idle. Failure of any involved \ncircuitry along the route will cause a TEST FAIL lamp to illuminate at \nthe originating point. This test can be performed at any time, even \nwith a protection switch in effect, and checks most of the circuitry \ninvolved with signaling.\n\nHach set of up to 11 coaxial lines in a given transmission direction \nhas an associated signaling transmitter and an associated signaling \nreceiver. At the transmitting end, a switch signal distribution unit \ncontrols the application of the switch signals to the transmitting \ncoaxial lines. The signals are normally introduced on to all outgoing \nregular lines but, as explained in Section IV, the signals are sometimes \ntemporarily introduced on to only the standby line or one regular line. \nThe signaling receiver is electrically connected to only one line by \nthe signaling receiver switch unit, which normally provides connection \nto the lowest-numbered \u2018good\u2019 regular line. In this case, \u2018\u2018good\u201d\u2019 \nimplies the line is in service (e.g., not manually switched out of service) \nand not failed. Under certain conditions, the receiver is temporarily \nconnected to the standby line.\n\nIn this section, we examine the operations of Lpss-3 in detail through \nthe use of flow charts. Line switching, termination, and message cutoff \noperations are discussed after additional preliminary concepts are \nestablished.\n\nA switching section consists of line switches and associated switching- \ncontrol circuits at both the transmitting and receiving ends of the \ncoaxial lines. The receiving Lpss-3 circuitry has primary control of all \nswitching actions in that switching section. The transmitting LPss-3 \ncircuitry merely responds to directions from the receiving circuitry. \nSwitching activity, manual or automatic, is never initiated at the \ntransmitting end.\n\nIn upss-3 jargon, the term, \u201cpriority,\u201d is associated with internal \nLPss-3 operations, which establish the right for an operation to take \nplace. There are two types of priority. A failed regular coaxial line \nattempts to establish switching priority as a first step in executing \nan automatic switch. Any operation requiring the use of the signaling \nsystem must obtain signaling priority before being given access to \nthe signaling transmitter. In both cases, priority avoids the system \nconfusion that would otherwise result when independent, nearly \nsimultaneous operations take place. The priority circuits in LPss-3 \n(Fig. 3) are N-input, N-output logic circuits designed so that a driven in-\n\nFig. 8\u2014Priority circuit. Only one output (corresponding to a driven input) may \nbe operated regardless of the input time sequencing. Inhibit prevents new outputs, \ncancel inhibits new or established outputs.\n\nput will result in the corresponding output going to the operated state. \nWith multiple driven inputs, there will be only one operated output.\n\nIf the inhibit input is activated, no new output can become acti- \nvated ; if the cancel input is activated, any previously activated output \nis cancelled. Consider, for example, the case in which regular line No. 5 \nfails at approximately the same time as line No. 2. This will result in \nthe No. 2 and No. 5 inputs of the switching priority circuit being \ndriven. Depending on the input sequence, either the No. 2 or the No. 5 \noutput will be driven, but both can never co-exist, even for short \nintervals. Thus, only one line will obtain clearance to establish a \nswitch. At certain times (e.g., standby line failed) Lpss-3 will refuse \nto grant new switching priority, but will not defeat established priority. \nThis causes the inhibit to be driven. Both new and established priority \nmay be defeated by the cancel input. Lock normal (Section 4.8) is \none such condition.\n\nThe major steps in accomplishing an automatic transfer of service \nfrom a regular line to the standby line are diagrammed in Fig. 4. The \nline bay switch initiator detects a line failure and automatically causes \nan LPpss-3 switch request for that line. Switching priority will be \nestablished if Lpss-3 can provide switching. Action will stop if another \nline has switching priority, if the standby line is failed, or if lock \nnormal is in effect. Signaling priority is established next. If another \noperation is in the process of using the signaling transmitter, another \ndelay is encountered until the transmitter is available. At that time, \na specific digital code word called the 1p (there are 10 such code words, \none for each regular line) is sent to the transmit end of the switching \nsection over all regular lines in that direction. The 1p is decoded by \nthe signaling receiver, causing the appropriate transmit switch to \noperate. An Lpss-3 indication Tso (transmit switch operated) is \nilluminated. The message is now being introduced to both the regular \nand standby lines. Signaling priority is sought at the transmit end so \nthat the answering command verifier may be sent back to the receiving \nend. This one command is used in common by all 10 lines. Receipt of \nthe verifier is taken as evidence that the transmit switch has operated \nproperly. The receive switch is operated, completing the switching \nsequence. An Lpss-3 indication osa (out of service automatic) is \nactivated.\n\nThe route of the verifier command from transmitting to receiving \nends of the switching section deserves special attention. The verifier is\n\nintroduced into the transmit line-connecting circuit of the line to be \nswitched. The verifier then propagates through the operated transmit \nswitch on to the standby line. Meanwhile, at the same time that the \nID was sent out from the receiving end, the signaling receiver was \nconnected to the standby line in anticipation of the verifier returning \nfrom the transmit end of that line. This routing gives an added measure \nof confidence in the serviceability of the path to which the message is \nto be transferred. If the verifier is not successful in traversing this path, \nthe receive switch will not operate.\n\nElapsed time from failure to switch complete is typically less than \n12 milliseconds:\n\n2.2 ms Signaling, verifier (zero length system) \n0.8 ms Verifier propagation (150 miles) \n0.3 ms Delay at receiving end\n\nCompletion of an automatic switch results in a minor office alarm, \nsince switching protection has been automatically used. If, for any \nreason, a switch request does not result in a completed automatic \nswitch within 1 second, an Lpss-3 TF (total fail) indication and ac- \ncompanying major office alarm result, since service is presumably lost.\n\nIn many cases, the 42.880-MHz pilot that is detected for switch \ninitiation traverses more than one switching section. The switch \nblocking feature prevents several tandem sections from switching on a \nsingle failure.\n\nThe temperature pilot level in the regular transmitting line-con- \nnecting circuits is detected by the switch initiator. When the pilot \nexceeds limits, the initiator so informs the Lrss-3 transmitting switch \ncontrol circuitry. The transmitting switch is then inhibited from opera- \ntion for the next 50 milliseconds. The receiving switch initiator circuits \nat the next office will also detect the failure, causing an 1p to be sent \nback to the transmit end. Since a block is in effect in the transmit \nswitch control circuitry, the transmit switch will not operate. Under \nthis condition, the BLOCK ON command is sent back to the receive end \nof the switching section instead of the verifier. At the receive end, the \napparently failed line is deprived of switching priority for 100 ms. \nDuring this interval, the switching system is free to execute other \noperations. After the 100-ms interval, the involved line is allowed to \nseek switching priority if it is still failed. Note that the section with \nthe actual failure is allowed to switch, since its transmitting detector \nexperiences no failure. Subsequent switching sections undergo a 12-ms \nfailure (typical) at both transmitting and recciving detectors and try \nto switch, but are blocked. After the 100-ms lockout, the lines are no \nlonger failed because of the completed switch in the failed section, and \nno further action results. If the switch does not complete in the failed \nsection, the next section will switch after the 100-ms interval.\n\nWhen a failed line that has an automatic switch in force returns to \na nonfailed condition as indicated by the switch initiator, an automatic \nswitch release sequence starts. Figure 5 summarizes the operation.\n\nFor the first 30 seconds after the line has returned to normal, no \naction takes place. If the line momentarily fails during this interval, \nthe full 30-second count is restarted. After the line has been nonfailed \nfor the full interval, the receiving switch is released. This transfers \nservice back to the regular line, since the operated transmit switch did \nnot remove the message from the regular line, but merely caused a \ndual feed of message on the regular and standby lines. The remaining \nsteps are taken to clear the switching system for the next operation. \nSignaling priority is obtained to send the RELEASE command to the\n\ntransmit end of the section, causing the transmitting switch to be \nreleased. The release verifier is then sent back to the receiving switch \ncontrol circuitry to verify the successful release of the transmitting \nswitch. If the release verifier fails to return within 5 seconds of the \nreceiving switch being released, the transmit switch release fail (TsRF) \nindicator is activated, and the minor office alarm is sounded in the \nreceiving office. With this condition in effect, the system will not \nattempt to establish another protection switch, since a transmitting \nswitch may still be operated. The rsrr condition is cleared by the \nsuccessful completion of a manual release (Section 4.7).\n\nManual switching capability is provided by Lrss-3 so that nonfailed \nregular L5 lines may be taken out of service for equalization, main- \ntenance, or measurement. The steps taken to establish a manual \nswitch are identical to those summarized in Fig. 4 for automatic \nswitching, except that the action is started manually rather than \nautomatically, the resulting indication is osm (out of service manual), \nnot osa, and no office alarms result.\n\nIn general, manual operations on Lpss-3 override automatic opera- \ntions. A manual switch may be executed on one line while another has \nan automatic switch in force. The manual switch initiation causes \nthe established automatic switch to be released before the manual \nswitch is executed. This capability allows operations personnel to \ncontrol which of several failed lines is to be switched to the standby \nline. If the manual switching procedure is executed on a line that has \nan automatic switch in force, the control of the protection switch is \nmade manual (osa extinguishes, osm lights), and the line switch will \nnot release automatically after the failure clears.\n\nThe release of a manual switch is identical to the automatic release \nsequence of Fig. 5 and Section 4.5, except that the action must be \nstarted manually and the 30-second delay is bypassed. If a manual \nswitch on a nonfailed line is released with another regular line failed, \nan automatic switch on the failed line will result. The release of a \nmanually switched line that is failed results in a transfer of control \nto the automatic mode (osm extinguishes, osa lights), but does not \ncause a release. A manual release may be performed at any time, not \nonly when a manual switch is in effect. This allows operations personnel \nto clear the TsRF condition, should it occur (Section 4.5).\n\nThe lock normal condition is manually initiated to prevent the use \nof the standby line for service. This feature is used while the standby \nline is being equalized, for example. The essence of lock normal is that \nswitching is inhibited. Restoration lock normal also prevents switching, \nbut is intended for use when the standby line is being used to carry \nservice from a failed facility that is not normally associated with that \nstandby line. Unlike lock normal, restoration lock normal is activated \nand released from the restoration patch bay and not from the Lpss-3 \nbay.\n\nLock normal may be activated at any time that neither a manual \nswitch nor restoration lock normal is in effect. Restoration lock normal \nis inhibited by a manual switch, lock normal, or a standby line termina- \ntion (Section 4.9).\n\nActivation of either lock normal or restoration lock normal results \nin the release of any automatic line switch that may be in force. \nRelease of lock normal or restoration lock normal allows the failed \nline to again complete an automatic switch.\n\nThe line termination capability of Lpss-3 provides a check against \nthe propagation of overloads on the Ld lines. Each line has a terminate \nswitch located on the output side of the receiving switches (Fig. 1). \nEach regular line terminate switch may be set in the manual or auto- \nmatic mode. Under the normal automatic control, any terminate \nswitch that is connected to an overloaded line by the receiving switch \nmatrix is operated, thereby removing the overload from subsequent \nsystem components. As previously discussed, an overload on a regular \nline will cause an attempt to switch that line out of service. The \nautomatic line termination will result only if the switching action (or \nlack of it) results in a persistent overload condition at the output of \nthe receiving switch matrix. When the manual termination mode is \nselected, the particular terminate switch involved may be manually \noperated at any time that the terminate switch is connected to a \nfailed line, where the failure may be due to either pilot deviation or \nsystem overload. Automatic terminations result when switching action - \ncannot stop an overload, while manual terminations are enabled when \nswitching action cannot remedy a failure, either pilot or overload.\n\nThe L5 manual equalization procedure is accomplished out of service, \nwith the message removed from the line facility. A message cutoff\n\nswitch is therefore located in each transmit line connect panel. Since \nuntimely operation of this switch while the particular line is in service \nwould cause a service loss, activation of a regular line message cutoff \nswitch is enabled only when the line is manually switched or manually \nterminated. The normal procedure for removal of the message from a \nregular line for equalization or measurement is to manually switch \nthe line, then operate the message cutoff control. Enablement of the \nmessage cutoff feature for a manual termination is intended to allow \nthe removal of an overload condition from a line to facilitate measure- \nment of that line. Message cutoff capability for the standby line is also \nprovided. Enablement of the feature is caused by lock normal or \nstandby terminate.\n\nWhenever the message cutoff feature is in force, none of the pre- \nconditions for message cutoff can be released. For example, if a regular \nline is manually switched and the message cutoff is activated, the line \nwould be isolated at both transmitting and receiving ends. If the \nmanual switch is inadvertently released, service would be lost. To \nprevent this, the manual release is inhibited while the message cutoff \nand manual switch conditions exist on the same regular line.\n\nThe steps taken for a regular line message cutoff operation are shown \nin Fig. 6. Provided one precondition is met when the control is acti- \nvated, signaling priority is established and the appropriate command \nis sent to the transmitting end. The command is decoded, causing the \noperation of the message cutoff switch and the illumination of an \nindicator Mcso (message cutoff switch operated). Signaling priority is \nthen sought to return the answering signal (message received) to the \ncontrolling receiving end of the switching section. When this command \nis decoded, the Mcco (message cutoff control operate) lamp for the \ninvolved line is illuminated, and the release of a manual switch or a \nmanual termination on that line is inhibited.\n\nRelease of a message cutoff condition is also executed from the \nrecelving end of the switching section. The message cutoff release opera- \ntion is never inhibited and may, in fact, be exercised without a message \ncutoff in effect.\n\nSwitching activity often has a direct effect on the presence or \nabsence of pilots on the L5 system. The most critical pilot is the \n42.880-MHz temperature pilot. Resupply of the temperature pilot is \nalways provided when that pilot is disrupted by switching, specifically \nfor any regular line message cutoff and for a regular line termination\n\nat offices where the pilot is not blocked and reinserted in the line- \nconnecting circuit. The temperature pilot is always blocked and re- \ninserted on the standby line on the line side of the message cutoff \nswitch. This eliminates the need for standby line temperature pilot \nreinsertion. The other three E3 equalization pilots are often passed \nthrough the office. Without special provision, these pilots would be lost \nin the transmitting switching section on the standby line each time a \nreceiving section switch is completed. Because of the relatively high \nactivity of such occurrences, the three E3 pilots on the standby line are \nreinserted whenever disrupted by receiving switching activity. Except \nfor the 42.880-MHz temperature pilot, E3 pilots are not reinserted on \nthe regular lines, since a line termination is the only means by which\n\nthey are disrupted. The combination of the lack of necessity of con- \nstant presence of the three E3 pilots and the relative infrequency of \nline terminations allows this mode of operation.\n\nOne of the 10 regular lines in each direction of transmission may be \ndesignated the key line. The key line is different from all others in that \na failure on the key line causes the release of an automatic switch on \nanother line, so that the key line may be switched out of service auto- \nmatically. The key line reacts as do the other lines in all other ways. \nThe key line will not automatically overtake a manual switch, can \nitself be overtaken by manual action, and will not attempt to switch \nto a failed standby line. The circuit module with the key line feature is \nprovided only if specifically ordered by the customer. When provided, \nthe capability may be easily disabled by using a switch on the module \nor transferred from one line to another by simple module interchanging. \nFigure 7 illustrates the key-line-switching process. If switching priority \nis available when the key line fails, the line switch to the standby line is \nexecuted in the conventional manner. With switching priority un- \navailable, a release operation will be executed if the standby is not \nfailed, no manual switch or lock normal is in effect, and another line \nhas completed a line switch. When the release is complete, the key \nline is allowed to establish a line switch. Typical time for a key line \nswitch with another line previously switched is 21 ms from time of \nfailure to key line switch complete.\n\nThe terss-3 bay is illustrated in Fig. 8; one bay is required per \ncoaxial cable end at each office.\n\nPowering equipment is provided in the form of regulated de-to-de \nconverters, fuse panels, and power alarm circuits. The converters \nprovide regulated outputs of +25, +12, and +6 volts de using the \n\u201424 volts de office supply as a source. The various logic circuits \nthroughout the bay have decentralized voltage regulators to provide \nthe proper voltage levels for the integrated circuits.\n\nSignaling circuitry is housed in four shelves; terminal strip and \nsignaling receiver switch unit, switch signal distribution unit, signaling \ntransmitter, and signaling receiver. These components contain all the \nhigh-frequency circuitry associated with the generation and distribu-\n\ntion of the signaling commands and also contain modular logic circuits \nfor the control of signaling.\n\nThe indicator and control panel provides a centralized input-output \ncapability for routine switching activity. Section 5.2 provides additional \ndiscussion of this important panel and associated operations.\n\nPer-line circuits are located on the three shelves directly below the \nindicator and control panel. The lowest of these three shelves is as- \nsociated with receiving operations, the middle shelf is associated with \ntransmitting operations, and the top shelf provides interfacing for \nboth receiving and transmitting circuits with the indicator and control \npanel, the office alarms, and the telemetry systems. Whenever a new \nregular line pair is added to an existing route, a new working module is \nadded to each of these shelves. Positions for lines not yet equipped are \nfilled by special modules that allow proper switching operation for the \nequipped lines. Both the transmitting and receiving per-line modules \nhave logic disable circuits with the associated control key and indicator \non the module face plate. Activation of this feature prevents Lpss-3 \nfrom responding to the normal stimuli that cause automatic line- \nprotection switching. The intended use is for cases in which automatic \nswitching is to be prevented on a per-line basis for switch initiator \nmaintenance or repair and for lines equipped but not yet in service.\n\nCommon control circuitry is located in the bottom panel in Lrss-3. \nThe coaxial line to be connected to the signaling receiver is selected \nby circuitry on the left-hand module. A set of 11 lamps provides visual \nindication of which line is being accessed. Other functions performed \nby the common control include switching priority, signaling priority, \nreceive switch timing, release control, standby line receiving control, \nsignaling retransmission, and signaling system test control.\n\nThe indicator and control panel (Fig. 9) is the focal point of LPss-3 \nmanual switching activity and visual indications. The controls and \nindications are organized in rows and columns by functions. The top \neight rows are associated with receiving functions, and the next two \nrows are associated with transmitting functions. The next-to-last row \nprovides for signaling system control and visual displays. The bottom \nrow is a set of lamp test keys that allow a rapid check of the lamps in \nthe panel.\n\nThe columns contain per-line functions, with the exception of the \nsignaling system row and the left column, which provide common \nfunctions. The second column is associated with the standby line, \nwhile the next 10 identical columns are associated with the regular \nlines. Each position on the panel contains a key, a lamp, or a key and \na lamp. The key designations are stamped on the panel to the left of \nthe key position, with the exception of LP rst (lamp test), which is \nbelow the keys. The designations on the plastic covers are for the \nlamps under the covers. To illustrate this, the top position in the \nthird column is the sEL L MAN key; the position also contains three \ndistinct indicators, SLM, osa, and osm. The upper left position contains \nthe rus key but no indicator, while the fourth position in the left \ncolumn contains the TsRF indication but no control key. The control \nidentifications are not repeated for the regular lines. Table I defines \nthe indicator and control panel abbreviations.\n\nThe indicators provide office personnel with information regarding \nthe present state of the switching system, while the keys allow the \nstate of line switching, termination, and message cutoff to be manually \ncontrolled, as discussed in Section IV. Disconnect keys and correspond- \ning indicators are provided for receiving, transmitting, and message \ncutoff line switches (e.g., TRMTG L SW Disc and TLsp\u2014third row from \nthe bottom). When a disconnect key is activated, the associated in- \ndicator is illuminated and the corresponding line switch is forced to its \nnormal nonoperated state. This allows Lpss-3 maintenance to be \ncarried out without concern for loss of service resulting from improper \nline switch operation under abnormal switch control bay conditions, \nsuch as having circuit modules removed. A message register at the top \nof each per-line column indicates the number of completed line switches \nfor each regular line and the number of failures for the standby line. \nThese registers are nonresettable, so that the change in readings over \na time span is an accurate measure of switching activity.\n\nTwo control panel design features reduce the probability of service \nloss because of accidental manual operations. Particularly sensitive \ncontrols, such as the switch disconnect keys, are mechanically in- \nhibited from accidental operation. In addition, manual operations that \ncould interrupt service require simultaneous operation of the involved \nkey and a master key, or the sequential operation of two keys, depend- \ning upon the operation.\n\nMCCO message cutoff control operated \nMCSD message cutoff switch disconnect \nMCSO message cutoff switch operated \nMSTR master\n\nSTOA standby terminate operate automatic \nSTOM standby terminate operate manual \nST OVRD standby terminate override \nTERM terminate\n\nThe signaling system control and indication appearances are in the \nnext-to-last row. The first three positions are associated with the \nsignaling transmitter. A normal operation of the transmitter results \nin a momentary flash of the two diagnostic lamps, TRMTR sT and \nREG LOAD. An alarm lamp is associated with each oscillator, one for \neach of the two signaling channels. An alarm lamp is also provided for\n\nthe clock signal that times the pulsed high-frequency output of the \ntransmitter.\n\nThe fourth position contains the signaling system test key and the \nTEST FAIL lamp. Depression of the key causes the signaling test opera- \ntion to start; if the operation has not completed in 1 second, the \nTEST FAIL lamp is illuminated.\n\nPositions 5, 6, and 7 in the signaling row are associated with the \nsignaling receiver. The alarm lamp com FAIL indicates that the receiver \nwas unsuccessful in decoding the last command. COMPL FAIL and PTY \nFAIL indicate a command lacking those properties. Loss of timing \nability in the receiver is indicated by CLK FAIL. RCVR ST and DR ENAB \nare the diagnostics for the receiver that flash momentarily with each \nnormal operation. The last two positions indicate failures of either \nsignaling channel and provide indication and control of which channel \nis being decoded by the receiver. The channel not being decoded is used \nfor the complement check.\n\nThe upss-3 switching system provides protection against service \nloss because of line failures and protection against overload propagation \nfrom any source. Line maintenance is aided through manual switching \ncontrols, including service transfer and message cutoff capabilities. \nOne switch control bay provides switching capability for 10 regular \nlines and one standby line, both transmitting and receiving.\n\nThe trss-3 bay is modular in design and utilizes a pom signaling \nsystem to maintain interstation switching coordination. Several design \nfeatures, both electrical and mechanical, minimize the risk of in- \nadvertent service loss from untimely or accidental control activation.\n\nCopyright \u00a9 1974 American Telephone and Telegraph Company \nTue Beut System TEecunicat JOURNAL \nVol. 53, No. 10, December 1974 \nPrinted in U.S.A.\n\nA centralized and automated transmission surveillance system has been \ndeveloped for the LS Coaxial-Carrier Transmission System as a means \nfor attaining desired transmission reliability. Additional benefits are \nextensive and include accurate data-processing capabilities plus substantial \ncost and time economies. The system consists of two basic measuring and \ncontrol facilities: (2) a transmission surveillance center, located at a desig- \nnated main station, originates all control operations and accumulates and \nprocesses all measured data through use of a small computer, and (12) \ntransmission surveillance auxiliaries, located at all other main stations, \nperform measurement functions as directed and return the resulting raw \ndata to the controlling transmission surveillance center. Digitally operated \ntest equipment makes desired measurements under local or remote pro- \ngrammed control, and the E2 Status Reporting and Control System provides \ninterstation transmission of commands and data through time-sharing of \na four-wire data transmission link.\n\nVerification of the transmission integrity of a complex, broadband \nnetwork such as the L5 Coaxial-Carrier Transmission System requires \nmany measurements at different locations and considerable processing \nof the measured data. To minimize the manpower and time require- \nments for these maintenance operations while providing a high degree \nof system reliability, a centralized, automated transmission surveil- \nlance system has been developed and forms an integral part of the L5 \nnetwork.\n\nThe transmission surveillance system (Tss) takes advantage of a \ntechnology evolving from the availability of inexpensive, flexible, small \ncentral processors (minicomputers) and programmable test sets. A \nsurveillance network, consisting of a computer-controlled center\n\nlocated at a strategic L5 main station and a set of remotely controllable \nauxiliaries located at other stations and having digitally operated test \nsets with automatic access to key test points, provides the following \nadvantages on a large segment of an Ld route:\n\n(c) Transmission performance overview that is not possible with \nlocal individual station maintenance. \n(22) Early warning of slowly developing troubles. \n(iit) Rapid localization of trouble by automatic techniques. \n(tv) Efficient use of manpower through computer control of routine \ntasks. \n(v) Accurate, flexible records as a result of computer processing \nand teletypewriter printout.\n\nA four-wire data-transmission system transmits all remote control \ncommands and retrieves remotely measured data in digital form. This \ndata transfer capability is realized by utilizing a new feature included \nin the E2 Status Reporting and Control System.\n\nThe L5 system also has a built-in fault-location capability for re- \nmotely identifying a defective repeater in a coaxial line. It consists of \nan oscillator unit associated with each repeater and a logic unit located \nin the manhole. Commands from an adjacent main station activate an \noscillator unit, and signals injected at the repeater input and output \nports are monitored at the receiving end of the line to verify proper \nrepeater operation.\n\nThe transmission surveillance system (Tss) for the L5 carrier system \nprovides computer-controlled transmission-measuring capability utiliz- \ning technology evolving in the field of digitally controlled transmission- \nmeasuring equipment. A typical Tss, shown in simplified block schema- \ntic form in Fig. 1, serves a segment of several hundred route miles in \nan L5 system. The main functional units are as follows:\n\n(z) One transmission surveillance center (Tsc), the focal point of \neach surveillance system, originates all automatic operations.\n\n(it) Several transmission surveillance auxiliaries (Tsa\u2019s), preferably \nnot over 10 to 12 per system, are controlled remotely by the \nTsc and are located at all main stations other than the Tsc site.\n\n(zi) A modular, coaxial, switched access network is associated with \neach Tsc and TSA.\n\n(iv) A fault-location facility provides test oscillators at each re- \npeater, remote-control circuits, and power for operating the \noscillators.\n\n(v) A data-transfer facility is used for transmission of remote- \ncontrol commands and measured data, and is provided by the \nE2 Status Reporting and Control System on a time-shared basis \nwith other services such as alarm surveillance.\n\nThe Tsc is generally located at a well-manned terminal main station \nhaving significance in the overall maintenance and operation of the L5 \nsystem area covered by the tss. The Tsc functions as a nerve center for \noriginating and processing automatic measurements on L5 line and \njumbogroup multiplex (smx) equipment. Upon diagnosing a trouble \ncondition, TSC personnel may request maintenance action at other \ndistant main stations, which may normally be unmanned or partially\n\ncnicaco FA > ne aries fo \" \nl i A | . \n? CLEVELAND | \n| | | LILLYVILLE \nILLINOIS \u2014 | INDIANA OHIO WAYNESBURG \n| a \n| 7 HILLIARD | PITTSBURGH \n: INDIANAPOLIS | \n| \n|\n\nmanned. The tsc for the initial L5 carrier installation (Fig. 2) is at \nthe Xenia, Ohio, station, which is on the main backbone route and has \nthree sidelegs and mx equipment. In this installation, the Tsc monitors \n13 other stations and 815 miles of repeatered line. Placing the tTsc \nat a station with the largest number of sidelegs and mx signal-process- \ning equipment maximizes the \u201clocal-category\u201d\u2019 control and measure- \nments. This hastens surveillance operations by minimizing the amount \nof data-link time sharing required for remote control through the E2 \nStatus Reporting and Control System.\n\n(t) Power-feed main. \n(iz) Switching power-feed main. \n(vt) Terminal/terminal main. \n2.4 Transmission measuring system\n\nRecently developed 90-type digitally controlled test equipment? \nassociated with the Tsc and each TsaA measures transmission at the\n\nvarious stations in a transmission surveillance system. The equip- \nment consists of a signal generator providing an adjustable sine- \nwave output in the amplitude range of \u201499.9 to 0 dBm and fre- \nquency range of 10 kHz to 100 MHz. The selective detector measures \nthe amplitudes of received signals in the range of \u2014119 to 0 dBm and \n10 kHz to 100 MHz. Most transmission measurements are made auto- \nmatically under computer control in the surveillance system. A digital \ncontrol unit (Dcu) associated with the 90-type equipment operates \nthe signal generator and selective detector in this mode. Four 16-bit \nbinary words set the output signal amplitude and frequency of a signal \ngenerator. Five 16-bit words set the frequency, sensitivity, bandwidth \n(250 or 2500 Hz), and noise-distortion mode of a selective detector. \nAn analog-to-digital converter in the pcu transforms the analog \nmeasurements of a selective detector to digital form suitable for input \nto the computer or to a data-transmission facility such as the E2 \nStatus Reporting and Control System.\n\nDuring automatic surveillance operation, an E2 Status Reporting \nand Control System associated with the Ld system transmits remote- \ncontrol commands from a Tsc to the various Tsa\u2019s and raw measured \ndata from the tTsa\u2019s to the computer in the tsc. The E2 system, as \narranged for L5 use, provides interstation communication under the \nfollowing categories or modes:\n\n(t) Alarm polling. \n(17) Status reporting. \n(177) Remote switching. \n(iv) Data transfer or remote callup.\n\nAlarm polling, the principal E2 function, continues automatically until \ninterrupted by a request for one of the other functions. Data transfer \nrequired for Tss operation is interleaved on a time-shared basis with \nthe other E2 operations. If not inhibited periodically, the remote \ncallup processing of most Tss programs would interrupt the alarm- \npolling cycle longer than the permissible alarm-updating period. There- \nfore, the E2 system suspends callup operation every 30 seconds and \npolls all remote stations for alarms. Callup resumes after a complete \nalarm-polling cycle, which takes two to four seconds, depending on \nwhere the sequence starts. \nThe E2 system consists of the following units:\n\n(z) A central station, which initiates, supervises, and controls most \nof the E2 system operations.\n\n(tit) Remote call-up units (RcU\u2019s) (one provided as an integral part \nof each remote station having general-purpose data-transfer \ncapability).\n\n(iv) A four-wire data link for interconnecting the remote stations \nand central stations. ; \nThe manual-type E2 central station provided for the initial L5 installa- \ntion is located at Williamstown, Ky. (Fig. 2). It accommodates one \ndata link, which may interconnect a maximum of 16 remote stations. \nAt least one remote station is provided at each of the 14 Ld stations \nfor alarm reporting and Tss control purposes.\n\nDuring alarm polling, the central station automatically interrogates \neach remote station in succession and registers the station location and \ncategory of an alarm should one occur. When an E2-central operator \ndesires detailed status information about a station, he manually ini- \ntiates the status reporting operations. The frequency and duration of \nthe alarm-polling interruptions for this purpose depend on the number \nof statuses assigned to each alarm indication and the amount of infor- \nmation needed. Remote switching is used in Tss operations to control \nthe switching of remote line sections for out-of-service measurements.\u00ae \nEach remote switch command interrupts the polling cycle for approxi- \nmately 0.3 second.\n\nDuring any of the first three E2 system operating modes, the central \nstation and the addressed remote station communicate only with each \nother. The rcv provides the remote-to-remote communications or data \ntransfer capability needed by a Tsc for sending commands to a distant \nTSA and retrieving remotely measured data. The E2 central station \nsupervises the callup operation, which involves communication among \nremote stations and between the remote and central stations. The \nRCU accepts and delivers commands and data in bit-parallel, word-serial \nform, with each word containing 16 information-bearing bits. As used \nwith the L5 carrier, the E2 system transmits bits serially through the \ndata link at a rate of 600 bits per second.\n\nFigure 3 shows the functional arrangement of the principal parts in \na Tsc. A desk-type console (Fig. 4) contains all the items except two. \nThe standard teletypewriter console is located at the left. The switched \naccess equipment is mounted in the surveillance distribution bay, \nwhich is at the left rear in this particular installation.\n\nVo CONTROL OSCILLATOR = = \n7\u201d FACILITIES\u00bb DIGITAL \nCONTROL FUNCTION \nUNIT AND TO LINE \n= ROUTE LEVEL OF \nWRITER, AUXILIARY BETECrOe LEVELS OF -\u00a9Q SWITCHED \nDIGITAL SWITCHED ACCESS \nCONTROL ACCESS NETWORK \nUNIT NETWORK\n\nTAPE COMPUTER \nUNIT TO FAULT- \nLOCATION \nOSCILLATOR \nCONTROL \nPUNCHED COMPUTER- TO E2 \nTAPE TELEMETRY TELEMETRY \nREADER INTERFACE SYSTEM\n\nThe small general-purpose digital computer (minicomputer) in the \nTSC serves as the central processor in a Tss. The extensive computational \nand control capabilities in it provide much flexibility for scheduling \nand sequencing measurements, processing data, and presenting results. \nAmong the salient features of the computer are core memory for \n8192 16-bit binary words, a comprehensive program-interrupt system, \ninterface circuits for peripheral equipment, and an internal power\n\nThe tTsc operator uses the teletypewriter for entering commands \nand program parameters. The teletypewriter also prints measured \nresults or processed information derived from measured results as \ndirected by the programs.\n\nThe punched-tape reader provides direct computer loading of \nprogram material from punched tape. Depending on the type and \noperational phase of a program, it may be used when initializing the \nsystem, when adding information on magnetic tape, when running \nspecial programs, or during diagnostic procedures.\n\nSmall magnetic tape cassettes, having a maximum storage of 180,000 \ncomputer words each, are used in the digital magnetic-tape unit for \nconvenient storage of programs and data. An operator loads tape con- \ntents into the computer by typing a command at the teletypewriter.\n\nThe digitally programmable transmission-measuring system makes \nmeasurements at the Tsc location when directed by the computer. The \nswitched access network described in Section V automatically connects \nthe oscillator and/or selective detector to the desired measuring points. \nBinary-coded instructions pass through the auxiliary digital control \nunit (AUX pcU) and are held in memory in the pcv to control test set \nparameters such as frequency, output power, sensitivity, and band- \nwidth. An analog-to-digital converter in the pcu encodes the measure- \nments into digital form suitable for input to the computer.\n\nLogic in the aux pcu transfers and steers control commands from \nthe computer or manual control circuit to test equipment, fault \nlocation, or switched access control circuits. Flip-flop memory holds \nthe fault location or switched access commands until the command \nselection is changed or released.\n\nFront-panel pushbutton keys and associated programmed logic in \nthe manual control provide manual backup of local switched-access \nand fault-location oscillator control. Numeric readouts display the \nselections made with these keys. Other keys permit generation of any \n16-bit binary word for testing or limited operation of the transmission \nmeasuring system. Presently, manual control is possible only on the \nlocal-office operated equipment. Subsequently, remote manual control \ncapability will be provided for operating fault-location control logic \nat adjacent stations and one beyond the adjacent stations.\n\nThe computer-telemetry interface circuit and a software program \ndriver adapt one 16-bit computer input/output (I/O) channel to the \ncontrol format of the E2 Status Reporting and Control System. The \ntime multiplexing in the interface circuit and software programming \nenable the computer channel to receive 16-bit information-bearing \nwords from the E2 system and, in addition, to monitor six E2 system \ncontrol leads. The interface circuit transfers outgoing 16-bit words \nfrom the computer without modification, but does stretch the outgoing \nstrobe pulse (device command) to the duration required by the E2 \nsystem.\n\nThe tsa has all the functional units of a Tsc except the computer \nand associated I/O peripherals (Fig. 5). A bay framework mounts all \nequipment except the transmission-measuring system, which is as- \nsembled in a rolling console and plugged into the bay. The console \nmay be disconnected temporarily from the Tsa and used for general- \npurpose measurements anywhere in the office during visits by main- \ntenance personnel. The tsa bay also contains portions of the switched \naccess network.\n\nDIGITAL FUNCTION \nCONTROL AND \nUNIT ROUTE TO LINE \nLEVELS OF ENTRED \nAUXILIARY SELECTIVE Lanter NeCESG \nDIGITAL DETECTOR A NETWORK\n\nThe tsa operates the local switched access network, fault-location \noscillators, and transmission-measuring equipment similarly to the \ntsc. However, all automatic commands originate at the computer in \nthe Tsc and are delivered to the tsa through the E2 data link. When \ndirected by an operational program, the Tsa sends raw measured data \nin digital form through the E2 data link back to the computer in the \ntsc. The local switched access network and locally powered fault- \nlocation oscillators may be operated manually by means of the Tsa \nmanual control, which is the same as that in the Tsc.\n\nSignal paths between the measuring equipment and desired Ld \nsystem test points are established under computer or local-manual \ncontrol through a three-level, dual array of multiport, coaxial, ferreed- \ntype switches (Fig. 6). The transmitting array conveys outgoing \nsignals from the oscillator in a TSA or TSC, and the receiving array con-\n\nveys the signals from certain designated measuring points to the \nselective detector. Interconnection with the L5 line-protection switch- \ning system safeguards working channels from strong test tones.? A \nswitched-access-network transmitting path is not completed to an L5 \nline unless the line is out of service.\n\nThe circuit and equipment groupings of three-level switching \n(function, route, and line) offer much flexibility and accessing capacity \nand permit modular growth. The basic arrangement of Fig. 6 yields a \nmaximum of 528 transmitting line-level ports and 1584 receiving line- \nlevel ports, not including those used for network calibration. Only 110 \ntransmitting ports and 330 receiving ports are used for switched access \nto L5 line facilities in a maximally equipped Tsc location (10 routes). \nMost of the remaining capacity will be used for accessing jumbogroup \nmultiplex equipment if the station has much of it.\n\nA port on the function switch determines the class of an accessed \nmeasuring point, such as a particular place in the L5 lines or the mmx \nequipment. For example, port 1 of the receiving function switch \nalways gets signals from a test point designated receiving line test \n(RCVG LINE TST), regardless of the selected route or line. A port on the \nroute switch determines the cable entrance, and a port on the line \nswitch determines the coaxial line.\n\nThe route- and function-level switches, with an associated power \nenabler and decoder circuit for controlling all switches, are mounted in \nthe tsa bay at Tsa locations and at a Tsc in the nearby distribution \nbay. A line access bay associated with each L5 transmission bay \n(transmit-receive) lineup contains the line-level switches, which make \nconnections to the L5 line test points. A transmit-receive bay lineup \nis associated with each cable entrance, which is designated as a route \nfor Tss administration and control purposes. A station without sidelegs \nand located along a through backbone system has two of these routes \ndefined above, which result from the two cable entrances. Each sideleg \ncontributes one route to a station, since it has one cable entrance.\n\nThe switched access network is controlled either automatically or \nmanually by registering specific binary-coded commands in the aux \npeu memory, which outputs this information as continuous de signals.\n\nThe power enabler and decoder circuit decodes and converts these de \nsignals to decimal arrangements with current capacity and voltage \nnecessary to energize the switch windings. Switch-selection control \nprogresses through the levels as follows:\n\n(z) A signal from the aux pcv enables the function switch. \n(iz) Operating a function-switch crosspoint enables the route switch \nconnected to that crosspoint. \n(iz) Operating a route-switch crosspoint enables the line switch \nconnected to that crosspoint.\n\nInterlock logic prevents simultaneous closure of more than one cross- \npoint at the same switching level to assure adequate crosstalk isolation \nin the access network.\n\nSeveral circuit features and operating procedures combine to mini- \nmize the contribution of the switched access network to measuring \nerrors. For receiving measurements, the losses are effectively calibrated \nout in the paths from the test access points in the transmission bays \nto the selective detector input. The implementation is as follows:\n\n(z) All B cables are made equal length between the 12 X 1 line- \nlevel switch and the associated receiving measuring points in \nthe transmission bays (Fig. 6).\n\n(t7) The loss-frequency slope from the level-control output through \nthe C cable to the 12 X 1 line-level switch input is made equal \nto that of the B cables.\n\n(iit) As part of the measuring program, a reference value is obtained \nat each frequency by sending a signal from the oscillator, \nthrough the level-control unit to the 12 X 1 line-level switch, \nand then down through the route and function switches to the \nselective detector.\n\n(zv) The computer programming takes the reference into account \nin determining the absolute amplitude of the received signal.\n\n(v) The attenuation in the signal-splitting arrangement at the \nlevel-control output brings the calibration-signal amplitudes \nnear those of frequently measured signals to minimize or \neliminate attenuator ranging in the selective detector.\n\nFor transmitted test signals, the circuit paths from the level-control \nunit to the transmitting point, where signals are applied to the line, are \nequalized to a nominal 50-dB flat loss. The accumulation of small errors\n\nin the switched access network, together with the inherent accuracy of \nthe transmission-measuring equipment, should yield an overall mea- \nsuring accuracy of better than 0.2 dB for received signals. Straightaway \nmeasurements should have slightly less accuracy because of switching \nand measuring equipment involvement at two locations.\n\nLocation of failed or degraded repeaters along an L5 coaxial line is \nbased on measurement of tones transmitted through the line from a \nfour-oscillator fault-location unit associated with each repeater (Fig. \n7a). During fault location, de power and control commands are sent \nthrough interstitial wires in the cable from the nearest Tsc or Tsa to \nthe manhole locations to energize one oscillator group at a time. Two \nsimultaneously closed coaxial switches in an energized oscillator unit \nconnect the four different test frequencies to the input and output of \nthe repeater, as shown in Fig. 7a. The separations between the two \nlow frequencies and between the two high frequencies permit measure- \nment of each individual tone with a selective detector in a distant Tsc \nor TSA.\n\nBefore the oscillators are installed in a repeater manhole, the out- \nputs are adjusted so that signals leave the location as follows:\n\n(z) The two low-frequency signals have nominally equal amplitudes \nif the low-frequency gain of the repeater is normal.\n\n(iz) The two high-frequency signals have nominally equal ampli- \ntudes, but not the same as the two low frequencies, if the high- \nfrequency gain of the repeater is normal.\n\nTherefore, the signals received from a basic repeater indicate a gain \nabnormality if the low-frequency pair or the high-frequency pair \ndiffers in amplitude. Signals from a regulating repeater require differ- \nent interpretation because the gain varies with seasonal temperature \nchanges. Equal amplitudes should be expected in the received signal \npairs only when the cable temperature is at the seasonal mean.\u00ae\n\nA rectifier power supply and control unit associated with each route, \nor cable entrance, at a Tsc or TSA station directly control one fault- \nlocation unit at a time through interstitial wires in the coaxial cable \n(Fig. 8). In straightaway runs, direct control extends halfway to the \nadjacent station, which in turn controls the far half of the power\n\nam \n| APPARATUS \n| apes CASE \n/ \nPn ee me: Seen eee sack dy \n! es Tee aes cs fC \n(a) J | nteneteet\u2014T tt an \n! | eae Se ts Lae | \n| \ni | \nI \n! \n! | \n| | \nt \n| \nbo t\u2014\u2014 \u2014 \u2014 \u2014 \u2014 \u2014 \u2014_\u2014\u2014_\u2014_ = 4728 \neae ees 7 APPARATUS \neres eee aR Soe ey eee CASE \nCONSTANT] 4,0 | al \nCURRENT \nSOURCE \nE \nD \nc#, | | LOAD AND LINE c \n7 SELECTION LOGIC a al a0 \nrae ptf PT A | tocations \nCONSTANT | \nVOLTAGES \nH | H\n\nFig. 7\u2014Fault-location oscillator. (a) Manhole signal arrangement. (b) Manhole \ncontrol arrangement.\n\n16-BIT \nBINARY \nSIGNAL \nFROM \nCOMPUTER \nOR \nMAN. .CONT \nAT TSC, \nFROM E2 OR \nMAN. CONT \nAT TSA\n\nPART OF POWER \nAUXILIARY DIGITAL SUPPLY \nCONTROL UNIT \n\u2014 \npigeilea +1 | CONSTANT \nSTROBE 9 OTHER \nSTEERING FAULT LOCATION CURRENT \nPOWER CONTROLS \nBITS +OR\u2014 \nSUBSECTION A : \nMEMORY at . \nzz cL LINE \nlee SELECTION \nB \nMANHOLE ra i FAULT a A To \nMEMORY F LOCATION MANHOLES \nPOWER \nCONTROL \nE \nH \n: = g |: MANHOLE \nSELECTION \nB | ; \nA \nssi \n5 \u2014\u2014_\u2014\u2014\u2014\u2014\u2014\u2014 \n; | POWER \n> [RETURN \n\u00e9 (1 OF 6) \nSs6 \nI \n15 \nINTERSTITIAL \nWIRES\n\nsection. The 48-location capability (maximum) of the fault-location \npower arrangement permits one-end control in a significant fraction \nof the sidelegs and end links. Voltage drop in the interstitial wires and \nthe permissible sending-end voltage limit the powering and control \ndistance.\n\n(t) Upon receiving an appropriate two-word sequence from the \ncomputer or manual control at a Tsc location, or from the E2 \nsystem or manual control at a tsa location, the aux pcu \nregisters the selection (route, line, subsection, and manhole).\n\n(72) The aux pcu sends continuous de voltages through office \nwiring to operate a fault-location-oscillator power-control \ncircuit.\n\n(tit) The operated power-control circuit makes connections to an \nassociated power supply, applies constant current to one in- \nterstitial wire (+J), applies binary-coded combinations of \nplus and minus voltages to eight interstitial wires (A to H), \nand establishes a connection to one of six subsection wires \n(SS1 to S86) that provides a return path for the oscillator-load \nand control-signal currents.\n\nWith one possible exception, each subsection wire serves a group of \neight consecutive locations. The last or highest numbered wire serves \nthe most distant group, which may have fewer than eight locations. \nMost rss control words use the four most significant bits (NV, R, S, and \nT\u2019) of the 16 for strobe steering or addressing. Fault-location-oscillator \nturn-on also uses a fifth bit (/) for this purpose. The first fault-location \ncontrol word conveys the route selection via bits A to D; and the second \nword conveys the selections of line via bits A to H, manhole via F to H, \nand subsection via J to L. A binary-to-decimal decoder provides a de \nsignal on one of 10 outputs to enable the power-control circuit asso- \nciated with the desired route.\n\nTransistor logic in the enabled power-control circuit operates minia- \nture relays, which apply the appropriate control voltages to outgoing \nwires A to H and complete connections from the power supply to the \n+J constant current lead and the desired subsection lead, SS1 to SS6.\n\nThe constant-voltage control signals applied to leads A to H are \nreferenced to the subsection leads at the sending end. Positive voltage \nsignifies a logical 1 and negative voltage a logical 0.\n\nAt each controlled location, a combinational logic circuit (381A gate) \nbridged across the interstitial control wires decodes the binary-coded \ncombinations of positive and negative voltages on these leads and \nperforms desired switching (Fig. 7b). The circuit consists of a tree of \nmany simple transistor gates and transistor switches. A mechanical \nrotary switch assigns one of eight binary codes appearing on wires F, \nG, and H to each manhole in the subsection. The load- and line-selec- \ntion logic part of the 31A gate responds to voltages on the A to E leads \nand selects a desired oscillator load. Coaxial windings in series with the \noscillator loads enable directing the outputs to a desired line. Of the \n32 binary combinations available with the five A to E bits, 22 only are \nused for oscillator turn-on and switching, corresponding to the maxi- \nmum number of coaxials in a cable. Inhibit logic in the power-control \ncircuit at the station prevents sending out an unused code, which could \noperate the 31A gate ambiguously.\n\nThe low current drains in the 31A gate decoding circuits permit \nindividual control of fault-location oscillators at many different loca- \ntions over a few wires of relatively small diameter. The turned-off \nlocations draw almost negligible current. Since the coded control in- \nformation is applied continuously while a load is turned on, the inter- \nstitial wires need not be loaded or delay equalized as for pulse signals. \nUse of memoryless logic contributes to reliability by minimizing the \npossibility of a sustained lockup at a remote point during a malfunction \nof the control system.\n\nCommercial 117-volt, 60-cycle power energizes the entire Tss. Should \nthis source fail, an \u201c\u2018essential\u2019\u2019 supply such as a local engine- or turbine- \ndriven alternator carries the load after a few seconds of interruption. \nSince the Tss is off-line equipment not in the service transmission paths, \nthe five-second or longer start-up time permitted in an \u201cessential\u201d \ncategory supply is tolerable.\n\nSolid-state rectifier supplies provide the nominal 5- and 25-volt de \nneeded to operate the office logic and control circuits in the tss. Fuses \nin the distribution paths to the various circuits provide overcurrent \nprotection and convenient power-disconnect capability for main- \ntenance. A blown fuse operates an alarm relay, which activates the\n\nlocal office alarms and furnishes an indication to the E2 Status Re- \nporting and Control System for transmission to the E2 central.\n\nThe tTsc equipment is assembled in a desk-type console having a \nwriting area and a turret-type structure with three vertical panel \nregions just beyond the writing area (Fig. 4). Frequently used units \nhave front-panel access. The left turret has the computer, the center \nturret the manual control panel, and the right turret the transmission- \nmeasuring system. The punched-tape reader is accessible in the left \nfront, behind a hinged door under the writing area. The teletypewriter \nconsole is conveniently located to the left of the Tsc operator position. \nInfrequently accessed units such as logic circuits and power supplies \nare mounted in the rear of the console behind hinged doors. Slides on \nthe logic unit shelves facilitate access for maintenance.\n\nA surveillance distribution bay located near the Tsc console contains \nthe function- and route-level coaxial switches for the switched access \nnetwork, the associated control circuitry, and terminal blocks for \nconnecting numerous station control wires to the Tsc and line access \nbays.\n\nThe tsa bay contains essentially all the equipment of both the Tsc \nconsole and the surveillance distribution bay except the computer, the \npunched-tape reader, and the transmission-measuring system. Person- \nnel normally stand while operating Tsa equipment locally. The same \ntype of transmission-measuring system as that in a TSc is provided in \na rolling console.\n\nOne line-access bay is associated with each route or cable entrance. \nIt is located in the associated lineup of L5 transmit-receive bays and \ncontains principally fault-location power and control equipment, the . \nline-level switches and level-control unit of the switched access net- \nwork, and the 31A gate for controlling fault-location oscillators in the \nstation.\n\nExcept for the 31A gate, which uses discrete solid-state components, \nthe various logic control circuits throughout the Tss contain mostly \ncommercial integrated circuits assembled on plug-in printed wiring \nboards. Most mounting frameworks and panels are fabricated sheet \naluminum.\n\nSoftware has been developed to exploit the computational and \ncontrol facilities of the Tss computer (Section 3.1). Features are pro- \nvided to minimize operator interaction and to process and present\n\nresults in an easy-to-use format. Discussion of the software has been \ndivided into the following categories.\n\n(7) Application programs. \n(177) Data-base generation and administration. \n(tv) Diagnostic system.\n\nComputer programming is done in two languages, Assembly and \nFORTRAN. Assembly language is used primarily to write software \ndrivers (routines to control peripheral devices), where the pseudo- \nmachine-language code is needed to set up and complete input/output \noperations. The higher-level FoRTRAN language is used for applications \nprograms (complete routines for each surveillance activity), where \nadvantage is taken of its speed and flexibility in program development.\n\nA teletypewriter is used to enter system commands and program \nparameters. The teletypewriter is vendor-modified to allow bit-parallel \ntransfer of information and independent control of the input, print, \nand punch functions.\n\nThe fault-location subsystem, switched access network, and test \nset are controlled directly by the computer at Tsc locations. The com- \nputer remotely controls identical equipment at Tsa locations via the \nK2 telemetry facilities.\n\nTo relieve the Tsc operator of much punched-tape handling, a \ncassette-type digital magnetic tape unit is also interfaced with the \ncomputer; and programs for surveillance system operations are stored \non magnetic tape. The magnetic tape unit can read or write on a data \ntrack or an address track and access tape files quickly. Tape positioning \naddresses are prerecorded on the address track of each cassette in \nproportion to the number of spindle revolutions. To access a file, the \ntape unit counts spindle revolutions as the file approaches at high speed \nand then reads the address track only during the last few revolutions \nto verify that the desired tape position has been reached.\n\nA 64-word program, used to load other programs into the computer \nvia the high-speed punched-tape reader, resides in a protected area of \ncomputer memory. This program is used, for example, to load the Tss \nbootstrap program, which, in turn, loads the operating system from \nmagnetic tape.\n\nBasic Binary Loader \nTss Bootstrap \nDirective Processor \nInput/Output Control \nDrivers for: \nTeletypewriter \nPaper Tape Reader \nMagnetic Tape Unit \nTest Set \nFault Location Oscillator \nSwitched Access Network \nE2 Telemetry \nFormatters for: \nTeletypewriter \nMagnetic Tape Unit \nArithmetic Routines\n\nGenerators for: \nPilot Measurement \nFault Location \nLine Measurement \nJmx Measurement \ntss Diagnostic\n\nComputer Diagnostic for: \nComputer Instructions \nComputer Memory \nHardware Functions\n\ncatable loader, and magnetic-tape storage programs, used to prepare \nand manufacture those programs that are a part of the Tsc.\n\nThe Tss operating system processes a small set of operator directives \n(Table II); and, to economize magnetic-tape space and reading time, \nit contains a set of subroutines expected to be used by all applications\n\nprograms. \nTable I! Operator directives \nCommand Function \nLOAD (NO.) To recall and start an applications program stored on \nmagnetic tape. \nBEGIN To restart the program currently in computer core. \nRUN To continue a program after a programmed pause for \nmanual operations, or \nTo retry certain input/output operations resulting in \nerror messages. \nREWIND To rewind the magnetic tape cassette to clear leader.\n\nApplications programs are stored on magnetic tape and can be re- \ncalled by file number into the computer for execution upon an operator- \ninitiated Loap (file number) command. At the termination of a pro- \ngram or after a programmed pause, the operator may request another \nprogram to be loaded and automatically started. He may restart the \nprogram currently in the computer core with the BEGIN command; or, \nin the case of a programmed pause for manual operations, he may cause \nthe current program to continue with the RUN command. Also, using \nthe RUN command, he may request retries on certain input/output \noperations that result in error messages. All commands for these opera- \ntions are given by the operator at the teletypewriter.\n\nVendor-supplied, interrupt-compatible software drivers for the \nteletypewriter and magnetic tape and an input/output control pro- \ngram for policing requests to these devices are required to perform the \noperator directives. Software drivers have been developed by Bell \nLaboratories to control the E2 telemetry system, the fault-location \nsubsystem, the switched access network, and the programmable test \nsets.\n\nVendor-supplied library routines such as a formatter for teletype- \nwriter operations and arithmetic subroutines are included in the \noperating system because of frequent use by applications programs. \nAlso included are Bell Laboratories developed rortTRAN-callable rou- \ntines for teletypewriter plotting of measurement results and for con- \ntrolling magnetic-tape operations. The last-mentioned routine permits \nreference measurement files and data-base files to be stored on magnetic \ntape and/or recalled under program control.\n\nAll the above routines are stored on magnetic tape as file number 2, \nwhich is automatically recalled by loading a short paper tape referred \nto as the Tss bootstrap. The teletypewriter prints the message *WHAT? \nwhen the computer is ready to accept an operator directive. At that \ntime, the operator can request, for example, that a specific application \nprogram be loaded from magnetic tape and executed. File number 1 is \nreserved for a tape address directory, which contains the tape position \nof up to 67 other files for application programs, reference-measurement \nfiles, and data-base files.\n\nUse of a programmable computer permits quick and easy imple- \nmentation of various maintenance activities. In general, an applica- \ntions program is a complete sequence of computer instructions designed \nto monitor and evaluate a particular function of the L5 transmission\n\nThe surveillance library serves as a memory jog for the operator. \nIt prints a list of program names and numbers and optionally will print \ndetailed instructions regarding a specific program.\n\nAfter starting the automated fault-location program, the operator \nenters the set of cable routes (one or more power-feed sections), man- \nholes, and the lines containing the repeaters to be interrogated. The \ncomputer recalls a data base from magnetic tape and checks for the \npresence of requested repeaters.\n\nFor each section of coaxial line requested, the computer initiates the \nfollowing: (2) calibration of office trunks and coaxial switches in the \nline-access network between the RCVG LINE TsT point (Figs. 6 and 9) \nand the receiver, (77) access of the test point, (777) turn-on of the oscil- \nlators for one repeater at a time, and (v) measurement of the received \nset of tones. Repeater gain deviations, as derived from tone measure- \nments, are compared with allowed limits and are printed only for out- \nof-limit repeaters unless the operator specifies, when starting the \nprogram, that all measured and derived data be printed. As each section \nof line is completed, the program releases the switched access network \nand the fault-location oscillators.\n\nFigure 10 is a sample printout depicting a repeater in trouble. The \nteletypewriter bell and special characters in the printout provide \naudible and visual indications of anomalies and out-of-limit conditions. \nThe operator may optionally obtain a teletypewriter plot of the \nreceived tone amplitudes as a function of distance along the cable.\n\nIn a composite message signal, pilot tones may be monitored by the \npilot measurement program. The operator may select from the set of \nline, jumbogroup, mastergroup, and supergroup pilots. Pilots can \nautomatically be measured at both ends of all smx sections or optionally \non a set of lines between two stations. The computer recalls a data \nbase from magnetic tape and checks for allowable combinations of \ntransmitting and receiving stations.\n\nFor each section requested, calibration of office trunks and switches \nis performed at both stations. At the transmitting end of a section, \npilots are accessed and measured via the TRMTG LINE TsT point (Fig. 9)\n\nand, at the receiving end, via the RcvG LINE TsT point. The difference \nin pilot amplitudes between the two stations is a measure of the gain \ncharacteristics of the line. Out-of-limit conditions or, optionally, all \npilot amplitudes and differences are printed as a function of frequency. \nThis printout can serve as an early indication of the need to re-equalize \na line.\n\nThe line measurement program is used to determine the gain- \nfrequency characteristic of a coaxial line with better accuracy and \nfiner granularity than with pilot measurements. A sequence of tones \nis injected at the EQL sIG IN point in the station at the transmitting \nend of the desired line and measured at the RcvG@ LINE TST point in \nanother station. Using the line-protection switching system, the opera- \ntor must first take the line out of service and operate the message cut- \noff switch.\n\nAfter taking the line out of service, the operator selects the trans- \nmitting and receiving stations, the line, and the set of frequencies, and\n\ndecides whether or not a teletypewriter plot of the gain characteristic \nis required. Careful study of the printout can be used to evaluate the \nneed for or results of equalizer adjustment. Measurements can be \nstored on magnetic tape for future comparison with historical data.\n\nUsing the smx measurement program, the operator can access the \nremote test switch for each jumbogroup and calibrate office trunks and \ncoaxial switches. The program then measures carrier or jumbogroup \nand mastergroup pilots at 11 test points per jumbogroup. The data \nare compared with limits and with historical data. The historical data \nare updated on magnetic tape, and a printout of out-of-limit conditions \nis made.\n\nThe general-purpose loss measurement program operates the trans- \nmission-measuring system in a TSC or a distant TSA under computer \ncontrol to observe directly connected equipment or facilities in the \nselected station. Loss vs. frequency characteristics are printed and, \noptionally, plotted.\n\nThe operator can select the station address of the test set and all \nprogrammable functions of the test set, including the transmit ampli- \ntude and set of frequencies. Measurement averaging and reuse of \ncalibration measurements are available.\n\nSeveral application programs use data bases that are prepared by \nthe data-base-generator program and stored on magnetic tape. The \ndata bases determine the arrangement and extent of surveillance \nequipment to be controlled by a center Jocation, as well as testing \nlimits, etc. Separate data-base files are created for the automated fault \nlocation and Jmx measurement programs, whereas the pilot and line \nmeasurement programs share a data-base file.\n\nEach data-base-generator program is stored on magnetic tape and \nrecalled by the operator using the same procedure as for other applica- \ntions programs. The program simply asks questions that the operator \nmust answer. For many questions, the operator may elect to use the \npreprogrammed answers. Questions are answered to define and cor- \nrelate manhole designations, route numbers, coaxial lines equipped, \ntelemetry addresses, the allowable range of measurement levels, test \npoints available, etc. When all questions are answered, a separate data- \nbase file containing this information is automatically created on\n\nWhen new or additional equipment must be accounted for in the \ndata-base file, area engineers provide the craft operator with completed \nforms for updating those files.\n\nA separate set of programs, stored on a second magnetic tape, is \ndedicated to checking transmission-surveillance-related equipment for \nproper operation. These diagnostics are called into the computer using \nthe Tss operating system and, if no errors are detected, control is \nreturned to the operating system. Errors are indicated by a computer \nhalt or message on the teletypewriter.\n\nSeparate programs are used to test the teletypewriter, high-speed \npunched-tape reader, digital control unit, and aux pcu. The magnetic \ntape unit diagnostic is loaded from punched tape.\n\nThe computer diagnostic control program supervises the running \nof a large set of vendor-supplied computer diagnostics. There are \ndiagnostics for testing all instruction types, core memory, and other \ncomputer hardware. Each computer diagnostic has an overlay as- \nsociated with it to allow nonstop testing and multiple execution of \neach diagnostic. The control program interacts with the operator to \ndetermine the diagnostics to be run.\n\nThe Tss diagnostic program provides an indication of the health of \nthe entire surveillance system by exercising all the hardware at the \ntsc and the Tsa\u2019s. To narrow down possible trouble causes, each new \nphase of testing involves a minimum of previously untested equipment. \nWhen a problem is detected, more detailed testing is performed in the \narea affected. To determine what facilities are equipped in each office, \nthe program shares a data base with the automated fault location \nprogram.\n\nThe degree to which the diagnostic system localizes troubles is \nestablished principally by the number and significance of the places \nwhere the statuses of transmission and logic signals can be determined \nautomatically and remotely. The Tss automated diagnostic procedures \ngenerally localize a fault to a subsystem small enough for knowledge- \nable craft personnel to find the actual fault quickly.\n\nSince the transmission surveillance system provides the potential \nfor maintaining the L5 system on a network basis rather than as a\n\nseries of independent stations, many somewhat complex and inter- \nrelated factors must be considered in the evolution of an overall main- \ntenance plan to effectively realize this advantage. To begin with, a Tss \ncontrolled by a single tsc should be assigned to a portion of L5 route \non a basis compatible with the overall administration of the system. \nThe association of surveillance operations and alarm reporting, as \nprovided by the E2 system, is another important consideration. A \nthird consideration is that of defining the boundary between adjacent \nsurveillance systems along a route. Finally, in developing an overall \nmaintenance plan, intervals for performing routine tasks must be \ndetermined by carefully considering such factors as system size, \nmeasurement speed, need for update, and output requirements. Of \ncourse, considerable blocks of free time must be left for performing \ndemand-type measurements to locate a trouble or to characterize \nperformance, as a result of an external stimulus such as a line switch, \nE2 alarm, or a line-equalization adjustment.\n\nThe rules for the layout of a transmission surveillance system are \ndetermined as much by administrative and management aspects as by \nany hardware constraints. Presently, the E2 data facility has a capacity \nof 16 remote stations, which limits a Tss to 15 Tsa stations and one con- \ntrolling Tsc station. The resulting system bound of this number of \nstations, likely near 800 to 900 route miles, probably approaches the \nlimit that can be monitored effectively by a single rsc. Other con- \nsiderations include: (\u00a2) AT&T-Long Lines area boundaries, (77) \nplanned E2 alarm reporting and other maintenance arrangements, \nand (iz) location of strategic stations along the route.\n\nOnce a TSS arrangement has been established within a region, it is \nimperative that a single E2 data facility interconnect all the stations \nserved by the Tss to provide data-transfer capability to and from the \ntsc. Generally, this data facility will be time-shared with normal E2 \nalarm polling operations.\n\nJudicious tsc placement enhances the effectiveness of a Tss. Several \nfactors should be considered in selecting the station. It should (2) be \non the associated L5 route, (77) be a major, manned facility, (277) have \nsignificance in the area maintenance plan, and (zv) be a junction point \nwith signal processing (multiplex). Examination of the Tss area and \nconsideration of these points should lead to an appropriate Tsc place- \nment. Another highly desirable goal is to colocate the Tsc and E2 \ncentral, if at all possible, since this would centralize all major mainte- \nnance operations. It is important that the Tsc location be manned 24\n\nhours a day. Although it is possible to control the tsc computer over a \ndata set/rry link from another office, perhaps off the route, it is not \nconsidered desirable unless craft personnel thoroughly familiar with \nLd operations are available to interpret performance data.\n\nThe criteria for obtaining a viable tss layout have been given \n(Section 10.1), and, if availability of computer software is assumed, the \nrole of the surveillance system in the overall maintenance scheme for \nthe L5 system can be defined. As previously discussed, the Tss performs \ntransmission measurements for the purposes of maintaining a high level \nof performance and locating troubles. These operations fall into two \ncategories:\n\nThe demand measurements are unscheduled and the need is gen- \nerally spontaneous, resulting from a trouble condition. The routine \noperations, where the words transmission surveillance apply, are sched- \nuled to realize continued monitoring of performance and to spot any \ndeteriorating conditions. Procedures and associated software for an \ninitial L5 maintenance plan have been based on experience gained \nduring the L5 field trial, in the early stages of the initial system turn-up, \nand in discussions with AT&T personnel. This plan, when implemented, \nwill assist in maintaining a high quality of service on the L5 system. \nIn simple terms, the plan consists of:\n\n(7) Pilot measurements : Transmitted and received line and master- \ngroup pilots will be measured daily between all signal-processing \npoints (terminal stations) to characterize the transmission per- \nformance of the coaxial lines.\n\n(iz) Fault location runs: All line repeaters will be checked weekly, \nand out-of-limit conditions and regulating repeater gain \nchanges will be printed out.\n\n(ziz) Jumbogroup multiplex measurements: Test points in all mmx \nequipment will be monitored on a weekly basis and out-of- \nlimit conditions will be printed out.\n\n(iv) Line gain/frequency measurements: Detailed measurements, \non an out-of-service basis, of the frequency characteristic of a \ncoaxial line will be made following any repair or realignment \noperation to verify proper performance.\n\n_(v) Tss diagnostics: Proper operation of the surveillance system \nitself will be routinely checked on a weekly basis. This includes \nan overall test at each station to pinpoint any problems in the \naccess network, control circuits, or test equipment as well as \ndiagnostic tests of the computer and associated peripherals at \nthe TSsc.\n\nAccording to calculations for the 815-route-mile initial system be- \ntween Lillyville, Pa., and Hillsboro, Mo., these routines should use \nabout 50 percent of the total hours in a seven-day week when all 22 \ntubes are in service. The calculations have assumed that only a few \npilot measurements are needed beyond those in the daily overall check \nbetween terminal stations to associate a trouble with a particular \npower-feed section. Methods are being investigated to reduce the \nmeasurement times before systems reach full capacity so as to free the \ntsc for troubleshooting operations when necessary. Furthermore, the \npresently provided facilities are probably just the beginning of auto- \nmated transmission-measuring capability. As experience is gained with \ncentralized and automated maintenance and more sophisticated test \nequipment becomes available, the role of the Tss will grow\u2014perhaps \ninto the areas of acceptance testing, trouble shooting of individual \nequipment units, and interfacing with other evolving maintenance \nsystems.\n\nRealization of the transmission surveillance system for the L5 \ncoaxial system has required the coordinated efforts of many persons. \nThe authors hereby acknowledge the contributions of the following \nimmediate associates: P. M. Berard, M. A. Leveille, and J. P. Russo \nfor circuit design and performance verification ; R. A. Noel for diagnos- \ntic software design; M. A. Plante for application planning; and W. P. \nFrawley and C. J. Rimas for physical design.\n\n1. F. C. Kelcourse and F. J. Herr, \u201cL5 System: Overall Description and System \nDesign,\u201d\u2019 B.S.T.J., this issue, pp. 1901-19383.\n\n. N. H. Christiansen, \u201cNew Instruments Simplify Carrier System Measurements,\u201d\u2019 \nBell Laboratories Record, 48, No. 8 (September 1970), pp. 232-238.\n\n. J. H. Green and R. W. Sanders, \u201cL5 System: Line-Protection Switching,\u201d B.S.T.J., \nthis issue, pp. 2011-2034.\n\n. E. H. Angell, Y.-S. Cho, K. P. Kretsch, and M. M. Luniewicz, \u2018\u201cL5 System: \nRepeatered Line,\u2019 B. ST. J., this issue, pp. 1935-1985.\n\nCopyright \u00a9 1974 American Telephone and Telegraph Company \nTuE BELL System TECHNICAL JOURNAL \nVol. 53, No. 10, December 1974 \nPrinted in U.S.A.\n\nThe jumbogroup multiplex (sux) translates basic jumbogroup signals \nto and from the L& line spectrum. Each of three jumbogroups provides \n8600 4-kHz voice circuits. In addition to frequency translation, the sux \nprovides regulation and equalization on a sumnbogneup basis and auto- \nmatic one-for-one protection.\n\nPerformance objectives are discussed and interpreted Gah the perspec- \ntive of their anfluence on the design philosophy and approach used and \nthe resultant physical realization. All subsystems are presented in detail. \nParticular mention is made of the double-balanced diode-ring modulator \ndeveloped for sux application.\n\nWhile gross performance objectives for analog multiplex equipment \nare available, the detailed statement of design objectives is part of the \ndevelopment task. The basic guiding principle is that the resultant \nsystem, the aggregate of line and terminal equipment, must provide \ntransmission paths of quality and reliability appropriate to the Bell \nSystem communication network, and it must do so at a cost sufficiently \nlower than other alternatives to warrant development by Bell Labora- \ntories, manufacture by Western Electric, and purchase by Long Lines \nand the operating telephone companies.\n\nThe basic function of the jumbogroup multiplex (4x) terminal is to \ntranslate basic jumbogroup signals to and from the Ld line spectrum. \nEach of the three jumbogroups is composed of six mastergroups. \nThese signals must not be corrupted or distorted by the smx in a \nmanner or degree that would compromise the quality of transmission. \nThe equipment must be reliable and maintainable. Derivable from \nthese considerations are a host of performance objectives, including \nthose discussed in the following sections.\n\nThe overall objective on noise, including thermal noise and inter- \nmodulation, for a 4000-mile, transcontinental, L5 system is 40 dBrnc0.? \nOf the total, 39.4 dBrnc0 is allocated to the L5 line, and the remaining \n31.2 dBrncO is allocated to the aggregate of all terminal equipment. \nWhile the L5 line is, in a sense, a totally new design entity, such is not \nthe case in the associated terminal equipment. The jmx has to share \nthe terminal noise allocation with an existing hierarchy of equipment, \nincluding channel, group, supergroup, and mastergroup banks. \nConsider a representative transcontinental connection including two \nchannel banks, five supergroup multiplex (LMx) terminals, seven \nmastergroup multiplex (mmx) terminals, and eight mmx terminals. \nAssuming 18 dBrncO performance for the tmx and mmx terminals \nand 10 dBrncO for the channel banks, a maximal allocation of 18 \ndBrnc0 results for the ymx. Allowing 1 dB for misalignment and other \nvagaries, including the possibility of other connections more demanding \nthan the one cited, the ymx design objective was set at 17 dBrncO, with \nthe realization that an even lower number would be desirable. In \nessence, it appeared that at least initially the design philosophy should \nbe to obtain the lowest noise possible. |\n\nTo put the 17-dBrncO objective into perspective, it is helpful to \nview the smx challenge relative to the most comparable multiplex \navailable, the mmx-2.2 In an optimally designed multiplex terminal, \nthe controlling noise sources are the modulators and the associated \namplifiers following the modulators. Normally, the intermodulation \ndistortion of the modulators is greater than that of the other trans- \nmission apparatus, so that lowest signal levels are found at modulator \noutputs. Since levels are low at such points, the noise figure of the \nfollowing amplifier is of greatest importance. Levels are chosen to \nminimize the combined noise from these critical nodes. Since the smx \nuses multiple steps of modulation and demodulation, while mmx-2 \nuses single steps, the smx has twice as many critical nodes in each \ntransmission path as the mmx. Furthermore, the bandwidth of the \njumbogroup signal is six times that of the mastergroup signal. For \nthird-order intermodulation distortion alone, the number of products \nthe ymx must contend with is 16 dB greater than in the mmx. In a \nsense, JMX must be many decibels better than mMmx-2 to yield com- \nparable noise performance.\n\nBased on studies of typical system configurations, with the multi- \nplicity of occasions for crosstalk, an objective of 85 dB equal-level\n\ncoupling loss (ELcL) has evolved for analog multiplex equipment. This \nobjective was assumed for the Jmx.\n\nThe sources of crosstalk in multiplex equipment are many and varied, \nand include many modes not encountered elsewhere. Crosstalk may \noccur between different jumbogroup signals, between different portions \nof the same jumbogroup, between signals at different stages in the \nmodulation-demodulation process, and in an intelligible or noise-like \nfashion. Coupling may exist between signals, between carriers, between \npilots and carriers, etc. While an exhaustive discussion of crosstalk \nis inappropriate here, one typical illustrative example will be \nmentioned.\n\nGiven a carrier with a tone 4 kHz removed from the carrier at a \nlevel 79 dB lower than the carrier level, intelligible adjacent channel \ncrosstalk appears in almost every channel at a level approximately \n85 dB below the level of the interfered channel. Because of this mode \nof crosstalk, spectral purity of mmx carriers is important. Since Jmx \ncarriers are as high in frequency as 91.648 MHz, filtering, shielding, \nand grounding are critical. Common ground impedances of very small \nmagnitude are sufficient to cause unacceptable crosstalk performance.\n\nTones falling in a voice channel are particularly annoying. Often, \ntones generated from several sources add coherently (voltage addition). \nBecause of the multiplicity of sources of tones, the design objective \nfor mx was that no tone falling in the jumbogroup passband should \nhave a level exceeding \u201470 dBm0. This becomes particularly challeng- \ning when, in the sox, the source of the tone may be a 15-dBm, 91.648- \nMHz carrier, and transmission levels are as low as \u201443 dB. The \nresulting implications on isolation of separate paths, carrier balance, \nand filtering of carrier leak signals are significant.\n\nThe transcontinental objective on frequency offset is less than 2 Hz. \nThis implies an accuracy requirement of one part in 10\u00b0 on the effective \ncarrier frequencies used in the gmx.\n\nThe overall misalignment objective for a transcontinental L5 \nconnection is less than +4 dB. This requires that the mx passband \nbe flat to +0.2 dB. In addition, since the mx may carry digital \nsignals, delay distortion must be limited.\n\nThe L5 line,\u2019 with its automatic line-protection switching system,\u2018 \nis highly reliable. If terminal equipment is not to add noticeably to the \nmean outage time for a transcontinental connection, based on the \nsystem configuration considered earlier, the objective for each smx \npath (transmit or receive) becomes less than 0.14 minute per year. \nThis corresponds to an equivalent failure rate of about 200 Fit\u2019s \n(mean time to failure is 500 years), assuming that the mean time for \nrepair is one hour.\n\nWhile it is difficult to become quantitative relative to maintain- \nability, several observations of a qualitative nature may be made. \nEach piece of transmission equipment can handle as many channels \nas a whole L4 system. The equipment may be housed in unmanned \nmain stations. Many modes of failure are of a subtle nature. It there- \nfore seems desirable to use automatic, remote diagnostics extensively. \nRoutine measurements at the bay should be held to a minimum. Most \ntroubleshooting on site should require no removal of modules prior to \ntrouble isolation. Status indicators should be provided for the con- \ntrolled and trouble states of the equipment. In short, the equipment \nshould be easy to maintain and difficult to operate incorrectly.\n\nThe cost of the smx on a per-channel basis may be reasonably viewed \nfrom two perspectives: (7) its contribution to the total terminal cost \nfor the terminal equipment required to take a voice signal from voice \nfrequency to basic jumbogroup frequency and back, including signal- \ning, and (22) its contribution to the cost of an L5 system for a system \nlength of a few hundred miles or more. It has been estimated that the \nper-channel cost of the mx amounts to less than 2 percent of either \nthe terminal or line cost; therefore, the mx cost would have very \nlittle influence on the total cost of providing a voice channel.\n\nAn overall block diagram of the transmitting arrangement, typical \nof any jumbogroup, is shown in Fig. 1. Basic jumbogroup signals, \ncomposed of six mastergroup signals, and a 5.888-MHz jumbogroup \npilot are fed to the smx from the basic jumbogroup trunk bay (Bser).\u00b0\n\nThese signals, contained between 0.564 and 17.548 MHz, first enter \nthe ymx at the transmitting jack field. Two input ports are provided. \nThe first or regular input is that through which the signal normally \npasses. The second or spare input is provided for emergency patching \npurposes and is normally terminated. These two inputs are combined \nin an input hybrid transformer and then connected to a test pad that \nprovides a low-loss (0.6 dB) through path and a high-loss (30 dB) \nbridging path. The high-loss or low-level output is fed to a splitting \ntransformer, and provides a local and a remote test point, both of \nwhich are isolated from the transmission path by about 33 dB of loss. \nLocal test points are accessed via jacks mounted at the front of the \nbay. Remote test points are provided to allow access for automated, \ncentralized test equipment. The low-loss pad output is fed to a splitting \nhybrid that provides signals to the redundant A and B paths of the \ntransmitting side.\n\nThe basic jumbogroup signal then passes through the restoration \naccess switches which, when operated, are capable of providing an \nappearance of the basic jumbogroup signal and access to the input of \nthe associated modulator at the restoration patch bay. These switches \nare connected in a manner that allows monitoring of the interbay \ncabling when the switches are in the unoperated condition.\n\nThe basic jumbogroup signal then goes to the modulators, where it is \nfiltered and translated to any one of three jumbogroup line assign- \nments. Multiple steps of modulation are used, and the required \ncarriers are provided by the associated carrier supply. Carrier supplies \nare redundantly provided for each jumbogroup. Each carrier supply \nprovides the carrier signals for an associated modulator and de- \nmodulator. Carriers are generated by appropriately mixing internally \ngenerated signals with signals provided by the sync-distribution \ncircuit. There are two sync-distribution circuits per bay, each of which \nuses redundantly provided jumbogroup frequency supply (sFs)\u2019 signals \nto generate a number of highly stable reference signals for trans- \nmission to all of the A or B carrier supplies.\n\nThe jumbogroup signal, at line frequency, is then fed through the \noutput restoration switch to the cable equalizer. The output restora- \ntion switch, connected in a manner similar to that of the input \nrestoration switch, provides access to the modulator output. The \ncable equalizer, located in the transmitting line-interface unit, compen- \nsates for the loss of the cable connecting the mmx to the L5 line bay.\u00ae \nLevels are adjusted in the transmitting line interface to provide a \nstepped preemphasis for transmission over the repeatered L5 line.\n\nThe cable equalizer output passes through a test pad which provides \na high-loss access for local and remote testing and a low-loss path to \nthe transmitting protection switch. The transmitting switch accepts \nthe A and B output signals as inputs, and provides a working and idle \noutput. The state of the switch, set by the switch-control circuitry, \ndetermines which input will be connected to the working and idle \noutputs. Both outputs are fed to unequal-ratio hybrid transformers \nwhich provide a high-loss (7 dB) output for monitoring purposes, \nand a low-loss (1 dB) output that is connected to the transmitting \njack field. The working output signal is then cabled to the Ld line bay, \nand the idle output, available for emergency patching through spare \ncabling, is normally terminated.\n\nOn the receiving side (Fig. 2), the complete L5 spectrum is re- \ndundantly cabled to the receiving jack field of each jumbogroup \nequipment. The signal first passes through a test pad that provides \nhigh-loss access for testing and a low-loss connection to the associated \ncable equalizer. The cable equalizer compensates for the loss of the \ncable between the L5 line bay and the smx. The signal is then fed to an \ninput restoration switch which is capable of providing an appearance \nof the L5 line signal and access to the demodulator input at the restora- \ntion patch bay.\n\nThe received signal then goes to the demodulator where the desired \njumbogroup is selected and translated to basic jumbogroup frequency. \nAfter passing through an output restoration switch, which can be \nused to gain access to the demodulator output for the restoration \npatch bay, the signal is fed through a test pad to the basic jumbogroup \nequalizer, a manually adjustable, multistage equalizer that will \nprovide* the capability of correcting for misalignment accrued over \nmany miles of the repeatered L5 line on a jumbogroup basis. The \nsignal then goes to the regulator which monitors the level of the \nreceived jumbogroup pilot and modifies its gain accordingly.\n\nThe regulator output is fed through another test pad to the receiving \nprotection switch. Working and idle outputs are connected to unequal- \nratio hybrid transformers that provide access points for automatic \nmonitoring in addition to output signals connected to the receiving\n\n* Space and power have been provided for the jumbogroup equalizer in the ymx \nbay. Determination of the appropriate characteristics of this equalizer awaits evalua- \ntion of misalignments for working L5 systems.\n\npatch field. Additional equal-ratio hybrid transformers are used to \nprovide dual appearances of the working and idle output signals. One \nof the working output signals is cabled to the Bsat. The remaining \nports, available for emergency patching, are normally terminated.\n\nThe modulator and demodulator subsystems for all three jumbo- \ngroups are shown in Fig. 3; transmission levels are as indicated. In \neach case, the subsystems used in the A and B sides are identical.\n\nThree types of amplifiers are used in the modulator and demodulator \nsubsystems. The 9-dB fixed-gain amplifiers and the 12 + 2-dB adjust- \nable amplifiers are transmission quality amplifiers with a controlled \ntransmission band running from 0.5 to 100 MHz. They are realized \nin the hybrid integrated circuit (HIc) technology. The 9-dB amplifier \nis designed for low-level application and has a 5-dB noise figure. The \n- 12-dB amplifier has been designed to handle higher signal levels \nwithout introducing appreciable intermodulation distortion. Both \namplifiers are multistage, major-loop feedback designs with 75-ohm \ninput and output impedances. The 15-dB carrier drive amplifier is \ndesigned to provide a 15-dBm carrier signal for the modulator. In \naddition to providing highly linear gain for carriers up to 91.648 \nMHz, this amplifier provides at least 40 dB of reverse isolation to \ncontrol crosstalk that might otherwise be established through the \ncarrier drive path. A multistage local-feedback design with printed- \ncircuit realization is used. All amplifiers use feed-through filters for \nbattery connections.\n\nThe 17A modulators are double-balanced diode-ring modulators \nusing Schottky barrier diodes and 75:300-ohm center-tapped trans- \nformers.\u2019 These are highly linear modulators providing at least 40- \ndB carrier and signal balance. Nominal port impedances are 75 ohms. \nA single design is used throughout the gmx. More will be said about \nthe 17A modulator in Section 3.16.\n\nThe filters* have been designed on a system basis; i.e., the loss \nrequirements have been specified in an interactive manner in an \nattempt to reach a global optimum without placing undue stress on \nany one design. Often the attenuation required for some undesired \nsideband can be obtained more easily through the combined effect \nof two or more filters than in a single filter. The systems of filters were \nspecified to provide combined attenuation to undesired modulation\n\nFig. 3\u2014Modulator and demodulator subsystems for jumbogroups 1, 2, and 3. (a) Jumbogroup 1 modulator. \n(b) Jumbogroup 2 modulator. (c) Jumbogroup 3 modulator.\n\nFig. 3 (cont.)\u2014(d) Jumbogroup 1 demodulator. (e) Jumbogroup 2 demodulator. (f) Jumbogroup 3 demodulator.\n\nproducts and carrier leak sufficient to keep them at least 85 dB lower \nthan the level of the desired sideband. All filters have nominal input \nand output passband impedances of 75 ohms.\n\nIdeally, the basic jumbogroup signal fed to the smx should contain \nno energy above 17.548 MHz. If some disturbance is present above \nthe desired input spectrum, it could corrupt the message spectrum \neither by being superimposed as a result of linear transmission through \nthe modulator (signal leak) or by being translated into the message \nband via the modulation process. Accordingly, after the input signal is \nraised to the appropriate transmission level via the input amplifier, \nit is passed through a low-pass filter to reject any spurious high- \nfrequency disturbance. The basic jumbogroup spectrum is then \ntranslated in frequency using a 42.496-MHz carrier. The signal is \namplified and its level is adjusted to a more optimal value for a subse- \nquent step of modulation. Next, a bandpass filter selects the lower \nsideband between 29.948 and 41.932 MHz. The signal is modulated \nonce more, using a 45.056-M Hz carrier. After amplification, the lower \nsideband located at 3.124 to 20.108 MHz is selected by a second \nbandpass filter. The modulator subsystem output level is then achieved \nthrough the use of an adjustable 12-dB amplifier, a fixed-loss pad, \nand a test pad that introduces a 0.6-dB loss in the signal path and \nprovides a bridged test point with 30-dB isolation.\n\nThe net frequency shift provided by the jumbogroup 1 modulator \nis 2.56 MHz. Since the input spectrum overlaps the output spectrum \nfor jumbogroup 1, multiple steps of modulation are mandatory. Signal \nleak prohibits using one step of modulation (or demodulation) for \njumbogroup 1.\n\nJumbogroup 2 is processed in a manner similar to that for jumbo- \ngroup 1. The first step of modulation uses a 70.144-MHz carrier. The \nfirst bandpass filter selects the lower sideband between 52.596 and \n69.580 MHz. The second carrier frequency is 91.648 MHz, and the \njumbogroup 2 signal occupies the band of 22.068 to 39.052 MHz. The \nnet shift for jumbogroup 2 is 21.504 MHz.\n\nIt is often more desirable to use the lower sideband, since many of \nthe undesired byproducts of modulation corrupt the upper sideband. \nIn jumbogroup 2, using a single step of modulation and selecting the \nlower sideband would cause the most difficult filtering\u2014attenuation\n\nof carrier leak and rejection of the upper sideband\u2014to be done in the \nneighborhood of 40 MHz. Using two stages of modulation greatly \nfacilitates this task by shifting the chore to the region of 20 MHz.\n\nJumbogroup 3 uses three steps of modulation. The first two steps \nare identical to those of jumbogroup 1. The third step of modulation \nuses a 40.448-MHz carrier. The third bandpass filter selects the upper \nsideband between 43.572 and 60.556 MHz. The net frequency shift \nis 43.008 MHz.\n\nWith either one or two steps of modulation, the attenuation of \ncarrier leak and the rejection of the nearby sideband would require \nfilters of questionable realizability for jumbogroup 3. Thus, three steps \nof modulation were necessary. The first two steps may be viewed as \nproviding a baseband signal with significant separation between the \nlowest signal frequency and de. The result is that adjacent sidebands \nare separated by about 6 MHz after the third step of modulation. \nThis, coupled with the use of the upper sideband (in this case, no \nspurious modulation products overlapped the upper sideband), greatly \nfacilitates filter synthesis.\n\nFollowing a slight level adjustment to place the signal at a more \noptimal level for the first step of demodulation, the jumbogroup 1 \nsignal is selected from the L5 line signal by the first bandpass filter \nwhose passband extends from 3.124 to 20.108 MHz. Using a 45.056- \nMHz carrier, the first step of demodulation shifts the input spectrum \nup in frequency. Following amplification and level adjustment, the \nsecond bandpass filter selects the lower sideband signal between \n24.948 and 41.932 MHz. The filter output goes through a 9-dB ampli- \nfier-pad combination to provide isolation between the filter and the \nfollowing modulator. This signal passes through a second stage of \ndemodulation using a 42.496-MHz carrier. The signal is amplified, \nand the lower sideband signal at basic jumbogroup frequency is \nselected by the output low-pass filter. The basic jumbogroup signal is \nfurther amplified, then fed to a test pad that provides a local bridged \ntest point.\n\nThe jumbogroup 1 demodulator is similar in most aspects to the \njumbogroup 1 modulator. It uses the same carriers and uses filters \nwith identical passbands (although the detailed nature of the rejection \nbands is not always the same).\n\nModulators of the type used in the smx are bilateral. That is, a \nsignal applied to the output port will be translated in frequency and \nappear at the input port in the same manner that signals applied to \nthe input are translated and appear at the output. Consider a modu- \nlator with conversion loss C' terminated on its output by a load whose \nreturn loss, with respect to the average modulator output impedance, \nis R,, and terminated on its input by a source whose output return \nloss, relative to the average modulator input impedance, is R;. Signals \nappearing at the modulator output will have some energy reflected \nback through the modulator, and some of this energy will, in turn, be \nreflected from the source output through the modulator once more. \nThe result will be an echo-like signal superimposed on the original \nsignal at the modulator output. The reflected signal will be attenuated \nrelative to the original signal by 2C + Rk, + AR; dB.\n\nThe modulator input and output impedances are complex and \nperiodically time-varying, and are not capable of being carefully \ncontrolled and manipulated without significant decrease in conversion \nefficiency. In practice, the average input impedance is controlled to \nsome nominal value. For the smx, this nominal value is 75 ohms. \nSimilarly, amplifiers, pads, and in-band filter impedances are designed \nto be 75 ohms. Filters are designed to pass in-band energy and reject, \nprincipally through reflection, out-of-band energy. Consequently, \nfilters typically have rejection-band terminal impedances correspond- \ning to open or short circuits.\n\nAssume that a modulator is driven from a filter and operates into \nan amplifier. If the amplifier input impedance has an average return \nloss of R, = 18 dB, relative to the modulator output impedance, the \nfilter out-of-band return loss is R; = 0 dB, and the conversion loss \nis C = 6 dB, then the reflected signal will be suppressed 30 dB relative \nto the signal with which it adds, assuming that the signal that ap- \npeared at the filter input was an out-of-band signal. The interfering \nsignal will be at the same frequency as the interfered signal and will \nadd on a voltage basis, depending on the phase relationship between \nthe two signals. If the angle is constant across the band, a slight level \nshift will occur. If the angle is not constant, but changes slowly with \nfrequency, passband distortion, gradually changing with frequency, \nwill be introduced. This can be compensated for through the use of \nsimple deviation equalizers designed to attend to the aggregate \ndistortion introduced by filters and other apparatus. If, however, the \nangle of the reflected signal changes rapidly, as it would if the angle \nof the filter output impedance changed in the region of resonance of a\n\ncrystal, then sharp, unequalizable notches could be introduced into \nthe jumbogroup passband. The reflected energy could cause variations \nin passband loss as large as +0.27 dB for the relative levels indicated \nabove. This was, in fact, experienced in initial measurements of \njumbogroups 1, 2, and 38 receiving equipments, and required the \nintroduction of 9-dB pad-amplifier combinations to improve the return \nloss of driving impedances for selected modulators, as indicated in \nFig. 3.\n\nThe jumbogroup 2 demodulator is functionally identical to that of \njumbogroup 1. It uses the same carriers and has filters with the same \npassbands as the jumbogroup 2 modulator.\n\nThe jumbogroup 3 signal is demodulated in a single step; thus, all \nthree jumbogroups traverse four modulators during the modulation/ \ndemodulation process. After level adjustment, the jumbogroup 3 \nsignal is selected from the Ld line spectrum by a bandpass filter with a \npassband between 43.572 and 60.556 MHz. The signal then goes \nthrough a 9-dB amplifier and pad to provide isolation between the \nfilter and the following modulator. The jumbogroup 38 signal is then \ndemodulated with a 43.008-MHz carrier and, following amplification \nand level adjustment, is sent through a low-pass filter to isolate the \nbasic jumbogroup signal. Following additional amplification, the signal \npasses through a test pad that provides local test access.\n\nA major and critical aspect of the design of an Fpm terminal involves \nthe ordering of apparatus and the settings of internal levels in the \nmodulator and demodulator subsystems. The output of a double- \nbalanced diode-ring modulator contains the dominant upper and lower \nsidebands, carrier leak, and a multitude of other signals.? The energy \nin either sideband is of a magnitude comparable to that of the energy \nin the carrier leak. If the modulator output is fed through a bandpass \nfilter into an amplifier, then the undesired sideband and the carrier \nleak can be attenuated so that the total signal power carried by the \nfollowing amplifier is greatly reduced. This tends to reduce the inter- \nmodulation distortion generated by the amplifier. It also tends to \ndegrade the effective noise figure of the amplifier by an amount \nequivalent to the filter passband loss. Furthermore, should the signal\n\nnext be fed to a subsequent stage of modulation, the next modulator is \npresented with a broad band of noise covering not only the message \nband, but the image band as well. If the order of the amplifier and \nfilter is reversed, then the thermal noise problem is mitigated at the \nexpense of increased intermodulation distortion in the amplifier. The \nmore beneficial arrangement depends upon a number of parameters, \nincluding filter loss, amplifier noise and intermodulation distortion, \nmodulator intermodulation distortion, and carrier leak. While pre- \nliminary analysis of the gross characteristics of the various trans- \nmission elements is helpful in establishing a starting point, the optimal \nsubsystem configuration can only be obtained through extensive, if \nnot exhaustive, examination of the various possible configurations. \nThis has been done for all modulator-demodulator subsystems using \nnoise-loading techniques.\u2019\u00b0\n\nIt has been found that the setting of internal levels is equally \nimportant. Signal and carrier levels also were optimized using noise \nloading.\n\nJumbogroup signals are translated to their L5 line-frequency \nallocation through multiple steps of modulation. The net frequency \ntranslation they experience may be viewed as having been achieved \nthrough a single modulation step using an equivalent carrier frequency \nas follows. Consider the first step of modulation to have used a carrier \nof frequency fi. Following the first step of modulation, a signal of \nfrequency f, would be translated to fi \u2014 f,. The second step of modu- \nlation, using carrier frequency fe, would translate the signal to \nfe -\u2014fitf.. The net translation is f. = fe \u2014 fi. In jumbogroups 1 \nand 2, where two steps of modulation are used, the above applies. \nFor jumbogroup 3, a third modulation step, using a third carrier \nfrequency f3, is employed. In this case, the equivalent carrier is \nfe =f3s+fe \u2014 fi. To demodulate the jumbogroup signals accurately, \nonly the equivalent carrier frequencies need be generated at the \nreceiver. This principle is used extensively in the mx. The equivalent \ncarrier frequencies are 2.560, 21.504, and 43.008 MHz for jumbogroups \n1, 2, and 3, respectively. The frequency stability of these frequencies \nis determined solely by the stability of the jumbogroup frequency \nsupply (sFs).\u2019\n\nThe srs generates three reference signals at frequencies of 20.480, \n2.560, and 1.024 MHz. The second and third reference signals are \nobtained from the first through division of a 20.480-MHz signal by 8\n\nand 20, respectively. These signals are further processed by the sync- \ndistribution circuits to yield the 2.560- and 21.504-MHz signals which \nare fed to the carrier supplies. The carrier supplies, using the ap- \npropriate input signals and locally generated signals, generate the \ncarriers for modulation and demodulation. The functional separation \nbetween the srs and sync-distribution circuit is somewhat arbitrary. \nThe counters that provide the division by 8 and 20 operations could \nhave been located in the sync-distribution circuit, or much if not all \nof the sync-distribution circuit could have been housed in the Jrs. \nSince it may be feeding up to 20 smx bays, failure of the srs could be \ncatastrophic; therefore, the yrs should be as uncomplicated as possible, \nwith every effort made to maximize its reliability. The only exception \nto this rule was the inclusion of the count-down circuitry in the Jrs. \nThese counters are implemented using high-speed emitter-coupled \nlogic. It was expected that locating such circuitry in the smx could \ncause the generation of high-level spurious tones for which appropriate \nshielding and filtering might prove unachievable.\n\nThe sync-distribution circuit is shown in Fig. 4; amplitudes are \nindicated in dBm. Three reference signals at 1.024, 2.560, and 20.480 \nMHz are cabled to the input from the srs. The 2.560-MHz signal is \namplified, filtered, and passed through splitting hybrids to provide \nthree \u20145-dBm output signals, which are then cabled to the appropriate \n(A or B) carrier supplies, where they are used as needed. The 20.480- \nand 1.024-MHz signals are amplified, filtered, and mixed in a 17A \nmodulator to provide an output at 21.504 MHz. This signal is filtered, \namplified, and passed through splitting hybrids to provide three \n\u20145-dBm output signals, which are cabled to the appropriate carrier \nsupply inputs. Level adjustments are provided as required. Local \ntest points are provided for the combined three-tone input signal, for \nboth output signals, and for both intermediate signals through 30-dB \nbridging test pads. All these test points appear at jacks mounted in the \nface plate of the sync-distribution circuit drawer.\n\nThe gmx bay is arranged to process three jumbogroup signals in any \narrangement; that is, it can process three jumbogroup | signals, one \neach of jumbogroups 1, 2, and 3, etc. There is no fixed assignment of \nthe equipment to limit the flexibility with which the smx is used. Since \nit is not known a priort which jumbogroup any given position will\n\nhandle, the sync-distribution outputs are fed to all locations. Those \nwhich are required are used. Those not required are automatically \nterminated when the carrier supply drawers are installed.\n\nThe jumbogroup 1 carrier supply terminates the 21.504-MHz input \nsignal, as shown in Fig. 5; amplitudes are indicated in dBm. This \ncarrier supply contains a temperature-compensated crystal oscillator \nthat generates a \u201415-dBm, 42.496-MHz signal whose frequency \nstability is typically good to one part in 10\u00b0. As with all de-powered \napparatus used in the mx, this oscillator uses feed-through filters on \nthe battery leads to prevent coupling between circuits through common \nde paths. The oscillator output is amplified and passed through a \nsplitting hybrid. One of the hybrid outputs passes through a fixed \npad, a level adjustment, a phase-adjust network, a low-current 14-dB \namplifier, a crystal bandpass filter, a 15-dB amplifier, a bridging test \npad, and a splitting hybrid whose outputs each experience further \nnoncrystal filtering to yield two 0-dBm output signals.\n\nThe phase-adjust network is provided to facilitate the matching \nof the A and B side signals from the modulators (demodulators) \nprior to manual operation of the protection switch. The match is made \nto minimize the mean-squared error between the output signals to \nachieve hitless switching. This will be covered in more detail later.\n\nAs indicated earlier, the spectral purity of the carrier signals and \nisolation between carrier ports are of special importance. It may \nappear desirable to reverse the order of the crystal filter and the \n15-dB amplifier which follows it. The amplifier introduces both thermal \nnoise and harmonic distortion. However, the signal level into the \ncrystal filter would then be on the order of 10 dBm. The reliability \nof a crystal degrades significantly when it is exposed to levels exceeding \n0 dBm. Accordingly, the arrangement shown was used. While the \ncrystal filter provides sufficient rejection in the immediate vicinity of \nthe carrier, achieving sufficient rejection elsewhere requires further \nfiltering. This is accomplished through the use of noncrystal tc \nfilters placed in both output legs. This output arrangement provides \na net isolation between carrier supply outputs equivalent to the trans- \nhybrid loss of the transformer plus twice the output filter loss.\n\nThe second output of the input splitting hybrid is mixed with the \n2.560-M Hz input in a 17A modulator to yield a product at 45.056 MHz. \nThis signal is similarly amplified, crystal filtered, further amplified, \nand passed through a 30-dB bridging test pad. The low-level output is \nfed to a 6-dB splitting pad that provides two test appearances, one \nlocal and one remote. The higher-level output is fed to a splitting\n\nhybrid, where its outputs are further filtered to yield two 0-dBm \ncarrier signals.\n\nThe carrier signals are cabled to the appropriate modulators and \ndemodulators as indicated. On the transmitting side, the first modu- \nlator uses the 42.496-MHz carrier, while the second uses the 45.056- \nMHz carrier. Each of these carriers is only as stable as the stability \nof the local oscillator. However, the effective carrier frequency, which \nis the difference between the actual carriers used, is orders of magni- \ntude better since it depends only on the stability of the srs. For \npurposes of accurate demodulation, the stability of the local oscillator \nis not important. Other considerations\u2014namely, the matching between \nfilter rejection peaks and the location of the tones to be rejected\u2014 \nrequire that the actual carrier frequencies do not wander too far from \ntheir nominal values. The carriers must be at least as stable as a few \nparts in 10\u00b0,\n\nFunctionally, the jumbogroup 2 carrier supply is very similar to \nthat for jumbogroup 1, as shown in Fig. 6. In this case, the 2.560-MHz \ninput is terminated and the 21.504-M Hz input is mixed with the locally \ngenerated 70.144-MHz signal. For jumbogroup 2, the phase-adjust \nnetwork is located in the path of the input reference signal prior to \nmixing with the locally generated signal, while for jumbogroup 1 the \nphase-adjust network was located in the through path of the locally \ngenerated signal. Functionally, either location is acceptable. The \nrelative positions shown were chosen because they facilitated network \nrealization.\n\nJumbogroup 3 requires three carriers for modulation and one \ncarrier for demodulation; the first two steps of modulation for jumbo- \ngroup 3 are identical to those for jumbogroup 1. As shown in Fig. 7, \nthe 42.496- and 45.056-M Hz carriers are generated in the same way as \nfor jumbogroup 1. Since only one appearance of each signal is required, \nno splitting hybrids are needed at the outputs. Following the second \nstep of modulation, the jumbogroup signal has a net translation of \n2.560 MHz. The third step of modulation and the one step of demodu- \nlation must both use carriers dependent solely on the srs. The desired \nnet translation for jumbogroup 3 is 43.008 MHz. This requires the \ngeneration of a 40.448-MHz carrier for the third step of modulation, \nand a 43.008-MHz carrier for demodulation. Both sync-distribution\n\nsignals are used. The 43.008-MHz carrier is obtained by doubling the \n21.504-MHz input. The 43.008-MHz signal is then mixed with the \n2.560-MHz input to form the 40.448-MHz carrier. Once again, all net \ntranslations depend only upon the stability of the mrs. Two phase- \nadjust networks are required for jumbogroup 3, one for modulation \nand one for demodulation.\n\nThe jumbogroup regulator, shown in Fig. 8, compensates for level \nvariations that might be introduced in the jumbogroup passband by the \nLd line or terminal equipment between the remote transmitting BJaT \nand the regulator input. A bridged-T variolosser, using indirectly \nheated thermistors in the shunt and bridging arms, provides a level \ncontrol of +5 dB dynamic range. Control currents are automatically \nadjusted in a complementary manner in accordance with the level of | \nthe received 5.888-MHz jumbogroup pilot. In the event that the \njumbogroup pilot is lost, the regulator automatically assumes a mid- \nrange setting. An operational amplifier, connected as an integrator, \nprovides high de gain in the control loop. This results in a residual \nerror of less than 0.01 dB over the full regulation range.\n\nTo facilitate maintenance, the jumbogroup regulator is equipped \nwith a manual reset feature. Operation of the reset switch locks the \nregulator in the midrange state. This feature is particularly valuable \nwhen noise loading tests are being made. An indicator lamp is lighted \nwhen the regulator is manually placed in the reset state.\n\nFor each jumbogroup, both transmitting and receiving, A and B \npaths are provided. These paths emanate from a splitting hybrid and \nterminate on a 2 X 2 protection switch. It is the principal function of \nthe protection-switching eircuitry to monitor the integrity of the A \nand B paths, to assure that a good signal is fed from the protection \nswitch when possible, and to activate alarms as appropriate. The \ncircuitry also provides for the control of restoration switches and the \nmanual operation of the protection switch in a hitless manner.\n\nAs shown in Fig. 9, both working and idle output jumbogroup pilot \nsignals are monitored by the protection-switching circuitry. The \nswitch control detector emits a de signal proportional to the level of \nthe jumbogroup pilot. On the transmitting side, the pilot frequencies \nare 8.448, 27.392, and 48.896 MHz for jumbogroups 1, 2, and 38, \nrespectively; receiving, all jumbogroup pilots are at 5.888 MHz. If \nthe idle signal is lost for more than 0.1 second, a minor alarm condition \nis indicated. If the working output is lost and the idle output is present, \nthe switch changes state. If this causes a good signal to be present at \nthe working output, and the signal is not present at the idle output, \na minor alarm is indicated. If both outputs are lost simultaneously for \nmore than 0.1 second, the switch does not change state, but a major \nalarm is indicated. Alarms are indicated both in real time and with \nmemory. Switching is nonrevertive; that is, once a service-providing \ncondition has been established, the switch will remain in that state; \nneither A nor B output is preferred. This arrangement avoids un- \nnecessary hits on jumbogroups that might be heavily loaded with \ndigital signals.\n\nIn the event that the working output appears lost while the idle \nsignal is present and operation of the protection switch fails to correct \nthe situation, no further switching occurs. Such a condition would most \nlikely be due to failure of the circuitry monitoring the working output. \nA major alarm would be indicated.\n\nManual control of the protection switch is provided. If no trouble \ncondition is indicated, manual operation of the appropriate control \nswitch will cause the protection switch to change state. To facilitate\n\nthis operation in a hitless manner, a variable-sensitivity monitor of the \ndegree of match is provided. The idle and working outputs are sub- \ntracted in a special hybrid transformer, yielding an error signal. This \nerror is amplified and detected, and the resultant de signal is fed to a \nmeter. The meter reading is indicative of the energy in the error signal. \nGiven that the gain of the A and B sides are closely matched, the \nmeter reading is indicative of the phase difference between the A and \nB sides. Adjustment of the phase network in an associated carrier \nsupply corresponds to varying the phase, constant with frequency, of \none side relative to the other. By altering the phase network adjust- \nment to null the meter reading, the best possible phase match is \nachieved. Since this procedure uses the actual message loading present \nat the time, it results in the best match over the most meaningful \nportion of the jumbogroup passband. Inability to reach a very low \nmeter reading is indicative of gain mismatch. This can be corrected \nby making gain adjustments as appropriate, through the use of the \nbridged test jacks provided on all modules.\n\nOnce A and B paths have been matched, the protection switch is \noperated manually. In this mode of operation, the switch contacts are \nsequenced in a make-before-break manner to avoid the momentary \nopen that might otherwise result.\n\nIf the manual operation would cause the protection switch to select \nan apparently failed side, the control logic will not normally respond \nto the manual command. In this instance, an additional, recessed \nbutton must be depressed by the craftsperson to allow him to override \nthe inhibit feature of the control logic.\n\nRestoration access switches are operated via the switch-control \nlogic. Upon command for restoration switch operation, the switch- \ncontrol logic locks the protection switch in its existing state. It then \noperates those restoration switches associated with the unused side. \nWhen the control logic is in this state, an indicator lamp is lighted \nat the bay.\n\nFrequency translation is achieved in the ymx through the use of a \ndouble-balanced diode-ring modulator, the 17A. The important \nperformance parameters of such a modulator include conversion loss, \npassband distortion, intermodulation distortion, carrier balance, and \nsignal balance.\n\nConversion loss, which gives rise to the difference in level between \nthe input signal and the desired output signal, has a theoretical mini-\n\nmum value of 3.9 dB as a result of the partitioning of input energy \ninto the multitude of sidebands generated by the modulation process. \nAdditional loss is introduced principally through diode dissipation, \ntransformer loss, and whatever additional padding is introduced \nowing to other considerations. While conversion loss can be offset by \nadding appropriate gain in the amplifiers preceding and following the \nmodulator, the tendency of increasing conversion loss ordinarily is to \nresult in increased noise, either because levels are increased internal \nto the modulator, or because the effective noise figure of the following \namplifier is degraded, or both.\n\nPassband distortion is introduced principally through transformer \nroll-off and circuit parasitics. Mutual coupling decreases with de- \ncreasing frequency, and shunt capacitance becomes dominant as \nfrequency is increased. Although transformer design can be optimized \nfor a specific frequency range, the design of a single transformer for \nthe multitude of input and output spectra encountered in ymx modu- \nlators is particularly challenging. While passband distortion is usually \nequalizable, prudent design requires that such distortion be minimized \nat its source.\n\nIntermodulation distortion is influenced by diode nonlinearities, \ndiode match, transformer and circuit balance, carrier waveform, and \nsignal level.? Intermodulation distortion may fall in-band or out-of- \nband, relative to the desired sideband. Unlike most other parameters \nof interest, in-band intermodulation distortion cannot be mitigated \nonce it is generated. Consequently, this type of distortion must be \ncontrolled at its source.\n\nIdeally, in a double-balanced modulator, neither input signal nor \ncarrier should appear at the output. Circuit imperfections preclude \nrealization of these objectives. Carrier balance and signal balance \ndepend on diode match, transformer balance, and circuit layout. \nTypically, the carrier input level is 70 to 80 dB higher than the level \nof the output signal. These relative levels are chosen because they \nresult in optimal intermodulation performance. Furthermore, Jmx \ncarriers run as high in frequency as 91.648 MHz. Consequently, \nsizable levels of carrier signals may appear at the output. Since they \nmay subsequently fall within a message spectrum, such carrier leak \nmust be attenuated. While carrier leak is theoretically filterable, \nrejection peaks well in excess of 100 dB are required. Furthermore, \nunfiltered carrier leak may significantly increase the total power that \nan amplifier directly following a modulator must handle. This can give\n\nrise to further nonlinear distortion in the amplifier. Finally, any noise \nentering the carrier port from the carrier-drive amplifier will appear \nat the output to the extent that carrier balance allows. Should a \nportion of this noise overlap the output message spectrum, it will \ncontaminate the signal in an unfilterable manner. Accordingly, good \ncarrier balance is critical, second only in importance to intermodulation \ndistortion.\n\nSignal leak, by design of the modulation process (i.e., the selection \nof modulation steps and carrier frequencies), is always filterable. \nNevertheless, signal leak at times gives rise to extremely demanding \ndistrimination objectives.\n\nThe 17A modulator is shown in Fig. 10. The center-tapped trans- \nformers have a 75:300-ohm impedance ratio with a 7:7 + 7 turns \nratio. These transformers introduce less than a 0.05-dB passband \nshape over any jumbogroup band. Transformer balance exceeds \n55 dB.\n\nShottky barrier diodes are used. They provide essentially instan- \ntaneous switching at all jmx carrier frequencies, and appear to be \npurely resistive up to 350 MHz. Their I-V characteristics are extremely \nlinear in the region running from 2 mA to 20 mA. In this region, their \nresistance is about 40 ohms. Diodes are selected to yield matched \nquads. Each quad has less than 15 mV mismatch at 2 mA and at 20 mA.\n\nThe 1-dB pads at both signal and carrier inputs were found experi- \nmentally to result in improved intermodulation distortion, carrier and \nsignal balance, passband distortion, and return loss.\n\nCircuit layout was optimized experimentally. Topological symmetry \nwas found to be critical. This was achieved, for example, by using four \nmanually inserted crossovers as opposed to partial use of printed- \ncircuit land connection.\n\nThe 17A modulator has a conversion loss of 6.2 dB, passband \ndistortion not exceeding 0.1 dB, less than 0 dBrnc0 of noise at optimal \nsignal levels, and, when driven by its associated carrier-drive amplifier, \nsignal and carrier balance of at least 40 dB.\n\nThe smx is housed in a shop-wired, shop-tested, 11-foot 6-inch, \nunitized double bay. It is arranged to contain all the equipment to \nprocess three jumbogroup signals. Each jumbogroup position is in- \ndependent of other positions and contains transmitting and receiving \njack fields, transmitting and receiving switch-control logic units, and \nredundant modulator and demodulator subsystems, carrier supplies, \nregulators, equalizers, transmitting and receiving line interface units, \nand detectors. Common equipment includes redundant sync-distri- \nbution circuits at the lower left of the bay, and a summary alarm \ncircuit at the upper right. Sixteen dc-to-de converters are located at \nthe top of the bay, just above the fuse panels. Alarm cutoff and lamp \nbuttons are at the lower right just above the writing shelf. Redundant \nbattery feeds and srs inputs are provided. Several photographs of the \nJMX equipment are contained in Ref. 11 of this issue.\n\nNo at-the-bay routine maintenance is planned for the mx. Rather, \neach jumbogroup is automatically and remotely monitored by the \ntransmission surveillance system\u201d associated with L5. Eleven critical \ntest points are provided for each jumbogroup position. These include \nthe transmitting input, the transmitting A and B outputs, the receiving \nA and B inputs and outputs, the A and B regulator inputs, and two \ncarrier supply test points. These eleven access points are concentrated \nin a remotely controllable 1 X 12 switch. Through this arrangement, \ndetailed analysis of smx performance is possible with little intervention \nby a craftsperson.\n\nAll the above-mentioned test points, in addition to others, are also \nprovided at the bay. It is possible to verify the integrity of any trans- \nmission-related module without removing it from the bay, through the \nuse of the test points provided.\n\nVerification of the integrity of the many status and alarm indicators \nis facilitated through the lamp-test feature. Depressing the three \nlamp-test buttons allows the location of any failed lamp in seconds.\n\nThe gmx has undergone extensive field evaluation. Both functionally \nand operationally, the equipment operated superbly. The poorest \njumbogroup noise performance obtained was 12 dBrnc0.\n\nThe poorest equal-level coupling loss obtained was 90 dB. This was \nmeasured between A and B sides of the same transmitting or receiving \njumbogroup, and was due to coupling through the subtracting hybrid \ntransformer of the circuitry used to evaluate the degree of match \nbetween the associated A and B sides. This is a well-controlled coupling, \nnot expected to vary from system to system. Most other forms of \ncoupling were so low as to be unmeasurable.\n\nExtensive testing of regulation and protection switching showed \nthese features to operate without fault.\n\nTen gmx bays were constructed by Western Electric for use in the \nLillyville, Pennsylvania-Hillsboro, Missouri route. This equipment \nhas been providing commercial service since January 1, 1974.\n\nTaking into consideration the bandwidth of the signal it processes \nand the frequencies at which it operates, the sx is the quietest analog \nmultiplex terminal ever designed. The success of the design is a tribute \nto the technical excellence and total dedication of its design team \nwhich included J. S. Young, T. B. Merrick, G. W. Kattke, B. B. Garg, \nS. J. Davis, A. G. Favale, and W. D. Radwill.\n\n1. F. C. Kelcourse and F. J. Herr, \u201cL5 System: Overall Description and System \nDesign,\u201d\u2019 B.S8.T.J., this issue, pp. 1901-1933.\n\n2. W.G. Albert, J. B. Evans, Jr., T. J. Haley, T. B. Merrick, and T. H. Simmonds, \nJr., \u2018\u201cL4 System: Terminal Arrangements,\u201d B.S.T.J., 48, No. 4 (April 1969), \npp. 993-1040.\n\n3. E. H. Angell, Y.-S. Cho, K. P. Kretsch, and M. M. Luniewicz, \u201cL5 System: \nRepeatered Line,\u201d B.S.T.J., this issue, pp. 1935-1985.\n\n4. J. H. Green and R. W. Sanders, \u201cL5 System: Line-Protection Switching,\u201d\u2019 \nB.S.T.J., this issue, pp. 2011-2034.\n\n5. R. E. Maurer, \u2018\u201cTerminal and Main Station Functions,\u2019\u2019 IEEE Transactions on \nCommunications, COM-22, No. 2 (February 1974), pp. 226-229.\n\n6. R. K. Bates and D. J. Zorn, \u201cL5 System: Signal Administration and Inter- \nconnection,\u201d B.S.T.J., this issue, pp. 2129-2145.\n\n7. J. F. Barry, S. Narayanan, and J. F. Oberst, \u201cL5 System: Jumbogroup Fre- \nquency Supply,\u201d B.S.T.J., this issue, pp. 2109-2127.\n\n8. J. L. Garrison, A. Olsen, Jr., and T. H. Simmonds, Jr., \u2018\u201cL5 System: Transmis- \nsion Networks and Magnetic Components,\u201d B.S.T.J., this issue, pp. 2203-2248.\n\n9. R. BE. Maurer, \u2018Analysis of Intermodulation Distortion in Double Balanced \nDiode Ring Modulators,\u2019\u201d\u2019 IEEE 1973 Int. Conf. Commun., Conf. Record, II, \nJune 11-13, 1973, pp. 51.21-51.26.\n\n10. B. B. Garg, \u201cFDM Synthesis Using Noise Loading,\u2019? IEEE 1974 Int. Conf. \nCommun., Conf. Record, June 17-19, 1974, pp. 7F-1 to 7F-3.\n\n12. J. L. Thomas, R. E. Anderson, and P. J. Baun, \u2018\u2018L5 System: Centralized Trans- \nmission Surveillance,\u201d B.S.T.J., this issue, pp. 2035-2064.\n\nCopyright \u00a9 1974 American Telephone and Telegraph Company \nTue Bey System TECHNICAL JOURNAL \nVol. 53, No. 10, December 1974 \nPrinted in U.S.A.\n\nThe 39A precision oscillator is designed to operate at 40 watts from a \n24-V +10-percent power source. The reference frequency can be set digitally \nto 6.12 MHz +8 X 10 over a tuning range of +4 X 107\". The output \nsignal 1s a 1-mW sinewave into a 100-ohm load. This performance is \nassured over a temperature range of 0 to 60 degrees Celsius. Oscillator out- \nline dimensions for length, width, and height are 35.8, 22.1, and 21.8 cm \nrespectively.\n\nThe 39A oscillator was developed to meet the precise frequency- \ncontrol requirements of the L5 jumbogroup frequency supply. The need \nfor frequency control as exemplified by the 39A oscillator is explained \nin the paper on the jumbogroup frequency supply in this issue. This \n5.12-MHz precision oscillator is characterized by long-term stability \nof <1 X 10- per day, and by short-term stability of <1 X 10-8 for \na 1-millisecond sampling time or <2 X 10~\" for a 10-second sampling \ntime. External frequency control is accomplished utilizing a compatible \ndigital word applied to the oscillator\u2019s 14-bit, TTL, precision, digital-to- \nanalog converter. Setting range of the converter is >+4 X 107\u2019, \nwhich gives a least-significant-bit resolution of 5 X 107\"; the con- \nverter section incorporates a linearizing network to reduce the fre- \nquency-set deviation from a straight-line function to <+5 X 107\" \nover a +4 X 107\u00b0 tuning range, and to a maximum variation of \ntuning sensitivity of +30 percent over the entire range of +4 X 107\u2019, \nwith less than +1 X 10~\u00b0 departure from a smooth curve.\n\nIl. GENERAL DESCRIPTION \nThe 39A oscillator can be separated into three major subassemblies:\n\ncircuit, a tuned feedback amplifier detector circuit (aac) \nfunctioning to provide constant low-level crystal-current \ndrive, and an output amplifier instituting sufficient load isola- \ntion for the output signal.\n\nA thermal and electrical noise-isolation subassembly which is \ncomposed of temperature-control circuits, power-supply regula- \ntors, and redundant electrostatic shielding amply protecting \nall major circuitry.\n\nThe digital frequency-control subassembly with a precision \ndigital-to-analog converter and a frequency-linearizing net- \nwork for remote steering of output frequency.\n\ncrystal-oscillator circuit in the 39A oscillator is the modified \ncommonly used in precision oscillators for many years.)?\n\nvaractor diode in the crystal network is used to control the operating \nfrequency of the network. Earlier circuit designs have varied the \noscillator-transistor de operating point to control the loop gain, but \nhere low-noise performance is achieved by fixing the de operating point \nof the oscillator transistor and adjusting its positive loop gain. Oscilla- \ntor loop gain is controlled by varying the forward bias of variolossers \nCR; and CR4,, which controls the degeneration of this transistor stage.\n\nThe acc amplifier circuit is shown in Fig. 2. This is a three-stage \ndirect-coupled feedback amplifier selected for low noise, simplified \ntuning, and having a minimum number of components. Passband \ntuning is accomplished by an uc network in the feedback path. This \namplifier operates with 38-dB in-band gain and has a 3-dB bandwidth \nof about 200 kHz. The rectified output from diodes CR: and CRez is \nused to control the oscillator gain for a crystal-operating current of \n100 to 200 wA.\n\nThe output signal of the acc amplifier is coupled to the final amplifier \ncircuit by a variable attenuator which is used to adjust the oscillator \noutput level to 1 mW into a 100-ohm load. The design of the output \namplifier is essentially the same as the acc amplifier.\n\nThe crystal oscillator, aac amplifier, and output amplifier combine \nto make up the oscillator subassembly shown in Fig. 3. The unit at\n\nthis stage in fabrication can be production tested as a functional sub- \nassembly. These tests include a thorough mechanical inspection, a \ntest of acc and output-level, and a check on harmonic distortion. \nThe long-term frequency of this precision oscillator is determined \nprimarily by changes of the crystal unit and directly associated Lc \ncomponents. In this respect, these components have been selected for \nminimum long-term drift. Short-term frequency variations (times of \nless than 0.1 second) can be attributed to transistor noise and small \nfluctuations in crystal current. The effects of crystal-current variations \nhave been reported\u2019 and are typically 1 X 10-\" per 0.02 dB for current \nlevels in the range of 100 to 200 wA at a frequency of 5.12 MHz.\n\nThe 5.12-MHz crystal units*:\u00ae used in the 39A oscillator operate at \n76\u00b0C minimum to 81\u00b0C. At this operating temperature, the frequency \ndependence on temperature is approximately 7 X 10-\u00b0/\u00b0C if the operat-\n\ning temperature of the crystal is <-\u00a30.25\u00b0C from the frequency turn- \ning point. In addition to the effects of long-term temperature variations, \nshort-term changes in temperature cause perturbations in frequency \ndue to thermal gradients within the crystal. A 1 X 10-\" frequency \nchange results from a 4 X 10-\u00b0\u00b0C/hour rate of change in crystal \ntemperature.\n\nTo reduce frequency variation effects due to ambient temperature \nchanges, the oscillator subassembly is housed in a two-stage tempera- \nture-stabilized oven. This two-stage oven has an ambient ratio of \n> 105:1 resulting in short-term temperature variations being measured \nas <l X 10-*\u00b0C per hour. The outer oven reduces ambient effects on \nthe temperature-control and voltage-regulator circuits to <5 X 10-2\u00b0C. \nTwo identical voltage regulators are used in the oscillator to reduce\n\nthe effects of power-supply variation on reference voltages, RF circuits, \nand control circuits. Figure 4 shows the environmental isolation \nsubassembly.\n\nI'requency control is accomplished by a varactor diode in the \ncrystal circuit. The precision voltage supplying this diode is derived \nfrom the precision digital frequency-control subassembly shown in \nFig. 5. This subassembly consists of a 14-bit digital gate network, \nwith strobing option, fabricated from standard Tru logic. The network \ncontains latching circuits, relay drivers, and relays (5 V dc) controlling \na constant-impedance voltage divider. Figure 6 gives the basic circuit \ndiagram for this digital-to-analog converter.\n\nFigure 7 is a simplified schematic of the divider excluding the control \nlogic. Resistors Rvi, Rye, and Ruz (or Rae if required) are selected by \ncomputer program at time of manufacture to precisely give the \nrequired nonlinear binary-voltage allowing a linear binary-frequency \nfunction of oscillator output. The loading resistor, Rumi or Re, de-\n\ntermines the network output voltage function and is selected for each \noscillator. A typical set of output-voltage characteristics for this net- \nwork are shown in Fig. 8 for values of Rai or Rue from 20,000 ohms \nto \u00a9 plotted as output voltage versus ratio of the programmed binary \ndivider.\n\nFigure 9 shows the tuning characteristics of a typical 39A oscillator \nover the frequency range from F,, (1 \u2014 4 X 10-8) to F, (1 + 4 X 107-8) \nand plotted relative to the frequency setting. The mean slope achieved \nover this range is within a few percent of the objective, with slope \nuniformity within about +10 percent over the range. Over the entire\n\nFig. 8\u2014Typical digital-to-analog converter output-voltage characteristic.\n\ntuning range of +4 X 107\u2019, slope variations are no more than +30 \npercent, and deviation from a smooth curve is no greater than \n+1 X 107.\n\nThe circuit also provides an LED indicator for each input bit, which \nallows a visual check of the frequency-setting word. This, in addition \nto an electrical output provision for automatically sensing the tuning \nstatus, gives a convenient check on the complete functioning of the \ndigital-to-analog converter.\n\nThe overall electromechanical design makes maximum use of current \ninstrument-fabrication methods. All printed-circuit boards are of a \nplug-in configuration simplifying assembly and testing by a large \nfactor. Hand wiring has been held to a minimum throughout the entire \nassembly. Polyurethane foam both isolates and supports the inner and \nouter oven assemblies. RFI or constant K bandpass filtering has been \nincorporated between major subassemblies where deemed necessary.\n\nFigure 10 illustrates the entire oscillator disassembled to show inner \nand outer ovens plus the inner and outer card frames. The front cover \nis mechanically attached to the digital-to-analog converter. The two \nconnectors seen on the precision digital-to-analog converter circuit- \nboard serve to provide a major part of the wiring interconnecting the \nrear input panel and the oscillator circuitry. The right front vertical \nedge of the external housing is designed to allow viewing of the LED\n\nindicator. All connections are made through the rear oscillator panel \nto the digital-word inputs, strobe pulses, 5- and 24-V de power inputs \nand to the sense and rF outputs.\n\nThe 39A oscillator has been developed as a secondary frequency \nstandard and is characterized by long-term stability of better than \n1 X 10~\u00b0/day, and short-term stability of better than 1 X 10-8 for a \n1-millisecond sampling time and 2 X 10-\" for a 10-second sampling \ntime. The total range of digital-to-analog frequency control is \n+4 X 10-7 from a nominal 5.12 MHz having slope variations < +30 \npercent with a maximum deviation of any point from a smooth curve \nof less than 1 X 10-'\u00b0. Midrange frequency control has a nominal \nstraight-line tuning characteristic of 5 X 10-\"/bit, with an average \nslope tolerance of +10 percent and maximum deviation for a smooth \ncurve of less than +5 X 10-\". This oscillator has been designed to \nminimize the effects of thermal, mechanical, and electrical noise to a \nhigh degree and is fabricated using quality hardware and reliable \ncomponents.\n\n1. W. L. Smith, \u2018Miniature Transistorized Crystal-Controlled Precision Oscillators,\u2019 \nIRE Trans. on Instrumentation, J-9, No. 2 (September 1960), pp. 141-148.\n\n2. H. S. Pustarfi, \u201cAn Improved 5 MHz Reference Oscillator for Time and Fre- \nquency Standard Applications,\u2019 IEEE Trans. on Instrumentation and Mea- \nsurement, [M-15, No. 4 (December 1966), pp. 196-202.\n\n3. T. C. Anderson and F. G. Merrill, \u2018\u201cCrystal-Controlled Primary Frequency \nStandards: Latest Advances for Long-Term Stability,\u201d IRE Trans. on In- \nstrumentation, J-9, No. 2 (September 1960), pp. 136-140.\n\n4. A. W. Warner, \u201cUltraprecise Quartz Crystal Frequency Standards,\u201d IRE Trans. \non Instrumentation, [-7, Nos. 3 & 4 (December 1958), pp. 185-188.\n\n5. A. W. Warner, \u2018Design and Performance of Ultraprecise 2.5 MC Quartz Crystal \nUnits,\u201d B.S.T.J., 39, No. 5 (September 1960), pp. 1193-1218.\n\nCopyright \u00a9 1974 American Telephone and Telegraph Company \nTue Bey System TEcHNICAL JOURNAL \nVol. 53, No. 10, December 1974 \nPrinted in U.S.A.\n\nThe jumbogroup frequency supply (srs) provides accurate and reliable \nsynchronization signals to L& jumbogroup multiplex carrier supplies. \nThe srs output-signal frequency is held to within several parts in 10\" of \nan input reference frequency. This extraordinary accuracy is achieved by \ndigital frequency control of a precision quartz oscillator. The excellent \nreliability required when the srs serves over 100,000 two-way voice circuits \nas achteved by redundant power and signal feeds, one-for-one protection \nof key circutts, and automatic protection switching.\n\nIn addition to its L& application, the srs will also serve as a regional \nfrequency supply in a new synchronizing frequency distribution network \nthat will upgrade the entire Bell System transmission plant.\n\nIt is the goal of the Bell System to maintain frequency offset in any \ntranscontinental circuit to within +2 Hz. This requirement is imposed \nby the most sensitive message signals, such as those in program- \nming, telegraph signals, and data signals. In an analog amplitude- \nmodulated system, such a frequency offset can be introduced into the \nmessage signal if the transmitting and receiving multiplex carriers are \nnot synchronized. In a single-sideband suppressed carrier multiplex \nformat as used in jumbogroup multiplex! terminals (Jmx), the multi- \nplexing carriers are not transmitted. Thus, precision frequencies must \nbe provided for the multiplex terminal at each geographical location.\n\nThe top channel frequency of the long-haul, analog, frequency- \ndivision multiplex system has been steadily increasing. With the in- \ncrease in top channel frequency (F), the frequency precision required \nof the multiplex carriers increases. This is because the frequency offset\n\n(AF) is limited to 2 Hz, whereas the ratio AF\u2019/F decreases as F in- \ncreases. In the L5 coaxial system, the top channel frequency is at 60 \nMHz, and the 2-Hz requirement results in a ratio of 3.3 parts in 108 \n(3.3/108). Because a typical signal may traverse through several ymx \nterminals, carrier frequencies must be synchronized with an accuracy \nof about 1 part in 10%.\n\nThe frequency precision demanded by the L5 system could not be \nprovided by the synchronizing network existing at the time L5 design \nwork was initiated. Several alternatives were studied. One alternative \nwas to provide a cesium frequency standard at each multiplex location. \nThe cost of the cesium standard, the need for providing at least two \nsuch standards for reliability at each location, and the need for training \npersonnel to maintain them made this option unattractive. An alterna- \ntive was to provide a primary frequency standard at a central location \nand transmit reference-frequency signals to different locations. Such a \nprimary frequency standard, called the Bell System Reference Fre- \nquency Standard? (BsrFs), has been designed. It comprises three \ncesium standards and provides highly accurate reference signals \n(better than one part in 101).\n\nAt the receiving location, several alternatives were open. One option \nwas to use a precision voltage-controlled crystal oscillator, whose fre- \nquency drift could be manually corrected periodically. The manual \ncontrol implied trained personnel and periodic maintenance. Moreover, \nfrequency-control circuitry is required for alarming and switching \npurposes in case of oscillator failure. Thus, it was decided to control \nand monitor frequency automatically, and the precision frequency \nsupply so designed (the jumbogroup frequency supply, or Fs) is the \nsubject of this paper.\n\nThe reference-frequency signal, even though highly accurate, may \nsuffer short term impairment like noise, switching transients, etc. It \nmay even be lost completely due to transmission difficulties. Because \nof the large number of circuits (more than 100,000) dependent upon \none JFs, it is imperative that precision frequencies be supplied to the \nJMx even when the reference signal is lost or impaired. To accomplish \nthis objective, a precision crystal oscillator capable of accurately free- \nrunning for a short period of time is employed in the srs. The precision \noscillator is loosely coupled to the reference signal such that it can \nautomatically correct itself when the unimpaired reference signal is \npresent and disconnect from the reference signal if the reference signal \nis impaired or lost. A digital memory is provided in the digital fre- \nquency-control circuit so that the oscillator can remain at the correct\n\nfrequency even when the reference signal is disconnected. The precision \noscillator and the precision digital frequency-control circuits are housed \nin a unit called the jumbogroup frequency generator (JFG).\n\nAn important consideration in the design of the srs is reliability. \nMore than 100,000 two-way voice circuits may depend upon one JFs. \nTo ensure reliability, automatic protection switching is provided for \nits key circuits. Several visual alarms are provided so that preventive \naction can be taken before protection switching is needed. The working \nof the frequency-control circuitry can be checked without any inter- \nruption of service. Redundant power feeds connect each srs to the \noffice battery and two cables are provided to each mx.\n\nHuman engineering has been given careful consideration. A precision \ntest unit to measure accurately the difference between two signals has \nbeen incorporated in the srs. This will enable the craftsperson to \ntroubleshoot the bay if necessary. A manual patching arrangement and \nmanual override have been provided to ensure that the craftsperson \ncan take full control of the bay if needed.\n\nThe srs furnishes three precision frequencies to the JmMx carrier \nsupply, where further processing occurs to generate all the carriers \nneeded for frequency multiplexing. The three frequencies are at \n20.48 MHz, 2.56 MHz, and 1.024 MHz. An overall block diagram of \nthe srs is shown in Fig. 1. It consists of three Jra@\u2019s arranged in a master-\n\nslave configuration, two frequency dividers, switch and alarm circuitry, \ntwo output buses, and a precision frequency-measuring test unit.\n\nThe master sra directly receives the 20.48-MHz reference signal. \nIn some locations, the reference signal will be at 2.048 MHz; a fre- \nquency converter has been designed to convert 2.048 MHz to 20.48 \nMHz. The master Jra compares its precision-oscillator frequency with \nthe incoming reference signal and corrects its oscillator based on the \nlong-term average of the frequency difference. The master src output \nsignal is at 20.48 MHz and is fed to the regular and standby sra\u2019s.\n\nThe incorporation of three sr@\u2019s in a JFs ensures detection of a JFe@ \nfailure even when the reference signal is noisy or absent. Such a detec- \ntion would quickly initiate alarm and switching circuits to take ap- \npropriate action. If only two sre\u2019s were provided in a srs, the failed \nJF@ could not be isolated if the failure occurred when the reference \nsignal was absent or noisy.\n\nThe regular and standby sre\u2019s use the output of the master sre as \ntheir reference. In this master-slave arrangement, the input signal to \nthe regular and standby sre\u2019s should be good all the time unless the \nmaster sFG fails. In contrast, the master sra reference signal, while \nhighly accurate when present, may be absent or noisy at times. For \nthis reason the master sFa does not produce frequency alarms unless \nthe frequency offset exceeds 1/10\u00ae and lasts at least two hours; the \nregular and standby sra@\u2019s produce frequency alarms if the offset \nexceeds 2/10\u00b0 and lasts for several hundred milliseconds (see the sre \ndescription in Section III).\n\nAs shown in Fig. 1, the output of the regular yra (20.48 MHz) is fed \nto the regular frequency divider; similarly, the standby sre feeds the \nstandby frequency divider. The function of the frequency divider is to \ngenerate 2.56-MHz and 1.024-MHz signals and combine all three \nfrequencies (20.48 MHz, 2.56 MHz, and 1.024 MHz) together for \ntransmission on one coaxial cable. The output of the regular divider \nis transmitted to the two output buses through two coaxial switches. \nThe coaxial switches are operated by the alarm-and-switch-control \nunit which processes information from the three jre\u2019s and the level \ndetectors. The output bus has 32 taps, and because of reliability con- \nsiderations one JFs may serve up to 22 Jmx\u2019s. Each smx is connected \nby two cables, one from bus A and the other from bus B. Thus, the \nregular divider feeds both sides of the sux. The output of the standby \ndivider is fed to the two coaxial switches. If a regular sra@ or frequency \ndivider fails, the alarm and switch control unit will operate the coaxial \nswitches so that the standby sre and divider will supply signals to\n\nboth sides of the ymx. A phase comparator is provided to compare the \nphases of signals from the regular and standby dividers; this enables \nthe craftsperson to manually transfer output signals from the regular \ngra to the standby sre with minimal disturbance.\n\nSeveral features have been included in the srs to facilitate trouble- \nshooting and maintenance. A precision frequency-measuring test unit \nhas been designed and incorporated as a part of the srs. A patch panel \nhas been included to provide a convenient and non-service-affecting \nway to rearrange the remaining JFe\u2019s if a ura fails. Several test points \nand a level meter have been provided. The coaxial switches can be \npatched around without affecting service; the alarm and switch control \ncircuit could then be exercised for routine checkup of the coaxial \nswitches and control circuitry.\n\nThe srs is shown in Fig. 2 in a double 9-foot bay. A 7-foot-bay ar- \nrangement is also available. At the top of each side is a distribution \nbus; these buses provide 32 redundant (64 total) outputs. The three \ngrq@\u2019s are on the left side with the master on top and below it the \nregular and standby jrae\u2019s with the ura patch panel between them. \nEach sre has its alarm display panel above its oscillator panel on which \nthe 14 bits of the oscillator frequency-control word are displayed. On \nthe right side, from top to bottom, are shown: three sets of de-de con- \nverters, test unit and associated dc-dc converters, regular frequency \ndivider, phase comparator, standby frequency divider, and alarm and \nswitch control unit.\n\nMost of the srs circuitry consists of digital integrated circuits (1c\u2019s) \nand various discrete components mounted on two- or four-layer\n\nprinted-wiring boards. Additional apparatus such as thin-film, hybrid, \nintegrated-circuit amplifiers and crystal filters are also employed.\n\nThe wide use of digital 1c\u2019s in the Jrs made it possible to achieve \nsophisticated control and alarm functions that would be impractical \nwith discrete components. Western Electric-manufactured 101-type \nemitter-coupled logic (EcL) was chosen for reasons of speed and \navailability. Since only small-scale integration (sst) members of this \nlogic family were available during the srs design phase, a large number \nof \u201c\u2018chips\u2019\u2019 was necessary to implement all srs functions\u2014about 1100 \nlogic chips in a full yrs. A typical two-sided printed-wiring board is \nshown in Fig. 3.\n\nThe sre is discussed in detail in the next section. The frequency \ndivider, protection switching, and test unit are discussed in the suc- \nceeding sections.\n\nThe sFa produces an accurate 20.48-MHz sinusoidal output signal \nwith a AF/F normally less than 1/10\" relative to its input signal. This \nsmall AF/F is achieved by combining an extremely stable crystal \noscillator with precision frequency-control circuitry. Frequency ac- \ncuracy is provided by reference to a signal whose ultimate source is \nthe BSRFs.\n\nThe oscillator is the 39A digitally controlled crystal oscillator.\u2019 It \nhas a drift rate less than 1/10\" per day, low enough to operate without \ncorrection for several weeks and still meet srs frequency-accuracy \nrequirements. This oscillator\u2019s frequency is controlled by a 14-parallel- \nbit binary-control word developed by the sre frequency-control circuit. \nThe control word is changed when comparison of the JFe\u00a2 input and \noutput signals indicates a need to correct the oscillator frequency.\n\nFigure 4 is a block diagram of the src showing the relationship be- \ntween oscillator, input, output, and frequency-control circuits. The \n39A digitally controlled crystal oscillator (pcxo) has a 5.12-MHz \noutput which is multiplied to 20.48 MHz in the sre output circuit. The \n20.48-MHz input signal, after being filtered and amplified in the sra \ninput circuit, is compared with the 20.48-MHz output signal in the \nfrequency-control-and-alarm circuit, which produces the frequency- \ncontrol word and frequency alarms. The logic-interface circuit processes \nfrequency, level, and voltage alarms generated within the sre for \ntransmission to the srs alarm-and-switch-control circuit and for dis-\n\nplay on the sre itself. This alarm processing is controlled by a status \ninput from the srs alarm and switch control.\n\nThe major consideration in designing the Jra was the character of \nthe signal supplied to the srs as a frequency reference. As mentioned \nearlier, this signal could be intermittent, noisy, or absent due to trans- \nmission difficulties. Furthermore, at the time of Jra development, the \nproposed reference-signal-transmission path contained phase-locked \noscillators that could introduce frequency offsets as great as several \nparts in 10\u00b0 if they lost phase-lock. These phase-locked oscillators were \nnot used in the final Bsrrs distribution network. Because of these \nreference transmission difficulties, the JrF@ was designed to rely on \nonly one characteristic of the reference signal\u2014its average frequency \nas measured during times of good transmission.\n\nThe low drift rate of the 389A pcxo allows a sre to ignore the refer- \nence during times of poor transmission. Although this drift rate is low, \nit is not zero and cannot be ignored. In the sra the pcxo frequency- \ncontrol word is held constant until the average of a large number \n(256 minimum) of short (16 seconds), valid frequency comparisons\n\nbetween input and output signals indicates a frequency offset greater \nthan 1/10'\u00b0. Then the control word is changed to move the pcxo \nfrequency 2/10\" in a direction opposite to the offset. This frequency- \ncontrol operation can be separated into three functions: frequency- \noffset detection, averaging, and frequency correction.\n\nFigure 5 shows, in general terms, the basic portion of the circuit by \nwhich the ura detects small values of AF/F between two frequencies \nA and P. In the first portion of this circuit, divider D and the two \nmixer-bandpass filter circuits develop a frequency A/D + AF. The \nfinal portion of the circuit consists of a frequency divider N whose \noutput controls a gate, allowing the frequency A to be counted for one \ngate period, resulting in the count C. C, is the count when AF = O. \nThe count AC contains the AF/F information obtained in the time T.\n\nImplementation of this circuit involves the selection of values for \nD and N and the number of stages in the counter. To maximize the \nAC/T ratio, D is made as large as possible within the constraints im- \nposed by the bandpass filter requirements and the short-term frequency \nstability of the signals. In the ura, D = 320; hence, A/D = 64 kHz. \nN is selected to provide the necessary precision (AC) in a reasonable \ntime (7). N is also considered in the counter design. A counter large \nenough to count to C would be larger than necessary. In the sre fre- \nquency-control circuit the largest AF /F that must be measured by the \ncounter is 1/108; this limits AC. If one selects N and the number of\n\ncounter stages so that after multiple overflowings C, leaves a remaining \ncount of all zeros, the number of counter stages need only be sufficient \nto detect the sense and magnitude of AC. In the sra, N = 2\u201d\u00b0 and the \ncounter has 19 stages. This results in a circuit capable of detecting \nsense and magnitude of frequency offsets in the 1/10\" to 2/10 range.\n\nFigure 6 shows the application of this circuit in the src. During the \n16.384-second gate period provided by cate 1, the count signal (the \nJFG output frequency) is counted. At the end of that period, AC is \nexamined and if it indicates a AF/F > 1/10\", high or low, a pulse is sent \nto the high or low input of the accumMULATOR. The INHIBIT 2 signal will \ninterrupt this process and reset the circuit should other, quicker- \nacting circuits determine that the frequency comparison is not ac- \nceptable for frequency control.\n\nAveraging of the results of the 16.384-second frequency comparisons \nis done in the accumuLatTor. This circuit is a nine-stage up-down \nbinary counter. Starting from its midrange, or reset, position, this \ncounter counts up or down depending on the sense of AF/F. When this \ncounter has received 256 more pulses in one direction than in the other, \nit sends two pulses to the memory by way of the FREQ CONT MODE \nSELECTOR.\n\nThe normal position of the FREQ CONT MODE SELECTOR will connect \nthe ACCUMULATOR to the Memory. The memory provides the 14-bit \nfrequency-control word to the 39A pcxo. In the 39A a D/A converter \nplus varactor diode circuit provide a linear relationship between \noscillator frequency and digital control word. The least significant bit \n(usB) of the control word corresponds to AF/F = 5/10\"; so the 14-bit \nword provides a frequency-control range of +4/10\u2019. The mMmEmMoRyY \nkeeps the LtsB at 0 and increases the higher bits by one for each pulse \nit receives. Thus, the two pulses from the ACCUMULATOR result in a \n2/10 frequency change.\n\nThe maximum correction rate of the frequency-control circuit is \n2/10!\u00b0 per 256 frequency comparisons made at a rate of one every \n20.48 seconds; this converts to 3.4/10\u00b0 per day.\n\nFigure 6 shows a number of frequency alarms being generated. The \ncircuits generating these alarms are interconnected to utilize the inverse \nrelation between the value of AF'/F and the time needed to detect it.\n\nThis is the coarsest frequency alarm and utilizes the 30-Hz band- \nwidth of the 64-kHz bandpass filter. At the output of this filter a AF \nof 15 Hz or greater causes a 3-dB or greater drop in signal level. This \ndrop corresponds to a AF /F of 7.5/10\" or greater. A level-detector \ncircuit, DET, detects this and produces the 1/10\u00b0 aum and iNursIT 1 \nsignals. (The 1/10\u00b0 designation is nominal.)\n\nGATE 2 is called the frequency-alarm gate, and the circuits generating \nit and using it comprise the frequency-alarm circuit. This circuit uses \nthe frequency offset detection scheme shown in Fig. 5. N = 2 and the \ncounter has 12 stages; so the circuit has a range from 1/10\u00b0 to 2.5/108. \nThe 1nHIBIT 1 signal interrupts and resets this circuit which makes \na AF/F measurement every 320 ms. The main function of this circuit \nis to produce the 1/108 nz atm and 1/108 Lo aLm and shut down the \nfrequency-control circuit when AF/F > 1/108.\n\nSince the counter used in the frequency-control circuit can detect \n1/10\u00ae offsets, it is also used to produce the 1/108 nr aum and 1/108 \nLO ALM. This provides protection in case of frequency-alarm-circuit \nfailure.\n\nThe 2/10\u00b0 nr and to alarms are produced in the frequency-control \ncircuit. These alarms alert maintenance personnel to problems before \nthey affect service.\n\nA frequency drift less than 3.4/10\u00b0 per day can be compensated for \nin the gra. An oscillator with a drift rate greater than 3.4/10\u00b0 per day \nwill eventually produce a frequency alarm. An oscillator with a lower \ndrift rate will not cause such an alarm and no action would be taken \nuntil the oscillator reached the end of its control range and had to be \nreplaced. The pRiFT ALARM forestalls this. The DRIFT ALARM circuit \nnotes each frequency-correction pulse from the ACCUMULATOR, and \ngenerates the DRIFT ALM signal when the memory has received 64 more \npulses in one direction than in the other. The DRIFT ALM signal notifies \nmaintenance personnel, who manually reset the circuit and record the \nalarm. The record of prirT ALM times and oscillator frequency-control \nwords indicates oscillator drift rate.\n\nIf the oscillator is at the end of its control range; i.e., its control \nword is all ones or all zeros, the MEMorRy produces the END OF RANGE \nALM and inhibits itself from further changes. The oscillator must then \nbe replaced. However, long before this happens, maintenance personnel \nshould have predicted it from the drift-alarm records.\n\nMost functions within the Jr@ are automatic, but installation and \nmaintenance operations must be performed manually. The sre design \nhas tried to facilitate these operations.\n\nAt time of installation, or following a loss of power, the oscillator \nfrequency might need to be changed so much that the 3.4/10\u00b0 per day \nrate would be far too slow. Shown in Figs. 6 and 7 are three features \nto provide more rapid frequency correction. First, there is the MEM \nRST switch which can put the frequency control word in its midrange \nposition. Second, there are the 1/10\" up and pown switches which \nchange the oscillator frequency by 1/10\" with each operation. Third, \nthere is the speed-up circuit which uses the 320-ms period of the\n\nfrequency-alarm circuit to correct the oscillator at a 6/10 per second \nrate. To operate the speed-up circuit, the FREQ CONT MODE switch is \nmoved from the Nor to SPEED-UP position and then the SPEED-UP \npushbutton switch is operated. The frequency-alarm circuit automat- \nically stops sending pulses to the MEMoRyY when the AF\u2019/F is reduced \nbelow 1/10\u00b0. Figure 7 shows that all these controls are located behind \ncovers to forestall their misuse. Note the angle bracket on the cover \nof the FREQ CONT MODE switch; this bracket forces one to return the \nswitch to its Nor position before the cover can be closed.\n\nIntermittent alarms have always been the bane of maintenance \npersonnel. In the src, alarm indications are held until they can be \nnoted or until their cause has been corrected, whichever comes later. \nFigure 7 shows the Atm rsT switch which will extinguish an alarm lamp \nif the trouble has gone. If the trouble still exists, the lamp cannot be \nextinguished. An exception to this lock-up is that all alarms associated \nwith the srs reference signal do not lock up.\n\nA failure of the frequency-alarm circuit could go unnoticed. The sre \nprovides a simple way of checking this. Operation of the FREQ ALM \nTEST switches, HI or LO, artificially shortens or widens the 256-ms gate \nby 1.6 ms, causing the effect of a 1/108 ur or xo offset. If the circuit \ndetects this, a green lamp lights; if it does not, the red rain lamp lights. \nNo other indications are generated in this operation. But if one \ndesires to check the interaction of this alarm circuit and the srs alarm \nand switching circuits, operation of the yrs ALM & Sw TEST switch \n(see Fig. 7) simultaneously with the H1 or Lo switch will allow 1/10\u00b0 \nalarms to be transmitted from the sre to the srs. Note that these \noperations do not disturb the sra@ output frequency.\n\nAs shown in Fig. 1, the sFs contains two frequency dividers. Called \nthe regular divider and standby divider, these units are connected to \nthe outputs of the regular yra and standby sre, respectively. Each \nfrequency divider provides as an output 1.024-MHz, 2.56-MHz, and \n20.48-M Hz signals multiplexed onto one coaxial cable.\n\nFigure 8 is a block diagram of the basic circuit showing that phase \ncontrol is provided for all frequencies. The 20.48-MHz phase-adjust \nnetworks provide a 180-degree control range. The fine phase adjust-\n\nment for the 1.024-MHz and 2.56-MHz output signals is done at 20.48 \nMHz to avoid component problems associated with lower-frequency- \nphase-adjust networks. The 180-degree phase adjust at 20.48 MHz \nyields 22.5 degrees at 2.56 MHz and 9 degrees at 1.024 MHz. Coarse- \nphase-adjust circuits are included in the two dividers to increase the \nrange of phase adjustment at the two lower frequencies. Each divider \nskips a single input pulse for each push of its associated pushbutton, \nthus producing a step change in its output phase. From each divider, \nthe test outputs, shown in Fig. 8, are connected to the phase-compara- \ntor panel, which provides one meter for each frequency. A meter null \nis an indication of phase coincidence. All three meters must be nulled \nto indicate that the signals from the two dividers are in phase. In \nadjusting phase to produce these nulls, the coarse-phase adjust of the \nin-service divider should not be used.\n\nProtection switching plays a vital role in ensuring the continuity \nof sFs output signals. Figure 9 shows the switching arrangement; the \nswitches are fabricated with coaxial, dry-reed contacts. The normally \nclosed contacts of these switches are magnetically biased so there is no \nneed for switch coil current while the switch is released. Since the \nswitch is normally released, a loss of switch power will normally not \naffect the srs output signal. As shown in Fig. 9, with the switches\n\nreleased, the regular divider feeds detectors A and D and output \nbuses A and B, while the standby divider feeds detectors B and C. \nWhen the switches are operated, the standby divider feeds detectors \nA and D and output buses A and B, while the regular divider feeds \ndetectors B and C.\n\nThe switch-control circuit energizes (to operate) or deenergizes \n(to release) the switch coils in a \u2018\u2018make before break\u2019\u2019 sequence. This \nfeature has little importance if the switching is being done automat- \nically due to a signal fault. However, if the switching is being done \nmanually for any nonemergency reason, the switch input signals can \nbe adjusted to be of equal phase and level; this combined with the \n\u2018\u201c\u2018make before break\u201d feature minimizes perturbations of the JFs out- \nput signal caused by nonemergency switch operation.\n\nManual control is provided in the srs alarm-and-switch-control \npanel by means of pushbuttons. Accidental manual switching is pre- \nvented by requiring multiple simultaneous pushbutton operations for \nswitching. It is possible to override automatic-switch-control signals \nmanually.\n\nThe alarms received from the four level detectors (Fig. 9) and the \nJFq\u2019s are processed by the switch-control circuit, which sends a com-\n\nmand to the switches if an automatic switch is warranted. The switch \nwill operate if detector A or D indicates an unusual change in signal \nlevel or if the regular sre indicates it is impaired sufficiently to require \nbeing switched out of service. Should detector B or C indicate an un- \nusual change in signal level or the standby sre indicate it is impaired, \nswitch operation will be inhibited.\n\nIf a sre failure occurs, the sFs may have to operate for some time \nwith only two sra@\u2019s. Patching facilities are provided for restructuring \nthe sra interconnections (Fig. 10). For example, if the master sre \nfails, the standby sra temporarily serves the master sFa@ function. The \nalarm-and-switch-control circuit must be modified to recognize the new \narrangement. This modification is achieved by selecting a position on \na rotary switch corresponding to the new arrangement, and then \noperating an associated pushbutton switch. Because the srs can func- \ntion for some time with only two sFe\u2019s, the Jrc is spared on a regional \nrather than a per-office basis. For most of the other units in the srs, \nspares are included either in the srs itself or in each office.\n\nA jack-and-patch-plug arrangement has been provided in the alarm- \nand-switch-control panel so that the srs output signals can be routed \naround the coaxial switches. Using this arrangement to take the \nswitches out of service produces small phase and amplitude \u2018\u2018hits\u201d on\n\nFig. 10\u2014Jumbogroup frequency supply operation using two jumbogroup fre- \nquency generators.\n\nthe JFs output signals. When the patching operation has been com- \npleted, the signal through the patched path is within 2 degrees and \n0.2 dB of what it had been through the switch path.\n\nA rotary-switch-and-power-meter arrangement has been provided \non the alarm-and-switch-control panel to enable the craftsperson to \nread the level of each frequency (1.024, 2.56, and 20.48 MHz) at \ndetectors A and B shown in Fig. 9.\n\nThe JFrs incorporates a test unit to measure small frequency offsets, \nsince a typical office does not have precision-frequency-measuring \nequipment.\n\nThe test unit can compare two frequencies of approximately 20.48 \nMHz, and display on numeric indicators the fractional frequency offset \nin parts in 10\" or 10\". The test unit has two modes of operation, either \nof which can be selected by a switch on the front of the panel. The \n6 sEc, or normal, mode measures frequency differences of parts in \n10\u00b0. The 30 sEc, or extended precision, mode measures frequency \ndifferences of parts in 10\". The frequency-offset information is dis- \nplayed on LED numeric readouts.\n\nSignals to be compared must have a level greater than \u201430 dBm \ninto 75 ohms, and the frequency difference must be less than one part \nin 10\u00b0. If either of these conditions is not met, the OUT OF RANGE \nindicator will light, and the test unit will not operate. The counTING \nindicator is lit while the test unit is counting. When the couNTING \nindicator goes off, information is transferred from the counter to the \nnumeric readouts. After approximately six seconds, a new count is \ntaken.\n\nThe internal functions of the test unit can be divided into three \nsections: gate generator, counter, and display. The gate generator \nprovides a gate whose period is a function of AF, the frequency differ- \nence between the two input signals. In the counter, the duration of the \ngate is measured by counting proportional to the fractional frequency \noffset (AF/F). This is similar to the frequency-offset-detection tech- \nnique used in the src. The information in the detection counters is \ntransferred to a digital display unit.\n\nThe first sFs production unit was shipped from the factory in June \n1973. There are presently more than ten srs bays installed at L5 main\n\nstations. The operational performance of these units has been excellent, \nand they have met all of their design objectives.\n\nThe availability of these accurate and reliable frequency supplies \nhas stimulated the modernization of the entire Bell System synchroni- \nzation network.\u2018 Beginning in 1974, the United States will be divided \ninto approximately twenty synchronization regions, and telephone \noffices in each region will receive their synchronization signals from a \nregional frequency supply (RFs). Certain L5 main stations will become \nregional syne centers, with the L5 srs serving as the rFs and providing \n64-kHz and 512-kHz sync signals to telephone offices in its region. In \nareas where no Ld facility exists, a special two-sFre version of the Jrs \nwill serve as the rFs. Thus, the entire frequency-division multiplex \nplant, including both radio and coaxial transmission systems, will soon \nbe synchronized through srs equipment.\n\nIn addition to serving L5 and the Bell System synchronization net- \nwork, the srs has sufficient frequency precision to serve future systems \nhaving ten times the bandwidth of L5.\n\nThe authors acknowledge the many contributions of R. E. Benjamin \nto the srs development; he has been involved with the srs since its \ninception. Several others contributed to the development of the sFs, \nin particular, W. W. Brown, L. J. Finelli, D. A. Lane, W. G. Albert, \nD. M. Pierce, and W. C. Schmidt. Special mention is due W. L. Smith \nand H. 8. Pustarfi whose expertise in the field of precision frequency \nsources led to the development of the 39A oscillator.\n\n2. J. F. Oberst, \u2018\u201cKeeping Bell System Frequencies on the Beam,\u201d Bell Laboratories \nRecord, 52, No. 3 (March 1974), pp. 84-89.\n\n3. A. F. Flint and H. S. Pustarfi, \u201cL5 System: 39A Precision Oscillator,\u2019\u2019 B.S.T.J., \nthis issue, pp. 2097-2107.\n\n4. R. E. Powers, \u201cReference Frequency Transmission Over Bell System Radio and \nCoaxial Facilities,\u2019 Proc. of 28th Annual Frequency Control Symposium, \nFort Monmouth, N.J., U.S. Army Electronics Laboratories, May 29-31, 1974.\n\nCopyright \u00a9 1974 American Telephone and Telegraph Company \nTue BELL SystEM TECHNICAL JOURNAL \nVol. 58, No. 10, December 1974 \nPrinted in U.S.A.\n\nEach L5 coaxial line is capable of carrying three 8600-channel basic \njumbogroup signals that are translated to and from the L\u00e9 line spectrum \n(3.124 to 60.556 MHz) through jumbogroup multiplex equipment. The \nLS line signal also includes several line pilots and switching, maintenance, \nand reference signals. Administration of all these signals ts performed by \nthe L5 line connecting circuits, which vary in complexity with the type of \nmatin station they serve.\n\nIn addition to the circuits required to handle the various components \nof the Ld line spectrum, signal administration is also required at basic \ngumbogroup frequencies, before the jumbogroup multiplexing step, to \nallow interconnection to lower-order multiplex or other systems using the \nbasic jumbogroup frequency format of 0.564 to 17.548 MHz. To perform \nthis function, the basic jumbogroup trunk bay was developed and provides \ninterconnection flexibility hitherto unavailable for direct connections to \nother long-haul systems, such as L4, LMp, Tp, or TH radio, or LS systems \nof another route.\n\nThe L5 coaxial line is a transmission facility with a message capacity \nof three jumbogroups. These jumbogroups are placed on the line in the \nfrequency format shown in Fig. 1. Each jumbogroup begins as a \nbasic jumbogroup signal, formed by the mastergroup multiplex-2 \n(mMx-2) frequency-division multiplex or the basic jumbogroup trunk \nequipment. This basic jumbogroup consists of six 600-channel basic \nmastergroups and has a frequency assignment identical to that of the \nL4 line assignment.! Each of three 3600-channel basic jumbogroups \nis translated to the L5 line spectrum through the jumbogroup multiplex\n\n(gmx) frequency-division multiplex equipment? to form the 10,800- \nchannel line signal.\n\nseveral pilots used for the dynamic equalization and temperature \nregulation of the line,\u2019 switching signals required for the control of the \nline-protection switching system-3 (LPss-3),4 fault-locating signals \nused for line maintenance tests in the transmission surveillance system \n(rss),5 and a reference frequency signal used by the jumbogroup \nfrequency supply (srs).\u00b0 Several rules and administrative functions \nmust be applied to the message and other signals forming the L5 line \nspectrum which vary under specific situations. The line-connecting \ncircuits described in Section II perform all of these functions.\n\nIn the evolution of the various long-haul systems such as the L- \ncarrier and TD and TH radio systems, the need for more direct inter- \nconnection between these systems in large channel blocks became \nincreasingly evident. Prior to the development of L5, the predominance \nof intersystem message interconnection was done on a basic master- \ngroup basis,* which requires the use of costly mmx terminals and \nmastergroup connectors. The need for a simpler and less expensive \nmeans of intersystem interconnection was recognized early in the L5 \ndevelopment, especially in view of the large 180-mastergroup capacity \nof this new system.\n\nThis need was met with a new bay, designated the basic Jumbogroup \ntrunk bay (BseT), which allows interconnection of single- or multi- \nmastergroup signals in the basic jumbogroup spectrum; i.e., before the \njumbogroup multiplexing step to the L5 spectrum through sux \nequipment. Interconnections may be made directly to radio systems \n(using 3A wire line entrance links), L4 systems, L-carrier mastergroup \ndigital (LMp) terminals, mmx-2 terminals, or other L5 systems. The \nBJGT circuits are covered in detail in Section ITI.\n\nAs mentioned in the introduction, the function of the line-connecting \ncircuit is to process the Ld line signal in accordance with circuit require- \nments and certain administrative rules. Although many options are \nrequired to handle the various conditions that arise, there are only \nthree basic line-connecting arrangements, one for power-feed main \n(PFM) stations, another for switching power-feed main (sPrM) stations \nand, finally, one to cover terminal stations or terminal main (TM) \nstations.!\n\nAll line-connecting equipment is located in the line transmit-receive \nbay. With the exception of line-connecting equipment, transmit-\n\n*The basic mastergroup is the \u201cU600\u201d output of the L-type multiplex terminal \n(Ref. 7). \nt The distinction among the four types of main stations is covered in Ref. 8.\n\n\u2014_\u2014\u2014\u2014_\u2014_\u2014\u2014\u2014_\u2014 \nTO LINE \nRCVG LINE CONN | TRMTG LINE CONN i eaonn \nDETR FROM \n(MAJOR ALM) FLO \nTO \nTO LPSS-3 PILOT TO \nTSC DETR RESUP TSC \nFROM SW \nTRMTG PIL | | \nFROM TST GEN TRMTG \nRCVG LINE TST \nEQUALIZER \n/ mco \n\u201co'une + o \nTO LINE \nJFS PILGEN | INTERCONN \n[ DETR LH +] i \n(IDLE ALM) TO \nTRMTG \nEQUALIZER \nEQL SIG \n! \u2018 J \nFROM | FROM \nTo | stBy | RSTN LH | a FROM \nLPSS-3 | LINE TRMTG \u00a7fsc \nDETR SW \nTO TO LINE PILOT \nRSTN INTERCONN ae TO - GENERATOR \nDETR moar eet le BT Ie \\\\rrom STBY LINE ec CKT \n(INPUT ALM) TO JMX TO ANOTHER FROMANOTHER \u2018MX Poe gs fo \nTERMINAL L5 ROUTE L5 ROUTE TERMINAL a aL Baa \nPS ee a See ae SS ee ] \n| LINE INTERCONN DETR CKT Tine INTERCONNECTING CKT (THRU OR BRANCH)! PILOTS | | ule \nFrom (INPUT ALM To AEF FREO | Reve \nDETR LINE ] SIGNAL \nBRIDGE INTERCONN | FROM \nPOINTS LPSS-3 \nSWITCHES | Sanat \nee | ie Sins SF Sees Ie ee De reel ]\n\nreceive bays are virtually identical for all types of stations.*\u00ae The \nline-connecting circuits are simplest at Prm stations where no signal \nprocessing is performed, and all message continues through the office. \nA pad and hybrid, for fault-location oscillator (FLO) access to the line, \nare the only apparatus in the line-connecting circuit. At sprm stations, \nall message is again connected through; however, switching apparatus, \nfiltering, and additional access to the transmission surveillance \ncenter (Tsc) are added to the line-connecting circuits for system \nadministration.\n\nTerminal or Tm stations require similar basic equipment units in \nthe line-connecting circuits, but the terminal station options within \nthat equipment are simpler. Each Tm station transmit-receive bay \nrequires six basic circuits to perform the line-connecting functions: the \nreceiving line-connecting circuit, transmitting line-connecting circuit, \npilot generator circuit, line-interconnecting circuits, line-interconnect- \ning detector circuit, and, optionally, the line-branching circuit. These \ncircuits are shown in the block diagram found in Fig. 2, and are \ndiscussed more thoroughly below.\n\nThe signal administration role of the line-connecting circuits in T \nstations may be reduced to six major functions:\n\n(t) Message Administration: Provides routing of the message \nsignal for any combination of the three jumbogroups.\n\n(it) Pilot Administration: Introduces equalizing and temperature \npilots to the line and provides pilot blocking when required.\n\n(i172) Switching Administration: Provides line switches, introduces \nswitching signals to the line, and provides detection access and \nsignal blocking when required.\n\n(tv) Transmission Surveillance Administration: Provides access to \nthe line at strategic points for Tsc analysis, and furnishes signal \nblocking when required.\n\n(v) Reference Frequency Signal Administration: Introduces a \nreference frequency signal to the line, and provides signal \nblocking when required and a distribution network for circuits \nrequiring the use of a reference frequency.\n\n(vt) Restoration Access: Provides access to the regular transmitting \nand receiving lines for restoring L5 over other facilities and \naccess to the standby transmitting and receiving lines for \nrestoring other facilities over L5.\n\nThe simplest of the Tm station line-connecting circuits is required \nwhen all message passes through the office; i.e., no jumbogroups are \nbranched to other L5 lines or are dropped to mx equipment. In such a \ncase, the line signal entering the line-connecting circuit from the \nreceiving equalizer (Fig. 2) would only connect (via splitting hybrids) \nto the A and B Turv cxtT modules at the output of the receiving line- \nconnecting circuit. No message blocking filters are required in the \nTHRU CKT modules for this application.*\n\nIn the example shown in Fig. 2, the message administration is \narranged to split the jumbogroup signals into three directions: one \njumbogroup connected through, one branched, and one dropped to \njmx. Assuming JGl is connected through, s@2 branched, and Je@3 \ndropped, blocking filters would be assigned to block sa2 and Ja3 in \nthe THRU CKT and Jel and Ja3 in the BRANCH cKT. No blocking filters \nare required in the drop circuit, since the smx circuitry selects that \njumbogroup for which it is equipped.\n\nAny combination of jumbogroups can be passed in the through, \nbranched, or dropped path. The blocking is accomplished through the \nuse of five arrangements of filter designs developed for L5 use. The \nrequirements for jumbogroup blocking are outlined in Fig. 3.* In all but \njumbogroup 2 blocking, single high-pass or low-pass filter designs are \nused ; however, in the case of jumbogroup 2, filters d and e are paralleled \nwith a \u201c\u2018split-apart\u2019\u2019 filter at the input and output of the paralleled \npair. The advantage of split-apart filters, as compared to hybrids, \nallows better return loss with lower in-band loss.\n\nA minimum of 80-dB out-of-band discrimination for each blocking \narrangement is required. Return loss is approximately 26 dB (75 ohms). \nMaximum insertion loss is less than 4.0 dB.\n\nIdeally, it is desirable to maintain continuous-line pilot continuity \nthroughout a frogging\u00ae section (approximately 800 miles). Two factors, \nhowever, preclude this possibility. When jumbogroup blocking filters \nare required at TM stations, one or more line pilots are attenuated to \nsome degree because they fall within the attenuation region of the \nfilters. In addition, the stability of each line pilot is adversely affected \nas the number increases of regulating repeaters or dynamic equalizers \nthrough which they must pass. Therefore, the pilots are blocked and \nreinserted at specific intervals.\u2019\n\nSince the administration of individual pilot blocking and reinsertion \nwould be unwieldy, all pilots are blocked when any pilot or jumbogroup \nsignal is blocked. Thus, rules for pilot blocking were established such\n\n* A more detailed description of filter design techniques for L5 is given in Ref. 10.\n\nthat all pilots are blocked when either (7) jumbogroup blocking is \nprovided or (77) pilots will otherwise pass through more than four E3\n\nequalizers. \nMinimum requirements for the pilot band-elimination filters are as\n\nThe line-connecting circuits provide several functions for the \nadministration of the Lpss-3. A bridged access point is provided at two \nlocations, (z) on the line side of the receiving switch and (27) on the \nline side of the transmitting-line switch hybrid (Fig. 2). The switch- \ninitiator circuit monitors the 42.880-MHz pilot at these points for \nLPss-3 operations.\n\nUnlike L3 and L4, the L5 line switches are located within the line- \nconnecting equipment to reduce the cable length in the through signal \npath. This was made possible through the use of parallel switch \ncabling, as opposed to the series arrangement in L3 and L4 systems.\u2018\n\nTwo switches are provided in the receiving line-connecting circuits, \ndesignated the RcvG sw and the TERM sw. The Rcv6G sw furnishes access \nfrom the standby line when the associated regular line is out of service \nfor any reason. This switch also provides a termination to the out-of- \nservice line. The TERM sw is either automatically operated as a result \nof a line overload condition, thereby preventing noise propagation to \nother systems, or manually operated under certain abnormal line \nconditions. The transmitting line-connecting circuit also includes \ntwo switches, the TRMTG sw and the mco (message cutoff) sw. The \nTRMTG sw provides access to the standby line when the associated \nregular line is out of service. The mco sw, also controlled by the \nLPss-3, provides a means for opening the transmitting end of a regular \nline, primarily as an access for line equalization.\n\nAs indicated in Fig. 1, the line-switching signals occupy the band- \nwidth between 68.76 and 68.78 MHz. These signals are blocked in the\n\nline-interconnecting circuit at every switching station to prevent \ninterference with switching functions between switching sections.\n\nEach transmit-receive bay at Tm and sPFrM stations has six measure- \nment points which can be accessed by the tTsc for automated remote \nmeasurements of the Ld line status.\u00b0 In addition to the measurement \npoints, an input access is also provided in the transmitting line- \nconnecting circuit so that out-of-service frequency characteristic \nmeasurements may be remotely performed by the Tsc on a switching \nsection. This point, equalizing signal in (EQL sic IN), is also used for \nline equalization described in an earlier section.\n\nAnother transmission surveillance system administrative function \nperformed by the line-connecting circuits is that of fault-locating \nsignal blocking. Fault-locating signals at 1.59, 1.60, 68.6, and 68.65 \nMHz must be blocked at each switching section to avoid interference \nwith fault-location routines in other switching sections. A high-pass \nfilter blocks the low-frequency fault-locating tones a minimum of 50 \ndB. An alternate high-pass filter design will also block the 2.976-MHz \nequalizing pilot (Section 2.2) when required.\n\nSince switching (Section 2.3) and fault-locating signals are always \nblocked at each switching station, a single filter accomplishes both \nfunctions. Because of the relative closeness of the 66.048 equalizing \npilot\u2014which is not always blocked concurrently with the switching \nand fault-locating signals\u2014a low-pass filter was not a practicable \ndesign. Instead, a band-elimination filter with a minimum of 50 dB of \nsuppression was developed for this purpose. In those cases where the \n66.048-MHz pilot is also blocked (Section 2.2), a low-pass filter is \nfurnished to block all signals above the message band.\n\nThe smx requires a very stable frequency for the synchronization \nof its carrier supply circuits. This frequency, 20.480 MHz, is generated \nby the Bell System Reference Frequency Standard (BsrFs) located at\n\n\u201cFigure 4 represents a simplified block diagram of the line-connecting circuits for \na SPFM station. The circuitry is essentially similar to the Tm station, with the exception \nof the line-interconnecting circuits, which are much less complicated.\n\na \nRCVG LINE CONN | TRMTG LINE CONN \nREGULAR sae \naptie ov | ae EQUALIZER \nSW SW FLO & an \nng sta o Clear \nSIGNAL \nEQUALIZER Se \nEQL \nSIG IN TRMTG \nLINE \nTST \nTo FROM \nLPSS.3 TSC \n- To \nPIL RESUP TSC \nLess FROM SW \n= | : re} \nLINE CONN FROM GEN i a \nDETR PIL GEN - \nan TSC \nOTHER \nLINES Paha TRMTG \nLINE \nSTANDBY LINES ah \n1 10 o ft \nFROM 42.880 \nRCVG BEF \nEQUALIZER RCVG CVG TRMTG TRMTG \nDIR SW sw ODI \nSW <c | SW t TO \nSei TRMTG \n| EQUALIZER \nTO A RCVG TRMTG sae \nLPSS-3 ! TST t TST PIL rH | a \nDETR GEN 0 \nTO TO (TEMP PIL)\n\nTSC TSC \nFig. 4-Simplified block diagram of spr station line-connecting circuits.\n\nHillsboro, Mo.\u00ae The ssrrs will supply the reference frequency signal \nfor all L5 systems, either directly, as in the case of the initial L5 \nLillyville-Hillsboro route, or indirectly via other long-haul systems \nin subsequent routes.\n\nSince the reference frequency signal is only required on the first \nregular working line and in only one direction, the line-connecting \ncircuit administration of that signal is limited to that line. As in the \ncase of the equalizing pilots, it is desirable to maintain the integrity \nof the reference signal throughout the route without restriction; \nhowever, if jumbogroup 1 or 2 is blocked in the through route, some \nattenuation of the signal occurs that requires that a 20.480-MHz band \nelimination filter be added to the receiving circuit of the through route \nto block the signal sufficiently to prevent its interfering with the \nreference signal reinserted in the transmitting line-connecting circuits \nof that route. In such cases, the 20.480 signal is bypassed around the \nthrough-line-connecting circuit, as described in the next paragraph. \nThe characteristics of the filter are similar to those of the 20.992-MHz \nband-elimination filter described in Section 2.2.\n\nIn those offices requiring a reference signal for mx terminals (via \nthe srs circuitry), for reinsertion in a through route or for insertion \non an Ld branch route, a synchronizing receiving panel is provided. \nThis panel, which accepts the entire line signal from a port in the \nreceiving line-connecting splitting-hybrid circuit (Fig. 2), is furnished \nwith a 20.480-MHz bandpass filter, amplifiers, an adjustable attenu- \nator, and a hybrid tree that provides eight outputs for distribution as \nrequired. The bandpass filter provides a minimum of 80-dB discrimi- \nnation approximately +75 kHz from the reference frequency signal.\n\nThe line-connecting circuits provide access for message service \nrestoration both for L5 restoration and for the restoration of other \nlong-haul systems over standby Ld lines.\n\nAccess to restore a failed L5 facility is provided at each regular \nline at the TERM sw (Fig. 2) in the receiving line-connecting circuit \nand the TRMTG sw in the transmitting line-connecting circuit. These \naccess points connect\u2014via the restoration patch bay\u2014to the standby \nL5 line of another facility or, through idle smx terminals, to spare \nequipment of other long-haul facilities. Another access point is provided \nat the TERM sw to allow monitoring of the failed lines at the restoration \npatch bay. Similar points are provided with the standby line connecting\n\nequipment to allow direct restoration access to other failed L5 facilities \nor other types of long-haul systems through idle ymx equipment.\n\nDuring the early development of the L5 line-connecting circuits, it \nwas recognized that considerable gain would be required at Tm stations \nto offset the loss of the filters required for specific administrative \nfunctions. Since as many as 18 mastergroups may be served on each \nline-connecting circuit, redundancy was provided to assure reliability \nconsistent with the overall L5 system. Although without redundancy \nthe reliability of the amplifiers in isolation was considered adequate, \nthe associated power, fusing, and related cabling associated with the \nactive elements made redundancy necessary.\n\nA beneficial byproduct of the redundancy in line-connecting circuits \nis the ability to provide all optional equipment in a modular package, \nthereby allowing in many cases in-service rearrangements of circuit \nassignments with less danger of service outages.\n\nThe line-interconnecting circuit modules are depicted in Fig. 2 as \nthe A and B Turvu ckt or A and B BRANCH cKT. The A and B paths \nterminate in a coaxial switch that provides the automatic protection \nfor the A or B unit, whichever is serving the line. Jumbogroup pilots \ncontrol the switch logic circuitry, which can be rather complex because \nof the wide number of options available in the line-connecting circuits. \nThree points of detection are required to satisfactorily implement the \nswitching logic: (2) beyond the output of the switch to indicate a \nmajor alarm\u2019 at loss of pilot; (\u00a27) at the output of the alternate leg of \nthe switch to monitor the condition of the idle line-connecting circuit ; \nand (27) at the input of the modules to assure that the jumbogroup \npilots are present at the proper level. Since as many as three sets of \nline-interconnecting modules may be required, i.e., one through circuit \nand two branch circuits, up to nine detector circuits are required as a \nmaximum for monitoring the three jumbogroup pilots at the three \npoints of detection. The simplest case, of course, is when all jumbo- \ngroups are \u201cthrough,\u201d in which case only one jumbogroup pilot need \nbe detected (at the three points) for the full message load.\n\nLine-interconnecting circuits for the standby line are considerably \nsimpler than the regular lines, since no jumbogroup blocking or syn- \nchronization administration is required. All intermediate stations are \ntherefore \u2018\u2018through routes\u201d on the spare line for an entire frogging\u00ae\n\n* All line-connecting alarms and status indications may be connected to remote \nalarm systems.\n\nsection. Although active circuitry is furnished with the standby \nline-connecting circuits, no automatic protection is required, inasmuch \nas service does not normally traverse an office on the standby line. An \nalarm is provided, however, to monitor continuity through the standby \nline-interconnecting module, and a spare module furnished may be \nmanually patched in to replace it.\n\nAs shown in Fig. 2, the pilot generator circuit combines all the pilots \nand maintenance signals so that only one input port is required for \ntheir application to the transmitting line-connecting circuit. Several \noptions within the generator, as noted in Fig. 5, are consistent with the \nvarious options required for line-connecting circuit administration.\n\nThe pilot-generator circuit provides the means for combining the \nswitch signals from the Lpss-3 system, the reference signal from either \nthe local standard or the receiving reference circuit (Section 2.5), and \nthe four locally generated equalizing pilots. The four pilots are derived \nfrom free-running crystal oscillators mounted within the generator\n\npanel, which are designed to maintain a frequency accuracy of approxi- \nmately one part in 10\u00b0 per year under normal office temperature \nvariations. Harmonic distortion is required to be at least \u201440 dB \nrelative to the fundamental.\n\nIn all sprm and T\u00bb stations,* the minimum options required are \nthe 42.880-MHz oscillator and the switch signal combining circuits. \nThe 42.880-MHz signal is required for regular lines to connect to the \npilot resupply switch (Fig. 2), which automatically reinserts the pilot \nshould (7) the pilot be lost in the through path, or (7) the mco switch \nbe operated for equalization or Tsc measurements, thereby sustaining \ntemperature regulation of the line during out-of-service conditions. \nSince the 42.880-MHz pilot for standby lines is always blocked and \nreinserted (Fig. 4), the oscillator is always required with standby \ntransmit-receive bay pilot generator panels.\n\nAnother feature provided for standby lines is a switch for reinserting \nthree of the equalizing pilots. This switch is provided at sprm and T\u2122 \nstations where pilots are not blocked in the through path. Since there \nmay be times when the standby line is being used for protection for \nsustained periods, the stations with through pilots would otherwise \nbe without dynamic equalization; thus, when a receiving line switch \noccurs, three equalizing pilots are automatically reinserted for the \nduration of the switch.\n\nAs discussed in the introduction, the basic jumbogroup trunk bay \ndesign is unique in long-haul system development in that it provides \na direct standard interconnection between the ymx and certain multi- \nplex terminals and systems of multi-mastergroup capacity. Any \ncombination of contiguous mastergroups in the basic jumbogroup \nformat may be interconnected from L4 lines, MMx-2 or LMD terminals, \nradio systems using 3A wire line entrance links, or other mx terminals \nas indicated in Fig. 6a.\n\nActually, a maximum of four inputs per jumbogroup is available \nto the Biat bay, so that a more practical arrangement with contiguous \nmastergroup assignments is shown in Fig. 6b. Through judicious \nplanning of the mastergroup assignments, line engineering personnel \ncan more economically engineer facilities; however, there will still be \ncases where some interconnecting systems will carry identically\n\n*prMm stations are not furnished with pilot generator panels; however, a special \nportable, equalizing-pilot generator unit is available for installation and maintenance \ntests.\n\nFig. 6\u2014(a) Basic jumbogroup trunk function. (b) Example of basic jumbogroup \ntrunk arrangement with four-system input.\n\nnumbered mastergroups, thus requiring multiplex equipment to \nreassign their frequency allocations.\n\nThe ssat bay is required whenever a Mx terminal is furnished, since \nit provides not only the transmitting and receiving trunk circuits for \ninterconnecting the gmx to the various terminal arrangements, but \nalso the 5.888-MHz jumbogroup pilot. The Byet bay provides more \nthan 250 different list structures to accommodate all combinations of \ncontiguous mastergroups to associated systems.\n\nThe transmitting and receiving trunk circuits consist of signal- \nprocessing panels and switch and logic panels. An example of the \ntransmitting trunk circuit (transmitting to smx), shown with a two- \ninput circuit arrangement, is given in Fig. 7a. The signal from mmx, \nradio, tMD, L4, or another jmx terminal is split into two paths to \nidentical, redundant signal processing circuits consisting of filters,* \namplifiers, cable equalizers, and, in the case of L4, de-emphasis. Each \nprocessing circuit connects to a coaxial switch controlled by the logic \ncircuitry in the switch and logic panel. The logic circuitry is activated \nwhen the mastergroup pilot level associated with the highest-numbered \nmastergroup in that particular circuit arrangement changes by a\n\n*The filters are low-pass, high-pass combinations, similar to those used in L4 \nblocking and branching arrangements,! and are required for removing unwanted \nsignals in the various spectra of interconnecting systems.\n\nouTPuT 1 | TO \nMMX, \nFROM TO 3RD AND 4TH RADIO, \nJMX RECEIVING OUTPUT LMD, \nCIRCUITS AS REQUIRED OUTPUT 2 oe\n\npredetermined amount. In addition, the logic circuitry provides local \nand remote alarming.\n\nThe transmitting trunk circuit also contains the 5.888-MHz basic \njumbogroup pilot oscillator. The insertion of the jumbogroup pilot \nat the Bsat bay allows the interbay trunk to the mx bay to be alarmed, \nthereby offering further protection against personnel error or cabling \nproblems.\n\nThe receiving basic jumbogroup trunk circuits are very similar in \ndesign to the transmitting circuits, as shown in Fig. 7b; however, in \naddition, a 5.888-MHz band elimination filter is provided to prevent \ninterference with connecting systems, and pre-emphasis is furnished \nwhen connecting to L4.\n\n10. J. L. Garrison, A. Olsen, Jr., and T. H. Simmonds, Jr., \u201cLS System: Trans- \neat Networks and Magnetic Components,\u201d B.S.T.J., this issue, pp. \n2203-2248.\n\nCopyright \u00a9 1974 American Telephone and Telegraph Company \nTue Beut System TEecuHnicaL JOURNAL \nVol. 53, No. 10, December 1974 \nPrinted in U.S.A.\n\nThe L& Coaxial-Carrier Transmission System equipment was designed \nto provide long-term reliable service in diverse environments. The line re- \npeaters, for instance, are housed in apparatus cases located in manholes \nthat are subjected to flooding, whereas the main-station repeaters and \nterminal multiplexing equipment are located in the controlled environment \nof underground or aboveground buildings, often only partially attended by \nmaintenance personnel. In addition to describing novel physical designs, \nthis paper covers other very important considerations such as thermal \ndesign, manufacturability, incorporation of hybrid integrated-circurt \ntechnology, efficient \u2018\u201c\u2018building-block\u201d\u2019 system growth capability, and long- \nterm reliability.\n\nProposed approaches to the physical design of L5 repeaters covered \na wide array of possibilities ranging from the buried repeater concept \nto that of an ideal conversion. The former would capitalize on the high \nreliability and trouble-free performance of solid-state repeaters to gain \ncost advantages in outside plant for new routes, but would require \ninitially equipping all tubes of a coaxial cable with repeaters. The \nlatter approach would feature repeaters configured to closely resemble \nexisting L4 system repeaters.! It would encourage conversions from \nthe L4 to the L5 system and take advantage of as much existing out- \nside plant as possible. The range of proposals also included such alter- \nnatives as limited-access enclosures, underground lockers, and single- \nrepeater apparatus cases, all of which would have required extensive \noutside-plant apparatus development. \n2147\n\nExperience with the L4 system showed that accessibility to line \nequipment is essential during initial installation of a system. This \nprecluded burying L5. Perhaps, after a shakedown period, buried \nrepeaters can be developed for new-route applications. One-mile \nrepeater spacing and continuing improvements in reliability of solid- \nstate devices may create a wish to bury L5 in the future. In this case, \nthe basic repeater, the least complicated and most numerous of the \nL5 repeaters, is the most likely first candidate.\n\nClearly, the pursuit of a conversion design was the most logical \ninitial approach to the physical design of the L5 repeaters. There will \ncertainly be a continuing need to convert L4 routes to L5 when the \ndemand for circuits exceeds L4 route capacity. Also, the L5 trial in \nmid-1970 was only realized by the use of existing plant designs where \npossible. In addition, by pursuing a conversion design, the development \nof the L5 system did not depend first on the acquisition of knowledge \nof a wholly new repeater environment. The existing L4 environment \nwas usable for evaluation of the L5 repeaters during the development \nperiod. Conversely, it was also understood that pursuing a conversion \ndesign would not preclude the use of newly developed L5 outside plant \nin applications at the 1-mile repeater points where new manholes are \nrequired, and along new routes, if the need for new plant could be \njustified. Thus, the established schedules for development, trial, and \ncommercial service for the L5 system, together with the economic ad- \nvantages of reusing existing outside plant, all favored the conversion- \ndesign approach to L5 physical design.\n\nAn early plan was directed toward ideal conversion\u2014the situation \nin which the L5 repeater shape would correspond closely to that of the \nL4 repeater, allowing easy replacement in the existing L4 apparatus \ncase housing and chassis assembly. However, the proposed L5 system \nfault-location scheme requires 14 copper-wire pairs in the coaxial \ncable, whereas the L4 fault-location scheme requires only 5 wire pairs. \nThis became the controlling factor in causing a change in the L4 ap- \nparatus case chassis. Once it was recognized that a chassis change was \nneeded, the degree of change was academic, and a modified conversion \nplan was adopted. This plan was directed toward developing a repeater \nconfiguration that would still fit the L4 apparatus case, but required \nthat the chassis of the apparatus case be changed to accommodate \nadministration of the L5 fault-location pairs, to promote good heat \ntransfer, and to simplify repeater installation. Since the Ld basic \nrepeater dissipates some 18 watts (compared to 13.5 watts for its L4\n\ncounterpart), it was necessary to design a more efficient thermal \ncircuit in the L5 repeater.\n\nFigure 1 shows the L5 apparatus case chassis, the outside-plant item \nrequired to convert an L4 apparatus case to L5. This is simply a base \nplate holding guide pins that line up plug-in repeaters. The guide pins \nserve also as retaining elements for the repeaters. Thus, the repeaters \nare bolted on the plate by means of the retainer guide pins, eliminating \nthe need for the cantilevered guide and clamp arrangements used in\n\nL4 (sometimes a source of problems because of the degree of precision \nrequired to insure proper mating of the plug-in units). The repeater \nis designed so that, when bolted in place, its leading surface is in \nintimate contact with the apparatus case chassis. The repeater-ap- \nparatus case chassis interface becomes an important link in the thermal \npath from the repeater to the manhole.\n\nFigure 2 illustrates the L5 basic repeater apparatus case arrange- \nment featuring four basic repeaters located on the periphery of the \nchassis. The center space is occupied by the non-heat-producing fault- \nlocation oscillator unit. Coaxial patch cords are used for insertion of \nfault-location oscillator tones at both the input and output of the \nrepeaters via twin coaxial jacks located on the apparatus case chassis. \nThe twin jacks serve also as repeater by-pass points in the high- \nvoltage, series-powered L5 line to permit a power patch around a \nrepeater, without turning power down on a line, in the event repeater \nreplacement becomes necessary.\n\nApparatus cases are mounted in precast concrete manholes (6 by 12 \nby 64 feet high inside the basic repeater manhole) that are not intended \nto be impervious to water. The apparatus cases are watertight and are \nfilled with dry nitrogen to 9-psi cable pressure to protect against water \nseepage. Six apparatus cases are required in a basic repeater manhole \nto fully equip a 22-tube coaxial cable. Adaptors, or \u2018dummy\u2019 ap- \nparatus cases (one of which is shown at the top of Fig. 2), are installed \nto permit an entire manhole to be racked and spliced initially. Ap- \nparatus cases are installed only to cover the number of tubes being \nequipped in order to defer expenditures until necessary.\n\nThe cross-connect apparatus case, an additional watertight ap- \nparatus case that is somewhat different in design, is also located in the \nmanhole. It houses both the logic unit (used to turn on fault-location \noscillators) and the appearances of all the coaxial cable\u2019s copper wire \npairs used for support systems such as fault location, order-wire, alarm, \nand surveillance. A logic-unit support-shelf kit was designed to permit \nconversion of this type of L4 apparatus case for use on Ld.\n\nFigure 3 is an L5 basic repeater with covers removed. Elements of \nthe basic repeater are the fundamental building blocks of the L5 \nsystem and appear in every repeater in the L5 equipment.? These basic \nelements, the preamplifier, the power amplifier, the low-frequency \nnetworks, and the line-build-out (LBO) network, are assembled to an \ninner chassis located inside a rugged die-cast aluminum outer housing. \nFigure 4 is a simplified repeater cross section showing the relative \nphysical positioning of the apparatus on the inner chassis which is \nassembled to a flange protruding from the inner surface of the outer \nhousing. The low-frequency networks protrude into the power separa- \ntion filter (psr) cavity through a cut-out in the flange.\n\nBecause of the series-powering arrangement for the L5 repeatered \nline, repeater circuit ground can be as much as +1150 volts de (1750 \nvolts de under trouble conditions) from earth or sheath ground. In \nFig. 4, the cross section of the inner chassis assembly which is at circuit- \nground potential is shown clear to contrast with the heavy black cross \nsection of the outer housing, which is at earth-ground potential. \nInsulation between these two items is provided by a conformal coating \nof epoxy, 0.015-inch thick, applied to the inside cavities of the outer \nhousing. A thin layer of epoxy sheet adhesive is used to bond the inner \nto the outer. The controlled-thickness epoxy interface provides the \nstructural strength required, preserves the integrity of the insulation \nfor the safety of personnel and the protection of precision, low-voltage \nelectronic components, and constitutes a very low impedance to the \nthermal path.\n\nThe insulating and bonding techniques are the same as those de- \nveloped for the L4 repeaters. As in L4, the bond joint was designed to \ncover a relatively large area for strength and for enhancement of \nthermal conduction, which is further enhanced by the massive cross \nsections of the repeater\u2019s structural elements. The epoxy is applied by \nthe fluidized bed process. The process is carefully controlled to eliminate \nvoids, thereby minimizing the likelihood of high-voltage corona\n\nThe individual apparatus units are contained in die-cast aluminum \ncans fastened to the inner chassis by threaded inserts integral to the \ninner chassis. The repeater covers are coated with epoxy on the inside \nto complete the continuity of the insulation of the two cavities. They \nare secured using one-way screws as a safety feature. In effect, \u2018\u2018sealed\u2019\u2019 \nrepeaters are shipped to the field. There are no field adjustments to \nbe made on a basic repeater, and repairs, if necessary, are made at the \nfactory. Therefore, there is never a reason to open a repeater in the \nfield. They are sealed to avoid the exposure to high voltage that would \noccur if a coverless repeater were accidentally inserted into a powered \napparatus case. Thus, safety features are not compromised by the use \nof the simplified apparatus case chassis design (Fig. 1). High voltage \nappears on the center conductor of the coaxial jacks that are located \ndeep inside the apparatus case, and sealed repeaters keep the high \nvoltage from the reach of craftspeople.\n\nThe preamplifier and power amplifier are the heart of the L5 \nrepeaters. They are thin-film hybrid integrated circuit amplifiers, and \nin the basic repeater they are located side by side in the center part of\n\nthe repeater as shown in Fig. 3. Even though the amplifiers require \nmany discrete components such as transformers, inductors, capacitors, \nand transistors, it was established very early in the L5 development \nthat conventional printed-wiring arrangements could not be used. \nThe precision required for L5 amplifier circuits was such that thin-film \ncircuits were essential to the amplifiers. However, L5 required large \nsubstrates, holes in the substrate to mount discrete components, low- \nvalue precision resistors, and high-conductivity conductor paths. Each \nof these requirements suggested a departure from generally accepted \nthin-film practice. Together, they constitute a unique set of require- \nments and resulted in uic\u2019s that are unique in their physical and \nelectrical attributes.\n\nAlthough most uic\u2019s are small enough to fit many patterns on a \nsingle 32 by 43-inch ceramic mirror, the L5 uic\u2019s are large enough \n(23 by 33 inches) to allow only one per mirror. Conventional, leaded \ncomponents are almost never mounted on a ceramic substrate, but \nthis was required here. This gave rise to the need for about 50 holes \nper substrate. Holes complicate the manufacture of. substrates and are \navoided, if possible. To produce the low-value precision resistors, the \nresistive metal layer was deposited to a thickness of 25 ohms per square, \nas opposed to the conventional thickness of 50 ohms per square. This \nallowed longer resistors and a larger proportion of anodized-to-un- \nanodized area. The higher percent of the area anodized results in \ngreater long-term stability. Furthermore, thicker metal film is more \nstable because resistance changes are caused by surface changes, and \nthe surface constitutes a smaller portion of a thicker metal film than \nof a thinner metal film. An operating temperature of less than 90\u00b0C \nis necessary to maintain end-of-life tolerance for the precision resistors. \nPower resistors, on the other hand, would stay within their required \nresistance tolerance at film temperatures as high as 110\u00b0C. The power \nresistors and the discrete transistors are virtually the only heat pro- \nducers on the substrate. Therefore, the layouts were designed so that \nthe power resistors skirted the periphery of the substrate. The transis- \ntors, however, had to be centrally located because of circuit require- \nments for short connecting conductor lengths. Massive heat sinks were \ndesigned to divert the transistor heat from the precision resistors and \nother temperature-sensitive components.\n\nFigure 5 is the L5 power amplifier. The top view shows the two heat \nsinks, each housing a matched pair of transistors, surrounded by other \ndiscrete components. The bottom view, the thin-film side of the hybrid \nintegrated-circuit amplifier, illustrates the film topology and a number\n\nof reflow-soldered barium titanate chip capacitors. A special conductor \nmetal system was developed for L5 to produce the needed high-con- \nductivity paths and to provide solderable pads for the chip capacitors \nand the leaded conventional components. The high conductivity is \nobtained by depositing \u2018\u2018thick\u201d gold. Electroplated rhodium is added \nto prevent the gold from dissolving in the solder. A gold flash over the \nrhodium protects the rhodium from oxidation, thereby retaining its \nsolderability. By selectively etching the gold to expose a rhodium strip \naround the solder pads, solder \u2018\u201c\u2018dams\u2019\u201d\u2019 are formed after the exposed \nrhodium strip becomes oxidized and nonsolderable when exposed to \nthe elevated temperature of the film pre-aging process. Therefore, \nduring the solder reflow process, when the chip components are con- \nnected into the circuit, the exposed rhodium strips confine the solder\n\nto the area they surround. This avoids wetting and possibly degrading \nthe high conductivity paths. By precisely defining the solder area, \nthereby limiting the amount of gold that can dissolve in the solder, \nembrittlement of the solder joints is avoided. Complementary develop- \nments determined the correct amount of solder to be supplied as an \nintegral part of the chip component terminations and the precise \nreflow soldering cycle, thereby assuring reproducible, reliable Hic \nassemblies.\n\nIt was necessary to develop a structural support for the large, thin \nceramic substrates that would permit efficient thermal conduction \nfrom the transistors, but that would keep the substrate free of me- \nchanical stress. The substrates are not perfectly flat, and the amount \nof bowing or warping is limited by die-screening subsequent to firing. \nIdeally, the substrate would be fixed at three points (to define a plane) \nand contact to the transistor would be by a flexible heat conductor. \nHowever, the requirements are inconsistent for providing, simulta- \nneously, a massive heat sink for adequate heat dissipation and a flexible \nheat sink for stress-free mounting. Consequently, the structural support \ndeveloped was to hard-mount the substrate on the massive heat sink \nand merely stabilize its outer periphery on three bosses of the die-cast \naluminum amplifier housing. The location of the three bosses and the \nstabilizing screws is shown in Fig. 5, and a partial cross section of an \namplifier assembled in a repeater is shown in Fig. 6.\n\nIn Fig. 6 the large substrate, in cross section, appears virtually as a \nbeam with a fixed support in the center. The end supports (only one \nshown) are designated \u2018\u2018grommet suspension.\u201d At these points, the \nsubstrate is held between two soft, rubber grommets confined in a large \nhole in the substrate by a shoulder screw anchored to the boss in the \namplifier housing. The role of the grommets is to stabilize the nonflat \nsubstrate in whatever position it may happen to fall and have it \nremain there, stress-free.\n\nThe substrate\u2019s fixed support is the massive transistor heat sink, \nalso shown in cross section in Fig. 6. The discrete transistor, located \ninside the heat sink, has a beryllium-oxide header hermetically sealed \nto a Kovar* case. Heat sinking is to the beryllium oxide, which has the \nhigher coefficient of thermal conductivity, and thermal flow through \nthe transistor-heat-sink interface is enhanced by the force exerted by\n\nthe preloaded spring confined within the heat sink. While high, com- \npressive pressure is applied through the transistor to the heat sink, \nnone of it is transmitted to the delicate thin-film substrate to which \nthe transistor is soldered. The result is a stress-free structural support \nfor the large L5 substrates and an effective thermal circuit. As shown \nby the arrows in Fig. 6, heat is conducted from the transistor through \nthe heat sink, through the amplifier cover into the inner repeater \nchassis, and through the thin layer of epoxy to the outer repeater \nhousing.\n\nClose spacing of the transistors on the substrate was required to \nminimize parasitic inductance and capacitance in the electrical circuit \nand resulted in a unique heat sink design. Spring-loading the transistor \nwithin the confines of a massive heat sink was developed for use in the \nL4 system amplifiers. However, the L4 scheme was not applicable to \nL5 because of the combination of space limitations and the application \nof the heat sinks to delicate, large, ceramic substrates in L5, in contrast \nto the rugged epoxy-glass printed wiring boards used in L4. The L5 \nheat sink is a precision assembly that, for good heat transfer, presents \nthe largest possible metal cross-sectional area within the allowable \nlimits for the transistor lead lengths and transistor spacing. The screw \nthat preloads the spring against the transistor serves as the threaded \nnut for mounting the heat sink against the flat amplifier cover. To \nminimize thermal impedance, the heat sink\u2019s functional faces have a\n\n32 micro-inch finish, and an indium washer is inserted at the transistor- \nheat-sink interface. This assures intimate contact at the thermal \ncircuit joints.\n\nThe heat sink\u2019s cavity, large enough to accept the transistor with its \nflanged case, is constricted at the threaded portion because of the \ntransistor spacing constraints. This constriction makes it necessary to \nhave a side opening in the heat sink to load the transistor and precludes \nthe use of conventional screw-machine methods of manufacture. \nDouble heat sink designs are used in the preamplifier and power ampli- \nfier to maximize the amount of metal usable for heat sinking within the \ntransistor spacing constraints. This works out well, because the transis- \ntors are used in matched pairs and can be preassembled into the heat \nsinks for subsequent assembly onto the substrates.\n\nThe thin-film substrates are housed in die-cast aluminum cans. \nThese cans require a certain degree of precision to provide accurate \nstress-free positioning of the substrates, as shown in Fig. 6. Substrate \nsupporting surfaces are held within tight tolerances with respect to \nthe cover supporting surface. The bottom amplifier cover is controlled \nfor flatness and is designed to be overflush to present maximum contact \narea to the repeater inner chassis for heat transfer.\n\nHaving top and bottom amplifier covers removable provides ac- \ncessibility to both sides of the substrate during shop processing and also \nfor trouble shooting, and the open die-cast frame of the amplifier \nhousing serves as a convenient carrier and substrate protector during \nhandling in the shop.\n\nConnections to the substrate are made through 75-ohm miniature \ncoaxial connectors that clearly define electrical and mechanical \nboundaries for the amplifiers. This substantially facilitates amplifier \ntesting, repeater assembling, and, if needed, repair and replacement.\n\nTo ensure reliable performance of silicon solid-state devices, the \nphysical design goal was to limit the maximum junction tempera- \nture to 125\u00b0C and, in the case of the elements of the L5 repeatered \nline, can be expressed as\n\nTz = transistor junction temperature, \nATac = temperature rise in the repeater apparatus case, \nATr = temperature rise in the repeater, \nAT, = temperature rise from case to junction of the transistor \nbecause of its thermal impedance, \nTy = maximum manhole ambient temperature.\n\nThe maximum junction temperature for transistors in the L5 basic \nrepeater power amplifier has been determined by tests to be 128\u00b0C, \nand comparative performance of L4 and Ld is shown in Fig. 7. Re- \ncalling that an L5 repeater dissipates some 18 watts in comparison to \n133 watts for L4, it is evident that the thermal conductivity enhance- \nment was realized as anticipated because of the apparatus case chassis \ndesign and the repeater packaging techniques developed for L5. At \n128\u00b0C, maximum transistor junction temperature for L5 transistors, \nthermal results of L4 and L5 are equivalent and, based on L4 per- \nformance to date in the field, it appears that high-reliability per- \nformance of L5 devices is ensured.\n\nThe \u2018\u2018modified-conversion\u201d design concept as applied to the L5 \nregulating repeater requires that two repeaters fit in one apparatus \ncase. The two repeaters are bolted to a newly designed apparatus case \nchassis, similar to that for the basic repeater illustrated in Fig. 1, ina \nlayout that positions the fault-location oscillator unit between the two \nrepeaters. A maximum of 11 apparatus cases are required to accom- \nmodate an L5 22-tube coaxial cable, and these cases are arranged on \none wall of a precast concrete manhole 6 by 12 by 9 feet high on the \ninside, 23 feet higher than a basic repeater manhole.\n\nThe regulating repeater package is one of the most difficult of the \nL5 physical designs. A number of design constraints were imposed, \nsuch as field access for adjustments, addition of constituent apparatus \nand in-service testing, heat dissipation, high-voltage insulation, and \npersonnel safety.\n\nThis repeater contains the same circuitry as a basic repeater with \nthe addition of regulating circuitry, space for field insertion of two \nline-build-out networks and a deviation equalizer, and a low-voltage \ntest point for in-service measurements. Like the basic repeater, a \nrugged, epoxy-insulated outer aluminum die-cast housing holds the \ncaptive retainer screws for bolting the repeater to the apparatus case \nchassis. Unlike the basic repeater which has a flat-plate inner chassis, \nthe inner chassis of the regulating repeater is compartmented to accept \nindividually packaged circuit sections and units that do not re- \nquire their own individual shield cans (see Fig. 8). Because of the \npackaging density required to have two repeaters cantilevered from \nthe apparatus case chassis in a single apparatus case, the center flange \nof the outer housing had to be kept to a minimum, allowing just \nsufficient area to ensure a good bond to the inner chassis. Therefore, \ncompartmenting, in combination with individual apparatus housings \nthat act as stiffeners, produces an array of ribbing on both sides of the \nlarge inner chassis and provides the rigidity required over the large \ncross-sectional area of this repeater, which measures 103 inches wide \nby 173 inches long.\n\nThe regulating repeater uses five thin-film hybrid-integrated circuit \n(gic) amplifiers in addition to the preamplifier and power amplifier. \nThe five are flat-gain amplifiers, and their ceramic substrate size is \nmore nearly conventional, measuring 0.8 inch by 2.9 inches. Actually, \nthree designs, two of which appear twice as part of the two Vf net-\n\nworks,\u2019 are used in the regulating repeater. In the Vf network, the \nHIc\u2019s are connected as slave-boards to epoxy-glass printed-wiring \nnetwork boards by means of 26-gauge leads on the uic\u2019s. They are \nformed to fit into holes in the printed-wiring board and are clinched to \ntouch land areas to which they will be subsequently soldered. The \nsubstrates are structurally supported on transistor heat sinks that \nenclose the discrete, heat-producing transistors and are fastened \ndirectly to the can that houses the entire network. The heat sinks are \nsimilar to those described for the preamplifier and power amplifier, ex- \ncept that they are the single-cavity style. Here, too, conduction is the \nprimary mode of heat transfer. The heat from the transistors is \ndirected away from the substrate, through the heat sink into the \ndie-cast network housing and to the repeater framework, etc. Top \nand bottom network covers are removable to provide accessibility\n\nfor trouble shooting but, more important, to allow assembly in a \nprescribed sequence. The printed-wiring board is first mounted in its \nframe, then the uic is fastened to the frame by the heat sink and finally \nthe nic leads are soldered to the board Jand areas. This assembly \nsequence is essential to a stress-free assembly. Figure 9 shows the \ncollection of thin-film integrated-circuit substrates used throughout \nL5. The 55-types and the 57-types are the ones described earlier for \nuse in the repeaters, while the remaining ones are used in flat-gain \namplifiers for terminal equipment described later in this paper.\n\nTo compensate in part for changes in cable loss because of changes \nin cable temperature,? the regulating repeater requires access to \nground-temperature-sensing thermistors that are buried 15 feet from \nthe manhole at cable depth. The thermistors are contained in a cast- \nepoxy cable stub and, at a distance of 15 feet, are unaffected by man- \nhole temperature. A plug-and-jack connection through the apparatus \ncase chassis provides the thermistor connection to the repeater.\n\nWhile the basic repeater is shipped to the field as a sealed unit, the \ninside of the regulating repeater must be accessible for the addition of \nline-build-out networks and for adjustments when the repeater is ac- \nceptance-tested in the field.? For this reason, the regulating repeater\u2019s \ntwo large covers are hinged as a safety feature to prevent exposure to \nhigh voltage by the accidental insertion of a coverless repeater into a \nlive apparatus case. Manipulation of the bulky, hinged covers is \nawkward, at best, during manufacture and field-acceptance testing, \nbut amounts to a relatively effective trade-off for a necessary safety \nprecaution.\n\nHigh-voltage insulation lines the cavity on the inner chassis that \nhouses the coaxial jack used as the low-voltage access point for the \nrepeater. Even though the jack is electrically isolated from high voltage \nby a transformer and is mechanically bonded to the outer housing \nwhich is at earth or sheath ground, the insulation liner is provided as \na safety feature to protect personnel and equipment in the event a \nfault should occur.\n\nWith equalizing repeater points located at 34-mile (maximum) \nintervals, or roughly midway in the 75-mile (maximum) L5 power \nspans, it is unlikely that, in conversion from L4 to Ld, they would \ncoincide with L4 equalizing repeater points spaced at 54-mile (maxi- \nmum) intervals.* Therefore, the only constraint on the L5 equalizing \nrepeater, to comply with the modified-conversion design concept for L5, \nis that similar apparatus cases be used for the sake of standardization.\n\nHowever, L5 uses manually adjustable equalizers with less circuitry \nthan is required for the remotely adjustable equalizer used in L4. This \nsimplifies manhole arrangements in that the entire L5 equalizing \nrepeater was designed to fit in a single apparatus case, in contrast \nwith the two cases required to house the L4 equalizing repeater. For \na 22-tube coaxial cable, an equalizing repeater pre-cast manhole \nmeasures 6 feet wide by 24 feet long by 8 feet high on the inside to \naccommodate the layout for the 22 apparatus cases required to equip \nthe entire cable.\n\nThe equalizing repeater comprises three plug-in units\u2014a regulating \nrepeater, a manhole E1 equalizer (mE1), and a fault-location oscillator \nunit arranged in that order from left to right in the apparatus case. \nAs with the previous repeaters, these units are bolted to the apparatus\n\ncase chassis and are cantilevered inside the apparatus case. Equalizing \nrepeater points also provide access to buried-cable-temperature- \nsensing thermistors.\n\nThe mel equalizer is a larger unit than the regulating repeater, and \nwas designed as a fabricated unit to keep its weight down. The unit \nmeasures 12 inches wide by 17 inches long by 44 inches deep, and is \nillustrated in Fig. 10. The array of individually packaged amplifiers \nand Bode networks is visible. These use printed-wiring technology, \nare interconnected by cables terminated in miniature coaxial connec- \ntors, and are made accessible to the knobs of the mel faceplate by \ninsulated shafts that permit manual control of potentiometers for \nequalizer adjustments. The knobs are \u2018\u2018push-to-turn\u2019\u201d\u2019 to prevent \ninadvertent changes in adjustments, and universal couplings prevent \nlateral forces from being transmitted to the delicate potentiometer \nshafts. A power separation filter is contained in the lower cavity of the \nMEI equalizer, and high-voltage insulation is provided by phenolic \nstrips. This unit, like the basic repeater, need not be opened in the\n\nfield and, for safety, it is sealed by having its covers secured with one- \nway screws. The mounting base of the mE] is a one-piece aluminum \ncasting that permits the precise location of the mounting holes and \nthe coaxial plugs within the tolerances allowable for mating with the \napparatus case chassis.\n\nMain stations in an L5 \u201cbackbone\u201d route are single-story buildings \ncontaining four general categories of main-station equipment\u2014high- \nfrequency repeatered line, signal processing terminal, transmission \nsurveillance, and a fourth, broad or peripheral category comprising \nsuch equipment as order wire, restoration, repeater acceptance test, \nand equalizer adjustment. The main-station equipment, intended for \nthe central office environment, is bay-mounted in contrast to the \nrugged, cast plug-in type equipment designed for the manhole en- \nvironment. The main-station equipment is generally in shelves, \ndrawers, or panels in unequal-flange duct-type bays, on a 2-inch \nmounting plate modular spacing, and forming a bay-front that is flush \nwith the bay uprights. The equipment space is 15 inches deep, the \nfront and rear guard rails are 2 inches and 3? inch, respectively, and the \nduct formed by the uprights of two adjacent bays constitutes shielded \ncabling space closed at the front and accessible from the wiring aisle \nat the rear of the bays through the opening created by the smaller \nflange (see Fig. 11). The space directly behind the cabling duct is \navailable for shelf interconnections, for overflow cabling if required, \nand, in the case of a double bay, for additional equipment.\n\nThe deep equipment space is required because of the relatively \nbulky, well-shielded apparatus assemblies used throughout the system.\n\nThis depth, in combination with the width of the standard bay, gives \nrise to three-dimensional, high-density, equipment packaging which \nsaves coveted floor space in a main station but limits accessibility of \nindividual constituents of bay subassemblies. However, the limited \naccessibility is tempered by the connectorized assemblies used through- \nout the system to facilitate manufacture and maintenance.\n\nMiniature coaxial connectors terminate individual pieces of ap- \nparatus (amplifiers, filters, networks, etc.) to form precise, definable \nboundaries for specific electrical requirements. These individual units \nare arranged on a shelf and interconnected by miniature coaxial cables \nterminated in the mating coaxial connectors. The shelf is screw- \nfastened to the rear flange of the bay uprights, and is itself inter- \nconnected with small Bell System coaxial plugs and jacks to the bay \nwiring harness for signal connections and with multicontact connectors \nfor power and alarm connections.\n\nIn some instances, plug-in modular printed wiring board (pws) \nassemblies are arranged on a shelf and interconnected by gold-plated \nfingers on the pwB and multicontact connectors that are part of the \nback-plane wiring of the shelf.\n\nHigh-density packaging tends to accentuate thermal considerations, \nand limited accessibility coupled with high system capacity accentuate \nreliability considerations, both prime considerations in the design of \nL5 equipment.\n\nThe L5 bay height was standardized at 9 feet to accommodate \npower-feed main and switching power-feed main-station buildings. \nTerminal and terminal-main stations have higher ceilings and require \nbays 10 feet, 6 inches and 11 feet, 6 inches high. For these, bay ex- \ntenders are added to have the 9-foot standard bay fit in 10-foot, 6-inch \nand 11-foot, 6-inch bay line-ups. Specific bays, intended for use only in \nterminal and terminal-main stations and requiring the additional space \nafforded by the higher ceilings, are 10-foot, 6-inch and 11-foot, 6-inch \ndesigns. The new equipment building system (NEBS) requires that \nfuture equipment be standardized on a 7-foot bay height, and a series \nof 7-foot bays have been designed for L5.\n\nFor the L5 high-frequency repeatered line, the interface between \nmanhole repeaters and main-station equipment\u201d is the power-separa- \ntion filter (esr) cabinet (see Fig. 12). This cabinet confines all the\n\ntransmission area high-voltage equipment in one locked enclosure for \nsafety, and provides a single point to terminate the lead-covered, \npressurized, air-dielectric cable stubs coming from the multitube, \ncoaxial-cable splice in the station. This obviates the need to run the \nlead cable and the high voltage in front of the line bays, as has been \ndone in the past. The run from the psF cabinet to the line bays is at \nlow voltage and uses single-unit air-dielectric and/or solid-dielectric \ncoaxial cable, depending on the length of the connection. Initially, it \nwas intended that the psr cabinet be located in close proximity to the \nbuilding cable entrance to minimize the run of lead cable in the \nstation. However, placing the psF cabinet at the recommended location \nat the head-end of an L5 bay line-up optimizes the interbay cabling \nfor that line-up and, in most cases, precludes the need for the air- \ndielectric coaxial-cable-run portion of the connection from psF cabinet \nto the line bays.\n\nThe 9-foot-high esr cabinet can hold a maximum of 22 power \nseparation filter shelves required to equip an entire 22-tube cable. The \nshelves are contained in the lower portion of the cabinet seen from the \nrear in Fig. 12. The cable terminals are bolted to brackets in the top \nportion. Connections are made by screw-type coaxial connectors for \nthe high-voltage runs and by plug-and-jack connections for the low- \nvoltage runs. The cabling is dressed in the ducts formed on the sides of \nthe cabinet by the standard bay uprights (used to structure the \ncabinet) and the fabricated metal enclosure is assembled around the \nbay. Bay extenders accommodate the 9-foot cabinet in higher than \n9-foot equipment line-ups. The psr shelves are modular and can be \nequipped on an as-needed basis.\n\nThe line-protection switching system (LPss-3)*\u00b0 bay is the second \nelement in an Ld line-up, and is required at each end of an L5 switching \nsection (150 miles maximum), but not at the power-feed main stations \nlocated at the 75-mile (maximum) points. This 9-foot bay contains a \npreponderance of logic-type circuitry that is packaged in modular \nfashion on plug-in PwB assemblies arranged on fixed equipment shelves \nin the bay. The myriad interconnecting wires are formed into a sizable \ncable harness, dressed in the cable duct, and connected to a terminal \nstrip at the top of the bay. Bay power is derived from plug-in dc-to-de \nconverters located and fused at the top of the bay. Status indications \nand controls are concentrated in a large display panel located centrally \non the bay. Human-engineering design considerations predominantly\n\ninfluenced both the physical arrangement of this panel and the selec- \ntion of the lighted indicators and keys so that the system\u2019s switching \nstate can be clearly communicated to the craftsperson. The presence \nof plastic hinged protective covers on the more important controls \nalso provide the \u2018\u2018keying\u2019\u2019 necessary to avoid inadvertent manipulation \nof controls so as not to jeopardize service. Wire verification to ensure \nthe presence of all intrabay connections is required before shipment \nof a bay containing the common equipment. Modular plug-in assemblies \ncan be shipped separately on an as-required basis as the system grows.\n\nThe line transmit-receive bays are the mainstay of the Ld line-up. \nOne line bay is required for each transmit-receive pair of coaxials \nequipped in the system. This bay is essentially an equalizing repeater \nwith manually adjustable equalizer units, El and E2, in both the \ntransmit and receive sides and with the addition of a dynamic equalizer, \nE38, in the receive side. This, together with the elements of line-con- \nnecting, powering, and alarm equipment, comprises the line transmit- \nreceive bay which appears at every main-station location. Four designs \naccommodate the 9-foot requirements at power-feed and switching \npower-feed main stations and the 10-foot 6-inch and 11-foot, 6-inch \nrequirements at terminal and terminal-main stations. The difference \namong the four designs is the varying complexity of the elements of \nline connecting, ranging from the very simplest in the power-feed \nmain-station design, where the transmit-receive bay is essentially a \none-way repeater, to the more complex one-way repeater of the \nswitching power-feed main-station design, which includes switches and \nswitch-initiator circuitry for Less, to the most complex terminal and \nterminal-main station designs where elements of line connecting also \ninclude blocking, adding, dropping, and branching circuitry for the \nsignal-processing features. This hierarchy of line bay designs is il- \nlustrated in Fig. 18, with shading used to contrast the variable element \nwithin the otherwise repetitive structures. Note that the only difference \nbetween the two \u2018\u201c\u2018terminal or terminal-main\u201d\u2019 designs is that the 11- \nfoot, 6-inch bay has room for one additional line-branching unit.\n\nThe main-station transmitting and receiving repeaters together \nequate to a regulating repeater used in manholes. The transmit and \nreceive elements have been split and repackaged in an equipment \nshelf for bay-mounting (Fig. 14). Similarly, the main-station manually \nadjustable equalizer (E1) contains the same elements as its plug-in \nunit, manhole counterpart (Fig. 10), re-packaged in a fixed shelf for\n\nGASES he PILOT GENERATOR \nLD TRMTG LINE CONN \nY, LINE INTERCONN \nDETECTOR \nUi} \nYG a \nZ Y Y 3 CEIVING \nLZ, LINE CONNECT \nWEESELGZ \nFAULT LOC &\n\nbay-mounting (see Fig. 15); and this unit, expanded to include addi- \ntional networks and amplifiers required for additional bump shapes, \nyields the design of E2 (see Fig. 16).\n\nTwo line bays are shown at the right in Fig. 17 as they appear at a \nterminal or terminal-main station in an L5 five-bay line-up that con- \nstitutes the equipment required to provide service on the first working\n\npair of tubes, with the first line bay being the protection bay and the \nsecond the first regular bay.\n\nThe final step of multiplexing required to stack jumbogroup (sq) \nsignals for transmission over the L5 coaxial line is accomplished by \nthe jumbogroup multiplex\u00ae (sax). The smx is a completely solid-state \nmultiplex that utilizes thin-film hybrid-integrated circuit (Ic) ele- \nments contained in modular plug-in assemblies. The smx equipment \nis mounted on a unitized bay framework shown in Fig. 18, 11 feet, \n6 inches high by 52 inches wide by 15 inches deep. A complete bay \naccommodates a maximum of three transmitting and receiving jumbo- \ngroups. Jumbogroup multiplex designs are also available in 10-foot \n6-inch, 9-foot, and 7-foot bay heights to meet the needs of telephone \noffices with lower ceilings.\n\nThe smx is a completely shop-assembled, shop-wired, and shop- \ntested bay. It contains transmission, shaping, regulating, patching, \nlogic, alarm, carrier supply, and dc-to-de converter equipment and, \nsince 3600 voice circuits could be affected by a service interruption, \nautomatic protection switching is also provided for each jumbogroup.\n\nJumbogroup multiplex bays are generally located in the maintenance \naisle of a central office or main station where a minimum aisle spacing\n\nof 4 feet, 6 inches is required to permit the use of rolling test consoles \nwithout restricting the operating personnel. Since this area is a center \nof office activity, a good appearance, without significantly increasing \nthe cost, was considered a design objective.\n\nThe bay cabling consisting mainly of office-type coaxial cable is \ncarefully segregated, primarily to minimize crosstalk and secondarily \nto facilitate installation and shop wiring. Transmitting cables are con- \ntained in the left-hand cable duct and receiving in the right-hand duct,\n\nand, with few exceptions, all other wiring and cabling is confined to the \ncenter duct. Office cable connections to the bay are made through \nconnectorized cables located at the top of the bay, making it unneces- \nsary for the installer to enter the cable ducts.\n\nThe basic building block of the gmx bay is the jumbogroup shelf \nassembly shown in Fig. 19, which contains the modular plug-in units. \nFor example, the receiving shelf contains the regulator, equalizer, \ndemodulator, de-emphasis, transmission protection switch, logic \nswitch control, and jack field. The framework and module construc- \ntion for the transmitting side is the same.\n\nSince all equipment modules performing similar functions are the \nsame physical size, a variety of bay configurations can easily be ac- \ncommodated. For instance, at a typical end office, the gmx bay with \njumbogroups 1, 2, and 3 could be provided, whereas at intermediate \nstations the bay could contain all jumbogroups 1, 2, or 3 or any com- \nbination of jumbogroups 1, 2, or 3 up to the bay capacity of three.\n\nA lattice-type lightweight aluminum construction is used for the \njumbogroup shelf to optimize the strength-to-weight ratio and vir- \ntually eliminate shelf deflection under load. To avoid scraping when \nthe aluminum modules are inserted in or removed from the shelf, a \nlow-friction tape is bonded to the load-bearing surface.\n\nEach transmitting and receiving shelf assembly contains a built-in \njack field for jumbogroup patching or testing. To facilitate main- \ntenance, a functional schematic diagram is provided directly above the \nactual jack location. The jack field is recessed to minimize the pos- \nsibility of a circuit patch plug accidentally being removed or damaged, \nthereby causing a service interruption. To aid the craftsperson in \nidentifying jumbogroup equipment with its associated jack field, each \njumbogroup position is framed with a colored plastic strip. Each of the \nthree jumbogroup bay positions is identified with a different color. \nMatching color route assignment cards covered with glare-resistant \nplastic are also provided.\n\nDirectly above the writing shelves is a miscellaneous panel that \nprovides access to office trunks, alarm cutoff switches, and indicators \nand the means for simultaneously testing all lamps in the bay. In the \npresent bay, incandescent lamps are used for alarm indications; how- \never, in the very near future, these will be replaced by light-emitting \ndiodes (LED).\n\nModules are inserted from the front of the bay and held by a captive \nscrew in the rear of the shelf. These units are essentially plug-in, with- \nout fixed mating connectors on the shelves. In this way, tolerance \nproblems associated with mating connectors are avoided. Most as- \nsemblies are contained in 10-inch wide by 4-inch high by 12-inch deep \naluminum modules, as shown in Fig. 20. Coaxial and pin connectors \nare provided at the rear for transmission and power connections.\n\nFilters, amplifiers, and other types of apparatus have been mounted \nin separate housings for shielding, to facilitate manufacture and test- \ning, and for efficient field repair. Interconnections within the modules \nare made by miniature coaxial cables.\n\nDuring the development of the smx and other L5 main-station \nequipment, it became apparent that there was a need for a family of \nvery small amplifiers. To meet this need, the 509-, 510-, 511-, and \n512-series of HIc amplifier codes were developed. More than 230 of \nthese amplifiers are in a fully equipped ymx bay. The outside dimen- \nsions of the amplifiers are 2.75 by 1.5 by 2 inches high. Input and out- \nput connections are made through miniature 75-ohm coaxial connec- \ntors. The circuitry of the amplifiers is a HIc with some components\n\nappliqued, such as transistors, inductors, or chip-capacitors. Special \nground terminals are swaged into the chassis to ensure effective \ngrounding.\n\nConsiderable care was taken to limit the mechanical stresses be- \ntween the ceramic substrate and the chassis. For example, all connec- \ntions are made with soft gold-coated copper ribbon wire, looped to \nallow slight mechanical movement of the substrate during thermal and \nmechanical shock. As shown in Fig. 21, the substrate is mounted by \nfitting the ends with silicone rubber gaskets and supporting them at \nopposite ends of the chassis with adjustable aluminum blocks. To \nefficiently conduct heat from the transistor, a spring-loaded heat sink \ndescribed earlier connects the transistor mounted on the substrate \ndirectly to the cover of the amplifier.\n\nAs part of the overall smx development, a complete thermal analysis \nof the bay was performed. Where possible, units dissipating the most \npower were arranged in the bay to avoid hot spots. A 4-inch space is \nprovided for each jumbogroup shelf through the recessed jack field for \na front-to-rear air flow. Temperature measurements indicate a 55\u00b0F \nmaximum gradient between a transistor case in the smx and room \ntemperature. The complete thermal analysis shows that, even under \nthe worst condition, the equipment in the bay is well within the \ntemperature limits necessary for reliable performance.\n\nThe 24-volt power for the smx is supplied by the office power plant \nover two separate feeders to the 24-volt to 25-volt regulated converters.\n\nThe feeder-converter assignments have been arranged so that the loss \nof a feeder or a single converter will not result in a loss of service.\n\nThe interface between smx and the lesser units in the multiplex \nhierarchy is the basic jumbogroup trunk bay\u2019 (BseT). This bay offers \na very high degree of flexibility for the user in that it is designed to \nallow a maximum of four inputs per jumbogroup and to include designs \nfor inputting radio signals, L-carrier mastergroup digital (LMD) signals, \nL4 coaxial-carrier system signals, etc. The active circuits are re- \ndundant and are packaged in pull-out drawers and fixed shelves. \nThe combinations of circuits available are so numerous that normal \nbay coding or even normal panel coding is virtually impossible. \nOnly typical bay layouts are offered as a guide for ordering, and the \nspecific short- and long-range requirements of an office dictate the \ncomposition of any BseT shipped from the factory. Intra-bay cabling \nis laid out to accommodate multi-input options and suggested bay \nlayouts allow space for modular future growth.\n\nThe jumbogroup frequency supply (srs) furnishes three highly \nreliable and precise signals to the ymx to generate the necessary car- \nriers.\u2019 These signals, at frequencies of 1.024, 2.560, and 20.48 MHz, \nhave an accuracy well within the one part in 108 requirement imposed \nby the sax. This accuracy is realized by comparison with a reference \nsignal from a primary frequency source; however, even if the reference \nsignal is lost for several weeks, the inherent stability of the srs will \nmaintain the required accuracy.\n\nThe srs, shown in Fig. 22, is mounted in a unitized double 9-foot \nhigh framework. A front aisle is required for normal maintenance and \na rear aisle for installation wiring. Since the number of srs bays re- \nquired for coaxial systems is relatively low, the 9-foot bay arrange- \nment is provided for use in current design of main stations. A 7-foot \nversion of JFs is also available for future offices with the lower ceiling \nheight.\n\nThe srs is a fully shop-assembled, shop-wired, and shop-tested bay. \nIt also is completely solid-state, utilizing digital and thin-film hybrid- \nintegrated circuit (Hic) technology. The most important design ob- \njectives were precision and high reliability. To accomplish this, very \nstable oven-controlled 39-type oscillators and one-for-one redundancy \nautomatically switched are employed.\n\nThe jumbogroup frequency generator (sre) is the basic building \nblock of the srs bay, and the 39-type oscillators are the heart of the \nJrG. Each srs contains three sra\u2019s (master, regular, and standby). \nThe master ura serves as a buffer between the reference signal and \nboth the regular and standby sre\u2019s, thereby keeping the srs output \nrelatively immune from loss and/or disturbance of the reference signal. \nAs shown in Fig. 23, an integral part of the sre is the alarm display \npanel, which indicates whether the sre is operating as the master, \nin-service, or idle. It also indicates frequency offset, level failure, and \nthe frequency alarms. Test access points are provided for measuring \nthe 25-volt, 6-volt, and 5-volt power.\n\nThe sre patch panel provides patching facilities not only to remove \na failed sre from service but to rearrange the sra@\u2019s by function (master, \nregular, standby) to realize the best use of the remaining two. When a \nJFG is removed from service, the protection switching logic must be \nrestructured to accommodate rearrangement of jre@ functions. This is \naccomplished by operating the sre \u201c\u2018out-of-service\u201d switch on the alarm \nand switch control panel when the patching has been completed. The \nfaceplate of the ura patch panel is a schematic block diagram with\n\nappropriately positioned jacks interconnected with color-coded paths \nto indicate the placement of patch plugs for the src, master, regular, - \nor standby failure conditions. As in the mx, the panel is recessed to \nprevent damage and possible circuit interruption.\n\nThe control center for the srs is the alarm and switch control panel. \nIt is a visual continuation of the patch panel schematic block diagram \nand contains the level detectors and coaxial switch and the logic \nrequired for the alarm and switch functions.\n\na ps \n\u201cCATION! {ED OEE ; CAUTION \nSo CRPRC AL oo PS dnt ep Bm penises wt : CRITICAL \nSWITCHES 6\u00b0 e HITCHES \neet ad Beg \u00a9 a pes fe i\n\nMost plug-in units in the srs follow the general main-station equip- \nment format; however, to meet very stringent circuit requirements, a \nmultilayer printed wiring board was used for the logic I unit shown in \nVig. 24 and the logic II and sre alarm units. The logic I printed wiring \nboard consists of four layers; the first two layers contain the signal \ncircuitry, the third a ground plane, and the fourth the power wiring. \nThis board contains over fifty integrated circuit packs and discrete \ncomponents. Connections are made through four multipin connectors \nmated to the gold tabs provided on the board.\n\nThe complete power converter panels are provided, one per JFG. \nEach converter receives power from a separate office feeder lead. This\n\nensures that, even if a feeder is accidently damaged or a main fuse \noperates, the srs will continue in service.\n\nThe Ld surveillance\u2019 physical designs cover the widest cross section \nof types of equipment, ranging from the pure functional types located \nin manholes to the sophisticated equipment located in the center of \nactivity in a control terminal or terminal-main station. In addition to \nbeing influenced by functional design considerations, the latter type \nof equipment was strongly influenced by human-engineering design \nconsiderations and, since it is constantly in the forefront and under \nscrutiny, was designed to have an aesthetically pleasing appearance. \nWhile the repeaters may be the heart of the L5 system, the trans- \nmission surveillance system (TSS) equipment is its nervous system, \nfeeding back information on the health and well-being of the system.\n\nThe fault-location oscillator (FLO) unit and the 31A gate are the \nmanhole-mounted units of the Ttss. A FLO unit is installed in every ap- \nparatus case, a unit with eight outputs located centrally in a basic \nrepeater apparatus case (see Fig. 2, Section 1.1) and a unit with four \noutputs for use with regulating and equalizing repeaters. The Frio is a \nrugged unit made up of an inner chassis and an outer, fabricated, \nshell-like housing. The inner chassis supports the face plate and con- \ntains the four tone generators (two high frequency, two low frequency), \nthe combining circuitry, and the coaxial switches and associated logic \ncircuitry to turn on the unit. Connections within the unit are made \nwith miniature coaxial connectors and connections for the tone out- \nputs are by way of large Bell System coaxial jacks located in the face \nplate. The coaxial patch cords used to connect the FLO tone outputs to \nthe input and output of every repeater (through the twin coaxial-jack \nassemblies located on the apparatus case chassis) are shipped with the \napparatus case and provide continuity through the apparatus case for \ncable acceptance and corona testing. For this reason, the cords are \nrequired to be high-voltage designs, even though they are primarily \nintended to be used at low voltage. The protective, outer covering of \nthe FLo unit consists of a long, thin-walled, rectangular, aluminum \nhousing closed at one end, where it is welded to a rugged, cast, mounting \nbase. This base holds the multicontact connector for input power and \nlogic connections, and contains the large retainer guides that mate \nwith the retainer pins on the apparatus case chassis.\n\nThe 31A gate, the logic unit for the FLO\u2019s, is used on a one-per- \nmanhole basis and is located in the cross-connect apparatus case \n(Section 1.1). It consists of a large epoxy-glass PwB assembly housed in \na sheet-metal, fabricated can mounted on a hinged shelf that swings \nout to allow access to wire terminals and protector blocks inside the \ncross-connect apparatus case.\n\nRectifiers used to power the FLo units and control circuitry for the \nFLO units are located in the main station in the line access bay that \nserves as the interface between the line and the surveillance\u00ae equipment. \nConnections to the surveyed functions in the line bays are made with \nsolid-dielectric, coaxial-cable runs to the line access bay where pro- \nvisions are made to store the excess cable created by the requirement \nfor equal-length, balanced connections. Switched access to the line \nbays is provided by 1-by-12 dry-reed coaxial switches mounted on a \npanel together with installer-wired arrays of coaxial jacks located on \njack strips designed to be removable from the panel for easy access. \n\u201cHairpin-type\u201d plugs are used for interconnections, and the entire \npanel is recessed behind the bay uprights to protect the plugs from \ninadvertent disconnects. The recessing is evident in Fig. 17, where the \nline access bay appears as the third bay in the lineup. The lower part \nof the bay contains the order-wire-access panel followed by several \nfixed shelves containing cable equalizers. The last element in the bay \nis a removable drawer-type storage shelf for the equalizer adjustment \nunit (BAU) described below.\n\nThe transmission surveillance system is made up of the transmission \nsurveillance center (Tsc) and distribution bay combination for control \noffices along an L5 backbone route and the transmission surveillance \nauxiliary (TsaA) for other than control main stations.\n\nThe tsc is a large desk-type console that houses, in its sides, a \ncommercial computer, a transmission measuring test set, and associated \nequipment. The central portion of the console contains the manual \ncontrol panel that displays an array of keys and status indicators within \nconvenient reach of an operator. The Tsa is virtually a Tsc without the \ncomputer and is a standard 26-inch wide unequal flange duct-type bay.\n\nConduction and natural convection were inadequate to transfer heat \nfrom the thermal sources within the tsc. To avoid local hot spots,\n\nforced convection is required, and a fan is installed as part of the rsc \nequipment.\n\nAssociated with a project of the magnitude of an L5 system is a \nwhole series of peripheral items of equipment too numerous to cover in \nfull detail here but of obvious importance to the system and its \nworkings.\n\nOrder-wire bays have been coded and, for the first time, make \nconvenient mounting arrangements available to the field where, \nheretofore, only miscellaneously mountable equipment was provided. \nIn addition, connectorized versions of these bays have been designed \nto avoid difficulties encountered during the Ld initial installation to \naccomplish the myriad wiring connections for complete order-wire \nfan-out in an office.\n\nMulti-mastergroup restoration and zero-loss trunk bay designs in- \nvolve the use of inexpensive \u2018\u2018pseudo-plug-in\u201d construction where \nsimple modules are arranged on a shelf and connected at the rear of \nthe bay with jack-terminated cable arms that are part of the bay \nwire harness.\n\nBecause of the relative inaccessibility of repeaters in manholes and \nthe expense involved in entering a manhole, repeaters that are fully \ntested before leaving the factory are again tested at field maintenance \ncenters before being dispatched to the manholes. For this purpose, a \nspecial repeater test stand and associated warm-up rack were designed \nfor repeater acceptance testing. The test stand simulates the apparatus \ncase chassis details, which make provisions for mounting the L5 man- \nhole plug-in units for verification prior to installation. The verification \nprocedure is speeded up by pre-heating the repeaters on the warm-up \nrack, which is comprised of five mounting panels, each of which can \nhold two basic repeaters or one regulating repeater, yielding a total \ncapacity of ten basic or five regulating repeaters. The same retainer \nguide pins that are part of the apparatus case chassis are used to secure \nthe repeaters on the warm-up rack panels. Powering is by a commercial \npower supply, and the entire unit is mounted on casters for additional \nflexibility in those offices designated as maintenance centers.\n\nRequired settings on manually adjustable E1 and E2 equalizers are \nmade using the equalizer adjustment unit (EAU) designed as a compact, \nlightweight, portable test set (see Fig. 25). The unit uses pwB plug-in \nmodules, is designed for ease of assembly, offers accessibility for main-\n\ntenance, and is stored in the pull-out drawer at the bottom of the line \naccess bay.\n\nWith the Bell System providing the major long-haul communica- \ntions networks for the country, Bell Laboratories designs have always \nbeen motivated toward reliability. Systems like the J- and K-carrier \nsystems designed in the 1930\u2019s and installed in the 1940\u2019s are still \nproviding reliable service. These systems, however, were limited to \ntransmitting 12 two-way telephone conversations over a cable pair, \nwhereas the new L5 carrier system transmits 10,800 conversations per \ncoaxial pair and 108,000 conversations on a fully loaded 22-tube \ncoaxial cable. To accomplish this, new more sophisticated technology \nwas required and to complement this effort a more definitive reliability \nprogram was initiated.\n\nWe define reliability as ensuring that a given component, equipment, \nor system will perform a required function, under stated conditions \nfor a needed period of time.\n\nThis comprehensive reliability program was instituted almost from \nthe start of the L5 carrier system development and has been proceeding \nconcurrently with it. This overall program covered the following \nnine functional steps, some of which must still be completed and, there- \nfore, will not be part of this report:\n\n(t) Derive early estimates of equipment and system failure rates, \nmean time to failure (mTTF), and availability.\n\n(zi) Monitor the actual laboratory prototype experience. \n(4i7) Analyze field-trial production experience. \n(tv) Acquire and analyze actual field-trial data. \n(v) Review and revise, if required, early estimates based on \nlaboratory and field-trial experience. \n(vt) Monitor the Western Electric manufacturing and installation \ninitial route experience. \n(vit) Conduct a full-scale reliability study of the initial service \nroute. \n(vizi) On a continuing basis, obtain and analyze data on all Ld units \nreturned for repair. \n(tx) Conduct a reliability study on a subsequent route, installed \nabout two years after the initial route.\n\nThe purpose and objectives of this program are to obtain accurate \ndata on the reliability performance of the L5 carrier system as early \nas possible, to aid in formulating the early system design concepts, to \nuncover potential problem areas so that corrective action can be \nexpedited before the first system is turned up for service, to compare \nthe actual system reliability performance with the early estimates, and \nto review the reasons for any major differences, so that more accurate \nestimates can be made on future systems.\n\nVery early in the development program, a reliability study of the \nL5 system design was made. As a first step, \u201c\u2018black box\u2019\u2019 failure rate \nestimates were made on all the repeaters and equipment units. While \nit is impossible to predict for any individual electronics part either its \nlife or its rate of degradation under known stress, it is possible to treat \nlarge populations of such parts on a probabilistic basis with acceptable \nresults. These results can then be related to the statistical behavior of \nthe system. The \u201c\u2018black box\u2019 failure rates were calculated by summing \nthe failure rates of the components and parts. The component rates \nused were based on experience in well-defined systems, operating under \nnormal Bell System conditions and environment. For new devices or \ntechnology with little or no previous experience, estimated failure rates \nwere derived after consultation with the device or component designers. \nIn these cases, the rates reflected a very conservative estimate.\n\nUsing the basic repeater which is the most numerous and least \nsophisticated of the line repeaters as an example, it was estimated that \nthe failure rate would be about 900 rit\u2019s, with a Fir defined as one \nfailure expected or experienced in 10\u00b0 hours of service. This results in \nan estimated mrtr of over 125 years for each basic repeater. The\n\nregulating, equalizing, and main-station repeaters being substantially \nmore complex have shorter TTF estimates.\n\nTo obtain overall estimates, two basic system models were selected. \nBoth were 4000 miles long and employed automatic protection \nswitching for the coaxial tubes. In one case, the calculations were based \non switching section lengths at a maximum of 150 miles and, for the \nsecond, a nominal 120-mile spacing was used, based on the L4 coaxial- \nsystem route layout experience. There were some obvious trade-offs in \nthe two models selected. For instance, using the maximum length \nsection resulted in more line repeaters per switching section ; however, \nin a 4000-mile route the number of main stations are reduced from \n33 in a route with normal switching section lengths to 27 for the \nmaximum length sections.\n\nAn important consideration in the system availability estimates is \nthe average time to repair or restore service. Since it was difficult to \narrive at a time that was acceptable, calculations were made based on \noptimistic and pessimistic restoral times. The optimistic times were \ndefined as an average of 15 minutes for main-station equipment \nand three hours for manhole equipment, and the pessimistic times as \nthree hours for main-station equipment and six hours for manhole \nequipment. Recent data on the service restorals for existing coaxial \nsystems show that the average repair times actually being experienced \nare very close to the optimistic figures.\n\nIt is estimated that, for a pair of coaxial tubes in a fully equipped \n22-tube cable, with a 1-for-10 automatically switched redundancy, the \navailability of the electronics and switching contained in a 4000-mile \nroute using the optimistic restoral times would be 99.986 percent for \nan average outage time of about 1 hour and 15 minutes per year. If \nthe pessimistic repair times are assumed for the same system, the \navailability would be 99.948 percent per year for an average of about \n4.5 hours per year outage.\n\nOne very early system question was whether, with double the \nnumber of repeaters in a L5 switching section as compared with the \nL4 system, the ratio of one spare coaxial automatically protecting ten \nworking coaxials was adequate to ensure reliable service. Availability \ncalculations were made on the system model with 120-mile switching \nsections and optimistic repair times employing 1-for-10 and 2-for-9 \nprotection ratios. With the 2-for-9 arrangement, the probability of \nthree simultaneous failures in a switching section was remote; how- \never, the estimated outage attributable to the more sophisticated \nswitching system resulted in outage times for both strategies being\n\nrelatively close. Based on previous experience, cable failures constitute \na greater outage hazard in terms of system unavailability than do the \nelectronics. Therefore, using additional coaxials within the same cables \nas protection for most cable faults would be counter-productive. \nTaking into account these factors and the revenue lost by dedicating \na second coaxial for protection, the 1-for-10 protection ratio was con- \ntinued for the L5 system.\n\nThe field trial of the L5 carrier system started on July 1, 1970 and \ncontinued through June 1972 on a 125-mile cable route between \nCedarbrook and Netcong, N. J., shown in Fig. 26. The purpose of the \nfield trial was to test the L5 as a complete system operating in the \nfield environment before going into full-scale production and regular \nservice. The 240 basic repeaters and 48 regulating repeaters required \nfor the trial were constructed by Western Electric on a special basis \nbecause the quantities exceeded Bell Laboratories model shop capa- \nbilities. Since problems in the field can often be anticipated from \nexperience gained during manufacture, a data acquisition program was \ninitiated by Western Electric for the field trial equipment and, even- \ntually, standard production.\n\nA reliability study was made on the performance of the field trial \nroute. This was intended as prelude to a study of the initial L5 \nservice route between Lillyville, Pa. and Hillsboro, Mo. Data were \ncollected on all failures of equipment and all system outages. A \nthorough investigation was made on any units that failed. A case in \npoint was the failure of a basic repeater traced to the earth-ground \nfilter. After further analysis, including the X-ray photograph shown \nin Fig. 27, it was found that high-voltage shorting was caused by a \ncapacitor unit assembled off-center without outer insulating material. \nTo prevent this in the future, high-voltage test screening was specified \nand tighter quality controls were instituted by the supplier. Failure \nrate calculations were made based on the field data; however, it was \nrecognized that the population was small, the equipment was not \nfrom standard production, and, because of the level of field activity, \nit was difficult to keep accurate data on every single occurrence.\n\nOur experience with previous systems such as the L4 coaxial system \nand the Ld field trial pointed out the need for even greater emphasis \non obtaining accurate data on the manufacturing testing results, the \ninitial defect rates in the field, the infant mortality rates (the infant \nmortality period is defined as the first six months of service), and, \nfinally, the steady-state system reliability performance more commonly \nknown as the normal use period of the \u2018\u2018bathtub\u201d curve. To accomplish\n\n@ MAIN STATIONS \nmw POWER FEED STATIONS \n@ EQUALIZING REPEATER \n\u00a9 REGULATING REPEATER \nLIMITED ACCESS AND \nDIVIDED HIGHWAYS \n\u2014\u2014\u2014 IMPROVED HIGHWAYS \n\u2014 LOCAL AND CONNECTING \nROADS\n\nthis, an elaborate data collection program was initiated. Subunits such \nas coded amplifiers and even Hic substrates were serial-numbered. \nSpecial envelopes and data forms were attached to each unit at the \nstart of manufacture. As the unit was processed through the various \nstages of manufacture, the test results were recorded on punched tapes \nand stored in the envelopes. If components are removed at some stage, \nthey are stored in the envelopes and the data recorded on the forms. \nWhen, for instance, a repeater is ready to ship, a complete pedigree \nhas been established for analysis and future reference. The components \nremoved are forwarded to the experts for analysis to determine the \nfailure mechanism. Through this program, we are able to detect \nfailure trends and initiate corrective action.\n\nAn important part of the reliability program is to establish the \ninitial field defect rate. In Fig. 28, a block diagram traces the pro-\n\ncedures set up to process units returned for repair. Note that a \u201chot \nline\u201d was established by Western Electric to expedite the replacement \nof any units returned for repair so that the service schedules would be \nmet. When units are returned for repair, the designers are able to \nparticipate in diagnosing the problems. Data are recorded on all \nrepairs, and the components removed are kept for further analysis. For \nthe initial route, about 3200 basic repeaters were tested and installed \nin manholes. Approximately 1.5 percent were returned for all reasons. \nAfter further testing, it was determined that about 35 percent of these \nmet all test requirements, resulting in a 1-percent initial return rate. \nOn January 3, 1974, the L5 route between Lillyville and Morgan- \ntown, Pa., was put into regular service, and the Ld field reliability \nprogram was started. On January 25, the section from Morgantown \nto Hillsboro, Mo., was put into service, and this section was added to \nthe reliability study. Before the start of service, visits were made to \nall offices on the route to discuss the objectives of the study and the \ntype of data required. During the early service period, each office is \ncontacted on a weekly basis to report all field problems, even those \nnot affecting reliability. The purpose is to get immediate feedback on \nthe system performance. A complete listing of all equipment in the \ninitial route by serial number was made so that the hours of service \nnecessary for availability calculation can be accurately computed.\n\nJ SWITCH TO PROTECTION COAX \u2018one minute on mone! \n(hutomatic (7 wanuat AUTO. LINE TERMINATE  [L] YES 0) no \nLINE NO- REGULAR PROTECTION\n\nREASON LINE FAILURE = [o> WAINTENANCE =] ResTOoRATION (] \n(EXPLAIN IN IT) (EXPLAIN IN IIT)\n\nIL.__CAUSE_OF LINE FAILURE IIL. MAINTENANCE IV. OUTAGE TIME CAUSED BY \n(CINE EQUIPMENT FAILURE () woot ication CIN AODITION Te seer) \nCI capie FArLuRE ( otter ([] pror.coax.IN USE \n() TeRMINAL EQUI PuENT PROT.COAX.IN USE FOR \n(JH V CONVERTER FAILURE RESTORATION \n(JotHer causes (DO swITCHING FAILURE\n\nDATE/TIME of FAILURE_DE> @ _I1G@2O0 * SEE INSTRUCTIONS FOR \nare e320 EQUIPMENT TO BE REPORTED \nDATE/TIME restoreo_Ub=2__ -- \nATE \nFAILURE SERVICE AFFECTING? Oves [Aft \nNATURE OF FAILURE: CODE____ OTHER\n\nLINE(S) FAILED LF a 4 DATE/TIME \u2014\u2014\u2014_ \u2014\u2014 \nLINE(S) RETURNED TO NORMAL oarte TIME \nee DATE/TIME \u2014\u2014_ \nCe DATE/TIME ieee \n_ Ly DATE/TIME \u2014\u2014\u2014\n\nMAIL TO. MR. W)C. WESTPHAL, ROOM 3B13 \nBELL TELEPHONE LABORATORIES \n4600 OSGOOD STREET \nNO ANDOVER, MASS. 01845\n\nA data form shown in Fig. 29 is filled out and returned to Bell \nLaboratories for failures of any type, even if service was not affected, \nand for the use of the spare line facilities for protection and for main- \ntenance. These forms are returned to Bell Laboratories weekly for \ncompilation. Arrangements have been made so that complete data will \nbe recorded by Western Electric when the units are returned for \nrepair. At the time this paper was written, the system was in service\n\nfor a very short period of time and the field data were limited; there- \nfore, no reportable results were available. This reliability study is \nintended to continue for a minimum of one year.\n\nAs part of our continuing transmission system reliability program, \nthe general performance of the L5 system will be continually monitored \nby our computer-aided reliability program (cARP), which utilizes data \nreceived from the service centers on all units returned from any route \nfor repair. In addition, in about two years we intend to conduct a \nreliability study similar to the one presently in operation on another \nroute.\n\nCopyright \u00a9 1974 American Telephone and Telegraph Company \nTue Beuu SysteEM TECHNICAL JOURNAL \nVol. 53, No. 10, December 1974 \nPrinted in U.S.A.\n\nA family of ultralinear npn transistors has been developed for use in \nthe Ld coaxial-carrier system. These 3-GHz devices are characterized by \nextremely low distortion and noise figure. The transistor comprises an \ninterdigitated base-emitter structure with a heavily doped base grid con- \nnected to the peripheral base metal contact. The emitter contact ts overlaid \non the base-emitter region. Contact metallurgy consists of a platinum \nsilicide, titanium, platinum, and gold system. The transistor ts a highly \nreliable device and meets all the performance requirements of the Ld \nsystem.\n\nTo realize the circuit performance anticipated by raising the coaxial \ncarrier\u2019s highest message frequency from the 17.5 MHz of L4 to the \n60.5 MHz of L5, it was necessary to develop a new family of npn \ntransistors, the 76 and 77 types. The gain bandwidth product fr had \nto be nominally 3,000 MHz; three-tone third-harmonic distortion \nM 3x had to be less than \u201488 dB below 1 mW and the high-frequency \nnoise figure (NF) kept below 2.4 dB at 30 mA. Power dissipation was to \nbe 1.7 watts, which amounted to a considerable improvement in the \nstate of the art.\n\nThe 76-type transistors are used in the equalizing and regulating \nrepeaters and the 77-type in the basic repeater. The 77\u2019s are used as \nmatched pairs to lower intermodulation distortion and the noise figure.\n\nOne basic design is used to meet all of the equalizer and repeater \nrequirements. Six codes were developed in all, each of which was \nchosen for a specific purpose, such as low noise for the input stage of \nthe basic repeater, and low distortion for the output stage. Typical \ncharacteristics for two codes are given in Table I.\n\nSome early studies indicated that of the common geometries of \npower transistors, an interdigitated structure would have the best \ndistortion performance; thus, the present devices are made with \ninterdigitated base and emitter diffusions, but with an overlaid emitter \ncontact. Base current is carried by the heavily doped base grid, under \nthe overlaid contact, and out to metallized regions at the sides. See \nFig. 1. This structure avoids metallizing problems caused by tight \nmetal-contact tolerances, and it minimizes the total area. Emitter \nbonds are made over the active region, eliminating the emitter pad \nand its parasitic capacitance. The emitters are 2.5 um wide and 105 um \nlong.\n\nThe epitaxial thickness is a compromise between being large enough \nto provide sufficient series collector resistance to minimize second- \nbreakdown problems and yet not so large as to degrade device \nperformance.\n\nFollowing the practice set by the L4! system, the transistors are \nencapsulated in a small metal-ceramic package especially designed for \nrf use. It comprises a kovar-beryllia structure that minimizes parasitic \ncapacitance and lead inductance and provides a relatively low thermal \nimpedance of 30\u00b0C/watt. The latter characteristic is important if \nhigh reliability and low distortion (see Fig. 2) are to be obtained, since \nthe output transistors in the basic repeater will operate at 1.7 watts in a \npeak ambient of 85\u00b0C.\n\nThe principal contributor to system noise from the output stage \nof an amplifier is third harmonic distortion, M3z. In a transistor, this \nnoise arises from two main sources: from the essential nonlinearity of\n\nthe emitter current-voltage characteristic and from the falloff of fr with \ncurrent? owing to conductivity modulation of the collector-junction \nspace-charge region, or Kirk effect. The emitter nonlinearity is the \nsame exponential behavior of the forward-biased diode used, for \nexample, in mixers and for harmonic generation. The nonlinearity is \nreduced by increasing the total emitter current. Poon? has shown that \nif the emitter area and collector current (effectively the emitter \ncurrent) are both doubled, M32 decreases by 12 dB (see Fig. 3). If the \ncurrent density becomes too large, however, distortion arising in the \ncollector starts to degrade the device performance. Whenever the \ncurrent density exceeds qvN, where q is the electric charge, v the \ncarrier velocity, and N the density of fixed impurities at the collector, \nthe collector-base junction will move in towards the collector substrate \nand cause fr to decrease. According to Poon,? in the Ld case there is a\n\nIt can be seen that any mechanism, such as the Kirk effect, which \nincreases the curvature of the fr vs J, characteristic, will contribute to \nharmonic distortion. This means, of course, that as the emitter current \nis increased to minimize the emitter-distortion term, the collector area \nmust increase proportionately. At some point it ceases to be economic\n\nFig. 3\u2014Harmonic distortion as a function of collector current for several device sizes.\n\nto increase the emitter current because both capacitance and dissipated \npower become excessive. For the transistor size chosen, a current of \n110 mA provides the lowest distortion.\n\nThe sustaining voltage for the L5 transistor turns out to be bounded \nat both limits. At the lower limit, it is necessary to keep above the \nsupply voltage, so a value of 18 volts minimum is set. At the upper\n\nlimit the doping may become too low and a Kirk effect may occur if the \nvalue of the sustaining voltage is made too high. This result can be \nseen from the following relationships\u2018\n\nwhere \nNezc is the collector epitaxial doping \nN, is the surface impurity concentration \nx; is the junction depth \nn is a constant ~ 4 for the transistor concerned.\n\nOne of the basic differences between L4 and L5 transistors is brought \nabout by the increase of the highest message frequency from 17.5 \nMHz to 60.5 MHz. To accommodate the increased frequency, it is \nnecessary to design the transistor to have a nominal gain bandwidth \ncutoff, fr, of 3 GHz. The solution to fr is given by\n\n1 \na fe th Fase Fe \nwhere \nTe = emitter time constant \nT\u00bb = base time constant \ntT; = collector depletion region time constant\n\nwhere \nK = Boltzmann\u2019s constant \nT = absolute temperature \nq = electronic charge \nJ. = collector-current density \nCre = emitter capacity \nW, = base width \nW. = collector width \nD, = diffusion constant for electrons \n\u2014_ Veo + ae \nJ = peW. \nVe\u00bb = applied voltage \nV, = built-in junction potential \nv = carrier velocity \nSince r, dominates the frequency response, the necessity of keeping \nthe current density low can be readily seen. \nUnder this condition, rz, 72 and 7, reduce to \nWe , Ws \nOe MDs a Vs \na a 1 \u2014J./Jo \n= 20, | 1 \u2014 J./qu:Nec \n| W. 1\u2014 fe | \nEpc \u2019\n\nwhere \nv; = scattering limited velocity \nXmo = depletion layer width \ne = dielectric constant.\n\nOnce the current density has been chosen to be suitably low, the only \navailable design factor for lowering fr is the actual base width. To \nachieve an fr of 3 GHz, it is necessary to lower the base width to \n0.27 um nominal.\n\nNeilsen\u2019s relation\u2019 for the noise figure, NF, provides a simple means \nof determining the choice of design parameters for minimizing the\n\nR, = generator impedance of 50 ohms \nre = emitter resistance = KT/qI. \nfe = frequency at which a is 0.707 of its de value\n\nIt is apparent that once current, gain, and operating frequency have \nbeen chosen, 7; is the only parameter at our disposal and, in fact, for \nthe L5 transistor, the r, term is the largest contributor to the noise \nfigure. The term 7; is made up of two parts. One is the lateral resistance \nunder the emitter stripe, which is minimized by using extremely \nnarrow emitters 2.5 um in width. The outer component consists of \nthe series resistance to the external base contact. This resistance is \nreduced to as low a level as possible by the use of a very heavily doped \nbase grid.\n\nDuring development of the 76- and 77-type transistors, several \nhundred devices were subjected to accelerated aging of both shelf- \ntemperature and power at junction temperatures up to 300\u00b0C. Analysis \nof the data predicts that the transistors will have a failure rate of \nconsiderably less than 10 fits at an operating temperature of 150\u00b0C. \nUnder these conditions, it is no longer reliability that is the determinant \nfor system success, but rather, the temperature dependence of the \nvarious device parameters.\n\nA highly reliable ultralinear family of transistors has been developed \nfor coaxial-carrier-system use. Its high gain-bandwidth product, low \nnoise, and low third-harmonic distortion allow the L5 system to \nsuccessfully meet its operational requirements with an upper master- \ngroup frequency of 60.5 MHz.\n\n1. N. J. Chaplin, G. A. Dodson, and R. M. Jacobs, \u2018\u2018L-4 System: Solid State De- \nvices,\u201d\u2019 B.S.T.J. 48, No. 4 (April 1969), pp. 983-992.\n\n2. H. C. Poon, \u201cImplication of Transistor Frequency Dependence,\u201d IEEE Trans. \nElectron Devices, ED-21, No. 1 (January 1974), pp. 110-112.\n\n3. C. J. Kirk, Jr., \u201cA Theory of Transistor Cutoff Frequency (fr) Falloff at High \nCurrent Densities,\u201d IRE Trans. Electron Devices, ED-9, No. 2 (March 1962), \npp. 164-174.\n\n4. A. B. Phillips, Transistor Engineering and Introduction to Integrated Semicon- \nductor Circuits, New York: McGraw-Hill, 1962.\n\n5. E. G. Nielsen, \u201cBehavior of Noise Figure in Junction Transistors,\u2019\u2019 Proc. IRE, \n45, No. 7 (July 1957), pp. 957-963.\n\nCopyright \u00a9 1974 American Telephone and Telegraph Company \nTue Bei System TECHNICAL JOURNAL \nVol. 53, No. 10, December 1974 \nPrinted in U.S.A.\n\nThe development of transmission networks and magnetic components \nfor the LS system represents the largest network development project of \nits type ever undertaken within the Bell System. Over 200 different coded \ndesigns of networks, requiring in excess of 40 man-years of effort, were \nrequired to meet the frequency-selective and signal-shaping requirements \nof the system. Despite this effort, neither systems requirements nor systems \nschedules could have been met without significant contributions from allied \ntechnologies. This article identifies those technologies and describes design \ntechniques that have advanced the state-of-the-art capabilities in trans- \nmission network and magnetic component design.\n\nAs in other analog systems, transmission networks of the L5 Coaxial- \nCarrier Transmission System perform the indispensable functions of \nfrequency selection and signal shaping. Without the filtering functions \nprovided by certain of these networks, the basic multiplexing arrange- \nments on which the system depends could not be realized, nor could \nthe various fault-locating, equalizing, regulating, and switching pilots \nbe effectively separated from the message portion of the line signal. \nSimilarly, without use of the wide variety of equalizers provided, the \namplitude distortions introduced as the signal traverses the line could \nnot be as effectively corrected. Examples of such equalizers include \nline-build-out (LBO) networks to compensate for cable spans shorter \nthan the nominal one-mile spacing, fived equalizers to compensate for \ntime-invariant and predictable effects, and adjustable equalizers to \nreduce the unpredictable, statistical variations of the system.\n\nTo complete a listing of system functions performed by the trans- \nmission networks of the L5 system would require a tabulation of all \nmajor system functions. Over 200 transmission networks, requiring \nin excess of 40 man-years of effort and representing the largest network \ndevelopment of its type ever undertaken within the Bell System, were \ndeveloped for L5. Despite this large and sustained effort, the required \ndesigns could not have been produced without significant advances in \ncomputer-aided design, precision measurement, magnetic components, \nand monolithic crystal filters. The purpose of the transmission net- \nworks section of this paper is to identify those technologies that have \ninfluenced network developments for the L5 system and to describe \ncertain techniques that have advanced the state-of-the-art capabilities \nin network design.\n\nThe majority of networks for the L5 system were developed using \nclassical network synthesis procedures. Insertion-loss synthesis tech- \nniques were used exclusively for filter realization, Bode\u2019s semi-infinite \nslope approximation for LBo networks, cable equalizers and artificial \nlines, and constant-resistance bridged-T networks, in various topolo- \ngies, for fixed and variable amplitude equalizers. State-of-the-art \nadvances stemmed primarily from the stringent requirements imposed \nby the system rather than from network theoretic developments. \nStrong interaction between systems and network engineers, in frequency \nallocation and the setting of reasonable guard bands, assisted greatly \nin the problem of filter realizability. In practically all cases, develop- \ning techniques for reducing circulating currents, and developing prac- \ntical design procedures for the compensation of parasitics were a \nnecessary adjunct to the physical realization of manufacturable \nnetworks.\n\nThe development of modulation filters for the jumbogroup multiplex \nterminal (smx)! is typical of that for the many filter designs produced \nfor the L5 system. Figure 1 is a simplified block diagram of the gmx \nterminal as viewed by a networks designer. The modulation scheme \nis typical of most analog multiplex systems except for the multiple- \nmodulation steps. Although the double and triple steps of modulation \nincrease the number of filter designs required, they reduce the com- \nplexity of the individual filters. The final circuit and networks objec- \ntives were the result of extensive discussions between systems and\n\nFig. 1\u2014Simplified block diagram of mmx terminal emphasizing transmission networks.\n\nnetworks engineers with the goal of optimizing the trade-off between \nfrequency-spectrum utilization and the setting of adjacent jumbogroup \nguard bands.\n\nAs a result of such discussions, typical filter transition bands ap- \nproximated a reasonable 10 percent, which allowed the use of small \nslug-tuned, air-core inductors having a maximum quality factor (Q) \nof approximately 150. All 14 of the filter designs\u201410 complex bandpass, \n3 low-pass, and 1 low-pass-high-pass filter\u2014are equal-ripple insertion- \nloss designs. The degree of the filters ranged from a minimum of n = 7, \nfor a simple low-pass filter having a relatively low stopband objective \nand a wide transition band, to a maximum of n = 20, for a more \ndifficult bandpass design having an 80-dB discrimination objective \nacross the unwanted sideband. Figures 2 and 3 show the schematic dia- \ngram and measured characteristics, respectively, of one of the more \ncomplex bandpass filters.\n\nThe initial design phases were concerned with the optimization of \nthe filter designs for practical element values, low passband delay \ndistortion, good out-of-band performance, and acceptable inband \nreturn loss. Versatile insertion-loss synthesis, pole-placing, and analysis \nprograms were available.? Some of the features of the programs found \nparticularly useful during the design stage were: (\u00a2) the ability to \nchoose arbitrary stopbands with Chebychev passband behavior, (72) \nthe ability to order loss peak removal to optimize element values, and \n(tit) the automatic computation of network characteristics subsequent \nto the solution of the synthesis problem.\n\nExtensive use was also made of available programs to minimize \nthe delay distortion contributed by the modulation filters for any \nfuture application of the mmx terminal to data systems. Theoretical \ncutoffs were placed as far into the transition region as possible without \nunduly increasing the degree of the filters or making the element \nvalues impractical. This modification of the designs achieved up to a \n50-percent reduction in delay distortion over the desired portion of \nthe passband with only a 10-percent increase in the complexity of the \nfilters.\n\nAs a result of the mx modulation arrangement, the carrier, in many \ncases, fell close to the passband edges, making it impossible (with the \nextended theoretical cutoffs) to adequately suppress the carrier leak. \nQuartz crystals, as shown in the schematic of Fig. 2, were therefore \nintroduced in various shunt branches to suppress the unwanted \ncarrier. Highly stable, high-Q, AT-cut crystals, designed for maximum \nsuppression of unwanted responses, were used. Other components\n\nFig. 3\u2014Characteristics of typical mx bandpass filter. (a) Component of loss \nSar pags by crystals at 42.496 MHz. (b) Stopband loss. (c) Return loss. (d) Pass- \nand loss.\n\nincluded adjustable air-core inductors and dipped-mica and tubular \ncapacitors. These components were mounted on printed wiring boards \n(pwp\u2019s) and packaged in sealed enclosures.\n\nLaboratory development of the smx networks was a difficult and \nlengthy process because of the wide bandwidths and high frequencies \ninvolved. Precision scanning network analyzers were used widely in \nthe initial stages of development, but in all instances the computer- \noperated transmission-measuring set? (corms) was used for final \ncharacterization.\n\nMonolithic crystal filters\u2018 have found wide application in the L5 \nsystem as narrow-band pick-off filters and as wide-passband band-\n\nelimination structures. The monolithic crystal filter was first intro- \nduced in the Bell System plant as simple two-resonator bandpass \nfilters for carrier-supply applications\u2019 in the L4 coaxial system. For \nthe L5 system, over 40 designs, ranging in frequency from less than 3 \nMHz to nearly 100 MHz, were required for both bandpass and band- \nelimination applications. To cover this frequency range, fundamental \nand third-overtone crystal designs were necessary, and multiresonator \nfiltering arrangements were provided to meet the high out-of-band \nsuppression requirements.\n\nThe monolithic crystal filter, in its most elementary form, consists \nof a quartz plate with an array of two-electrode pairs placed on its \nmajor surfaces, with each pair establishing a resonator system. The \nresonators are acoustically coupled via the elastic quartz plate. If \nthe quartz plate, electrode pairs, and acoustic coupling are properly \nproportioned, a self-contained, two-port piezoelectric device, having \nprescribed filtering characteristics, results. Higher-order filter func- \ntions may be obtained by deposition of multielectrode pairs on an \nextended quartz plate, with adjacent resonators acoustically coupled. \nAs an extension of this arrangement, multiple quartz plates may be \nused, with adjacent resonators (on facing quartz plates) coupled \nelectrically (capacitively). A third alternative, the composite mono- \nlithic crystal filter, uses Lc filter sections to couple individual two- \nresonator monolithics. The latter approach was used exclusively in \nthe design of more than 30 monolithic crystal bandpass filters for L5 \napplications. Impedance-matching sections necessarily were included \nin these designs to match the impedances of the coupled-resonator \ndesigns to specified source and load impedances. This combination \nof Lc coupling networks, Lc impedance-matching networks, and two- \nresonator monolithics led to the concept of the monolithic crystal \nfilter (MCF) as a two-port, building-block device available for inclusion \nin some larger frequency-selective network.\u00ae?\n\nBefore describing those methods of design peculiar to the composite \nMcF, a more detailed summary of methods of deriving higher-order \nfilter functions is required. Figure 4 shows the three different tech- \nniques described above. In each instance, an impedance-matching \nnetwork is indicated at the input and output ports of the networks. \nThese may or may not be required. Figure 4a shows the multiresonator \nMCF on a single piece of quartz with the interresonator coupling \nsupplied by the built-in acoustic coupling properties of the device \nitself. To avoid acoustic coupling of certain unwanted modes, the \napproach indicated in Fig. 4b is sometimes advisable. In this instance, \nthe single plate of Fig. 4a has been split into two parts and the facing\n\nFig. 4\u2014Methods of deriving higher-order filter functions. (a) Multiorder mono- \nlithic filter with acoustic interresonator coupling. (b) Multilithic filter with acoustic \nand electrical interresonator coupling. (c) LC network coupling of two identical Mcr\u2019s.\n\npairs of electrodes electrically coupled by the capacitor are shown in \nthe figure. The present design of the A6 channel-bank filters\u00ae uses this \ntechnique. The third method of realization is represented in the \nschematic of Fig. 4c. This schematic emphasizes the concept of the \nMcF as a two-port, building-block device employing supplementary \nLc filter sections (coupling networks) and impedance-matching net- \nworks for realization of a particular transfer function. In all three \nexamples shown in Fig. 4, the classical work of Dishal\u00ae provides closed- \nform solutions in the design of narrow-band bandpass filters to exact \namplitude characteristics.\n\nIn view of the importance of the building-block concept to the \ndesign of composite McF\u2019s, devices and filter sections required for \nfilter realization are shown in Figs. 5 and 6. Emphasizing the role of \nthe mcr as a two-port device in composite-filter design, schematics \nfor the basic and reversed-phase connections of a symmetrical, two- \nresonator McF and their equivalent circuits are shown in Fig. 5. The \nassociated Lc networks required in a composite filter design are shown\n\nin Fig. 6. These include both impedance inverters and impedance- \nmatching networks. The impedance inverters are shown in their T- and \npi-configurations and consist of positive and negative capacitive \nelements. In the final network design, the negative capacitances shown \nin the figure are either absorbed in positive capacitances existing \nwithin the network, or are approximated in a narrow-band sense by \ninductances.\n\nThe impedance-matching networks of Fig. 6 are shown in their \nL-configurations. It is important to note that the shunt elements of \nthese networks must also be designed to \u2018\u2018absorb\u201d\u2019 the input or output \nstatic capacitances of the associated McF\u2019s.\n\nGiven the equivalent circuits of Figs. 5 and 6, design techniques \napplicable to multiresonator bandpass filters become readily apparent. \nA familiarity with the theory of coupled-resonator design, however, is \nassumed. References 9, 10, and 11 provide excellent introductions to \nthe theory. Starting with a general n-resonator structure, the equiv- \nalent, composite McF is easily derived by a series of network de- \ncompositions as shown in Fig. 7. A filter design having equal inductance \nvalues throughout the structure is indicated. As shown in the figure, \nimpedance inverters, having element values corresponding to the \nshunt capacitances of the structure, have been introduced into the \ncircuit at each of the shunt branch points. Certain sections of the\n\n~ PLATE Co \n7 aaa \n{ \\ \nstleaet. sees lvate M yasonton a Rte szsouaon \nPAIR NO. 1 PAIR NO. 2 COUPLING \n(a) \nCy -Cy -Cy Cy\n\nFig. 5\u2014Schematics and equivalent circuits of monolithic crystal filters. (a) Basic \nbandpass filter. (b) Reversed-phase bandpass filter.\n\nresulting network can be identified as individual, two-resonator \nmcr\u2019s. Impedance inverters coupling the mcr\u2019s are also easily identified. \nIf a composite filter design is sought, the impedance inverters indicated \nin the figure must be changed to their equivalent pi-configurations and \nthe negative capacitance appearing in the series arms replaced by \nequivalent inductances on a narrow-band approximation basis. The \nfinal circuit arrangement is shown in the lower schematic of Fig. 7. \nIt should be noted that, for the composite filter structure comprised \nof individual two-resonator mcr\u2019s, the design must be constrained to \nan even number of resonators. .\n\nThe insertion-loss characteristic of a composite monolithic crystal \nbandpass filter, designed for use as a pick-off filter in the regulating \nrepeater of the L5 system, is shown in Fig. 8.\n\nimMpEDANcE\u2014 | | | COUPLING | | | IMPEDANCE\u2014 \nMATCHING R, NETWORK Ro MATCHING \nNETWORK NETWORK\n\nFig. 7\u2014Decomposition of an n-resonator bandpass filter to an equivalent, com- \nposite, monolithic crystal filter structure. (a) n-resonator bandpass filter. (b) \nEquivalent, composite, monolithic crystal filter. (c) Final composite filter.\n\nAlthough the development of the monolithic crystal bandpass filter \nwas first reported in early 1965, it was not until approximately three \nyears later that the first disclosure of the monolithic band-elimination \nfilter (BEF) was made. Two types of monolithic BEF\u201d may be derived \nfrom monolithic crystal bandpass filters simply by the addition of \ninductive or capacitive elements connected across the ungrounded \nelectrodes of conventional two-resonator monolithies. Different connec- \ntions of the mcr plates, however, must be used to effect proper phase \nrelations in the two types of filter. These phase relations are obtained \nby use of the basic or the reversed-phase connections previously\n\nshown for the bandpass filter of Fig. 5. Network schematics and \nequivalent circuits for the inductor- and capacitor-derived BEF\u2019s are \nshown in Fig. 9.\n\nFor the inductor-derived filter, the reject frequency occurs when the \nmagnitude of the reactance of the bridging inductor equals that of \nthe acoustic-coupling capacitors of the mcr. This reject frequency \noccurs at the center frequency of the bandpass filter. For the capacitor- \nderived filter, the reject frequency is obtained when the magnitude of \nthe bridging capacitance is made equal to that of the acoustic-coupling \ncapacitors.\n\nThe band-reject properties of either type of BEF are easily demon- \nstrated by application of Bartlett\u2019s bisection theorem. The equivalent \nlattice of either network is first derived from the open and short- \ncircuited impedances of the bisected network. Combining the re- \ndundant elements in the series and lattice arms of the lattice and\n\nplotting reactance as a function of frequency, the passbands and \nstopbands can be readily identified. For the inductor-derived filter, \nthe insertion loss characteristic includes a passband on either side of a \nreject band, with a second reject band extending to infinity beyond \nthe second passband. The inductor-derived monolithic BEF is, therefore, \na bandstop filter imbedded in a low-pass filter. The capacitor-derived \nfilter, however, is a bandstop filter imbedded in a high-pass structure.\n\nFor practical applications, the inductor-derived filter exhibits \nsuperior passband performance compared to its capacitor-derived \ncounterpart. Consequently, the remaining discussion of the monolithic \nBEF will be restricted to the inductor-derived circuit. The initial \ndesign procedures, used successfully on both single-section and multi- \nsection monolithic BEF\u2019s, were based on dividing the frequency \nspectrum into two regions: (7) the stopband, over which the motional \nreactances of the mcr change rapidly as the crystal passes through its \nresonances, and (72) the passband, over which the motional reactances \nof the mcF are sufficiently high to be considered as open circuits. The \nextreme stiffness of the resonators makes it possible, therefore, to \nattribute the frequency response, above and below the narrow stop- \nband, solely to the low-pass filter section of the network. The narrow\n\n9\u2014Monolithic crystal, band-elimination filters. (a) Inductor-derived filter. \n(b) Ganache filter.\n\nstopband, however, is controlled by the interactions of the mcr and \nthe bridging inductor.\n\nSingle-section and multisection, inductor-derived, monolithic crystal \nBEF\u2019s, shown schematically in Fig. 10, may therefore be designed \nusing standard low-pass synthesis techniques to meet passband \nrequirements. The narrow reject band is formed by adding an esti- \nmated number of mcr\u2019s, with each monolithic bearing the proper \nrelationship to the corresponding bridging inductor. The stopband \nperformance can then be evaluated from an equivalent-circuit model \nof the entire network with crystal units being added or removed until \nthe desired characteristic is obtained.\n\nAn alternative design technique to that outlined above results in a \nclosed-form synthesis procedure in which a set of filter requirements \nsuch as passband ripple, cutoff frequency, and stopband bandwidth \nare related directly to the set of equivalent circuits shown in Fig. 11. \nThis figure clearly delineates the low-pass and monolithic-crystal \nsections of the composite filter.\n\nFig. 10\u2014Inductor-derived, monolithic crystal Brr\u2019s. (a) Single-section filters. (b) \nMultisection filters.\n\nWriting the cascade matrix for each of the component networks \nof the composite structure, the complete matrix may be formed by \nconverting each of the component matrices to its equivalent admittance \nform, adding the two resulting matrices and converting the sum back \nto the cascade form. Such a procedure is essentially routine and \nprovides little insight into the synthesis procedure. If, however, these \nmatrices are expressed in terms of dimensionless parameters\" and a \nnormalized frequency variable, the elements of a synthesis procedure \nbecome apparent. In each instance, the several dimensionless param-\n\nEQUIVALENT CIRCUIT OF INDUCTOR\u2014DERIVED, MONOLITHIC \nCRYSTAL, BAND-ELIMINATION FILTER\n\nQ/K,2 WHERE Q = f/fy - fosg = 2(f-fol/fp * \nAND K,9= 1/(wo? Cy YLiL2 ) \n* NARROWBAND APPROXIMATION \nDIMENSIONLESS PARAMETERS AND DEFINING NETWORK SECTIONS \n(b)\n\nFig. 11\u2014(a) Equivalent circuit of inductor-derived, monolithic crystal, band- \nelimination filter. (b) Dimensionless parameters and defining network sections.\n\nA complete discussion of the cascade matrix approach to the design \nof composite monolithic crystal BEF\u2019s is beyond the scope of the \npresent paper; however, the usefulness of the method in describing \nseemingly anomalous behavior will be illustrated. A particularly \ninteresting example is that of a two-resonator monolithic BEF having \nan asymmetric stopband and a Butterworth passband with selected \nparameters \u00ab = 0.5, D = 1 (see Fig. 11), and with f, < f.., where f. \nis the reject frequency and f.. is the cutoff frequency of the low-pass \nfilter. The resulting characteristic of this design is shown in Fig. 12. \nThe solid-line plot is the insertion-loss or stopband model predicted \non the basis of the foregoing development. The dashed-line charac- \nteristic is that of the Butterworth low-pass filter, and the circled \npoints indicate the characteristic as derived from computer-based \nmesh analysis of the complete equivalent circuit. Two transition re- \ngions are to be noted: the first, below the reject frequency, f., and the \nsecond, above the reject frequency. In the region below the lower \ntransition region and above the upper transition region, the circled \npoints fall almost exactly on the computed characteristic of the low- \npass filter by itself. In the region indicated as the stopband model \nregion, the characteristic predicted by the cascade-matrix method of \nanalysis is duplicated. In the transition regions, however, the approxi- \nmation is not nearly as good. To simplify the analysis, it was assumed \nthat the reactances of the low-pass filter remained constant in a \nnarrow-band approximation. This approximation breaks down, how- \never, over the transition and passband regions of the per. The results \nof the mesh-analysis computer runs, nevertheless, agree with the \npredicted characteristic in the stopband to at least three significant \nfigures. The smoothest transition regions are attained in symmetric \nstopband designs when the reject frequency is placed at an insertion \nloss zero of a Chebychev low-pass filter. This is illustrated in Fig. 13 \nfor the case of a 0.1579-dB ripple Chebychev passband.\n\nRequirements for some of the BEF\u2019s specified for the L5 system \nincluded both high-loss stopbands (50 to 80 dB) and very wide pass- \nbands (approximately 70 MHz), coupled with low-distortion and \nhigh-return-loss requirements. Using Chebychev low-pass filters to\n\nFig. 12\u2014Characteristic of band-elimination filter with asymmetric stopband and Butterworth passband. Transition regions are\n\nFig. 13\u2014Characteristic of band-elimination filter with symmetric stopband and 0.1579-dB-ripple Chebychev passband. Transition \nregions are indicated. Stopband center frequency f, is at insertion loss zero of the low-pass filter.\n\nmodel the low-pass section of the composite filter, it was found possible \nto meet all systems requirements. A specific example of design capa- \nbility is given in Fig. 14 for a 42.880-MHz BEF, using three two- \nresonator monolithics imbedded in a ninth-order Chebychev low-pass \nfilter. This low-pass filter was designed at an impedance level of 450 \nohms to achieve optimal inductance values for the mcr\u2019s. The filter \nwas then matched to the 75-ohm source and load impedances by use \nof auto-transformers, as shown in the figure. Impedance-matching \nsections, previously described for bandpass designs, could not be used \nbecause of the narrow-band approximations involved. To meet the \npassband insertion-loss deviation requirements of +\u00a30.05 dB from flat \nresponse over the entire frequency range of the L5 system, a simple \nbridged-T amplitude equalizer was also required.\n\nIn the L5 coaxial system, a total of 28 adjustable bump shapes are \nprovided to correct for time-invariant amplitude distortions.!\u201d Ten of \nthese shapes are allocated to the E1 equalizer for coarse correction of \ndistortions introduced by variations in the manufactured product. \nThe remaining 18 shapes are provided by the E2 equalizer to correct \nwhatever residual distortions remain after employment of all lower \nlevels of equalization, including the E1 equalizer.\n\nThe majority of bump shapes are realized by series-type, ae \nbump, adjustable Bode equalizers. The remainder are supplied by \nshaping networks introduced into the feedback paths of isolating \namplifiers. The amplifiers buffer the input and output ports of each \nBode network and supply gain to offset the losses introduced by the \nequalizers. Each of the multibump equalizers, however, introduces no \nmore flat loss than would have been introduced by any one of the several \nsingle-bump equalizers required to duplicate its performance. With \na reduction in the number of Bode networks, the number of isolating \namplifiers is also reduced. This, in turn, results in a reduction in inter- \nmodulation and thermal noise. Finally, the introduction of shaping net- \nworks into the feedback of the isolating amplifiers also reduces the \nnumber of Bode networks required. The net effect of all these consider- \nations is to improve reliability, to reduce cost, and as previously in- \ndicated, to reduce intermodulation and thermal noise.\n\nAll of the multibump Bode networks supplied for the E1 and E2 \nequalizers were derived from extensions of the simple series-type \nadjustable network of Fig. 15a. The basic shaping network, as indicated \nin the figure, is the bridged-T equalizer. Figure 15b shows the range \nof insertion-loss shapes that may be obtained by variation of load \nresistor, Ra.\n\nAs demonstrated by Lundry,!\u00ae the insertion loss of the generalized \nBode network of Fig. 16a can, to a good approximation, be expressed as\n\n6 \u2014 00 = K*pae*?, (1) \nwhere \n@ = the image transfer constant of the bridged-T network \n6 = the insertion loss of the adjustable equalizer \n6. = the insertion loss of the equalizer when Rz = R, \neo \u2014 ] \nef + 1\n\nFig. 15\u2014Series-type, single-bump, adjustable, Bode equalizer. (a) Schematic. \n(b) Characteristic.\n\nwhere p(f) is the input reflection coefficient of the shaping network \nrelative to R, and is a function of frequency. Considering only the real\n\nFig. 16\u2014(a) Series-type, single-bump, Bode equalizer. (b) Series-type, double- \nbump, Bode equalizer. (c) Bump characteristic, H (f).\n\nwhere a, is the insertion loss of the structure when R, = R,. Since \na, and x are constants, it is only necessary to consider p(f).\n\nTo realize a single-bump shape with the network of Fig. 16a, \nRe |p(f)| is required to have the form\n\nwhere H(f) is a bump-shaped function of frequency (Fig. 16c), and \np\u2019(R.) is a function of a single resistor, independent of frequency.\n\nSimilarly, to realize a double-bump shape with the network of Fig. \n16b, p(f) will have the form\n\nwhere the subscripts a and b refer to two separate bump shapes. Both \nH,(f) and H,(f) have the shape indicated in Fig. 16c. For the type of \ndesigns discussed here, it is required that these shapes be in distinct \nfrequency bands. R, and R, provide independent control of the two \nshapes.\n\nOne method of realizing eq. (5) is to use bridged-T sections as \nshaping networks. To demonstrate this, a general expression for the \ninput reflection coefficient of a bridged-T network is required. Referring \nto the schematic of Fig. 17b, which is an unconventional arrangement \nof the circuit of Fig. 17a, the scattering matrix! of the network, with \nZ1:Z2 = Rj, is given by\n\nwhere a; and 5; refer to the incident and reflected voltage waves at the \nith port, respectively, and\n\nThe reflection coefficient p(f) = b:/a1, looking into port 1 of the \nnetwork, must now be determined. Ports 2 and 4 are terminated in \nimpedances whose reflection coefficients are pe(f) and p4(f), respec- \ntively. Assume port 3 is terminated in R,, and as a result, p3(f) = 0.\n\nFig. 17\u2014Bridged-T networks with ports 1 through 4 open-circuited. (a) Con- \nventional. (b) Unconventional.\n\ndz = po(f)bz \nas = pa(f)bs (7) \na3 = 0. \nThese port relationships, together with eq. (6), lead to \no(f) = Si(f)-p2(f) + Sta(f) -pa(f). (8)\n\nwhere p.(R.) has been substituted for p.(f), since a resistive element \nhas replaced a frequency-dependent impedance. This has the form of \nthe reflection coefficient for a single-bump shape defined by eq. (4). \nThe single-bump network, based on the unconventional bridged-T \nnetwork of Fig. 17b, is shown in Fig. 18a.\n\nAssume that the shaping network is such that Re [Si4(f) ] is a bump \nshape centered at frequency f,. As a result, Sis 2 1 in the vicinity of \nfa and s14 & 0 elsewhere. From eqs. (8) and (9)\n\nThe term Sis(f)-p4(Ra) represents a bump shape. As a result, half \nof the desired equation, (5), has been realized. To realize the second \nbump shape, p2(f) is made the input reflection coefficient of a similar \nshaping network whose center frequency is f,, with fs << fa or fr > fa. \nNetworks having the assumed properties for Si, are shown for a two- \nbump equalizer in Fig. 18b. For this network,\n\nEquation (10) is now in the proper form to realize the defining \nequation, (5), of the double-bump equalizer. The equivalence of the \ntwo equations becomes apparent when the following substitutions are \nmade in eq. (10):\n\nFig. 18\u2014Shaping network for adjustable bridged-T Bode equalizers. (a) Single \nbump. (b) Double bump. (c) Triple bump.\n\nTo compensate for changes in cable loss relative to ground-tempera- \nture variations, regulating repeaters are placed in the Ld line approxi- \nmately every 5 to 7 miles.\u201d Relative to nominal cable loss at mean \nground temperature, both gain and loss compensation are required \naccording to whether the temperature is above or below mean tem- \nperature. In either case, the shape, on a dB (log) basis, is proportional \nto the square root of frequency. The networks that provide the compen- \nsation, as in the L4 system,\u201d are thermistor- and pilot-controlled, \nwideband, square-root-of-frequency Bode networks. They differ from \nthe L4 designs, however, in that buffering amplifiers are included as \nintegral components of the networks, as shown in Fig. 19. Design\n\nconsiderations, however, are basically the same and are not \nreconsidered.\n\nThree measurements are required to fully characterize the network: \n(t) at high-loss setting, (77) at nominal setting, and (777) at low-loss \nsetting (see Fig. 20). The dynamic errors may be defined as\n\nwhere An(f), Aw(f), and Azx(f) are loss measurements in the high-, \nnominal-, and low-loss conditions, respectively, and Ky and Ky are \ndetermined by requiring the error to be zero, by definition, at the pilot \nfrequency of 42.88 MHz. The static error may be defined as\n\nAr(f) is the expected loss in the nominal condition, which, because of \nparasitics, gain shaping in the amplifiers, and interaction effects, is not \nflat with frequency. Deviations from the ideal flat shape, however, \nhave the same effect as deviations from the nominal characteristic \nof other components in the transmission path and may be corrected \nby fixed and adjustable equalizers available in the system. Since \ncomparable dynamic equalizers are not available, more stringent \nrequirements are placed on the dynamic performance of the Bode \nequalizer than on the flat or reset shape.\n\nThe necessity of requiring three loss measurements, corresponding \nto three thermistor settings, to characterize network performance, \nprecludes the use of conventional methods for compensating parasitic \nvariations in the amplifiers and the network during manufacture. As \nit is not possible to guarantee network performance from easily \nmeasured amplifier parameters, the alignment procedure must correct \nfor both network and amplifier variations. Since the amplifier is not \nadjustable, the network must provide tuning capabilities for both \nshaping and amplifier compensation. Simple prescription of element\n\nvalues or adjustment of resonances would not provide this capability. \nFurthermore, an adjustment in a network component usually affects \nall three shapes, making adjustment on a scanner too difficult. Thus, a \nmore sophisticated procedure is needed.\n\nSince the amount of adjustment needed is small, but the allowed \ndeviations even smaller, linearization has proved to be valid. The \nerror after adjustment, ea(f), can therefore be expressed as:\n\nwhere ey1r(f) is the initial error, Aysz(f) is the normalized deviation \ncaused by the jth component, and 2; is the amount of the deviation \nused for adjustment. Similar relations hold for the nominal- and low- \nloss errors. The composite error may be defined as the summed weighted \nsquares:\n\nA weighting W less than unity allows a larger error in the nominal or \nreset shape than in the dynamic errors ey or ez. Given an initially \ndetermined set of errors for the three conditions, over an appropriate \nfrequency band, a set of parameter deviations 7; may be found that \n_minimizes e?, using the so-called least-sum-of-squares procedure. This \nprocedure is well known and only requires the solution of a set of \nsimultaneous linear equations.\n\nUnfortunately, in practice, the unconstrained solution often requires \nan adjustment greater than the adjustable components can provide. \nThus, it was necessary to include a search for the best constrained \nsolution. The method chosen uses gradient information to determine \nif a parameter must violate the constraints to lessen the summed- \nsquares error, e?. Other than being iterative, the method does not \nrequire any information beyond that obtained from the least-sum-of- \nsquares procedure. While inclusion of the constraint algorithm con- \ntributed to the success of the program in manufacture, a complete \ndescription is omitted because of space limitations. It is necessary, of \ncourse, to preset the adjustments to a prescribed condition, prior to \nalignment, to correspond to bounds stored in the program.\n\nProgramming for the constrained least-sum-of-squares algorithm \nwas added to a streamlined measurement program for cotms. The \ncomputer controls the measurement sequence and the alignment \nprocedure automatically. The only manual tasks required during \ntuning are to switch the network into the different conditions and to \nturn the slugs on the adjustable components. Measured, rather than\n\nFig. 21\u2014Deviations in error with element adjustment in square-root-of-frequency \nBode network. (a) Li. (b) C4. (ce) L5. (d) L3.\n\ncomputed, deviations from a typical network are used due to the \nparasitic modeling problem. Because adjustments are effected on a loss \nbasis, it is neither necessary to count turns of the adjustable inductor \nslugs nor necessary to measure inductance.\n\nThe deviations in the high-, nominal-, and low-loss shape errors, \ncaused by variations of four components, are shown in Fig. 21. The \ncurves for inductor L1, for example, show the deviation in the errors \ncorresponding to a 0.1-dB change at 70 MHz in the nominal- or flat-\n\nloss-setting response. The result of the alignment algorithm is a \npercentage of this deviation. For example, if the multiplier x; of eq. \n(13) is determined by the least-sum-of-squares procedure to be 1.32, \nmultiplication of inductor L1 sensitivity (0.1 dB) by this number \nresults in 0.132 dB. Inductor L1 must, therefore, be adjusted for a \n0.132-dB change in loss, at 70 MHz, to effect the optimum tuning of \nthat element. For inductor L3, the maximum deviations from flat \nresponse occur at approximately 20 MHz over the 1- to 70-MHz \nfrequency range. This inductor is, therefore, tuned at that frequency. \nSix adjustable components are included in the program. Only one \nmeasurement, adjustment, and remeasurement iteration proved \nnecessary for the adjustment of the networks for optimum performance. \n\u201cBefore\u201d and \u2018\u2018after\u2019\u2019 adjustment error curves are shown in Fig. 22.\n\nFig. 22\u2014Network errors before and after adjustment. (a) Before adjustment. (b) \nAfter adjustment. (c) Before adjustment. (d) After adjustment.\n\nThe computer-aided-adjustment program has proved its value \nduring the several years of manufacture of the regulating network. \nWithout corms, this approach would not be feasible. Without com- \nputer-aided adjustment, the dynamic regulating objectives for the L5 \nsystem could not have been met.\n\nIn a system having as complex an equalization arrangement as L5, \nthe possibility of advantageous interactions among equalizers in the \ncomplete hierarchy of equalization must be considered in any one \ndesign. An example of such design considerations is the design of the \nrelatively simple, fixed-loss, deviation equalizer. This equalizer \ncompensates for the average error in the match of average-line-repeater \ngain to nominal cable loss. Three deviation equalizers are installed \nin each power-feed section.\n\nInitial evaluation of the design led to the conclusion that require- \nments could be met with an equalizer consisting of two valley-shaped \nand two bump-shaped bridged-T equalizer sections connected in \ntandem, as shown in Fig. 23. It was necessary, however, that the \ndesign be optimized so that the equalizer, in combination with the E1 \nand E2 adjustable networks, would provide optimum compensation. \nThis was achieved using the general-purpose optimization program \n(@popr) tied together with an E1, E2 simulation program.\u201d!\n\nThe E1, E2 program contains the measured response of all the bump \nshapes included in the El and E2 equalizers. With approximately 15\n\nFig. 23\u2014Deviation equalizer. (a) Valley-equalizer sections. (b) Bump-equalizer \nsections.\n\npercent of each El and E2 bump allotted to the deviation equalizer, \noptimization resulted in an equalizer characteristic having deviations \nfrom ideal placed at frequencies where the El and E2 adjustable \nequalizers could best supply shaping to reduce the overall misalignment.\n\nThe E1, E2 simulator package simulated the iterative process that a \ncraftsperson would use in the field in making system equalization \nadjustments. This yielded an effective design which has led to excellent \nresults on equalized lines.\n\nAlthough essentially every transmission network for the L5 system \nprovided its own challenges in physical realization, the majority of \nchallenges related to compensation for parasitic inductance and \ncapacitance, to minimization of ground loops, and to reduction of \ncoupling between components. Environmental protection was provided \nby drawn or fabricated metal enclosures with internally mounted \nPwp\u2019s for supporting and interconnecting the individual circuit com- \nponents. External connections were, in most instances, made by use \nof moisture-resistant plugs or jacks.\n\nIn several instances, however, substantially greater physical-design \neffort was involved. Two examples of such design effort are: (z) the \ndesign of the shaping networks of the line repeaters, and (77) the \nphysical design of the earth-ground filter.\n\nThe shaping networks of the basic, regulating, and equalizing \nrepeaters were specifically designed to reduce excess low-frequency \ngain, to furnish surge protection, and to provide the power-separation \nfiltering requirements of the L5 repeaters. Figures 24a and b show the \nschematics of the low-frequency A and B networks. The circuit \ndiagrams identify those portions of the networks concerned with the \ndifferent circuit functions. The network schematic of the bridged-T \nequalizers used to obtain the desired insertion loss is shown in Fig. 24c.\n\nThe physical design of the shaping networks required considerable \ninteraction with system engineers. For overall efficiency and economy | \nof space utilization, the surge-protection and power-separation \ncircuits were included in the shaping-network package. As indicated \nin the photograph of Fig. 25, the two networks were placed side by \nside in the lower section of the basic repeater housing. Requirements \non RF isolation, between input and output ports of the repeaters, led \nto the use of cast-aluminum housings for packaging the shaping \nnetworks. These housings, in turn, furnished the necessary mechanical\n\neS \nFROM | SHAPING 70 \nEARTH | SURGE ! NETWORK PRE- \nGROUND, (@ PROTEC- | AMPLIFIER \nFILTER , \n(a) \nPOWER SEPARATION \nBio hte ee 5 \n| \n| \nJ \nBRIDGED 1 \na OUTPUT \nSHAPING \nFROM \nPOWER NETWORK \nAMPLIFIER \n(b) (c)\n\nFig. 24\u2014Low-frequency shaping networks. (a) Network A. (b) Network B. (c) \nBridged-T network schematic.\n\nrigidity to meet the +0.025-dB tolerance on insertion-loss shape as \nspecified for manufacture.\n\nThe earth-ground filters of the line repeaters, shown in the photo- \ngraph of Fig. 25, with their coaxial jacks protruding from the repeater \nhousing, perform several important functions. Despite the seemingly \nsimple electrical circuit shown at the input and output ports of the \nsimplified block diagram of the basic repeater of Fig. 26, the earth- \nground filter must\n\n(22) Maintain a low-impedance return path between the grounds \nfor all frequencies in the L5 band.\n\n(izi) Provide a low-pass filter characteristic with a minimum of \n120-dB isolation between repeater input and output (through \nthe ground return) for signals in the L5 band.\n\n(tv) Carry the L5 line frequencies and the de powering current with \na minimum of amplitude distortion.\n\nThe physical design of the earth-ground filter\u201d presented a design \nchallenge. As shown in Fig. 27, the filter contains two coaxial chokes in \nthe form of ferrite toroids and three high-voltage, high-reliability, \ndual-dielectric capacitors packaged in a sealed enclosure. The filter, \nminus the coaxial plug, is impregnated with polyisobutylene and \nhermetically sealed to prevent the entrance of moisture or the leakage \nof impregnant, which would contaminate the connectors.\n\nTo insure the high reliability of this design, considerable emphasis \nwas placed on the compatibility of materials, structural integrity, \nsealing, and manufacturing control and testing.\n\nThe L5 system required the development of a wide variety of mag- \nnetic components numbering in excess of 100 different experimental \ndesigns. This number includes original designs, subsequent improved \nversions, and redesigns that reflected changing circuit concepts or \nrequirements. Ultimately, about 50 apparatus codes were issued to \nspecify these designs for manufacture. Many additions to existing \ncode series of magnetic components were also made available for L5 \napplications, but these are not discussed. The adjustable, air-core, \nprecision inductors specified for the manufacture of many of the trans- \nmission networks developed for L5 are an example of magnetic com- \nponent designs based on existing code series.\n\nCoded transformers and inductors appear in all segments of the L5 \nrepeatered line. Frequently, these components are associated with \nthin-film circuits and, as a result, miniaturized designs had to be made \navailable. Development of inductor structures for such applications, \nhowever, predated the development of the L5 system, and, as a result, \nemphasis is placed here on the development of various types of trans- \nformers. Many transformer designs for the line repeaters required \npin-type terminals for interconnecting and mounting on PWB\u2019s or \nthin-film substrates. These transformers are discussed in Section 2.2, \nRepeater Components.\n\n/ FLANGE \n/ SOLDER JOINT \nJs \nsoe / FINGER STOCK \n4 7s \nff / \ned FIRST ie ,SECOND CAPACITOR \u2014_, THIRD CAPACITOR \n/ ue // CAPACITOR / , i \n/ 7 \ni // ,CORE TUBE/\u201d 7 MINI-COAX Fa / KRAFT PAPER \n7 7 fe / /\n\nMagnetic components intended for terminal applications are fre- \nquently used in complex arrays to develop hybrid trees arranged for \ncoaxial interconnection. The different methods used to establish an \ninterface between the magnetic components and the circuitry of the \nsystem reflected not only on the problem of physical design, but on \nthe problem of electrical realization as well. Requirements on the \nindividual transformers constituting the hybrid trees had to be \ntightened to meet overall system performance criteria. As a result, \nthese magnetic components had to be built out with resistors and \ncapacitors to form what are, in effect, special types of transmission \nnetworks. Although this type of component is also used to a limited \nextent in the regulating and equalizing repeaters, a representative \ndesign is described in Section 2.3, Terminal Components. Finally, since \nmeasurement limitations and techniques were important to all classes \nof designs, these are discussed separately in Section 2.4.\n\nThe amplifiers in the basic and regulating line repeaters make \nextensive use of bridge-type feedback,\u201d? which offers the advantage \nof making the uf loop independent of the line impedance Ordinarily, \nthe disadvantages of bridge-type feedback are the extreme impedance \nlevels and the power consumed in the bridge arms used to secure a \nbridge balance. Both of these limitations, however, may be overcome \nby use of hybrid transformers. Because of the very tight limits, \ntypically +0.02 dB, placed on reproducibility over the 1- to 70-MHz \npassband, transmission-line design techniques** were extensively \nadapted to the design of hybrid transformers. These designs specify \npairs, triplets, or quadruplets of insulated magnet wire twisted uni- \nformly together and wound around a ferrite core. At low frequencies, \nthis arrangement behaves like a conventional transformer, but at high \nfrequencies, where parasitic elements predominate, the device behaves \nlike a transmission line. If the winding inductance and capacitance \nhave been properly proportioned, this approach results in greater \nbandwidth than can be obtained from a conventional transformer \ndesign.\n\nIn the case of a two-winding transmission-line transformer operating \nbetween equal impedances, the characteristic impedance of the line, \nZ. = VL/C, should be made equal to the terminating impedance. \nUnder these conditions, the transformer would theoretically have \ninfinite high-end bandwidth. However, the output hybrid of the basic- \nrepeater power amplifier was required to operate between an amplifier\n\nimpedance of 50 ohms and a line impedance of 75 ohms with the feed- \nback-port impedance equal to 150 ohms. Since the transmission line \nused to realize the hybrid could not simultaneously assume all these \ndifferent values, the best compromise was to make Z, equal to the geo- \nmetric mean of the output impedances, or 106 ohms. The ratios of the \nterminating impedances also forced the use of a trifilar winding. The \nimpedance properties of such windings were studied for various wire \ngauges, twist rates, insulation thicknesses, and materials. Ultimately, a \ntransmission line made from three strands of awe 40 polyurethane- \ninsulated wire, combined at a rate of 35 twists per inch, was selected. \nAlthough other combinations would have produced the same Z,, the \nhigh twist rate was used because it resulted in the best uniformity from \nmodel to model. The schematic diagram and transmission charac- \nteristics of the output hybrid are shown in Fig. 28. The divergence in \ninsertion-loss characteristics at high frequencies is the result of the \ncompromise value of impedance selection for Z,.\n\nAlthough transmission-line transformers were preferred from a \ncomponent-design standpoint, they could not always be used because \ntheir feedback-port impedance must be an integral multiple of both \nthe amplifier and line impedances. To overcome this constraint, \nconventional layer-wound hybrid transformers were used at the input\n\nFig. 28\u2014Schematic and insertion-loss characteristic of output hybrid for basic \nrepeater power amplifier.\n\nFig. 29\u2014Transformer structure for conventional layer-wound hybrids using D- \ncores.\n\nand output of the basic-repeater preamplifier. The initial estimates of \nthe impedances both transformers would be required to match were \n75 ohms at the line port, 60 ohms at the amplifier port, and a \u201clow\u201d \nimpedance at the feedback port. Various ratios were evaluated, and \nthe final ratio selected was 75:65 + 28 ohms, which results in a feed- \nback-port impedance of 19.6 ohms.\n\nTo provide the best consistency in a layer-wound transformer, a \nstructure that includes a winding bobbin is preferred. Bandwidth \nconsiderations, however, dictate that the structure would have to \napproach the efficiency of a toroid having a winding wrapped directly \non the core and extending around its entire periphery. Core and \nwinding dimensions, therefore, are nearly coincident. The D-core \nshown in Fig. 29 was chosen because the large area of the outer shell \nproduces a low magnetic reluctance. This means that the center-leg \nlength and area control the magnetic properties of the core. Since the \nwinding bobbin fits directly over the center leg, the winding dimensions \nare only slightly larger than the effective core dimensions, and the \nstructural efficiency approaches that of the toroid.\n\nEvaluation of amplifier models revealed that their high-frequency \ngains were extremely sensitive to parasitics in the transformer wind- \nings. Requirements specified a +0.02-dB reproducibility from model to \nmodel up to 70 MHz. Stringent winding procedures and in-process \nchecks were required to obtain this control in a conventional trans- \nformer. Wire gauges and pitches were selected to provide smooth, \neven, single-layer windings. Carefully controlled paper insulation was \nused between the most critical windings to insure that the parasitic \ncapacitance could be held to within +5 percent, which is three times \nmore stringent than normal manufacturing tolerances. Several different \ntypes of winding machines were evaluated to obtain one capable of \nproviding the controlled-pitch and constant-tension features required \nto maintain performance tolerances.\n\nOne of the more difficult designs required for terminal applications \nwas a 75:300-ohm, unbalanced-to-balanced transformer intended for \nuse in the Jmx modulators. The insertion loss of this transformer was \nto be held flat to within +0.1 dB from 0.5 to 70 MHz, and the balance \nof the center-tapped winding was to exceed 50 dB from 0.5 to 90 MHz. \nOriginally, separate designs were proposed for each jumbogroup \nbecause no core material was available that would permit simultaneous \nrealization of these two requirements. In addition, an ability to main- \ntain a balance of the required magnitude over this broad bandwidth \nhad not been demonstrated at these high frequencies. After carefully \nstudying the balance problem, however, it was felt that if an adequate \ncore material could be developed and the mounting structure rede- \nsigned, it would be possible to cover the entire range with a single \ndesign by employing carefully positioned toroidal windings. The \nbalanced winding consists of seven turns of a bonded pair of wires \nspaced evenly around the periphery of the core. These windings were \nthen connected in a series-aiding fashion to complete the center- \ntapped 300-ohm winding. The 75-ohm winding was spaced between \nadjacent bonded turns and all leads carefully dressed, resulting in a \nsufficient degree of structural symmetry to maintain the balance. \nSimultaneously, a new core having the necessary properties was \ndeveloped and the required mounting structure realized in a new \nphysical design.\n\nHybrid transformers, to provide combining and splitting functions \nat a 75-ohm impedance level, were required for several different \napplications in the L5 system such as Lpss-3, the E38 equalizer, line- \nconnecting arrangements, and smx circuits. Mounting and coaxial-\n\nconnector variations resulted in four distinct physical designs, as \nshown in the photographs of Fig. 30. The basic electrical building \nblock consists of a transmission-line hybrid and an autotransformer \ninterconnected as shown in Fig. 30a. With reference to that schematic, \nautotransformer Tl matches the 75-ohm input at jack J1 to the \n37.5-ohm impedance at the center tap of hybrid T2. It is wound with a \ntwisted-pair wire whose characteristic impedance Z, = V75 X 37.5 \n= 53.0 ohms to optimize performance. Since the impedances to be \nmatched differ by 1:2, the tap must be placed at a position correspond- \ning to 1:v2. This means that the windings cannot consist entirely of \ntwisted-wire transmission line, but must include some free turns. \nThese are created by partially decomposing the twisted bundle, with \nbest results obtained by minimizing the number of free turns.\n\nThe hybrid T2 has a 2:1 turns ratio and is wound with a twisted \npair having a characteristic impedance of 75 ohms. The network \nterminating resistance R has a value slightly greater than the ideal \n150 ohms to account for transformer core losses, and the capacitor C \nenhances the high-frequency response of the device. The electrical \nperformance achieved by these components over the 1- to 70-MHz \nfrequency range is as follows: (2) the transmission loss from input to \neither output is 3.10 + 0.05 dB, (72) the trans-hybrid loss between \nconjugate output ports is greater than 30 dB, and (777) the return loss \nat any port is greater than 26 dB. In addition, the two output ports \nof any individual hybrid track to within 0.02-dB loss and 0.5-degree \nphase shift. An ideal hybrid would have 3.01-dB transmission loss, \ninfinite trans-hybrid and return losses, and perfect flatness and track- \ning, with both outputs exactly in phase with each other and the input.\n\nIn addition to their use as individual hybrids, particular pairs of \nthese transformers may be grouped to form large n-port arrays or \n\u201c\u2018trees.\u201d\u201d The transformers in Fig. 30b use standard Bell System \ncoaxial plugs and jacks so arranged that they may be interconnected \ndirectly. The transformers in Fig. 30c use miniature coaxial connec- \ntors and may be formed into trees by cabling them together. To reduce \nthe number of interconnections required, the hybrid at the right of \nFig. 30c, a dual hybrid, has an input, one \u20143-dB output, and two \n\u20146-dB outputs. The hybrid at the left of Fig. 30c is a single hybrid \nof the type shown in Fig. 30a.\n\nThe +0.02-dB reproducibility of insertion loss required for many \nL5 magnetic-component designs could not be guaranteed initially by \ndirect measurements because neither the test set nor the apparatus\n\nSINGLE HYBRID SINGLE HYBRID \n(THREE-JACK DESIGN) (b) (TWO-JACK, ONE~PLUG DESIGN)\n\nFig. 30\u2014Hybrid transformers for L5 terminals. (a) Building-block schematic. \n(b) Hybrids with Bell System coaxial connectors. (c) Hybrids with miniature coaxial \nconnectors.\n\nunder test exhibited a return loss compatible with this extremely \ntight limit. Instead, a system was used in which a set of transformers \nknown to have the proper characteristics in an amplifier was used to \ncalibrate a test fixture. The calibrated fixture was then used to measure \nproduct. To guard against changes in the fixture or measurement \nsystem, however, these measurements were correlated to those made \non reference transformers whose histories were well known. When \neither fixtures or reference units needed replacement, the appropriate \nsteps in the calibration and correlation processes were repeated. \nMeanwhile, a program to provide improved measurement capability \nfor these devices was begun.\n\nTo have a reproducible base line for loss measurements on coTms,? \na simple strap between the transformer input and output ports of the \ntest fixture was used. While this introduces reflections because these \nports often have different impedances, it is the simplest and most \nconsistent method of eliminating differences in test sets, connectors, \nand cables. Furthermore, with sufficient padding close to the corms \nports of the fixture, satisfactory base-line reproducibility can be \nobtained. Accordingly, alternate measurement schemes were discarded \nin favor of upgrading the existing approach.\n\nThe test fixtures were originally constructed with ordinary Pwp\u2019s. \nThe contacts to the pin-type transformer terminals were made with \nminiature spring sockets, and the interconnection to the test set used \nBell System coaxial connectors. The resistive pads used to match the \ntransformer impedances to the coTms ports were made with various \ntypes of discrete resistors for the different transformer codes. It was \nrecognized very early that the required reproducibility of measure- \nments would be difficult to maintain in a production environment with \nthis type of test fixture. Coaxial connectors can introduce impedance \ndiscontinuities, and spring sockets can become contaminated or wear \nout with frequent use. When replaced, resistive pads are often damaged \nby the heat of soldering and change value. In addition, the associated \nparasitic electrical element values are a function of exact mechanical \nconfiguration and placement which are difficult to control.\n\nImproved designs of the spring contacts and of the coaxial connectors \nwere introduced to improve the performance of the test fixtures. \nKnife-edge contacts with force-free insertion were used to replace the \nformer, while the latter were superseded by precision 50-ohm versions. \nAs the next step, the PwB was replaced by an alumina substrate \nhaving controlled dielectric properties. Tests quickly indicated new \nproblems. Because of the higher dielectric constant of the substrate,\n\nthe effective electrical length of the connections increased significantly, \ncausing difficulty with crosstalk and reflections. The chip resistors, \nwhich were chosen for the matching pads, changed value on soldering \nand caused reproducibility and rework difficulties. A final design \nevolved in which the path lengths of the microstrip connections on \nthe substrate were reduced to a minimum and the resistors realized \nas thin-film elements. This arrangement permits high accuracy in the \nadjustment of resistance values, makes parasitic elements closely \nreproducible, and eliminates the repair operations that had been \nrequired with the original fixture design. External connection problems \nwere further reduced by coupling the coaxial connectors to the substrate \nvia precision spring-loaded microstrip launchers. These new fixtures, \nshown in the exploded view of Fig. 31, have attained a reproducibility \nof better than +0.01 dB.\n\nTransmission networks and magnetic components are among the \nleast visible components of the L5 system. Despite this lack of visi- \nbility, these components serve vital systems functions. Many of these \nfunctions are reviewed in this paper. The principal objectives of the \npaper, however, are to record those state-of-the-art advances that \nhave contributed to the success of the L5 system, and to identify those \nallied technologies that have influenced network and magnetic com- \nponent development. The influence of the computer in analysis and \nsynthesis, precision measurement, and optimization is stressed, and the \nadded influence of development in the piezoelectric device and mag- \nnetic materials areas has been noted. In many instances, developments \nin these allied technologies predated the L5 system and resulted from \na general philosophy of \u2018\u2018tool building\u201d initiated, specifically, to have \nthese technologies keep pace with future systems and component- \ndevelopment needs. Without such a philosophy, L5 requirements \ncould not have been met on schedule.\n\nMany individuals, too numerous to mention, have made significant \ncontributions to the development of transmission networks and \nmagnetic components for the L5 system. These contributions have \ncome from individuals in the authors\u2019 own departments and from \nco-workers at other Bell Laboratories locations. Special thanks, \nhowever, must be given to R. L. Adams and G. J. Mandeville for \ncontributing to the sections on computer-aided tuning of networks \nand the design of multibump, adjustable Bode equalizers, respectively.\n\n2. G. Szentirmai, \u2018\u2018A Filter Synthesis Program,\u201d in System Analysis by Digital \nComputer, ed. F. F. Kuo and J. F. Kaiser, New York: J. Wiley and Sons, \n1966, pp. 130-174.\n\n3. W. J. Geldart, G. D. Haynie, and R. G. Schleich, \u2018\u2018A 50-Hz to 250-MHz Com- \nputer Operated Transmission Measuring Set,\u201d B.S.T.J., 48, No. 5 (May-June \n1969), pp. 1839-1381.\n\n. R. A. Sykes, W. L. Smith, and W. J. Spencer, \u201cMonolithic Crystal Filters,\u201d \nIEEE International Convention Record, 1967, pp. 79-93.\n\n. W. G. Alberts, J. B. Evans, T. J. Haley, T. B. Merrick, and T. H. Simmonds, Jr., \n\u201cL4 System: Terminal Arrangements,\u201d B.S.T.J., 48, No. 4 (April 1969), pp. \n1024-1025.\n\n. J. L. Garrison, A. N. Georgiades, and H. A. Simpson, \u2018\u2018The Application of Mono- \nlithic Crystal Filters to Frequency Selective Networks,\u201d Digest of Technical \nPapers, 1970 International Symposium on Circuit Theory, 1970, pp. 177-178.\n\n7. H. A. Simpson, E. D. Finch, Jr., and R. K. Weeman, \u2018\u2018Composite Filter Struc- \ntures Incorporating MCFs and LC Networks,\u2019 Proceedings of the 25th \nAnnual Symposium on Frequency Control, Fort Monmouth, N.J.: U. 8. \nArmy Electronics Laboratories, 1971, pp. 287-290.\n\n. P. Lloyd, \u2018\u2018Monolithic Crystal Filters for Frequency Division Multiplex,\u201d ibid., \npp. 280-286.\n\n9. M. Dishal, \u2018\u201cTwo New Equations for the Design of Filters,\u2019\u2019 Electrical Communi- \ncation, 30, December 1952, pp. 324-337.\n\n10. A. I. Zverev, Handbook of Filter Synthesis, New York: J. Wiley and Sons, 1967, \nChapter 6.\n\n11. \u201cFilters, Modern Network Theory Design,\u2019\u2019 Reference Data for Radio Engineers, \ned. H. P. Westman, Fifth Edition, ITT, 1968, Chapter 8.\n\n12. J. L. Garrison and A. N. Georgiades, \u201c\u201cBand-Elimination Filters,\u2019\u2019 U. S. Patent \n3,704,433, November 28, 1972.\n\n13. A. A. Comparini, \u2018\u2018Analysis and Synthesis of a Band-elimination Filter Using \nMonolithic Crystal Filters,\u2019\u2019 unpublished memorandum.\n\n14. L. Brier, \u201cDer Entwurf von HF-Bandfiltern und Mechanischen Filtern mit \nDampfungspolen nach dem Betriebsparameterverfahren,\u2019\u2019 Guest Lecture at \noe International Science Colloquium, University of Ilnienan, September\n\n16. F. C. Kelcourse, W. G. Scheerer, and R. J. Wirtz, \u201cL4 System: Equalizing and \nMain Station Repeaters,\u2019 B.S.T.J., 48, No. 4 (April 1969), pp. 907-911.\n\n17. E. H. Angell, Y.-S. Cho, K. P. Kretsch, and M. M. Luniewicz, \u201cL5 System: \nRepeatered Line,\u201d B.S.T.J., this issue, pp. 1935-1985.\n\n18. W. R. Lundry, \u201cAttenuation and Delay Equalizers for Coaxial Lines,\u2019\u201d\u2019 Transac- \ntions of AIEE, 68, 1949, pp. 1174-1179.\n\n19. H. J. Carlin, \u2018\u2018The Scattering Matrix in Network Theory,\u201d IRE Transactions on \nCircuit Theory, CT-3, No. 2 (June 1956), pp. 88-97.\n\n20. J. L. Garrison, L. P. Labbe, and C. C. Roch, \u201cBasic and Regulating Repeaters,\u201d \nB.S.T.J., 48, No. 4 (April 1969), pp. 871-876.\n\n21. R. M.-M. Chen, C. F. Hempstead, Y. L. Kuo, M. L. Liou, R. P. Snicer, and \nKE. D. Walsh, \u201cL5 System: Role of Computing and Precision Measure- \nments,\u201d B.S.T.J., this issue, pp. 2249-2267.\n\n22, R. A. Thatch and F. D. Waldhauer, \u2018High Frequency Ground Isolation Filter \nfor ine Powered Repeater Circuits,\u2019\u2019 U.S. Patent No. 3,530,393, September \n22, 1970.\n\n23. H. W. Bode, Network Analysis and Feedback Amplifier Design, Princeton, N.J.: \nD. Van Nostrand Company, 1945, Chapter III, pp. 37-38.\n\n25. R. K. Bates and D. J. Zorn, \u201cL5 System: Signal Administration and Intercon- \nnection,\u201d B.S.T.J., this issue, pp. 2129-2145.\n\nThe L& Coaxial-Carrier Transmission System 1s the first long-haul, \nhigh-capacity transmission system for which the design was strongly \ninfluenced by extensive application of computer-aided design (cap) \ntechniques. A tight time schedule required a parallel effort of (1) improving \nand using the somewhat limited capabilities of existing caD programs \nand (12) developing new programs having increased capability and effi- \nciency. Similar development of computer-controlled measurement tech- \nniques provided necessary device and component characterization and \nsubsystem evaluation. The result is a powerful group of tools that are \nindependently important, but whose combined use helped make possible \nthe temely completion of the Ld system design. These tools, now \u2018proved \ntn,\u201d will profoundly tnfluence the next system design philosophy.\n\nAt the beginning of the L5 project in 1968, the set of computer- \naided design (cAD) tools and the specialized computer-operated trans- \nmission measurement set (corms) at the Merrimack Valley location of \nBell Laboratories were in an early stage of evolution. The tight system \ndevelopment schedule prevented the thorough, leisurely development \nof sophisticated programs that could provide the complete analysis and \ncharacterization needed for high confidence in design. Rather, a \nparallel effort was undertaken to augment existing design aids and to \ndevelop improved ones. The result was immediate answers generated \ninefficiently for early designs and a powerful group of programs that\n\nhave been \u2018\u2018proved in\u201d by use during the later stages of L5 develop- \nment. These were, however, not completed early enough in the design \ncycle to allow their full impact to be exerted on the overall design \nprocess in the manner that is now possible.\n\nThe following sections illustrate typical uses of computer programs \nfor seven phases of the design sequence. These include:\n\n(\u00a2) Small signal ac analysis of circuits having many nodes, in- \ncluding sensitivity and tolerance analyses. \n(iz) Optimization of element values in a circuit. \n(177) Statistical measurements of early production models of re- \npeaters for use in equalizer design. \n(tv) Nonlinear distortion analysis. \n(v) Component characterization. \n(vt) Overall system analysis. \n(vit) Equalizer characterization and alignment during manufac- \nturing.\n\nLimitations as well as successes are included to illustrate why and \nhow the evolution of programs took place.\n\nSeveral general-purpose circuit analysis programs,! as well as \noptimization? and Monte Carlo* programs, were in wide use in the \ncommunications industry by 1968. However, none was able to handle \na circuit of the size (over 80 nodes) and complexity of the L5 repeater. \nWhile the next generation programs were being developed, modifica- \ntion of the existing programs and segmentation of the circuit were \nnecessary to solve the immediate circuit analysis problems.\n\nSince the L5 system uses amplitude modulation, the repeater must \nbe a very linear device; this places strong emphasis on accurate, \nsmall-signal ac analysis of all circuits in the signal path. The basic \nrepeater amplifier alone contains 80 nodes, 111 passive components, \n8 transistors, and 4 wideband transformers. These numbers do not \ninclude \u201c\u2018parasitic\u2019\u2019 elements found to be essential for accurate modeling \nso that, in practice, the actual node and component count must be \nsignificantly increased.\n\nTo overcome the size problem, the circuit was at first partitioned \ninto blocks that could be analyzed with existing programs. One output \nof each of these analyses was a multiport admittance matrix charac- \nterizing the block at those nodes where it interconnects with other \ncircuit blocks. These block representations were stored in a library for \nsubsequent retrieval when the performance of a multiblock circuit was \nto be calculated. This approach of using a frequency-dependent matrix \nto represent subcircuits proved useful, not only for reducing node and \nelement count, but also for allowing the direct use of measured data \nfrom cotms to characterize components not easily or accurately \nmodeled by lumped element values.\n\nThe circuit segmentation approach also aided the calculation of \nsensitivities to element values. As each element in a given block was \nvaried to determine overall circuit sensitivity to that element, the \nmatrix representations of other blocks remained unchanged, saving \nmuch calculation time for the overall problem. Sensitivity data proved \nuseful in many ways. Since a full-scale Monte Carlo analysis of the \ncircuit was impossible during early development, a linear sensitivity \nmodel was used to set component tolerances based on their predicted \neffects on frequency response. The sensitivity data on parasitic element \nvalues also identified critical areas of the circuit where careful modeling \nand control were essential. It further identified those elements to be \nsubjected to optimization and smaller-scale Monte Carlo analysis.\n\nThe segmentation approach also permitted use of the early optimiza- \ntion and Monte Carlo programs.\u2018 These were particularly helpful in \nthe design of the feedback networks which, though small compared \nto the overall repeater, were critical in determining the frequency \nshaping of the repeater.\n\nAlthough the \u201cdivide and conquer\u201d method did allow some success \nin the early stages of analysis, several difficulties became apparent \nthat were corrected in the next generation programs. The many steps \ninvolved in the piecemeal approach led to analysis turnaround times \nof the order of one day for a complete circuit. This severely limited the \nnumber and type of \u2018\u2018what if\u2019? questions that could be raised and an- \nswered. Not only was the overall turnaround time relatively long, but \nall the segments had to be run in a batch mode, as no interactive \nfacilities were available for the designer to pose such \u2018\u2018what if\u2019\u2019 ques- \ntions quickly or effectively. These inefficiencies, coupled with the non- \nstatic nature of the actual circuit as various engineers wanted to\n\nimplement changes suggested by earlier analysis steps, made the \nturnaround time a serious problem. The multiple passes through \nvarious analysis segments also led to interfacing problems and in- \ncreased probability of coding errors.\n\nThe program developed to overcome most previous limitations is \ncalled circuit analysis program for efficient computer-optimized design \n(cAPE-coD). It: achieves its power and efficiency through a sparse \nmatrix technique and special, machine-generated coding. It has \nanalyzed circuits with 120 nodes, provides a wide variety of engineering- \noriented outputs, allows simple problem formulation by engineers, and \nprovides interactive capability on a graphics terminal.\n\nThe formulation technique chosen for cAPE-cop uses the tableau\u00ae \napproach. This method of arranging the system of equations that \ndefines the topology and behavior of a circuit not only yields an \nefficient analysis scheme, but also gives flexibility and simplicity to the \nprogram design while minimizing the restrictions placed on the circuits \nthat can be analyzed. For example, the following types of elements \nthat are seldom allowed in ac analysis programs are allowed in CAPE- \ncop: controlled voltage and current sources; zero-valued resistors, \ncapacitors, and inductors; and devices that have no Y or Z matrix \nrepresentations, but have an S matrix representation (e.g., an ideal \ntransformer or a three-port ideal circulator). By inserting zero-valued \ncomponents into the original circuit where one suspects parasitics to \nbe important, the program can then systematically investigate their \neffects without recoding, simply by varying the values from zero on \nsubsequent runs.\n\nThe elimination order for the tableau can produce, at one extreme, \nessentially a nodal formulation; at the other extreme, a mesh formula- \ntion; in general, a hybrid formulation results. This flexibility gives the \ntableau method the capability of producing more accurate results than \nother methods. Such accuracy improvements are important when an \nelement or subcircuit has a Y or Z matrix representation that is ill- \nconditioned, e.g., the Y matrix of a very small resistance.\n\nTo minimize core memory requirements of the computer and to \nallow analysis of large circuits having over 100 nodes, sparse matrix \nreduction techniques are used throughout the program. For efficiency \nin execution time, a loopless machine code program to solve the \ntableau system of equations is generated first for each problem. This \nefficiency is most important when the analysis package is used as a\n\nsubroutine in an iterative program such as optimization or tolerance \nanalysis, since the time spent generating the machine code is insignifi- \ncant compared to the time for repeated analyses. In addition to this \nmachine code, dynamic memory storage allocation and memory \npaging schemes are used to increase efficiency further.\n\nInput data for cAPE-cop can be entered in completely free format, \nwhich not only provides great convenience for the user, but also reduces \nthe number of coding errors in large circuits. The input \u201clanguage\u201d \nwas developed to be natural for an engineer, easily understood and \nremembered. The caPE-cop currently accepts the following circuit \nelements: resistors, capacitors, inductors, controlled or independent \nvoltage or current sources, two forms of transmission lines, two models \nfor transistors (one of which is used for distortion analysis, described \nlater), and N-port black-box devices. The latter can be characterized \nby directly measured Y, Z, or S parameter data, or by a user-supplied \nsubroutine that calculates these parameters. In addition to having \nfixed values, elements can be optionally defined as a tabular function \nof frequency. This black-box capability was essential to the L5 repeater \nanalysis, since no sufficiently accurate lumped element model existed \nto represent the wideband transformers used in the forward and feed- \nback paths. Instead, two-, three-, and four-port matrix representations \nwere formulated from frequency-dependent measured data\n\nPerhaps the most important improvement over previously used \nprograms which caPH-cop provides is the wide variety and flexibility \nin its computed outputs. These include any branch voltage or current, \nany node voltage, or any of the following transmission related quanti- \nties: insertion gain or loss; voltage, current, or power gain or loss; \nreturn loss or reflection coefficient; load impedance or admittance; \ndriving point impedance or admittance; u8 gain and yw impedances; \nand Y, Z, or S parameter matrix values.\n\nThe 8 gain, i.e., the loop gain of a feedback amplifier, is especially \nuseful for verifying stability in the amplifier. Since many of these \noutputs are complex numbers, they can be listed in any of the following \nforms: real and imaginary parts, magnitude and angle (in degrees or \nradians), dB[ = 20 logio(magnitude) |] and envelope delay (the deriva- \ntive of the angle with respect to angular frequency).\n\nSensitivity studies are facilitated by another feature of CAPE-cOD. \nAfter an analysis with nominal circuit values has been completed, it \nis possible to modify the values of any (one or several) element, sub- \nject to constraints if desired. A second analysis is then done, and the \noutput quantities can be the difference in any normal output that results\n\nfrom the modification of the circuit element values. This provides a \ndirect measure of circuit sensitivity to any change, large or small, in \nany element values.\n\nFor a system as complex as L5, no exact design procedure exists at \nany level, either device, circuit, or system. Many iterations are required \nfrom an initial to a \u2018\u2018final\u2019\u201d\u2019 design. Various tradeoffs are weighed, merit \ncriteria are assigned, and design parameters are adjusted to yield a \nresult that is optimum under given assumptions.\n\nThe crop is capable of minimizing a criterion function F(x) by \nadjusting the set of n parameters 21, %2, --+, %n (or in vector notation \nx) which may be bounded (above and/or below), unbounded, or fixed. \nA simplified program flow chart is shown in Fig. 1. Built into the \nprogram is a least-pth error criterion function that has the following \nform:\n\nyij(X) = y;(X, a:) is the jth response function value evaluated for \nthe ith independent variable a;. For example, y;, y2, and \nyz; may be loss, delay, and input impedance of a given \ncircuit, while x represents the circuit parameters and a; is \nthe ith frequency point. \nrij is the requirement of the jth response function for the zth \nindependent variable. If a range of requirements is re- \nquested instead of single value, and if y,;(x) falls outside \nthe range, the r;; in the formula is the value closer to y:;(x). \np is any positive even integer. \nW:; is the weighting constant of the jth response function at \nthe ith requirement point.\n\nINCREMENT COUNTER \nAPPLY STRATEGY TO \nFIND BETTER SET Xi \nCALL USER\u2019S SUBROUTINE \nTO EVALUATE F(Xi)\n\nIf the least-pth error criterion is not adequate for an application, \nthe apor program also accepts a user-defined criterion function in the \nform of a subroutine. A user-supplied subroutine must be given to \nevaluate y:;(x) for the above error criterion. This subroutine is often \nin the form of a complete ac circuit analysis done by another program, \nsuch as CAPE-copD, but now treated as a subroutine.\n\nA variety of strategies or algorithms exists for finding the optimum. \nSince the efficiency of these strategies is problem-dependent, it is \ndesirable to have many available from which the user (or program \nitself) can choose. The @roP is equipped with Fletcher-Powell,\u00ae steepest \ndescent,\u2019 Nelder-Mead,\u00b0 least-pth approximation\u00ae (least-squares when \np = 2), and many others. Stopping the search or switching to another \nalgorithm may be done either after a given number of iterations* or \nafter the percent improvement is less than a specified value for five \nconsecutive iterations.\n\nSince the above algorithms work only for unconstrained optimization \nproblems, the optimization of a bounded parameter problem (which is\n\n*In Gpop, an iteration corresponds to an improvement; i.e., in one iteration, the \nalgorithm searches until a better criterion value is found.\n\na simple form of constraint) is accomplished by transforming the \nbounded parameters into free parameters. A variety of transforma- \ntions have been built into the program. A user may also supply a \nspecific transformation for a particular application. The gradient of the \nresponse function can be computed either from a numerical difference \napproximation or from a user-supplied expression. Input and/or output \nsubroutines for pre- and/or postoptimization computation add another \nuseful capability. In some problems, it is necessary to apply parameter \nconstraints (such as interrelationships between parameters) inde- \npendently of any parameter bounds. The input or output routines in \nGpoP allow the user to apply sequential unconstrained optimization \ntechniques to constrained optimization problems.\u201d\n\nThe application of a@popr to a system design problem is given in a \nlater section; here, a circuit problem, the optimization of a fixed \ndeviation equalizer, is used for illustration. In practice, the fixed \nequalizer applies an average correction for the difference between \nrepeater gain shape and cable loss shape.\"'? Residual errors are then \ncorrected in the field, using adjustable E1-E2 equalizers. The criterion \nfunction does not (and should not) consider the error at the output \nof the fixed deviation equalizer, but rather the final error after all\n\nOVERALL EQUALIZATION COMPUTE ERROR \nREQUIREMENT \nAPPLY E1 \nDEVIATION EQUALIZER ZERO-\u2014FORCING DESIGN \nINITIAL DESIGN \nCOMPUTE MEAN SQUARE \nERROR AND ADJUST E1 \nCOMPUTE ERROR \nCHANGE DEVIATION APPLY E2 \nEQUALIZER PARAMETERS ZERO-\u2014FORCING DESIGN \nCOMPUTE TOTAL ERROR COMPUTE MEAN SQUARE \nTRANSFER TO GPOP ERROR AND ADJUST E2\n\nFig. 3\u2014Performance of fixed deviation equalizer before and after optimization.\n\nthree equalizations have been applied. Thus, the program must \nsimulate the field adjustment process before it can optimize the \nparameters of the fixed equalizer. Figure 2 shows the flow chart for \nthis process, and Fig. 3 shows the result of optimization. Four stages \nof bridged-T equalizers, having 32 elements, were needed in the final \ndesign for the excellent match to requirements.\n\nIV. STATISTICAL TRANSMISSION MEASUREMENTS OF REPEATERS \nFOR EQUALIZER DESIGN\n\nThe equalization hierarchy of the L5 system defined two levels of \naccuracy needed for transmission measurements on subsystems such \nas the basic repeater, regulating repeater, and equalizing repeater. \nThese levels must be kept in mind to appreciate the measurement \nproblems that had to be solved. First, the basic repeater provides a gain \nshape to compensate for cable loss (proportional to Vf), going from \n4.97 to 32.04 dB over the range from 1.6 to 66 MHz. The manufacturing \ngoal was that basic repeater gains would not deviate from target shape \nby more than +0.1 dB at the two-sigma points of the statistical dis- \ntribution. Thus, a transmission measurement accuracy of about 0.01 \ndB at each frequency was needed. The wide dynamic range (30 dB) \nand frequency range (nearly two decades) made this 0.01-dB target \ndifficult to achieve. Since measurements could be made on only a \nlimited number of basic repeaters, but the mean shape had to be \nextrapolated to apply to many repeaters in the system, the accuracy \nof individual measurements should, if possible, exceed the 0.01-dB\n\ntarget. Only a test set such as the cotms was able to provide this \ncapability with the speed needed for a thorough job. Typical runs \nincluded 100 discrete frequencies across the band and took one minute \nper run. .\n\nThe second level of measurement accuracy was needed for the re- \nmaining steps in the hierarchy, the regulating and equalizing repeaters. \nSince these provide mainly second- or third-order corrections, the \ndynamic ranges to be covered were relatively small, but accuracies \nneeded approached 0.001 dB over the same wide frequency range. \nNothing can be above suspicion at the 0.001-dB level (i.e., one part \nin 10). Mistermination or mismatch errors are the most frequent and \ninsidious problem. Over this frequency range, even the best available \npads needed to mask the test set impedances from those of the unknown \nshowed significant frequency shaping of both transmission and im- \npedance characteristics. Inexpensive mass-produced connectors, as \nused in field installations, have poor connection repeatability and im- \npedance in terms of these accuracy requirements. But the temptation \nis strong for the infrequent user to assume that an instrumental \nresolution of 0.001 dB will guarantee him an absolute accuracy of \n0.001 dB in any measurement. So regular accuracy verification of the \ntest equipment, use of selected highest quality test fixtures and masking \npads, and careful error analysis of each measurement was essential ; \nwhen these checks were omitted, serious errors were often discovered \nlater.\n\nA third complication arose from the need for detailed temperature \ncharacterization of the repeaters. The maximum temperature coefficient \nof 0.0016 dB per \u00b0F for the basic repeater is a value critical in the design \nof the dynamic equalizer and the regulating repeater. A computer- \ncontrolled environmental test chamber was used with cotms to ac- \ncumulate sufficient data to determine this. Continuous control and \nautomatic data logging minimized the operator interaction, but round- \nthe-clock measurements required months to complete. Long intercon- \nnecting cables and a network of multiplexing relays introduced serious \nline reflections, requiring special care in selecting and locating masking \npads. When the measurements were self-consistent, a standard sta- \ntistical analysis program in another computer completed the tempera- \nture characterization.\n\nNonlinear distortion is a serious problem for a long-haul, frequency- \ndivision-multiplexed, analog communications system using solid-state\n\ndevices. Most serious are the second- and third-order intermodulation \ndistortions, caused mainly by the transistors. Allowable distortions \nmust be far below fundamental signals in practice; for example, the \nthird-order distortion should not exceed a level 90 dB below funda- \nmental. Calculation of such small distortions requires not only very \naccurate transistor modeling, but also appropriate mathematical tools \nto extract such a small fraction of the main signal.\n\nCircuits having only a few transistors connected in cascade and a \nlinear feedback path have been analyzed by 8. Narayanan,\" using the \nVolterra series approach. While answers are obtained in closed form, \nthe computer implementation is not general and cannot handle the \nlarge complex circuits in an L5 repeater. So a new program was written \nto treat the second- and third-order distortions of transistor circuits \nunder small signal conditions. It is called the nonlinear distortion \nanalysis program (NODAP) and is based on the perturbation method.\n\nFor this type of problem, successful analysis depends on an accurate \nactive device model. The integral charge control model (icm) of \nGummel and Poon, which contains many nonlinear effects not in- \ncluded in conventional models, was chosen. Some more important \neffects include conductivity modulation, base push out, Early voltage, \nand avalanche effects. Expanding the 1cm equations around a given \nbias point and expressing terminal currents in terms of junction \nvoltages leads to an appropriate small-signal model. This is readily \nseparated into linear and nonlinear parts, the linear part being repre- \nsentable by a hybrid-7 equivalent circuit, or simply a two-port ad- \nmittance matrix Y(jw) (Fig. 4). Two controlled current sources, ty. \nand iy., represent the nonlinear part. These together represent the \nintrinsic transistor, to which must be added appropriate parasitic \nelements; the parasitics are treated linearly except for the inactive \ncollector capacitance.\n\nThe perturbation method works well (two iterations are sufficient), \nsince the nonlinearities are small for small-signal conditions. The \ncomputational algorithm!\u00ae uses any linear circuit analysis program \n(in this case, CAPE-coD). The result is an accurate, efficient, general- \npurpose program. The only nonlinear elements in the circuit are iy. \nand iy-, and they can be shown to be equivalent to two distortion cur- \nrent generators having amplitudes and phases determined by the \nsecond- and third-order nonlinear coefficients and the linear charac- \nteristics of the transistor circuit. This approach for a single transistor\n\ncan be readily extended to multitransistor circuits, since the distortion \ncomputation is carried out by a quasilinear approach and superposition \nholds.\n\nThe sequence of calculations in NopAP is interesting, showing how \nmajor programs can be combined. Given the circuit topology with the \nIcM parameters, bias point of each transistor, and sets of three funda- \nmental frequencies (fi, fe, and f;), NoDAP computes the linear and non- \nlinear coefficients for the transistor model. The linear characteristics \nare passed to cAPE-cop to calculate output voltages, load admittance, \nand junction voltages. These are returned to NopaP to calculate the \nsecond-order intermodulation current sources at fi + fo, fi + fs, and \nfe + fs. These sources go back to caPE-cop to calculate second-order \ndistortion voltages. This type of looping is similarly repeated for the \nthird-order distortion. The final step, by NopapP, prints out the cal- \nculated second- and third-order distortion indices Merz and Mz.\n\nA \u201clinear\u201d and \u201cnonlinear\u201d option in NopaP allows the \u201c\u2018turning off\u201d\u2019 \nor \u201cturning on\u201d of the distortion sources in any stage in an amplifier, \nto determine the contributions from that stage. The distortion effects \nof various nonlinearities can thus be isolated, giving insight impossible \nfrom direct measurements alone. Also, distortion vectors are computed, \nallowing the design of distortion cancellation circuits.\n\nThe calculated results for several L5 circuits have been in very good \nagreement with measured data, within 2 dB over a wide range of bias\n\nconditions and frequencies. Figure 5 shows a comparison of predictions \nand measurements over the L5 frequency band.\n\nAccurate device modeling is essential to the success of NODAP in \npredicting nonlinear distortion. A program called extraction of parame- \nters for the Icm of transistors (EPIC) has been developed to extract 30 \nIcM parameters from three sets of measured data. The raw data include \nthe de V-I characteristics, the junction capacitance characteristics, \nand the small-signal ac characteristics measured over a wide range of \nbias conditions. The previously described Gpor program is used to \nobtain the best Icm parameters by optimizing the match to the data. \nThose parameters associated with a single data set are matched first,\n\nprogressing to parameters that affect multiple data sets. This gives \naccurate results (physically realistic) quickly.\n\nThe main L5 transistor parameters were carefully extracted, using \nEpic. In addition to their usefulness in predicting distortion, the linear \nmodel parameters were used in cAPE-cop for normal ac circuit analysis, \ngiving excellent results. Agreement between computed and measured \nS parameters for the transistor was within 2 dB over the entire Ld \nfrequency range.\n\nAlmost all the electronic subassemblies are built with components \nthat are treated as classical \u201clumped elements\u201d such as resistors, \ncapacitors, inductors, or diodes, or as multiport devices such as trans- \nformers or transistors. Near the top L5 frequencies (66 MHz), the \nlumped element approximation is often inadequate and transmission \nline delays become important. Small capacitances and inductances, \nnormally considered \u2018\u201c\u2018parasitics,\u201d\u201d may have substantial effects in a \ncircuit, especially over the extreme bandwidths covered by feedback \namplifiers.\n\nSimple passive components, such as two-terminal impedances, were \nmeasured automatically over frequency ranges up to 1000 MHz. Not \nonly was the variation of element value with frequency obtained \n(caused by inadequate models), but derived properties were also ob- \ntained, such as the permeability and loss of magnetic materials, \u2018\u2018Q\u201d \nvariation with frequency, self-resonant frequencies of reactive com- \nponents, and linear two-element equivalent circuits of diodes. Such \nmeasurements were made quickly and very accurately on the coTms, \nusing an impedance calculating program. Impedances from milliohms \nto megohms can be measured over the range from audio to microwaves. \nAccuracies of the order of 0.1 percent are realized for real and im- \naginary components between 1 and 1000 ohms. A recent addition to \nthe measurement tools is a computer-operated impedance bridge \n(cozy) that provides even greater resolution (approaching 10 ppm) \nand impedance range, though at frequencies limited to below 30 MHz.\n\nThe characterization of transformers and transistors as multiport \ndevices was at a more sophisticated level. Basic measurements of \nvoltage scattering (S) parameters were made automatically on coTMs, \nusing a special program and calibration technique which compensates \nfor the mismatch errors caused by imperfect test set impedances.\u201d \nThese corrections are especially important since test set impedances \ndo not remain constant over wide frequency ranges. From the basic S\n\nparameters, other matrix parameters such as H or Y were obtained \nby computer data transformation. The H parameters are particularly \nuseful for transistors, while the Y parameters fit naturally into circuit \nanalysis programs based on nodal techniques. The multiport trans- \nformers used in the L5 amplifiers were characterized by merging several \nsets of two-port measurements into a 4-by-4 Y matrix representation.\n\nMany of these characterization jobs were not carried out until \ninexplicable problems appeared during the development of circuits. \nHowever, with the presently available tools that are quick, accurate, \nand easy to use, all critical component types can be characterized \nautomatically before use in model circuits. Such routine checks can \nprevent many problems and save considerable time in the long run. \nA component as simple as a resistor has many nanohenries of body and \nlead inductance that can play havoc in high-frequency circuits, yet \nit takes only a few minutes to characterize the component with modern \ntools.\n\nIn repeatered analog transmission system design, the optimal al- \nlocation of total system noise caused by various sources and subject \nto certain constraints is an important but tedious task, if the con- \nventional approach is used. Manual trial-and-error modification of \ntransmission levels and calculation of the resulting noise at various \nsystem points are time-consuming processes. All these are interde- \npendent functions of repeater gain, noise figure, modulation coefficients, \netc. A nonoptimum assignment may lead to failure to meet system \ndesign objectives and to possible redesign of basic elements, starting \na new system design cycle. Since the process requires optimizing a \nperformance function subject to a given set of system constraints, \nmodern optimization techniques are applicable.\n\nAn intermodulation noise computation program called NEWwMoD was \nwritten at Bell Laboratories in the 1960\u2019s for the development of sub- \nmarine cable systems. This program employs a modified version of \nBennett\u2019s product count method!\u00ae in a formulation that takes into \naccount the effect of the nonlinear phase characteristic of repeaters. \nThe NEWMOD program was used extensively in the development of the \nL5 system, in which the introduction of phase shaping networks effec- \ntively randomized the addition of third-order intermodulation prod-\n\nucts. This allowed the channel capacity of the system to be increased \nby about 20 percent for the same noise performance.\n\nThe product count method requires too much computer storage and \nprocessing time to allow the coupling of Newmop with erop for system \nnoise optimization. This difficulty can be avoided by modifying the \nearlier formulation so that a high-order Chebyshev numerical-integra- \ntion formula is employed to replace the product count approach. Based \non this new formulation, a program named noise optimization for \nrepeatered analog transmission systems (NORATS) has been imple- \nmented and is briefly described.\n\nThe main objective of this optimization is to select the best repeater \noutput transmission level as a function of frequency for the given \nconditions. For a noise-limited system such as L5, the total system \nnoise should be minimized. For an overload limited system, the \naverage power output of repeaters should be minimized so that the \ntotal system noise still satisfies the requirements. In both problems, we \nmay consider the modulation coefficients as additional design parame- \nters subject to certain limitations. All these problems can be considered \nspecial cases of minimizing the following criterion function\n\n2,-++,m\u2014 8, \nYm\u20142(X) = average value of the second-order modulation coefficient \n(in dB), \nYm\u20141(X) = average value of the third-order modulation coefficient \n(in dB), \nYm(X) = equivalent rms power of a single tone at repeater output \n(in dBm),\n\nr; and w; are the corresponding requirements and weighting constants, \np is a positive even integer, and the components of x are the coefficients \nof the polynomial representations of the transmission level and modula- \ntion coefficients as functions of frequency. The total system noise \nfunction in the equation is the power addition of thermal, second-order,\n\nSolutions to various design problems can be carried out using NORATS \nby choosing appropriate weighting constants in the criterion function. \nFor a noise-limited system, the weighting constant w, corresponding \nto the output power is set to zero, since this allows the optimum trans- \nmission level to be determined by minimization of total system noise, \nregardless of output power. A relatively large wn corresponds to an \noverload-limited system. The weighting constants wm_1 and/or Wm_2 \nare set to zero if the corresponding modulation coefficients are given \ninstead of being treated as design parameters. Recent NoRATS runs \nhave confirmed that earlier designs approached an optimum.\n\nThe strategy of using distributed corrections for the separate tem- \nperature variations of the cable and the repeaters (regulation), and for \ndeviations from nominal frequency shaping caused by manufacturing \ntolerances (equalization), is a complex one. The strategy depends on \naccurate characterization of the various equalizer and regulator cir- \ncuits to be effective, and this was considerably simplified by the use of \ncorms. One special program for alignment during manufacture deserves \ndescription. It illustrates that, in some problems, even the most \nsophisticated measurement equipment alone is not enough to fill the \nneed. By incorporating within the computer-controlled test set an \nalgorithm for adjusting a set of nonorthogonal controls, considerable \ntime and cost savings are achieved.\n\nThe 4211A network generates a Vf shape for use in regulating the \nL5 system against changes in cable loss with temperature.\u201d Six ad- \njustable elements, which are not independent, are needed to match the \nresulting shape to requirements. While their individual effects are \nreadily measured, the interdependence makes final adjustment during \nmanufacture nearly impossible by trial-and-error methods. The problem \ncould be solved on a separate computer, but this would require vol- \numinous data transfer and lost time. The logical time and place to \ncarry out the adjustment were simultaneously with the measurement \nprocess. The details of the procedure and its effectiveness are described \nin another article in this issue.\" This particular combination of pre- \ncision measurement, data reduction, computer prediction, and output\n\n\u201cinstructions\u201d to an operator, all done in a single machine, will be the \nway of the future for such complex tasks.\n\nThe complex design problems to be solved in the development of the \nL5 system led to the writing of a series of new computer programs \nthat are highly interrelated and yet very useful independently. carr- \ncop provides ac circuit analysis with flexible input and output, speed \nof computation, and the ability to handle large circuits with very few \nlimitations of types of elements. Continued development is under way \nto extend its range of application into the microwave region and to \nimprove its interactive and statistical capabilities. The Grop program \nimplements modern optimization techniques through a variety of \nalgorithms and considerable flexibility in types of constraints allowed. \nBoth epor and caPsE-cop contain links to simplify their combined use, \nmaking an extremely flexible combination. The NopAP introduces a \nnew method of handling nonlinear distortion problems, working to- \ngether with carE-cop. Input data for NopAP are provided by EPIc, a \ntransistor model parameter extraction program that works together \nwith cpop. A new system design program, NoRATs, also works with \nGPop to optimize the noise performance of repeatered analog trans- \nmission systems.\n\nThe usefulness of computer-operated transmission measurement \nequipment was enhanced by the development of advanced impedance \nmeasurement and modeling programs. Automatic measurement of \nmany repeaters under computer-controlled environmental conditions \nwas facilitated, along with data links to transfer measurement and \ncharacterization data directly to another computer where they could \nbe used in caPE-cop. A specialized alignment program was written for \nuse in COTMS.\n\nWhile each of these new tools is important in its own right, their \ncombination and coordinated development provided an extremely \npowerful set of tools. Clearly, the L5 project could not have been \ncompleted on schedule without these computer aids to design and \nmeasurement.\n\nMany individuals have contributed to the development and ap- \nplication of computer and measurement aids for the design of the L5 \nsystem. This paper covers only the key aspects of the total involve-\n\nment. The authors wish to thank all the other contributors for their \nenthusiastic support.\n\n1. M. A. Murray-Lasso, \u2018\u2018Analysis of Linear Integrated Circuits by Digital Com- \nputer Using Black-box Techniques,\u201d in G. Herskowitz, ed., Computer-Aided \nIntegrated Circuit Design, New York: McGraw-Hill, 1968.\n\n2. P. E. Fleischer, \u201cOptimization Techniques in System Design,\u201d in F. F. Kuo \nand J. F. Kaiser, eds., System Analysis by Digital Computer, New York: \nJohn Wiley, 1966.\n\n. L. A. O\u2019Neill, \u201cInteractive Tolerance Analysis with Graphic Displays,\u2019\u2019 Proc. \nSpring Joint Computer Conference, 1969, pp. 207-213.\n\n. C. L. Semmelman, E. D. Walsh, and G. T. Daryanani, \u201cLinear Circuits and \nStatistical Design,\u201d B.S.T.J., 60, No. 4 (April 1971), pp. 1149-1172.\n\n. G. D. Hachtel, R. K. Brayton, and F. G. Gustavson, \u2018The Sparse Tableau \nApproach to Network Analysis and Design,\u2019 IEEE Transactions on Circuit \nTheory, CT-18, January 1971, pp. 101-113.\n\n. R. Fletcher and M. J. D. Powell, \u201cA Rapidly Convergent Descent Method for \nMinimization,\u2019\u2019 Computer Journal, 6, June 1963, pp. 163-168.\n\n. R. Fletcher, \u201cA Review of Methods for Unconstrained Optimization,\u201d R. \nFletcher, ed., Optimization, New York: Academic Press, 1969.\n\n. J. A. Nelder and R. Mead, \u2018\u2018A Simplex Method for Function Minimization,\u201d \nComputer Journal, 7, January 1965, pp. 308-313.\n\n10. A. B. Fiacco and B. P. McCormick, \u2018\u201c\u2018Programming Under Nonlinear Constraints \nby Unconstrained Minimization: A Primal-dual Method,\u2019 Tech. Paper \nRAC-TP-96, Research Analysis Corporation, Bethesda, Md., September 1963.\n\n11. E. H. Angell, Y. S. Cho, K. P. Kretsch, and M. M. Luniewicz, \u2018\u201cL5 System: \nRepeatered Line,\u201d B.S.T.J., this issue, pp. 1935-1985.\n\n12. J. L. Garrison, A. Olsen, Jr., and T. H. Simmonds, Jr., \u2018\u2018L5 System: Transmission \nNetworks and Magnetic Components,\u201d B.S.T.J., this issue, pp. 2203-2248.\n\n13. W. J. Geldart, G. D. Haynie, and R. G. Schleich, \u2018A 50-Hz to 250-MHz Com- \nputer-operated Transmission Measuring Set,\u2019 B.S.T.J., 48, No. 5 (May- \nJune 1969), pp. 1839-1381.\n\n14. S. Narayanan, \u201cApplications of Volterra Series to Intermodulation Distortion \nAnalysis of Transistor Feedback Amplifiers,\u2019\u201d\u201d IEEE Transactions on Circuit \nTheory, CT-17, November 1970, pp. 518-527.\n\n15. H. C. Poon, \u2018Modeling of Bipolar Transistor Using Integral Charge-control \nModel with Application to Third-order Distortion Studies,\u201d IEEE Trans- \nactions on Electron Devices, HD-19, June 1972, pp. 719-731.\n\n16. Y. L. Kuo, \u2018Distortion Analysis of Bipolar Transistor Circuits,\u2019 IEEE Trans- \nactions on Circuit Theory, (7-20, November 1973, pp. 709-716.\n\n17. J. G. Evans, \u2018Measuring Frequency Characteristics of Linear Two-port Net- \nworks Automatically,\u201d B.S.T.J., 48, No. 5 (May-June 1969), pp. 1813-1338.\n\n18. W. R. Bennett, \u2018Cross Modulation Requirements on Multichannel Amplifiers \nBelow Overload,\u201d B.S.T.J., 19, No. 4 (October 1940), pp. 587-610.\n\nW. G. Albert, Bell Laboratories, 1951\u2014. Mr. Albert was initially \nengaged in the physical design of terminal equipment for the L3 \nCoaxial System and microwave transmission systems. He also worked \non the physical design and development of the A4 and A5 channel \nbanks, L multiplex, mastergroup multiplex, and L4 and L5 coaxial \nsystem line repeaters. He is now supervisor of the Reliability and \nDigital Banks Physical Design Group. Member, IEEE.\n\nRobert E. Anderson, B.S.E.E., 1940, University of Wisconsin; E. I. \nduPont de Nemours and Co., 1940-48; Radio Research Laboratory, \nHarvard University, 1948-45; Bell Laboratories, 1945\u2014. At Bell \nLaboratories, Mr. Anderson has worked on the development of video \ntransmission systems for television, the L3 carrier, automatic-restoral \nalarm circuits for N2 and N38 carrier, the equalizer remote control \nsystem for L4 carrier, and most recently, on the transmission surveil- \nlance system for the Ld carrier system. Member, IEEE, Eta Kappa Nu.\n\nE. H. Angell, B.S.E.E., 1964, Union College; M.S. (E.E.), 1965, \nHarvard University; Bell Laboratories, 1964\u2014. Mr. Angell has been \nengaged in circuit design work for coaxial systems. He is currently \nsupervisor of the L5 repeatered line group. Member, JEEE.\n\nJohn F. Barry, B.S., 1961, M.S., 1963 (Electrical Engineering), \nNortheastern University; Bell Laboratories, 1961-1972; Western \nElectric, 1972\u2014. Mr. Barry initially worked on the development of the \nBellboy\u00ae personal signaling receiver, the 3A Fm terminals, and the \n2A 1F restoration switch system. Later he worked on the development \nof the jumbogroup frequency supply. Since 1972, he has been a member \nof a special design group.\n\nR. K. Bates, AT&T Long Lines, 1941-1953; Bell Laboratories, \n1953\u2014. At Bell Laboratories, Mr. Bates has worked on private-line \ntelegraph switching systems including the 82B1 system for the U.S. \nNavy and the Delta Airlines reservation system. In long-haul carrier \nwork, Mr. Bates has worked on the L4 system and, most recently, on \nthe basic jumbogroup trunk bay for the L5 system.\n\nPhilip J. Baun, Jr., B.S.E.E., 1959, University of Wisconsin; \nM.S.E.E., 1961, Northeastern University; Bell Laboratories, 1959\u2014. \nMr. Baun, a member of the Coaxial Systems Department, is responsible \nfor the L5 transmission surveillance software. He has had previous \nexperience in the design and development of passive networks and in \nthe analysis and characterization of electronic circuits for toll trans- \nmission systems via analog and digital computation. Member, IEEE, \nTau Beta Pi, Eta Kappa Nu.\n\nRichard Ming-Ming Chen, Dipl. El. Eng., 1959, Chiao Tung Uni- \nversity, China; M.S., 1965, Pratt Institute; Ph.D., 1968, Rutgers \nUniversity; Faculty Member, El. Eng. Dept., 1959-1962, Peking \nConstructional Industrial College, Peking, China; Bell Laboratories, \n1968\u2014. Mr. Chen has been engaged in the areas of optimization, \nsimulation, tolerance assignment, and partitioning techniques for \nlarge scale problems. Member, IEEE, Eta Kappa Nu, Sigma Xi.\n\nYo-Sung Cho, B.S.E.E., 1962, Seoul (Korea) National University ; \nM.S., 1966, and Ph.D., 1968, Yale University; Honeywell E.D.P. \nDivision, 1964-1965 and 1967-1969; Bell Laboratories, 1969\u2014. \nMr. Cho made equalization studies of the L5 coaxial transmission \nsystem employing manual and automatic equalizers. He was also \nengaged in the development of the equalizer adjustment system \nwhich is used for the equalization of the L5 line. His subsequent \nwork includes design of exploratory repeater amplifiers for a future \ncoaxial transmission system. He is presently supervisor of the group \ndeveloping terminal multiplexing equipment for coaxial and radio \ntransmission systems. Member, IEEE.\n\nFred A. D\u2019Altroy, B.A., 1949, M.A., 1951, University of British \nColumbia; Ph.D., 1956, Purdue University ; Bell Laboratories, 1955\u2014. \nMr. D\u2019Altroy has been engaged in the development of semiconductor \ndevices since 1956. He is currently supervisor of a group having \nresponsibility for the Triac pnpn devices, pnp transistors, and npn \nultralinear transistor.\n\nAlbert F. Flint, A. E., 1962, Wentworth Institute; Bell Telephone \nLaboratories, 1962\u2014. Mr. Flint has worked on the design and develop- \nment of precision crystal oscillators. Since 1971, he has worked mainly \non the development of clock oscillators for several systems.\n\nJ. L. Garrison, B.E.E., 19384, and M.E.E., 1936, Polytechnic \nInstitute of Brooklyn; Bell Laboratories, 1936\u2014. Mr. Garrison has \nworked on the design of transmission transformers, on the final de- \nvelopment of transistors, and on technical publications. He now \nsupervises a group engaged in development of transmission networks. \nMember, Sigma Xi, Tau Beta Pi, registered professional engineer in \nNew Jersey and New Hampshire.\n\nJohn H. Green, B.S.E.E., 1966, and M.S.E.E. 1968, Northeastern \nUniversity ; Bell Laboratories, 1966\u2014. Mr. Green has been involved \nin various terminal system developments for L-carrier systems in- \ncluding mastergroup multiplex and line protection switching systems. \nHe is currently involved in digital channel bank development. Member, \nTau Beta Pi.\n\nB. H. Hamilton, B.S.E.E., 1949, University of Kansas; Bell Labora- \ntories, 1950\u2014. Mr. Hamilton has worked on development of power \nconversion and control circuits, new energy conversion techniques, \nand medium- to high-voltage dc-to-de converters. He is presently \nsupervisor of a circuit-development group. Senior member, IEEE, \nmember, Kappa Eta Kappa, Sigma Tau, Tau Beta Pi.\n\nCharles F. Hempstead, B.S., 1949, Northwestern University; \nPh.D. (Physics), 1955, Cornell University; Bell Laboratories, 1954\u2014. \nMr. Hempstead designed millimeter wavelength backward-wave oscil- \nlators and studied solids for maser applications until 1961. He was then \nconcerned with Type II superconductors for high magnetic field \ngeneration, followed by research on visual perception of motion. \nPresently he supervises a group developing new applications of wide- \nband computer-controlled test equipment for precise component \ncharacterization. Member, IEEE, Sigma Xi, Phi Beta Kappa.\n\nFred J. Herr, B.S.E.E., 1942, Cooper Union; M.S., 1952, Stevens \nInstitute of Technology; Bell Laboratories, 1936\u2014. At Bell Labora- \ntories, Mr. Herr was first engaged in the development of measuring \nequipment for coaxial transmission systems. He was later concerned \nwith system design analysis of long-haul coaxial and type-N short-haul \ncarrier systems. He participated in early feasibility experiments on \ncolor television transmission on coaxial systems and worked on the \ndesign specification of new video measuring equipment. He participated \nin the laying of the Alaskan and second transatlantic submarine cable\n\nsystems and did the system design analysis and terminal maintenance \nplanning for the sp submarine cable system. Currently, he supervises \nthe test equipment planning and application group for coaxial trans- \nmission systems. Member, Tau Beta Pi.\n\nRichard M. Jacobs, B.S. (Chemistry), 1954, Brooklyn College; \nB.S.E.E., 1959, University of Wisconsin; M.S.E.E., 1961, Lehigh \nUniversity; Bell Laboratories, 1959\u2014. Mr. Jacobs has been engaged \nin the development of transistors and integrated circuits since 1959. \nHe is currently head of a department responsible for the development \nof discrete devices and integrated circuits.\n\nFrank C. Kelcourse, B.S.E.E., 1959, M.S.E.E., 1962, Northeastern \nUniversity ; Bell Laboratories, 1959\u2014. Mr. Kelcourse worked initially \non Fp terminals, including the design of transmission amplifiers, \nmodulators, and carrier supplies, and subsequently on the design of \nwideband feedback amplifiers for the L4 line repeaters. He has super- \nvised groups responsible for the final development of the L4 equalizing \nand remote control systems and for the equalization, planning, and \ndesign of the L5 system. He is currently supervisor of a systems \nstudies and applications group responsible for analyses and appli- \ncations related to analog, digital, and hybrid transmission systems. \nMember, Tau Beta Pi.\n\nKenneth P. Kretsch, B.S.E.E., 1959, Pennsylvania State University ; \nM.E.E., 1961, New York University; Bell Laboratories, 1959\u2014. Mr. \nKretsch began at Bell Laboratories in the switching research area, \nparticipating in research into time-division switching systems. He \nalso participated in the development of a message switching system and \nwas responsible for system design of high-speed processors for use in \ntelephone switching systems. Currently, he is responsible for equaliza- \ntion of the L5 coaxial cable system. Member, IEEE.\n\nY. L. Kuo, M.S. (E.E.), 1961, Oklahoma State University; Ph.D. \n(E.E.), 1966, University of California; Assistant Professor, 1966-1970, \nPurdue University; Bell Laboratories, 1970\u2014. Mr. Kuo\u2019s primary \ninterest is in the area of active device modeling and computer-aided \nanalysis of nonlinear networks.\n\nM. L. Liou, B.S., 1956, National Taiwan University; M.S., 1961, \nDrexel Institute of Technology; Ph.D., 1964, Stanford University ;\n\nBell Laboratories, 1963\u2014. Mr. Liou is presently supervisor of the \nAnalysis and Interactive Computing Group providing analytical and \ncomputational support to the transmission system development at \nMerrimack Valley. His fields of interest have included system theory, \nnumerical analysis, optimization, and computer-aided design of circuits \nand various components in radio and cable transmission systems. \nMember, IEEE, Eta Kappa Nu, Sigma Xi.\n\nMichael M. Luniewicz, B.S.E.E., 1958, University of Massachusetts ; \nM.S. (engineering), 1961, Northeastern University ; Bell Laboratories, \n1960\u2014. Mr. Luniewicz has been engaged in circuit design and develop- \nment for multiplex terminals and coaxial lines.\n\nRobert E. Maurer, B.S.E.E., 1962, M.8.E.E., 1964, Ph.D., 1968, \nNortheastern University; Bell Laboratories, 1962\u2014. In addition to \nsupervising a group responsible for the development of analog multi- \nplex equipment, Mr. Maurer has worked on the design of the equalizing \nrepeater for the L4 system, the analysis and modeling of intermodu- \nlation distortion, the equalization of random channels, and analysis \nand exploratory development related to the transmission of high-speed \ndigital signals over sharply band-limited analog channels. He presently \nsupervises a system-planning group working on a new baseband digital \nsystem for exchange area application. Associate Editor, JEEE Trans- \nactions on Communications. Member, Tau Beta Pi, Eta Kappa Nu, \nPhi Kappa Phi, Sigma Xi.\n\nSamuel Mottel, B.S.M.E., 1950, City College of New York; \nM.S.M.E., 1968, Newark College of Engineering; Bell Laboratories, \n1952\u2014. Mr. Mottel has been concerned with physical design of power \nequipment. He has worked on power for carrier systems, microwave \nsystems, submarine cables, key telephones, ringing and tone plants. \nHe supervises a group responsible for physical design of a variety of \npower equipment.\n\nJoseph M. Nacci, B.S. (Physics), 1956, University of Rhode Island ; \nBell Laboratories, 1956\u2014. Mr. Nacci has been active in the design \nand development of a wide variety of silicon transistors and integrated \ncircuits. These include pnp transistors of all types, npn ultralinear \ntransistors, silicon based capacitors, and pnp integrated circuits.\n\nSundaram Narayanan, B. Tech., 1960, Indian Institute of Technol- \nogy, Kharagpur, India; M.S., 1963, and Ph.D. (Electrical Engineer- \ning), 1965, Carnegie-Mellon University ; Bell Laboratories, 1965\u2014. Mr. \nNarayanan has worked on nonlinear distortion in transistor amplifiers \nand the use of high-speed digital signals over band-limited analog \nchannels. He was supervisor of a group developing precision signal \nsource and is presently supervisor of a group working on a new multi- \nplex arrangement for L5 and a basic repeater design for an advanced \ncoaxial system. Member, IEEE, Sigma Xi.\n\nJames F. Oberst, B.E.E., 1964, Manhattan College; M.S., 1966, \nand Ph.D. (Electrical Engineering), 1969, Polytechnic Institute of \nBrooklyn; Assistant Professor of Electrical Engineering, Polytechnic \nInstitute of Brooklyn, 1968-1969; Bell Laboratories, 1969-\u2014. Since \njoining Bell Laboratories, Mr. Oberst has worked on various aspects \nof pcm transmission over cable and synchronization for FpM terminal \nequipment. He is presently working on pcm channel banks.\n\nArthur Olsen, Jr., B.S.E.E., 1959, Worcester Polytechnic Institute ; \nM.S.E.E., 1961, Northeastern University; Bell Laboratories, 1959\u2014. \nMr. Olsen has been responsible for the design and development of \ntransmission networks and magnetic components. He _ presently \nsupervises a magnetic components group.\n\nEdward J. Panner, B.S.E.E., 1962, Lafayette College; Bell Labora- \ntories, 1949\u2014. Mr. Panner has been engaged in device development \nprincipally for transmission systems involving technologies ranging \nfrom klystrons and general-purpose tubes to transistors and integrated \ncircuits. Member, Tau Beta Pi, Eta Kappa Nu, Phi Beta Kappa.\n\nHenry S. Pustarfi, A.E., 1955, Newark College of Engineering; \nBell Laboratories, 1951\u2014. Mr. Pustarfi has worked on the development \nof quartz crystal filters, temperature-control circuits, and thermoelec- \ntric ovens. He is presently engaged in the development of crystal- \ncontrolled oscillators, temperature-control devices, and precision fre- \nquency standards.\n\nRichard W. Sanders, B.S.E.E., 1959, University of Vermont; \nM.S.E.E., 1961, Northeastern University; Bell Laboratories, 1959- \n1972; Western Electric Company, 1972\u2014. At Bell Laboratories, Mr.\n\nSanders was engaged in the development of coaxial transmission sys- \ntems, specifically protection switching for the L4 and L5 Coaxial- \nCarrier Transmission Systems. At Western Electric, he is presently \ninvolved in test engineering for the voice-band interface frame and the \ndigroup terminal for the No. 4 Ess project. Member, IEEE, Tau \nBeta Pi.\n\nT. H. Simmonds, Jr., B.S.E.E., 1954, University of Virginia; \nM.S.E.E., 1961, Northeastern University; Active Naval Reserve, \n1954-1958; Bell Laboratories, 1954 and 1958\u2014. Mr. Simmonds\u2019 \nearly work was on a variety of filters and networks for long- and \nshort-haul carrier transmission systems. As supervisor of a networks \ngroup in the Transmission Systems Networks Department he is \nresponsible for work on transmission filters and networks for carrier \nand radio transmission systems. Member, IEEE, Tau Beta Pi.\n\nRobert P. Snicer, B.S.E.E., 1966, and M.S.E.E., 1967, Massachu- \nsetts Institute of Technology; Bell Laboratories, 1967\u2014. Since joining \nBell Laboratories, Mr. Snicer has been engaged in transmission system \ndevelopment and computer-aided design. He presently supervises a \nnetwork design group. Member, IEEE, Eta Kappa Nu, Tau Beta Pi, \nSigma Xi.\n\nJohn L. Thomas, B.S.E.E., 1957, University of Maine; M.E.E., \n1960, New York University; Bell Laboratories, 1957\u2014. Mr. Thomas \nengaged initially in circuit design work associated with special applica- \ntions of submarine cable systems. He worked on systems analysis and \nsupervised a group responsible for the circuit design of shore terminal \ntransmission facilities associated with the sF submarine cable system. \nHe later supervised a group responsible for repeater, equalizer, and \nspecial test set circuit design for submarine cable systems. He is \npresently responsible for the design of transmission surveillance and \nfault location circuitry for the L5 coaxial system. Member, Phi Kappa \nPhi, Tau Beta Pi.\n\nEdward D. Walsh, B.S.E.E., 1965, Gannon College; M. Eng., \n1966, and Ph.D., 1968, Rensselaer Polytechnic Institute; Bell Labora- \ntories, 1968\u2014. Mr. Walsh engaged initially in the frequency domain \ncharacterization of high-frequency active and passive devices. He has\n\ndeveloped a general-purpose frequency domain simulation program \nfor transmission circuits. He is presently working on Monte Carlo \nsimulation programs for transmission circuits. Member, IEEE.\n\nR. J. Wirtz, B.S. (M.E.), 1950, Brown University ; Bell Laboratories, \n1956\u2014. Mr. Wirtz was at first involved in resistor development and the \nphysical design of the N38 Carrier System. He later supervised the \nphysical design of the L4 Coaxial System. He is currently supervisor of \nthe Coaxial System Physical Design Group responsible for the design of \nlong-haul carrier systems.\n\nDonald J. Zorn, A.A. (electronic engineering), Wentworth Institute, \n1959; B.S. (industrial technology), Northeastern University, 1965; \nBell Laboratories, 1959\u2014. Mr. Zorn has worked in the development \nof A5 channel banks, L-multiplex, mmx-multiplex, and the L4 coaxial \nsystem, and is currently working on the L5 coaxial system. In addition, \nhe has worked on several special development projects, including a \ncarrier supply for channel banks used in the initial Telstar operations \nand a special pilot supply for the Norap headquarters carrier system. \nMember, Sigma Epsilon Rho.\n\nTHE BELL SYSTEM TECHNICAL JOURNAL is abstracted or indexed by Abstract \nJournal in Earthquake Engineering. Applied Mechanics Review, Applied Science \n& Technology Index, Chemical Abstracts, Computer Abstracts, Computer & Con- \ntrol Abstracts, Current Papers in Electrical & Electronic Engineering, Current Papers \non Computers & Control, Electrical & Electronic Abstracts, Electronics & Communi- \ncations Abstracts Journal, The Engineering Index, International Aerospace Ab- \nstracts, Journal of Current Laser Abstracts, Language and Language Behavior Ab- \nstracts, Mathematical Reviews, Metals Abstracts, Science Abstracts, and Solid State \nAbstracts Journal. Reproductions of the Journal by years are available in micro- \nform from University Microfilms, 300 N. Zeeb Road, Ann Arbor, Michigan 48106.", "title": "magazine :: Bell System Technical Journal :: BSTJ V53N10 197412", "trim_reasons": [], "year": 1974}
{"archive_ref": "bitsavers_BellSystemJV60N02198102_7638344", "canonical_url": "https://archive.org/details/bitsavers_BellSystemJV60N02198102_7638344", "char_count": 288600, "collection": "archive-org-bell-labs", "doc_id": 676, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc676", "record_count": 477, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bitsavers_BellSystemJV60N02198102_7638344", "split": "test", "text": "R. J. Canniff A Digital Concentrator for the SLC\u2122 -96 System 121 \nJ. F. Reiser Compiling Three-Address Code for C Programs 159 \nA. S. Acampora Rain Margin Improvement Using Resource Sharing in 167\n\nD. E. PROCKNOW, President, Western Electric Company \nIl. M. ROSS, President, Bell Telephone Laboratories, Incorporated\n\nG. E. SCHINDLER, JR., Editor \nPIERCE WHEELER, Associate Editor \nJEAN G. CHEE, Assistant Editor\n\nTHE BELL SYSTEM TECHNICAL JOURNAL is published monthly, except for the \nMay-June and July-August combined issues, by the American Telephone and \nTelegraph Company, C. L. Brown, Chairman and Chief Executive Officer; W. M. \nEllinghaus, President; V. A. Dwyer, Vice President and Treasurer; F. A. Hutson, Jr., \nSecretary. Editorial inquiries should be addressed to the Editor, The Bell System \nTechnical Journal, Bell Laboratories, 600 Mountain Ave., Murray Hill, N.J. 07974. \nChecks for subscriptions should be made payable to The Bell System Technical \nJournal and should be addressed to Bell Laboratories, Circulation Group, Whippany \nRoad, Whippany, N.J. 07981. Subscriptions $20.00 per year; single copies $2.00 \neach. Foreign postage $1.00 per year; 15 cents per copy. Printed in U.S.A. Second- \nclass postage paid at New Providence, New Jersey 07974 and additional mailing \noffices.\n\nThe SLC\u2122-96 subscriber loop carrier system is a digital subscriber \ncarrier system serving up to 96 single-party customers. The system \ncan be configured with an optional plug-in which digitally concen- \ntrates two standard, 1.544 megabit, serial pulse-code modulation \n(pcm) bit streams into a single stream, thereby concentrating 48 \ncustomers onto a single T1 digital line. The concentrator employs a \ncustom NMOS LSI chip providing a full-access, time-slot interchange \nfunction. It has microcomputer controllers at the two ends of the \nsystem to control time-slot assignments. The development of the \nconcentrator involved challenges in chip design, software design, and \nperformance testing.\n\nAs the state of the art in electronics has advanced, so have the \ninroads of electronics into the loop plant as a supplement to or \nreplacement for cable.\u2019 The application of pair-gain systems to loops \nhas been of particular importance in recent years. The SLC\u2122-96 \nsubscriber loop carrier systems is the latest in a line of digital loop \ncarrier systems to provide improved features and reduced cost.\u201d The \nSLC-96 system serves up to 96 single-party subscribers between a \nCentral Office Terminal (cot) and Remote Terminal (RT) using stan- \ndard pulse-code modulation (PCM) coding over facilities such as T1. \nSystem features include a variety of channel units, channel and drop \ntesting provisions, and spare digital line switching. System versatility \nis further enhanced by three modes of operation (each mode uses one \nadditional T1 line for protection):\n\nThis article discusses the Time Assignment Unit (TAU) which is the \nconcentrator employed in Mode II operation. In a fully equipped Mode \nII system there are actually two identical, but independent, concentra- \ntors, each concentrating 48 lines onto 24 time-slots of a T1 line, and \nthereby eliminating two main T1 lines (see Fig. 1). In many cases the \nsaved T1 lines provide more advantage than is apparent. The benefits \narise in applications limited by apparatus case size for holding repea- \nters, by small cables where the purpose of pair-gain systems is to avoid \nadding new cable, and, similarly, in situations that use several systems \nin parallel along the same route where the number of saved pairs can \nbe very significant. For long-distance systems, of course, the saved \nlines result in important cost savings.\n\nSection II of this article discusses the traffic-handling ability of the \nTAU and the traffic administration provisions. This is followed by a \nfunctional description of the concentrator operation in Section III. \nSection IV discusses first the hardware features provided by the \ncustom Time-Slot Interchange (Ts1) chip and then enumerates the \nfeatures provided by software. Selected implementation details for \nboth hardware and software are expanded upon in Section V. Perform- \nance testing, that is, the equipment and resources that were used to \nadequately stress the concentrator to confirm its design, is discussed \nin Section VI.\n\nThe concentrator performs a full-access, digital time-slot inter- \nchange function. The very conservative two-to-one concentration ratio \nthat was chosen provides for the inclusion of a limited number of \ndedicated (unconcentrated) special-service circuits as well as a large \nnumber of multiparty circuits. It is important to note that a single \nchannel unit for special service occupies the same physical space as a \nstandard unit for dual voice frequency and thus reduces the number of \nlines to be concentrated by two. Also, the special-service unit requires \nonly a single T1 time slot. Thus, regardless of how many special-service \nunits are inserted, the concentration ratio remains fixed at two to one, \nbut the group size is reduced. The blocking probability for concentrated \nlines increases as more special-service units are added, because of the \nreduced group size. By design, the number of special-service channel \nunits is limited to eight and must be physically inserted in the last four \nchannel unit positions of each shelf.\n\nOFFICE \u00b0 \nSWITCHING \nNETWORK DIGITAL DIGITAL * \nCONCENTRATOR CONCENTRATOR\n\nFor heavy loading or mandatory monitoring purposes, six traffic \nadministration aids are provided:\n\n(z) A two-digit numeric display is provided on the faceplate of the \nCOT TAU which displays, upon demand, the peak traffic in hundred call \nseconds per hour (ccs), per concentrated T1 time slot since the internal \ntraffic register was last cleared. The register is cleared by means of a \npin jack on the faceplate.\n\n(ti) The same display will also indicate the number of blocked calls, \nup to a maximum of 15, since the internal blocked calls register was \nlast cleared. This register is cleared by means of the pin jack.\n\n(uit) A traffic lamp is provided on the faceplate, in conjunction with \na minor alarm, that indicates there has been two or more blocked calls \nfor two out of three weeks running. This alarm must be manually \ncleared through the pin jack.\n\n(tv) A relay closure is provided to remote to an electronic switching \nsystem (Ess) office that all T1 time slots are in use. The office can \nthen divert terminating calls and provide its own reorder tone while \nkeeping individual line blockage statistics. (In a non-Ess Central Office \nthis diversion will not occur and the TAU will provide a digital reorder \ntone.)\n\n(v) Asecond relay is provided that outpulses peak weekly traffic in \nCCS, once per week, to a remote traffic-monitoring register at the rate \nof one pulse per second.\n\n(vi) A third relay is provided that outpulses the number of blocked \ncalls as they occur to a remote blocked calls monitoring register at the \nrate of 1 pulse per second. There is no saturation limit here as there \nwas in the faceplate display.\n\nThe following definitions are given to clarify all discussion that \nfollows: A line refers to a subscriber loop at the RT or a wire pair \nappearance at the cot. A channel unit is the physical plug-in serving \none or two lines and providing the per-line circuit functions. A channel \nis the electrical path from a channel unit to the Transmit Receive Unit \n(TRU) serving it, and the time slot reserved for the line into and out of \nthe TRU on the 1.544 megabit serial Pcm busses. Channels enter or \nleave the concentrator as time slots from or to the TRUs. Time slots on \nthe T1 line interfacing to the concentrator are referred to as trunks, \nbecause they provide a limited number of shared paths between the \ntwo terminals of the system, in analogy with traditional trunking \nfacilities between central offices.\n\nFigure 2 shows a simplified system block diagram of the TAus. The \nmicrocomputer controllers are realized using the Bell Laboratories\n\ndesigned MAC-8 microprocessor.\u00ae The MAC-8 microcomputers main- \ntain control over the Time-Slot Interchange (TsI) chips and talk to \neach other over a data-link channel derived from the subframe bits on \nthe T1 line. The transmit TsI selectively combines two 1.544 megabit, \nserial pcm bit streams, one from each of the TRUs it serves. (A TRU \nperforms the A/D, D/A, and framing tasks for 24 channels.) The \nresulting T1 signal is sent to the receive TsI at the other end of the \nsystem where it is expanded into two streams, for sending to corre- \nsponding TRus. The TsIs provide full access, meaning that any one of \nthe incoming (outgoing) 48 lines or time slots has access to any one of \nthe 24 outgoing (incoming) trunks, where trunks are time slots on the \nT1 line. The cot Tau handles trunk assignments and, in general, \ncontrols the concentrator. The RT TAU acts more as a slave. When the \nMAC-8 controller assigns a line to a specific trunk, that line will keep \nthe trunk for the duration of the call and in no way inhibits any of the \nother lines from accessing any of the other trunks.\n\nA line-service request, called \u201cactivity,\u201d is picked up by the transmit \nTsI by accessing the A and B bit signaling busses on the system \nbackplane. A and B bits are the standard nomenclature for per-channel \nsignaling that indicates off-hook, ringing, etc. The activity is stored in \nmemory in the Ts1 from which it can be retrieved by the MAC-8 \nthrough the TsI microcomputer address and data ports. Activity at the \nRT is passed over the data link to the coT where the line/trunk \nassignments are determined.\n\nThe transmit and receive TSIs must be synchronized to the TRUS \nthey serve and appropriate signals are provided for this purpose. No \nsynchronization is assumed between transmit and receive TsIs. Fur- \nther, no synchronization is assumed between the MAC-8, the TsIs, and \nthe data link. The TsIs are accessed by means of a handshake proce- \ndure. The data-link frame signals are polled to determine when mes- \nSage processing is needed.\n\nIf all 24 trunks are busy, provision is needed for feeding a fast-busy \n(overflow) tone to the cot channel units. This is done digitally through \nthe receive TSI, resulting in significant cost savings in the channel \nunits. The receive TsI allows the assignment of up to 24 lines to \u201cbusy \ntrunks\u201d whereby the selected lines receive a fast-busy tone in PCM \nform as read from a code-word table stored in the MAC-8 program \nmemory. Signaling information (A and B bits) is also stored in these \ncodes and thereby allows the channel unit to recognize that it is getting \nthe fast-busy tone and accordingly trip ringing without charge, prior \nto applying the tone on the blocked customer\u2019s line.\n\nFigure 3 shows a simplified schematic for the coT TAU. The RT TAU \nis nearly identical with the basic exceptions that Output Port 2 is \nremoved and there is no Read Only Memory 2 (RoM2). The Random \nAccess Memory (RAM), Read Only Memory (rom), Input Port, and \nOutput Ports connect to the MAC-8 bus as in any normal microcom- \nputer. The custom TsI chips were also designed to connect directly on \nthe bus and appear to the MAC-8 as programmable peripheral chips. \nThe MAC-8 talks to the Ts1 chips by means of a handshake procedure\n\nDATA LINK FRAME SANITY MONO \nPCM PCM TRUNK TRUNK PCM PCM BANK LOOP BACK \nPCM PCM OUT\u2014OF\u2014F RAME\n\nusing the MAC-8 Data Ready line. The clock frequency for the TsI \nand MAC-8 is 1.544 MHz. Three to nine wait states result, as required \nand generated by the TsI on a read or write. The Ts1 chip generates \nand responds to the necessary signals for system synchronization.\n\nOutput Port 1 provides four alarm light emitting diodes (LEDs) on \nthe faceplate and three system alarms. For the CoT TAU, the LEDs \nindicate alarms for coT, RT, traffic, and special-service channels. The \nRT TAU has only two LEDs indicating RT alarm and unit alarm for \nspecial-service channels. The system alarms are minor, major, and \nmajor with channel unit disable. Output Port 1 also provides an output \nfor strobing the sanity monostable, a timeout device that checks on \nproper sequencing of the program. Output Port 2 provides the dual- \ndigit numeric display for indicating traffic and blocked calls on the \nfaceplate and also services the three relays used for remoting concen- \ntrator status. The Input Port provides for the display selection switches \nand the internal register-clearing pin jack, all mounted on the faceplate. \nIt further provides inputs for the data-link frame signals which are \npolled to determine data-link message requests. Also, the Input Port \nallows accessing a receiver out-of-frame signal and a bank loop-back \nsignal.\n\nThe coT TAU has 4K bytes of program memory and the RT TAU has \n2K bytes. Each are realized using 2K byte RoM chips. At both coT and \nRT, 1K bytes of RAM are provided, though the coT uses less than one- \nthird of the available memory and the RT uses only about one-fourth. \nThe TAU plug-ins are printed circuit cards measuring approximately 4 \ninches by 10 inches as pictured in Fig. 4. It was required that the TAUs \nbe sized to physically replace the Line Interface Units for the T1 lines \nthat are not needed in the concentrated mode. Power supplies for the \nTAU are 5V and 12V, with typical dissipation being about 3 watts. \nSpecial requirements had to be met for the RT TAU so that it could \nwork over the temperature range of \u201440 to +85 degrees Celsius.\n\nThe Ts1 chip was designed to be universal, in the sense that it is \npackage lead and microcomputer programmable for use at COT or RT, \nfor either the transmit or receive function. In making a universal chip \nwith all the features mentioned below, it is possible, and very desirable, \nto reuse pieces of the hardware inside the chip. Thus, for example, \npieces of RAM and other hardware used for busy trunk assignments in \nthe receive TSI mode are alternatively used in the transmit TsI mode \nfor the activity and \u201cTNEN\u201d\u2019 collection (discussed later). Similarly, \nmany input and output package leads serve dual purposes. In addition \nto the basic features, several very important additional features are\n\nprovided by the Ts1 chip. Most of these features cost very little \nhardware, that is, they increase chip complexity very little, while \nproviding useful features.\n\nFigure 5 is a block diagram of the TsI chip. The heart of the chip is \na RAM used for PcM data, trunk assignment information, busy trunk- \nassignment information, activity data, \u201cTNEN\u201d data, per-line test bits, \n1 byte of fast-busy tone buffer, and 2 bytes of RoM. Incoming and \noutgoing serial pcm data is handled in bytes inside the chip by feeding \nthe data through serial-to-parallel registers (Registers 1 and 2) and \nparallel-to-serial registers (Registers 2 and 3). The use of Register 2 \ndepends on whether the TsI is performing in the transmit or receive \nmode. The frame bit is normally stored in the FR FF (frame flip-flop). \nThe main-control logic provides the signals for controlling all the chip\u2019s \nregisters, multiplexers, and RAM. Multiplexers are included for selecting \naddress and data for the RAM and for selecting output onto the \nmicrocomputer data bus.\n\nThe time-slot interchange is executed by the method of writing data \ninto a RAM sequentially and reading it selectively for the transmit \nconcentrator function and vice versa for the receive function. The \ntrunk assignments (24 bytes), which are addresses of the desired PCM \nline data, are stored in the RAM, yet are used to address the RAM\n\nselectively by a feedback register (Register 4) to the address bus. Each \nof the trunk (busy trunk) assignments are fed back in turn to address \nthe desired line memory. In the transmit mode, if a trunk is unassigned, \nthe ROM locations are accessed to send idle code on the T1 line. In the \nreceive mode, if a line is unassigned, idle code from the PCM memory \nis read out for the line or idle is forced by means of a cleared \u2018\u2018enable \nbit\u201d (discussed later).\n\nSince the concentrator is positioned between the TRUs and the T1 \nline, A and B bit-signaling information is already contained in the PcM \nbit streams when they are received by the TsI. It is necessary, therefore, \nthat the TsI insure integrity of signaling frames. Signaling frames are \nthe 6th and 12th frames of a 12-frame sequence and contain the per- \nchannel A and B bit signaling information in the least significant bit \nof the pcm code words. Integrity is maintained by having two RAM \nsections of pcm data (48 bytes each). While writing (reading) in one \nsection sequentially, the other section is read (written) selectively; the \nroles are reversed every frame. The frame bit is correspondingly \ndelayed to match up with the outgoing data. The result is that \nsignaling-frame integrity is maintained while the data experiences a \nfixed delay of approximately two frames (250 microseconds) end-to- \nend regardless of the line or trunk.\n\nThe microcomputer has access to the TsI memory through the \naddress and data ports of the TsI chip (see Fig. 5). This allows all the \nmemory locations, including the pcm data memory, to be written and \nread. The main control block of the TsI contains a frame counter for \ncontrolling all chip sequencing. When the counter is in a state not \nneeded for a specific internal function, that clock cycle can be used to \nrespond to a microcomputer read or write request. The result is a \nvariable number of wait states, as mentioned earlier, because the \nmicroprocessor can request a TSI access at an arbitrary time. The \nmicrocomputer talks to the TsI by means of Read/Write, Select, and \nBusy leads that connect to the microcomputer interface circuitry. On \na read operation the data is held in Register 6 so that the processor \ncan use as much time as necessary to recognize the data. Some of the \naddress space of the TsI is reserved for addressing the data-link register \n(Register 7) and activity-mode control register (Register 8).\n\nThe Ts1 must be synchronized to the TRUs to know where every bit \nin the incoming and outgoing bit streams is located. For the transmit \nfunction, the TsI puts out a superframe synchronization signal which \nthe TRUS can accept and lock to. For the receive function the TsI\n\naccepts an out-of-frame and synchronization signal from the TRU. The \nreceive function is complicated, however, by the reframing process. \n- Since the TsI exists on the T1 line side of the TRU, the reframing \nexecuted by the TRU is carried out after the bit stream has passed \nthrough the Ts1. This implies the need for special modes of operation \nwhile the system is out-of-frame to assure reframe.\n\nA feature of the TsI in the receive mode is the ability to feed a \ndigitally generated, fast-busy tone to a terminating connection when \nall trunks are busy. In order to receive the fast-busy tone, a line must \nbe assigned to a \u201cbusy trunk.\u201d The assignment mechanism is nearly \nidentical to assigning a line to a normal trunk. The assignment consists \nof an address of the line location into which the fast-busy tone bytes \nare to be placed. These assignments are fed back in turn through \nRegister 4 to select the desired lines. Up to 24 lines can receive the \ntone byte simultaneously. The source of the current PcM tone byte is \na holding register (Register 5) which is indirectly updated every frame \nby the microcomputer, through the fast-busy tone buffer byte in RAM. \nTwo signaling codes (A=B=0 or A=B=1) can be sent out with the \ntone byte by making the least significant bit always 0 or 1. Only one of \nthese codes is used in the TAU for signaling the channel units. The \nfast-busy tone is simulated by a sequence of 48 pcm coded bytes stored \nin the program ROM which emulates the dual-tone frequency needed.\n\nOne very important side feature of the TsI in the transmit mode is \nthat it gathers A and B bit information to supply the microcomputer \nwith line activity information. The simplest and fastest way to collect \nthis data is to tap into the A and B bit busses on the system backplane. \nThe A and B bit data are available there every frame and, because the \nTSI and the TRU are synchronized, the precise time for each line\u2019s A \nand B bit data is known.\n\nSince the concentrator is interested only in activity and not 1n the \nprecise A and B bit signaling states, the A and B information is \ncondensed. No activity (idle) is signaled by the channel units as A = \nB = 0 or A = B = 1, depending on whether the location is cor or RT. \nThus, the A and B bit collection hardware just looks for a deviation \nfrom the idle pattern. It is further desirable to have an elementary \nfiltering effect so that if there is any activity within a certain time \nperiod during which the chip is told to collect activity, that activity is \ncaught and held, with the result that the microcomputer is not required \nto make numerous, closely time-spaced searches for activity.\n\nsift for zeroes, sift for ones, or no collection. Since the A and B bit \ninformation is combined, all 48 lines of activity information are stored \nin 48 bits or six bytes of the RAM. The collection mode control is set by \nthe microcomputer in Register 8 (see Fig. 5). Selector 2 and Register \n5 are used to collect each byte of data before being transferred to the \nRAM for storage.\n\nAnother important side feature of the TsI chip is that it gathers the \nso-called \u201cTNEN\u201d bits. The per-channel TNEN bit tells the TRU to \nencode PcM data for the channel or to send digital data off a backplane \nbus. Since two channels are associated with a single physical channel \nunit, the two corresponding TNEN bits can be used as an indication of \nthe class of service desired by a channel unit. Thus, four different \ntypes of channel unit can be identified based on the permutations of \nthe two TNEN bits, and the concentrator can then take the required \naction. For example, standard dual voice-frequency units get concen- \ntrated service while single voice special and single data special units \nget a permanent trunk.\n\nThe Ts1 chip in the transmit mode collects TNEN bits in almost the \nsame way it collects activity (using Registers 5 and 8 and Selector 2), \nbecause the TNEN bits are available on the backplane bus in the same \nformat as the A and B bits. The only difference is that there is only \none TNEN bit per line and no sifting is performed, that is, they are just \ncollected. The TsI cannot collect activity and TNEN simultaneously \nsince they use common hardware. The TNEN bits are stored in different _ \nmemory bytes in the RAM, however; thus, by means of a fourth \ncollection mode (in addition to the three activity modes), the micro- \ncomputer can get a snapshot of the TNEN bits between normal activity \ncollection. This can be done in just a little more than one frame of \ntime so that no significant activity is lost.\n\nto an idle channel. Since the enable bit feature does not inhibit the \nline from being assigned to a trunk, it allows, with the aid of the \nmicrocomputers at both coT and RT, PCM test codes to circulate around \nthe entire connection loop prior to customer cut through. This checks \napproximately 90 percent of the hardware involved in a connection \nand, as implemented, adds only about 15 ms to the connect delay time. \nThis feature employs pairs of enable bits as they are read out to \nRegister 4 in conjunction with the trunk assignment for that time slot. \nWhile in Register 4, the enable bits can cancel writing data to the RAM \nfrom Register 1 (Transmit Mode) or force Registers 2 or 3 to be loaded \nwith idle code (Receive Mode) for the particular lines that correspond \nto the time slot in question.\n\nAnother auxiliary feature built into the TsI to relieve real-time \nconstraints on the processor is an 11-bit data-link shift register (Reg- \nister 7). The data link, as seen by the concentrator, consists of 11-bit \npackets of data, every 9 ms, in a serial format. The data-link register \nis loaded or unloaded in parallel by means of the microcomputer \naddress and data ports, after which it is shifted asynchronously by the \ndata-link clock. The microcomputer polls the data-link frame signals \nseparately to determine when to read or write the shift register.\n\nA final feature of the TSI is a power-up and initialization sequence. \nBy means of an external Rc network, a latch internal to the Ts! is set \nupon power up. Then after the clock starts, a 12-frame sequence must \nbe passed through before the chip comes out of initialization. This dual \nmethod of Rc timeout and clock timeout assures a robust initialization \nsequence that assures all memory is initialized and all trunks and busy \ntrunks are deassigned. The initialization sequence is also very useful \nfor manufacturing testing, as is an additional lead that allows breaking \nup the main-control counter sequence.\n\nThe software for the Tau MAC-8s was developed on a UNIxX\u2122 time- \nsharing-system (see Ref. 4). A Bell Laboratories microprocessor de- \nvelopment tool for the MAC-8, called PLAID, was used for debugging \nand testing the code.\u00b0\n\nThe cot and RT programs are written in MAC-8 assembly language \nand are designed to fit into the available 4K bytes and 2K bytes of \nROM, respectively. Assembly language was used, not only because of \nlimited program capacity, but also because of stringent real-time \nconstraints, which exist in part because of the decision to keep hard-\n\nInitialization and alarm filtering (8%) \nTraffic and blocked calls recording 17% \nChannel unit identification 5% \nOther . . 3%\n\nware at a minimum where software can do the job. For this particular \napplication, these decisions are still justified after the fact. It is also \ntrue that understandable code can be written easily in MAC-8 assem- \nbly because of its C-like syntax, thereby negating some normal aversion \nto assembly-level programming.\u2019 A characteristic of the code is that \nit is very heavy in register instructions, as would be expected for byte \nand time-efficiency reasons. (The MAC-8 employs 16 general-purpose \nregisters residing in RAM.) Careful attention was given to register usage \nso that data required over long segments of a routine, or even between \nsubroutines, could remain as register variables.\n\nTable I shows a usage breakdown of the coT TAU code. The funda- \nmental job of call processing and message handling represents only 38 \npercent of the code. The importance of self-diagnostics is obvious and \nreflects the concern and effort that was expended in this area.\n\nTable II shows a breakdown of the code in terms of routines, \ninstructions, and bytes. A large percentage of the code runs in response \nto interrupts generated by data-link message requests. For example, \nall trunk assignments and deassignments initiated by the COT TAU are \ntriggered by the need to form a new data-link message for transmittal \nto the RT. A total of 3894 bytes (95 percent) of the available 4096 bytes \nare used. Table III lists the bytes of RAM required by the coT TAU \nMAC-8 program. Only 286 bytes (28 percent) of the available 1024 \nbytes are used.\n\nProcessing Processing Total \nRoutines 28 6 34 \nInstructions 923 418 1341 \nCode (bytes) 2644 1199 3843 \nBytes/instruction 2.86 2.87 2.87 \nPercent of code 69% 31% 100% \nData table (bytes) 49 2 51\n\nProgram variables 64 bytes \nLine/trunk data base 90 bytes \nStack allowance 100 bytes \nMAC-8 registers 32 bytes \nTotal 286 bytes\n\nThe COT TAU program can be envisioned as a Main (background) \nroutine that runs when no other processing is needed and interrupt \nroutines that respond to real-time call-processing requests.\n\nThe Main routine performs most of the sanity and memory-checking \ntests. Another job is to manipulate the \u201cTNEN\u201d bits into masks used to \nforce trunk assignments or no assignments in response to the types of \nchannel units that are inserted in the SLC\u2122-96 bank. Several alarm \nfilters are maintained; the Main routine examines these filters and \noutputs the correct system and faceplate alarms. The traffic and \nblocked-call calculations are also performed by the Main routine. The \nresults are passed to an interrupt routine, which times the output for \ndisplaying on the cot TAU faceplate and for outpulsing via the relays. \nThe Main routine also performs some housekeeping chores.\n\nAn interrupt from the receive TSI occurs every frame (125 us). \nBecause of the digitally generated, fast-busy tone feature, whereby \nPCM code words are read from the processor ROM to the receive TSI, \nthis high-speed interrupt is needed. This interrupt is then counted \ndown to time other functions for the interrupt routines (see Fig. 6). \nEvery 2 ms the Data-Link Polling and pcm Test routine (Poll routine) \nis entered. This routine polls the data-link frame signals to determine \nwhen a data-link message must be transmitted or received. As a result, \nthe Transmit and Receive Message routines are executed every 9 ms. \nThe Pc test portion of the Poll routine refers to a function performed \nat the time of assigning a line to a trunk. Prior to customer cut- \nthrough, test PcM codes are circulated coT to RT to coT. The transfer \nof the test codes from receive TsI to transmit TSI is done by the \nprocessor by sampling at the 2 ms rate. Finally, the two-digit numeric \ndisplay on the cot Tau faceplate is multiplexed at 6 ms intervals by \nthe Display Mux routine.\n\nAs mentioned previously, the concentrator data link consists of 11- \nbit framed messages. These 11-bit messages are grouped together in a \nprotocol providing error protection by means of redundancy. All mes- \nsages except \u201cIdle\u201d are 33-bit messages made up of three sequential \npackets of 11 bits. For digital central-office compatibility, normal\n\nmessages communicate change-in-state information. Provision is made \nfor updating of assignments and activity as required.\n\nThe coT transmits a trunk/line assignment (deassignment) by three \nidentical, sequential 11-bit submessages. The rT looks for a two out of \nthree match to respond, thus providing error protection. The RT \nsimilarly sends activity as three identical submessages. Update infor- \nmation is needed to periodically assure that the COT and RT TAUs data \nbases are in agreement. Updates are sent as a header plus the message \nand its complement. This biases these messages toward getting \nthrough correctly, or not getting through at all, a desirable condition \nfor update messages. The cor sends assignment updates whenever it \ndesires and at the request of the RT by an \u201cAssignment Update \nRequest\u201d message. The RT sends activity updates only at the request \nof the cot by means of an \u201cActivity Update Request\u201d message. A \n\u201cLooping Test\u201d message is a periodic message initiated by the coT to \ntest the continuity of the data link. A \u201cNo Alarm\u201d message is sent by \nthe RT periodically as a fail-safe way of sending an alarm message to \nthe cot. Care had to be taken in selecting the code words for the \nmessages to assure that message boundaries could be determined, and \nalso to provide error protection across message boundaries since all \nmessages butt end-to-end.\n\nActivity filtering is provided in both the coT and RT, in addition to \nthe rudimentary filtering provided by the TsI chip. Two bit-up/down \ncounters are employed as filters using unequal attack-decay, with \nvariable thresholds providing hysteresis. These filters provide noise \nimmunity and delay for bridging over dial pulsing, switch-hook flashes, \nand silent intervals of ringing so that trunks are not deassigned and \nreassigned during these intervals. These filters negate the need for \nanalog filters on the channel units to perform this function.\n\nThe cot controls trunk (busy trunk) assignments and deassignments \nbased on the coT and RT activity. As mentioned earlier, a connection \ntest is done when a line is assigned to a trunk, whereby pcm test codes \nare circulated coT to RT to COT prior to cut-through of the line. If no \ntrunks are available, the line is assigned to a busy trunk at the cor \nonly and the line then receives the digitally generated fast-busy tone. \nIn the case of a blocked call caused by activity from the RT, that call \nis transferred to a trunk when it becomes available. Thus, in this case, \nan RT customer would experience delayed dial tone.\n\nTo assure that all pieces of the data base relating to line and trunk \nstatus are correctly correlated, consistency routines have necessarily \nbeen implemented. If a conflict arises (for example, a line is assigned \nto two trunks), corrective action is taken. Such conflicts only arise \nbecause of glitches or memory faults, but must be guarded against. In \nfact, detecting and reacting to \u201csoft\u201d and \u201chard\u201d errors was one of the \nmost challenging aspects of the software work.\n\nAnother related aspect was the requirement that the processor be \nable to recover from an arbitrary R/W memory state, because it is a \nstand-alone computer. This required careful consideration and thor- \nough testing to determine that, for instance, the processor would not \n\u201chang\u201d if a flag bit accidentally flipped. The ability to detect and react \nsanely to a genuine fault is also related to these problems. Verifying \nthat software works correctly under the above-mentioned conditions \nis difficult. The TAU programs were tested by observing the reaction to \nrandom data in the processor RAM, and also by forcing bit faults by \nmeans of special hardware. These tests were excellent in pointing up \nseveral software bugs.\n\nThe TAv software is designed to be very tolerant of R/w memory \nfaults because approximately 65 percent of the R/w memory that is\n\nused (including TSI) is dedicated to per-line functions. Being able to \nisolate a fault to a single line allows the system to continue operation \nwith a minor alarm condition. The effect is.equivalent to a reduced \n\u201csystem crash\u201d failure rate for the TAU plug-in. It is estimated that the \nequivalent TAU lifetime will be increased by approximately 40 percent \nbecause of the fault-responding software. That is, this software would \noften allow the unit to be replaced before it caused a system crash. \nThis software represents only about a 10 percent overhead in code (cf. \nTable I).\n\nAnother software function is processing the \u201cTNEN\u201d information. By \nassociating two TNEN bits, the type of channel unit plugged into a \nparticular physical slot can be determined. The transmit TsI picks up \nthis information and stores it in its own memory from which the \nprocessor can obtain it. The result of the processing is essentially two \nmasks. One mask, when combined with the A and B signaling bit \nactivity, generates permanent activity. The other mask forces no \nactivity. Thus a special-service unit plugged into the correct physical \nslot can be given permanent service or, if plugged into an illegal slot, \ncan be denied service and the condition alarmed. The channel-unit \ninformation is also used to condition traffic calculations, since traffic \n(in ccs) applies only to concentrated trunks.\n\nAs mentioned previously, a function of the cor software is to \ncalculate and store information related to traffic and blocked calls. A \nbasic software consideration is that some of this information is held \nfor very long periods. The faceplate traffic alarm is based on two or \nmore blocked calls for two out of three weeks running. The traffic and \nblocked calls displayed on the faceplate are stored indefinitely. Thus, \nit was necessary to provide storage protection for these pieces of \ninformation. For simplicity, the approach used was to triple-store the \ndata and recover them by a two out of three match. This includes not \nonly data, but also the long-term software timers.\n\nThe TAU has the ability to output both minor and major system \nalarms and to light alarm LEDs on its faceplate. To control these alarm \noutputs, the software maintains several alarm filters. These filters are \nup/down counters with a natural decay (down count) built in. To \nmaintain an alarm condition, the appropriate filter must be incre- \nmented periodically or fail to be incremented, depending on its use. \nThese alarm filters are maintained for various purposes. For example, \none filter checks that the interrupt routines are periodically serviced.\n\nFinally, the software also includes necessary diagnostic routines. A \nROM checksum and a processor maze test are performed continuously. \nA RAM test is also done continuously, but without slowing down call \nprocessing or missing activity. This is done by testing one byte of RAM \nat a time from an interrupt routine. A few bytes of RAm that are \ndirectly involved in the highest-speed interrupt are tested by an \nindirect method. As alluded to earlier, methods are also used to detect \nthat the processor periodically passes through various portions of the \ninterrupt routines. A sanity monostable is also employed and is strobed \non each cycle of the main routine. Power-up initialization routines are \nprovided based on duplicated bytes that are set in the transmit TsI \nafter a power-up or a power supply glitch. Initialization of the data \nbase is not done based solely on a sanity monostable timeout or \nprocessor reset since this could be caused by a glitch, which would not \nbe a reason to take down all line/trunk connections.\n\nV. SELECTED IMPLEMENTATION DETAILS \n5.1 The time-slot interchange chip \n5.1.1 Organization of the TSI RAM\n\nThe NMOS TsI chip uses a custom-designed RAM in conjunction with \npolycells (standard catalog gate functions) for the register and control \nlogic. Figure 7 shows a functional block diagram of the RAM. The \noperation of the RAM is slaved to a Master Clock (Mc) input signal, as \nis nearly all on-chip circuitry. Address, input data, and output data are \nall latched internal to the RAM. Separate data input and data output \nbusses exist. The memory is broken into three sections of 49 bytes \neach. The upper two address bits are decoded to select one of the three \nsections. Since four possible combinations of the two address leads \nexist, the fourth combination, not needed for addressing a memory \nsection, is used for addressing the data-link register (Register 7, Fig. \n5), activity/TNEN control register (Register 8), and two test bits, thus \nmaking these isolated circuits appear as part of the RAM memory \nspace.\n\nThe bottom six address leads are decoded to select one of the 49 \nbyte locations within a memory section. To avoid unnecessary transis- \ntors and a resulting slowdown of memory operation, a full decoding of \nthe six bits was not done. Valid addresses for the six least-significant \nbits are decimal 0 to 48 and 63. Addresses 0 to 47 access the 48 bytes \nused for pcm storage in Sections 1 and 2 of the memory and access the \n48 bytes used for trunk and busy-trunk assignments in Section 3. \nAddress 48 or 63 selects the remaining byte. This byte is an all-zero \nROM byte in Sections 1 and 2 and a read/write byte in Section 3. The \nROM bytes are used to send all-ones (by an inversion) on the T1 line\n\nfor idle trunks. Unassigned trunks are written to \u201cline\u201d number 48 or \n63 which, when fed back, will address one of the Rom bytes. The 49th \nbyte in Section 3 is used as a buffer store for the fast-busy tone bytes \nwritten to the TsI by the microcomputer. Since the tone byte must be \nstable during an entire frame, a buffer is needed to allow the micro- \ncomputer the flexibility of writing the next tone byte any time during \na frame.\n\nThe memory presents a timing interface for the remaining circuitry \non the TsI chip (see Fig. 8). For a read or write operation, the address, \nread/write signal, and input data must be stable at the end of the first \nhalf cycle of Master Clock (Mc). The memory latches the address, \nread/write, and input data on the rising edge of mc. For a write\n\noperation, the write will be completed by the end of the Mc cycle (Fig. \n8). On a read operation, a peculiarity of the memory is that the data \nappears in the next Mc cycle after the cycle that commanded the read. \nThe data becomes valid some time after the start of the cycle and \nremains valid for the rest of the cycle. Each cycle of Mc can be used to \nperform either a read or write. The TsI design always mandates one or \nthe other. In memory cycles when no useful function is needed, a \nmemory read is performed but the data is not accepted by any of the \nregisters attached to the memory output data bus.\n\nFor receive TSI operation at the cot, the odd-address locations in \nSection 3 contain the busy trunk assignments. Again, as with the \ntrunks, each location is written to a number between 0 and 47 to assign \na line to a busy trunk, or written to 48 or 63 if the busy trunk is not \nassigned. The upper two bits of the busy trunks have no special use. If\n\nthe TSI is operating as a transmitter, then there are no busy trunks, \nbut some of the same locations are used for holding activity and TNEN \nbits. Each of the activity or TNEN bytes holds eight data bits, one for \neach of the eight lines. Thus there are six bytes of activity and six \nbytes of TNEN to hold data for the 48 lines.\n\nThe control logic on the TsI chip regulates the flow of data between \nregisters and memory within the TsI chip, and the selection of times \nappropriate for microcomputer accesses. The main element of this \ncontrol is a 12-frame counter. By assigning states of this counter to \nvarious functions, control of the chip is achieved.\n\n5.1.3.1 Trunk assignments. To understand the assignment of states \nto specific functions, first consider the steps that must be performed to \naccomplish the basic time-slot interchange function for a TsI operating \nin the transmit mode. Serial unconcentrated data is coming in from \nmaster and slave TRUs and is being converted to eight bit parallel data \nin Registers 1 and 2 (see Fig. 5). Every eight cycles of Mc (1.544 MHz \nMaster Clock), two pcm words must be written to the memory, one \nfrom Register 1 and one from Register 2. Similarly, every eight cycles, \none PCM word must be read from the memory and written to Register \n3 for transmittal on a T1 trunk. (As noted earlier, the trunk data is \nread from one section of the PCM memory while the unconcentrated \ndata is being written into the other section.) To read out the trunk \ndata, however, the line assignment for that trunk must first be read \nout to Register 4, to provide the address for the line data to be read \nout. Thus, these basic operations consume four out of every eight Mc \ncycles. For the receive TsI operation the trunk data comes into Register \n1 and exists unconcentrated through Registers 2 and 3, similarly \nrequiring four cycles.\n\n5.1.3.2 Busy trunk assignments. In the receive TSI operation at the \ncot where busy trunks are needed, two additional cycles out of every \neight are used, one for reading the busy trunk assignment to Register \n4 and one for writing the current fast-busy tone byte from holding \nRegister 5 to the Pcm line memory of the addressed line. The fast-busy \nbyte is then read out to Register 2 or 3 via the normal sequential \nreadout. The current fast-busy tone byte is read from the fast-busy \nbuffer location in memory to holding Register 5 during the frame bit \nstate (193rd count of the control counter). Thus the receive TSI at the \nCOT uses a total of six out of every eight cycles plus the one extra state \noccurring each frame.\n\n5.1.3.3 Activity/TNEN. In the transmit TSI mode, states other than \n\u2018the four basic states are used for transferring activity and TNEN \ninformation from/to Register 5. These states occur once every 32\n\nstates rather than once every 8 states, because it takes that long to \ncollect a byte of activity or TNEN data.\n\n5.1.3.4 Microcomputer access. In all modes of operation, no more \nthan six out of the eight states are used for internal operations; the \nother two states are reserved for potential use by the microcomputer \nto perform a read or write of the memory. Waiting for an available \nstate gives rise to the wait states encountered by the microcomputer \non a TSI access. The two internally unused states out of every eight are \nevenly spaced to minimize the delay seen by the microcomputer.\n\n5.1.4.1 Transmit TSI mode. In the transmit mode the TsI chip estab- \nlishes its own reference based on its internal counter state. It then \nforms a super-frame system synchronization signal for sending to the \nTRUS (a unique pattern over 12 frames). The signal consists of two \npulses, one in frame 2 and one in frame 12. Each pulse must be only \n162 ns wide to synchronize the 6.176 MHz count-down circuits in the \nTRUs (6.176 MHz is the fundamental crystal-controlled clock in the \nSLC-96 system from which the T1 rate is derived). The TRUs and the \nTAU receive the same clock and then the TAU must carefully time the \npulse with respect to the clock edges. Because of the short delays \ninvolved here, the pulse had to be timed with TTL external to the \ncustom NMOS chip. The TsI chip puts out a pulse that spans two 6.176 \nMHz periods which is then gated externally to produce the desired \npulse. The effect of the pulses that are sent to the TRus is to synchro- \nnize the counters in the TRUs.\n\nAs the slave digroup PCM enters the TsI, it is delayed by one clock \nperiod by means of an FF (see Fig. 5) so that it arrives one MC cycle \nlater than the master pcm. The outgoing trunk Pc\u00bb is also delayed one \ncycle by means of an FF, which exists for matching outgoing PCM \nstreams in the receive mode. The frame bit from the master digroup \nis picked off and saved in the FR FF for re-insertion into the outgoing \ntrunk pcm bit stream (which is why this stream is called the master).\n\n5.1.4.2 Receive TSI mode. Synchronization in the receive mode is \nvery different from synchronization in the transmit mode, because the \ntiming is controlled by the reframing circuit contained in the TRu. This \ncircuit locks on the frame bit in the received T1 bit stream and thus \ncan provide the timing reference and reframe signals needed by the \nreceive TSI in the TAU.\n\nThe timing reference is provided in the form of a 4 KHz (two frame) \nclock signal from the master digroup TRU. The edge of this clock \nwaveform is used to trigger the TsI control counter to a predetermined \nstate chosen to synchronize the TsI to the incoming bit stream from \nthe T1 line interface unit. When calculating the position of this bit\n\nstream relative to the synchronization signal provided by the TRU, the \ndelay in passing through the TsI must be taken into account.\n\nReframing is a very important consideration for receive TSI opera- \ntion. Since the reframing process occurs downstream from the TsI, the \nTSI must assume a special mode of operation during the reframe \nprocess. This mode must pass the concentrated trunk bit stream \ndirectly through the Ts1 unchanged to assure that the frame bit is \nreceived by the TRus (the timeslot interchange function can lose the \nframe bit if the Tst chip is out of sync). Once the TRus reframe, the \nreceive TSI can resume normal operation. To assure that the TRU is \nnot thrown out-of-frame when the receive TSI returns to its normal \nsequencing after a reframe, it is necessary that the relative position of \nthe frame bit not change. That is, during the time that the TsI is \npassing the received concentrated bit stream directly through, it must \ninsert the same number of cycles of delay as it will when it is operating \nnormally. Also note that the scrambled time slots sent to the channel \nunits during reframe are of no consequence because the channel units \nare not enabled to receive the information.\n\nA further consideration during the reframe process is that the 4 KHz \nsynchronization signal slips as the TRU searches for the frame bit. Thus \nthis signal cannot be allowed to preset the TsI control counter during \nthis time or confusion can result. The TRU provides an out-of-frame \nsignal that can be used to gate the synchronization clock and also force \nthe TsI chip into its special reframe mode of operation. Once the out- \nof-frame signal indicates the TRU is reframed, the synchronization \nclock is allowed to once again force the counter state in the receive \nTSI.\n\nDuring the reframe mode the TsI performs as follows. All incoming \ntrunk data, instead of being selectively written to the assigned line \nmemory locations, is sequentially written to all the slave PCM memory \nlocations in the same section of PCM memory as the selective writes \nwould have been done. The sequential readout of each line\u2019s data in \nthe alternate section of PCM memory continues as normal, except that \nthe slave data is also forced out to the master digroup. In this way the \nconcentrated bit stream is passed directly through the TsI with the \nsame delay experienced by the bit streams in the normal mode of \noperation. Note that if the receive TSI is out of synchronization (as it \nis assumed to be if a reframe is needed), then the frame bit is not being \nput into the frame flip-flop (FR FF). Instead, an arbitrary bit out of the \nframe sequence is being inserted in the FR FF, depending on the relative \nstate of the TsI control counter. The actual frame bit, then, is being \nstored somewhere in the PCM memory. The bit stream passes un- \nchanged through the TsI, however, with the bit stream being put back \ntogether correctly as it exists from the TsI.\n\nAfter the reframe process terminates, the control counter is resyn- \nchronized as mentioned above. Since the frame bit was arbitrarily \nlocated in memory, the resynchronization will, in general, cause the \nloss of one frame bit. The lost frame bit will be in error with 50 percent \nprobability assuming a random bit is inserted in its place. By selecting \nthe rising edge of the synchronization clock, the frame bit that is lost \nis a signaling frame bit. Thus there is no concern, during the resyn- \nchronization, about approaching the master frame bit error threshold \nthat would throw the TRU back out-of-frame. A signaling frame bit \nerror will, at most, delay the reframing of the signaling bit extraction \ncircuit in the TRU.\n\nOne other step must be taken in the TsI to complete the reframing \nprocess. Since the slave PCM memory locations were used to hold data \nwhile the TsI was passing the concentrated bit stream directly through, \nit is desirable to clean out this memory, that is, write these memory \nlocations to idle pcm code. This is done automatically by the hardware \nduring the two frames that follow the control counter resynchroniza- \ntion (two frames are needed to initialize both sections of PCM memory).\n\nThe tsi adds approximately four frames of time to the reframe \nprocess for pcm data. This is an increase of 0.5 ms to an average \nreframe time of 25 ms. The TsI may also add up to one super frame of \ndelay to the receipt of A and B signaling bits by the channel units \nbecause of the erroneous signaling frame bit. This is equivalent, worst \ncase, to the loss of one additional A and B signaling bit.\n\nThe development of the Time-Slot Interchange (TsI) chip was \njustified on the basis of cost, power, and space. Further the chip is a \nuniversal design that can function as a transmitter or receiver, at COT \nor RT, so that only a single custom design is required. The breadboard \nfor the TsI was built using 96 off-the-shelf integrated circuits. The chip \nis realized with NMOS polycells and a custom NMOS RAM using 5 micron \nrules on a chip 258 X 367 mils, packaged in a 40-pin pip. Typical power \ndissipation is 750 mW. The design cycle from the first paper design to \nfirst chips took about two years. The chip contains 432 polycells, 145 \nbytes of static RAM, and two bytes of RoM. There are about 10,000 \ntransistors, 70 percent of which are used in the RAM. Extensive logic \nand timing simulations were required to verify the design.\n\nThe Ts clock frequency of 1.544 MHz mandated careful considera- \ntion of timing delays in the design of subcircuits for the TsI. \u201cRegular \npower\u201d NMOS polycell gates can give delays of approximately 50 ns. \n\u201cHigh-power\u201d and \u201csuper-power\u201d gates must be used for shorter delay\n\ntimes with corresponding increases in power dissipation. Certain crit- \nical paths in the Ts1 did require the use of high-power and super-power \ngates.\n\nThis method of meeting timing constraints, however, was only \nsupplementary to the decision to build a highly synchronous design. \nBy clocking nearly all subcircuits directly with Mc (Master Clock), a \nvery clean timing plan was developed. That is, all propagation delays, \nsetup times, and hold times could easily be calculated with respect to \nan MC edge. Memory operations were also specified with respect to McC \n(previously discussed). Such a design avoids accumulating long strings \nof delays that can cause problems at the expense of, in some cases, \nadditional gates. For example, all the registers have their FFs directly \nclocked by mc. Gating around and between FFs determine the function \nthe register performs such as hold, shift, or load. With this type of \ndesign, where all action takes place on a clock edge, the remaining \nportion of a cycle can be used for propagation delays of the signals \nthat determine what function will be performed when the next clock \nedge arrives. This design technique is very valuable for high-speed \ndesigns (high speed relative to the technology limitations).\n\nAnother timing consideration that should be mentioned is the ex- \npected clock duty cycle variations in the receive TSI mode. MC in the \nreceive mode is the recovered T1 line clock. As such, duty cycle \nvariations can be expected. Final assumptions were for a 60/40 duty \ncycle worst case in either direction. This implies a need for the TSI \nchip to work, equivalently, at a higher frequency (viz. 40/40). Timing \nmust apply for worst-case device, power supply, and temperature \nvariations, as well, leading to the test-clock rate at room temperature \nof 2.6 MHz.\n\nA basic problem in structuring the software was to determine a \nmethod for handling the real-time constraints. All the interrupt rou- \ntines run indirectly off the fast-busy tone interrupt which occurs once \nevery eighth of a millisecond (recall Fig. 6). The message routines can \ntake milliseconds to execute, however. Thus it was necessary to enable \nthe interrupt upon leaving the Fast-Busy Tone routine to enter the \nother interrupt routines. The result is that several levels of interrupt \ncan exist on the stack. The MAC-8 microprocessor automatically saves \nthe condition register and return address on the stack when an inter- \nrupt occurs, and thus conveniently allows nesting of interrupts. Since \nthe Main routine may have been nested in subroutines when it was \ninterrupted and the interrupt routines, themselves, may have been \nnested in subroutines at the time of another interrupt, the stack must \nbe large enough to hold all the return addresses. A few data values are\n\nalso sometimes stored on the stack. Fifty bytes was determined to be \nadequate by considering worst-case nesting levels and including margin \nfor unexpected levels caused by an interrupt glitch or an accidental bit \nflip (a subroutine call stores two bytes on the stack, an interrupt three \nbytes).\n\nTo get a better feeling for how the processing gets time-shared, refer \nto Fig. 9. This time diagram is an example of the routines that might \nbe processed in response to a sequence of interrupts. The Fast-Busy \nroutine is executed after every interrupt, followed by a return to the \nroutine which had been executing. This time line starts out by assum- \ning the processor is in the Main routine when an interrupt occurs. The \ninterrupt causes the Fast-Busy routine to be executed. After 16 exe- \ncutions of the Fast-Busy routine, a 2 ms timeout occurs causing the \nPoll routine to be entered (recall Fig. 6). The Poll routine can be \ninterrupted several times, before it finishes, by the Fast-Busy routine. \nAfter the Poll routine finishes, it is assumed that the Transmit Message \nroutine needs to be executed. This routine runs for milliseconds and so \nwill be interrupted not only by the Fast-Busy routine but also by the \nPoll routine. After the Transmit Message routine terminates, it is \nassumed necessary to process the Receive Message routine and so \ncontrol switches to it. Finally, the Receive Message routine finishes \nand control returns to the Main routine.\n\nThe Fast-Busy, Poll, and Display Mux routines are fast enough to \nfinish before they would be called upon to re-execute. However, they \nwould not cause serious problems even if they were re-entered because \nof some glitch. The message routines are much more complex, however, \nand can cause some serious consequences (confusing trunk assign- \nments, for example) if they are re-entered. Normally these routines \nwould be finished before being called upon again (as necessary to meet \nthe demands of the data links for messages every 9 ms). However, to \nprovide a degree of protection from the havoc that could occur if they \nwere re-entered, a flag is maintained that indicates the message routine \nneeds processing or is in the process of being executed. This flag is set \nby the Poll routine when it determines that the message routine needs \nprocessing, and is cleared at the end of the message routine after all \nprocessing is completed. These flags also provide the means by which \ncontrol can be transferred directly from the Transmit Message routine \nto the Receive Message routine or vice versa (recall Fig. 6).\n\nThe data link provided the TAU by the SLC-96 Data Link Unit \n(DLU) consists of 11 bit packets of serial data every 9 ms. The TAU is \nprovided with a frame signal that remains high for 2.75 ms and low for\n\n2ms ENTER 2ms RETURN TO \nTIME OUT XMIT MESSAGE TIME OUT XMIT MESSAGE \nROUTINE ROUTINE\n\nXM TRANSMIT MESSAGE \nEXIT XMIT EXIT REC \n: RECEIVE ME E \nENTER REC MESSAGE ROUTINE, Am SEEM Eae as \nMESSAGE ROUTINE RETURN TO MAIN\n\n6.25 ms. During the high portion of the frame, serial data is transmitted \nor received at a 4 KHz rate (1 bit every 0.25 millisecond), by combining \na 4 KHz clock provided by the DLU with the frame signal. As Fig. 10 \nindicates, the time while the frame signal is low is available for loading \nor unloading, in parallel, the shift register provided in the TsI chip. \nFigure 10 also indicates that the frame signal is polled every 2 ms. The \npolling is asynchronous and so only one possible phasing of the polling \nwith respect to the frame signal is shown. The polling scheme works \nby noting 1 to 0 transitions of the frame signal and using this event as \na trigger for processing the message routine and loading (unloading) \nthe shift register. Both the Transmit Message routine and the Receive \nMessage routine write or read, respectively, the shift register immedi- \nately upon entry; thus, if the routine is delayed or takes a while to \nprocess, it is assured that the data-link shift register is loaded or \nunloaded prior to the next rise in the frame signal.\n\nThe actual data-link polling is complicated by the fact that there \nare two data-link frame signals, one for transmit and one for receive. \nThe data links are asynchronous with respect to each other (or at least \nout of phase with any random phase), and so all relative phasings must \nbe considered. The message routines are designed to run to completion \nbefore transferring control to the other message routine, if needed. \nTherefore, a message routine can be delayed and real-time constraints \nmust be considered. Since a request to process both message routines \nmay occur at the same time, a priority of processing had to be \nestablished.\n\nFigure 11 shows two possible phasings of the data-link frame signals \nand the processing sequences that could correspondingly result (the \nReceive Message routine is given priority). Note that message routine\n\nCASE 2 \nREC DATA Es 1] rl \nLINK FRAME \nREG UNLOAD f t t \nXMIT DATA \nLINK FRAME \nREG LOAD \u2018 { t\n\n*NOTES: \n1. M = MAIN ROUTINE \nRM = RECEIVE MESSAGE ROUTINE (3 ms RUN TIME) \nXM = TRANSMIT MESSAGE ROUTINE (4.5 ms RUN TIME)\n\n2. RM HAS PRIORITY OVER XM \n3. LOW LEVEL INTERRUPTS IGNORED IN REPRESENTATION\n\nprocessing need not alternate; instead, the Transmit Message routine \nmay be processed twice in a row followed by the Receive Message \nroutine being processed twice in a row as the result of 2-ms polling of \nthe 9-ms frame interval.\n\nIf both frame signals are aligned and the polling was such that the \ntransition from 1 to 0 is not noted until almost 2 ms after it occurs, \nthen the Receive Message routine is constrained to be less than 4.25 \nms. Enough time must be allowed to enter the Transmit Message \nroutine and put out the new submessage, prior to the rise of the \ntransmit data-link frame signal.\n\nSince only a single flag is used with each message routine to indicate \nthat the routine needs processing or is in processing, and since requests \nfor processing can occur 8 ms apart, the sum of the execution time of \nthe Transmit and Receive Message routines is constrained to less than \n8 ms. If, under some unusual circumstance, the processing takes more\n\nthan 8 ms, the result is that the next occurrence of the Transmit or \nReceive Message routine is not processed and a data-link submessage \nis therefore lost. But because of the redundancy and error protection \nbuilt into the data-link messages, a message will probably not be lost.\n\nThe most time-critical routine is the Fast-Busy routine. Great care \nwas taken to make this routine as fast as possible; two MAC-8 registers \nare used, one for an auto-increment pointer to the fast-busy word data \ntable (register b4) and one as a temporary holding register for the tone \nbyte (register a5), since a memory-to-memory transfer instruction does \nnot exist. (In MAC-8 assembly code, \u201cb\u201d registers are 16-bit general- \npurpose registers and \u201ca\u201d registers are 8-bit general-purpose registers.) \nFor the purpose of saving bytes and improving speed throughout the \nentire TAU program, two eight-bit registers (a3 and al3) are used for \nflags. One of these flags is used for the silent interval of the fast-busy \ntone (the tone is pulsed on for a 0.25 s, off for a 0.25 s).\n\nTable IV gives the cycle count on an immediate return path of the \nFast-Busy routine. The variable number of cycles for the write instruc- \ntion is due to the variable number of wait states for a TSI access. \nAssuming that 67 cycles of each frame (out of the available 193) are \nused for the Fast-Busy routine implies that 35 percent of the real time \nis used in this routine.\n\nA 2-ms software timer is derived from the data table pointer (register \nb4) in the fastest possible way by a test on bit 4. For this scheme to \nwork, the Poll routine, when it is entered, immediately increments the \npointer by 16 so that only a single 2-ms timeout is indicated by the bit \ntest on the pointer. Thus the fast-busy code table of 48 bytes is actually \nstored in ROM as three tables of 16 bytes, separated in address space \nby 16 locations. Worst case, the data table pointer must be set to its \nnew value in less than one frame of time from the interrupt, to assure \nit has the correct value when the next interrupt occurs. The values for \nthe fast-busy code words were derived by calculating linear samples of \nthe tone and then using a translation table to companded PcM codes. \nThe least significant bit was forced to always be one as needed for A\n\nCycles Instruction Comment \n10 (interrupt preamble) \n7 if (bit(7, a3)) goto ial; /* jump if silent interval */ \n15 a5 = *b44+4; /* fetch busy word */ \n16-22 BSYWD = a5; /* write wordtorec TSI */ \n7 if (bit(4, a4)) goto ia2; /* check polling timer */ \n9 ireturn( );\n\nand B bit signaling to the channel units. The code inverse actually \ngets sent to the channel units because of a data inversion in passing \nthrough the TSI.\n\nThe Poll routine is less critical than the Fast-Busy routine by a \nfactor of 16, but it is still important to consider worst-case paths. The \nroutine is held to a minimum, again, by careful use of register instruc- \ntions and arranging the conditional branches for minimum worst-case \npath. The execution time of the Transmit and Receive Message \nroutines must also be considered, because they need to be fast enough \nto service the data-link requests upon demand, to avoid further com- \nplication of the algorithms. An example of what can be done to \nminimize worst-case timing paths is the execution of the consistency \nchecks and updates in the Transmit Message routine on the passes \nthrough this routine that do not require the calculation of a new \ntransmit data-link message. That is, since a message is composed of \nthree 11-bit submessages which are calculated and saved until needed, \nno new message calculation is required while the first two of these \nsubmessages are being sent, and so there is real time available for the \nconsistency and update routines to use.\n\nThe Main routine has no significant real-time constraints; however, \nit is still necessary to know the approximate cycle time of the routine \nto allow an adequate timeout for the sanity monostable. The cycle \ntime of the Main routine is, of course, strongly affected by the running \ntime of the interrupt routines. (It is necessary to strobe the sanity \nmonostable from the Main routine, rather than an interrupt routine, \nsince the latter might continue to strobe the monostable while allowing \na return to an unknown loop at an arbitrary location, rather than to \nthe Main routine.)\n\nThe processor RAM is used for several purposes as was given in \nTable III. The line/trunk data base stores copies of the 24 trunk and \n24 busy-trunk assignments. These copies are used during assignment \nsearches and acted upon by the consistency routines. The transmit \nand receive TSI assignments are updated from this copy. These copies \nare maintained because they can be accessed more quickly than the \ninformation in the TsIs.\n\nThe line/trunk data base also holds information concerning the 48 \nlines the concentrator serves. This information is split into six data \ngroups, each group holding information for eight lines. The information \nis stored in bitwise correlation with the eight bits of line activity as \ncollected in a single byte by the transmit Ts1. For each group, seven \nbytes are stored. Bytes 0 and 1 are the TNEN masks for forcing activity \nor no activity. Bytes 2 and 3 are the activity filters, stored as least-\n\nsignificant bits and most-significant bits of two bit up/down counters. \nByte 4 stores the status of activity from the RT as received over the \ndata link. The trunk and busy-trunk status in bytes 5 and 6 indicate \nwhether a line is assigned to a trunk or busy trunk, respectively. These \nbits are maintained to provide a fast method of determining whether \nan assignment (deassignment) is needed. Without them, one would \nhave to scan all the trunk assignments for every line that had activity \nto determine which line needed assigning. This could not be done \nwithin the time constraints of generating a data-link message \u201cupon \ndemand.\u201d All the bytes of data in each group are arranged so that all \neight lines can be processed simultaneously by byte operations, thus \nperforming activity filtering, TNEN masking, and service request deter- \nmination very quickly.\n\nA significant feature of the software is the ability to allow concen- \ntrator operation in the presence of partial faults. Before assignment to \neither a trunk or busy trunk, extensive local memory checks are made \non the trunk (busy trunk) assignment, enable bit, pcm data, and status \nbit memory locations. A failure causes an alarm to be raised and \npossibly a line/trunk fault to be stored. If local tests pass, then the \nPCM looping test is set up for a trunk assignment. A failure of this test \ncan also cause a fault to be stored.\n\nThe routines that are provided allow for the detection of memory or \nconnect (disconnect) failures at the time of line/trunk assignment \n(deassignment). If a particular line/trunk combination fails, that com- \nbination is put in a fault store and periodically retried. One such fault \ngives rise to a minor alarm, two such faults shut down the system. The \nquestion is always raised, why not try to assign the line to a different \ntrunk if the first one fails? This is not as easy as it sounds. The basic \nproblem is one of fault isolation. The line/trunk combination is fun- \ndamental in finding the fault; splitting them up could easily cause a \nloss in the ability to refind and retest the fault, to maintain an alarm. \nThe result could easily be intermittent alarms or a faulty trunk that \nwanders from line to line, possibly causing random customer com- \nplaints. In any case, it would require a lot more software with dimin- \nishing returns. It is also true that, for a given line/trunk combination \nfault, hardware considerations give a higher probability to the line \nbeing at fault than the trunk. In short, it seems very acceptable, and \nis fairly straightforward in software, to keep the line/trunk combina- \ntion as a means for maintaining an alarm while allowing all other \ncustomers normal service.\n\nA fault consists of a trunk and line pair. The stored trunk number \nis the lower byte of the address to the trunk or busy-trunk assignments\n\nin the processor RAM. A line number is stored as 0 to 47. If the fault \nwas trapped during a deassignment, the \u201cline\u201d number may be 48 or \n63. Both 48 and 63 are used because TsI will allow either one as a no \nassignment number. If memory fails and will not allow deassignment \nto 63, then 48 is tried. An empty fault store is designated by all ones \nfor the line number and all zeroes for the trunk number.\n\nIf a fault is to be retried, the trunk and line numbers are checked for \nvalidity and then a jump to the appropriate trunk (busy trunk) \nassignment (deassignment) algorithm is made. Also, to assure that the \nfault is not lost because of changes in the faulted line\u2019s activity, \npermanent activity is maintained by setting the activity filter for that \nline to a count of 3. A fault will be retried approximately every 1.8 s \nand thereby continue to increment an alarm filter if the fault persists.\n\nFor a trunk assignment, the pcm looping test is always executed. \nFirst, the test is initiated in the Transmit Message routine. This \nprocess consists of clearing the enable bits in transmit and receive TsIs \nfor the line under test, so that the test codes will not be overwritten or \nsent to the channel units. Also, the first Pcm test code (alternating 1\u2019s \nand 0\u2019s) is written into the transmit TSI PCM memory locations and a \nPcM test timer is set for an 80-ms timeout. Since the line is assigned to \na trunk at the cot, the test codes are received at the RT as soon as the \nRT receives the trunk assignment message over the data link. The \nenable bits at the RT are automatically cleared upon receipt of the \nassignment message.\n\nThe operation of the test at the coT is then picked up by the pcm \ntest portion of the Poll routine. This routine will sample the receive \nTSI every 2 ms looking for the Pc test code that should be returned \nby the rT. When the code is received, it is complimented and sent to \nthe RT. When the complimented code is received at the cot, the cor \nsends a test termination code (all 1\u2019s) to the RT for 10 ms and then sets \nthe coT TsI enable bits, thereby cutting through the customer at the \ncot. The RT will correspondingly set the enable bits at the RT when it \nreceives the test termination code.\n\nIf a timeout occurs, the coT will deassign the trunk and store the \nline/trunk combination for later retry. If the RT fails to see the test \ntermination code after a timeout from receipt of the assignment \nmessage, it will simply deassign the trunk (only the cor records the \ntrunk and line that gave rise to a fault). If a trunk (busy trunk) \ndeassignment is requested, the disconnect is performed in a straight-\n\nforward manner. Some memory tests are performed and may result in \nthe storing of a fault. This fault may involve only a trunk if the \nproblem is in writing the unassigned \u201cline number.\u201d\n\nPerformance testing of the TAU was a very big part of the project. \nWhen all the hardware and software effort expended in testing is \nconsidered, the testing job was nearly as big as the basic job of \ndesigning the TAU circuits, custom chip, and software. Because of the \ncomplexity and compactness of the design, thorough and sophisticated \ntesting was essential.\n\nThe TAU project began with the design of the TsI. TsI breadboards \nwere built and debugged using a scope and logic analyzer. Because the \nTSIs use a handshake arrangement in talking to the microcomputer \nbus, manual switches can be used to simulate the bus signals, and thus\n\nthe TSI design was well on its way to being realized as a chip, the \nmicrocomputer designs were finalized and built. The microcomputers \nwere debugged using simple programs and a logic analyzer.\n\nOnce serious programming began, a versatile test set was needed in \naddition to the MAC-8 development system (PLAID). A MAC-8-based \ntest set was designed that allows the display of all 24 trunk assignments \non numeric displays by examining the data link messages that flow \nfrom COT TAU to RT TAU. It also employs A and B bit and TNEN bit \ngenerator cards that store information for all 48 lines that the concen- \ntrator is working with and allows the operator to manually set these \nbits to simulate channel units. It also displays the received A and B \nbits for the selected channels.\n\nThis test set was invaluable for tracing bugs in the TAU software as \nit developed. Some of the intermittent and transitory phenomena was \nespecially visible on the displays. It was also very nice for observing \nthe results of simulated memory faults, which was done with another \npiece of test hardware, consisting mainly of EPROMs with the selected \nbits to be faulted marked in the EPROM. Most of the software for the \nTAU was written and debugged using this MAC-8-test set in combina- \ntion with the PLAID.\n\nAnother capability that was developed later and was very useful in \ndebugging the code was a data-link monitor. This allows the concen- \ntrator data-link messages in both directions to be displayed on a CRT \nin a correlated fashion. The program allows the storage of messages \noccurring over approximately 2.3 s and allows triggering on a particular \nmessage pattern with \u201cdon\u2019t care\u201d conditions. The stored messages \n_can then be viewed by scrolling forward and backward. The data-link \nmonitor was especially useful for checking out the software that\n\ndetermines message priority. It was also useful in noting response \nmessages, such as a trunk assignment message leaving the COT TAU in \nresponse to an activity message received from the RT TAU. Another \nuse was in viewing widely-spaced periodic messages such as the data- \nlink \u201cLooping Message.\u201d\n\nA development that paralleled the TAU and which proved very useful \nto us for system tests and final Tau software tests was the \u201cTraffic \nGenerating System\u201d (TGs). This development was initiated to simulate \nrealistic traffic on digital lines for the testing of a digital switch. It was \ndecided to develop the hardware and software necessary to use the \nSLC-96 carrier system for simulating traffic on a T1 line. This hardware \nand software also serves as a debugging tool for the TAU, while the TAU \nalso provided a shakedown test for the traffic-generating system.\n\n\u201cSignaling interface boards\u201d that perform similar to the original \nMAC-8 test set are used to simulate channel activity in TGs, that is, A, \nB, and TNEN bits for all channels are stored in a RAM and read out in \nthe proper sequence. Similarly, received A and B bits are stored. One \nsignaling interface board serves 24 channels. The RAM is writable and \nreadable through an I/O port to a DEc-LSI-11. This hardware is \nflexible, portable, and used simply by plugging the simulator card into \na channel unit position in the SLC-96 system bank. The LsI-11 is tied \nthrough a satellite processor link to a host UNIX system (see Fig. 12).\n\nBy using TGs and the data-link monitor, the final TAU boards were \nexercised very thoroughly. One program that was written measured \nthe connect delay time from an A or B signaling bit change. This \nprogram was expanded to make thousands of random calls, measure \nthe delays, and store the results in a UNIX file for later graphing of \ndelay distributions. Most of the delay and distribution of delay is \nattributable to the delay and asynchronism of the data link. Originating \ncalls are delayed more than terminating calls because of the delay in \ntransmitting RT activity to the coT. Average connect delay from the \ncoT is 70 ms, with 100 ms from the RT.\n\nDelay measurements were repeated with a random-error generator \nused to insert errors on the T1 lines. At an error rate of 2 x 10~*, which \nis worse than a functioning SLC-96 system will see, the only noticeable \nchange in the connect delays was an increase of a few milliseconds in \nthe delay. This checked out the error protection built into the data- \nlink messages.\n\nAnother program that was written simulated two simultaneous calls \nand measured the connect delay of the delayed call. These measure- \nments showed an increase in the connect delay of approximately 50 \nms for the second call. Programs were also written to manipulate the \nTNEN bits, with and without A and B activity, and thereby simulated \neach type of channel unit in each physical position. Other programs \nchecked blocked-call functioning, the generation of fast-busy tone, and \nnormal ringing and dialing. The traffic generating system was undoubt- \nedly very important in establishing confidence in the final hardware/ \nsoftware design.\n\nThe SLC-96 carrier system TAU demonstrates that modern electron- \nics economically provide improved features in the loop plant. The \nability to integrate the Time-Slot Interchange function onto a single \nchip made this development possible. By digital concentration, the \nTAU reduced the number of T1 lines needed by the SLC-96 system \nfrom five to three. Because provision is made for special-service circuits \nto be given unconcentrated trunks, a separate system is not needed to \nprovide a few special interfaces. Traffic measurement and extensive\n\nmaintenance features were also successfully integrated into the firm- \nware control of the TAU.\n\nThe author wishes to especially acknowledge Lary Range for per- \nforming extensive simulations on the Ts1 chip and John Beck for his \nsoftware-design assistance. The layout and design of the Ts1 chip were \ndirected by Gil Mowery. Sam Arnold, Brian Redman, and Doug Corey \nwere responsible for the software design of Tas. Credit is also due to \nmany other individuals who were associated with all phases of this \nproject.\n\n3. H. D. Rovegno, \u201cA Support Environment for MAC-8 Systems,\u201d B.S.T.J., 57, No. 6, \nPart 2 (July-August 1978), pp. 2251-63.\n\nCopyright \u00a9 1981 American Telephone and Telegraph Company \nTue BELL SysTEM TECHNICAL JOURNAL \nVol. 60, No. 2, February 1981 \nPrinted in U.S.A.\n\nThis paper describes a post processor that improves the assembly- \nlanguage code generated by the portable C compiler. The novel ability \nto change a sequence of two-address instructions into an equivalent \nthree-address instruction distinguishes this particular code improver \nfrom other \u201cpeephole\u201d improvers. The combined compiler-improver \ngenerates good three-address code for the Digital Equipment Cor- \nporation vAx-11\u00b0 computer without requiring extensive changes in \nthe compiler itself, which was designed to accommodate machine \narchitectures with at most two addresses per instruction. For typical \nprograms the improver reduces the number of bytes in the instruction \nstream by 10 to 23 percent. This paper emphasizes the technique used \nto transform two-address code to three-address code.\n\nThe portable C compiler\u2019 is an effective tool for quickly constructing \na C compiler\u2019 for a general purpose digital computer. With reasonable \neffort the resulting compiler generates correct code, and the quality of \nthe translation into assembly language is acceptable. However, users \nfrequently demand better code if they anticipate prolonged or exten- \nsive use of programs written for a particular application. A post \nprocessor that reads the assembly language generated by the compiler \nand writes better assembly language having the equivalent effect can \nsatisfy much of the demand. (Here \u201cbetter\u201d code requires fewer bytes \nfor instructions or less time to execute, or both.) This paper describes \na program that improves code generated for the Digital Equipment \nCorporation vax-11\u00ae computer, paying particular attention to the \ntechnique used to transform two-address codes into three-address \ncodes. :\n\nOne reason why a code improver can be effective is that the portable \nC compiler often generates code in the easiest possible correct manner, \neven if such a code is suboptimal over a wide range of machines. The\n\ncompiler expects that a post processor will clean up after it. For \nexample, the compiler translates the C program fragment\n\nwhich contains a conditional jump around an unconditional jump. It \nshould not be difficult to compile the original fragment as if it were\n\nbut the compiler does not do this, so one of the standard tasks for a \ncode improver is to replace \u201cskips over jumps\u201d with jumps on the \nnegated conditions.\n\nAnother reason that a code improver can produce better code is that \nthe compiler\u2019s model of code generation may ignore or not take full \nadvantage of architectural features found on a specific machine. The \nportable C compiler understands one-address instructions and two- \naddress instructions, but does not understand three-address instruc- \ntions or instructions which use an address as an immediate operand. \nSimilarly, the compiler thrives on certain addressing modes (register, \npointer, displacement from a register base) and has difficulty fully \nexploiting others (auto increment, double indexing).\n\nA code improver can also be effective because C-language statements \nor compilation on a statement-by-statement basis may be too low \nlevel. The concept \u201cturn off bit 15\u201d may have a direct hardware \nimplementation, but must be expressed in C language as a Boolean \nAND operation. The portable C compiler attempts no analysis of \ninterstatement information flow, nor does it always take advantage of\n\nhardware idioms. A code improver can often perform some flow \nanalysis and recognize more hardware idioms.\n\nThe code improver described here makes the portable C compiler \nusable as the workhorse compiler in a serious production environment. \nMeasurements indicate that for typical programs the improver reduces \nthe number of bytes in the instruction stream by 10 to 23 percent; the \nnovel technique reported here accounts for as much as one-third of \nthe reduction. The time required to execute the code is also reduced \nby 4 to 8 percent. The improver produces good three-address code \nfrom the two-address code generated by the compiler.\n\nAn existing improver of code compiled for the ppp-11 served as a \nmodel and outline for the vAx-11 code improver. The improver reads \na file of assembly language and divides the file into segments corre- \nsponding to C procedures. For each procedure it constructs a doubly- \nlinked list of the instructions and label definitions, with additional \nlinks for references to labels. The improver then combs the list, \nrepeatedly trying to apply any one of several incremental transfor- \nmations. The transformations satisfy a principle of optimality: Any \nlocal improvement is guaranteed to be a global improvement at least \nas large, and conversely, if the program as a whole can be made smaller \nor faster, then there is a collection of local changes which will account \nfor the improvement. When no further transformation can be made, \nthe improver prints the list and moves on to the next procedure. Many \nof the transformations depend little on the particular machine. \nStraightforward adaptation of the old program yielded code to transi- \ntively close jumps to jumps, delete instructions that immediately \nfollow unconditional jumps, delete jumps to the immediately following \ninstruction, remove unreferenced or redundant labels, merge common \ntail sequences, move basic blocks to the point of sole use, and inter- \nchange physical order of the consequent and alternative to a test. \nSimple modifications also produced a program to rotate loops to place \na single conditional jump at the bottom, handle skips over jumps, \neliminate redundant setting of the condition code, move common \nantecedents of jumps into the merged tail, eliminate constant tests or \ntests which are subsumed by a preceding test, exploit add-compare-\n\nFully utilizing the three-address instructions available on the vax- \n11 presented a new challenge. Table I illustrates a common opportunity \nto use a three-address instruction. In this example the variables a, 8, \nc are assumed to reside in memory (either global or local) and not in \nregisters. The first column gives a translation for the Ppp-11 that \ncannot be improved in either time or space. (If some of the variables \nreside in registers, then improvements are possible.) Both the produc- \ntion and the portable C compiler for the Ppp-11 produce this translation \nwithout the aid of a code improver. The second column contains the \ncode generated by the portable C compiler for the vAx-11. The com- \npiler saves one instruction by doing the work of the first two ppp-11 \ninstructions in one three-address vAx-11 instruction. However, it will \nnot generate the code in the right-most column, where a single instruc- \ntion suffices for the whole statement. Internally the portable C com- \npiler uses a binary tree to represent each parsed statement. The height \nof a binary tree with three external nodes (each explicit variable is \nrepresented by an external node) must be at least two. Furthermore, \nthe pattern-matching algorithms used by the compiler are restricted \nto subtrees of height one. (The pattern match has since been general- \nized to match subtrees of arbitrary height.) Thus the compiler gener- \nates two separate instructions for this case. It does have the flexibility \nto use an instruction with three addresses, but the destination operand \nof a three-address instruction must always be one of the compiler\u2019s \ntemporary locations, usually a register. The challenge to the code \nimprover is to recognize situations like this one and change the code \nappropriately.\n\nTable II illustrates a complication. Here the addition and assignment \nare embedded as an expression whose value is passed as an actual \nargument in a procedure call. Although the same addlI3 and movl \ninstructions appear together, the value in r0 is needed later and rO \ncannot be elided. In standard terminology, the value in register r0 is \nlive, or alternatively register r0 is busy. The improver can elide register \nusage only when the value in the register is known to be dead, or the \nregister is free.\n\nFor an arbitrary program, determining which registers are free at a \ngiven point requires a fair amount of work. The register usage and \nflow of control through any part of the program can effect whether or \nnot a register is busy in any other part of the program. Code generated \nby the portable C compiler has a property that makes busy/free \nanalysis much simpler. All registers are free any time the compiler \ngenerates a backward branch instruction. The portable C compiler \ngenerates code on line, completely translating the current expression \nor statement before proceeding to the following expression or state- \nment. The use of a temporary expression always occurs physically \nafter its generation. Thus the entire busy/free analysis can be done in \na single backward scan over the generated code. The backward scan \nmarks a register busy each time the register is read or used as a source \noperand. Some instruction occurring closer to the front of the file must \nhave put a live value into the register, or else the register would \ncontain garbage. Analogously, the backward scan marks a register free \neach time the register is written or used as a destination operand. \nSince the write destroys whatever used to be in the register, no one \ncould have wanted that dead value.\n\nThe backward scan must take precautions to record each use of a \ntemporary register, including the implicit uses. The return instruction \nret implicitly reads r0, the register in which C code returns function \nvalues. Thus r0 is busy just before each ret. The overall code-gener- \nation strategy of the compiler assumes that each procedure call instruc- \ntion calls writes all the temporary registers. Thus all the vemporary. \nregisters are free just before a procedure call.\n\nThe busy/free information can also be used to eliminate dead code. \nAn instruction that writes only into free registers does no useful work, \nexcept possibly for the side effects it causes. If the address computa- \ntions contain no side effects, then only the condition code could matter. \nThe condition code is set by each nonbranch instruction, so the \ncondition code itself is free unless the instruction which logically \nfollows is a conditional branch.\n\nThe backward scan must also be careful with code generated from \nconditional expressions. There can be no busy registers at the time of \na backward jump, as noted earlier. Since the compiler performs no\n\nPDP-11 vax-ll (but wrong) vAx-11 \nmov b, r0 \nadd c, r0 addl3 b, c, r0 \nmovr0,a movl r0,a addl3 b, c, a \nmov r0, (sp) pushl r0 pushl r0 \njsr pe, f calls $1, f calls $1, f\n\nTable III\u2014Translation of x =a?b:c; \ntestl a \njeql L100 \nmovl b, r0 \njbr L101 \nL100: movl c, r0 \nL101: movl r0, x\n\ninterstatement data-flow analysis (and in particular does not recognize \ncommon subexpressions), there can be no busy registers at the time of \na forward jump generated from an entire C statement. Since labels \nexist only because jump instructions branch to them, these two facts \nmight suggest that a register cannot be busy at any label, either. A \nregister can, however, be busy at a forward jump (and thus at a label) \nwith one of the values of a conditional expression. Table III illustrates \none such situation.\n\nEven though the instruction movl c,r0 writes r0, the register is \nbusy at the jbr because (if a is true) it contains the value of b to be \nstored into x. Thus the busy/free status of each register must be \nassociated with each label as the label is passed during the backward \nscan, and retrieved from the corresponding label at each jump. This \ncan be done efficiently by keeping a bit vector associated with each \nlabel, initializing all the bits to \u201cfree,\u201d and recording busy registers as \nlabels are passed. Because backward jumps have no busy registers and \nthe backward scan encounters the destination label of a forward jump \nbefore seeing the jump itself, the bits will always be correct.\n\nIn general the code improvements other than insertion of three- \naddress instructions and elimination of dead code by consulting the \nbusy/free information destroy the property that no temporary register \nis busy at a backward jump. This implies that using a single backward \nsweep over the code for the entire procedure to determine busy/free \nis valid only once, at the beginning before other improvements are \ntried. Fortunately, once is enough.\n\nThe backward prescan is also a good time to recognize hardware \nidioms. The vAx-11 has a number of instructions to set, clear, and test \nsingle bits, and to extract contiguous bit fields of arbitrary size. \nAppropriate uses of these instructions are often concealed in C with \nvarious Boolean or shift-and-mask operators or sequences of operators. \nComputing with the addressing modes by using instructions in which \nan address is used as an immediate operand often saves time and \nspace. Powerful addressing modes often depend heavily on register \nusage, and the backward pass is already computing this information. \nSince the backward scan is performed only once, time will not be\n\nwasted searching for hardware idioms more than once, as part of the \ngeneral iterative improvement strategy. Table IV gives some example \nimprovements.\n\nOn the vax-11, the control and data registers for input/output \ndevices lie in the memory address space. Programs manipulate the \nregisters in much the\u2019same way as they manipulate memory, and the \nassembly-language code for a device driver cannot be identified solely \nby its form. However, certain instructions and addressing modes do \nnot work properly when addressed to device registers. Generally these \nare exactly the instructions and addressing modes that the code \nimprover wants to introduce. For example, neither of the first two \nimprovements in Table IV is legal on a device register. Thus the code \nimprover must be told when it is improving the code for a device \ndriver, so it can avoid those improvements that cause problems. \nReading or writing a device register typically has side effects that are \ndifferent from reading or writing a memory location, and other hard- \nware considerations such as bus widths, circuit board area, or number \nof words of microcode are often important. Yet from a software \nviewpoint such special cases are irritating and error prone, and it \nwould be desirable to get rid of the complication.\n\nA single backward scan enables the code improver to determine \nregister usage and introduce three-address instructions where appro- \npriate. The backward scan takes advantage of the fact that all registers \nare free at each backward jump, a property that would otherwise be \nconsidered a weakness in the compiler. The single backward scan also \nrecognizes hardware idioms at a lower cost than previous algorithms.\n\nC Code Raw Translation Improved Translation \nint a; \na | = 0x8000; bisl2 $0x8000, a jbss $15, a, L100 \nL100: \nnee \nbie > 12) & OxF; ashl $\u201412, a, r0 extzv $12, $4, a, b\n\n1. S. C. Johnson, \u201cA Portable Compiler: Theory and Practice,\u2019 Conference Record \nFifth Ann. ACM Conf. Principles of Programming Languages, Tucson, AZ (Jan- \nuary 1978), pp. 97-104.\n\n2. B. W. Kernighan and D. M. Ritchie, The C Programming Language, Englewood \nCliffs, New Jersey: Prentice-Hall, 1978.\n\n4, A.V. Aho and J. D. Ullman, Principles of Compiler Design, Reading, Massachusetts: \nAddison-Wesley, 1977.\n\n5. W. Wulf, R. K. Johnsson, C. B. Weinstock, S. O. Hobbs, and C. M. Geschke, The \nDesign of an Optimizing Compiler, New York: Elsevier, 1975.\n\nCopyright \u00a9 1981 American Telephone and Telegraph Company \nTHE BELL SYSTEM TECHNICAL JOURNAL \nVol. 60, No. 2, February 1981 \nPrinted in U.S.A.\n\nRain Margin Improvement Using Resource \nSharing in 12-GHz Satellite Downlinks\n\nin this paper we consider ine effectiveness of sharing a small pool \nof reserved time slots of a Time-Division-Multiple-Access (TDMA) \nframe among a large number of ground stations to overcome rain \nfading. With this approach, a system dynamically assigns time slots \nfrom the reserved pool to ground stations experiencing fade depths \nabove the built-in margin. Powerful error correcting codes can be \nintroduced to occupy the extra time slots, providing 10 dB or more of \nextra fade margin. Because a large number of ground stations are \ncompeting for the limited reserved pool, blockage can occur if the \nnumber of simultaneous fades exceeds the maximum number that can \nbe accommodated. Some factors that influence the effectiveness of \nresource sharing are the mutual fade statistics at the various sites, \nthe traffic distribution within the network, the number of earth \nstations, the size of the reserved pool, and the rain outage objective. \nSince the mutual fade and traffic statistics are unavailable, we \ndevelop models that can be used to find a conservative bound on the \nrequired size of the reserved pool. The rain model accounts for \ndiurnal, seasonal, and geographical correlation among attenuation \nevents. Results for a maximum resource-sharing gain of 10 dB show \nthat reserving six percent of the time slots ensures a realized fade \ngain in excess of 9 dB for a down-link outage objective of 0.005 \npercent if there are more than 50 ground stations in the network, each \nwith two percent or less of the traffic.\n\nbuilt-in fade margin, and are relinquished when the fade event has \nended. Error-correcting coding is introduced to occupy the extra time \nslots, thereby reducing the carrier-to-noise ratio (CNR) required to \nmaintain the threshold bit-error-rate (BER). Low rain outage is there- \nfore achieved without radiating excessive down-link power. Not only \ndoes this conserve satellite power, but, also, interference into the \nsystems of other users of the geosynchronous orbit is minimized. In \nsuch an application, the operating speed of the decoder is much lower \nthan the transponder data rate by virture of the low TDMA duty cycle \nassociated with each ground station.\n\nBecause of the infrequency of simultaneous deep fading at multiple \nsites, a small pool of reserved time slots can often protect a large \nnumber of ground stations. The degree of protection so provided is the \nsubject of this current work; we shall evaluate the reduction in rain \nmargin required to achieve a given outage objective when all ground \nstations in the network are competing for a limited number of shared \nresources. We restrict our attention to the power-limited down-link \nsince up-link fading can usually be overcome by means of up-link \npower control.\u2019 A convolutional code yielding a maximum power saving \nof 10 dB is assumed throughout. Results are directly applicable to \neither a wide-area coverage system or a single-scanning beam system,\u201d \nbut the modeling and analytical approach can be extended to study \nTDMA systems that are fixed-beam satellite-switched, multiple scan- \nning beam, or hybrid-fixed scanning beam.**\n\nFigure 1 shows a typical sequence for interconnecting the various \nspot-beam footprints. Each interconnection contains one or more time \nslots during which ground stations within the connected regions com- \nmunicate on a sequential basis. Although all TDMA time slots can be \nmade available to accommodate normal network traffic demand, the \napproach taken here specifically reserves a certain number of time \nslots, shown at the end of the frame, for use exclusively during the rain \nfade events. By so doing, we have reduced the traffic-handling capa- \nbility of the system by a small percentage, while guaranteeing that\n\nFig. 1\u2014Typical switching frame showing the interconnections between the system \nground stations. An unused pool of reserved time slots is also shown.\n\nsome extra time slots will be available as needed to maintain reliability \nof circuits already in use when fade events occur.\n\nThe number of reserved slots necessary to provide a specified degree \nof protection at all stations is dependent upon several factors. Clearly, \nthe utility of resource sharing is dependent upon the joint fade statis- \ntics at multiple geographically remote sites; a small number of reserved \nslots cannot provide protection if the probability of simultaneous \nexcess attenuation at many sites, conditioned upon the occurrence of \nexcess attenuation at any one site, is high. Unfortunately, experimental \njoint-fade statistics at multiple remote sites are unavailable, and we \nmust resort to modeling to obtain quantitative results.\n\nThe utility of resource sharing also depends on the number of ground \nstations in the network and the traffic distribution among those ground \nstations, If the number of ground stations is small, then the fraction of \nTDMA time slots which must be reserved to protect even one site must \nbe large, resulting in an inefficient solution to the rain fade problem. \nAlso, if the traffic distribution is highly nonuniform such that a few \nground stations carry a disproportionately large volume of traffic, then \nagain it becomes impractical to reserve enough time slots to protect \nthis small number of large users. In such an event, it might be desirable \nto protect the large users by some other technique, such as larger \nantennas or site diversity, and employ resource sharing for the exclu- \nsive protection of the much larger number of small users.\n\nSimilarly, the effectiveness of resource sharing depends on the \ngeographical distribution of ground stations relative to the profiles of \nhigh rain-attenuation regions, and upon the volume of traffic carried \nby ground stations in high rain-attenuation regions. An additional \nfactor is the relationship between the busy hour, when all time slots \nnot reserved for resource sharing might be expected to be in heavy \ndemand, and the time-of-day occurrence of significant rain-attenuation \nevents.\n\nWe shall present both a multiple-site rain-attenuation model and a \npopulation-dependent traffic model, upon which is based the subse- \nquent predicted performance of the resource sharing concept. All \nassumptions implicit in this modeling are addressed in detail in Section \nII. Section III contains the mathematical analysis of outage based on \nthe rain and traffic models. Section IV contains numerical results of \nthis analysis; the effects of the various factors are presented paramet- \nrically. A typical result shows that for a 12/14-GHz network of 100 \nidentical ground stations, a shared resource reservation equal to six \npercent of the total transponder time slots will provide an outage of %- \nhour per year with 9 dB less rain margin than otherwise needed; \ndiurnal, seasonal, and geographical dependencies of joint rain-fade \nstatistics are accounted for in this prediction.\n\nH. JOINT RAIN ATTENUATION AND TRAFFIC MODELS \n2.1 Joint rain attenuation model\n\nFigure 2 shows statistics for single-site rain attenuation. It plots the \nfraction of time that rain fading exceeds the level of the abscissa \naveraged over one contiguous 12 month interval. This fraction can be \ninterpreted as the probability that any given rain-attenuation level is \nexceeded. From such single-site curves, we develop a model to be used \nfor predicting joint outages at multiple geographically remote sites.\n\nAt first glance, one might assume that if two sites are widely \n_ separated, then rain events at the two sites occur independently. \nHowever, this cannot be the case because in the 12-GHz satellite band, \nrain attenuation in excess of 5 dB is typically associated with thunder- \nstorm activity which produces intense rainfall. Periods of thunder- \nstorm activity are typically restricted to a four-month interval lasting \nfrom June through September, and to an interval of six hours each \nday lasting from 1 PM to 7 pM. Thus, if we are told that the rain \nattenuation at one of the sites is, say, 10 dB, then the probability of \nsimultaneous deep fading at the second site must be higher than its \nyearly average because, at that moment, we are likely to be in the \ninterval when thunderstorm activity normally occurs. Thus, knowledge\n\nFig. 2\u2014Typical single-site attenuation curve showing the fraction of time in one year \nthat the attenuation exceeds the abscissa.\n\nof rain attenuation at one site affects the probability of simultaneous \nattenuation at the second site, and the two events are not independent. \nSuppose, however, that we assume that all of the probability of Fig. \n1 is attributed to the period of thunderstorm activity. Then, if we \nrestrict our attention to this period only (it is only during this period \nwhen resource sharing is needed to combat fading), the probability \nthat attenuation exceeds a given level is about 12 times the yearly \naveraged value of Fig. 2 (4 out of 12 months per year, 6 out of 24 hours \nper day). Let pi(A) be the yearly averaged probability that attenuation \nexceeds level A at site No. 1. Thus, at a given instant of time \u00a2,\n\nIn (1), for the illustrative example given above, a = 12. Choosing a \nlarger value of a enhances the probability of multiple simultaneous \nfades since the yearly-averaged probability is then attributed to a \nnarrower interval. In what follows, we conservatively assume the \nthunderstorm periods to be 3 out of 12 months and 4 out of 24 hours, \nyielding a = 24. The factor a applies at all sites in the satellite network.\n\nNow, given that \u00a2 is within the thunderstorm-activity period, the \nevents of attenuation exceeding level A at site 1 and level B at site 2 \nmay be assumed to be independent if the two sites are widely sepa- \nrated. For example, knowledge that attenuation exceeds level A in \nNew York during the thunderstorm-activity period provides no infor- \nmation concerning the event that attenuation exceeds level B in \nDenver, since different independent storms are involved. Thus, for \u00a2 \nwithin the period of thunderstorm activity,\n\nwhere pi,2(A, B|t) is the probability that attenuation at site 1 exceeds \nlevel A and attenuation at site 2 exceeds B simultaneously at time \u00a2. \nEquation (4) is readily generalized for the case of an arbitrary number \nof widely separated sites.\n\nFor two closely spaced sites, another degree of attenuation event \ncorrelation is assumed beyond the seasonal and diurnal correlations \njust addressed. Here, attenuation at the two sites may be produced by \nthe same storm. We assume that, for closely spaced sites, all fades in \nexcess of some thunderstorm characteristic level Ao always occur \nwithin an H-hour interval of each other, where H is much larger than \nthe typical several minute duration of deep fades. Suppose that a fade \nof level A > Ao occurs at some site. Then, at a neighboring site, the \nprobability that a fade of level B > Ao is simultaneously occurring is \ngiven by C(B), a fade level dependent constant over the H-hour\n\nwindow. At this second site, the yearly averaged probability of a fade \nof level B > Ap is attributed exclusively to the H-hour intervals \nsurrounding the events A > Ao at the original site. Then,\n\nwhere T' is the number of hours in a year and x is the average number \nof events per year that the attenuation level exceeds Ao. Now,\n\nThus, defining 8 = T/\u00abHa, \npi2lA, Bt) = Ba\u2019pi(A)p2(B). (8a) \nFor L closely spaced sites, (8) generalizes to \nPi,...,t(Ar, As, +++, Az) = BY 'a'pi(Ar) +++ pr(Ax). (8b)\n\nIn the following, we will allow the \u201cgeographical correlation factor\u201d \nB to vary between 1 and 6. The extreme value 8 = 1 implies that \u00abHa \n= T, and that, within the thunderstorm period, fades occur indepen- \ndently. Recognizing that a is equal to the number of hours in a year \ndivided by the annual number of hours in the thunderstorm-activity \nperiod, we see that for 8 = 1, the average number of fades per year \nmultiplied by the uncertainty window H equals the number of hours \nin the thunderstorm-activity period. Hence, knowledge that a fade of \nlevel A > Apo is occurring at some site does not restrict the interval \nwithin the thunderstorm-activity period when a fade of level B > Ao \nmay occur at a neighboring site. Similarly, the extreme value B = 6 \nimplies, for example, that knowledge of deep fading at a given site \npinpoints the two out of four thunderstorm hours per day and the \naverage of one out of three days during the thunderstorm period when \nintense rainfall might occur in the region surrounding that site. Con- \nditioned upon an attenuation event at some site, the probability of an \nattenuation event at a second site within the surrounding region is \nthen six times higher than would be expected from diurnal and sea- \nsonal-correlation considerations alone.\n\nThe conservatism of the values a = 24 and 8 = 6 can be demonstrated \nby applying eq. (8a) to experimental attenuation data obtained with \nsite diversity such as appears in Ref. 7. Averaging (8a) over one full \nyear, we obtain\n\nFor attenuation greater than 5 dB, the site diversity measurements \nare somewhat more optimistic then predicted by eq. (9). This obser- \nvation, coupled with ground station separation much wider than used \nfor the site diversity measurements and the physical interpretations \ngiven to a and \u00a3 above, confirms the conservatism of the approach.\n\nTo apply the above rain-attenuation event model, we need to know \nthe yearly-averaged attenuation statistics at the location of each site \nin the network. The following simplification is invoked: We divide the \ncontinental United States into three regions such that the yearly \naveraged attenuation statistics that apply at a representative site in \none region are typical for all sites in that region (see Fig. 3). Los \nAngeles is chosen as representative of the western region (Region 1), \nwhere rain attenuation occurs infrequently. New York is chosen as \nrepresentative of the northeast central region (Region 2), throughout \nwhich rain attenuation is moderate. Finally, Atlanta is chosen as \nrepresentative of the southeast central region (Region 3), where rain \nattenuation occurs frequently. When applying the model, we make the \npessimistic assumption that the \u201cgeographical correlation factor,\u201d \u00a3, \napplies at all sites throughout an entire region; for sites located in \ndifferent regions this factor is neglected. Thus, for this model, \u201cclose- \nness\u201d of two sites is defined by whether both sites are within the same \nregion. The errors incurred because of sites located near each other \nbut on opposite sides of a regional boundary are more than offset by \nthe large \u201ccorrelation distances\u201d assumed.\n\nThe traffic model used in this analysis is based on a rank ordering \nof the 100 most populous continental United States cities, as shown in\n\nFig. 3\u2014Three regional maps of the United States. Regional boundaries are selected \nsuch that an attenuation curve for a representative ground station in each region applies \nthroughout that region.\n\nTable I. We assume that the traffic between any two of these cities is \ninversely proportional to the product of their indices. Traffic between \ntwo cities closer than 500 miles is excluded from the network. Under \nthese assumptions, Region 1 offers 24 percent of the total satellite \ntraffic, while Regions 2 and 3 offer 63 percent and 13 percent, respec- \ntively.\u2019\n\nThe number of ground stations serving a given region is assumed to \nbe proportional to the traffic offered by that region. For example, if \nthe network contains 100 ground stations, then 24 are located in \nRegion 1, 63 are in Region 2, and 13 are in Region 3. All ground \nstations are assumed to carry identical traffic cross sections. Thus, for \nthe purposes of analysis, the traffic differences among regions or among \ncities within each region are accounted for by assigning more or fewer \nidentical capacity ground stations, as the case may be, to accommodate \nthe traffic. The modeling and analysis can be extended to networks \ncontaining a mixture of large and small traffic ground stations if the \nlarge users are protected by site diversity and if resource sharing is \napplied to protect only the larger number of small users. This extension \nhas not been carried out, however, and the numerical results to be \npresented are valid only for the former case.\n\nTwo additional assumptions are also made. First, all ground stations \nin a given region are assumed to be identical, i.e., have the same built- \nin rain fade margin; the built-in margins of ground stations in different \nregions need not be the same, however. Second, we conservatively \nassume that the traffic-busy period, when all time slots not reserved \nfor resource sharing are in full-time use, coincides with the thunder- \nstorm-activity period. Thus, the results are intentionally made to be \npessimistic since it is precisely when resource sharing is needed that \nextra time slots not contained in the reserve pool are unavailable; \nduring traffic off-peak hours, when additional slots are available, \nresource sharing is not needed because rain fading does not occur.\n\nConsider the three-regional map shown in Fig. 3. Let there be N; \nground stations or sites located in Region 1, each of which has a built- \nin fade margin of A; dB. Similarly, let there be N2 and N3 sites located \nin Regions 2 and 3, respectively, with fade margins of Az and A; dB. \nThe extra fade margin provided by resource sharing if extra time slots \nare available is M dB, that is, M dB is the maximum extra margin \nprovided by the coding approach employed. All sites carry the same \nvolume of traffic, and enough time slots are reserved to accommodate \nK simultaneous fades. We wish to find the yearly-averaged probability \nthat a given site is operational when all sites in the network are \ncompeting, as needed, for the reserved resource-sharing slots. With no\n\nSan Francisco \nWashington \nBoston \nPittsburgh \nSt. Louis \nBaltimore \nCleveland \nHouston \nMinneapolis \nDallas \nSeattle \nAnaheim \nMilwaukee \nAtlanta \nCincinnati \nSan Diego \nBuffalo \nMiami \nKansas City \nDenver\n\nFt. Lauderdale \nGreensboro \nSalt Lake City \nAllentown \nNashville \nOmaha \nGrand Rapids \nYoungstown \nSpringfield \nJacksonville \nRichmond \nWilmington \nFlint\n\nTulsa \nOrlandoa \nFresno \nTacoma \nHarrisburg \nCharlotte \nKnoxville \nWichita \nBridgeport \nLansing \nMobile \nVentura\n\nYork \nBakersfield \nLittle Rock \nColumbia \nLancaster \nBeaumont \nAlbuquerque \nChattanooga \nTrenton \nCharleston \nBinghamton \nGreensville \nReading \nAustin \nShreveport\n\nloss in generality, we find this probability for a site located in Region \n1; a simple permutation of indices enables this result to be applied in \neither of the two remaining zones.\n\nAt any site within Region 1, three disjoint events may occur at any \npoint in time: (1) the fade depth F' may be less than Ai, (2) F may be \nbetween A; and A; + M, and (3) F may exceed A; + M. Call these \nevents \u00a3, Eo, and E3. Then\n\nClearly, \nP (operational | F,, \u00a2) = 1, (11) \nP (operational | E3, t) = 0, (12) \nP(E,|t) \n1 \u2014 ap(A)), t within thunderstorm period, \noe ; (13) \n1, otherwise, \nP(E2 | t) \n_ Jap(Ai) \u2014 ap(Ai+ \u2122), t within thunderstorm period, (14) \n~ 10, otherwise.\n\nThus, for \u00a2 within the thunderstorm period, \nP (operational | t) = 1 \u2014 ap(A,) \n+ aP (operational | \u00a32, t)[p(A1) \u2014 p(A1+ M)], (15) \nand for \u00a2 outside the thunderstorm period, \nP (operational |\u00a2) = 1. (16) \nFrom (15) and (16), we obtain the result that, averaged over an entire \n\u201cyear, \nP(not operational) = p(Aj) \n\u2014 P(operational | F2, t)[p(A1) \u2014 p(Ai+M\u2122)], (17)\n\nNow, for \u00a2 within the thunderstorm period, and conditioned on the \nevent E>, a particular site will be operational if the number of other \nsites which need to use the reserve pool is less than K \u2014 1. Also, if the \nnumber of other sites which need to use the pool is equal to j = K, \nthen the probability that a particular site is one of the K sites permitted \nto use the pool is equal to K/(j + 1).\n\nConditioned on the event EF; at a particular site, the probability that \ni; additional sites in Region 1 need to use the reserved pool, where \n0=1= N, \u20141, is given by\n\nConditioned on the event EF\u00bb at a particular site, the probability that \ni2 sites in Region 2 need to use the reserved pool, where 0 S iz = Np, is \ngiven by the unconditional probability during the thunderstorm-activ- \nity period since the events are assumed to be independent. For i2 = 1, \nwe find this probability by first defining the three events:\n\nLz: {an additional site in Region 2, not included in the particular set \nof L,, needs to use the reserved pool},\n\nThe probability that none of the sites in Region 2 need to use the \nreserved pool is given by\n\nClearly, conditioned on the thunderstorm period, the event {i2 sites in \nRegion 2 need to use the reserved pool} and the event {i3 sites in \nRegion 3 need to use the reserved pool} are independent, 0 S i2 = No, \nOsis 5 N3.\n\nReturning to (15), for \u00a2 within the thunderstorm period, the proba- \nbility that a particular site is operational, given attenuation between \nA; and A; + M dB, can be expressed as the union of the events \n(Vii nignigs}!\n\n{.4;,} = {i sites in Region 2 need to use the reserved pool}, (33) \n{.4;,} = {iz sites in Region 3 need to use the reserved pool}, (34)\n\n{S} = {the particular site has been assigned time slots from \nthe reserved pool}. (35)\n\nClearly, for ij \u00a5 i\u2019 or 12 # iz or 13 \u00a5 13, the events {V;j.i:,:,.} and \n{ Viv,ix,iz,s} are disjoint, and the probability of the union of events (31)\n\nwhere P(i; in Region 1| #2), P(i2 in Region 2| E2), and P(i3 in Region \n3| #3) are given, respectively, by eqs. (18), (25), and (28).\n\nSubstituting (37) into (17) yields the desired result for the yearly- \naveraged probability that a particular site is unavailable versus the \nbuilt-in fade margins Aj, Ao, and A3.\n\nWe now apply the model of Section II and the analysis of Section \nITI to investigate the utility of resource sharing for a 12-GHz satellite. \nInitially, we shall assume a satellite location of 100 degrees West \nlongitude. A total of 100 ground stations is assumed, and the thunder- \nstorm-activity factor a = 24. Results are obtained for various values of \nK and f\u00a3. The yearly-averaged attenuation data used for Regions 1, 2, \nand 3 are based upon S. Lin\u2019s attenuation model\u00ae \u00ae for converting long- \nterm rain-rate data, obtained from the United States Weather Bureau, \ninto attenuation predictions. Figure 4 shows derived plots for Los \nAngeles (Region 1), New York (Region 2), and Atlanta (Region 3). \nSensitivity of the results to regionally dependent built-in fade margins, \nnumber of ground stations, thunderstorm-activity factor, and satellite \norbital position are investigated later on. A coding gain of 10 dB is \nassumed.\n\nFigure 5 shows predicted values of yearly fractional outage with \nresource sharing for a site in Region 2 versus the built-in rain margin,\n\nFig. 4\u2014Yearly average 12-GHz attenuation curves for Los Angeles (Region 1), New \nYork (Region 2), and Atlanta (Region 3). The satellite is at 100\u00b0 W longitude.\n\nassumed to be common at all ground stations (i.e., all ground stations \nhave the same size antenna). The network contains 100 ground sta- \ntions, and the \u201cgeographical correlation factor,\u201d 8, is assumed equal to \nunity, that is, within a given region, the probability of attenuation \nduring the thunderstorm period conditioned upon an attenuation event \nat any site is equal to the unconditional probability. The number of \nsimultaneous fades which can be accommodated, K, varies between 1 \nand 3. Similar curves for B = 2, 4, and 6 appear in Figs. 6, 7, and 8, \nrespectively.\n\nFigures 5 through 8 show that resource sharing is of increasing \nutility as the outage objective becomes more stringent. For example, \nfor K = 2, B = 2, the fade margin gain is 8.8 dB for a down-link outage \nof 0.01 percent, and increases to 9.8 dB for an outage of 0.005 percent. \nWe see also that the ability to accommodate only a single fade (K = \n1) restricts the fade margin resource-sharing gain at a down-link outage \nobjective of 0.005 percent to 9.6 dB under the best of conditions (8 = \n1). For less favorable conditions (8 = 6), the gain shrinks to 7.6 dB. \nThus, the ability to accommodate only a single fade may severely \ninfluence the utility of resource sharing. However, if two simultaneous \nfades can be accommodated, then even for the unfavorable condition\n\nB = 4, the fade margin gain is 9.4 dB; whereas for 8 = 1, the gain \nbecomes 10 dB. If three simultaneous fades can be accommodated, \nthen, even for the extremely unfavorable case 8 = 6, a fade margin \ngain of at least 9.5 dB can be achieved. For this last case, the outage \nobjective of 0.005 percent can be achieved with a built-in margin of 8.8 \ndB, to compare against 18.4 dB required in the absence of resource \nsharing.\n\nFigure 9 shows results obtained in Region 3. Again, a 12-GHz \nsatellite located at 100 degrees West longitude and 100 identical ground \nstations, all with a common fade margin, are assumed. Curves in this \nillustration apply for B = 1, 2, 4, and 6 and for K = 2, 3, and 4, and \nshow that at an outage objective of 0.005 percent, the fade margin gain \nis 10 dB for all cases considered. This outage objective can be achieved \nwith a built-in margin of 11.3 dB, to be compared against 21.3 dB \nrequired in the absence of resource sharing.\n\nFig. 5\u2014Rain outage experienced with resource sharing at a site in Region 2 (northeast) \nversus the built-in fade margin. A maximum possible power saving of 10 dB is assumed. \nTime-of-day and seasonal thunderstorm activity factor a = 24. The geographical factor \n\u00a3 = 1. (Attenuation events at different sites within the thunderstorm activity period are \nindependent.) The satellite is at 100\u00b0 W longitude, and the network contains 100 identical \nground stations.\n\nFig. 6-\u2014Same as Fig. 5, except the geographical factor 8 = 2. (Given an attenuation \nevent at one site within the thunderstorm activity period, it is twice as likely that a \nsimultaneous attenuation event occurs at any other site within the same region.)\n\nof 11.3 dB is provided at all sites in Regions 1, 2, and 3. From Figs. 5 \nto 8, we see that in Region 2, a built-in fade margin of 11.3 dB provides \nan outage lower than the required 0.005 percent. (No results are given \nfor Region 1 because it was found that the objective of 0.005 percent \nwas always readily achieved by virtue of the low level of attenuation \nprevalent in that region). Thus, we are motivated to consider a case \nwherein sites with different size antennas are deployed in different \nregions. The goal here is to optimize the system (i.e., provide the \nsmallest antenna possible in each region) such that a down-link outage \nobjective of precisely 0.005 percent is achieved everywhere.\n\nWe consider a system wherein the antenna gain at each site in \nRegion 3 exceeds the antenna gain for Region 2 by 2 dB, and the \nantenna gain at each site in Region 1 is less than the antenna gain in \nRegion 2 by 2 dB. The results are shown in Fig. 10. Again, a 12-GHz \nsatellite at 100 degrees West longitude and a network of 100 ground \nstations are assumed. Outage is plotted as a function of the built-in\n\nmargin in Region 2. Parameters are K = 2 or 3, B = 1 or 4. We see that \nif K => 3, the outage objective of 0.005 percent can be achieved in \nRegion 3 if the Region 2 built-in margin is 9.3 dB, even for B = 4. At \nthis level, the outage actually achieved in Region 2 is 0.0044 percent \nfor B = 4, K = 3; an outage of 0.005 percent could have been achieved \nwith a built-in margin of 8.8 dB (within 0.05 dB of that needed to \nachieve 0.005 percent in Region 3). Thus, this system is close to \noptimum, except that again, the outage achieved in Region 1 is far \nlower than the objective. This indicates that the antennas used in \nRegion 1 are still far too large if outage is the only consideration.\n\nFor the case of a 500-MHz satellite transponder radiating a power \nlevel of 30 watts and a satellite aperture of 15 feet diameter, it has \nbeen estimated that a 5 meter diameter earth-station antenna would \nbe needed in Region 2 to provide a rain margin of 16 dB.\u201d Extrapolating \nto the parameters of Fig. 10 by holding the satellite power, bandwidth \nand aperture fixed, we find that an antenna diameter of 2.5 meters \nwould be needed in Region 2 to provide a built-in margin of 9.3 dB; \nthe Region 1 antenna diameter assumed in Fig. 10 is 2 dB smaller or \n2 meters. Thus, it may be impractical to deploy a smaller antenna in\n\nRegion 1 without seriously interfering with other satellites in the \ngeosynchronous orbit. Rather, it might be more advantageous to \nmaintain a larger-than-needed antenna in Region 1 to enable a reduc- \ntion in the required uplink transmitter power. In summary, the illus- \ntration portrayed in Fig. 10 is nearly optimum in a practical sense, and \nthe outage objective of 0.005 percent is achieved everywhere by pro- \nviding antennas which yield a built-in margin of 7.3 dB in Region 1, \n9.3 dB in Region 2, and 11.3 dB in Region 3. The capability to \naccommodate K = 3 simultaneous fades is required for a network with \n100 ground stations.\n\nIn Fig. 11, we consider a network containing 200 ground stations and \nplot the achievable outage in Region 2 versus the built-in margin, \nassumed to be the same at all 200 ground stations. Results for geo- \ngraphical factors 6B of 1, 2, and 4 and K = 2 to 5 appear. Because the \nnumber of ground stations has increased by a factor of two above \nprevious cases considered, the number of simultaneous fades which \nmust be accommodated is generally higher to achieve the same re- \nsource-sharing advantage. However, this factor is generally smaller\n\nK =NUMBER OF SIMULTANEOUS FADES \nWHICH CAN BE ACCOMMODATED \nA ee ae Ln a \n0 4 5 10 15 20 \nBUILT-IN FADE MARGIN IN DECIBELS\n\nFig. 9\u2014Rain outage experienced with resource sharing at a site in Region 3 (southeast) \nversus built-in fade margin. The geographical factor B = 1, 2, 4, and 6. Other conditions \nare the same as for Fig. 5.\n\nFig. 10\u2014Rain outage experienced with resource sharing for sites in Regions 1 (west), \n2 (northeast), and 3 (southeast) versus built-in fade margin at sites in Region 2. Sites in \nRegion 1 have 2 dB less built-in fade margin than sites in Region 2. Sites in Region 3 \nhave 2 dB more fade margin than sites in Region 2. The geographical factor 8 = 1 and \n4, Other conditions are the same as for Fig. 5.\n\nFig. 11\u2014Rain outage experienced with resource sharing at a site in Region 2 (north- \neast) versus built-in fade margin. The network contains 200 identical ground stations, \nand \u00a3 = 1, 2, and 4. Other conditions are the same as for Fig. 5.\n\nthan two, implying that the resource-sharing overhead, expressed as \nthe required number of reserved slots divided by the total number of \naccesses or interconnections, decreases as the number of users is \nincreased.\n\nSimilarly, as shown in Fig. 12, the number of simultaneous fades \nwhich must be accommodated to achieve a given level of resource- \nsharing advantage decreases for a network containing 50, rather than \n100, ground stations. Again, results for Region 2 are shown, and a 12- \nGHz satellite located at 100 degrees West longitude along with a \ncommon fade margin at all sites are assumed. The resource-sharing \noverhead is generally higher for the 50 station network than for the \n100 ground station network.\n\nFigure 13 is a composite plot for Region 2 showing the resource- \nsharing fade margin gain at an outage of 0.005 percent versus the \nrequired overhead for 8 = 1, 2, and 4 and for a total number of 50, 100, \nand 200 ground stations. The four-for-one time slot expansion of Ref. \n1, which allows use of a rate 4 convolutional code to provide a coding \ngain of 10 dB along with extended synchronization preamble, is as- \nsumed. Letting the number of ground stations be represented by N, \nthe overhead n, expressed as a percentage, is given by\n\nFig. 12\u2014Same as Fig. 11, except that the network contains 50 identical ground \nstations.\n\nAgain, a 12-GHz satellite located at 100 degrees West longitude is \nassumed. A similar composite plot can be derived for Region 3. We see \nthat for B = 4, an overhead of 4.3 percent will ensure a fade margin \ngain greater than 9 dB if the network contains 200 ground stations. An\n\nFig. 13\u2014Composite plot for a site in Region 2 (northeast) showing the resource- \nsharing fade margin gain at an outage of 0.005 percent versus the TDMA overhead for \nnetworks of 50, 100, and 200 ground stations. A maximum possible power saving of 10 \ndB is assumed. Time-of-day and seasonal thunderstorm activity factor a = 24 and the \ngeographical factor B = 1, 2, and 4. The satellite is at 100\u00b0W longitude.\n\nFig. 14\u2014Composite plot for a site in Region 2 (northeast) showing the resource \nsharing fade margin gain at an outage of 0.005 percent and a TDMA overhead of 5.66 \npercent versus the number of ground stations in the network. A maximum possible \npower saving of 10 dB is assumed. Time-of-day and seasonal thunderstorm activity \nfactor a = 24 and the geographical factor B = 1, 2, and 4.\n\noverhead of 5.66 percent is needed to achieve similar results for a \nnetwork of 100 ground stations. Of course, the required overhead in all \ncases is reduced if smaller values of 8 apply.\n\nFigure 14 plots a family of curves for a site in Region 2, showing the \nresource-sharing fade margin gain, for an outage of 0.005 percent, as a \nfunction of the number of ground stations in the network. The over- \nhead is kept constant at 5.66 percent (K = 1 for 50 ground stations, K \n= 2 for 100 ground stations, and K = 4 for 200 ground stations), and \nthe family parameter is 8. Again, we see that resource sharing becomes \nmore effective as the number of ground stations in the network \nincreases.\n\nIn Fig. 15, we investigate the effect of the thunderstorm-activity \nperiod factor, a, on the utility of resource sharing. For these curves, we \nassume that a = 12, rather than the value of 24 used for all previous \nresults, implying that the period of heavy attenuation is concentrated \nover an interval twice as large in time. Thus, at any site, the probability \n- that attenuation exceeds some value A, conditioned on the thunder- \nstorm period, is half its former value. Again, a 12-GHz satellite located \nat 100 degrees West longitude is assumed, and results apply at a \nground station in Region 2. A total of 100 ground stations is assumed, \nand all ground stations have a common built-in fade margin. We see \nthat, at an outage of 0.005 percent, the ability to accommodate two \nsimultaneous fades provides a fade margin advantage in excess of 9.8 \ndB for \u00a3 as high as four. Thus, under the less pessimistic and more \nrealistic assumption that a = 12 (4 out of 12 months/year, 6 out of 24 \nhours/day) rather than 24, we see that the resource-sharing advantage \nis very close to the maximum possible for an overhead of about six\n\nFinally, in Fig. 16, we plot the yearly average attenuation statistics \n(based upon Lin\u2019s model) for sites in Los Angeles (Region 1), New \nYork (Region 2), and Atlanta (Region 3) for a 12-GHz satellite located \nat 130 degrees West longitude rather than 100 degrees West as assumed \nbefore. Because of the greater slant range to the satellite, the atten- \nuations experienced at sites in Regions 2 and 3 are greater; again, \nRegion 1 experiences very little attenuation. We use these statistics to \nderive the curves of Fig. 17 which show the fractional outage versus \nthe built-in margin for a site in Region 2 and for various values of K \nand ~. A total of 100 ground stations of common fade margin is \nassumed, and a = 24. Figure 17 shows that for 6 = 4, capability for \naccommodating only one fade is sufficient to provide 10 dB of extra \nprotection via resource sharing at an outage objective of 0.005 percent. \nBecause the yearly averaged attenuation statistics are nearly the same \nin Regions 2 and 3 for this satellite longitude, it is safe to assume that \nFig. 17 is reasonably representative of a site in Region 3 as well. Also, \nbecause of this similarity between Regions 2 and 3, it appears that \nresource sharing is more effective for a satellite location at 130 degrees\n\nFig. 15\u2014Rain outage experienced with resource sharing at a site in Region 2 (north- \neast) versus the built-in fade margin. Time-of-day and seasonal thunderstorm activity \nfactor a = 12, and the geographical factor Pe = 1, 2, 4, and 6. Other conditions are the \nsame as for Fig. 5.\n\nFig. 16\u2014\u2014Yearly average 12-GHz attenuation curves for Los Angeles (Region 1), New \nYork (Region 2), and Atlanta (Region 3). The satellite is at 130\u00b0 W longitude.\n\nFig. 17\u2014Rain outage experienced with resource sharing at a site in Region 2 (north- \neast) versus the build-in fade margin. The satellite is at 130\u00b0W longitude, and the time- \nof-day and seasonal thunderstorm activity factor a = 24. Other conditions are the same \nas for Fig. 5.\n\nWest, compared against 100 degrees West, although the required built- \nin margin is much greater. This is because, when we position the \nsatellite at 100 degrees West, the higher attenuation experienced in \nRegion 3 relative to Region 2 presents more competition for the \nreserved time slots, which is disadvantageous for a site in Region 2.\n\nIn this paper, we have attempted to evaluate the effects of the size \nof the reserved pool, the number of ground stations, correlation among \nattenuation events, rain outage objective, and satellite position on the \nutility of resource sharing to provide 10 dB of extra down-link margin \nagainst rain fading. A single wideband satellite transponder such as \nmight be associated with a scanning spot-beam system was assumed, \nbut the joint rain attenuation and traffic modeling and the analytical \napproach can be extended to study multiple spot beam frequency reuse \nconcepts. In general, it was found that for the cases examined, a pooled \nresource overhead equal to six percent of the available TDMA time slots \nis adequate to ensure a fade margin gain in excess of 9 dB for an outage \nobjective of 0.005 percent if there are more than 50 ground stations in \nthe network. The difference between the actual gain realized and the \nmaximum gain of 10 dB possible with the coding approach employed \narises from the effects of many ground stations in competition for a \nsmall number of shared resources.\n\nThe traffic model employed for this study assumes that all ground \nstations carry similar traffic cross sections. Although the analytical \napproach can be modified to reflect the effects of users with different \ntraffic cross sections, such an approach might become numerically \nunwieldy if the number of user classes assumed becomes too large. \nHowever, a simple overbound on the required overhead can readily be \nobtained if for a given number of users, we apply the results obtained \nby assuming a smaller number of users. For example, suppose the \nnetwork contains 200 ground stations, and the amount of traffic carried \nby the ground stations are within a factor of four of each other. Then, \nthe required \u2018overhead calculated for a network of 50 ground stations \nis an overbound of that needed for the 200 ground station network.\n\nThe predicted results depend, of course, on the rain model assumed; \nmodeling is necessary because statistics of multiple fade events for a \nlarge number of geographically dispersed sites are unavailable. Fortu- \nnately, in a TDMA system designed on the resource-sharing concept, it \nis a simple matter to alter the size of the reserved pool in response to \nreal operating experience.\n\nThree final observations will be offered. First, it appears from these \nresults that resource sharing is most effective in the southeastern \nportion of the United States because the traffic demand from that\n\nregion is smaller than that presented by other regions. This is indeed \nfortunate because it is precisely in that region, where attenuation is \nhigh, that resource sharing can offer the most saving. Second, if the \nsatellite receives a western orbital-slot assignment, resource-sharing \nagain assumes an important role because of the large rain margins \n(and, therefore, the large satellite radiated power or large ground \nstations) required to provide a suitably low rain outage. Finally, we \nnote that resource sharing can provide high-system reliability (low- \nrain outage) while maintaining a sufficiently low-satellite effective \nradiated power to satisfy interference constraints imposed by the \npresence of other users of the geosynchronous orbit. It is perhaps in \nthis last regard that resource sharing will prove most valuable, by \nenabling coexistence of spot-beam systems with existing wide-area \ncoverage systems.\n\n1. A. S. Acampora, \u201cA Shared Resource TpMa Approach to Increase the Rain Margin \nof 12/14 GHz-Satellite Systems,\u201d B.S.T.J., 58, No. 9 (November 1979), pp. 2097- \n111.\n\n. D. O. Reudink and Y. S. Yeh, \u201cA Scanning Spot Beam Satellite System,\u201d B.S.T.J., \n56, No. 8 (October 1977), pp. 1549-60.\n\n. A. S. Acampora and B. R. Davis, \u201cEfficient Utilization of Satellite Transponders via \nTime-Division Multibeam Scanning,\u201d B.S.T.J., 57, No. 8 (October 1978), pp. 2901- \n14.\n\n4. A. S. Acampora, C. Dragone, and D. O. Reudink, \u201cA Satellite System with Limited- \nScan Spot Beams,\u201d IEEE Trans. Commun., Com-27, No. 10 (October 1979), pp. \n1406-15.\n\n. D. O. Reudink and Y. S. Yeh, \u201cThe Organization and Synchronization of a Switched \nSpot-Beam System,\u201d Fourth Int. Conf. Digital Satellite Commun., Montreal, \nOctober 1978.\n\n6. D. O. Reudink, A. S. Acampora, and Y. S. Yeh, \u201cMethods for Achieving High \nCapacity Universal Service Satellites,\u201d Nat. Telecom. Conf., Birmingham, Decem- \nber 1978.\n\n. S. H. Lin, H. J. Bergman, and M. V. Pursley, \u201cRain Attenuation on Earth-Satellite \nPaths\u2014Summary of 10-Year Experiments and Studies,\u201d B.S.T.J., 59, No. 2 \n(February 1980), pp. 183-228.\n\n. S. H. Lin, \u201cNationwide Long-Term Rain-Rate Statistics and Empirical Calculation \nof 11-GHz Microwave Attenuation,\u201d B.S.T.J., 56, No. 9 (November 1977), pp. \n1581-1604.\n\n. S. H. Lin, \u201cEmpirical Rain Attenuation Model for Earth-Satellite Paths,\u201d IEEE \nTrans. Comm., COM-27, No. 5 (May 1979).\n\nCopyright \u00a9 1981 American Telephone and Telegraph Company \nTHE BELL SYSTEM TECHNICAL JOURNAL \nVol. 60, No. 2, February 1981 \nPrinted in U.S.A.\n\nModeling Multipath Fading Responses Using \nMultitone Probing Signals and Polynomial \nApproximation\n\nWe show in quite a general way that highly accurate modeling of \nmultipath fading responses is possible using low-order complex poly- \nnomials. This applies to all terrestrial radio systems in the chan- \nnelized common carrier bands below 15 GHz, where channel widths \nare 40 MHz or less. The context of the study is a new multipath \nexperiment being conducted in New Jersey over a 23-mile path at 11 \nGHz. The transmitted signal consists of up to nine tones in a 40-MHz \nbandwidth. These tones are coherently processed, sampled, and dig- \nitized in the receiver and recorded, during fading events, for later \noff-line reductions. Simple routines can be used to determine poly- \nnomial coefficients from these recorded data. This paper describes \nthe signal processing and data reduction methods and analyzes them \nto assess the accuracy of polynomial fitting. The analysis uses a \nmean-square error measure and assumes a representative form for \nthe underlying response function. Our results predict that the vast \nmajority of multipath fading responses can be accurately approxi- \nmated over bandwidths of 40 (62) MHz using first- (second-) order \ncomplex polynomials.\n\nMultipath fading (hereafter abbreviated MPF) on terrestrial micro- \nwave paths can be a major cause of outage in digital radio systems.'\u201c \nNumerous efforts have been aimed at understanding, analyzing, and \ncorrecting this source of disruption, and some have led to new statis- \ntical models for MPF responses.\u201d\n\nThe particular model that inspires the present work approximates \nthe MPF response by a low-order complex polynomial in frequency.\u00ae \nFor a particular 26-mile path in Georgia, it was shown that a first- \norder polynomial suffices to characterize the fading response in a 25-\n\nMHz band centered near 6 GHz. The joint probability distribution for \nthe polynomial coefficients was derived for that path, thus permitting \na complete statistical description of the MPF response.\n\nNow another experiment is being instrumented, this time for a 23- \nmile path in New Jersey operating in the 11-GHz band. The aim of the \nnew experiment is to add to, and in several respects improve upon, the \ndata base used to quantify the earlier polynomial model. The improve- \nments include higher measurement signal to noise ratios (SNRs), higher \nsampling rates (20 measurements per second rather than 5), coherent \nprocessing to obtain phase information (previously absent), and a \nwider measurement bandwidth (40 MHz rather than 25 MHz).\n\nGiven the highly variable nature of multipath fading, such improved \nmeasurements for a new path in a different frequency band and locale \nshould add importantly to our knowledge of this phenomenon.\n\nThe basic design of the experiment can be simply stated: As many \nas nine coherently-phased tones within a 40-MHz bandwidth are \ntransmitted from Murray Hill and coherently demodulated in a re- \nceiver at Crawford Hill; the demodulated tones are sampled, digitized, \nand screened by a desktop computer/controller; and the digitized data, \nif deemed interesting, are recorded on magnetic tape for later off-line \nprocessing.\n\nThe recorded data will be in a form that facilitates polynomial \napproximation using simple, efficient computer routines. The data will \nbe quite general in form, however, i.e., amenable to modeling via any \nmathematical approximation considered promising.\n\nThe present study evaluates the accuracy of polynomial approxi- \nmation, relating it to the experiment parameters and to the methods \nof signal processing and data reduction. Section II describes the signal \nprocessing in the transmitter and receiver, and derives signal and noise \nrelationships used in the subsequent error analyses. Section III de- \nscribes the methods of polynomial fitting to be considered, and defines \nthe mean-square error measures that will be used to evaluate them.\n\nSection IV analyzes the errors in the polynomial fitting caused by \nmeasurement noise, and Section V analyzes the errors caused by finite \nsampling of the frequency response. In general, the errors increase \nwith the bandwidth over which the fitting is done. In analyzing the \nerrors caused by finite sampling, we assume a general form for the \nMPF response function that has been applied successfully in other data- \nfitting studies,\u201d and assume either worst-case or typical values for the \nfunction parameters.\n\nThe mean-square error calculations permit predictions of the maxi- \nmum bandwidths for which polynomial fitting is valid. Section VI \nsummarizes the results computed, under rather stringent mean-square \nerror requirements, for polynomial orders of one, two, four, six, and \neight.\n\nli. SIGNAL PROCESSING ANALYSIS FOR THE MPF EXPERIMENT \n2.1 Propagation path and radio channels\n\nMultipath fading responses are to be measured on a 23-mile path \nbetwen Murray Hill and Crawford Hill, similar to the one used by \nCrawford and Jakes in their earlier experiments.\u201d The transmitting \nantenna at Murray Hill is 655 feet above sea level, and the receiving \nantenna atop Crawford Hill is 425 feet above sea level. An experimental \nlicense has been obtained to operate over this path in three 40-MHz \nchannels within the 11-GHz common carrier band. These channels are \ncentered at 11.465, 11.545, and 11.625 GHz. The initial measurements \nwill be in the band centered at 11.545 GHz.\n\nThe transmitted signal is created by the two-stage upconversion of \na baseband signal having the form \nN/2 \nb(t) =do+ > d, cos(nAwt + On), (1) \nn=1 \nwhere N is even and the other parameters will be discussed. The up- \nconversion places the signal in an RF channel centered at radian \nfrequency w. = 27f. ( f- = 11.545 GHz). Hence, the transmitted signal \nis\n\nwhere p, is the power of the nth transmitted tone and a total of \nN + 1 tones are transmitted. From (1), we see that po is proportional \nto d\u00e9 and that pp (n \u00a5 0) bears the same proportionality to d? /4.\n\nThe variation of p, with n is clearly symmetrical about n = 0 because \nit derives from amplitude settings of the baseband tones. Nonuniform \nvariations of p, can easily be compensated for via baseband adjust- \nments in the receiver. In Section IV we consider nonuniform variations \nfor which receiver noise effects are minimized.\n\nThe frequency spacing between transmitted tones, Af, may be 5, \n10, or 20 MHz. Since the transmission is confined to a channel of 40- \nMHz width, and occupies a bandwidth NAf, we have the constraints \n(N + 1) = 9 tones when Af = 5 MHz; (N + 1) <= 5 tones when Af = 10 \nMHz; and (N + 1) = 3 tones when Af = 20 MHz. We will consider the \nfour particular combinations N = 2, Af = 20 MHz; N = 4, Af = 10 \nMHz; N = 6, Af = 5 MHz; and N = 8, Af = 5 MHz.\n\na common 5-MHz reference, and so can be relatively phased in any \nmanner desired. For purposes of analysis, we will assume all 0,\u2019s to be \nzero here; since any phasings in the transmitter are easily undone in \nthe receiver, no generality is lost. One criterion for choosing the actual \n#,\u2019s is minimization of the peak factor of the RF signal (2). The \nbaseband phase adjustments that accomplish this have been derived \nfor N = 2, 4, 6, and 8.\"\u2019 We will use the resulting minimized peak \nfactors in making noise calculations later.\n\n2.3 Response of the propagation medium \nWe denote the complex signal gain of the propagation medium by\n\nGain magnitude (go) during nonfading \nThe quantity go can be computed from familiar radio path equations. \nNote that w is measured from the center of the channel. During \nnonfading periods, | F(w)| = 1 throughout the channel bandwidth; \nduring multipath fading, F'(w) varies with w in a randomly time- \nvarying manner.\n\nThe function F'(w) contains two phase factors of no interest to us. \nOne is exp(jo), where \u00a2o is the phase shift through the medium at w \n= 0; the other is exp(\u2014/jwt,), where \u00a2, is the nominal propagation time \nalong the path. (For a 25-mile path, tf, = 0.13 ms.) The investigation of \nmultipath fading can be simplified, with no loss of information, by \nremoving these two factors. Thus the function of interest to us is\n\nThe aim of our modeling effort is to find suitable functions for approx- \nimating H(w), and to statistically characterize the parameters of those \nfunctions.\n\nWe will see in Section 2.5 that the response function actually \nsampled by the measurement system is\n\nG(w) = Pwexp| =i(\u00a5+ on)| (5) \nAw \nwhere \u00a5 and @ are the (possibly) random or unknown phases of \nfrequency references in the receiver. To obtain samples of the desired \nfunction [H(w)] from samples of the measured function [G(w)] will \ntherefore require performing the operation\n\nat each of the sampling frequencies [w = 0, +Aw, ---, + (N/2)Aw]. We \nwill show later how to accomplish this in the data processing.\n\nWe demonstrate here a useful decomposition for complex response \nfunctions such as F'(w), G(w), and H(w). We will treat only H(w), \nnoting that the same mathematics and notation apply to F(w) and \nG(w).\n\nSince w is measured from an arbitrary microwave frequency (27f.), \nthere is no physical reason to assume complex conjugate symmetry for \nH(w). In its most general form, H(w) can be expressed as\n\nwhere H.(w) and H,(w) are each functions having complex conjugate \nsymmetry. Accordingly, we can write\n\nReal, Imaginary, Real, Imaginary, \nEven Odd Even Odd \nBy transmitting and coherently receiving N + 1 tones spaced by Aw, \none could in theory obtain measurements of the two even functions \nat w = 0, Aw, --+, (N/2) Aw; and of the two odd functions at w = Aw, \n-, (N/2) Aw. [The total number of samples, 2(N + 1), is consistent \nwith measuring the amplitudes and phases of the N + 1 received \ntones.| In reality, the receiver obtains these samples for the corre- \nsponding set of G functions, which differ from the H functions if y and \n@ are not both 0. Obtaining H samples from G samples i is discussed in \nSection 2.6. \nAnother departure of the receiver outputs from the desired samples \nis the presence of measurement noise. We will defer the introduction \nof noise until Section 2.7.\n\nwhere | F,| and \u00a2, are the magnitude and phase of F(nAw); go is the \nnormal (nonfading) path gain, (3); and we have used (2) with all 9,\u2019s \nassumed to be zero.\n\nThe signal goes through a two-stage down conversion which amounts \nto quadrature demodulation. That is, two baseband outputs are ob- \ntained which correspond to mixing Vr(t) with 2 cos(w-t + W) and with \n\u20142 sin(w-\u00a2 + Wy). A nonzero value of y signifies that the RF and IF\n\nreferences in the receiver are not in phase synchronism with those in \nthe transmitter.\n\nEach of the baseband signals consists of a dc component plus \nsinusoids at w = Aw, ---, (N/2)Aw. The nth sinusoid in each of these \nsignals goes through quadrature demodulation, via the local references \ncos[n(Awt + @)] and \u2014sin[n(Awt + 6@)], to produce two more dc \noutputs. These references are all derived from a 5-MHz source in the \nreceiver, and nonzero @ signifies that this source is not in phase \nsynchronism with the one in the transmitter.\n\nUsing (5) and ordinary trigonometric identities, the following state- \nments can be proven:\n\nThus, the de outputs of the receiver correspond to evenly-spaced \nfrequency samples of the four G functions Gar(w), Gai(w), Gor(w), and \nG,i(w). Each of the 2(N + 1) dc components is low-pass filtered, with \na noise bandwidth of 100 Hz; time-sampled 20 times per second and \nquantized with 14-bit precision by a Datel System 256 acquisition \nsystem;* and passed to a HP 9845A computer for real-time screening.\n\nAny set of samples that seems interesting, or is part of a sequence that \nseems interesting, is recorded on magnetic tape for subsequent off-line \nprocessing.\n\nBy using (6), the H\u2018functions (H2,-(w), Hai(w), etc.) can be easily \nobtained from the corresponding G functions once A\u00ae and AT are \nspecified. Defining a new function G(w; A\u00ae) 4 G(w) exp(jA\u00ae), we first \nperform the matrix operation\n\nAn identical equation relates Gai(w; A\u00ae) and G,;(w; A\u00ae) to Gai(w) and \nG:;(w). The H functions are then obtained from these new G functions, \nfor specified AT, as follows:\n\nThe operation indicated by (13) leads to the result H,,(0) = 0, ie., \nthe phase response of the function to be analyzed is forced to zero at \nw = 0. [To see this, combine (6), (12) and (13) for w = 0.] This is a \nwelcome simplification in the data and entails no loss of useful infor- \nmation. Fortunately, sin A\u00ae and cos A@ are readily obtained from the \nmeasured G samples, for \nG,(0) cos A\u00ae = ene I (15)\n\nBy way of contrast, the value of AT to use in (14) is not so readily \nspecified or determined. Yet, to get the full benefit of polynomial \nmodeling (i.e., accurate fitting using low-order functions), AJ\u2019 must be \ncarefully chosen.* We have arrived at a criterion for choosing AT \nbased upon the following data reduction procedure: For a given AT, \n(14) is applied and the resulting H samples are fitted by a finite-order \ncomplex polynomial in jw. We consider that value of AT to be optimal \nfor which the polynomial fitting is best, in some least-squares sense\n\n* To see why, consider the example G(w) = exp(\u2014jwfo), where f ~ 1/B, and B is the \nbandwidth over which H(w) = G(w) exp(jwAT) is to be fitted. If AT = 0, H(w) = \ncos wty \u2014 j sin wf, and high-order polynomials are needed for accurate fitting over the \npar ne If AT = t&, however, H(w) = 1 + 70, and low-order polynomials are quite \nsufficient.\n\ndefined later. In Section 5.4, we will identify a data-derived measure \nthat accurately predicts the optimal AT.\n\nWe have shown that the dc receiver outputs are proportional to \nfrequency samples of the function G(w), and that the unwanted phase \nfactors that distinguish G(w) from H(w) can be removed in the data \nprocessing. Not so readily removed are the noises associated with the \ndigitized outputs. These consist of both additive Gaussian noise from \nthe input and components of the receiver and quantizing noise from \nthe 14-bit analog-to-digital conversions.\n\nWe shall now assume that each of the dc outputs in (10) to (12) is \nadjusted by a factor 1/(go V2pn); n = 0, N/2, before being digitized. \nAccordingly, the variance of the Gaussian noise associated with a given \noutput sample is\n\nTable I defines the quantities in (16) and gives values for each. Using \nthose data and assuming uniform tone powers, we obtain the following\n\nTable I\u2014System parameters used in noise analysis \nParameter Definition and Assumed Value(s)\n\nBo Power gain between transmitter output and first receiver amplifier stage; includes \nclear-air path loss and circuit and waveguide losses: \n10 log &o= = \u201474 dB \nkT Thermal noise density at receiver input: \n10 log kT = \u2014174 dBm/Hz \nb Noise bandwidth of receiver processing for each tone: b = 100 Hz \nNr Noise figure of first receiver amplifier stage: \n10 log Nr = 2 dB \nPp Peak transmitter power, constrained to meet out-of-band emission requirements: \n10 log P, = 14 dBm \nPr _\u2014 Transmitter peak factor, minimized via phase adjustments of the baseband tones \nbefore upconversion. For uniform tone powers,\n\n10 log P= 7.4dBm \nPn Averge power in transmitted tone a ene stnAw from center. \nFor uniform tone powers and N =\n\nresult: o2 in dB, for n # 0, lies in the range \u201479 dB + 2 dB, where the \nprecise value depends on N; for n = 0, o@ is 3 dB higher.\n\nAssuming the dc outputs are amplitude-adjusted as indicated, the \ninput to every 14-bit A/D conversion is precisely a sample of a G- \nfunction. During normal propagation, the G samples lie within + 1 [see \n(3) and (5)]. To provide some room for excess gain, we assume \nquantizer amplitude limits set at + 1.60 (4-dB margin). As a result, the \nquantizing error for each digitized sample can be characterized as an \nadditive noise uniformly distributed on [\u2014A/2, A/2], where A = 2 x \n1.6 x 274 = 1.95 x 10-*. The quantizing noise variance is then\n\nComparing this with o@ above, we find justification for ignoring quan- \ntization effects. Alternately, they can be accounted for using an ap- \nproximate correction factor given in Section IV.\n\nThese are the quantities produced by (13) and (14) for any specified \ncombination of A\u00ae and AT. We account for the noisiness of the H \nsamples via the notation\n\nHarn A Harn + Carns Horn & Hb,,.n + Corny etc., (19) \nwhere far,n, for,n, etc., are the random noise samples. Since the \u00e9s are \nproduced by phase rotations of the noises associated with the G \nsamiples, they are all Gaussian, zero-mean, and mutually independent, \njust like the original noises. Moreover, their variances are identical to \nOG, as given by (16). We thus have an accurate, simple description for \nthe noisiness of the data to be processed.\n\nill. POLYNOMIAL FITTING AND ERROR MEASURES \n3.1 Fitting polynomials to the H samples\n\nThe implicit assumption of the polynomial fitting approach is that, \nover some finite bandwidth B centered on f = 0, the response function \nH(w) can be accurately approximated by a low-order complex poly- \nnomial, 1.e.,\n\nM \nH(w) = H(w) 2 Ys (Ar +JBi)(Jw)*,  |wo| Ss 7B. (20) \nk=0 \nUsing (7) and (8), we can break this representation down as follows: \nM M M \nHi(w) = \u00a5 Ar(jwo)? = Y An(jo)* + Y Ax(Jo)*, (21) \nk=0 h=0 k=1 \nEven Odd \nFit to Har(w) Fit to jHailw)\n\nThe A,;\u2019s and B,\u2019s are slowly varying random coefficients; in any given \nmeasurement interval, they collectively characterize the short-term \nfrequency response of the propagation medium.\n\nThe fittings indicated above can be done, for every 50-ms measure- \nment interval, by using the 2(N + 1) H samples obtained in that \ninterval. The way the fitting is done depends on the values of M and \nNN. We now consider three possible cases.\n\nThe H samples obtained using N + 1 tones can be fitted precisely \nusing an Nth-order complex polynomial. Thus, when M = N, fitting \nconsists of matching each summation in (21) and (22) to the appropri- \nate H samples at the sample frequencies. The resulting equations for \nthe A;\u2019s are as follows (identical equations apply to the B,\u2019s, with \nAy,\u2019s and H;,\u2019s replacing the Ha,\u2019s and Hui\u2019s):\n\nwhere D{,, and D?m are the (J, m)th elements of the N/2 x N/2 \nmatrices [D\u00b0] and [D\u00b0], respectively; [D\u00b0] and [D\u00b0] are the inverses \nof the matrices [d\u00b0] and [d\u00b0], respectively; and the (m, /)th elements \nof [d\u00b0] and [d\u00b0] are\n\nThe matrices [D\u00b0] and [D\u00b0] for N = 2, 4, 6, and 8 are given in Table \nII. Note for future reference that the derived A;\u2019s and B;\u2019s are weighted \nsums of the H samples.\n\nSince N can be as high as eight, this method of fitting suggests the \npossibility of eighth-order polynomial modeling. Earlier studies, how- \never, suggest that this order is unnecessarily high for bandwidths of 20 \nto 40 MHz.\u00ae\u201d Reductions using M = N = 8 may therefore involve \nexcessive demands on data storage and analysis, and unduly complicate\n\nmodel development and usage. For this reason, we also consider the \ncombinations M = N = 6, Af = 5 MHz; M = N = 4, Af= 10 MHz; and \nM = N = 2, Af = 20 MHz.\n\nAnother possibility is to assume a low-order polynomial while using \nall nine tones, in which event the fittings indicated in (21) and (22) are \ndone by least-squares methods. Compared with the case M = N = 2, \nthis approach requires more data storage and analysis. On the other \nhand, it leads to better fitting accuracy and protects against loss of \ntones caused by equipment problems. Noise effects can be slightly \nworse for this case, despite the noise-averaging it affords, because the \npower per tone must be lower to satisfy the transmitter power con- \nstraint.\n\nThe least-squares polynomial fitting method is well known,\u201d and so \nwe give only the results:\n\nThe equations for Bo, B,, and Bz are identical, with Hy,,, and Hoi.n \nreplacing Harn and Hain. Again, the A,\u2019s and B,\u2019s are weighted sums \nof the H samples.\n\nThe case M = 1 represents the ultimate in modeling simplicity, \nnamely, a response function which is first order in jw. Strong experi- \nmental evidence for such a polynomial has been reported.** Assuming \nleast-squares fitting to nine tones, we find that\n\nA, is given by (27); and the same equations give Bo and B,, with Ho,n \nand H;;,, replacing Ha, and Hain. \n' 3.2 Error measures\n\nThe MPF response function, as reconstructed from the data-derived \nA,\u2019s and B;\u2019s, is\n\nSince the coefficients are weighted sums of noisy H samples, (19), we \ncan write H(w) in the form\n\nH(ww)= Aww) + Zw) (31) \nWeighted sum Weighted sum \nof true of random \nH samples noise samples\n\nWhereas H(w) is noiseless, it is not identical to the true response \nfunction, H(w), but is a polynomial approximation. \nWe now define the error function\n\nIf B is some bandwidth about f. over which H(w) is to be characterized, \nthen the mean-square error for a given polynomial fit can be defined \nas\n\nA useful normalizing quantity for these mean-square errors is the \nmean-square gain of the fading channel, i.e.,\n\nLet us now interpret B as the bandwidth over which the intended \nreceiver output (undistorted by MPF) has a roughly uniform power \nspectrum, and outside which the spectral content is small. Accordingly, \ne\u2019 accurately represents the mean-square error in predicting the output \nsignal using H(w) for the MPF response and H\u201d accurately represents \nthe true mean-square output. It is thus reasonable to say that a given \npolynomial fit is valid if e*(noise)/H\u201d and e\u201d(approx)/H\u201d are both 10~\u00b0 \nor less. Since \u20ac?(noise) is independent of H(w), and since H\u201d is seldom \nlower than 10, our requirement for e*(noise) is that it be 107\u201d or less.\n\nWe compute e*(noise) by forming Z(w) in (31), using (30) and the \ngoverning formulas for A; and B;. For simplicity, we use the first-order \nmodel (Case 3: M = 1, N = 8) to exemplify the approach.\n\nwhere 6Ao and 6A; are the noises in Ap and A;, which are computed \nusing (27) and (29); and similar definitions and equations apply for 6Bo \nand 6B,. There noises are related to the noise components fa,,0, ar,1, \netc., in the H samples of (27) and (29). Thus,\n\nand similarly for 6Bo and 6B,. We now proceed in the obvious way: \n6Ao, 6Ai, etc., are combined with (35); | Z(w) |? is formed; and the \nintegration in (33) is performed to obtain e*(noise). In doing so, we \nmake use of the statistical independence among all the noise samples. \nWe also use (16) for their variances. The result is of the form \nkTON\u00a2r *\u201d S,(B/Af)\n\nFor all of the other cases considered, the analytical approach and \nthe form of the result are as shown above; only the specific equations \nfor S,(B/Af) vary.\n\nThe final step in computing e7(noise) is to specify the values of the \ntransmitted tone powers. We have considered two approaches, namely \n(i) assuming all p,\u2019s to be equal, adding up to some specified value (P) \nof average transmitted power; and (ii) choosing the p,\u2019s so as to \nminimize \u20ac7(noise), subject to the same average power specification. \nIn the first approach, we apportion power according to the rule\n\nee 4 \n\u2014 N+1\u2019 \nIn the second approach, we use the method of Lagrangian multipliers \nto minimize (37), subject to the constraint\n\nwhere 5, is the Kronecker delta function. Note that the optimal \nvariation of p, with n depends on the bandwidth ratio, B/Af.\n\nWe have reduced (37) to numerical results using the parameter \nvalues in Table I. We now present our findings.\n\nFigure 1 shows curves of e\u2019(noise) vs. B for the various possibilities \nunder Case 1. We can make the following observations:\n\n(zt) The noise errors and their rate-of-growth with B depend strongly \non the polynomial order M, both being less for lower M. From a noise \nstandpoint then, M should be chosen to be low. The competing factor \ninfluencing this choice is the approximation error, which we consider \nlater.\n\n(it) Optimizing p, does not improve significantly on the use of \nuniform tone powers. There is no compelling reason, therefore, to \ntaper the p,\u2019s. On the other hand, doing so entails no price in com- \nplexity and may offer benefits, e.g., tapering of p, may reduce out-of- \nband spurious tones caused by transmitter nonlinearities. We will not \nexplore this topic here, but present in Table III some examples of \noptimal p, variations.\n\n(iii) Values of \u20ac*(noise) less than 10~\u2019 can be attained, using either \nuniform or optimal p,\u2019s, for bandwidths up to 34 MHz or more, \ndepending on M. For purposes of modeling, errors of this magnitude \ncan be regarded as negligible, as noted in Section 3.2.\n\nFig. 1\u2014Mean-square noise errors, \u20ac\u201d(noise) in dB, for Case 1 (M = N = 2, 4, 6 or 8). \nFor each N, the curve for optimal tone powers (dotted) merges with the one for uniform \ntone powers (solid).\n\nFigure 2 shows curves of e\u201d(noise) vs. B for Cases 2 and 3. These \nresults show the effects of noise when first- and second-order polyno- \nmials are formed via least-squares fitting to nine received tones. For \npurposes of comparison, Fig. 2 also repeats the results for M = N = 2 \nunder Case 1, i.e., second-order matching to three received tones. We \nnote the following: .\n\n(t) As before, fitting H(w) with a lower-order polynomial leads to \nsmaller noise-related errors.\n\nTable IIl\u2014Optimal tone powers for Case 1, B = 40 MHz (in dB \nabove power of central tone)\n\nFig. 2\u2014Mean-square noise errors, \u20ac7(noise) in dB, for Cases 2 and 3 (M = 2, N = 8 \nand M = 1, N = 8). Results shown are for uniform tone powers; results for optimal tone \npowers are lower by less than 0.5 dB.\n\nsometimes smaller when N = 2 and sometimes smaller when N = 8. \nThe explanation lies in two offsetting factors: Using nine tones leads \nto less power per tone, but affords averaging benefits not available \nwhen using three tones. Depending on B, one or the other of these \nfactors dominates.\n\n(vit) For all three low-order polynomial approaches considered, \ne*(noise) is less than 107\u2019 for bandwidths up to 90 MHz or more.\n\nWe emphasize that e*(noise) in (37) scales readily in the quantities \nbNr/Pg\u00e9 and B/Af. Therefore, the curves of Figs. 1 and 2, however \nparticularized to the present experiment parameters, can easily be \nscaled to reflect other conditions.\n\nWith this in mind, we can account in a simple and fairly accurate \nway for both quantization noise (previously neglected) and changes in \nP, go, b, and Nr. We assume the quantization and Gaussian noise \npowers to be additive, invoke (16) and (17), and obtain the following \nrule: If bNr/Pgo is changed by a factor C from the one obtained using \nTable I, then e?(noise) should be adjusted by the factor (C + 0.25). \nFor C = 1 (no change in bNr/Pg>), quantizing noise adds roughly 25 \npercent (1 dB) to the mean-square noise error.\n\nThe approximation error defined in (32) results from sampling H(w) \nat a finite number of discretely-spaced frequencies. The mean-square \nerror, as defined in (33), depends on B, Af, M, N, and\u2014unlike the noise \nerror\u2014on the specific function being approximated. We consider a \nhighly accurate polynomial approximation to be one for which 7, \ndefined by\n\nThe basic form assumed for H(w) is the one used by W. D. Rummler \nin previous studies.\u201d To fit measured amplitude responses in a 25.3- \nMHz bandwidth near 6 GHz, Rummler expressed H(w) by the function \na[1l \u2014 b exp(j@) exp(\u2014jwAr)], where a, b, 0, and Ar are function \nparameters to be chosen. For each_of 24,920 response measurements, \nthese four parameters were chosen to give a least-squares functional \nfit to the data. Error analyses of the results (Fig. 17 of Ref. 5) revealed \na high degree of fitting accuracy over nearly all of the data records. \nFor this reason, we use the same function here to represent the actual \nunderlying response of multipath fading.\n\nRummler also analyzed the effect of fixing the delay parameter Ar \nat 6.31 ns, and choosing (a, b, @) to fit each data record. While not as \ngood in all cases, the \u201cfixed delay\u201d model was found to be highly \naccurate over at least 98 percent of the data base. With this in mind, \nwe conjecture that the above function, with Ar <= 10 ns, can be used to \nrepresent the vast majority of MPF responses. We will invoke this \nconjecture later in tabulating bandwidths over which polynomials of \nspecified order can be used to model multipath fading.\n\nwhere A\u00ae and AT were defined in Section II. As noted there, the data \nreduction process chooses A\u00ae in such a way that H(0) is made real, \nand chooses AT in such a way that 7 is minimized for a given Mth-\n\norder polynomial fit. Accordingly, we specify that exp(jA\u00ae) = (1 \u2014 \nbe~/*)/V1 + b* \u2014 26 cos@; search over AT so as to minimize 7 for a \ngiven fitting polynomial and given (0b, @); and search over b and @ so as \nto maximize that minimized 7.* Thus, we obtain the measure\n\nThe worst-case combination of (b, @) for any B, At, and polynomial \norder is found to be (1.0, 0.0). This combination corresponds to total \nfading at the band center (w = 0). Also, the optimal AT for b = 1.0 is \nprecisely Ar/2.\n\nThe variations of no with BAt, with M and B as parameters, are \nshown in Fig. 3. As expected, 7 is a strongly decreasing function of M, \na strongly increasing function of BAt, and a weakly increasing function \nof B alone for given BAr.\n\nA useful empirical formula for these results, accurate to within 40 \npercent for B = 80 MHz and BAr \u20ac 1.2, is\n\nwhere a = 0.57 and k(M, B) is a fairly mild function of M and B. \nNumerical values for k(M, B) are given in Table IV.\n\nCurves of 7o for these cases are shown in Fig. 4, where the results for \nM = N = 2 are repeated for comparison purposes. There is an evident \nimprovement when second-order polynomials are derived using nine \ntones instead of three. An empirical formula for these results (again, \naccurate to within 40 percent for B = 80 MHz and BAr <= 1.2) is given \nby (45), where a = 0.32; the power of (BAr) is replaced by 4; and \nnumerical values for k(M, B) are given in Table IV.\n\nIn minimizing 7 with respect to AT, (44), we used computer search \nprocedures and exploited our assumed knowledge of the underlying \nH(w). In an actual measurement the latter is not possible, as the only \ninformation available for optimizing AT is the set of measured G \nsamples.\n\n* The amplitude factor a has no impact on fitting accuracy and so is set to unity for \npurposes of this study.\n\nFig. 3\u2014Relative mean-square approximation errors, jo in dB, for Case 1 (M = N = 2, \n4, 6, or 8). The variable is BAt, where Ar is the delay spread quantity in Eq. (43). \nBandwidth, B, and M are parameters.\n\nBy experimenting with different strategies, we have identified a \ncomputationally efficient procedure that uses these samples to opti- \nmize AT. It consists of computing C2 4 (Aj + B3) as a function of AT \nand choosing that AT for which C2 is a minimum. The procedure is \neven simpler than it might seem; for, if A,,. is An computed for AT = \n0, then A2 for any other AT is just\n\nAo = Aoo + AtoAT + %Ao,.AT\u201d\u2019, (46) \nand an identical relationship applies to B,. These results can be\n\nThus, \u201coptimizing\u201d AT amounts to finding the minimum of a fourth- \norder polynomial in that quantity. This can be done very efficiently\n\nFig. 4\u2014Relative mean-square approximation error, 40 in dB, for Cases 2 and 3 \n(M = 2, N = 8 and M = 1, N = 8). Same variable and parameters as in Fig. 3.\n\nusing computer algorithms. Our investigation of this procedure con- \nsisted of minimizing C2 with respect to AT for numerous (0b, 6, Ar), and \nusing the resulting values of AT to compute yn. We obtained remarkably \nsimilar answers, in most cases, to those obtained with AT rigorously \noptimized [i.e., via computer search using the known function H(w)]. \nThe worst departures occur when 7 is either so large as to be of no \ninterest or so small that the error increases do not matter.\n\nOther measures derivable from the data, such as [(A2 \u2014 AZ)? + (B \n\u2014 Bi)*], may be even more appropriate Quantities such as this, or C2 \nas defined above, are reliable indicators of the \u2018curvature\u2019 in H(w); \nchoosing At to minimize them should therefore maximize the fitting \naccuracy of low-order polynomials. Our results give empirical support \nfor this principle.\n\nThe major findings of this study are summarized in Table V. The \nfirst row gives, for each combination of M and N considered, the \nbandwidth below which polynomial fitting yields e?(noise) < 107\u2019. \nThese bandwidths are derived for uniform tone powers and the param- \neter values listed in Table I. They cannot be increased much without \ndramatic (and unlikely) improvements in transmitter power, noise \nbandwidths, and quantizing precision.\n\nThe second row of Table V gives the corresponding bandwidths \nbelow which 7 <= 107\u00b0. They are derived for the response function (43), \nwith 5 and @ having worst-case values and Ar = 10 ns. The third row \ngives the maximum bandwidth which satisfy both mean-square error \nrequirements. Since these requirements are quite stringent, we present \nin the fourth row the results of relaxing both of them by 6 dB (factor \nof four).\n\nThese tabulations suggest that the vast majority of MPF responses \ncan be accurately approximated, over bandwidths of 40 (62) MHz, by\n\nTable V\u2014Bandwidths, in MHz, over which H(w) can be accurately \ncharacterized\n\nNotes: \n(1) Results assume Art = 10 ns. \n(2) Results in excess of 80 MHz are extrapolated estimates.\n\ncomplex polynomial functions of just first (second) order. We surmise, \nthen, that the appropriate modeling approach is to form first- and \nsecond-order polynomials using nine-tone measurements (N = 8) and \nleast-squares data fitting. With these low orders, there is maximal \nsimplicity in developing and using the model, with no substantial \nsacrifice in accuracy. By using nine tones, moreover, polynomials of \nhigher order (up to eight) can also be deduced from the data, thereby \nenabling the theoretical predictions given here to be fully tested.\n\nVil. ACKNOWLEDGMENT \nWe are grateful for the helpful comments of W. D. Rummler.\n\n1. W. C. Jakes, Jr., \u201cAn Approximate Method to Estimate an Upper Bound on the \nEffect of Multipath Delay Distortion on Digital Transmission,\u201d IEEE Trans. \nCommun., COM-27, No. 1 (January 1979).\n\n2. L. J. Greenstein and V. K. Prabhu, \u201cAnalysis of Multipath Outage with Applications\n\nin 90-Mb/s PSK Systems at 6 and 11 GHz,\u201d IEEE Trans. Commun., COM-27, No. \n1 (January 1979). \nC. W. Anderson, S. Barber, and R. Patel, \u201cThe Effect of Selective Facing on Digital \nRadio,\u201d Int. Conf. Commun. 1978, Conf. Record, 2 (1978), pp. 33.5.1-6. \n. C. W. Lundgren and W. D. Rummler, \u201cDigital Radio Outage Due to Selective \nFading-Observation vs. Prediction from Laboratory Simulation,\u201d B.S.T.J., 58, No. \n7 (May-June 1979), pp. 1073-100. \n. W. D. Rummler, \u201cA New Selective Fading Model: Application to Propagation Data,\u201d \nB.S.T.J., 58, No. 7 (May-June 1979), pp. 1037-71. \n6. W. D. Rummler, \u201cExtensions of the Multipath Fading Channel Model,\u201d Int. Conf. \nCommun. 1979, Conf. Record, 2 (1979), pp. 32.2.1-5. \nL. J. Greenstein and B. A. Czekaj, \u201cA Statistical Model for Multipath Fading \nae Responses,\u201d Int. Conf. Commun. 1979, Conf. Record, 2 (1979), pp. 32.1.1-\n\n.L. Greenstein and B. A. Czekaj, \u201cA Polynomial Model for Multipath Fading \nChannel Responses,\u201d B.S.T.J., 59, No. 7 (September 1980), pp. 1197-225.\n\n9. W. D. Rummler, \u201cAdvances in Multipath Channel Modeling,\u201d Int. Conf. Commun. \n1980, Conf. Record, 3 (1980), pp. 52.3.1-5.\n\n10. A. B. Crawford and W. C. Jakes, Jr., \u201cSelective Fading of Microwaves,\u201d B.S.T.J., 31, \nNo. 1 (January 1952), pp. 68-90.\n\n11. L. J. Greenstein and P. J. Fitzgerald, \u201cPhasing Multitone Signals to Minimize Peak \nFactors,\u201d (to be published).\n\n12. L. J. Greenstein, \u201cA Multipath Fading Channel Model for Terrestrial Digital Radio \nSystems,\u201d IEEE Trans. Commun., COM-26, No. 8 (August 1978), pp. 1247-50.\n\n13. J. R. Green and D. Margerison, Statistical Treatment of Experimental Data, New\n\nCopyright \u00a9 1981 American Telephone and Telegraph Company \nTHE BELL SYSTEM TECHNICAL JOURNAL \nVol. 60, No. 2, February 1981 \nPrinted in U.S.A.\n\nThis paper describes the Quality Measurement Plan (QMP), a \nrecently implemented system for reporting the quality assurance audit \nresults to Bell System management. QMP replaces the T-rate system, \nwhich evolved from the pioneering statistical work of Shewhart and \nDodge during the 1920\u2019s and 1930\u2019s at Bell Laboratories. Box and \nwhisker plots are used for graphically displaying confidence intervals \nfor the quality of the current production. The confidence interval is \ncomputed from both current and past data and is derived from a new \nBayesian approach to the empirical Bayes problem for Poisson \nobservations. Here we discuss the rationale, mathematical deriva- \ntions, dynamics, operating characteristics, and many comparative \nexamples. We show that QmP reduces statistical errors relative to the \nearlier T-rate system.\n\nThe responsibility of the Bell Laboratories Quality Assurance Center \n(QAC) is \u201cto ensure that the communications products designed by Bell \nLaboratories and bought by Bell System operating companies from \nWestern Electric Company, Incorporated will meet quality standards \nand will perform as the designers intended.\u201d\u2019 This obviates the need \nfor each operating company to carry out its own acceptance inspection.\n\nTo meet this responsibility, the QAc works with its Western Electric \n(WE) agents, the Quality Assurance Directorate (QAD),\u201d and Purchased \nProducts Inspection (PPI) organizations. However, as stated in Ref. 1, \n\u201cThe primary responsibility for quality lies with the line organizations: \nBell Laboratories for the quality of design and Western Electric for \nthe quality of manufacture, installation, and repair.\u201d The quality \nassurance organizations conduct independent activities to assure qual- \nity to the operating companies.\n\nThe quality assurance organizations have two major activities. The \nfirst is to conduct quality audits where products change hands, either \nwithin WE or between WE and the operating companies. Examples \nare manufacturing, installation, and repair audits. The second concerns \na collection of field quality monitoring activities. Examples are the \nProduct Performance Surveys. These are designed sample surveys of \nreported field troubles.\n\nAn audit is a highly structured system of inspections done on a \nsampling basis. The ingredients of an audit are: (t) sampling method, \n(it) scope of inspection, (iii) quality standards, (tv) nonconformance \nprocedures, (v) defect assessment practices, (vi) quality rating method, \nand (vii) report formats.\n\nThe sampling method along with the scope of inspection determines \nwhat tests will be performed on what units of product or attributes of \nproduct. The statistics and economics of sampling, the engineering \nrequirements, and the field effect of defects play the central roles in \ndetermining the sampling and the scope of inspection.\n\nThe quality standards are numerical values expressed in defects, \ndefectives or demerits per unit. They are set by the QAc in consultation \nwith the QaD. The standards are target values, reflecting a tradeoff \nbetween first cost and maintenance costs.\n\nThe nonconformance procedures are rules for detecting and dispos- \ning of audited lots that are excessively defective with respect to a \nparticular set of engineering requirements.\n\nThe defect assessment practices are a set of transformations that \nmap defects found into defects assessed for quality rating purposes. A \nterminal strip may have all ten connections off by one position, but, \nthe consequences of these ten defects found are much less than ten \nindependent occurrences of this defect. Therefore, less than ten defects \nare assessed.\n\nThe quality rating method and report formats determine how the \nresults of the audit are presented to Bell System management. For \nexample, a product is reported as \u201cBelow Normal,\u201d when it fails a \nstatistical test of the hypothesis that the quality standard is being met.\n\nThe statistical foundations of the audit ingredients were developed \nby Shewhart, Dodge, and others, starting in the 1920\u2019s and continuing \nthrough to the middle 1950\u2019s. This work was documented in the \nliterature in Refs. 3 to 6.\n\nIn recent years, research has been carried out to evaluate the \napplication of modern statistical theories to the audit ingredients. An \nimportant idea is summarized in an article by Efron and Morris\u2019 which\n\nexplains a paradox discovered by Stein.\u00ae When you have samples from \nsimilar populations, the individual sample characteristics are not the \nbest estimates of the individual population characteristics. Total error \nis reduced by shrinking the individual sample characteristics part way \ntowards the grand mean over all samples. Efron and Morris used \nbaseball batting averages to illustrate the point. But the problem of \nestimating percent defective in quality assurance is the same problem. \nAnd you are always concerned with similar populations\u2014for example, \nthe population of design-line telephones produced for each of several \nmonths.\n\nThis idea was originally explored in Ref. 9. The idea has now evolved \ninto the Quality Measurement Plan (QMP). QmP is the recently imple- \nmented system for conducting three of the audit ingredients: defect \nassessment, quality rating, and quality reporting.\n\nAs a quick introduction to QmP, consider Fig. 1. This is a comparison \nof the QmpP reporting format (Fig. la) with the old T-rate reporting \nformat (Fig. 1b). Each year is divided into eight periods. On the \nbottom, the J-rate is plotted for each period and it measures the \ndifference between the observed and standard defect rates in units of \nsampling standard deviation (given standard quality). The idea is that \nif the T-rate is, e.g., less than negative two, then the hypothesis of \nstandard quality is rejected. Section II considers the exact rules for \nexception reporting under the T-rate system.\n\nUnder amp, a box and whisker chart is plotted each period. The box \nchart is a graphical representation of the posterior distribution of \ncurrent population quality on an index scale. The index value one is \nthe standard on the index scale and the value two means twice as \nmany defects as expected under the quality standard. The posterior \nprobability that the population index is larger than the top whisker is \n0.99. The top of the box, the bottom of the box, and the bottom \nwhisker correspond to the probabilities 0.95, 0.05, and 0.01, respec- \ntively.\n\nThe heavy \u201cdot\u201d is a Bayes estimate of the long run process average; \nthe \u201ccross\u201d is the observed value in the current sample; and the \u201cdash\u201d \nnear the middle of the box is the posterior mean of the current \npopulation index and is called the Best Measure of current quality. \nThe process averages, \u201cdots,\u201d are joined to show trends.\n\nAlthough the T-rate chart and the qmpP chart often convey similar \nmessages, there are differences. The Qmp chart provides a measure of \nquality; the T-rate chart does not. For example, in 7806 (Period 6 of \n1978) both charts imply that the quality is substandard, but the QMP \nchart also implies that the population index is somewhere between one \nand two.\n\nFig. 1\u2014@mP versus the T-rate. The box and whisker plot in (a) is the @mMP replacement \nof the T-rate. One is standard on the index scale; two is a defect rate of twice standard. \nThe box and whisker are 90 and 98 percent confidence intervals for production during \nthe period; the \u201ccrosses\u201d are the indices in the samples; the \u201cdots\u201d are process averages; \nand the \u201cdashes\u201d in the middle of the boxes are Best Measures of current quality derived \nfrom empirical Bayes theory. (b) is a time series of T-rates. Each point measures the \ndifference between observed quality and expected quality on a standard deviation scale. \nNotice that the sixth period of 1977 and the fourth period of 1978 are the same in the T- \nrate chart but quite different in the QmpP chart.\n\nuses the past sample indices, but makes an inference about current \nquality. The 7-rate system uses runs criteria based on attributes of \nthe T-rate, such as \u201cless than zero,\u201d and can make an inference about \npast quality. In Fig. 2, 7707, the T-rate signals an exception, because \nsix T-rates in a row are less than zero, indicating that quality has not\n\nbeen standard for all six periods. But for qomp, the standard is well \nwithin the box, indicating normal current quality. The different treat- \nment of past data is also illustrated in Fig. 1. Comparing 7706 with \n7804 reveals very similar T-rates, but QmMP box charts with different \nmessages.\n\nThe T-rate system is based on the assumption that the total number \nof defects in a rating period has a normal distribution. QmP is based on \nthe Poisson distribution. This difference is important for small audits, \nas shown in Section VII.\n\nQMP was on trial for two years and was applied to 20,000 sets of \naudit data. The relatively simple QmP algorithm published in Ref. 9\n\nFig. 2\u2014A weak T-rate exception. The seventh period of 1977 was reported as a quality \nexception because six 7-rates in a row were less than zero. For QMP, it would have been \nreported as normal. This is because QMP provides a statistical inference about current \nproduction only, even though past data is used.\n\nwas used originally. This simple algorithm worked for most data sets, \nbut not all (e.g., zero defects in every period). The relatively complex \nalgorithm discussed in Section IV is the result of a lengthy fine-tuning \nprocess, designed to make the algorithm work for every case. This is \nwhy the full power of Bayes theorem with empirically based prior \ndistributions had to be used.\n\nNote that in the Qmp box chart, the Best Measure always lies \nbetween the estimated process average and the current sample index. \nThe Best Measure is a shrinkage of the sample index towards the \nestimated process average. In 7706 of Fig. la, the shrinkage is away \nfrom standard; but, in 7804, it is towards the standard.\n\nThe Best Measure is related to the class of estimators described by \nEfron and Morris.\u201d\u00b0 In the cited reference, they provide a foundation \nfor Stein\u2019s paradox with an empirical Bayes approach. In Ref. 7, they \nused baseball data to illustrate Stein\u2019s paradox. There is a clear analogy \nbetween percent defective in a quality assurance application and a \nbaseball batting average. The data in Ref. 7 was for many players at \na given point in time. The qmp algorithm works with the data for one \nproduct over time. So a better baseball analogy would be one player \nover time.\n\nTable I contains batting average data for Thurman Munson from \n1970 through 1978. This data was collected and analyzed by S. G. \nCrawford and is displayed graphically in Fig. 3. The \u201ccrosses\u201d are \nMunson\u2019s batting averages reported on the last Sunday of April for \neach year. The \u201cboxes\u201d are Munson\u2019s batting averages at the end of \nthe season. The dashed line is the average of the \u201ccrosses.\u201d\n\nThe early season averages are analogous to the audit data. The \naverages are the results from small samples of the populations. The \npopulations are the finite populations of \u201cat bats\u201d for each season. In\n\nFig. 3\u2014Batting averages for Thurman Munson. For each year, the movement from \nthe early season average (the sample) to the season average (the population) is always \nin the direction of the time average of the samples. This suggests strongly that by \nshrinking the samples towards their time average, one can obtain improved estimates of \nthe populations.\n\nthe audit, we are interested in making a statistical inference each \nperiod about the current population. So our problem in Fig. 3 is to \nmake a statistical inference each year about the season batting average \nusing only the early season averages observed to date.\n\nAs an estimate, one would be tempted to use the maximum like- \nlihood estimate, the early season average. But, in every year, the \nmovement of the batting average from the early season to the season \nend is in the direction of the aggregate early season average over time. \nSo, paradoxically, the early season averages from other years seem to \nbe relevant to the current season average. It is clear from the data, \nthat a better estimate of the season average is some kind of shrinkage \nof the early season average towards the aggregate early season average \nover time. And the amount of shrinkage can depend only on the \navailable data\u2014the early season averages.\n\nWhat we really have here is a multivariate problem. We observe a \nnine-dimensional vector of observations whose mean is a vector of \npopulation characteristics, one of which we are particularly interested \nin. Stein\u00ae showed (for the normal distribution) that the maximum \nlikelihood estimate of the vector is inadmissible. Why this is true \nmanifests itself in baseball lore. A player that starts the season rela-\n\ntively hot, usually cools off; and a player that starts in a relative slump \nusually improves. This is due to the nature of sampling error. The hot \nplayer is usually partially lucky and the slumping player is usually \npartially unlucky.\n\nQMP is based on the concepts illustrated by the Munson data. We \nsaw in Fig. 1a that the Best Measure of the population index is between \nthe current observed index and the estimated long-run process average.\n\nThe approach used for QmpP is actually Bayesian empirical Bayes. \nThe shrinkage factor used is a Bayes estimate of an optimal shrinkage \nfactor. So the Best Measure has the form\n\nwhere W is a Bayes estimate of \n[sampling variance] \n[sampling variance] + [process variance]\n\nThe bigger the sampling variance is, relative to the process variance, \nthe more weight is put on the estimated process average.\n\nThere are two advantages to the Bayesian empirical Bayes approach \nover the approach in Ref. 10. One is that the weight, W, is always \nstrictly between zero and one. This is because W is a Bayes estimate \nof an unknown optimal weight, w, which has a nondegenerate posterior \ndistribution on the interval [0, 1]. The approach taken in Ref. 10 is to \nuse maximum likelihood estimates of w, which can be one; i.e., total \nshrinkage to the process average.\n\nThe second advantage is that an interval estimate of the current \npopulation index can be constructed from its posterior distribution. \nMost of the literature (e.g., Ref. 10) treats the estimation problem \nthoroughly, but it provides little guidance for the interval estimation \nproblem.\n\nThe qmp algorithm is applied to the Munson data and the qmp \nestimates of the season averages are given in Table I. The sums of the \nabsolute errors for the maximum likelihood estimates (April averages) \nand the qmP estimates are 0.603 and 0.331, respectively\u2014a forty-five \npercent improvement. Notice that the QmP estimates for 1970 and 1971 \nare close to the April averages. This is because there was no history on \nMunson. The reduction in total absolute error for the years 1973 \nthrough 1978 was sixty-five percent, because of the benefit of history.\n\nThis paper is intended to document QmP. It contains the rationale \nfor changing the rating system, a synopsis of QMP features, mathemat-\n\nical derivations of the rating formulas, the dynamics of QmMP, the \noperating characteristics of (MP, many examples, and the QmP report- \ning format.\n\nReaders who are interested only in the mathematics of gmp and how \nit relates to empirical Bayes, may skip Section II. Readers, who are \nnot interested in the mathematical derivation of QMP, may skip Section \nIV.\n\nTo understand the rationale behind QmP, one must first understand \nthe T-rate system. From this we shall see where things have changed \nand where things have remained the same.\n\nThe sampling methods along with the scope of inspection provide \nfor a sample of units of count for each set of inspections. A unit of \ncount is either a unit of product or a unit of a product\u2019s attribute such \nas solderless wrapped connections.\n\nThe result of conducting a set of inspections is a list of defects found \nand their descriptions. Frequently, underlying a defect is a variable \nmeasurement* that falls outside a range. QMP does not affect the \nprocess of finding defects.\n\nThe defects found sometimes occur in clusters for which the effect \nof the cluster is nonadditive; i.e., the effect is less than the sum of the \neffects of the individual defects occurring by themselves. In this case, \nthe number of defects assessed for rating purposes is less than the \nnumber found. The defect assessment practices for the T-rate system \nevolved over a 50-year period, so these practices were based on a \nvariety of criteria and engineering judgements. The defect assessment \npractices under QMP amount to a redesign of the practices using a \nsingle principle, which is described in Section 3.1.\n\nThe defects assessed are transformed into demerits or defectives or \nmay remain as simple unweighted defects. In an audit based on \ndemerits, each defect assessed is assigned a number of demerits: 100, \n50, 10, or 1 for A, B, C, or D weight defects, respectively. Guidelines for \nassigning demerit weights are contained in numerous general and\n\n*For rating transmission characteristics of exchange area cable, some variables \nmeasurements are used directly without conversion to defects. We do not treat this case \nhere.\n\nFor any set of inspections, the quality engineers in the QAc have \nestablished quality standards. To do this, they considered audit scope, \nshop capability, field performance, economics, complexity, etc. The \nphilosophy of standards is described in Ref. 3. For audits based on \ndefects or defectives, the standards are expressed in defects or defec- \ntives per unit. For audits based on demerits, the standards are derived \nfrom fundamental defect per unit of count standards for A, B, C, D- \ntype defects. In addition, we use Poisson as the standard distribution \nof the number of type A defects (for example).\n\nTo make this clear, let\u2019s consider a simple example. Suppose in a \nsample of size n, there are X4, Xz, Xc, Xp-type A, B, C, D defects. The \ndefinition of standard quality is that X4, Xz, --- are independent and \nhave Poisson distributions with means nA, nAg, ---. The number of \ndemerits in the sample is\n\nNote that U, is the demerit per unit standard and C; is a variance \nper unit standard. These are the numbers that would be published in \nthe official list of standards called the Master Reference list. The \nquality standards are not affected by QmP.\n\nFor the purpose of reporting quality results to management, the \nproducts are grouped into rating classes. An example is: ESS No. 1\n\nwired equipment, functional test, at Dallas.* The results of all the \ninspections associated with this rating class are aggregated over a time \nperiod called a rating period. A rating period is about six weeks long \nand there are eight rating periods per year. QMP does not affect the \nrating classes or periods.\n\nThe advantage of having quality standards is that observed quality \nresults can be statistically compared to the standards. In the T-rate \nsystem, this is done with a statistic called the T-rate.\n\nFor a given rating class, let Q denote the total number of defects, \ndefectives, or demerits that are observed in all the inspections con- \nducted on all the subproducts during a rating period. Because there \nare quality standards for each set of inspections on each product \nsubclass, it is possible to compute the standard mean and variance of \nQ, denoted by E(Q|S), V(Q|S). The T-rate is\n\nIt measures the difference between the observed result and its standard \nin units of statistical standard deviation.\n\nFor each rating period, the T-rate is plotted in the control chart \nformat shown in Fig. 1b. The control limits of +2 are reasonable under \nthe assumption that @ has an approximate normal distribution. Then \nthe standard distribution of Q is the \u201cstandard normal,\u201d and excursions \noutside the control limits are rare under standard quality. For large \naudit sample sizes, this approximation follows from the central limit \ntheorem. As we shall see, the approximation is poor for small sample \nsizes.\n\nThe fundamental reports to WE management are books of T-rate \ncontrol charts for all rating classes. However, every rating period, a \nsummary booklet is prepared. The summary consists of various aggre- \ngate quality performance indices and an exception report which lists \nrating classes that are having quality problems.\n\nThere are two kinds of exceptions: Below Normal (BN) and ALERT. \nThese are based on statistical tests of the hypothesis that quality is at \nstandard. The rules for BN and ALERT are based on six consecutive T- \nrates, t:, --- ts, where fs is the current T-rate. The rules use the\n\n* Technically, this is called a scoring class in quality assurance documentation. Here, \nrating class means scoring class.\n\nFinally, the rules for BN and ALERT are: \nBelow Normal (BN): One of the following two conditions is satisfied:\n\n(1) &<-3 \n(2) \u20143 s ts < \u20142 and at least one of the following three conditions \nhold: \n(t) SCAN \n(wz) 341 \n(zit) At least one of the set {te, ts, ts, ts} is less than \u20142. \nALERT: SCAN or 341 but not BN. \nIn Fig. 1b, examples of BN are 7806 and 7803. Examples of ALERT are \n7808 and 7804. \nBoth the fundamental report formats and the rules for BN and ALERT \nare different under QMP\n\nThe advantage of the T-rate is its simplicity. It can be calculated \nmanually. Exceptions can be identified by inspection. The fact that \nthe T-rate has been used for so long is a testimonial to its advantages.\n\nHowever, the 7-rate does have problems.* The T-rate does not \nmeasure quality. A T-rate of \u20146 does not mean that quality is twice as \nbad as when the \u00a2-rate is \u20143. The T-rate is only a measure of statistical \nevidence with respect to the hypothesis of standard quality. This \nsubtle statistical point is often misunderstood by report readers. Years \nof explanations have not cleared up the confusion.\n\nAnother problem is that the ALERT (SCAN and 341) rules are tests of \nhypothesis about quality trends, not current quality. Consider Fig. 2. \nYou can assert that quality was probably substandard sometime \nbetween 7702 and 7707. You cannot, however, assert that quality is \nsubstandard in 7707. The QmP result for 7707 is normal.\n\nIn addition, the rules for ALERT and to some extent BN depend on \nattributes of past T-rates rather than their exact values. For example, \nfive consecutive past T-rates at \u20141.0 are treated exactly like five \nconsecutive T-rates at \u20140.1. This was done for statistical robustness. \nBut statistical information is lost. There are no \u201coutliers\u201d in the audit \ndata. Defects assessed were in the product. Many defects assessed\n\n* Although the foundation of the T-rate system was laid by Shewhart* and Dodge,> \nthe details are the results of contributions by many people over 50 years.\n\nmean substandard quality at the time of assessment. It is possible that \nvery unusual circumstances caused the defects. But it is intended that \nthe audit flag such unusual circumstances.\n\nThe significance level of the T-rate hypothesis test depends on \nsample size and can be very large. Suppose that we have a simple test \ndefect audit with a sample size of 32 units and a standard of 0.005 \ndefects per unit. The expected number of defects is (32)(0.005) = 0.16. \nFor one defect observed, the T-rate is\n\nNow, assuming standard quality, the number of defects has a Poisson \ndistribution with mean = 0.16. The Poisson probability of one or more \ndefects is 0.15. So even when the standard is being met, there is a 15 \npercent chance of the T-rate dropping below \u20142.0. In statistical terms, \nwe have a biased test (i.e., there is no reasonable upper bound on the \nsignificance level).\n\nDefect assessment practices have two parts. Part one is a description \nof those situations where fewer defects are assessed than are found. \nPart two is a formula for the number of defects assessed.\n\nIn QmP, the principle for part one is: Normally all defects found in \nthe quality assurance audits are assessed. Occasionally a cluster of two \nor more defects is found for which the seriousness of the cluster is less \nthan the seriousness implied by individually assessing every defect in\n\nment is necessary. More specifically, a reducible cluster is a collection\n\n[2] almost surely discover in its entirety when a small part of the \ncluster is discovered and\n\n[4] total seriousness is better represented by assessing d. defects \n(computed by the assessment formula), rather than the number \nfound.\n\nIn [1], we use the word dependent in a statistical sense. Defects are \ndependent if they occur in a short interval of time and are systemati- \ncally introduced by a common feature of the production process.\n\nIdeally, the assessment associated with a reducible cluster of defects \nshould depend on the situation. Over time, lists of reducible clusters \nand their assessments could be catalogued and added to the demerit \nlists. But, for now, there is no list of reducible clusters, so an assessment \nformula is needed.\n\nwhere AN stands for \u201cAllowance Number.\u201d In turn, AN has the general \nform AN = e + 3Ve, rounded down to an integer or to the closest \ninteger, where e can be interpreted as an expected number of defects. \nThe computation of e and the rounding depend on the audit. For some \naudits, tradition has prevailed, and for other audits, methods of com- \nputing e were developed for QmP.\n\nAs an example, consider a single relay for which three contacts are \ndefective (class B defect). The traditional method of computing e for \nan apparatus audit is\n\nwhere 12 is the number of contacts in the relay and 0.005 is a traditional \ngeneric standard per unit of count for class-B defects. The traditional \nrounding is down, so AN = (0.06) + 3V0.06 = 0.79, rounded down to \nzero. Hence, one class-B defect is assessed.\n\nAnother example is a reducible cluster of loose terminations found \non a bay of equipment in a transmission installation performance \naudit. In this case, e is just the quality standard for the bay in defects \nper unit, and the rounding is to the nearest integer.\n\nA complicating factor in the analyses of audit results is that defects, \ndefectives, and demerits are different. But, are they really? The answer \nis no; because, for statistical purposes, they can all be transformed into \nequivalent defects that have approximate Poisson distributions.\n\nSuppose we have a quality measure @ (total defects, defectives, or \ndemerits). Let E, and V; denote the standard mean (called expectancy) \nand variance of Q. So the T-rate is T = (E; \u2014 Q)VV,.\n\nIf all defects have Poisson distributions and are occurring at 6 times \nthe standard rate, then it can be shown that\n\nhence, X has an approximate Poisson distribution with mean e@. \nAs an example, consider the demerits case. The total number of \ndemerits has the general form\n\nwhere the w,\u2019s are known weights and the X;\u2019s have Poisson distribu- \ntions. Assume that the mean of X; is e,6, where e; is the standard mean \nof X; and @ is the population quality expressed on an index scale. So @ \n= 2 means that all types of defects are occurring at twice the rate \nexpected.\n\nand \nV(D) = \u00a5 wiV(X) \n=) wi(e6) \n= OV,, \nwhere E, and V, are the standard mean and variance, respectively, of\n\nThe mean and variance of X are equal; so, X has an approximate \nPoisson distribution with mean e@. Of course, it is not exact; because, \nX is not always integer valued. But, this Poisson approximation for \nequivalent defects is better than the normal approximation implied by \nthe T-rate system. It is the Poisson approximation in QmP that obviates \nthe need for the modification treatments discussed in Section 2.8.\n\nA similar analysis works approximately for the defectives case. So, \nany aggregate of demerits, defectives, or defects can be transformed \ninto equivalent defects. Just use the standard expectancy and variance \nas illustrated above for demerits.\n\nFor rating period \u00a2, let x; = equivalent defects in the audit sample, \n\u20ac; = equivalent expectancy of the audit sample, @; = population index,\n\nas defined in Section 3.2. Based on the discussion in Section 3.2, we \nassume that the conditional disribution of x; given 6; is Poisson with \nmean e,6;; 1.e.,\n\nIn Fig. 3 we see that the season average varies from year to year. \nSome of that variation is due to the fact that the season is itself a \nsample from a conceptual infinite population of at bats. The rest of \nthe variation is due to changes in ability, competition, etc., that are \ncaused by numerous factors that may or may not be identifiable. The \nimportant concept is that the time series of season averages is a \nstochastic process. For QMP we assume that the time series of 6;\u2019s is an \nunknown stochastic process.\n\nFor reasons that are partly statistical and partly administrative, we \nhave decided to restrict our use of past data to five periods. The main \nadministrative reason is that the T-rate system used the past five \nperiods. So all of the T-rate administrative rules that dealt with \nmissing data and reinitialization of rating classes can be used in QMP. \nStatistically, gmp works as well for six periods as it does for eight \nperiods (one year).\n\nA consequence of using only six periods of data is that no useful \ninference can be made about possible complex structure in the sto- \nchastic process of @;\u2019s. So we assume simply that the @;\u2019s are a random \nsample from an unknown distribution called the process distribution. \nFurthermore, six observations are not enough to make fine inferences \nabout the family of this unknown distribution. So for mathematical \nsimplicity we assume it to be a gamma distribution with unknown \nmean = @ and variance = y\u201d (Appendix A); i.e.,\n\nThe parameters 0\u00b0/y\u2019 and y\u201d/@ are the shape and scale parameters of \nthe gamma distribution. We use the names\n\nThis choice of a unimodal distribution reflects our experience that \nusually many independent factors affect quality; so there is a central \nlimit theorem effect.\n\nWe are assuming that the process average is unknown but fixed. In \nreality, it may be changing. We handle this by using a moving window \nof six periods of data. But this treats the past data symmetrically. An\n\nalternative would be some kind of exponential smoothing or Kalman \nfiltering. My colleague M. S. Phadke is developing a generalization to \nQMP based on a random walk model for the process average.\n\nThe model so far is an empirical Bayes model.\u2019 The parameter of \ninterest is the current population index, 87, which has a distribution \ncalled the process distribution. Bayesians would call it the prior \ndistribution if it were known. But we must use all the data to make an \ninference about the unknown process distribution. So, the model is \ncalled empirical Bayes.\n\nEfron and Morris\u201d take a classical approach to the empirical Bayes \nmodel. They use classical methods of inference for the unknown \nprocess distribution. QMP is based on a Bayesian approach to the \nempirical Bayes model. Each product has its own process mean and \nvariance. These vary from product to product. By analyzing many \nproducts, we can model this variation by a prior distribution for \n(9, y\u2019).\n\nThis is a full Bayesian model. It specifies the joint distribution of all \nvariables. The quality rating in QmP is based on the posterior distri- \nbution of 67 given x = (x1, +++, xr).\n\nQuality rating in QP is based on posterior probabilities given the \naudit data. Of course these probabilities depend on the model. But \nhow do we know the model is right?\n\nIt is important to understand that we are not doing data analysis \nwith QmpP. In data analysis, each set of data is treated uniquely. \nProbabilities cannot be computed. Objective decisions cannot be made.\n\nA requirement of quality rating is a specific rule that defines quality \nexceptions and a figure of merit (e.g., a probability) associated with an \nexception. A statistical model provides both. qmMp could have been \nbased on a more elaborate model. Our model represents a compromise \nbetween simplicity and believability.\n\nSo our exception decisions are at least consistent with one simple \nmodel of reality. The probabilities are conditional on that model. \nOtherwise, they can only be interpreted as figures of merit.\n\nWe have imbedded the simple hypothesis of a Poisson distribution \nwith a standard mean into a class of alternatives. The alternatives are\n\nPoisson distributions with nonstandard means. Much more compli- \ncated alternatives can be included: e.g., the class of negative binomial \ndistributions, and our probabilities would change a little. But gmp has \nachieved a kind of empirical validity. The exceptions being identified \nare accepted by the managers being rated. And for the products \ndeclared normal, there is a model (i.e., our model) that affords the \nstandard hypothesis some credence.\n\nWe show in Section IV that it is computationally impractical to \nderive the exact posterior distribution of @r. The best we can do is \napproximate the posterior mean and variance of @r.\n\nThe posterior mean and variance of @r are derived in Section IV. \nThe posterior mean is\n\nThe posterior mean, 67, is a weighted average between the estimated \nprocess average, 6, and the defect index, Ir, of the current sample. It \nis the dynamics of the weight, w7, that makes the Bayes estimate work \nso well. For any \u00a2, the sampling variance of J; is\n\n= 6,/e:. \nThe expected value of this is \nE[6@:/e:] = @/e:. \nSo the weight, wz, is \n[expected sampling variance] \n[expected sampling variance] + [process variance]\n\nIf the process is relatively stable, then the process variance is \nrelatively small and the weight is mostly on the process average; but \nif the process is relatively unstable, then the process variance is \nrelatively large and the weight is mostly on the current sample index. \nThe reverse is true of the sampling variance. If it is relatively large \n(e.g., small expectancy), then the current data is weak and the weight \nis mostly on the process average; but if the sampling variance is \nrelatively small (e.g., large expectancy), then the weight is mostly on \nthe current sample index. In other words, wr, is a monotonic function \nof the ratio of expected sampling variance to process variance.\n\nIf the process average and variance were known, then the posterior \nvariance of 07 would be (1 \u2014 wr)6r/er (Appendix B). So the first term \nis just an estimate of this. But since the process average and variance \nare not known, the posterior variance has two additional terms. One \ncontains the posterior variance of the process average and the other \ncontains the posterior variance of the weight.\n\nThe first term dominates. A large @r (relatively stable process), a \nsmall 67 (good current quality), and a large er (large audit) all tend to \nmake the posterior variance of #7 small (the box chart short).\n\nIf. @ is small, then the second term is negligible. This is because the \npast data is not used much, so the uncertainty about the process \naverage is irrelevant.\n\nIf the current sample index is far from the process average, then the \nthird term can be important. This is because outlying observations add \nto our uncertainty as to what is happening.\n\nIf the process average and variance were known, then the posterior \ndistribution would be gamma (Appendix B). So we approximate the \nposterior distribution with a gamma fitted by the method of moments. \nThe parameters of the fitted gamma are\n\na = shape parameter \n= 0%/Vr, \n= scale parameter \n= V7/6r, \nand the posterior cumulative distribution function is \nPr{@r s y|x} = G.(y/rt) \n(Appendix A).\n\nFig. 4\u2014QmpP box and whisker chart. This is a graphical representation of the posterior \ndistribution for current production given the six most recent periods of audit data. The \nwhiskers display the 99th and Ist percentiles and the box displays the 95th and 5th \npercentiles. The Best Measure is the posterior mean or Bayes estimate. It is a weighted \naverage of the process average (\u201cdot\u201d) and the current sample (\u201c\u201cX\u201d). The weight is the \nratio of sampling variance to total variance. If all the variance is due to sampling, then \nthe production is stable and the process average is the Best Measure of current quality. \nIf the sampling variance is zero, then the current sample is the Best Measure.\n\n3.4.2 QMP Below Normal and ALERT definitions \nIn QP, a rating class is Below Normal (BN) if\n\ni.e., the posterior probability that the product is substandard exceeds \n0.99. Substandard means @7 > 1. A rating class is on ALERT if\n\ni.e., the posterior probability that the product is substandard exceeds \n0.95 but not 0.99.\n\nThese definitions are illustrated graphically in Fig. 5, which is \noriented like the location summary in Fig. 6.\n\nFig. 5\u2014QMP exceptions. Below Normal means that the probability of substandard \nquality exceeds 0.99. For ALERT, the probability exceeds 0.95 but not 0.99. Normal is not \na quality exception.\n\nThere are two report formats for QMP results. One is a time series of \nbox charts illustrated in Fig. 1a. The estimated process averages are \njoined. The other is a location summary for the current rating period. \nThis is illustrated in Fig. 6. It orders the rating classes by Best Measure \nfor the current period. Another ordering that will be used is by rating \nclass name.\n\nWestern Electric, Bell Laboratories, and American Telephone and \nTelegraph management will receive all QmP results. Operating com- \npany management will receive QmP results on those rating classes that \nare of direct interest to them. Examples of results provided to the \noperating companies are the quality of repaired telephone sets and \ninstalled switching systems.\n\nMany of the advantages of QP relate to the disadvantages of the T- \nrate system (Section 2.8). QMP provides a direct measure of quality. If \na rating class is Below Normal, one can tell how bad the quality is. \nQMP uses past and current data to make an inference about current \nquality not past quality. If a rating class is on ALERT, then it is over 95 \npercent probable that there is a quality problem now. qmp does not \nuse runs criteria, but uses the actual equivalent defects observed. This \nprovides more statisxtical efficiency and therefore shorter interval \nestimates. QMP is robust against statistical \u201cjitter.\u201d It does not over- \nreact to a few defects. Consequently, there is no need for special\n\nREAD ONLY MEMORY (ROM) \u2014 PLASTIC ENCAPSULATED \nSTANDARDIZED TANTALUM ACTIVE RESONATOR HIC\u2019S ** MAJOR \nSILICON DIFFUSED DEVICES - PNPN AND INTEGRATED CKTS \nTRANSISTORS\u2014SILICON DIFFUSED, KMH\n\nMOUNTED BEAM LEAD ICS TRANSISTORS & DIPS ** NORMAL AUDIT \nBEAM LEAD METAL OXIDE SEMICONDUCTOR CHIPS\n\nSTANDARDIZED TANTALUM ACTIVE RESONATOR HIC\u2019S ** COMPOSITE \nTRANSISTOR AND CAPACITOR BEAM LEAD CHIPS\n\nCERAMIC CIRCUIT PACKS \u2014 GENERAL AUDIT ** MAJOR \nTRANSISTORS-\u2014SILICON DIFFUSED, MED POWER\n\nMOUNTED BEAM LEAD ICS, TRANSISTORS & DIPS ** OPERATIONAL LIFE TEST \nHYBRID INTEGRATED CIRCUITS\u2014DIGITAL ** COMPOSITE\n\nTRANSISTORS\u2014NPN SILICON DIFFUSED (PLASTIC ENCAP) \nCERAMIC CIRCUIT PACKS\u2014 SEMICONDUCTOR MEMORY ** MAJOR\n\nmodification treatments. This way we retain statistical objectivity \nconditional on our model.\n\nAnother advantage compared to the T-rate is that QMP provides a \nlower producer\u2019s risk and consequently a more accurate list of excep- \ntions. This is supported by data presented in Section VI.\n\nFinally, QMP will allow us to unify our reporting to Bell System \nmanagement. In the past, the T-rate statistic did not meet the needs \nof the operating companies; so, we developed a collection of special \nreports for the operating companies. Since the QmP report format does \nmeet operating company needs, the relevant subset of all the results is \na useful report.\n\nFrom Appendix B, we know that the distribution of 67 given 0, y\u2019, \nand xr is gamma; so, Pr{@r = y| 0, y\u201d, xr} can be expressed in terms of \nan incomplete gamma function.\n\nwhere p(6, y\u201d) is the prior density of 0, y? and L(6, y\u201d) is the likelihood \nfunction. Since x; given 6; is Poisson and @ given 6, y\u201d is gamma, it \nfollows that x; given 0, y\u201d is negative binomial; hence,\n\nSo the posterior distribution of #7 is a complex triple integral that \nhas to be inverted to compute the QmP box chart. The posterior mean \nand variance of #7 can be expressed in terms of several double integrals. \nThere are more than 1,000 rating classes that have to be analyzed each\n\nperiod, so computational efficiency is important. This is why we have \ndeveloped an efficient heuristic solution to the problem.\n\nIt is clear from Section 4.1 that prior distributions for 0 and y\u2019 are \nneeded. In the fourth rating period of 1979, we applied an earlier \nversion of the QmP algorithm to over 1,000 rating classes. This provided \nover 1,000 estimates of 9 and y\u2019 and empirical distributions of these \nestimates. The empirical mean and variance of the @ estimates were \n0.75 and 0.17, respectively. The empirical mean, variance, and mode of \nthe y\u201d estimates were 0.28, 0.19, and 0.05, respectively.\n\nIn the remainder of Section IV, we use 1 as a mean value of @ instead \nof 0.75. This is because 1 is the desired standard value that minimizes \nfirst cost plus maintenance costs. Under QmpP, the shops will be able to \noperate on the average closer to 1, because the producer\u2019s risk (see \nSection 6.2) is smaller than for the T-rate. Also, more defects are \nassessed under QmP than for the T-rate (see Sections 2.2 and 3.1).\n\nConditioning on @ and y\u201d means that the process distribution is known. \nSo by Theorem B.1 in Appendix B,\n\n67 = E[wrd + (1 \u2014 wr)Ir| x]. (1) \nTo calculate this posterior expectation exactly requires a double \nintegral. But a posterior expectation, E[- |x], can be viewed as an \nestimate of the operand \u201c-\u201d, because it is the Bayes estimate. So all \nwe need are estimates of wr and @. \n4.3.1 Moment estimates of process parameters\n\nAs argued in Section 4.1, given 0 and y\u2019, x; has a negative binomial \ndistribution. We show in Appendix D, eqs. (56) and (58), that\n\nSo we have many independent estimates of @ and y\u201d. A general \nmethod of combining independent estimates of parameters is a\n\nweighted average, where the weights are proportional to the reciprocal \nof the variances of the individual independent estimates. Such esti- \nmates of 6 and y\u2019 are\n\nNow V(I,) and V(Y,) depend on the unknown parameters @ and y\u201d. \nThe important consideration in setting the weights p; and q; is their \ngeneral behavior as e; varies. So for simplicity (to avoid iteration), we \nchoose 0 = 1 and y\u201d = %, which are empirically-determined mean \nvalues of these process parameters (see Section 4.2).\n\nIn Appendix D, we derive formulas for V(J;) and V(Y;). Plugging 6 \n= land y\u2019 = % into eqs. (56) and (59) yields\n\nNote that for small e,, f, \u00ab< e:; but for large e;, the fs and therefore \nthe weights, p,\u2019s, are all about equal. This is because for any large e;, \nI, = 6, and we are trying to estimate the average of the @,\u2019s.\n\nIn the case J; = 0 for all \u00a2, there is a problem with the estimate @. If \nwe plug 6= 0 into (1), then #r = 0. But 6r is a posterior mean of a \npositive parameter, so it cannot be zero. The correct method of \nhandling this problem is to start with a proper prior distribution on \nthe process average, 9. But then the mathematics and the computa- \ntions become complicated.\n\nSo we assert that we have prior information that is equivalent to \nobserving some \u201cprior data,\u201d xo and eo. Then a Bayes type estimate \nhas the form\n\nwhich has the same form as the moment estimate, 9, but uses all the \ndata including the \u201cprior data.\u201d\n\nTo choose values for xo and \u00e9o, consider JT = 1. A generic form of a \nBayes type estimate of @ is\n\nSetting this generic form equal to (7) yields \nE(@) = x0/eo, \nV(O) = 1/eo + \u201c4. \nFrom Section 4.2, E(@) = 1; and we conservatively choose V(@) to be \n1.25 (we do not want our prior observations of # estimates to preclude \nlarge future values of 9). This implies x = eo = 1. \n4.3.3 Bayes estimate of weight\n\nas suggested by eqs. (2) and (4).* The first term is the total variance \nabout the process average and the second term is a weighted average \nof estimated sampling variances. [Recall from Section 3.3.3 that \nVU. | 6.) = @:/e:, which can be estimated by I,/e;.] We denote the \naverage sampling variance by\n\nThe problem with yj as an estimate of process variance is that it \ncan be negative. To solve this problem, we use the results in Appendix \nC. Assume o\u201d is a known constant, and define the unknown weight as\n\nTo apply Appendix C, we must find a statistic, ss, and a degree of \nfreedom, df, so that, approximately,\n\nOriginally, we just assumed approximate normality of J, and took ss \n= (df + 1) Yeo qi(I: \u2014 6)? and df = T. But we found unusual sets of \ndata for which the number of defects allowed (before declaration of \nBelow Normal) was a decreasing function of expectancy in short ranges \nof small expectancy. We dubbed this the \u201cqmpP wiggle.\u201d\n\nby a scaled chi-square with degrees of freedom deduced by the method \nof moments.\n\nwhere u is an unknown constant. The two moment equations that \nhave to be satisfied are\n\n2E7[Z] \nVIZ] df. (10) \nInspired by well-known normal theory, we use the approximations \ndf \nE(Z) = Lei E(Z,), (11) \n2E*(Z 2E? \n2k (@) =\u2014 2B (41) =\u2014 1, (12) \nV(Z) V(Zi)\n\nAs for eq. (12), the mean and variance of Z, depend on @ and y\u201d. So \nto avoid iteration, we now select 6 = 1 and y* = 0,* which were \nempirically determined in Section 4.2. Then by eqs. (12), (56), and (57),\n\nwhere df is given by eq. (18). . \nNow apply the Corollary to Theorem C.1 in Appendix C to get\n\n* The choice of y* = 0 here may seem inconsistent with choice of y? = % in the \ndefinition of f; and g:. There it was necessary to take a positive value (the empirical \nmean across products) of y? to get the correct behavior for large e;. Here it was not \nar ae so for simplicity, we took the approximate empirical mode across products, \ny = 0.\n\nTo determine the parameters (ao, bo) of the prior distribution of w, \nwe first develop an empirical distribution of estimated w\u2019s across many \nrating classes, which have a mean and variance of 0.6 and 0.03, \nrespectively. To be conservative, we inflate the variance, shrink the \nmean, and select the prior mean and variance of w to be 0.55 and 0.045, \nrespectively. The parameters ao and bo are then solutions to (see \nAppendix C)\n\nwhere F' and G are defined in Theorem C.2 in terms of ap and Ro = \nbo/aoo\u201d. A numerical analysis yields\n\nNow we define an estimate, 7\u201d, of the process variance by \n2 \no \nE (w | x) _ ore . \nSo by eq. (14) \n7 = FS? = co\u201d \n= (FR \u2014 1)o\u2019. (21) \nThis is our improvement of the moments estimate. The inflation factor \nF prevents 7\u201d from being negative or zero. It can be shown that if R is \nlarge, F' is approximately one; but if R is small, FR \u2014 1 is positive and \nF is large. \nWe are now in a position to estimate wr by\n\nConditioning on 6 and y\u201d amounts to the process distribution being \nknown. So by Theorem B.1, in Appendix B,\n\nVr = E[(1 \u2014 wr) E(6r| 9, y\u2019, x)/er|x] + V[wr(9 \u2014 Ir)| x]. \nConditioning on y\u2019 in the second term yields \nVr= E[(1 \u2014 wr) E(Or| 8, y\u2019, x)/er| x] \n+ E[V[wr(@ \u2014 Ir) | y\u2019, x]|x] ; \n+ V[E[wr(@ \u2014 Ir) | y\u2019, x]x], (25) \nso the posterior variance has three components.\n\nThe first component is approximated by regarding the posterior \nexpectation operator as an estimation operator, and it is\n\nTo approximate the second component, we first approximate V[wr(@ \n\u2014 Ir)|y\u2019, x]. Since wr depends primarily on y\u201d and ez, we shall \nconsider w7 a constant. So\n\nAgain, treating the posterior expectation operator as an estimation \noperator, we get for the second component of eq. (25)\n\nFor the third component in eq. (25), we first approximate E[wr(0 \n\u2014 Ir)| y?, x] by &r(@ \u2014 Ir), where \n= 6/ eT \nOT =z \n- O/er + y\u2019 \nSo the third component in eq. (25) is\n\nPutting eqs. (26), (30), and (35) together implies that the approxi- \nmate posterior variance of 97 is\n\nHere we summarize the QmP formulas. On the right side of the \nformulas are the section numbers or equation numbers where the \nformulas were derived.\n\nQ, = Attribute quality measure in the sample, period \u00a2 (total \ndefects, defectives, or demerits), . \nEs: = expected value of Q, given standard quality,\n\nVs: = Sampling variance of Q, given standard quality. \nFor each period compute the following:\n\nFor the \u201cprior data\u201d (t = 0), let x\u00bb = e0 = 1. \nFor t = 0, --- , T, compute the following: (Section 4.3.2) \nSample index: \nI, = x2 /er,\n\nOver all periods \u00a2 = 0, --- , T compute the following: \nProcess average: \n6 = (\u00a5 pelt) (7) \nDegrees of freedom:\n\nThe Best Measure and the box chart percentiles are nonlinear \nfunctions of all the data, so the dynamic behavior of these results can \nappear to be complex. But this complex behavior is desirable and can \nbe explained. This section characterizes the fundamental dynamics of \nQMP by example.\n\nSince QmP is partially based on a long run average, it is natural to be \nconcerned about responsiveness of the box chart to sudden change. If \nthere is a sudden degradation of quality, Quality Assurance would like \nto detect it. If the producer solves a chronic quality problem, they \nwould like their exceptions to disappear. Figures 7 and 8 illustrate the \nQMP dynamics of sudden change.\n\nThe history data in Fig. 7 is a typical history for a product that is \nmeeting the quality standard. The equivalent expectancy of five is \naverage for a manufacturing audit. The history is plotted on a T-rate \nchart along with six possible values for the current T-rate (labeled A \nthrough F\u2019). So the current period is anywhere from standard (T-rate \n= 0) to well below standard (Index = 3.24, T-rate = \u20145).\n\nThe right side of Fig. 7 shows the six possible current results plotted \nin QMP box-chart form. The box chart labeled A is the result of \ncombining current result A with the past five periods. The box chart \nlabeled F is the result of combining current result F\u2019 with the same \npast history.\n\nAs you can see, the QmP result becomes ALERT at about T-rate = \n\u20143 (letter D) and becomes BN at about T-rate = \u20144 (letter E). For the \nT-rate method of rating, you would have a BN at T-rate = \u20143. The \ngood past history has the effect of tempering the result of a T-rate =\n\nFig. 7\u2014Dynamics of sudden degradation. The six QP box charts (labeled A through \nF) result from the analysis of six time series of data, which all have the same past \nhistory, but have different current values, as shown in the T-rate chart. A QMP ALERT is \ntriggered at a T-rate of \u20143 (letter D) and a gmp Below Normal is triggered at a T-rate \nof \u20144 (letter E). So a good past history tempers an observed change. Notice that from \nA to F, the Best Measure swings towards the sample value. This results from increasing. \nevidence of an unstable process (expected number of defects equals 5 for this chart).\n\nindex, process average, and Best Measure as the current value goes \nfrom A to F. The current index changes a lot (from 1.00 to 3.24) and \nthe process average changes a little (from 1.00 to 1.38), both in a linear \nway. The Best Measure also changes substantially, but in a nonlinear \nway. It changes slowly at first and then speeds up. This is because the \nweight is changing from 0.71 to 0.32. The weight changes, because as \nthe data becomes more and more inconsistent with the past the process \nbecomes more and more unstable, while the current sampling variance \nchanges slowly in proportion to the process average.\n\nFigure 8 is the dual of Fig. 7. It illustrates the dynamics of sudden \nimprovement. For the first five periods plotted, the process average is \ncentered on an index of two. Then an improvement takes place and \nfrom the sixth period on, the sample index is at the standard value of \none.\n\nFor the first five periods plotted, the rate is BN four times and ALERT \nonce. In the sixth period there is a sudden improvement and the \nsample index goes to standard. Immediately, there is a jump in the \nBest Measure and the rate is no longer BN. Because of the increase in \nprocess variance, the weight changes from 0.69 to 0.61, putting more \nweight on the current good result. The posterior variance stays about \nthe same [6r gets smaller but (1 \u2014 dr) gets larger].\n\nthese periods both the process average and the Best Measure gradually \nmove up towards the standard.\n\nA Bogie chart is a graphical device for tracking quality assurance \naudit data during a rating period. Figures 9 and 10 are examples of \nBogie charts. The vertical axis is an index scale and the horizontal axis \nis an equivalent expectancy scale. During the rating period, as the \naudit sample size builds up, the sample equivalent expectancy in- \ncreases. So the horizontal axis can also be viewed as a time axis.\n\nThe Bogie curves labeled ALERT and BN are plots of the indices in \nthe current sample for which [95% and 199% (the 95th and 99th \npercentiles) are exactly one, respectively. So the Bogie curves depend \non the past history. The past histories associated with Figs. 9 and 10 \nhave average indices of 0.92 and 4.89, respectively. The variance of the \npast histories were 0.69 and 5.36, respectively.\n\nTo use the Bogie chart, you plot continuously through the period \nthe sample index as a function of the equivalent expectancy in the \nsample (see Fig. 9). Anytime this plot falls below the ALERT or BN \ncurve, the rate is ALERT or BN at the plotted. equivalent expectancy. \nThen to bail the rate out, the plotted sample index must get above the \nBogie curves before the end of the period. For example, in Fig. 9, if the \nperiod had ended at an equivalent expectancy of three, then the rate \nwould be ALERT. If it had ended at an equivalent expectancy of five,\n\nFig. 8\u2014Dynamics of sudden improvement. As soon as the sample value becomes \nstandard, the product is no longer in the quality exception report (expected number of \ndefects equals 5 for this chart).\n\nFig. 9\u2014Index Bogie chart for a good past history. Equivalent expectancy is a measure \nof how many defects are expected in the sample; so, equivalent expectancy increases \nwith sample size. During a rating period, as the sample size increases, one can track the \nee sample index (dotted curve) and compare it to Below Normal and ALERT \nthresholds.\n\nthen the rate would be BN. But the period ended at an equivalent \nexpectancy of eight and there is no exception.\n\nThe ALERT Bogie curve in Fig. 10 is interesting. It starts at zero, so \nyou start the period on ALERT. The past history is so bad, that in the\n\nFig. 10\u2014Index Bogie chart for a substandard past history. The Below Normal and \nALERT thresholds are very tight. At the beginning of the period, the product is on ALERT \nuntil proven otherwise.\n\nabsence of any current data the probability that the current quality \nwill be substandard exceeds 0.95.\n\nFor a fixed past history and current equivalent expectancy, there is \na BN Bogie for the current sample index. If the sample index is worse \nthan the BN Bogie, then the product is BN. Figure 11 is a contour plot\n\nFig. 11\u2014Below Normal Bogie contour plot. If the past mean is 0.8 and the past \nvariance is 0.7 (on an index scale), then the product is on the contour labeled 2.6. This \nmeans that if the current sample index exceeds 2.6, the product will be Below Normal \n(equivalent expectancy equals 5 for this chart).\n\nof the BN Bogie for an equivalent expectancy of five. The axes are the \nmean and variance of the five past values of the sample index; i.e.,\n\nwhere J; is the sample index in past period \u00a2. For given values of I, and \nS?, we used a standard pattern of I;\u2019s to compute the Bogie. The \nresults are insensitive to pattern. The dashed curve is an upper bound \nfor S32.\n\nTo see how the contour plot works, consider an example. Suppose \nI, = 0.8 and S?, = 0.7. The point (0.8, 0.7) falls on the contour labeled \n2.6. This means that if the current sample index exceeds 2.6, then the \nproduct will be BN. The contour labeled 2.6 is the set of all pairs (Jp, \nS?) that yield a BN Bogie of 2.6. The T-rate associated with a BN Bogie \nof 2.6 is \u20143.6, as shown in Table II.\n\nTable II\u2014Index to T-rate conversion \ntable \nT-Rate (T)* \nEquivalent Expectancy (e) \nIndex (J) 1 5 10\n\nalent expectancy of five. As J, gets larger than one, the BN Bogie gets \nsmaller. If J, exceeds 1.6, then the BN Bogie is smaller than 2.34, which \ncorresponds to a T-rate of \u20143. So in T-rate terms, BN triggers earlier \nthan a T-rate of \u20143.\n\nFor J, less than 1.4, as S} gets larger, the BN Bogie gets smaller. \nThis is because large S?, implies large process variance which makes \nan observed deviation more likely to be significant.\n\nFor very small S2, as you move from I, = 0 to J, = 1, the BN Bogie \nincreases from 2.6 (J-rate = \u2014 3.6) to 2.9 (T-rate = \u20144.2). This is an \napparent paradox. The better the process average, the less cushion the \nproducer gets.\n\nThis is not a paradox, but an important characteristic of QMP. \nRemember with QmP we are making an inference about current quality, \nnot long-run quality. If we have a stable past with J, = 0.2, and we \nsuddenly get a sample index of 2.7, then this is very strong evidence \nthat the process has changed and very probably become worse than \nstandard. If we have a stable past with J, = 1, and we suddenly get a \nsample defect index of 2.7, then the evidence of change is not as strong \nas with J, = 0.2. The weight we put on the past data depends on how \nconsistent the past is with the present.\n\nNotice that the maximum BN Bogie is 2.92 and occurs at J, = 0.85 \nand S?2, = 0. It would be a mistake for the producer to conclude from \nthe contour plot that he should control his process at J, = 0.85 and S?, \n= (0. He cannot achieve S? = 0. The sample index has substantial \nsampling variance that the producer cannot control.\n\nThe Bogie contour plots provide the engineer with a manual tool to \nforecast the number of demerits that will be allowed by the end of a \nperiod. So we have published a book of BN and ALERT Bogie contour \nplots for equivalent expectancies from 0.5 to 25. .\n\nIt is tempting to conjecture that if both the process average and the \ncurrent sample index for one rate are worse than for another, then the \nBest Measure will also be worse. This is because the Best Measure is \na weighted average between the process average and the current \nsample index. But, since the weight depends on the data nonlinearly, \nthe conjecture is not true.\n\nTo illustrate this, consider Fig. 12. The six sample indices in Chart \nB are uniformly worse than the six sample indices in Chart A. But the \nBest Measure in Chart B is better than for Chart A. The reason is that \nthe weight in B is 0.54 vs 0.12 for Chart A.\n\nThe T-rate and QmP methods of rating are similar in some respects, \nbut there are major differences. In this section, these differences are\n\nFig. 12\u2014Nonlinearity of gmp. The sample indices in Chart B are uniformly worse \nthan the sample indices in Chart A; but, the Qmp result in Chart B is better than for \nChart A. In Chart A, the data provides very strong statistical evidence of an unstable \nprocess, so the past data is used very little in estimating current quality. This is not as \npronounced in Chart B.\n\nexplored using operating characteristics. The differences are a result \nof different rating formulas and assessment practices.\n\nA @QmP analysis of a rating class provides a probability, ps, that the \nrating class is substandard. For a typical rating period analyzed in \ndetail, we computed ps for all T-rate Below Normals and ALERTS. \nTable III shows the results.\n\nSo we find that for T-rate BNs, the QMP PS is typically high (greater \nthan 0.97); but, there can be an occasional low Ps (e.g., 0.75). However, \nfor T-rate ALERT\u2019s, the QmP Ps is frequently low (e.g., 0.85). This is \nbecause the 7-rate ALERT is an indicator of long-run quality, not \ncurrent quality.\n\nRange of Outlier \nException Probability Probability \nBelow Normal 0.97 \u2014 1.00 0.75 \nALERT 0.83 \u2014 0.99 0.59\n\nAny list of rating classes that is put in an exception report has a \nproducer\u2019s risk. It is the fraction of rating classes on the list whose \npopulation quality meets the standard. For a given period, let 6; = \npopulation index, rating class i, i = 1, ---, I. Label the rating classes \nso that product 1 through product L are on the exception list.\n\nHaving done QmP for each rating class, we have a posterior distri- \nbution for each 6;. Now let\n\nThe number of rating classes on the list whose population quality \nmeets the standard is .\n\nIn qoP, there is an exception list for each threshold probability (TP). \nTP = 0.95 corresponds to the list of all amp BNs and ALERTs. Figure 13 \nshows the QmP producer\u2019s risk and number of exceptions as a function \nof TP for the manufacturing audits in a particular period. The smaller \nTP, the bigger the exception list and the bigger the producer\u2019s risk. \nAlso, note that the producer\u2019s risk must be less than 1-Tp.\n\nThe set of all T-rate BNs and ALERTs is another exception list, whose \nproducer\u2019s risk is 0.037. This is relatively large because some individual \nALERTs have relatively large probabilities (e.g., 0.15) of being standard. \nThe number of T-rate exceptions (BN + ALERT) is shown to be 34.\n\nOf course to implement QmP, a particular TP had to be chosen. The \nTP that would match the T-rate producer\u2019s risk is about 0.885. But \nthat would lead to an unreasonable (70 percent) increase in exceptions, \nand a producer\u2019s risk of 0.037 is considered too high for this type of \nexception reporting, because of the high cost of false alarms. So we \ntook TP = 0.95, a reasonable balance between producer\u2019s risk and size \nof the exception report.\n\nFig. 13\u2014Operating characteristics of gmp versus the T-rate. As the qmp threshold \nprobability for exceptions (currently set at 0.95) is lowered, the number of exceptions \nand the producer\u2019s risk (for a particular rating period) both increase. The number of \nexceptions and producer\u2019s risk for the T-rate were 34 and 0.037. For threshold probabil- \nities between 0.96 and 0.89, amp has more exceptions and lower producer\u2019s risk than the \nT-rate.\n\nIt should be recognized that these curves depend on the particular \nset of audits being analyzed. For example, the curves depend on the \naudit sample sizes. It would be possible to lower sample sizes, decrease \nthe threshold probability, and still maintain a comparably sized excep- \ntion report with a reasonable producer\u2019s risk.\n\nNote that consumer\u2019s risk is not analyzed in this paper. Consumer\u2019s \nrisk is more relevant to acceptance sampling than to an audit. The \nmain purpose of the audit is to provide quality results to management \nincluding a compact exception report of high integrity. The Western \nElectric quality control organizations have primary responsibility for \nthe quality of each individual lot of product. |\n\nHere we explore specific examples that illustrate the similarities and \ndifferences between QmpP and the T-rate. In the examples, both Qqmp \nand T-rate results are based on the same defect data. For the actual \nimplementation of Q@mMP, the defect assessment rules will be different \nthan they are for the T-rate as explained in Section 3.1. The intent of \nthis section is to compare how the two rating methods work on the \nsame data.\n\nThe examples are shown in Figs. 1, 2, and 14 through 17. These \nfigures show a comparison between the time series of T-rates and QMP \nbox charts. Table IV contains summaries of the gmp calculations for \nthe particular periods that will be discussed in the following text.\n\nThe QmpP calculations shown do not use 1976 data. The box chart for \nthe first period of data available is not shown except in Fig. 15. Period \n7706 is the first period for which five periods of past data are used in \nthe QmP box charts. So the comparisons made in this section will \ninvolve periods 7706 through 7808.\n\nFigure 14 illustrates a T-rate borderline* in 7806 preceded by a good \nhistory. Since the equivalent expectancy (2.78) is fairly small and the \nprocess is fairly stable, the Best Measure (1.81) is heavily weighted \n(0.65) towards the process average (1.32). The posterior variance (0.36) \nis fairly large, so 195% is better than standard. However, in the next \nperiod, the T-rate plummets to \u20144.8 and the process average drops to \n1.77. Now the rate is clearly BN.\n\nIn Fig. 15, 7802, the T-rate is \u20143.8 (BN) but there is no exception for \nQmMP. One reason is that QP is based on the assumption that equivalent \ndefects have a Poisson distribution. A T-rate of \u20143.8 is very significant \nfor a normal distribution, but not as significant for a Poisson distri- \nbution with an equivalent expectancy of 0.29. For a normal distribution, \nthe probability, given standard quality, of being below \u20143.8 is 0.000072. \nNow the observed number of equivalent defects in 7802 is 2.36. The \napproximate Poisson probability of exceeding 2.36 equivalent defects \ngiven an equivalent expectancy of 0.29 is 0.15\u2014very different from \n0.000072.\n\nAnother reason is that the QMP result for 7802 is based on one period \nof data. Rather than using the sample defect index (8.00) as the process \naverage, we use a Bayes estimate [eq. (7)] of 2.77.\n\nFigure 16 is a similar example. In 7708 the T-rate of \u20142.8 is BN \nbecause in 7705 the T-rate was \u20142.7. But again, the \u20142.8 T-rate \noverstates the significance. The equivalent expectancy is only 0.23. \nAlso, the weight (0.60) on the process average (1.81) adjusts the sample \nindex (6.81) to the more moderate Best Measure (3.83). This, together \nwith the large posterior variance (8.47), implies a comfortable [95% of \n0.56.\n\nFigure 1 illustrates how two similar T-rates, both on ALERT, can be \neither a QMP BN or normal. Compare 7708 with 7804. The sample\n\nindices of 1.57 and 1.50 are very similar, but the process averages of \n2.00 and 1.32 are very different and the weights of 0.51 and 0.67 are \ndifferent. Hence, the Best Measures are very different and the conclu- \nsions are very different.\n\nFigure 2 illustrates a \u201cweak\u201d ALERT under the T-rate. The T-rate in \n7705 through 7707 are \u20140.1, \u20140.2, and \u20140.1, respectively. Although it\n\nFig. 14\u2014Example of agreement. Throughout 1978, gmp and the T-rate are in agree- \nment. The drop in the sixth period was called \u201cborderline\u201d under the T-rate, because it \nwas the first excursion below \u20142 and it was moderate. The Qmp box chart conveys the \nsame borderline message. In the seventh period, the product was Below Normal for \nboth systems.\n\nFig. 15\u2014Poisson versus Gaussian assumption. In the second period of 1978, the \nexpected number of defects in the sample was 0.29 and the observed number of \nequivalent defects was 2.36. Under the Gaussian assumption, the observed significance \nlevel is 0.000072 (i.e., T-rate = \u20143.8); but, under Poisson, the level is 0.15. This explains \nwhy the qmpP box chart contains the standard.\n\nis unlikely that the quality standard was being met in every period \nfrom 7702 through 7707, it is not unlikely (probability of 0.23) that the \nquality standard was being met in 7707.\n\nThe T-rate system had modification treatments that resulted from \nthe statistical deficiencies of the T-rate (see Section 2.8). There are no \nmodification treatments in QMP. The Poisson model and the stabilizing \neffect of shifting the sample index towards the process average alleviate \nthe need for modification treatments.\n\nIn Fig. 17, the 7807 unmodified T-rate is \u20142.5. It is modified to +0.6 \nbecause of the \u201cisolated\u201d A weight (100 demerits) defect. Under Qmp, \nthe process average (1.10) is only slightly substandard, the weight \n(0.51) is medium, and the equivalent expectancy (1.40) is small. All \nthis implies a safe [95% (0.74) without modification.\n\nIn the Venn diagram of Fig. 18, BNs are shown by circles and ALERTS \nare shown by rectangles. QmP results are shown by dashed lines and T- \nrate results are shown by solid lines. Every rating class that is BN or\n\nFig. 16\u2014Statistical jitter in the T-rate. With small samples and zero defects, the T- \nrate is slightly larger than zero. Every time a defect is found, the T-rate jitters. The \nmessage in the qmP chart is that there is too much uncertainty to reach any conclusions.\n\nFig. 17\u2014A case of T-rate modification treatment. Because the T-rate is biased for \nsmall samples, modification treatments were needed to compensate (seventh period, \n1978). QMP mathematics obviates the need for modification.\n\nTen rating classes were BN under both methods of rating. Five rating \nclasses were BN under the J-rate but ALERT under QMP. There were 16 \nrating classes that were ALERT under the T-rate but normal under \nQMP. This indicates a major difference. ALERT under the T-rate is \nstrong evidence that the quality standards for the current period or \nsome of the past periods have not been met. But it does not necessarily \nimply strong evidence that the quality standard for the current period \nhas not been met. ALERT under QMP implies more than a 95 percent \nchance that the current quality standards have not been met.\n\nMany members of the Bell Laboratories Quality Assurance Center \nand the Western Electric Quality Assurance Directorate have made \nimportant contributions to the development of QmP.\n\nB. T. Howard and E. Fuchs originally gained the support of the \nsenior management of Bell Laboratories and Western Electric. R. A. \nPeters and L. E. Bray followed this with many presentations to \nWestern Electric management. The role of liason among all the orga- \nnizations involved was handled by J. M. Wier.\n\nAs for technical contributions, R. A. Senior did the original data \nanalyses that transformed the idea into a concrete proposal. C. S. \nSherrerd developed the prototype QMP reporting system. S. G. Craw- \nford developed the summary reports and numerical analyses algo- \nrithms that permit efficient computation of a thousand QmP charts \neach period. S. W. Roberts, Jr., developed the QmP defect assessment \npractices. M. S. Phadke made contributions to the mathematical \ntheory of QMP.\n\nSeveral members of the Quality Information Systems groups have \nbeen developing the data bases and software associated with the \nofficial implementation of gmp. They are E. W. Hinds, V. A. Partridge, \nG. D. Rosen, P. A. Douglas, J. H. Carey, P. D. Ting, and S. Kadakia.\n\nFig. 18\u2014Venn diagram of exceptions. The Venn diagram accounts for all gmp and T- \nrate exceptions for a particular period using defects assessed under the T-rate. The lists \nof ALERTS under the two systems are quite different (only 10 out of 32 in common).\n\nWestern Electric Quality Assurance Headquarters people adminis- \ntered and analyzed the qmpP trial and prepared the material for nu- \nmerous presentations. They are C. Popik, N. O. Dickerson, N. Linar- \ndakis, T. M. Ferme, D. Snyder, H. M. Cook, E. Hoffman, and S. Chory.\n\nA random variable X = rY has a gamma distribution with shape \nparameter a and scale parameter r. We write\n\nA chi-squared random variable with \u00bbv degrees of freedom has a \nGamma distribution; namely,\n\n0,)** \u2014e \nflxe| 6) mica Bic At (38) \nXt \nThe process (prior) distribution of 0, is*: \n_ eo\" Xo\u20141,,\u2014e0% \nPo) F(x) Oe re (39) \n0=xo/eo, y= xX0/es.\n\nBy Bayes theorem, the posterior density of @; is proportional to the \nproduct of equations (39) and (38), which is in turn proportional to\n\nWe recognize eq. (40) as proportional to a Gamma density. So the \nposterior distribution is Gamma with shape parameter xo + x; and \nscale parameter 1/(eo + e,). And the posterior mean and variance are\n\nNow multiply the numerator and denominator in both eqs. (41) and \n(42) by 8/ece;. Theorem B.1 follows. Q.E.D.\n\n= (ss) | w ~ x? (chi-square, v degrees of freedom) \no known, wunknown \nand \no \nwW ~ Gamma( a, =). ao, bo known. \n0 \nThen \na\u201d \nw|Ss ~ Gamma( a 5). \nwhere \nv \na=a+ 9 \nss \nb = bo + 9\u00b0\n\nBy Bayes theorem, the posterior density of w is proportional to the \nproduct of eqs. (43) and (44):\n\nDefinition: Let X ~ Gamma(a, r). Denote the conditional distribution \nof X given X = c by\n\n1. E. Fuchs and B. T. Howard, \u201cQuality Assurance Tradition and Change,\u201d Bell \nLaboratories Record, 56, No. 9 (October 1978), pp. 226-31.\n\n2. R. Peters and IJ. O. Karraker, \u201cWhy You Can Trust Your Telephone,\u201d Ind. Res. 17, \nNo. 11 (November 15, 1975), pp. 47-51.\n\n3. W. A. Shewhart, \u201cNature and Origin of Standards of Quality,\u201d B.S.T.J, 37, No. 1 \n(January 1958), pp. 1-22.\n\n4. W. A. Shewhart, Economic Control of Quality of Manufactured Product, New \nYork: D. Van Nostrand, 1931.\n\n6. H. F. Dodge and M. N. Torrey, \u201cA Check Inspection and Demerit Rating Plan,\u201d \nInd. Qual. Contr. 13, No. 1 (July 1956), pp. 1-8.\n\n8. C. Stein, \u201cInadmissibility of the Usual Estimator for the Mean of a Multivariate \nNormal Distribution,\u201d Proceedings of the Third Berkeley Symposium on Math- \nematical Statistics and Probability, California: University of California Press, \n1955, pp. 197-206.\n\n9. Bruce Hoadley, \u201cAn Empirical Bayes Approach to Quality Assurance,\u201d 33rd Annual \nTechnical Conference Transactions of ASAC, May 14-16, 1979, pp. 257-63.\n\n10. B. Efron and C. Morris, \u201cStein\u2019s Estimation Rule and Its Competitors\u2014An Empirical \nBayes Approach,\u201d J. Am. Statist. Ass., 68, No. 341 (March 1973), pp. 117-30.\n\n11. M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions, New York: \nDover, 1965.\n\n12. M. B. Wilk, R. Gnanadesikan, and M. J. Huyett, \u201cProbability Plots for the Gamma \nDistribution,\u201d Technometrics, 4, No. 1 (February 1962), pp. 1-20.\n\nCopyright \u00a9 1981 American Telephone and Telegraph Company \nTHE BELL SysTEM TECHNICAL JOURNAL \nVol. 60, No. 2, February 1981 \nPrinted in U.S.A.\n\nFractionally-Spaced Equalization: \nAn Improved Digital Transversal Equalizer\n\nHere we describe and demonstrate, via analysis and simulation, \nthe performance improvement of voice-grade modems which use a \nFractionally-Spaced Equalizer (FSE) instead of a conventional syn- \nchronous equalizer. The reason for this superior performance is that \nthe FSE adaptively realizes the optimum linear receiver; consequently \nit can effectively compensate for more severe delay distortion than \nthe conventional adaptive equalizer, which suffers from aliasing \neffects. An additional advantage of the FSE is that data transmission \ncan begin with an arbitrary sampling phase, since the equalizer \nsynthesizes the correct delay during adaptation. We show that an \nFSE combined with a decision feedback section, which can mitigate \nthe effect of severe amplitude distortion, can compensate for a wide \nrange of linear distortion. At 9.6 kbit/s, the FSE provides a 2 to 3 dB \ngain in output signal-to-noise ratio, relative to the synchronous \nequalizer, over worst-case private-line channels. This translates to a \ntheoretical improvement of approximately two orders of magnitude \nin bit error rate.\n\nAs is well known,\u2019\u201d high-speed (=4.8 kbit/s) voiceband modems \nmust employ some sort of adaptive equalization to achieve reliable \nperformance in the presence of linear distortion and additive noise. \nThe equalizers are invariably implemented using transversal filters, \nbut the question of how the taps should be spaced has been, and still \nis, of great theoretical as well as practical interest. Conventionally, the \nequalizer taps are spaced at the reciprocal of the signaling rate. While \nit has been known theoretically that this synchronous structure does \nnot, by itself, realize the optimum linear filter, it has up to this time \nprovided adequate performance. The continuing demand for improved\n\nperformance at 9.6 kbit/s has renewed interest in adaptive equalizers \nwhose taps are spaced closer than the reciprocal of twice the highest \nfrequency component in the baseband signal.*\u2019 As we shall demon- \nstrate, such Fractionally-Spaced Equalizers (FSEs) are able to compen- \nsate much more effectively for delay distortion than the conventional \nsynchronous equalizers. Consequently, we will show that the perform- \nance of a FSE, with a sufficient number of taps, is almost independent \nof the channel delay distortion, and thus of the receiver sampling \nphase. More generally, the FSE is able to adaptively realize, in one \ndevice, the optimum linear receiver, which is known to be the cascade \nof a matched filter and a synchronously-spaced equalizer.\u00ae\n\nThe purpose of this paper is to report the results of an in-depth \ncomparative analytical and simulation study of FSEs and the conven- \ntional synchronous equalizer. We also evaluate the performance of an \nequalizer which results when a decision-feedback section, which is \nparticularly effective in compensating for amplitude distortion, is \ncombined with an FSE. We present simulation results that compare \nthe performance of practical-length synchronous and fractionally- \nspaced equalizers over a variety of voice-grade private-line channels.\n\nMany years have elapsed between Lucky\u2019s invention of the adaptive \nsynchronous equalizer,\u2019 Gersho\u2019s\u00ae and Brady\u2019s*\u201d\u00ae early work on FSEs, \nand our present interest in fractionally-spaced equalization. This is \ndue to both the increased complexity required to implement the FSE, \nand the relatively satisfactory performance of the conventional syn- \nchronous equalizer. Recent investigators have regarded the FSE pri- \nmarily as a means for mitigating the timing jitter produced by an \nenvelope-derived timing recovery system.\u201d\"\u2019 Our viewpoint, however, \nis that this property is just an example of the salient feature of the \nFSE\u2014the ability to effectively compensate for an extremely wide range \nof delay distortion, and to deal more effectively with amplitude distor- \ntion than the synchronous equalizer.\n\nIn Section II we describe why an FSE has the ability to compensate \nfor an arbitrary receiver sampling phase. Performance, as measured \nby the equalized mean-squared error, of an infinitely-long passband \nFSE is derived in Section III, and the corresponding results for a finite- \nlength equalizer are described in Section IV. Simulation results, for \ntypical voice-grade channels, are presented in Section V, and these \nresults are used to compare the performance of the synchronous \nequalizer, the FSE, and the FSE with a decision-feedback section.\n\nWe begin with a brief discussion of the ability of an FSE to compen- \nsate for any receiver timing phase. To do this we need the transfer \nfunction of a baseband fractionally-spaced equalizer. Consider the\n\nwhere {a,} is the discrete multilevel data sequence, 1/T is the symbol \nrate, f(t) is the system pulse response, and v(t) is additive noise. As \nshown in Fig. 1, we denote excess bandwidth of the pulse f(t) by a. \nThe input to a conventional synchronous digital equalizer are samples \nof the filtered received signal at the instants t = nT +7, i.e.,\n\nThe noiseless output of this nonrecursive digital filter, with tap weights \n{c;}, is the sample sequence\n\nwhere the equalized pulse samples, h(nT), have a (Nyquist-equivalent) \nFourier transform\n\nHere, F'7(w) is the aliased spectrum of F(w), Cr(w) is the (periodic) \ntransfer function of the equalizer, and, ideally, the equalizer output is \nthe data symbol, i.e., u(nT +7) = Gn.\n\nRecall\u00ae that the Nyquist-equivalent or folded (aliased) spectrum is \nthe relevant transform when dealing with sampled-data systems. In \nparticular, since Cr(w) = Cr(w + k27/T), the synchronously-spaced \nequalizer can only act to modify F'r(w), as opposed to directly modify- \ning F(w)\u2019**. In other words, the synchronous equalizer cannot exercise \nindependent control over both sides of the rolloff region about w =\n\nFig. 2\u2014Representative transfer function of a fractionally-spaced equalizer [tap spac- \ning = T\u2019 T/(1 + a)].\n\na/T. If, because of a severe phase characteristic and a poor choice of \n7, a null is created in the rolloff portion* of the folded spectrum F'7(w), \nthen all the conventional equalizer can do to compensate for this null \nis to synthesize a rather large gain in the affected region; this leads to \na severe performance degradation because of the noise enhancement \nat these frequencies.\n\nConsider, on the other hand, a receiver which uses a fractionally- \nspaced equalizer with taps spaced T\u2019 < T/(1 + a) seconds apart. This \nequalizer has the (periodic) transfer function:\n\nNote that. if 7/T\u2019 = (1 + a)a/T, then the first repetition interval of the \ntransfer function Cr-(w) includes the rolloff portion of the spectrum, \nas shown in Fig. 2. We assume that, for digital implementation pur- \nposes, 7\u201d is generally an appropriate rational fraction of T. For an FSE \nreceiver the equalizer input is sampled at the rate 7\u201d, but the equalizer \noutput is still sampled at the rate T, since data decisions are made at \nsymbol intervals. The equalized spectrum, just prior to the output \nsampler, is periodic (with period 27/T\u2019) and is given by \n2 2 \nHr(w) = Cr(w) SY Flo +k = Jexp| jlo+k\u2014)F}, (6) \nk T f be \nand for systems where 77/T\u2019 = (1 + a)z/T only the k = 0 term survives, \nL.e., \nHr(w) = Cr(w)Flw)e\u2122, \u2014 al Sz. (7)\n\nThe salient aspect of (7) is that Cr-(w) acts on F(w)e\u2019\u201d* before aliasing, \nwith respect to the output sampling rate, is performed. Thus Cr\u2019(w) \ncan compensate for any timing phase\u2014or phase distortion\u2014by syn-\n\nthesizing a transfer characteristic of the form e\u2019\u2019*. Clearly, such \ncompensation is highly desirable since it minimizes noise enhancement \nand avoids the extreme sensitivity to timing phase associated with the \nconventional equalizer.* After sampling the equalizer output at the \nrate 1/T, the output spectrum is periodic with period 27/T and is \ngiven by\n\nNote that (8) differs from (4) in that it is the sum of equalized aliased \ncomponents rather than an equalization of an already-formed sum of \naliased components.\n\nIt is evident that an FSE is capable of much more than compensating \nfor a poor choice of timing phase. With a properly chosen tap spacing \n(T\u2019 = [1/(1 + a)]T), the FsE has the capabilities of an analog filter. \nHence the FSE can be configured as the best linear receiver. In Section \nIII we derive the structure and performance of such a receiver for a \npassband modem.\n\nThe receiver minimizing the mean-squared error is known to consist \nof a matched filter followed by a synchronous sampler.\u00ae Our discussion \nin Section II demonstrated the equivalence of an appropriately sam- \npled fractionally-spaced equalizer with an analog receiver. We begin \nby writing the transmitted signal s(\u00a2), in an in-phase and quadrature, \nor quadrature amplitude modulated (QAM), data transmission system \nas the real part of the analytical signal\n\nwhere d, denotes the complex} discrete-multilevel data sequence, an \n+ jb,, p(t) is the (generally real) baseband transmitter pulse shaping, \n1/T is the symbol rate, w, is the radian carrier frequency, and s(t) is \nthe Hilbert transform of s(t). In our presentation we will make exten- \nsive use of complex notation to denote either passband or in-phase \nand quadrature signals, as well as system pulse responses. A discussion\n\n* In the synchronous equalizer, a \u201cbad\u201d timing phase is one which produces nulls in \nthe folded spectrum of F\u2019r(w) of (4). In the FSE, a \u201cgood\u201d timing phase is generated, \nregardless of the input sampling epoch, such that the FsE does a minimum of amplitude \nenhancement.\n\nof this approach is presented in the appendix. As shown in Fig. 3, s(\u00a2) \nis transmitted through the passband (around w,) channel x(t), with \nimpulse response\n\n= Re{(xi(t) + jxo(t))e/\"} = Re{%p(t)e\u2019}, (10) \nwhere the complex baseband-equivalent channel is defined by \nXp(t) = x1(t) + Jx2(t). (11)\n\nThus the received analytic signal has the representation \nF(t) = r(t) + y(t) = Ys dafa(t \u2014 nT)e7@e'*\u00ae + x(t)e%!, (12)\n\nwhere r(\u00a2) and 7(\u00a2) are the in-phase and quadrature components of \nthe received signal (and are a Hilbert transform pair), fp(t) is the \nbaseband-equivalent received pulse which is given by the convolution \nof Xp(t) with p(t), @ is the channel phase shift, and \u00bb(t) is the complex \nnoise signal.\n\nAt this point we may consider either a passband equalizer, which \noperates directly on F(t), or a baseband receiver which processes \nF(t) exp[\u2014j(wet + 0)]\u2014assuming that carrier-phase coherence\u201d has \nbeen established. From a mathematical viewpoint both systems are \nequivalent, and here we find it convenient to filter the demodulated \nsignal,\n\nFig. 3\u2014Qam data transmission system. Variables with overtildes are complex, ie., \nthey have in-phase and quadrature components.\n\nIn writing (14) we have used the notation g;(\u00a2) and g2(t) rather than \ng(t) and g(\u00a2) to emphasize that the receiving filter does not correspond \nto an analytic pulse. Note, however, that the equalized signal at the \nfilter output is given by the analytic signal\n\ni.e., u(t) and u(t) are a Hilbert transform pair. As shown in Fig. 3 the \nin-phase and quadrature output signals u(t) and u(t) are synchro- \nnously sampled at \u00a2 = n7T' + 7 and quantized to provide the data \ndecisions d, and bn.\n\nOur attention now turns to finding the linear filter, g(t), which \nminimizes the output mean-squared error. The output or equalized, \nmean-squared error (MSE), which is the performance measure com- \nmonly used for in-phase and quadrature data transmission systems, is \ngiven as\n\nwhere \u00a3 denotes the ensemble average with respect to the data \nsymbols and the additive noise, \u00e9, is the complex error sample, and e,, \nand \u00e9, are the in-phase and quadrature errors, respectively. For con- \nvenience, we have absorbed the receiver\u2019s sampling phase, 7, into the \npre-equalizer pulse response by incorporating the transfer function \ne/\u201c? into the transform of fg(t). In terms of the equalized pulse, h(t), \ndefined by\n\nUsing the independence of the data symbols and the independence of \nthe noise samples, \u00bb(nT), the terms in (20) are readily evaluated. The \nfirst term is the quadratic form\n\nf x(t)y(t) dt. By straightforward evaluations, the second and third \nterms in (20) are seen to be\n\nwhich is a quadratic form, where .% is recognized as being a Hermitian \noperator (its kernel is conjugate symmetric).\n\nThe MSE, given by (24), is minimized by taking the gradient with \nrespect to g. The optimum filter is given as the solution of the integral \nequation,\n\ncan be explicitly determined. This is accomplished by writing the left- \nhand side of (25) as\n\nwhere Z, = f f&(nT \u2014 1)g(t)dr are the equalized pulse samples. \nEquating (27) to the right-hand side of (25) gives\n\nwhich indicates that the optimum filter has the representation \nEopt(t) = Y \u00e9nfa(nT \u2014 t), (29)\n\nmatched to fg(t), and a synchronously-spaced tapped delay line with \nweights {\u00a2,}. To solve for the {\u00a2,} we substitute (29) into (25), giving\n\n+ 0\u00b0) \u00e9nfalnT \u2014 t) = fa(\u2014t), (30) \nand if we define the channel-correlation function, \nfan = | fi(nT \u2014 1) fa(mT \u2014 7) dr, (31) \nthen we can rewrite (30) as \np> fr-n\u00e9nfa(nT \u2014 t) + 0\u00b0 \u00e9nfa(nT \u2014 t) =fa(-t). (32) \nTaking Fourier transforms on both sides of (32), with respect to the\n\nwhere Fp(w) is the transform of fg(t). Dividing through by \nF'\u00a7(w), over the region where the channel does not vanish, we can \nrewrite (33) as\n\nwhere the following Fourier transforms, with respect to the discrete- \ntime variables, are identified by\n\ne 2 \nFz (. +1 7) \nThe transform \u00a5( w) corresponds to the synchronously-sampled \nmatched filter pulse, fg(\u00a2) \u00a9 f#(t), and Cr(w) is the transform of the\n\ncoefficients of the synchronously-spaced tapped delay line. From (34) \nwe have that\n\nThe final transform of interest is that of the equalized baseband- \nequivalent pulse, which is \n[|F B(w) |\u2019\n\nSince H(w) is real, the real part /,(\u00a2) and imaginary part h2(t) of its \ninverse Fourier transform are even and odd functions of time, respec- \ntively. Moreover, as o\u201d \u2014> 0 it is also clear that Heq(w) = (1/T) x Hw \n+ k 27/T) = 1, i.e., not surprisingly, the equalized channel is Nyquist. \nFrom (24) to (26) it follows that the minimized MSE is\n\nIn summary, eqs. (37) to (40) give a complete description of the \nperformance and structure of the optimum linear receiver. The struc- \nture is equivalent to an infinitely-long fractionally-spaced equalizer \nwhose taps are spaced close enough to accommodate the bandwidth of \nthe transmitted signal. Finally note that, as expected, the phase \ncharacteristics of the channel, including the timing phase, do not enter \ninto the expression for the minimum MSE; thus, the steady-state system \nperformance is independent of these characteristics.\n\nHere we consider the mean-squared error of a finite-length fraction- \nally spaced equalizer. The demodulated received signal,* (13), is sam- \npled at the rate 1/T\u2019, and thus the equalizer input is\n\nWe make a slight change in notation by letting \u00a2c, denote the complex \nequalizer taps\u2014thus the transfer function G(w) of the previous section \nis replaced by C(w). The equalizer output, which is only needed at the \nsynchronous instants, is given by\n\nwhere the equalizer has 2N + 1 complex taps. For the finite-length \nequalizer the MSE is written compactly as\n\n* A passband equalizer, in which the demodulator follows the equalizer, has the same \nperformance against linear distortion as does the baseband equalizer. The passband and \nbaseband equalizer differ in their performance in the presence of phase jitter.\n\nqn = (Q(nT + NT\"), --+, Q(nT), \u00ab++, (nT \u2014 NT\u2019)), (44) \nand the vectors with an asterisk will denote the transposed conjugate \nvector. Performing the indicated expectation gives\n\nwhere the (2N + 1) X (2N + 1) Hermitian channel-correlation matrix, \nthe (2N + 1) X 1 channel vector, and the data power are defined, \nrespectively, by\n\nIt is interesting to compute the kith element, Aj, of the channel- \ncorrelation matrix; a direct calculation gives\n\nwhere 6,-; is the Kronecker delta. Note that in contrast to the syn- \nchronous equalizer, the channel-correlation matrix is Hermitian but \nnot Toeplitz. To explicitly see the non-Toeplitz nature of the A matrix \nwe can rewrite (47) in the frequency domain as\n\nand for systems with nonzero excess bandwidth the bracketed terms \ndepend on & and I individually, rather than on k \u2014 J. Recall that for \nthe synchronous equalizer T\u2019 = T and A;,; depends only on k \u2014 1.\n\nIn terms of the above parameters, it is evident from (45) that the \noptimum tap setting ist\n\nAs with the conventional passband equalizer,\u2019\u201d the adaptive control \nalgorithm makes use of the gradient of the sum of the squared in-phase \nerror and the squared quadrature error with respect to the tap weights. \nTaking these derivatives, and writing the result in complex notation \ngives the adjustment algorithm\n\nwhere \u00a2, is the complex tap vector at the nth iteration, and a is a \npositive number, called the step size, which affects the algorithm\u2019s rate \nof convergence and the fluctuation about the minimum-attainable \nsteady-state MSE. Note that the algorithm is updated once per symbol \ninterval, but it is conceivable that adjustments could be made more \nfrequently if the mid-symbol output levels can be interpolated reason- \nably well. Reference 12 gives a detailed analytic and experimental \ntreatment of the convergence rate and some of the dynamic aspects of \nFSEs.\n\n4.3 Does a finite-length fractionally-spaced equalizer have a unique tap \nsetting?\n\nTo answer the question posed by the title of this section we return \nto the baseband data transmission system discussed in Section II. The \ntransmitted spectrum, as shown in Fig. 1, is bandlimited to (1 + a) 7/ \nT rad/s, where the rolloff factor, a, varies from 0 to 1. From Fig. 2 it \nshould also be evident that when the.noise becomes vanishingly small, \nthere is legitimate concern as to what function(s) the equalizer will\n\n+ A little care must be exercised in differentiating the MSE with respect to \u00e9, since \n\u00a2*e is not an analytic function of 6. The most compact approach is to differentiate the \nMSE, with respect to the real and imaginary components of \u00e9, and to then interpret the \ngradient as a complex vector.\n\nsynthesize in the region (1 + a) 7/T < w < 22/T, where there is no \nsignal energy. In terms of the channel correlation matrix given by (47), \nwe note that the matrix A is the sum of two matrices, and, as will be \nevident from the discussion which follows, the channel-dependent \ncomponent of A is always positive semidefinite. Since the other com- \nponent of the channel-correlation matrix, oJ, is positive definite, then \nA will also be positive definite, and we can conclude that when there \nis noise present the optimum tap setting is unique.\n\nWe now consider the situation as the noise becomes vanishingly \nsmall; clearly from (50), the optimum tap setting is unique if and only \nif A is nonsingular. A sufficient condition for A to be nonsingular is the \nnonvanishing of the quadratic form u\u2019Au, for any nonzero test vector \nu with components {u;}. Let us consider in detail this quadratic form, \nwhich we write from (47) as\n\nThe above inequality establishes the positive semidefinite nature of \nthe matrix A, and we see from, (53) that u\u2019Au can vanish only if *\n\nN \nYX unf(IT \u2014 mT\u2019) = 0, 2=0,+1,+2,---. (54) \nm=\u2014N \nIf we define the periodic Fourier transform \nN \nUr(o)= Yume\", lol sz, (55) \nthen we can proceed further by noting that \nN N = ; . dw \nY% unf(IT-\u2014mT\u2019)= Y un | F(w)e\u2014\"\u2122) \u2014 \neet, m=\u2014N As 2a \n2 N \n= | x ime\" |layerm\u2122 dw \n_o Lm=-N 2a\n\n* The authors gratefully acknowledge discussions with J. E. Mazo which led to this \ndevelopment.\n\nThe right-hand side of (56) is recognized as the sample, at t = IT, of a \nfunction whose Fourier transform Z.,(w) is contained in the brackets. \nNow if (56) is to be zero for every value of J, then it must be that the \nFourier transform inside the integral vanishes completely, i.e.,\n\nFor less than 100 percent excess bandwidth, note that only the k = 0, \n+ 1 terms contribute to the above sum. However, in the nonrolloff \nregion, | w |< (1 \u2014 a) z/T, only the k = 0 term influences the sum. For \nchannels which do not vanish over the entire nonrolloff region, it is \nclear that for Z.q(w) to vanish it is required that U7(w) vanish at least \nover the entire nonrolloff region. Since Ur(w) is a finite-term Fourier \nseries, it cannot vanish over an interval without vanishing everywhere, \nwhich in turn would again make u = 0. Note that if the channel \nvanished over a portion of the nonrolloff region, then since Ur(w) is a \nfinite-term Fourier series, its energy could not be totally concentrated \nin the region where there was no channel energy. Thus, the solution \nwould still be unique. However, it is worth noting that in the extreme \ncase of 100 percent excess bandwidth, Z.q(w) can vanish. For example, \nconsider a constant F(w), with Ur(w) = cos (wT/2). It is apparent from \n(57) that Z.q(w) = 0. Thus for a finite-length FsE with an excess \nbandwidth of less than 100 percent, we can conclude that even as the \nnoise becomes vanishingly small the A matrix is nonsingular and \nthere is a unique optimum tap setting.\n\nNote that for a finite-length synchronous equalizer where T\u2019 = T, \n(57) indicates that since Ur(w + (k27/T)) = Ur(w), we can conclude \nthat if the folded-channel spectrum does not vanish completely then \nthere is always a unique tap setting. Reference 12 showed that as the \nequalizer becomes infinitely long there is an infinitude of equalizer tap \nvectors which achieve the same minimum mean-squared error. Con- \nsequently, it is to be expected that a FSE equalizer with a \u201clarge\u201d \nnumber of tap weights will have many tap vectors which produce \nessentially the same MSE.\n\nTo illustrate the advantages of fractionally-spaced equalization over \nsynchronous equalization, a number of computer simulation runs were \nmade for different equalizer configurations and for channel distortions \nof varying severity. The system tested was the 9.6 kbit/s QAM system, \nshown in Fig. 3, having a symbol rate of 2400/s and an excess band- \nwidth of 12 percent, and the transmitted-symbol alphabet, {+1, +3}. \nFor each run the steady-state mean-squared error was measured after \na sufficiently long period of adaptation, and the FSE was of the 7/2 \ntype.\n\nAmplitude and delay-distortion characteristics are illustrated in Fig. \n4 for the three linear channels which were simulated. The \u201cGood\u201d \nchannel is of low distortion, and well within the limits of standard \nconditioning, e.g., the \u201cBasic\u201d conditioning\u201d illustrated in Fig. 5. The \n\u201cBad-Phase\u201d and \u201cBad-Slope\u201d channels have, respectively, severe \nphase distortion and severe amplitude distortion, placing these chan- \nnels just outside the defining boundaries of basic-conditioned channels.\n\nFigures 6, 7, and 8 compare the performance of a 24-tap synchronous \n(T) equalizer and a 48-tap T/2 equalizer on the three test channels. \nPerformance is examined for five timing epochs within a symbol \ninterval, and is measured by the output signal-to-noise ratio, defined \nas\n\nSNRout = log ae (58) \nwhere Pz is the received baseband average signal power (a constant), \nand MSE is the measured output mean-squared error. The received \nsignal is normalized so that the ratio of the signal power at the output \nof the receiving filters to the power of the additive noise, at the same \npoint in the system, is 28 dB. Thus if the equalizer could \u201cundo\u201d the \nchannel distortion without enhancing the noise, then the output SNR \nwould be 28 dB. It is apparent that the performance of the fractionally- \nspaced equalizer is almost independent of the timing epoch, in sharp \ncontrast to that of the synchronous equalizer. This confirms the \nprediction of the analysis, culminating in expression (40) for the \nminimum mean-squared error, which is independent of the sampling \nepoch. It is also significant that the performance of the fractionally- \nspaced equalizer on the \u201cBad-Phase\u201d channel is significantly better \nthan that achieved by the synchronous equalizer even for the best \nsampling phase. The capability of the FsE for phase equalization, \nbefore folding the spectrum about the Nyquist frequency, is seen to be \nan important advantage on channels with severe phase distortion. \nWith the addition of a decision-feedback equalizer (DFE), with feed-\n\nback taps { fi}, shown in Fig. 9, compensation for severe amplitude \ndistortion is also improved, as illustrated in Fig. 8.\n\nThe simulation of an FSE with 37/4 tap spacing [still less than T/ \n(1 + a), where a = 0.12 was the percentage of excess bandwidth] \nresulted in performance comparable to that of the T/2 equalizer. A \n3T/4 equalizer needs only 2/3 as many taps as a T/2 equalizer to span\n\na given channel dispersion, which cannot only reduce implementation \ncomplexity, but also improve steady-state performance when digital \nresolution is a consideration.\u201d\n\nWe have shown, both analytically and by simulation, that the \nfractionally-spaced equalizer provides virtual independence from tim-\n\nFig. 6\u2014Performance versus sampling phase of 24-tap synchronous (7') equalizer and \n48-tap (7/2) equalizer on \u201cGood\u201d channel of Fig. 4. Results from computer simulation \nof 9600 bps QAM modem.\n\ning epoch, and significantly improves steady-state performance on \nseverely phase-distorted channels. Implementation of the FSEs in- \ncreased number of taps, with respect to a synchronous equalizer with \nthe same total time span, is well within the capabilities of current \ndigital signal-processing technology. The performance degradation \nintroduced in a digital implementation by using a larger number of \ntaps is more than compensated for by the FsEs capability of adaptively\n\nFig. 8\u2014Performances on \u201cBad-Slope\u201d channel. The top curve is for a receiver which \nincorporates both a 48-tap FSE and a 16-tap decision-feedback equalizer (DFE).\n\nFig. 9\u2014@am data receiver combining a fractionally-spaced equalizer and a decision \nfeedback equalizer.\n\nrealizing, in one structure, the optimum receiving filter consisting of a \nmatched filter followed by a delay line tapped at symbol intervals.\n\nThe purpose of this appendix is to review and organize the compact \ndescription provided by complex notation for the discussion of in-\n\nphase and quadrature data communications systems. In such a system \nthe transmitted waveform is of the form\n\nwhere {a,} and {b,} are the in-phase and quadrature data streams, \n1/T is the symbol rate, p(-) is a bandlimited pulse, and w. is the radian \ncarrier frequency. The signal s(t) can be written as the real part of the \ncomplex analytic signal\n\nRecall that a signal is analytic if it only has power at positive (or \nnegative) frequencies.\u201d The analytic signal with positive frequency \ncontent is\n\nthen the in-phase pulse response, x:(\u00a2), and the quadrature pulse \nresponse, x(t), are given by\n\nThus, x,(\u00a2) and x2(t) are determined by the positive spectral content \nof the real pulse x(t). In general, the baseband pulses x;(\u00a2) and x(t) \nare unrelated [except through (64)], but if the transfer function X(w) \nhas even amplitude symmetry and odd-phase symmetry about the \ncarrier frequency, then x2(\u00a2t) = 0. Also note that, in general, x,(\u00a2) and \nx2(t) are not a Hilbert transform pair.\n\nGiven the in-pulse and quadrature pulses x:(\u00a2) and x2(t), we define \nthe analytic pulse\n\nwhere, as noted above, the complex baseband-equivalent pulse, x(t) \n= x(t) + jx2(t), is not necessarily analytic. The pulse Xs(t) has a \nFourier transform X(w + w-), w > \u2014w\u00a2; 1.e., the transform is the positive \nfrequency portion of X(w) shifted down to the origin. The channel \noutput signal, s(t) \u00ae x(t), is of course the Re[s(t) \u00ae@ x(\u00a2)], and we have \nthat\n\nwhere we have allowed the transmitted pulse, p(t), to be \u201ccomplex.\u201d \nBy a complex transmitted pulse we mean that the pulse input is a two- \ndimensional vector and a cross-coupled operation defines the filter; \ni.e., if the filter input is the two-tuple vector (z2:(\u00a2), z2(\u00a2)), which we use \nto define an equivalent complex signal z(t) = 2:(\u00a2) + jze(t), then the \noutput vector is w(t) = p(t) @ Z(t), where the outputs, wi(\u00a2) and u2(t), \nare the real and imaginary parts of u(\u00a2).\n\nAt the receiver, coherent quadrature demodulation by cosw,t and \nsinw-t, followed by low-pass filtering, provides the in-phase and quad- \nrature signals; these component signals can also be derived by forming \nthe two-tuple vector composed of the in-phase and quadrature signals \n(the latter signal is simply the Hilbert transform of the received signal),\n\nand \u201crotating\u201d the two tuple by wt radians.\u2019\u00ae This latter operation \ncorresponds to simply multiplying (67) by exp(\u2014jw.t); ie., the de- \nmodulation signal, F(t), is given by\n\nThus, given the channel x(t), we see that from a linear distortion \nviewpoint, the system is characterized by the (equivalent) baseband \npulse fg(t) = p(t) \u00ae Xa(t). If the demodulated signal, F(t), is further \nfiltered, or equalized, by a lattice-type filter,t g(t), the overall pulse \nshape will be p(t) \u00ae xa(t) @ g(t).\n\n1. R. D. Gitlin, E. Y. Ho, and J. E. Mazo, \u201cPassband Equalization of Differentially \nPhase-Modulated Data Signals,\u201d B.S.T.J., 52, No. 2 (February 1973), pp. 219-38.\n\n2. D. D. Falconer, \u201cJointly Adaptive Equalization and Carrier Recovery in Two-\n\nDimensional Digital Communication Systems,\u201d B.S.T.J., 55, No. 3 (March 1976), \n. 317-34. \n. A Gessho, private communication, 1968. \nD. M. Brady, \u201cAn Adaptive Coherent Diversity Receiver for Data Transmission \nThrough Dispersive Media,\u201d Conference Record, ICC 1970 (June 1970), pp. 21- \n35-21-40. \n. G. Ungerboeck, \u201cFractional Tap-Spacing Equalizers and Consequences for Clock \nwal for Data Modems,\u201d IEEE Trans. Commun., COM-24, No. 8 (August \n1976). \n. J. E. Mazo, private communication. \n. O. Macchi and L. Guidoux, \u201cA New Equalizer and Double Sampling Equalizer,\u201d \nAnn. Telecomm., 30 (1975), pp. 331-8. \n. R. W. Lucky, J. Salz, and E. J. Weldon, Jr., Principles of Data Communication, \nNew York: McGraw-Hill, 1968. \n. R. W. Lucky, \u201cAutomatic Equalization for Digital Communications,\u201d B.S.T.J., 44, \nNo. X (April 1965), pp. 547-88. \n10. D. M. Brady, Adaptive Signal Processor for Diversity Radio Receivers, U.S. Patent \nNo. 3,633,107, January 4, 1972.\n\n11. S. U. H. Qureshi and G. D. Forney, Jr., \u201cPerformance and Properties of a 7/2 \nEqualizer,\u201d Conference Record, NTC 1977 (December 1977).\n\n12. R. D. Gitlin and S. B. Weinstein, \u201cOn the Required Tap Weight Precision for \nPie ence Adaptive Equalizers,\u201d B.S.T.J., 58, No. 2 (February 1979), \npp. 301-21.\n\n13. Bell System Technical Reference, \u201cData Communications Using Voiceband Private \nLine Channels,\u201d PUB 41004, October 1973.\n\n14. J. Salz, \u201cOptimum Mean-Square Decision Feedback Equalization,\u201d B.S.T.J., 52, No. \n8 (October 1973), pp. 1431-73.\n\n15. H. E. Rowe, Signals and Noise in Communication Systems, New York: Van \nNostrand, 1965.\n\n16. D. A. Spaudling, \u201cA New Digital Coherent Demodulator,\u201d IEEE Trans. Commun., \nCOM-21, No. 3 (March 1973), pp. 237-8.\n\n+ By a lattice filter, g(t) = gi(t) + jgo(t), we mean that if the in-phase and quadrature \ninputs are r(\u00a2) and F(t), then the filter outputs are r(t) \u00ae gi(\u00a2) \u2014 F(t) \u00ae go(t) and F(t) \u00ae \nBi(t) + r(t) \u00ae go(t), respectively.\n\nAnthony S. Acampora, B.S.E.E., 1968, M.S.E.E., 1970, Ph.D., \n1973, Polytechnic Institute of Brooklyn; Bell Laboratories, 1968\u2014. At \nBell Laboratories, Mr. Acampora initially worked in the fields of high- \npower microwave transmitters, radar system studies, and signal pro- \ncessing. Since 1974, he has been studying high-capacity digital satellite \nsystems. His current research interests are modulation and coding \ntheory, time division multiple access methods, and efficient frequency \nre-use techniques. Member, Eta Kappa Nu, Sigma Xi, IEEE.\n\nRonald J. Canniff, M.S.E.E., 1972, University of Michigan; Bell \nLaboratories, 1972\u2014. Prior to his development work on the digital \nconcentrator for the SLC\u2122-96 system, Mr. Canniff spent nearly five \nyears working on voiceband modems for digital subscriber carrier \nsystems. This included work on delta modulation and active filters for \nthe per-channel codecs of the SLC\u2122-40 carrier system. He is currently \nsupervising a group concerned with digital signal processing functions \nfor a digital switching machine. Member, Eta Kappa Nu, Tau Beta Pi.\n\nBarbara A. Czekaj, B.S. (computer science), 1980, Monmouth \nCollege; Bell Laboratories, 1978\u2014. Ms. Czekaj worked part-time for \ntwo years in the Radio Research Laboratory, involved in reductions of \nmultipath fading data and computer analyses of digital radio systems. \nIn May 1980 she became a member of the Business Communications \nSystems Engineering Center. Member, Association for Computing \nMachinery.\n\nRichard D. Gitlin, B.E.E., 1964, City College of New York; M.S., \n1965, and D. Eng. Sc., 1969, Columbia University; Bell Laboratories, \n1969\u2014. Mr. Gitlin is supervisor of the Data Techniques Group in the \nData Communications Laboratory. He is a member of the Communi- \ncation Theory Committee of the IEEE Communications Society, edi-\n\ntor for Communication Theory of the IEEE Transactions on Com- \nmunications, and is a member of the Editorial Advisory Board of the \nProceedings of the IEEE. Senior Member, IEEE; Member, Sigma Xi, \nEta Kappa Nu, Tau Beta Pi.\n\nLarry J. Greenstein, B.S.E.E., 1958, M.S.E.E., 1961, and Ph.D. \n(E.E.), 1967, Illinois Institute of Technology; Bell Laboratories, \n1970\u2014. Since joining Bell Laboratories, Mr. Greenstein has worked on \ndigital encoding and transmission. His most recent work is in the areas \nof satellites, mobile radio, and terrestrial digital radio. He currently \nheads the Satellite Systems Research Department. Member, Eta \nKappa Nu, Sigma Xi, Tau Beta Pi; Senior Member, IEEE; associate \neditor, IEEE Transactions on Communications.\n\nBruce Hoadley, B.S. (math), 1961, North Carolina State; Ph.D. \n(statistics), 1965, University of California; Assistant Professor, Statis- \ntics, University of California, Berkeley, 1965-1966; Bell Laboratories, \n1966\u2014. Mr. Hoadley was initially a member of the Statistics Depart- \nment. In 1970, he became a supervisor in the Operations Research \nDepartment. Since 1975, he has been supervisor of the Assurance and \nReliability Theory and Systems Group. Member, American Statistical \nAssociation, American Society for Quality Control, Phi Kappa Phi, \nTau Beta Pi, Pi Mu Epsilon, Sigma Xi.\n\nJohn F. Reiser, B.S. (mathematics), 1973, Michigan State Univer- \nsity; Ph.D. (computer science), 1977, Stanford University; Bell Labo- \nratories, 1977\u2014. Mr. Reiser is a member of the Interactive Computer \nSystems Research Department. His research interests include oper- \nating systems, programming systems and languages, and algorithms. \nMember, Mathematical Association of America, Association for Com- \nputing Machinery.\n\nSteven B. Weinstein, B.S.E.E., 1960, Massachusetts Institute of \nTechnology; M.S.E.E., 1962, University of Michigan; Ph.D. (E.E.), \n1966, University of California at Berkeley; Philips Research Labora- \ntories, Eindhoven, Netherlands, 1967-1968; Bell Laboratories, 1968- \n1980. Mr. Weinstein\u2019s technical interests include data communications,\n\ndata communications security, and microcomputer systems. Senior \nmember, IEEE.\n\nAccommodation and Formation of [1121] Twins in Co Single Crystals. S. \nVaidya and S. Mahajan, Acta Met, 28 (July 1980), pp 1123-31. \nAlpha-Particle-Induced Soft Errors and 64K Dynamic RAM Design Interac- \ntion. R. J. McPartland, J. T. Nelson, and W. R. Huber, Proc Int Reliability Phys \nSymp (1980), pp 261-7.\n\nAnnealing Technique for LEC Grown Twin-Free InP Crystals. W. A. Bonner, \nJ Electrochem Soc, 127 (August 1980), pp 1798-1800.\n\nBroken Hexagonal Symmetry in the Incommensurate Charge-Density Wave \nStructure of 2H-TaSez. R. M. Fleming, D. E. Mocton, D. B. McWhan, and F. J. \nDiSalvo, Phys Rev Lett, 45 (August 18, 1980), pp 576-9.\n\nBrownian Dynamics Study of Polymer Conformational Transitions. E. Hel- \nfand, Z. R. Wasserman, and T. A. Weber, Macromolecules, 13 (May-June 1980), pp 526- \n33.\n\nEpitaxial Crystallization from the Melt of Polyethylene and Polypropylene on \nMica. A.J. Lovinger, Am Chem Soc, Polymer Preprints, 21, No. 2 (1980), pp 253-4. \nGlide Ahead of Terminating [1012] Twins in Co. S. Vaidya and S. Mahajan, Scr \nMet, 14, No. 6 (1980), pp 623-6.\n\nKinetics of Conformational Transition in Chain Molecules. J. Skolnick and E. \nHelfand, J Chem Phys, 72 (May 15, 1980), pp 5489-5500.\n\nLow-Frequency Dynamics of a Cooperative Jahn-Teller Transition: \nTbVO,. R. T. Harley, K. B. Lyons, and P. A. Fleury, J Phys C, 13 (1980), p L447. \nMolecular, Electronic, and Crystal Structure of Naphtho[1,8-cd:4,5- \nc\u2019d\u2019]bis[1,2,6]selenadiazine. A. Gieren, V. Lamm, R. C. Haddon, and M. L. Kaplan, \nJ Am Chem Soc, 102 (July 16, 1980), pp 5070-3.\n\nA Note on the Effect of Incident Beam Convergence on Quantitative Electron \nEnergy Loss Spectroscopy. D.C. Joy, D. M. Maher, and R. C. Farrow, Proc \nMicrobeam Analysis 1980, 1 (August 1980), pp 154-7.\n\nObservation of Direct Excitation of High Orbital Angular Momentum High \nRydberg LEDs by Threshold Energy Electron Collision. S. Tarr, J. Schiavone, \nand R. Freund, Phys Rev Lett, 44 (June 23, 1980), pp 1660-3.\n\nPhotoelectron Spectrum, Including that of Auger Electrons, of Fano Reso- \nnances in Atoms. Y. Yafet, Phys Rev B, 21 (June 1, 1980), pp 5023-30. \nPicosecond Optelectronic Detection, Sampling, and Correlation Measurements \nin Amorphous Semiconductors. D.H. Auston, A. M. Johnson, P. R. Smith, and J. \nC. Bean, App! Phys Lett, 37 (August 15, 1980), pp 371-3.\n\nThe Preparation of Hindered Cuprates from Aldehyde Tosylhydrazones. S. H. \nBertz, Tetrahedron Lett, 27 (1980), pp 3151-4.\n\nRole of Twinning in Cavitation Erosion of Cobalt Single Crystals. S. Vaidya, \nS. Mahajan, and C. M. Preece, Met Trans, 11A (July 1980), pp 1139-50.\n\nThe Complexity of Coloring Circular Arcs. M. R. Garey, D. S. Johnson, G. L . \nMiller, and C. H. Papadimitriou, SIAM J Algebraic Discrete Meth, 1 (June 1980), pp \n216-27.\n\nBlock Coding of Graphics: A Tutorial Review. M. Kunt and O. Johnsen, Proc \nIEEE, 68 (July 1980), pp 770-86.\n\nCoding Method for Vector Representation of Engineering Drawings. K. Ra- \nmachandran, Proc IEEE, 68 (July 1980), pp 813-7.\n\nElectron Channeling Patterns in the SEM. D. C. Joy and R. C. Farrow, ISI \nElectron Opt News, 1 (July 1980), pp 4-10.\n\nAn L.C. Design Station Needs a High Performance Color Graphics Display. N. \nWeste and B. Ackland, 17th Design Automation Conf (June 1980), pp 285-91.\n\nPlanar Electrically-Symmetric n-Way Hybrid Power Dividers/Combiners. A. \nA. M. Saleh, IEEE Trans Microwave Theory Tech (June 1980), pp 555-63.\n\nPlanar, Multiport, Quadrature-Like Power Dividers/Combiners. A. A. M. Sa- \nleh, IEEE Int Microwave Symp Digest, MTT-S (May 1980), pp 483-6.\n\nReal Time Animation Playback on a Frame Store Display System. B. Ackland \nand N . Weste, Siggraph \u201980 Conf Proc (July 1980), pp 182-8.\n\nShort-Circuit Memory in Electrochromic Displays. G. Beni, Appl Phys Lett, 37 \n(July 1, 1980), pp 106-8.\n\nSimple Flying Spot Scanner for Electron Beam Lithography on a Scanning \nElectron Microscope Without Beam Blanking Capability. P. Grabbe, Rev Sci \nInstrum, 51 (July 1980), pp 992-4.\n\nSingle-Mode Stabilization by Traps in Semiconductor Lasers. J. A. Copeland, \nIEEE J. Quant Electron, QE-16 (July 1980), pp 721-7.\n\nTheorems on Match and Isolation in Multiport Networks. A. A.M. Saleh, IEEE \nTrans Microwave Theory Tech, MTT-28 (April 1980), pp 428-9.\n\nWarpage Reduction During Wave Soldering. R. J. Waltner, Conf Proc Inst \nInterconnecting Packaging Electron Circuits (April 1980), pp 1-19.\n\nApplication of Clustering Techniques to Speaker Independent Word Recogni- \ntion. S. E. Levinson, L. R. Rabiner, A. E. Rosenberg, and J. G. Wilpon, IEEE Trans \nAcoustics, Speech, and Signal Processing, ASSP-27 (April 1979), pp 134-40.\n\nCochear Models: Evidence in Support of Mechanical Nonlinearity and a Second \nFilter (A Review). J. L. Hall, Hear Res, 2 (1980), pp 455-64.\n\nCochlear Models: Two-Tone Suppression and the Second Filter. J. L. Hall, J \nAcoust Soc Am, 67 (May 1980), pp 1722-8.\n\nA Conversational Mode Airline Information and Reservation System Using \nSpeech Input and Output. 8S. E. Levinson and K. L. Shipley, Proc 1980 ICASSP, 1 \n(April 1980), pp 203-8.\n\nFrequency Selectivity of the Cochlea for Formant Peaks at High Signal \nLevels. J. L. Hall, J Acoust Soc Am, 68 (August 1980), pp 480-1.\n\nLoudness of Noise in the Presence of Tones: Measurements and Nonlinear \nModel Results. J. L. Hall and M. R. Schroeder, Psychophysical, Physiological and \nBehavioral Studies in Hearing, edited by G. van der Brink and F. A. Bilsen, The \nNetherlands: Delft University, 1980, pp 329-32.\n\nMeasuring and Modeling User Satisfaction With Telephone Switching and \nTransmission Performance. H. S. Cohen, Int Symp Human Factors in Telecom- \nmunications, 9th Proc (1980), pp 237-42.\n\nA New System for Continuous Speech Regulation\u2014Preliminary Results. S.E. \nLevinson and A. E. Rosenberg, Proc 1979 Int Conf Acoustics, Speech and Signal \nProcessing, J (April 1979), pp 239-43.\n\nSome Notes on Reading: Talkers, Material and Reading Rate. N. Umeda and \nA. M.S. Quinn, J Speech Hear Res, 23-1 (March 1980), pp 56-72.\n\nCollusive Behavior in Noncooperative Epsilon Equilibria of Oligopolies with \nLong but Finite Lives. R. Radner, J. Economic Theory, 22 (April 1980), pp 136-54.\n\nOn the Use of Dynamic Time Warping for Word Spot- \nting and Connected Word Recognition\n\nThe Material Dispersion Zero in Infrared Optical \nWaveguide Materials", "title": "magazine :: Bell System Technical Journal :: BSTJ V60N02 198102", "trim_reasons": [], "year": 1981}
{"archive_ref": "bitsavers_BellSystemJV62N09198311Part1_8280466", "canonical_url": "https://archive.org/details/bitsavers_BellSystemJV62N09198311Part1_8280466", "char_count": 298484, "collection": "archive-org-bell-labs", "doc_id": 710, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc710", "record_count": 491, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bitsavers_BellSystemJV62N09198311Part1_8280466", "split": "test", "text": "Penetration of Radio Signals Into Buildings in the Cellular Radio 2719 \nEnvironment \nE. H. Walker\n\nTransmission Errors and Forward Error Correction in Embedded 2735 \nDifferential Pulse Code Modulation \nD. J. Goodman and C.-E. Sundberg\n\nCCITT Compatible Coding of Multilevel Pictures 2765 \nH. Gharavi and A. N. Netravali\n\nPerformance of the Queueing Network Analyzer 2817 \nW. Whitt \nPAPERS BY BELL LABORATORIES AUTHORS 2845\n\nCalculation of Modes in an Optical Fiber Using \nthe Finite Element Method and EISPACK\n\nThis paper presents a method for computing the propagation modes of a \ncircular optical fiber. Finite element analysis reduces Maxwell\u2019s equations to \nstandard eigenvalue equations involving symmetric tridiagonal matrices. Rou- \ntines from the Eigensystem Package (EISPACK) compute their eigenvalues \nand eigenvectors, and from these the waveforms, propagation constants, and \ndelays (per unit length) of the modes are obtained. An extension allows loss \nof leaky modes to be calculated. Examples indicate that the method is reliable, \neconomical, and comprehensive, applying to both single and multimode fibers.\n\ndetermine pulse dispersion along the fiber). The technique combines \nthe Finite Element Method (FEM) and routines from the Eigensystem \nPackage (EISPACK)\u2019 to achieve an efficient calculation of these modal \nquantities for both single and multimode fibers.\n\nThe most popular approach to modal calculations for multimode \nfibers has been the Wentzel, Kramers, Brillouin (WKB) method, \nwhere Maxwell\u2019s equations are approximated by an easily integrated \nfirst-order differential equation.2 WKB analysis provides a simple \nmodel for understanding optical transmission through a fiber\u00ae and has \nguided fiber design.*\n\nAccuracy of the WKB method declines substantially with the num- \nber of propagating modes. For single-mode fibers the WKB method is \nnot suitable, and instead, various analytic and numerical techniques \nhave been used.\n\nAnalytic calculations have centered about the step-index profile and \nthe infinitely extending parabolic profiles. Modal quantities have been \nexpressed in terms of Bessel functions for the single step\u2019 and, also, \nthe double step.\u00ae Parabolic profiles, analyzed by analogy with the \nharmonic oscillator,\u00ae have well-known expressions for their modal \nquantities. Coupled with perturbation analysis,\u2019\u00b0 these results cover a \nbroad range of profiles. But numerical approaches permit an even \nmore comprehensive treatment.\n\nSeveral numerical procedures have extended the analysis of the \nstep-index profile. The solution in the core comes from a numerical \nintegration; the solution in the cladding (where the index is assumed \nconstant) is well known. Boundary conditions at the core-cladding \ninterface link the two solutions and lead to a set of linear equations \ninvolving the propagation constant as a parameter. A search of the \ncorresponding determinant for zeroes gives the propagation constants \nand, subsequently, the waveforms and delays.\n\nIn Ref. 11 this procedure is applied to Maxwell\u2019s first-order vector \nequations, and important results have been obtained for multimode\u2019\u201d\u201d? \nand single-mode\u2019*\u201d\u2019 fibers. A similar scheme\u2019\u00ae deals with two coupled \nsecond-order differential equations equivalent to Maxwell\u2019s equations. \nThe second-order scalar wave equation approximates Maxwell\u2019s equa- \ntions by neglecting the relatively small gradient index terms. In Ref. \n17 the basic procedure is applied to the scalar wave equation, and a \nvariety of results have been obtained for single-mode fibers. !\u00ae9\n\nfor the vector equations. Approximate solutions in the core and clad- \nding are matched at the interface by performing a determinant search, \nas described before. The FEM has also been applied\u201d? to diverse single- \nmode waveguides by approximating the index profile by piecewise \nconstant functions and then enforcing boundary conditions across the \nnumerous interfaces. The result is a matrix eigenvalue equation in \ngeneralized form. Both of these FEM approaches seem limited to a \nsmall number of modes for economical operation.\n\nThe approach in this paper is characterized by a sequence of three \nsteps. First, Maxwell\u2019s equations are transformed to ordinary differ- \nential equations in eigenvalue form. In one case, gradient index terms \nare neglected, and in the second, they are considered to first order so \nthat their effect can be monitored. Next, finite element analysis, using \nthe Galerkin technique,\u201d reduces these differential equations to matrix \nequations in standard eigenvalue form. The matrices are symmetric \nand tridiagonal, and their positive eigenvalues correspond to the \npropagation modes. The routine BISECT in the EISPACK? library \ndelivers the eigenvalues and TINVIT, also in EISPACK, delivers the \ncorresponding eigenvectors. The eigenvalues give the propagation \nconstants of the modes, the eigenvectors give the waveforms, and a \ncombination of the two give the delays.\n\nLike many other numerical techniques, this method can treat any \nuniform, circular fiber and meet usual standards of accuracy. Also, the \neffect of gradient index terms can be monitored. But by casting the \nequations in standard eigenvalue form, modern techniques of compu- \ntational linear algebra (as used in EISPACK) can achieve substantial \ncost advantage over other numerical approaches. Typically, this \nmethod will process for a multimode fiber 25 modes per second on the \nCray-1* computer.\n\nThe calculation procedure is derived in the next section. Effects of \nmaterial dispersion are incorporated into the analysis, and calculation \nof loss for leaky modes is also considered. Results are given in Section \nIII for a variety of single and multimode examples. These results, as \ndiscussed in Section IV, illustrate the reliability, economy, and scope \nof the method.\n\nThis section derives from Maxwell\u2019s equations an algorithm that \ncomputes the propagation modes of an optical fiber. The algorithm is \nstraightforward and can be carried out on any computer that has \naccess to the EISPACK' or similar routines.\n\nThe fiber is assumed perfectly straight and circular, and uniform \nalong its length. The cylindrical coordinate system (r, 0, z) is defined \nso that the z-axis coincides with the fiber axis. The index of refraction \ncan then be expressed by the function n(r), the index profile. The \nprofile can be any bounded function in the core (r < R,), but it is \nconstant (n,.) in the cladding. The geometry is shown in Fig. 1.\n\nwhere \u00ab, denotes the free-space permittivity. The permeability yu is \nassumed throughout to be yu,, the free-space value.\n\nMaxwell\u2019s equations relate the electric field HE and the magnetic \nfield H by\n\nwhere w denotes the frequency of excitation (in rad/s). Taking the \ncurl of the second equation eliminates E to give\n\nwith n the index of refraction, c the velocity of light in vacuum, and ) \nthe free-space wavelength of the excitation.\n\nwhere H, is a vector field independent of z, and 8 is the propagation \nconstant of the mode. Substituting this into the field equation gives, \nin cylindrical coordinates,\n\nwhere g, means dg/dr. The transverse (i.e., r and 8) components of H \nare uncoupled from the longitudinal (i.e., z) component and satisfy an \neigenvalue equation with eigenvalue 6\u201d. The corresponding operators \nare indicated as a 2 X 2 submatrix in the full 3 X 3 matrix.\n\nwhere the functions hy and h, depend only on r. The two forms \ncorrespond to different polarizations. Substituting these into eq. (7) \ngives\n\nwhere the + sign depends on the polarization. \nThe case where m\u2019 = 0 reduces to two uncoupled equations:\n\nfor the Transverse Magnetic (TM) modes for which H, is identically \nzero, and\n\ni a ee re \n(2,4 b+) b= Bh, (11) \nfor the Transverse Electric (TE) modes for which E, is identically \nzero. Often, gradient index terms (those involving g,) are neglected on \nthe basis that profile variations are relatively small.* The difference \nor splitting in 6? for the TM and TE modes measures the accuracy of\n\nwhere (A \u2014 B)fp = 6f2. The second term of eq. (12) is neglected \nbecause its first order perturbation contribution is zero, as in general\n\nrespectively. If two eigenvalues are equal or nearly equal, a degenerate- \nperturbation calculation may be required.\n\nTo first order, the modes when m\u2019 # 0 are either the EH, with \ntransverse H fields of the form\n\nkdrd_  (m\u2019 +1) (m' + 1) tH ; \n(' dr kdr r ae ie \nor the HE,,, with \n_ [f cos m\u20196 f sin m\u20190 \nHor = (/ sin m\u20196 and \u2014f cos m\u20196)\u2019 (19) \nwhere f satisfies \nRedo d? (m= 1)? m1) ale \u2014 92 \n(' dr k dr ez op 1 eas a \nWith m = m\u2019 +1, eqs. (18) and (20) can be combined as \nkdrd m_m 7, ee \n(pore Ry 4 elm oi (21)\n\nThe modes are indexed by the angular parameter m. For m = 0 \nthere are two polarizations for each HE,, mode. This group includes \nthe fundamental mode HE),;, which propagates in single-mode fibers. \nFor m = 1 there are two polarizations for the HE:, modes, also the \nTM modes and the TE modes. For m > 1 there are two polarizations \nfor the HE n+1,.. modes and two for the EH \u00bb-1,, modes.\n\nWhen gradient index terms are neglected, eq. (9) reduces to the \nscalar wave equation\n\n(:2, dr er ea (22) \nFor m = 0 each solution represents two polarizations, i.e., two modes; \nfor m # 0 each solution represents two polarizations and two angular \nharmonics, m\u2019 = m + 1, 1.e., four modes.\n\nThe scalar wave equation and its counterparts [eqs. (10) and (21)] \nassume the form of a time-independent Schroedinger equation in two \ndimensions. In particular, the scalar wave equation can be expressed \nas\n\nThe quantity (k? \u2014 k3) plays the role of a potential that is 0 in the \nouter cladding and beyond; (6? \u2014 k4) is an eigenvalue. \nResults on Schroedinger\u2019s operators\u201d imply that the propagation\n\nmodes correspond on a one-to-one basis to the positive eigenvalues \nand the number of such modes is finite (see Ref. 23, p. 366). This fact \nimplies that increasingly accurate estimates of the propagation modes \ncan be obtained by ever finer discretization of the equations. By \ncontrast, finer discretizations introduce ever more negative eigenval- \nues corresponding to the continuum of radiation or unbound modes.\n\nThe Finite Element Method (FEM) can solve to desired accuracy \nthe differential equations just derived. The present discussion special- \nizes to the scalar wave equation [eq. (23)]; modifications needed for \nthe equations containing gradient index terms are indicated in the \nappendix.\n\nThe solution function f(r) is approximated by a piecewise linear \nfunction. This can be expressed in terms of interpolation functions \n(shown in Fig. 2) as\n\nThe first and last terms of the series are affected by end conditions \non f(r). At the center (r = 0) of the fiber f(r) and its radial derivative \nmust be bounded and well defined; so f(0) = 0 when m > 0 and df/dr \n(0) = 0 or equivalently f(0) = f(6) to first order in 6 when m = 0.\n\nAt the other end, R = L\u00e9 represents a truncation radius in the \ncladding, and R + 6 represents the truncated part in a way that will \nnow be explained. Assuming that the cladding extends indefinitely, \nthe solution there is\n\nwhere K,, denotes the mth order modified Bessel function of the \nsecond kind,\u201d a is an unknown coefficient,\n\nf'/f = nKn(nR)/Kn(nR). (28) \nA Taylor\u2019s expansion of f to first order about r = R yields \nf(R + 6) = f(R) + 6f\u2019(R) \n= f(R)Q + dnKn(nR)/Kn(nR)). (29)\n\nMost modes of a multimode fiber are unaffected by the end condition \nin eq. (29) because their waveforms are negligibly small beyond the \ncore. But the waveform of a single-mode fiber can extend beyond the \ncore and then the end condition needs to be enforced.\n\nSample values of f(r), the coefficients in eq. (25), are estimated by \nthe Galerkin weighted residual method.\u201d Although a piecewise linear \napproximation cannot satisfy the scalar wave equation, it can satisfy \nweighted averages of the equation. In the Galerkin technique the \nweightings are chosen as the basis functions N,(r) j = 1, ---, L. In \nterms of inner products, defined for any functions p(r) and q(r) as\n\nfor j= 1, ---, L. The first term comes from an integration by parts. \nSubstituting the piecewise linear approximation of f [from eq. (25)] \ninto the Galerkin equations gives L equations for L + 2 sample values, \nbut the end conditions eliminate two of these values. The result is L \nequations in L unknowns, expressed as the matrix equation,\n\n(A \u2014 m?B + C)f = 67(8? \u2014 k2)Df. (32) \nThe column vector f denotes the sample values, \nf =[f(ri), \u00ab++, ADI. (33) \nThe L X L matrix A has jl element \n_ _(@N: aj). \naj = (a > dr ) (34)\n\nfor all j and / except, to accommodate the end conditions, ayo must be \nadded to a); when m = 0 and az,74:f14:/f, must be added to az, for all\n\nThe latter three matrices are evaluated by \u201clumping the masses,\u201d \nmeaning that the integrals are evaluated numerically and the integra- \ntion points are chosen as the sample points.\u201d\u201d The trapezoidal rule \nthen yields 0 for the off-diagonal elements and for the main diagonal \ngives\n\nbn = 6\u00b0/l Ci = I[k?(r1) = k2]64 dn = 16? (36) \nfor!=1,---,L. The matrix A can be evaluated exactly. It is symmetric \nand tridiagonal (i.e., a, = 0 if |! \u2014 7| > 1) and has\n\nforl!=1,---, L. When m = 0, ay is \u20141.56\u201d. \nEq. (32) converts to a standard symmetric eigenvalue problem by \nmultiplying both sides by the diagonal matrix D~'\u201d and putting\n\nstandard form relative to the vector g. \nThe key matrix, T, is symmetric and tridiagonal. For m # 0\n\nEquation (39) represents the FEM reduction of the scalar wave \nequation. The other differential equations, as indicated in the appen- \ndix, reduce to the same form, with T symmetric and tridiagonal. The \nsample distance 6 is discussed in Section III.\n\nThe propagation modes are calculated by solving for the positive \neigenvalues (4) of 7. The associated propagation constants are\n\nWhen the end condition is enforced, the T matrix depends on @ in its \n(L, L) entry, but a simple iteration procedure yields the proper 8. The \nassociated waveforms are given by f or g.\n\nThe modal delays per unit length (7,) are the reciprocal of the group \nvelocities,\n\nwhere g is assumed to be normalized (i.e., g-g = 1). This follows \nstandard procedure for taking the derivative of an eigenvalue with \nrespect to a parameter.\u201d\u00b0 Equations (40), (41), and (42) for T imply \nthat dT/dw\u201d is a diagonal matrix. Using the equivalence between w*d/ \ndw? and \u2014\\2d/d)?, eq. (47) becomes\n\nTo form the T matrix, the index profile n(r) must be available for \nany excitation wavelength (A). Sellmeier expansions\u201d\u2019 of the form\n\ni=1 \nhave been fitted to measured values of refractive index over the range \nof wavelengths 0.8 um to 1.5 um for bulk samples of pure SiOs, 13.5- \npercent Ge doped SiOz, and 1 percent F doped SiO, (denoted here as\n\na, b, and c, respectively). Also, results in Ref. 28 indicate that the \nindex is approximately linear with concentration. Therefore, the index \nprofile of an optical fiber having a graded Ge dopant is taken as\n\nwhere C,(r) denotes the concentration profile for Ge, relative to 13.5 \npercent. For dual dopants, Ge and F, the index profile is\n\nThe concentration profiles in these equations may be specified or \nmay be deduced from the index profile at a reference wavelength. Once \nthe concentration profiles are available, the index profile and its ? \nderivative can be determined from eq. (50) or eq. (51) for any wave- \nlength.\n\nThe T matrix can be expressed in dimensionless form by replacing \nthe sample spacing by\n\nwhere RF, denotes the core radius and L, the number of samples in the \ncore. This gives\n\nwhere now C,(r) represents the Ge concentration normalized with \nrespect to n,, the index at the center. The latter term in eq. (54) can \nusually be neglected because n; ~ nq and often, C,(r) ~ 1 for most r. \nThen the 7 matrix can be expressed in terms of the usual V number,\n\nwhere now an iteration on V, is required. A similar expression, involv- \ning three V numbers, holds for dual dopants. \nThe effective index (n,) is obtained from V, in eq. (56). The delay\n\nd \ncc \u2014 na) \u2014 d\u00b0 ne (nj \u2014 nap|t (58) \nMatrix T is always symmetric and tridiagonal. The routine BISECT \nin the EISPACK\u2019 library computes the positive eigenvalues for such \nmatrices and TINVIT, also in EISPACK, computes the associated \neigenvectors g. These routines operate efficiently and reliably.\n\nWhen the index profile drops below n,; over part of its range, then \nleaky modes may exist. These modes have n, <n, and complex prop- \nagation constants. The corresponding attenuation accounts for radia- \ntion loss as the waveform spreads out radially. As is well known, leaky \nmodes also can arise when m # 0 from the negative term, \u2014 (rm?/r\u2019), \nin the propagation equation.\n\nLeakage loss is calculated by truncating the waveform at the begin- \nning of the outer cladding and enforcing the proper end condition. In \nterms of the complex propagation constant y, eq. (27) for n changes \nto\n\nand the end condition [in eq. (43)] becomes complex. Now, the T \nmatrix has a real (7) and an imaginary (T;) part, but matrix T; \nconsists of all zeroes except the (L, L) diagonal element (call it x). \nFirst-order perturbation theory estimates an eigenvalue of T as\n\nwhere py, is an eigenvalue of T, and g; is the Lth component of the \nassociated normalized eigenvector g. Again, an iteration is required \nbecause T depends on y.\n\nIn this section modal calculations based on the results of Section II \nare illustrated in 10 examples. The next section summarizes and \nevaluates the results.\n\n3.1 Infinite parabolic profile \nThe parabolic profile that extends without radial limit has index\n\nfor all r, where R denotes the nominal radius of the fiber core. Under \nthe scalar wave approximation, its modal waveforms are Laguerre \nGaussian functions, its propagation constants are\n\nBum = [Rani \u2014 2Qkan,(2A)?/R]\u201d, (62) \nand its modal delays (neglecting material dispersion) are \nTum = (Bum + Ran}/Bym)/2w, (63) \nwhere the mode number is \nQ=2n+mt+1 m,n=0,1,--:, (64)\n\nwith m the angular harmonic and yu the radial harmonic. Each Q \nrepresents a group of modes that have the same propagation constant \nand delay (see Ref. 9).\n\nModal delays were calculated assuming parameter values A = 0.013, \nR = 25 um, and A = 1.3 wm. Convergence with respect to truncation \nradius (TR) was complete (within 0.1 ps/km) for each mode for TR = \n1.5R. Convergence with respect to the number (L;) of sample points \nin the core was within i ps/km in the rms delay for L, = 400, as \nindicated in Table I. An accuracy of 1 ps/km determines the bandwidth \nwithin 5 percent for a 10-GHz X km fiber and 0.5 percent for a 1-GHz \nxX km fiber. The percent errors of the computed delays (with L, = 500) \nare shown in Fig. 3.\n\nThe most accurately computed mode within a mode group had radial \nnumber yu = 0; their waveforms have no zero-crossings and are most \neasily approximated by piecewise linear functions. The least accurately \ncomputed mode (which was off by 5.5 ps/km) had the largest pv. As \nanother measure of accuracy, the spread in the computed delays for \neach Q is also given in Table I.\n\nTable I\u2014Convergence results for \ninfinite parabolic profile in units of\n\nFig. 3\u2014Computation errors in modal delays for infinite parabolic profile.\n\n0.005 and R = 5 um, but again \\ = 1.3 um. The calculated effective \nindex of the HE,, mode was accurate to seven decimal places and the \ndelay to 0.1 ps/km when TR was 2.5R and L, was 300. This precision \ntranslates to nm accuracy in the zero-dispersion wavelength. The \nbeam radius (BR) defined by the condition\n\nwas accurate to four decimal places, as indicated in Table II. The \nnumber of samples in the core can be less in the single-mode case \nbecause the HE,, waveform does not oscillate; the truncation radius \nneeds to be greater because the waveform extends farther into the \ncladding.\n\nAssuming parameter values of A = 0.013, R = 25 um, A = 1.3 wm, and \na = 2 (the parabolic case), effective index (n.) and delay (7z) were \ncomputed for each propagation mode under the scalar wave approxi- \nmation. The rms delay (trms) converged to 1.793 ns/km within 0.3 \npercent and the rms delay with the two highest-order-mode groups \ndeleted (7\u2018pms) converged to 113 ps/km within 0.9 percent when L, = \n300 and TR = 0.7R.\n\nThe delays spread almost 10 ns/km for Q = 14 and 3 ns/km for Q = \n13, but less than 1 ns/km for the others. The spread in n, is negligible, \nreaching a maximum of 0.01 percent.\n\nFigure 5 shows n, and 7, plotted jointly on an expanded delay scale \nwith outliers excised. To compare with WKB results, n, and 7, for the \ncorresponding infinite parabola are indicated. The reference delays \nexceed all in their mode group: by as little as 0.2 ps/km for each of \nthe first nine mode groups, but by more than 34 ps/km and 86 ps/km \nfor Q = 13 and 14, respectively.\n\nGradient index terms cause modal delays to spread still further. \nFigure 6 shows the delay differences caused by these terms for each \nmode of the parabolic fiber. Of the 112 modal delays only 14 differed \nby more than 20 ps/km from the corresponding scalar wave values. \nThe TM mode nearest cutoff had the largest difference (33 ps/km).\n\nEFFECTIVE INDEX \nFig. 5\u2014Plots of 7, versus n, for cladded and infinite parabolic profiles.\n\nFig. 6\u2014Absolute differences in modal delays due to gradient index terms for parabolic \nfiber.\n\nThe change in trms was 5.9 ps/km and in 7\u2019gms it was less than 1.5 \nps/km. The error is less than 1.5 percent in either case. In the examples \nthat follow gradient index terms are neglected.\n\nMaterial dispersion causes the bandwidth of power-law fibers to \ndepend sharply on the excitation wavelength (A). Here and in the \nfollowing examples, the profile is assumed known at the reference \nwavelength of \\ = 0.8 um; the concentration profile and the index \nprofile at other wavelengths are determined, as discussed in the \nprevious section.\n\nBandwidth is defined as the half-power frequency of the transfer \nfunction over some distance. When intramodal dispersion is neglected, \nthe transfer function is\n\nwhere 7, denotes the delay and a?, the power of the nth mode. Defining \nthe rms delay (7Rms) by\n\nTRMS = b in ie = Tag) / x oa) (68) \nand the average delay by \nTay =) AnTn / y an, (69) \nthen to second order in the w(t, \u2014 Tav), the bandwidth is \nBW = 1/(277Rs), (70)\n\nBandwidth was calculated on this basis over a range of wavelengths \nfor the parabolic fiber of example 2, assuming the tapered modal- \npower distribution shown in Fig. 7. Lower-order modes are weighted \nmore heavily than the higher and the highest are omitted as the higher \nmodes are increasingly vulnerable to microbending and cladding ab- \nsorption.\n\nFigure 8 shows the spectral plot of bandwidth. The peak value \noccurs at \\ = 0.963 wm, compared to 0.983 um reported in Ref. 13.\n\nThe spectral plot is also shown for the case where material dispersion \nis neglected. The figure shows that the bandwidth is lower and essen-\n\nFig. 8\u2014Plots of bandwidth versus wavelength for parabolic fiber with and without \nmaterial dispersion, using tapered modal-power distribution.\n\ntially constant except for small discontinuities associated with new \nmodes cutting in. Other calculation indicates that bandwidth increases \nto a peak at 9.5 GHz X km for a = 1.98 compared to an optimum a = | \n1.97 estimated by WKB analysis.\u2018\n\nMaterial dispersion causes the optimum a of power-law fibers to \ndepend on excitation wavelength. Figure 9 shows bandwidth versus a \nwhen A = 0.013 and R = 25 um for \\ = 0.82 wm and A = 1.32 um. The \noptimum a\u2019s are 2.07 for the former and 1.88 for the latter, compared \nto values of 2.081 and 1.884, respectively, reported in Ref. 13. The \npeak bandwidths are 5.48 GHz x km and 6.02 GHz Xx km, respectively.\n\nThe fractional power that propagates in the cladding can be used to \nderive an alternate modal-power distribution. Modes with more than \n0.1 percent of their power in the cladding will lose most of their power \nover 1 km for typical cladding losses of 1 dB/m. A realistic modal\n\nFig. 9\u2014Plots of bandwidth (in power-law fibers) versus a for \\ = .82 wm and \\ = \n1.32 um, using tapered modal-power distribution.\n\ndistribution would neglect these modes and for simplicity could weight \nthe others equally.\n\nFigure 10 shows bandwidth versus a for the two cases of example 5, \nassuming the new distribution. The optimum a is lower by 0.01 and \nthe peak bandwidth is increased by 40 to 50 percent in either case. \nTable III indicates the modes neglected in the bandwidth calculation \nwhen A = 1.32 wm and a = 1.88.\n\nLayer structure in actual fibers perturbs the ideal profile. In this \nexample power-law profiles are approximated by steps that span equal \nareas and match the ideal profile at the mid-area points of the layers \n(see Fig. 11). Bandwidth at \\ = 1.32 um (using the modal distribution \nin example 6) is shown versus a and the number of steps (NS) in Fig. \n12. The graphs indicate that the optimum a is essentially independent \nof NS, but the peak bandwidth and its sharpness decrease for smaller \nNS.\n\nFig. 10\u2014Plots of bandwidth versus a for \\ = .82 um and A = 1.32 um assuming equal \nexcitation of modes having less than 0.1 percent power in cladding.\n\nDelay (per km) of the HE,, mode was computed versus ) for an \nactual single-mode fiber based on its measured index profile. The \nprofile measurement was performed on the associated preform and is \nshown in Fig. 13. The depression of the inner cladding is due to F \ndoping.\n\nAs the calculation presumes radially symmetric profiles, the right \nand left sides were considered separately. About 750 sample points \nwere used in either case.\n\nThe radial scale of the fiber profile was assumed to depend linearly \non that of the preform. The fiber core radius (R) was estimated from \nthe preform core radius (RP), fiber diameter (DF), and preform \ndiameter (DP) as\n\nwhere the 1-percent addition estimates the SiO, loss in the outer \ncladding during the draw process.\n\nFigure 14 shows computed differential delay versus ) for the right \nand left profiles together with measurements. It was assumed that \nmeasurement matched calculation exactly at \\ = 1.32 um, because \nonly relative delays were measured. The zero-dispersion wavelength \n(at the delay minimum) was computed as \\, = 1.3127 um for the left \nprofile and \\, = 1.3113 um for the right, compared to measurement of \nho = 1.3114 pm.\n\nLeakage loss (in dB/m) of the TE mode of the single-mode fiber in \nthe preceding example was calculated for both sides of the profile as a \nfunction of \\ and is shown in Fig. 15. The difference between the left \nand right profiles becomes evident in this figure. A loss of 4 dB/m has \nbeen identified\u201d? with effective cutoff, and corresponds to cutoff wave- \nlengths, A, = 1.17 wm for the left profile and \\, = 1.23 um for the right, \ncompared to a measurement of 1.28 um. This discrepancy is discussed \nin the next section.\n\nThe equivalent step approximation (where the index of the core and \nthe inner cladding are area-weighted averages of the profile) gives \nAo = 1.808 um and i, = 1.21 um. The X, value straddles the previous \ncalculated values, but the A, value is somewhat less.\n\nThe dimensionless formulation allows design curves to be generated \nefficiently. Figure 16 shows V, versus V for the HE,, mode of three \npower-law profiles. The specific parameters (R, A, \\) determine V and, \nhence, V. and dV,/dV. Delay versus \\ is computed according to eq. \n(58) and X, is estimated at the minimum of the delay curve. Figure 17 \nshows A, versus R for two values of A for triangular profiles (a = 1).\n\nA method for computing modes of a circular optical fiber has been \npresented. The finite element method reduces Maxwell\u2019s equations to \nthe standard eigenvalue problem, involving tridiagonal matrices. Rou- \ntines from EISPACK exploit the tridiagonal form to compute the \neigenvalues and eigenvectors efficiently. From these the modal quan- \ntities are obtained.\n\nUsing a piecewise linear approximation of the waveform is necessary \nto get tridiagonal form. Although piecewise quadratic and cubic ap- \nproximations of the waveform can lead to smaller matrices, tridiagonal\n\nform is lost and the eigencalculations would be less efficient. Extension \nto elliptical and other nonradially symmetric fibers leads to similar \ndifficulty in the eigen calculations.\n\nThe method applies to any circular fiber and can account for \ngradient index terms to first order. As illustrated in the 10 examples \nof Section III, the calculations can provide information on the radial \ndistribution of propagating power, on pulse dispersion (arising from \nmaterial, intermodal, and intramodal dispersion), and on leakage loss.\n\nThe accuracy of the method was tested for the infinite parabolic \nprofile of the first example. After convergence was established, the \nmodal quantities were compared with the actual values known through \nanalysis. Agreement was excellent for both a multimode and a single- \nmode profile.\n\nThe WKB and scalar wave approximations were tested for the \nparabolic fiber in the next two examples. The cladding of the fiber \naltered the delays of the higher-order modes substantially, a facet \nmissed by WKB analysis. The gradient index terms contributed less \nthan 1.5 percent to the rms delay; and so, for most purposes the scalar \nwave approximation suffices for this fiber. The validity of these\n\napproximations may change in other fibers; the value of the method \nis that it permits the test.\n\nMaterial dispersion was incorporated into the calculation in the \nremaining examples. The variation of refractive index with wavelength \nand dopant concentration was modeled by Sellmeier expansions based \non measurements of bulk samples with certain dopant concentration.\u201d\" \nA linear dependence of index on concentration was assumed, but is \nnot essential to the method. Irreversible thermal and stress effects*\u201d \nmight also be incorporated.\n\nBandwidth was estimated in examples 4 and 5 for power-law fibers \nassuming a tapered modal-power distribution. With the higher-order \nmode groups deemphasized, results essentially agreed with WKB \nanalysis. The usual sharp peak in bandwidth occurred in the spec- \ntral plot for the parabolic (a = 2) fiber in example 4 and in the \na-dependence in example 5. Calculation of optimum a agreed reason- \nably well in these cases with other numerical work and WKB calcu- \nlations.\n\nThe bandwidth calculation was repeated in example 6 using a \ndifferent modal-power distribution. Modes with more than 0.1 percent \nof their power in the cladding were neglected, as being too lossy to \nmaintain power. The optimum a\u2019s were essentially the same, but peak\n\nFig. 14\u2014Measurement and calculations for left and right profiles of differential delay \nversus wavelength.\n\nbandwidths were somewhat higher. This example could be expanded \nto simulate differential mode attenuation, mode mixing, concatenation \nof dissimilar fibers, and other longitudinal variations.\n\nPerturbations of a profiles usually lower bandwidth. Example 7 \nconcerned the effect of layer structure on bandwidth and showed the \nexpected decline with a smaller number of layers. Other perturbations \nsuch as ripple can also be treated. As a measure of its efficiency, the \nmethod used about two seconds per profile on the Cray-1 for this \nexample.\n\nThe last three examples concerned single-mode fibers. The first two \nof these involved an actual fiber whose profile was measured in the \npreform stage. Calculation of delay (per km) versus \\ matched meas- \nurement extremely well, but calculated leakage loss exceeded meas- \nurement, giving a 7-percent average underestimate of cutoff wave- \nlength (A,). Other calculations of leakage involving only slightly de- \npressed inner claddings, where \\, matches measurement to within 1 \npercent, suggest that the discrepancy may involve material changes in \nthe F-doped glass caused by the draw process (consistent with obser-\n\nWAVELENGTH IN MICROMETERS \nFig. 15\u2014Plots of computed leakage loss versus wavelength for left and right profiles.\n\nFig. 16\u2014Plots of V. versus V for the HE,, mode of three power-law profiles. \n2690 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1983\n\nFig. 17\u2014Plots of computed zero-dispersion wavelength (\\,) versus core radius (R) \nfor two values of A, for triangular profiles.\n\nvations in Ref. 31). Such changes need to be understood when pre- \ndicting transmission performance from a preform profile. Similar \nvalues for \\, and \\, were obtained for the equivalent step profile (with \ndepressed inner cladding), but the lower i, indicates the effect of \nprofile structure.\n\nThe dimensionless formulation permits the greatest efficiency in \ngetting design curves. Zero-dispersion wavelength (\\,) for triangular \nprofiles is calculated in example 10 showing the expected shift\u00ae\u201d to \nhigher wavelengths. Spot size or cutoff wavelength can be obtained \nwith similar efficiency.\n\nIn summary, the calculation method is reliable, and relatively in- \nexpensive. In the context of circular fibers it is comprehensive, capable \nof simulating diverse effects.\n\nThe author is grateful for useful discussions with I. A. White and \nA. J. Ritger on multimode fibers, W. T. Anderson and P. F. Glodis on\n\nThe author is also grateful to J. S. Nobles and D. L. Philen for fiber \nmeasurements and to H. W. Friedrichsen for programming support.\n\n1. B. T. Smith et al., Matrix Eigensystem Routines\u2014EISPACK Guide, New York: \nSpringer Verlag, 1976.\n\n2. P. M. Morse and H. Feshback, Methods of Theoretical Physics, New York: McGraw- \nHill, 1953, p. 1092.\n\n4. D. Gloge and E. A. J. Marcatili, \u201cMultimode Theory of Graded-Core Fibers,\u201d \nB.S.T.J., 52, No. 9 (November 1973), pp. 1563-78.\n\n5. C. Pask, \u201cExact Expressions for Scalar Modal Eigenvalues and Group Delays in \nPower-Law Optical Fibers,\u201d J. Opt. Soc. Am., 69, No. 11 (November 1979), pp. \n1599-1603.\n\n6. G. A. E. Crone and J. M. Arnold, \u201cAnomalous Group Delay in Optical Fibers,\u201d Opt. \nQuant. Elect., 12, No. 6 (November 1980), pp. 511-7.\n\n8. S. Kawakami and S. Nishida, \u201cCharacteristics of a Doubly Clad Optical Fiber with \na Low-Index Inner Cladding,\u201d IEEE Trans., QE-11 (December 1974), pp. 879-87.\n\n10. R. A. Sammut and A. K. Ghatak, \u201cPerturbation Theory of Optical Fibers with \nPower-Law Core Profile,\u201d Opt. Quant. Elect., 10, No. 6, (November 1978), pp. \n475-82.\n\n11. G.E. Peterson, A. Carnevale, U. C. Paek, and D. W. Berreman, \u201cAn Exact Numerical \nSolution to Maxwell\u2019s Equations for Lightguides,\u201d B.S.T.J., 59, No. 7 (September \n1980), pp. 1175-96.\n\n12. G. E. Peterson, A. Carnevale, and U. C. Paek, \u201cComparison of Vector and Scalar \nModes in a Lightguide with a Hyperbolic Secant Index Distribution,\u201d B.S.T.J., \n59, No. 9 (November 1980), pp. 1681-91.\n\n13. G. E. Peterson, A. Carnevale, U. C. Paek, and J. W. Fleming, \u201cNumerical Calculation \nof Optimum a for a Germania-Doped Silica Lightguide,\u201d B.S.T.J., 60, No. 4 (April \n1981), pp. 455-70.\n\n14. U. C. Paek, G. E. Peterson, and A. Carnevale, \u201cDispersionless Single-Mode Light- \nguides with a Index Profiles,\u201d B.S.T.J., 60, No. 5 (June 1981), pp. 583-98.\n\n15. U. C. Paek, G. E. Peterson, and A. Carnevale, \u201cElectromagnetic Fields, Field \nConfinement, and Energy Flow in Dispersionless Single-Mode Lightguides with \nGraded-Index Profiles,\u201d B.S.T.J., 60, No. 8 (October 1981), pp. 1727-48.\n\n16. Y. Kokubun and K. Iga, \u201cMode Analysis of Graded-Index Optical Fibers Using a \nScalar Wave Equation Including Gradient-Index Terms and a Direct Numerical \nIntegration,\u201d J. Opt. Soc. Am., 70, No. 4 (April 1980), pp. 388-94.\n\n17. W. L. Mammel and L. G. Cohen, \u201cNumerical Prediction of Fiber Transmission \nCharacteristics from Arbitrary Refractive-Index Profiles,\u201d Appl. Opt., 21, No. 4 \n(February 1982), pp. 699-703.\n\n18. S.J. Jang, L. G. Cohen, W. L. Mammel, and M. A. Saifi, \u201cExperimental Verification \nof Ultra-Wide Bandwidth Spectra in Doubly-Clad Single-Mode Fiber,\u201d B.S.T.WJ., \n61, No. 3 (March 1982), pp. 385-90.\n\n19. L. G. Cohen, W. L. Mammel, and S. Lumish, \u201cTailoring the Shapes of Dispersion \nSpectra to Control Bandwidths in Single-Mode Fibers,\u201d Optics Letters, 7, No. 4 \n(April 1982), pp. 183-5.\n\n20. K. Ikamoto and T. Okoshi, \u201cVectorial Wave Analysis of Inhomogeneous Optical \nFibers Using Finite Element Method,\u201d IEEE Trans., MTT-26 (February 1978), \npp. 109-14.\n\n21. C. Yeh, K. Ha, S. B. Dong, and W. P. Brown, \u201cSingle-Mode Optical Waveguides,\u201d \nAppl. Opt., 18, No. 10 (May 1979), pp. 1490-504.\n\n22. D. S. Burnett, Finite Element Analysis, From Concept to Applications, New York: \nAddison-Wesley, to be published.\n\n23. M. Reed and B. Simon, Methods of Modern Mathematical Physics IV: Analysis of \nOperators, New York: Academic Press, 1978.\n\n24. M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions, Washington, \nD. C.: National Bureau of Standards, 1970, Chapter 9.\n\n26. T. Kato, Perturbation Theory a Linear Operators, New York: Springer-Verlag, \n1966, p. 125 and p. 391.\n\n27. J. W. Fleming, \u201cMaterial Dispersion in Lightguide Glasses,\u201d Elect. Lett., 14, No. 11 \n(May 1978), pp. 326-8.\n\n28. S. E. Miller and A. G. Chynoweth, ed., Optical Fiber Telecommunications, New \nYork: Academic Press, 1979, Chapter 7, p. 188.\n\n30. P. L. Chu and T. Whitbread, \u201cMeasurement of Stresses in Optical Fiber and \nPreform,\u201d Appl. Opt., 21, No. 23 (December 1, 1982), pp. 4241-5.\n\n31. M. J. Saunders, \u201cA Comparison of Single-Mode Refractive Index Profiles in Pre- \nforms and Fibers,\u201d 8th ECOC, paper C-17 (September 1982).\n\n32. K. I. White, \u201cDesign Parameters for Dispersion-Shifted Triangular-Profile Single- \nMode Fibers,\u201d Elect. Lett., 18, No. 17 (August 19, 1982), pp. 725-7.\n\nGradient index terms, involving the quantity g, = d/drink?, occur \nin eqs. (10) and (21). Although the index or its derivative may be \ndiscontinuous, they can be approximated by smooth functions, so k? \nand g, can be taken as well defined, continuous functions of r. This \nappendix concerns the changes needed in the T matrix to accommo- \ndate these terms.\n\nEquation (10) converts to symmetric form when h, = kf and both \nsides are divided by k. The result is\n\nin place of matrix A specified in eq. (34). The first quantity is aj as \nbefore, the second adds to the C matrix of eq. (35), and the last two \nwhen evaluated by the trapezoidal rule contribute only to the first side \ndiagonals. The term g,/r is handled in the same way as k?. Combining \nthese contributions gives a new 7 matrix that is symmetric and \ntridiagonal.\n\nThe rest proceeds in the same way to give a symmetric, tridiagonal T \nmatrix.\n\nMaterial dispersion is incorporated by expressing the index in terms \nof concentration functions as before. Radial derivatives needed for the \ngradient index terms themselves involve derivatives of the concentra- \ntion function. The w\u201d derivatives of these terms, needed for the modal \ndelays, are found by differentiating the coefficients of the concentra- \ntion functions.\n\nTerrence A. Lenahan, S.B. and S.M. (Electrical Engineering) 1964, Mas- \nsachusetts Institute of Technology; Ph.D. (Applied Mathematics) 1970, Uni- \nversity of Pennsylvania; Bell Laboratories, 1970\u2014. Mr. Lenahan has done \nstudies of various areas of mathematical physics including electromagnetic \npropagation, elasticity, and fluid mechanics. He has recently been interested \nin the propagation of light in optical fibers. Member, American Mathematical \nSociety, SIAM, Sigma Xi.\n\nMeasurements of 800-MHz Radio Transmission \nInto Buildings With Metallic Walls\n\nIn this paper we describe an experiment conducted to measure 800-MHz \nattenuation into buildings. This information is needed for refining the config- \nuration and design of portable radiotelephone systems that will accommodate \nlow-power portable sets. Signal levels have been measured in and around three \nsmall buildings and a house, using an instrumentation van with an erectable \n27-foot-high antenna. These buildings and the house all have metallic mate- \nrials in their walls and thus are expected to exhibit high attenuation. The van \nwas parked at one location with respect to the three buildings, and at nine \ndifferent locations, at distances ranging from 400 feet to 1600 feet, from the \nhouse. We found that small-scale signal envelope variations are approximately \nRayleigh distributed. For the house, large-scale distributions of the small-scale \nsignal medians are approximately log-normal. Median signal levels outside the \nhouse decrease as d~*\u00b0, where d is the distance from the van. Inside the house, \nlevels decrease as d~*\u00b0 for first-floor locations and as d~*\u00b0 for second-floor \nlocations. Average signal levels at 1000 feet, relative to free space, are \u201412.5 \ndB outside, \u201418.5 dB for the first floor, \u201416.5 dB for the second floor, and \n\u201428.9 dB for the basement. Other statistics of the signal levels and of \nattenuation into the house are also given in the paper. Cross-polarization \ncouplings of \u201410 dB to 0 dB were measured in and around the three small \nbuildings. The small-scale signal medians inside the three buildings range \nfrom 2 dB above to 24 dB below the averages of the signal medians outside \nbuildings.\n\nI. INTRODUCTION \nThe attenuation of radio signals propagating into buildings has a \nsignificant effect on the performance of portable radiotelephone sys-\n\ntems. Frequencies in the 800-MHz to 900-MHz range are good can- \ndidates for such systems. Most earlier propagation measurements at \nthese frequencies in the shadowing and multipath environment around \nbuildings were made for mobile radio systems.\u201d Ranges for the mobile \nradio data are generally greater than one mile. Battery power limita- \ntions on portable systems will likely restrict such systems to ranges \non the order of 1000 feet. Earlier measurements made in buildings are \nlimited in scope,** are oriented towards lower frequencies,\u201d or are \ndirected toward high-power portable sets with ranges greater than one \nmile.\n\nOne important portable radiotelephone environment comprises sub- \nurban residential areas characterized by discrete houses and other \nsmall buildings with densities ranging from less than one house per \nacre to a few houses per acre. Houses and small buildings with metallic \nmaterials in their walls are expected to exhibit high extreme values of \nattenuation.\n\nIn the residential environment, fixed radio terminals that commu- \nnicate with portable sets could be placed at convenient locations \noutside buildings. These fixed terminals will be referred to as Portable \nRadiotelephone Terminals or PORTs.\n\nWe have implemented an experiment to provide 800-MHz atten- \nuation information needed for refining the configuration and design \nof portable radiotelephone systems that will accommodate low-power \nportable sets. Measurements were made in and around three small \nbuildings and a house. The buildings and the house all have metallic \nmaterials in their walls. An instrumentation van was parked at differ- \nent locations to simulate different PORT with distances ranging from \n200 to 1600 feet. A 27-foot-high erectable antenna on the van simulated \nan unobtrusive PORT antenna. A portable signal source was moved \nin and around a building and signal levels were received and recorded \nin the van.\n\n*The actual frequency of the measurements is 815 MHz; however, the statistical \nresults are not sensitive to small changes in frequency. Therefore, when frequency is \nreferred to relative to the measurements, it will be rounded to 800 MHz.\n\nThe transmitting antenna is a half-wavelength coaxial sleeve dipole \nattached to the top of the hand-held unit. The de power is provided \nby a self-contained nickel cadmium battery through a voltage regula- \ntor. The regulator minimizes output power drift due to normal battery \ndischarge. The transmitter output is 0.8 watt. The output varies less \nthan 0.3 dB and 700 Hz over continuous one-hour periods that include \nambient temperature changes of 0\u00b0C to 25\u00b0C.\n\nThe instrumentation van (movable PORT) is a modified motor \nhome, containing a 5-kw ac generator. An uninterruptable power \nsupply isolates the instrumentation from generator voltage fluctua- \ntions that otherwise could affect measurement accuracy. The van is \nshown in Fig. 1.\n\nTwo 27-foot antenna masts are installed in pivoting mounts so they \ncan be stored horizontally for transport and can be erected at the test \nsite. They are 14.2 feet apart and are adjusted to vertical using built- \nin bubble levels. An 815-MHz collinear receiving antenna is mounted \non one mast. A bracket on the other mast holds the 815-MHz signal \nsource to provide a reference signal for calibrating the receiver. The \ncenters of the two antennas are at the same height when erected.\n\nThe measuring receiver is an 815-MHz frequency-modulated (FM) \ncommunications receiver, modified to detect the received signal enve- \nlope. The receiving antenna is a collinear array (four dipole elements, \n5.8-dB gain over dipole, 18-degree vertical beam width) mounted \nvertically at the top of the tilt-over mast on the instrumentation van. \nIn the receiver modification, an 11.7-MHz intermediate frequency (IF) \noutput is extracted before the limiter, down-converted to 13 kHz, \nbandpass-filtered (BW3as = 8 KHz), and linearly detected. The mod- \nified receiver has a \u2014123 dBm sensitivity for 0 dB output s/n from the \nlinear envelope detector and a 45-dB measuring range between levels \n6 dB above the noise level and 3 dB below saturation. The measuring \nrange is linear within +1 dB over a 35-dB range at high signal levels. \nThe entire receiver is enclosed in a sealed brass box to provide radio \nfrequency (RF) isolation. A variable RF attenuator reduces the input \nsignal in 1-dB steps to prevent overloading of the receiver.\n\nThe analog receiver output drives a 12-bit-resolution digital storage \noscilloscope and an integral 5-1/4-inch flexible disc drive for data \nstorage. The oscilloscope is set to record 2048 samples in a 20-second \nmeasurement period. Up to 16 tracks of 2048 samples each can be\n\nFig. 1\u2014The instrumentation van with the receiving antenna mast and the reference \nsignal mast erected. The center of the four-element collinear receiving antenna is 27 \nfeet above ground at the top of the mast mounted on the right side of the van near the \nfront. The signal source is at the top of its mast mounted on the left side near the rear. \nThe center of the signal-source dipole is also 27 feet above ground.\n\nrecorded on each disc. Data for each parked position of the van, i.e., \none PORT location, are stored on a separate disc.\n\nThe recorded data are transferred to a desktop computer. At the \ntime of transfer the following steps are performed:\n\n1. Hand-recorded log information is appended to the signal strength \ndata. The log information includes such items as the address of the \nhouse, the position of the van, the path azimuth, the path length, and \nthe receiver input attenuator setting.\n\n2. The data are scaled to convert the recorded signal voltages to dB \nrelative to 0 dB at the reference location 14.2 feet from the receiving \nantenna. The scaling takes into account recorded dc offsets, recorded \nreference levels, and the receiver input attenuator settings.\n\n3. Medians and cumulative distributions of signal level are calcu- \nlated from the scaled data.\n\n4, 'The scaled data are stored on flexible discs within the computer \nfor further analysis.\n\nVan locations were selected using tax maps covering the immediate \nvicinity of the house. Points were chosen equally spaced in azimuth \naround the house for each of three radii at about 400, 800, and 1600 \nfeet. Road layout and terrain irregularities influenced the final choices \nof vehicle placement. A single van location was used for the measure- \nments in the other three small buildings.\n\nMeasurements were coordinated between the van and the measure- \nment location over a 450-MHz voice link. The link comprises a 25- \nwatt FM transceiver in the van and a 2-watt handie-talkie carried by \nthe person making the measurements.\n\nA typical procedure for measuring a building starts with the van \nparked at an appropriate position. The van location must be fairly \nlevel. If necessary, wooden ramps are put under wheels to aid in \nleveling. The portable transmitter is installed on its mast as a local \nreference source. The mast is erected and is plumbed to vertical. \nSimilarly, the receiving antenna on the opposite side of the vehicle is \nerected and plumbed. Keeping both masts plumbed on the level van \nassures a fixed distance between the reference and receiving antennas \nfor calibration purposes. The linear detector IF input is grounded, and \nthe dc level is adjusted and recorded on a disc track. Next, the detector \ninput is ungrounded and the receiver RF attenuator is adjusted so the \nRF reference level is within the receiver operating range. This level, \nwhich serves as a calibration reference at the fixed 14.2 foot distance, \nis then recorded.\n\nThe signal source is removed from the mast and taken to the building \nfor signal level measurements. The unit is hand-held at arm\u2019s length,\n\nfor a scan height of 4.5 feet.\u00ae At a selected location within the building, \na 20-second raster scan is made by moving the transmitter in a \nhorizontal plane at 2.5 ft/s.\u00b0 The 4-foot square scanned area consists \nof 12 parallel linear scans separated by 4-inch increments. During the \nscan period 2048 data points are taken at 100 samples/s.\n\nDuring the scanning period, the oscilloscope is monitored to see \nthat signal amplitudes are within the receiver operating range. If not, \nthe RF input attenuation is adjusted and the scan is repeated. The \nremaining locations within the building and immediately outside are \nsimilarly scanned and recorded. After the measurements are made, \nthe transmitter is reinstalled on the reference mast, erected and \nplumbed, and the dc level and RF signal reference level are again \nrecorded. The closure error between beginning and ending reference- \nlevel recordings is usually <0.5 dB. If the closure is >1 dB, the \nmeasurements are repeated. Such high closure error occasionally oc- \ncurs when the transmitter battery has discharged below the regulation \nlimit of the voltage regulator.\n\nMotion of the signal source through the 4-foot-square areas inside \nand outside of buildings results in small-scale signal variations. The \nvariations are caused by multipath propagation.\u201d\u00ae Inside houses and \nin areas shadowed from the van, where propagation is dominated by \nreflection and scattering, the variations in the received signal envelope \nare approximately Rayleigh distributed (see Ref. 6 and Section IV of \nthis paper). Received signal minima are separated on the order of one \nhalf wavelength.\u00ae The medians (or means) of these small-scale varia- \ntions are approximately stationary over the small areas but the medi- \nans for areas in different rooms in a building can be significantly \ndifferent. Thus, the signal statistics can be modeled as a combination \nof a small-scale, quasi-stationary process (multipath) superimposed \non a large-scale process (shadowing). This model is like the models \nused for mobile radio propagation.\u201d\"\u201d\n\nOnly a single parameter is needed to describe the small-scale Ray- \nleigh-distributed signal variation. The median can be determined if \nsomewhat over half of the samples are above the measurement thresh- \nold. The mean, however, will be biased by the receiver noise level \nunless significantly more than half the samples are above the thresh- \nold. Therefore, medians of received signal variations for the 4-ft- \nsquare areas are used to characterize the small-scale signal variations \nat different measurement locations. The lowest median signal meas- \nured was 44 dB above the receiver noise level for the data presented \nin Sections III and IV. The lowest median cross-polarized signal \nmeasured was 38 dB above the noise level.\n\nBuilding attenuation for a location inside a building can be defined \nas the difference between the signal level at the location and a \n\u201crepresentative\u201d signal level outside the building. Both levels are taken \nto be small-scale medians, measured in decibels, relative to a common \nreference level.\n\nThere are at least two possibilities that could be used for the \nrepresentative level outside. The first choice is to use the average level \nof the signal in which the building is immersed. This level can be \napproximated by taking the average of several median levels, in \ndecibels, measured at different locations surrounding the building. \nThis decibel averaging is appropriate for large-scale variations of the \nmedian that are log-normally distributed. Large-scale variations are \nshown in Section IV to be approximately log-normally distributed. \nThis definition of representative level seems appropriate for system \nconsiderations because service would have to be provided all around \nthe exterior of a building.\n\nThe second possible choice for a representative level is to use the \nlevel of the signal incident on the building from the PORT (or received \nat the PORT from the incident region, since reciprocity holds). The \nincident level can be determined readily when the PORT is in line-of- \nsight of the building and there is little multipath from surrounding \nbuildings. This was the case for the measurements in Ref. 6. However, \nwhen there are many intervening buildings between the PORT and \nthe subject building and there is considerable multipath from sur- \nrounding buildings, the signal level incident on the building is not \neasily defined. This second choice of representative level appears least \nappropriate at the longest distances where the shadowing and multi- \npath are most significant. These are also the distances of most concern \nin system considerations.\n\nIn Section III, comparisons are made of attenuations determined \nusing both choices for representative levels. However, the more appro- \npriate average outside level is used for the statistics in Section IV.\n\nitl. ATTENUATIONS AND CROSS-POLARIZATION COUPLINGS FOR \nTHREE BUILDINGS\n\nThe first building, shown in Fig. 2, is 20 feet by 38 feet, of corrugated \nsteel construction and mounted on a concrete platform. There are \nmetal screened doors inside the main metal doors, and all of the nine \nwindows are metal screened, with the exception of three in the rear. \nThe building inside is a single large area and is about one-third filled \nwith equipment.\n\nThe second building is shown in Fig. 3. It is 21.5 feet by 28 feet, is \ncovered with aluminum siding and has one solid door and nine win-\n\nFig. 2\u2014Corrugated steel building on Crawford Hill as seen from the instrumentation \nvan position.\n\nFig. 3 SAntehin Control building on Crawford Hill. The photograph was taken \nlooking toward the building along the path from the van.\n\ndows. All except the three largest windows are metal screened. The \nbuilding interior is divided roughly in half on the long dimension and \nis about half filled with equipment.\n\nFig. 4\u2014Wooden building on Crawford Hill. The propagation path from the van was \ninto the corner to the left in the photograph.\n\nand contains vapor barrier foil insulation in the walls. This building, \nshown in Fig. 4, was empty. In front it has two doors with unscreened \nwindows and three other unscreened windows. The room with double \ndoors to the right in Fig. 4 is attached to an exterior wall. Measure- \nments were not made in the attached room.\n\nSignal levels were measured at four locations outside and two or \nthree locations inside of the three buildings from a single van (PORT) \nlocation. These levels are summarized in Table I. The distances \nindicated in the table (i.e., 225 feet, 330 feet, and 845 feet) are the \npath length from the van antenna to the fronts of the buildings. The \ncenters of the four outside locations for each building were about 5 to \n10 feet away from the midpoints of the four sides of the building. The \noutside locations are numbered clockwise looking down on the building \nfrom above, starting with the side closest to the van. The small-scale \nmedian signal levels in decibels and the decibel average for the four \noutside locations are tabulated in the column labeled \u201cLevel/dB\u201d. \nThese levels are relative to 0 dB at the signal level reference 14.2 feet \nfrom the van receiving antenna. The column labeled 1/r?/dB contains \nthe signal levels that would occur in free space (1/r?) at the distance \nof the measurement location. These levels are also relative to 0 dB at \nthe 14.2 foot reference location. The 1/r? level in the row labeled Avg. \nis the free space value at the building midpoint.\n\nFor the wooden building positioned at 330 feet from the van and for \nthe building with metal siding positioned at 845 feet, the signal \nmedians measured in front of the buildings are 1 to 2 dB greater than\n\nthe free-space values. The signal median in front of the all-metal \nbuilding positioned at 225 feet is about 1 dB below the free-space \nvalue. These values are all consistent with the effects of a single \nreflection from the relatively flat dry ground between the van and the \nbuildings.\u2019\u00ae The 27-foot van antenna height and a reflection coefficient \nphase angle of nearly 180 degrees, appropriate for small angles between \nthe incident wave and smooth dry ground, yield signal maxima at \ndistances above the ground of about 3 feet for 330 feet from the van \nand of about 8 feet for 845 feet from the van. For the same conditions, \na minimum occurs about 4 feet above the ground at 225 feet from the \nvan. Recall, the signal source is scanned about 4.5 feet above the \nground. Therefore, at 330 feet, the scan is near a broad maximum that \ncould be as much as 6 dB above free space if the ground were perfectly \nreflecting. The measured level at 330 feet is 2 dB above free space. At \n845 feet, the scan should be down on the side of an 8-foot-high \nmaximum. The measured level at 845 feet is 1.8 dB above free space. \nAt 225 feet, the scan is near the first minimum above the ground. The \nmeasured level at 225 feet is 1.2 dB below free space.\n\nThe interior locations of the buildings are arbitrarily numbered in \nthe column labeled \u201cLocation Inside\u201d. The column labeled \u201cRelative \nto Average Attenuation in decibels\u201d contains the differences between \nthe median signal levels measured at the locations and the average of \nthe four outside median levels for the building. Positive attenuation \nindicates the signal level inside is smaller than the average level \noutside.\n\nwere made at Location 1 inside the metal-sided building. The spread \nof the resulting three medians was only +0.5 dB around the average \nvalue listed in the table. Two separate measurements were made at \nLocation 2 outside the wooden building. The resulting two medians \nwere within +0.3 dB of the average value listed. Thus, the measure- \nment repeatability is good and is consistent with the +0.5 dB standard \ndeviation of the medians expected for the statistical fluctuation re- \nsulting from the limited number (~150) of independent samples in a \n4-foot-square area. Measurements were made at the same locations in \nand around the metal building on three different days that had \ndifferent ground moisture conditions, etc. The three medians for each \nlocation were averaged and the difference was taken between the \nlocation average and the individual medians. The standard deviation \nof the differences was 1 dB.\n\nThe column labeled \u201cRelative to Incident Attenuation in decibels\u201d \ncontains the differences between the median levels at the inside \nlocations and the median level for the outside location that is closest \nto the van. These outside closest locations have the largest signal \nlevels measured for their corresponding buildings. The outside level is \ncorrected for free space (1/r\u201d) for the distance between the outside \nlocation and the inside location being considered. This is the second \ndefinition of building attenuation described in Section 2.3. This second \ndefinition of attenuation is essentially the same definition of building \nattenuation that was used in Ref. 6 for the same all-metal building \nlisted first in Table I (the values in Ref. 6 were not corrected for 1/ \nr\u201d), The second definition attenuation values are within 1 or 2 dB of \nthe values in Ref. 6 for Locations 1 and 3 inside the metal building. \nAt Location 2, however, the attenuation in Table I is 6 dB less than \nthe earlier value. Items inside the building have been rearranged since | \nthe earlier measurements, but no reason for such a large change is \nevident. The second definition attenuation into the all-metal building \nis greater than attenuation from the front (van side) to the back of \nthat building (10.0 dB front-to-back attenuation including 1/r? cor- \nrection). The second definition attenuation into the other buildings is \nless than the front-to-back attenuation (21.7 dB for the metal-sided \nbulding and 15.0 dB for the wooden one). This suggests that the \ndominant mechanism for signal propagation behind the metal building \nis scatter and/or reflection from objects behind and to the side of that \nbuilding rather than passage through the building itself. The dominant \nmechanism for propagation behind the other two buildings is not \nevident from these simple measurements since either passage through \nmultiple walls or scatter and/or reflection is consistent with the result.\n\nFor Location 1 in the wooden building, the first definition attenua- \ntion is negative. This indicates that the median level inside at that\n\nlocation is greater than the average of the four outside medians. This \nis a reasonable situation because the signal levels at one side and at \nthe back of that building are much lower than the signal level inside \nLocation 1. This inside location has unscreened windows between it \nand the outside in the direction of the van. As mentioned in Section \nII, this first definition of attenuation seems more appropriate for use \nin system analysis since a system would have to serve the outside \nlocations at all sides of a building. Of course, the distribution of outside \nlevels relative to the outside average is also needed for a complete \nassessment of system performance. ,\n\nThe building attenuations by either definition are greatest for the \nall-metal building with metal screened windows. Since both of the \nother buildings have metal in their walls, attenuation into them \nprobably depends strongly on coupling through the nonscreened win- \ndows.\n\nCross-polarization coupling in multipath propagation is significant \nin reducing the effects of the random orientation of portable radiote- \nlephones.\u2019* Cross-polarization coupling is defined as 20 log(E,/F;), \nwhere F, is the average or median field magnitude of the polarization \naligned with the transmitted polarization and E, is the average (or \nmedian) field magnitude of the polarization orthogonal (crossed) to \nE,. An indication of the cross-polarization coupling can be obtained \nby orienting the signal-source dipole antenna horizontally and scan- \nning a measurement location with the dipole pointed towards the van \n(end-on orientation). The scan can be repeated with the dipole per- \npendicular to the direction of the van (broadside orientation).\n\nIf the multipath propagation were uniformly distributed in azimuth \naround the measurement location and were confined to a horizontal \nplane, horizontally polarized multipath would produce a median re- \nceived signal level in a scanned horizontal dipole that was 3 dB less \nthan the median that would be produced by the same multipath in a \nscanned loop oriented with its plane horizontal. The 3-dB decrease \nresults from the nonuniform directivity pattern of the dipole in any \nplane containing the dipole. The median level received by the loop in \nthe horizontally polarized multipath would be the same as the median \nlevel received by a scanned vertical dipole in vertically polarized \nmultipath of the same average intensity. The multipath signal varia- \ntions in all cases would be Rayleigh distributed.\n\nFor the measurement situation, equal median signal levels for end- \non and broadside horizontal scans would be consistent with multipath \nhaving a uniform azimuthal distribution. Then, under the assumption \nthat the propagation directions are confined to a horizontal plane, the\n\ncross-polarization coupling would be 3 dB greater than the signal \ndifference A = L, \u2014 Ly, where L, is the median level of the signal from \na scan with the source oriented vertically, and L;, is the median level \nof the signal from a scan with the source oriented horizontally. The \nmedians have a statistical fluctuation with a standard deviation on \nthe order of +0.5 dB because of the limited number of independent \u00b0 \nsamples. Therefore, for differences between the medians of the two \nhorizontal scans of one or two decibels, they can be taken as equal \nand their average value can be used to determine A.\n\nTable II summarizes the cross-polarization measurements made in \nand around the three buildings. The medians of the end-on and \nbroadside scans are tabulated in columns labeled End-on and Broad- \nside. The values are in decibels relative to the median of the signal \nscan at the same location with the source antenna oriented vertically. \nThat is, columns End-on and Broadside indicate A for end-on and \nbroadside scans. The column labeled cross-polarization is 3 dB greater \nthan the average A for end-on and broadside scans.\n\nThe locations inside the metal building would yield a small positive \nvalue for cross-polarization coupling. This probably indicates a break- \ndown of the assumption that the multipath propagation is confined to \na horizontal plane, so the coupling is taken as 0 dB. If the multipath \npropagation were uniformly distributed in all directions in three di- \nmensions, the orientation of the antenna would be irrelevant. Since \nthe data show only a small bias towards stronger median signal for \nthe vertical antenna in the metal building, an alternative assumption \nfor that building would seem to be uniformly distributed multipath \npropagation in all directions in three dimensions.\n\nThe cross-polarization values in Table II are all greater than \u201410 \ndB and most are greater than \u20146 dB. Another point worth noting is \nthat the locations with the lowest signal levels (greatest attenuation) \nalso have the largest cross-polarization coupling with values greater \nthan \u20146 dB and usually greater than \u20143 dB. Since cross-polarization\n\ncoupling increases the effectiveness of diversity in mitigating the \neffects of random portable set orientation and multipath propagation,'* \nthis trend could be significant in determining system performance.\n\nThe house is a two-story colonial located on a level, one acre lot. It \nis in a newly developed area with a density of one house per acre and \nwith relatively few trees. The house has an area of 2400 square feet, \nconsisting of living room, dining room, kitchen, and den on the first \nfloor, and four bedrooms on the second floor. It also has a basement \nand a two-car attached garage. It is constructed with wood, with \naluminum siding on three sides and nonmetallic siding on the front. \nAll exterior walls contain insulation faced with a metal foil vapor \nbarrier.\n\nThe cumulative distributions of the envelope variations of multipath \npropagation for a scan of a 4-foot-square location are expected to be \nRayleigh.\u201d* Figure 5 shows the distributions measured at four locations \nin and around the house. A Rayleigh distribution is a straight line \nwith the particular slope indicated on the figure. The distributions are \ngood approximations to the Rayleigh distribution for many locations \ninside the house and behind the house from the van. A few locations \non the same side of the house as the van and only 400 feet from it \nexperience essentially line-of-sight propagation. At these few locations \nthe signal variation is small and the envelope distribution significantly \ndeparts from Rayleigh, as indicated by the example plotted as squares \non the figure. The medians of the four distributions on the figure have \nbeen normalized to 0 dB. The two distributions that show the greatest \ndeparture from Rayleigh were selected as the extremes that have the \ngreatest and least spread in attenuation of any of the distributions in \nthe data set.\n\nThe open data points in Fig. 6 are the medians of the measured \nsignal envelopes for the small-scale locations outside the house. The \nmedians are plotted versus the distance between the van antenna and \nthe location. The median levels are in decibels relative to the signal \nlevel at the van reference. The outside locations are in front of the \nmidpoints of the four outside walls of the house. The decibel averages \nof the four locations for each van position are indicated by the solid \ndata points.\n\nFig. 5\u2014Measured signal envelope distributions from four small (4-ft square) areas. \nOn these coordinates, Rayleigh distributions are straight lines with the slope indicated. \nDistances between the van and the measurement location are indicated in the key.\n\nThe solid line is the linear-least-squares regression fit to the 36 \nmedians from the four locations for each of the nine van positions. \nThe least-squares regression fit to the averages of the four locations \nfor each van position yields the same solid line. Signal level varies \nwith distance as d~*\u00b0 for the solid line. The dotted line represents \nfree-space propagation (d~\u201d) relative to 0 dB at the van reference. At \n1000 feet from the van, the signal level represented by the solid line \nis 12.5 dB lower than the level would be in free space, i.e., the average \nexcess attenuation over free space is 12.5 dB at 1000 feet. The distance \ndependence exponent is probably reliable only to several tenths be- \ncause of the large spread in the data and the relatively small number \nof data points. The signal level outside was correlated to distance with \na coefficient of 0.8.\n\nThe lower dashed line is the linear regression line for the data near \n400 feet and 800 feet only. The upper dashed line is for the data near \n800 feet and 1600 feet. The distance dependence for the 400-foot to\n\nFig. 6\u2014Medians of the small-scale envelope variations for different outside measure- \nment locations plotted versus distances between the locations and the van antenna. \nOpen data points are medians for individual outside locations. Solid points are averages \nof the medians for the four measurement locations for each of the van locations. Points \nrepresented by the same symbols at nearly the same distance designate the four \nmeasurements associated with a particular van position. Signal levels are with respect \nto 0 dB at the signal reference located 14.2 feet from the van receiving antenna. The \ndotted line represents free-space propagation (1/r*) relative to 0 dB at the signal \nreference. The solid line and the dashed lines are linear-least-squares regression lines, \nas discussed in the text.\n\n800-foot distances is d~*'; the dependence for the 800-foot to 1600- \nfoot distances is d~*. Thus, the signal decreases faster with distance \nin the first several hundred feet than it does farther from the house. \nThis is reasonable because, for the first few hundred feet, there are \nfew obstructions along the path and propagation approaches line-of- \nsight. After several hundred feet, the number of intervening houses \nincreases. Reflection from the ground also causes the signal to decrease \nmore rapidly with distance than it does in free space.\n\nThe data points at 400 feet that are greater than free space values \ndeserve comment. These points are from locations that are not shad- \nowed from the van, i.e., they are within line-of-sight. A single ground \nreflection from a 27-foot-high antenna 400 feet away would produce a \nsignal maximum about 4 feet above the ground. The signal maximum \nwould be greater than the single-path free-space value. (It would be 6 \ndB greater for a reflection coefficient of unity.) The signal levels a few \ndB above free space are not unexpected 4.5 feet above the ground in\n\nlocations that have large reflecting surfaces (walls of houses) nearby \nin addition to the ground.\n\nFigure 7 is the cumulative distribution of the data from Fig. 6 after \nthe distance dependence, d~**, is removed. The calculated standard \ndeviation for the data points is 9.0 dB. A straight line on the coordi- \nnates in Fig. 7 represents a log-normal distribution of signal level. The \nmeasured distribution is a reasonable approximation to a log-normal \ndistribution.\n\nThe distance dependences for the linear-least-square fits to the \nmedians for measurement locations inside the house were somewhat \ndifferent for the different floors of the house. The variation with \ndistance was d~*\u00b0 for the first floor, d~*\u00b0 for the second floor, and d~*\u201d \nfor the basement. Since only one location was measured in the base- \nment, this value for the basement may contain considerable statistical\n\naoe 5 10 20 30 40 50 60 70 80 90 95 98 99 \nPERCENTAGE OF LOCATIONS WITH NORMALIZED LEVEL = ORDINATE \nFig. 7\u2014Cumulative distribution of the medians of the small-scale envelope variations \nfor outside locations after the values for the solid regression line in Fig. 6 are subtracted\n\nout. The solid line and the dashed line in this figure represent log-normal distributi \nwith standard deviations of 9 dB and 10 dB, respectively. Sere edna ve\n\nerror. The variation with distance for the composite set of data from \nthe entire house (eight measurement locations inside the house and \nnine van positions) was d~*\u00b0. The less rapid decrease in signal level \nwith distance for the second floor compared with the first floor is \nreasonable because the height of the second floor results in less \nattenuation from intervening houses. Since the decrease in signal level \nwith distance is less rapid for the locations inside the house than \noutside, the apparent building attenuation will decrease somewhat \nwith distance.\n\nAt 1000 feet, the excess attenuation over the free space values are \n18.5 dB for the first floor, 16.5 dB for the second floor, 29.9 dB for the \nbasement, and 19 dB for the composite data set for the entire house. \nThe average building attenuation at 1000 feet, which is the difference \nbetween the linear regression values outside and inside, is then 6.0 dB \nfor the first floor, 4.0 dB for the second floor, 16.4 dB for the basement, \nand 6.5 dB for the entire house.\n\nThe building attenuation considered in this section uses the first \ndefinition in Section 2.3, i.e., attenuation is with respect to the average \nof the four outside median levels. Building attenuations for the house \nare plotted versus distance in Fig. 8. The open data points represent \ndata from the first floor; the solid data points and the points marked \nB represent data from the second floor and the basement, respectively. \nAs noted in the previous section, the attenuation is weakly dependent \non distance. Also, the attenuation into the basement is significantly \ngreater than the attenuation into the other two floors. The linear \nregression lines are labeled on the figure for several combinations of \nthe data. The variation of attenuation with distance is d~\u00b0\u00ae for the \nfirst floor, d~}\u00b0 for the second floor, d~'* for the basement, and d7}\u00b0 \nfor the first and second floor together and for the entire house. The \nattenuation at 1000 feet taken from the regression lines is 6.0 dB for \nthe first floor, 4.0 dB for the second floor, and 16.4 dB for the basement. \nThese 1000-ft attenuation values are the same as those obtained in \nthe previous section from the separate linear regression lines to the \nsignal levels inside and outside.\n\nMeans (m) and standard deviations (c) for several different group- \nings of the attenuation data are tabulated in Table III. It appears that \nmost of the difference in the distance dependence for the first and \nsecond floors results from the different attenuation averages at 1600 \nfeet.\n\nThe different distance dependences and different average attenua- \ntion for the different floors present a dilemma if they are real effects \nnot confined to this data set. For radio system analysis, a single\n\nFig. 8\u2014Building attenuation by the first definition, i.e., relative to the average of the \nmedian levels for the four outside locations, plotted versus distance. Open data points \nand X points are from the first floor, solid data points are from the second floor, and \npoints marked B are from the basement. All points marked with a particular symbol \nare from the same measurement location in the house but are for different van positions. \nThe lines are linear-least-squares regression lines through different groupings of points.\n\nFig. 9\u2014Cumulative distribution of the building attenuations from Fig. 8 for the first \nand second floors after the values for the first- and second-floor regression line (solid \nline in Fig. 8) are subtracted out. The straight line represents a log-normal distribution \nwith a standard deviation of 4.4 dB.\n\ndistance dependence and average attenuation are desirable; however, \nif only the composite regression line for the whole house were removed \nfrom the data, the variation remaining in the residual attenuation data \nwould have a significantly larger standard deviation (o = 6.2 dB) than \nif the individual regression lines for each floor were removed separately \n(o = 4.1 dB). Also, the composite line for the entire house depends on \nthe number of measurements made on each floor. For systems analysis, \nthese numbers of measurements should be related to the probability \nof calls being made on those floors (or at those locations); but these \nprobabilities are unknown.\n\nThe separation in decibels between the regression lines on Fig. 8 for \nthe first and second floors is not extreme over the 4 to 1 distance\n\ncovered. Also, the attenuation averages for these floors for the three \ndistances are not grossly different (all within +2 dB). Therefore, it \nseems reasonable to simplify the description of the attenuation by \ncombining the first and second floor data. The composite regression \nline for this combination is also shown on the figure. When this \ncomposite regression line is removed from the first and second floor \ndata, the resulting cumulative distribution of attenuation is as shown \nin Fig. 9. The standard deviation of the attenuation data plotted in \nFig. 9 is 4.4 dB, not significantly different from the 4.1-dB standard \ndeviation for all the data with the regression line for each floor removed \nseparately. The distribution in Fig. 9 is a reasonable fit to a log-normal \ndistribution, especially considering the limited size of the data set. \nCombining the basement data with the data from the other floors is \nnot reasonable because of the large difference in average attenuations. \nSince this difference can be expected to exist for all houses, the \nattenuation into basements will probably have to be described sepa- \nrately from the attenuation for the rooms on other floors.\n\nSignal levels were measured in and around three small buildings \nfrom a single position of an instrumentation van. Measurements were \nalso made in and around a house from nine different van positions \nranging from 400 feet to 1600 feet away. The received signal envelope \nis approximately Rayleigh distributed over most small-scale areas \nabout 4 feet square. For the three buildings, building attenuation of \nthe small-scale signal medians ranged from \u20142 to 24 dB relative to \nthe average signal level outside each building.\n\nFor the house, some of the parameters of the medians of the small- \nscale envelope variations and of the building attenuation defined in \nthis paper are summarized in Table IV. After the distance dependence \nis removed from the median signal levels outside of the house, the \ndistribution of the residual large-scale signal-level variations is ap-\n\nthe house \nFirst \n& \nOut- First Second Base- Second \nParameter side Floor Floor ment Floors Units \nSignal-level distance dependence \u20144.5 \u20143.9 \u20143.0 \u20143.2 \u2014 \u2014 \nexponent \nSignal-level relative to free \u2014-125 -185 -\u2014-165  -\u201428.9 \u2014 dB \nspace at 1000 ft. \nBuilding attenuation exponent _ \u20140.6 \u20141.5 \u2014-14 -1.0 \u2014 \nAverage building attenuation at \u2014 +6.0 +40 +164 +5.1 \u00a3=dB \n1000 ft.\n\nproximately log-normal with a standard deviation of 9 dB. The stand- \nard deviation of the building attenuation for the first and second floors \nis 4.4 dB after distance dependence is removed. Because of the large \naverage attenuation into the basement of the house, it does not appear \nreasonable to combine the basement attenuation statistics with the \nstatistics for the rooms above ground level.\n\nCross-polarization coupling is strong inside and outside of the three \nbuildings. For all locations measured, the coupling is greater than \u201410 \ndB and for most locations it is \u20146 dB or greater.\n\nInitial design of the van masts and mounts was done by W. I. \nTohlman and H. H. Hoffman. We wish to thank H. W. Arnold for \npermitting the measurements reported in this paper to be made in his \nhouse. The continued support of L. J. Greenstein and D. O. Reudink \nis greatly appreciated.\n\n1. D. C. Cox, \u201cCo-Channel Interference Considerations in Frequency-Reuse Small \nCoverage-Area Radio Systems,\u201d IEEE Trans. Commun., COM-30 (January 1982), \npp. 135-42.\n\n. J. M. Durante, \u201cBuilding Penetration Loss at 900 MHz,\u201d Conference Proceedings, \nIEEE VTG Conference, 1973, p. 1-7.\n\n. P. I. Wells and P. V. Tryon, \u201cThe Attenuation of UHF Radio Signals by Houses,\u201d \nU. S. Dept. of Commerce Report, OT Report 76-98, August 1976; and IEEE \nTrans. Veh. Technol., VT-26 (November 1977), pp. 358-62.\n\n. J. Shefer, \u201cPropagation Statistics of 900 MHz and 450 MHz Signals Inside Build- \nings,\u201d Microwave Mobile Radio Symposium, March 7-9, 1973, Boulder, Colorado.\n\n. H. H. Hoffman and D. C. Cox, \u201cAttenuation of 900 MHz Radio Waves Propagating \nInto a Metal Building,\u201d IEEE Trans. Ant. Propag., AP-30 (July 1982), pp. 808-11.\n\n. D. Mitchell and K. G. Van Wynen, \u201cA 150 MC Personal Radio Signaling System,\u201d \nB.S.T.J., 40, No. 5 (September 1961), pp. 1239-57.\n\n. G. V. Waldo, \u201cReport on the Analysis of Measurements and Observations of New \nYork City UHF-TV Project,\u201d IEEE Trans. Broadc., 9 (1963), pp. 7-36.\n\n10. K. Tsujimura and M. Kuwabara, \u201cCordless Telephone System and its Propagation \nCharacteristics,\u201d JEEE Trans. Veh. Technol., VT-26 (November 1977), pp. \n367-71.\n\n11. M. Komura, T. Hogihira, and M. Ogasawara, \u201cNew Radio Paging System and Its \nPropagation Characteristics,\u201d IEEE Trans. Veh. Technol., VT-26 (November \n1977), pp. 362-6.\n\n12. D. C. Cox, \u201cMultipath Delay Spread and Path Loss Correlation for 910 MHz Urban \nMobile Radio Propagation,\u201d IEEE Trans. Veh. Technol., VT-26 (November \n1977), pp. 340-4.\n\n13. P. A. Matthews, Radio Wave Propagation, VHF and Above, London: Chapman and \nHall, Ltd., 1965, Chapter 2.\n\n14. D. C. Cox, Antenna Diversity Performance in Mitigating the Effects of Portable \nRadiotelephone Orientation and Multipath Propagation, IEEE Trans. Commun., \nCOM-31 (May 1983), pp. 620-8.\n\nDonald C. Cox, B.S., M.S., University of Nebraska, Lincoln, 1959 and 1960, \nrespectively; Ph.D., Stanford University, 1968, all in Electrical Engineering;\n\nHonorary Dr. of Science, University of Nebraska, Lincoln, 1983; Stanford \nUniversity, 1963-1968; Bell Laboratories, 1968\u2014. From 1960 to 1963, Mr. Cox \nworked on microwave communications system design at Wright-Patterson \nAFB, Ohio. From 1963 to 1968 he was at Stanford University doing tunnel \ndiode amplifier design and research on microwave propagation in the tropo- \nsphere. From 1968 to 1973 he was a Member of Technical Staff at Bell \nLaboratories, Holmdel, New Jersey, doing research in mobile radio propaga- \ntion and on high-capacity mobile radio systems. He is now Supervisor of a \ngroup doing propagation and systems research for portable-radio telephony \nand for millimeter-wave satellite communications. Fellow, IEEE; member, \nCommissions B, C, and F of USNC/URSI, Sigma Xi, Sigma Tau, Eta Kappa \nNu and Pi Mu Epsilon; Registered Professional Engineer, Ohio and Nebraska.\n\nRoy R. Murray, B.S. (Electronic Engineering), 1975, Monmouth College; \nBell Laboratories, 1965\u2014. Mr. Murray has worked on high-speed multi-level \ndigital modulators and, more recently, on UHF radio propagation into build- \nings. Currently, he is a member of the Telecommunication Systems Research \nDepartment. Member, Eta Kappa Nu.\n\nPenetration of Radio Signals Into Buildings in the \nCellular Radio Environment\n\nPenetration of radio signals from Advanced Mobile Phone Service cell site \ntransmitters has been measured in fourteen office and industrial buildings in \nthe Chicago area. Signal levels on the first floor of buildings averaged 14 dB \nless than reference levels in the adjacent streets. This penetration loss was \nfound to decrease with increasing height. Standard deviations of penetration \nloss ranging from 5 to 11 dB attest to the diversity of architecture and floor \narrangements. Other relationships useful in planning for portable phone \nterminal applications are derived from the data and are presented in this \npaper. The data include over 4000 measurements, each being the average of \n1024 samples of the local field taken over an eight-second period as the \ninstrumentation traveled about 20 feet over the measurement path. The \nmeasurement path in each building was traversed for each of the several cell \nsites that transmitted signals of adequate strength into the buildings.\n\nThe use of portable and hand-held terminals in 850-MHz cellular \nradio systems involves a radio field environment inside buildings that \ndiffers from the more familiar highway and street propagation envi- \nronment of the mobile terminal. The laws that define propagation \nover open and urban terrain are complicated by a penetration loss in \nthe transmission of the signal into the building\u2019s interior. The char- \nacterization of this loss, as encountered in large office and commercial\n\nbuildings, was the purpose of an extensive series of measurements \nmade in fourteen buildings served by the Advanced Mobile Phone \nService (AMPS) Developmental Cellular System in Chicago. These \nmeasurements were then compared with the penetration loss in a \nsingle-family residence.\n\nThe measurements were made in January 1980 to serve as a basis \nfor evolving system requirements related to the use of portable phone \nterminals in AMPS systems.\n\nIndoor portable service in cellular systems offers a possible improve- \nment over conventional systems. The multiplicity of serving cell sites \nshould in many cases illuminate all sides of those buildings that enjoy \na relatively exposed or unshadowed environment. It was expected that \nthis effect would be somewhat less pronounced in buildings that are \nmore sheltered in the urban core; the data support this conjecture. \nThe data also support expectations that areas with windows would \nhave lower penetration losses than areas without windows; the differ- \nence is about 6 dB. First-floor penetration losses averaged 14.2 dB. \nThe loss decreased with height at 1.9 dB per floor, very close to the \n2.0- and 2.5-dB rates reported in Refs. 1 and 2. The penetration loss \nmeasured in the aluminum-sided ranch house, as a matter of interest, \nis 7.3 dB, a value close to the findings of Cox et al. in Ref. 3.\n\nMost of the measurements made in the Cellular Test Bed (Ref. 4) \nand reported in Ref. 5 were of the signal as received by a mobile \ntransceiver. The measurements reported here include the entire dis- \ntribution of setup channel signal levels throughout the measurement \nareas for each cell site transmitter that qualified as a server. As such, \nthe data are perfectly applicable to conventional mobile portable and \npaging applications. The effect is that of taking measurements with \nthe base station in several different locations.\n\nThe Chicago AMPS Developmental Cellular System, as described \nin Ref. 4, uses ten separate setup channels for paging and mobile call \norigination in the Chicago service area. The channel signals are \nradiated continuously from omnidirectional antennas. The instrumen- \ntation used for the measurements reported in this paper can be tuned \nmanually to any of these frequencies to measure one channel at a \ntime. Figure 1 shows nominal coverage contours for the setup channels \nof the seven cell sites that served the selected buildings. The related \nvoice-channel coverage is shown in Fig. 2 of the paper by D. L. Huff.* \nThe center of each contour represents a cell site location. In general, \nthe cell sites are spaced about 15 miles apart.\n\nA group of buildings in the Chicago area was selected to represent \na range of physical characteristics including location, architecture,\n\n\u00a9 COMMERCIAL, INDUSTRIAL, NOMINAL SETUP \nOR OFFICE BUILDING CHANNEL COVERAGE\n\nFig. 1\u2014AMPS Developmental Cellular System locations of 15 buildings and seven \nserving cell sites.\n\nand local environment. Signal strength measurements were made \nalong planned routes on selected floors of these buildings. A Propa- \ngation Measuring Set (PMS) was used to measure the strength of \nsignals transmitted from cell sites.\n\nBuilding penetration loss for a given floor area is defined to be the \ndifference between the average of these measurements and an average \nof measurements made on the outside at street level.\n\nOutside signal strength was measured at street level around the \nperimeter of the building, along the closest available path to the \nbuilding\u2019s outside walls. These paths included driveways, streets, or \nparking lots, as required to achieve proximity to the building under \ntest.\n\nFig. 2\u2014Differences in level between the strongest and weakest serving channels on \nthe same floor.\n\nThe purpose of the measurement program, as stated previously, was \nto determine the extent to which signals available from AMPS cell \nsites penetrate buildings. A first-order determination was desired; \neffects of 5 dB or more on the loss measurements were considered to \nbe significant. A detailed investigation of the structure of interior \nfields in small buildings is reported by Cox in Ref. 3.\n\nThe measurements were subject to the effects of the following \ndifferent conditions:\n\n1. Different types of outside wall construction, such as steel-framed \nglass, brick, block masonry, etc.\n\n2. Urban vs. suburban areas to identify difference between buildings \nin exposed locations and buildings sheltered in the dense urban core.\n\n6. Different types of window treatment currently used to reflect \nsunlight and heat. =\n\nTime was an important factor in selecting the coverage objective for \neach building. Test runs were chosen to require about three hours of \noperations in any given building. The total time of recorded data is a \nsmall percentage of that..\n\nSix significant categories of interior areas were identified. These \nwere defined as open, enclosed, and hallways\u2014each with and without \nwindows. Typical of the areas that were considered to be \u201copen\u201d were \ncafeterias, lobbies, and large office spaces or conference rooms. Typical \nof \u201cenclosed\u201d areas were small, walled areas with space for only a few \noccupants, such as small office areas or workrooms.\n\nthought to represent a worst case. Higher floors, thought to represent \nbetter penetration, were included with less regularity. Test routes were \ndesigned to distribute measurements over the floor areas in a uniform \nmanner.\n\nThe buildings chosen in the Chicago AMPS service area were those \nwhere multiple cell sites could provide outside building signal levels of \nat least \u201490 dBm.\n\nIn the downtown section of Chicago, tests were made in three \nbuildings. These represent the urban group. The downtown buildings \nin which tests were made can be characterized as \u201csingle tenant \nbuildings\u201d.\n\nIn the suburban area around Chicago, tests were made in eleven \ncommercial buildings and one residential dwelling. Only the data from \ncommercial buildings were included in subsequent analyses.\n\nFrom the standpoint of architectural variety the program provided \nmeasurements in a warehouse, a manufacturing plant, open-space \noffices, cafeterias, lobbies, small offices, and hallways. Outside wall \nconstruction varied from unwindowed concrete slab to steel-framed \nwindows occupying about 90 percent of the outside surface.\n\nThe locations of the buildings where tests were conducted are shown \non the map of Fig. 1. The circles show the idealized boundaries of \nregions of cell site contro] channel coverage. Table I shows the distri- \nbution of floor levels in buildings where tests were conducted and the \ncell sites whose transmissions served the various floor levels. A more \ndetailed description of each of the buildings is given in the appendix.\n\nSeveral factors were considered when selecting floors and floor areas \nfor measurement. From a test sampling standpoint the first and second \nfloors were chosen when available since building penetration was \nexpected to be poorest at the lower levels. Higher floor levels were \nincluded in the tests to provide data on the sensitivity of building \npenetration to height. On any particular floor, areas were chosen that \nrepresent the different types of interiors; test paths were chosen to be \nuniformly distributed over the floor. In two instances measurements \nwere attempted in basement areas but signal strength levels were \nbelow the threshold of the PMS.\n\nA portable Propagation Measurement Set (PMS) designed for field \nstrength measurements was adapted for this test program by mounting \nit in a small \u201ctea\u201d cart equipped with a battery, battery charger, and a \nmobile antenna mounted on a 2-foot by 2-foot ground plane. The\n\nio aAmNye CW NF Pe WN RO PDH FH CONK NR OOWH OW WHF ANH \nPAPA PP PPPS PAM POPS OS\n\n* CVL-Cloverdale; EOL-Eola; LMT-Lemont; LNS-Lyons; CNL-Canal; MGV-Mor- \nton Grove; BEV-Beverly. \nground plane was positioned 4-1/2 feet above floor level, at approxi- \nmately shoulder height.\n\nThe PMS has an internal calibration system that calibrates the \nreceiver over an 80-dB range in 1-dB increments. In the measurement \nmode, the basic unit of measurement is a one-second power average \nderived from a log-amplifier output sampled at a 128-Hz rate. The \ndata output of the set can be adjusted to provide one average value, \nalso derived as a power average, for a group of 1 to 128 of these one- \nsecond averages, selectable in binary increments. The measurement \nrange of the set is \u201440 dBm to approximately \u2014122 dBm.\n\nThe PMS uses a paper tape printer for hard copy output. The output \ndata include frequency, a sequential \u201cmajor marker,\u201d the power aver- \nage, and the one-second sample variance for the group of one-second \nsamples contributing to each longer-term mean. In this application, a \nmean value was generated each eight seconds. The \u201cmajor marker\u201d is \nused to key measurements to the time of day and to logged operator \nobservations.\n\nPrior to a series of measurements, the PMS is calibrated and the \ncalibration results are printed on paper tape. A received signal strength \nmeter in the PMS provides a visual indication of the receiver output \nwithout operating the paper tape recorder. The measurement team \nuses it to verify test set operation and to choose the group of control \nchannels for measurement. The set also contains start, stop, and pause \ncontrols to limit data output to areas of interest.\n\nTo increase the confidence level of outside measurements, two \nmethods were used where possible to determine outside signal strength. \nOne method used the Mobile Telephone Laboratory (MTL) vehicle, \nwhich has the capability of power averaging instantaneous signal \nstrength samples. This facility is described by Huff in Ref. 4. The \nMTL receivers were sampled at a rate of 32 samples per second. The \nshort-term (1/2-second) MTL power averages were grouped in seg- \nments representing building faces and then further power averaged to \nprovide a single mean value for each face segment. The segment means \nwere then decibel averaged over all faces to develop the outside \nreference.\n\nThe second method used the PMS at the time of the in-building \nmeasurements to spot check MTL data as a guard against system \nvariations.\n\nMTL measurements were not available for the buildings in the \nChicago downtown area, the AMA building, the IBT Headquarters \nbuilding at Randolph, and the Bell Training Center at West Adams. \nThe PMS was used to collect both inside and outside measurements.\n\nThe PMS was transported to the building locations in a step van. \nWhile the PMS was in the van, the van rooftop antenna was connected \nto it, and street-level data were recorded along the building perimeter.\n\nThe measurements collected along the entire perimeter were decibel \naveraged to provide the outside reference.\n\nThe largest possible variety of areas was included in the routing \nchosen for the PMS measurements in each particular building. In \nmeasuring areas within a building the data were marked so that the \ndifferent interior types could be distinguished. The PMS has a provi- \nsion for including a marker number in the data output. As the PMS \nmeasurements were made along each route, an accompanying com- \nmentary was made on a cassette recorder. This commentary provided \ninformation about the type of building interior area being measured \nand the associated major marker number. The major marker infor- \nmation and the interior type were used subsequently in coding the \ndata for computer program input.\n\nAlso, this commentary was useful in flagging data that should be \ndiscarded and in coordinating the data with the floor area to which it \napplied.\n\nAs we noted earlier, the PMS has the capability to generate average \nvalues that can be selected to represent from 1 to 128 seconds of real \ntime. The 8-second average was chosen for output data in these \nmeasurements. That value represents a compromise between the vol- \nume of data points and the number of feet of travel per data point. \nWith the operational procedures used in these tests, the 8-second \naverages result in one data point for about each 20 feet of travel. A \ntotal data volume of approximately 4000 data points was accumulated \nfor the entire series of in-building measurements.\n\nDuring the initial planning phase, the expected setup channel signal \nstrength in the vicinity of each building was predicted from propaga- \ntion contours of the cell sites in the Chicago AMPS System (see Fig. \n1). Before measuring each building floor, the operator of the PMS \nnoted the received signal strength indicator (RSSI) value at several \nlocations on the floor. If it was determined that a high percentage of \nthe area was served at signal strengths above the PMS receiver \nthreshold, the operator proceeded to collect data at each setup channel \nfrequency over the entire prescribed route for the selected building \nfloor.\n\nIn the measurements made in the first building in the measurements \nprogram (Bell Laboratories, Naperville), a high degree of repeatability \nwas noted when the measurements from PMS output tapes were \ncompared for repeated runs. Based on these visual observations it was \ndecided that a single pass at each frequency of interest would provide \nsufficient accuracy for measurements in the remaining areas.\n\nA comparison of repeated measurements over a common route is \nshown in Table II. The results indicate run averages are repeatable \nwithin approximately 1 dB.\n\nAs the first step in data processing, all measurements are coded to \nassociate them with major marker numbers. The coding identifies the:\n\n. Serving frequency \n. Time of day \n. Number of 1-second samples included in each average \n. Type of interior area (six possible types) \n. Flags indicating: \n(a) repeated measurements \n(b) data to be rejected because of operational trouble \n(c) omission of N data points (beginning or end) for reasons of \nambiguity. \nThe time-of-day information permits verification of the major marker \nsequence and data group duration.\n\nData become ambiguous when the transition from one major marker \nto the next is accomplished without halting the measurements system. \nThe ambiguous data point (8-second sample) contains portions of data \nfrom two major markers; it is deleted from the database.\n\n1. Initial processing of the data (Pass I) develops statistics for all \ndata for each separate category (building, floor, area type, and channel \nfrequency). Invalid data points are eliminated. A summary listing is \nprovided that includes the number of entries, the number of entries \nbelow PMS threshold, the average power level, and the standard \ndeviation for each group of data from categories of areas. In addition \nto this summary, histograms for each group are developed.\n\n2. The second pass introduces the outside street-level reference \ninformation and combines it with the Pass I data. This produces a \ndisplay of the data for each frequency with its differential level relative \nto the outside reference (penetration loss).\n\n3. The third pass combines the penetration loss for all the meas- \nurement frequencies on common floors, as generated in Pass II. This \nallows penetration loss to be used as a base for further combining to \nclassify losses by area type, floor level, etc.\n\n4. The fourth pass combines the penetration loss data from Pass \nIII across different buildings. In the combining process similar data \nfrom different buildings are weighted to compensate for differences in \nthe data volume collected in each specific building. The weighted \naverage loss for a class of data from several buildings represents equal \ncontribution from each building. Pass IV generates information about \nthe effects of classes of building locations (suburban vs. urban) and \npenetration loss by floor and area type, for all buildings.\n\nThe conclusions made as a result of our extensive measurements \nare as follows:\n\n1. Urban vs. suburban\u2014First-floor penetration loss measurements \nwere processed for three different groupings of buildings. The urban \n(downtown Chicago) building penetration for three buildings shows \nan average value of 18.0 dB. The penetration loss distribution for \nsuburban buildings has an average value of 13.1 dB. These statistics \nare listed in Table III. The comparison indicates a loss for the urban \ngroup that is approximately 5 dB greater.\n\n2. Penetration loss sensitivity to transmitter location\u2014Different \npenetration losses are measured on common floors of common build- \nings as a function of the specific serving-site identity. These differences \narise when a building is served by multiple cell sites. The distribution \nof the \u201cmax-min\u201d decibel values for such multisite penetration loss \ndifferences is plotted in Fig. 2. The average max-min penetration loss \ndifference for all the data is 5.4 dB with a sigma of 4.6 dB. This\n\nsuggests that building penetration is a relatively strong function of \nthe direction of illumination.\n\n3. Penetration loss sensitivity to interior type\u2014Table IV compares \nall interiors with and without windows for all buildings and all floors. \nThe presence of windows is shown to reduce the average penetration \nloss by 6 dB, with a sigma of 5.2 dB. Not included in the data is the \noffice area of the Chrysler building in Elk Grove where the copper- \nsputtered glass windows blocked out all measurable signal levels. For \nthe windowed data category, Table IV lists penetration loss for three \ncombinations of interior types. Open interior areas are shown to have \n3 GB less penetration loss than hallways.\n\n4, Floor height effect\u2014A list of building penetration loss values \nranked by floor height is shown in Table II. The high values of loss \nfor the twelfth and fifteenth floors of a single building are the result \nof the shadowing effect of adjacent buildings. While these buildings \neffectively sandwiched the subject building, the adjacent street areas \nwere relatively open. This may be a relatively common occurrence in \nthe urban core and would appear to substantiate the observations of \nDurante.\u2019 Figure 3 is a scatter diagram with a straight line fit for \npenetration loss vs. floor level. The dots indicate data points that \nrepresent decibel averages of penetration losses. For each floor, there \nis a data point for each channel that served that floor. Only floor \nlevels having three or more data points are included in the diagram \nand the building described above was omitted. The mean values for \nall data on each floor are indicated by \u201cX\u2019s.\u201d The \u201cleast-squares\u201d \nstraight line fit to these means is also shown. The slope of the line is \n\u20141.9 dB per floor. The first floor intercept is 10.4 dB. While this loss \nrate agrees closely with the findings in Refs. 1 and 2, the first-floor \npenetration loss is about 10 dB lower. Subsequent measurements by \nBell Laboratories in the Newark, N.J., Cellular Test Bed have con- \nfirmed the ranges of loss values and loss rates presented in this paper. \nIt is assumed that differences in penetration loss with respect to values \nreported previously are probably due to. differences in the methods of \nestablishing the outside street-level reference.\n\nTable 1V\u2014Penetration loss comparisons \nWindowed/nonwindowed areas = 6.0 dB (Standard Deviation = 5.2 dB) \nFor windowed areas:\n\nEnclosed/Hallways = 0 dB (Standard Deviation = 6.0 dB) \nwhere open areas are in conjunction with enclosed areas \nOpen/Enclosed = 1.0 dB (Standard Deviation = 4.8 dB) \nwhere open areas are in conjunction with hallways \nOpen/Hallways = 3.1 dB (Standard Deviation = 5.2 dB)\n\nThe test team members for operations and data measurement were \nM. Patarca, L. F. Smith, and J. G. Rebele. Data keypunching and \ncoding from PMS output tapes and corresponding annotation was \nperformed by M. Patarca. The software and the processing of data to \nproduce the results of the measurements program was completed by \nA. Some. K. K. Kelly collaborated in the selection and location of \ncandidate buildings. Messrs. J. T. Kennedy, G. A. Lothian and J. R. \nNevarez arranged basing the operations in Chicago and obtained the \npermissions necessary to access the buildings for measurements.\n\n1. J. M. Durante, \u201cBuilding Penetration Loss at 900 MHz,\u201d Conference Proceedings, \nIEEE VTG Conf., 1973, pp. 1-7.\n\n2. J. Dietz et. al., \u201cExamination of the Feasibility of Conventional Land Mobile \nOperations at 950 MHz,\u201d FTC Office of the Chief Engineer, Research Division,\n\n3. D. C. Cox, R. R. Murray, and A. W. Norris, \u201cMeasurements of 800-MHz Radio \nTransmissions into Buildings With Metallic Walls,\u201d B.S.T.J., this issue. \n4. Special issue entitled \u201cAdvanced Mobile Phone Service,\u201d B.S.T.J., 58, No. 1 (Jan-\n\n1-11, transmitted to the FCC by Illinois Bell, June 8, 1977-November 26, 1979. \n6. L. P. Rice, \u201cRadio Transmission into Buildings at 35 and 150 mc,\u201d B.S.T.J., 38, No. \n1 (January 1959), pp. 197-210.\n\nMasonry with some corrugated siding. The \ninterior has some windowed office space but \na large portion is devoted to cereal processing \nequipment.\n\nLow light transmission glass in metal frame- \nwork. Mostly windowed halls with interior \noffice space. _\n\nOpen area without nearby structures. Lo- \ncated at the bottom of a hill to the East. \nBrick and glass with high-rise dormitory area. \nWindowed office space.\n\n5th floor\u2014Cloverdale, Eola, Lemont, Lyons \n9th floor\u2014Cloverdale, Eola, Lemont, Lyons\n\nStand-alone with other multistory buildings \nspaced several hundred yards distant.\n\n1st floor\u2014Cloverdale, Eola, Lemont, Lyons \n2nd floor\u2014Cloverdale, Eola, Lemont, Lyons.\n\nStand-alone with other multistory buildings \nspaced several hundred yards distant. \nNarrow windows with concrete in a ratio of \napproximately 50 percent. Interior office \nspace uses half-height dividers. First floor is \na lobby area.\n\nConcrete slabs, only one floor. The interior \nis a large warehouse with metal storage bins \nfor automobile parts. An attached office \nbuilding had no measurable signal because \ncopper-sputtered windows constituted 100 \npercent of outside surface.\n\nCorner buildings in area with other multi- \nstory buildings of about same height.\n\nCorner building without nearby structures. \nBrick and smaller glass windows typical of \nfactory construction. First floor is open area \nwith heavy machinery; second floor is open \narea with assembly lines; third floor is office \nspace with windowed space.\n\nCorner building in area with other multistory \nbuildings of similar height.\n\nReflective glass windows and concrete. The \ninterior office area is open with three-quarter \npartitions.\n\nBounded on four sides by city streets. Neigh- \nboring buildings are multistory.\n\nConcrete and glass with lobby on first floor \nand windowed work and office spaces on up- \nper floors. Office partitions are three-quarter- \nhigh dividers.\n\nNeighboring multistory buildings in a busi- \nness district. The building front is open to \nthe street and both sides abut neighboring \nbuildings. The rear of the building opens into \na service courtyard.\n\nCell site signals: 1st floor\u2014Canal \n2nd floor\u2014Canal \n12th floor\u2014Canal \n15th floor\u2014Canal.\n\nThe outside reference measurements for this building were incom- \nplete because a complete perimeter path was not available. The \nmeasurements for this building are not combined with the results \nfrom other buildings.\n\nThe penetration loss for this building based on the available \nstreet-level reference measurements is 7.3 dB with a sigma of 6.7\n\nEdward H. Walker, B.S.E.E., 1956, Newark College of Engineering; \nWestern Electric Company, 1940-1957; Bell Laboratories, 1958-1981. \nMr. Walker has been involved in the design of radio circuits and the \ndevelopment of military radar systems. Since 1973 he has been respon- \nsible for the experimental evaluation of audio quality of the AMPS \nSystem and of mobile and portable terminals. His work on building \npenetration was completed prior to his retirement from Bell Laboratories \nin 1981. Member, IEEE.\n\nTransmission Errors and Forward Error \nCorrection in Embedded Differential \nPulse Code Modulation\n\nWe have derived formulas for the combined effects of quantization and \ntransmission errors on the performance of embedded Differential Pulse Code \nModulation (DPCM), a source code that can be used for variable-bit-rate \nspeech transmission. Our analysis is more general and more precise than \nprevious work on transmission errors in digital communication of analog \nsignals. Special cases include conventional DPCM and Pulse Code Modulation \n(PCM). Our main result is a signal-to-noise ratio formula in which the effects \nof source characteristics (input signal, codec design parameters) and the effects \nof transmission characteristics (modulation, channel, forward error correction) \nare clearly distinguishable. We also present, in computationally convenient \nforms, specialized formulas that apply to uncoded transmission through a \nrandom-error channel, transmission through a slowly fading channel, and \ntransmission with part or all of the DPCM signal protected by an error- \ncorrecting code. Numerical results show how channel coding can have different \neffects on conventional and embedded DPCM. They also show how the binary- \nnumber representation of quantizer outputs influences performance.\n\nEmbedded coding can play a valuable role in variable-bit-rate speech \ntransmission. With an embedded code the analog-to-digital (a/d) and\n\ndigital-to-analog (d/a) converters operate at a constant, high bit rate, \nand the transmission system controls the instantaneous rate. Proposed \napplications for variable-bit-rate operation include a digital private \nbranch exchange,\u2019 digital speech interpolation,\u201d packet-switched voice \ntransmission,\u00ae and mobile radio.*\n\nSophisticated versions of Differential Pulse Code Modulation \n(DPCM) are promising speech codes for these and other environ- \nments.\u00b0\u2019 However, conventional DPCM is not suited to variable-bit- \nrate transmission because the decoder amplifies the effects of bit-rate \nadjustments. On the other hand, a slightly modified form of DPCM \navoids this problem and produces an embedded code.\u00ae\n\nFigure 1 shows the codec (coder, decoder) structure of embedded \nDPCM. Although up to E bits/sample can be transmitted, the signals \npresented to the two integrators have a resolution of only M bits/ \nsample, the minimum bit rate of the channel. While Fig. 1 is a useful \nguide to practical implementations, Fig. 2, which is equivalent, is \neasier to analyze. It shows the quantizer at the encoder as a successive- \napproximation combination of two quantizers: a \u201cminimal\u201d quantizer \nwith M bits/sample and a \u201csupplemental\u201d quantizer with E-M bits/ \nsample, operating on the error signal of the minimal quantizer.*\n\nIn embedded DPCM, all of the bits from the minimal quantizer \narrive at the decoder; the transmission system can delete some or all \nof the supplemental bits. With S bits/sample of the supplemental \nquantizer transmitted to the decoder, the rate is D = M + S bits/ \nsample, and the quantizing distortion is very close to that of a conven- \ntional codec with D bits/sample.\n\nErrors in the two bit streams have different effects on the decoder \noutput. Errors in the M, minimal bits, are enhanced by the decoder \nintegrator, which has no effect on errors in the S, supplemental bits. \nThis situation compares favorably with conventional DPCM, where \nall errors are integrated at the decoder. It also has implications for \nforward error correction in embedded DPCM. Figures 1 and 2 will be \nfurther explained in Section II.\n\nOur principal contribution in this paper is an analysis of the com- \nbined effects of granular quantizing distortion and transmission errors \non the mean-square error of embedded DPCM. The analysis \nis quite general: special cases include Pulse Code Modulation (PCM) \n(M = 0) and conventional DPCM (S = 0). The formulas for the noisy-\n\nFig. 1\u2014Embedded DPCM encoder and decoder. Because both predictors operate on \nthe same signal (with resolution M bits/sample) performance is unaffected by errors \ndue to bit-rate adjustment. The channel can transmit between M bits/sample and E \nbits/sample.\n\nis DELETE D/A SUPPLEMENTAL \nE-M BITS = E-M BITS r \nBITS PCM CODEC \nTRANSMISSION \nD=S+M \nBITS \nPREDICTOR\n\nA/D \u2014 ANALOG-DIGITAL \nD/A \u2014 DIGITAL-ANALOG \nDPCM CODEC \u2014 DIFFERENTIAL PULSE CODE MODULATION CODER-DECODER \nPCM CODEC \u2014 PULSE-CODE MODULATION CODER-DECODER\n\nFig. 2\u2014Embedded DPCM encoder and decoder. E-bit analog-to-digital conversion is \nshown as a two-stage, successive-approximation process. All M minimal bits are trans- \nmitted. S, the number of supplemental bits transmitted, can range from 0 to E-M.\n\nchannel performance of conventional DPCM are new, and the spe- \ncialization to PCM is more precise than previous work on the sub- \nject,!\u00b0-!? which includes several approximations that are accurate for \nmultibit (= 6-bit) quantizers, but are rather imprecise at lower bit \nrates. The method of analysis has the advantage of separating source \neffects from transmission effects. The source effects include the char- \nacteristics of the analog input signal and the codec design parameters. \nThe transmission effects include modulation and demodulation, the \nchannel, forward error correction, and diversity reception.\n\nThe main result is eq. (61), in which the transmission effects are \ncontained in the discrete probability function P(l), where / is an index \nof binary error patterns. The other symbols in (61) are source param- \neters and functions of source parameters. After deriving (61) we apply \nit to specific transmission environments and present, in Table VI, \nspecialized formulas that are convenient for numerical computation.\n\nIn all, there are 78 formulas in Sections III through VI, most of \nthem intermediate steps in derivations of a few key results. Anticipat- \ning that few readers will require all these details we provide here a \nsummary of the analysis and we display a few numerical results. \nSections I, II, and VII contain the main ideas of our work and sufficient \ninformation to allow readers to perform, on hand calculators, compu- \ntations similar to the ones we present.\n\nSection III introduces the notations for the signals and errors in the \nM-bit minimal DPCM codec and the S-bit supplementary PCM codec \nof Fig. 2. The analysis of Section III leads to (2), which expresses the \nsampled-data error sequence as a function of quantization errors, \ntransmission errors, and integrator characteristics. Section IV begins \nthe analysis of the mean-square value of (2) by deriving (35), the ratio \nof the mean-square codec input to the mean-square value of the \nencoder difference signal. Section V defines A factors, which are \nconditional mean squares of the errors due to specific binary error \npatterns, and derives (61), the general signal-to-noise ratio (s/n) \nformula. Section VI adapts (61) to specific transmission models and \nprovides guides to numerical computation.\n\nFigure 3 shows the performance of embedded DPCM in four trans- \nmission environments, all of them employing Coherent Phase Shift \nKeying (CPSK) modulation at 32 kb/s in a white-Gaussian-noise \nchannel. The encoder operates at 32 kb/s (8-kHz sampling, 4 bits/ \nsample), and in format 1 all of this information is transmitted. Figure \n3 indicates that when the channel s/n falls below 10 dB, the audio \ns/n deteriorates rapidly. In format:2, the least significant bit of each \nDPCM code word is deleted, and the remaining 3 bits/sample are\n\nprotected by a rate 3/4 convolutional code. Although there is more \nquantizing noise than in format 1 (the s/n is 6 dB lower in the absence \nof transmission errors), the convolutional code permits accurate re- \nception of the transmitted bit stream at channel s/n\u2019s down to 3 dB. \nGoing one step further with this approach to channel coding, we have \nformat 4, in which 16 kb/s of speech data are transmitted under the \nprotection of a rate 1/2 code. The threshold of essentially error-free \nperformance is now extended down to a channel s/n of about 0 dB.\n\nIn code format 3 the speech transmission rate is 24 kb/s, as in \nformat 2, but now only 2 of the 3 bits/sample are protected by the \nconvolutional code, which has rate 2/3. The threshold of curve 3 in \nFig. 3 is about 1 dB lower than that of curve 2. On the other hand, \nformat 3 is slightly worse than format 2 in intermediate channel \nconditions (s/n\u2019s between 3 and 5 dB). Over this range, format 2 is \nessentially error free, while format 3 is affected by errors in the \nunprotected third bit of each code word. The effect is small, however, \nbecause these errors are not amplified at the decoder.\n\nWith conventional, rather than embedded, DPCM, the correspond- \ning picture, Fig. 4, is rather different, especially with respect to format \n3. Here channel errors in the unprotected third bit are amplified by \nthe integrator at the decoder. The result is a noticeably lower output \ns/n relative to format 2 (all three bits protected) when the channel \ns/n is between 3 and 6 dB. On the other hand, in clear channels the \ngreater accuracy of prediction in the conventional encoder causes the \noutput s/n of conventional DPCM at 24 kb/s (formats 2 and 3) and\n\nFigure 5, which applies to 24 kb/s speech transmission with a rate \n2/3 code, summarizes the performance differences between conven- \ntional and embedded DPCM. Conventional DPCM has somewhat \nlower quantizing noise, which is reflected in the higher s/n in good \nchannels. In intermediate conditions, when errors in the unprotected\n\nFig. 4\u2014Performance of conventional DPCM in four transmission environments.\n\nFig.5\u2014A 24-kb/s speech transmission with a rate of 2/3 code (Format 3) used to \nsummarize performance differences between embedded and conventional DPCM.\n\nbit influence performance, embedded DPCM is better because the \neffects of these errors are not amplified at the decoder. In very difficult \nchannels, errors in the two coded bits dominate performance and the \ns/n\u2019s of conventional and embedded DPCM are virtually equal.\n\nFigure 1 shows the signal processing operations that take place in \nembedded DPCM encoding, transmission, and decoding. While the \nanalog-to-digital converter at the encoder generates E bits/sample, \nthe resolution of the signal presented to the integrator is limited to M \nbits/sample, where M is the minimum bit rate of the transmission \nsystem. The transmitted bit rate, D, can vary between M and E. At \nthe receiver E\u2014D filler bits are appended to the incoming signal. As in \nthe encoder, E-M bits are deleted at the integrator input so that in \nthe absence of transmission errors the encoder integrator and the \ndecoder integrator produce the same approximation signal. When this \nsignal is added to the full-resolution (D bits) quantizing error, the sum \nhas nearly the quality of a conventional DPCM signal with D bits/ \nsample.\n\nWhile Fig. 1 demonstrates practical implementations, Fig. 2, which \nis equivalent, is easier to analyze. It represents the analog-to-digital \nconversion as a two-step, successive-approximation process. First the \ninput to the converter is represented by M bits/sample. Then the error \nof this representation is processed by another analog-to-digital con- \nverter with E-M bits/sample. Taken together, the two digital signals \ncomprise an E bits/sample representation of the DPCM difference \nsignal. All of the M bits of the minimal analog-to-digital converter are \ntransmitted. The other E-M bits are subject to deletion by the trans- \nmission system. At the receiver the minimal, M-bit signal is processed \nby a conventional DPCM decoder. The result is added to the supple- \nmental, S = D \u2014 M bit representation of the DPCM error signal to \nproduce the system output.\n\nTo analyze Fig. 2, we introduce Fig. 6, which shows the signals that \nappear in the analysis and defines their notations. We are interested \nin the overall error signal\n\nthe difference between decoder output and encoder input. In particular \nwe will derive the formula\n\nestk) = q'slk) - as{k) A/D \u2014 ANALOG-DIGITAL \nep lk) = q'p(k) - ap (k) D/A \u2014 DIGITAL-ANALOG Vk = aq (k-1) + acgag(k-2) +\u00bb + + + aay (k-K)\n\nFig. 6\u2014The signal-processing operations of Fig. 2 and the notation used to analyze \nthem. Three new digital-to-analog converters define signals that appear in the analysis.\n\nwhere p(k) is the quantization noise of the two-stage, D-bit analog- \nto-digital conversion, ep(k) is the effect of a transmission error on the \nentire D-bit transmitted code word, and ey(k) is the effect of a \ntransmission error on the minimal, M-bit DPCM code word. The \ncoefficients b; are related to the predictor coefficients ai, a2, --- ax \naccording to (12). \nFormally,\n\neu(k) = qu(k) \u2014 gm (k), (3) \nthe difference between the quantized inputs to the decoder and encoder \nintegrators. To define np(k) and ep(k), we view the combined code \nword with M + S = D bits as a digital representation of &(k) = x(k) \u2014 \ny(k). A D-bit digital-to-analog converter would produce the quantized \nsignal gp(k), and so we have the definition of quantization error:\n\nAt the receiver, where the D bits/sample are possibly corrupted by \ntransmission errors, a digital-to-analog converter would produce qp(k). \nThe transmision error is\n\nep(k) = qn(k) \u2014 gn(k). (5) \nIn the remainder of Section III we derive eq. (2); in Section IV and V \nwe analyze its mean square. \n3.2 Derivation of the error sequence\n\nHere the signal analysis is facilitated by the transform notation of \nTable I. The reader may verify that the output of the minimal encoder, \nQwu(z), is related to input and quantizing noise\u00ae by\n\nQmu(z) = [X(z) + Na(z)][1 \u2014 F(2)]. (6) \nTable |\u2014Transform notation for codec signals \nEncoder Signal Description Decoder Signal \nX(z) Encoder input, decoder output X\u2019(z) \nMinimal decoder output \u2018u(z) \nY(z) Approximation signal Y\u2019(z) \nQu(z) M-bit representation of X(z) \u2014 Y(z) Q\u2018u(z) \nNa(z) Qu(z) \u2014 [X(z) \u2014 Y(z)] \nQuantizing error in Qy(z) \nQ\u2019ulz) \u2014 Qu(z) En(z) \nTransmission error in Q\u2018y4(z) \nQs(z) S-bit representation of \u2014Ny(z) Q's(z) \nNo(z) Qs(z) + Nu(z) \nQuantizing error in Qs(z) Es(2) \nQ(z) \u2014 Qs(z) ny \nTransmission error in Q(z) \nK \nF(z) Predictor }az\u2122 F(z) \ni=1\n\n\u00bb 7... Q(z) \n5 A eae (7) \nwhich leads to \nXG OR aN ee, (8) \n1 \u2014 F(z) \nIn the supplemental encoder, \nQs(z) = \u2014Na(z) + No(z), (9) \nand at the decoder \nQs(z) = \u2014Nu(z) + Np(z) + Es(z). (10) \nCombining (8) and (10) we have the output of the entire decoder, \nE \nX'(z) = Xu(z) + Qs(z) = X(z) + No(z) + Es(z) + 7 are (11) \nTo transform (11) to time-domain notation, we defined 6,, i = 0, 1, \n2, ---, to be the inverse z transform of 1/(1 \u2014 F), such that \n1 ae \nfre iy) \nThen we have \nx\u2019(k) = x(k) + np(k) + es(k) + Y bieu(k \u2014 i), (13) \ni=0\n\nTo analyze the mean value of (16), we assume that the sequence {x(k)} \nis drawn from a stationary ergodic random process. In our derivations \nwe ignore all correlations in (16) between nonsimultaneous samples. \nThat is, we assume \nE{[no(k) + en(Blem(k- i} x0; i= (17) \nand \nElem(k \u2014 it)em(k-\u2014j)}= 0; 1 Fj. (18) \nEquation (17) indicates that the overall error (quantizing plus channel \ndistortion) in the kth sample is uncorrelated with errors in other \nsamples of the minimal M-bit quantized samples. Equation (18) states \nthat errors in different minimal samples are uncorrelated. These \napproximations are accurate because the sequence of samples at the \ninput to a DPCM quantizer is decorrelated by the differential coding \nprocess and because transmission errors affecting different code words \nare independent or only weakly correlated. \nThe approximations, (17) and (18), remove the last two sums from \nthe expected value of (16), leaving\n\nThe expectations in (19) are related to the quantization and trans- \nmission of &(k), the DPCM difference signal. In Section V, we present \na complete theory of the errors due to these operations. While this \ntheory relates these errors to a = E{\u00a3*(k)}, we are ultimately interested\n\nin the s/n of the codec input, x(k): \ns/n = E{x*(k)}/E{e*(k)} = o3/oe. (21) \nTo find this quantity, we will now derive o2/o; and then combine it \nwith the results of Section V. To begin the derivation, we refer to Fig.\n\nThe first term in (24) depends on the spectral properties of x(k) and \non the predictor. The ratio of o2 to this quantity is called the prediction \ngain,\n\nK K \nG = \u00a5; > ajajr;-;, (26) \nj=0 i=0 \nin which ag = 1, aj = \u2014a;,, i = 1, 2, --- , K, and r, is a normalized \ncovariance coefficient of the stationary input, \nrn = E{x(k)x(k + n)}/o%. (27)\n\nIn evaluating the second and third terms of (24), we ignore corre- \nlation between different quantizing-noise samples and correlations \nbetween quantizing-noise samples and samples of the codec input. \nThus we use (18) and the approximation\n\nThe noise component of (29), which depends on the quantizer \noverload point and on the statistical properties of &(k), is analyzed in \nSection V, where we restrict our attention to granular quantizing noise \nand derive o2(B), the noise power of a B-bit quantizer with unity \noverload point. If the actual overload point is Emax, the noise power of \nthe M-bit minimal quantizer is\n\nand the approximation would be exact if ny,(k) were uniformly distrib- \nuted over the range \u2014Ay,/2 to Ay/2. Table II presents o2(B) numeri- \ncally and indicates the fractional error due to the above approximation. \nA parameter in Table II is the dimensionless load factor\n\nL = Emax/o\u00a2. (33) \nCombining (29), (31), and (33), we arrive at \not = G'o2 + apL?o2(M)o:, (34) \nor the quantity we set out to derive: \no;/o% = G[1 \u2014 apL?o4(M))]. (35)\n\nV. QUANTIZATION NOISE AND TRANSMISSION NOISE IN PCM \n5.1 Granular and overload conditions\n\nTo analyze (19), we study, statistically, the quantization and trans- \nmission of the DPCM difference signal \u00a3(k). In this type of study it is \ncustomary to separate the quantizing error into two components: \noverload distortion and granular noise. In speech communication this \ndistinction is valuable for predicting subjective quality.\u2019*\"* Moreover, \nin analyzing DPCM the distinction is essential because, except for a \ncodec with an ideal integrator,\u201d (F(z) = z~', which is pathologically \nvulnerable to transmission errors), there is no theory for computing \nthe mean-square slope-overload distortion. Thus our analysis sepa- \nrates the transmission of clipped samples of &(k) from samples subject \nto granular distortion. Our theory pertains only to the transmission \nof unclipped samples. For those samples we add two different distor- \ntions, quantizing noise and noise due to transmission errors. Unlike \nslope overload, both of these impairments are essentially uncorrelated \nwith the signal. This gives us confidence that the mean-square sum is \na reasonable quality measure.\n\nwhere \u00e9(k) is the quantizer input and &,,,. is the overload point of the \nuniform DPCM quantizer. The probability of overload is poy and pz, = \n1 \u2014 Poy is the probability of granular quantization. By definition, the\n\nquantizer is overloaded at time k if the quantization error exceeds half \nof a quantization step, i.e., if\n\nOur remaining analysis will be confined to the second expectation \non the right side of (36) and in particular to the ratio,\n\nTo be concise in the remainder of this paper, we will omit the granular \ncondition, | &(k)| < &max, from our notation of expected values.\n\nTo facilitate numerical evaluation of s/n\u2019s, we will present three \ntables of normalized error terms. The normalization relates these \nerrors to a quantizer with a unity overload point and an input with \nprobability density function p,(-). If the quantizer of interest has an \noverload point of &max and the input has the probability density p;(-), \nthe relevant errors are table entries scaled by \u00a3ax. The two probability \ndensities are related by\n\nTo confine our attention to the granular quantization condition, we \nperform our averages with respect to the conditional probability den- \nsity\n\nThe model is illustrated in Fig. 7. The signal u = \u00a3/\u00e9max is processed \nby a B-bit analog-to-digital converter with overload point 1 and step \nsize\n\nThe digital output of the a/d is i, and the corresponding quantized \nsignal is u;, which is related to u by the graph in Fig. 7 and by\n\nFig. 7\u2014Model for analyzing quantizing noise and the effects of transmission errors. \nThe graph and table define the quantizer and two binary number representations.\n\nwith the transformation of i to 1\u2019 characterized by a binary-error \npattern with index 1, | being an integer in the range 0, 2? \u2014 1. To relate \nthe error effect to l, we refer to the natural-binary representation of | \nand specify that a 1 in the bth least-significant position of | causes an \ninversion of the bth most-significant* bit in the binary representation \nof 1. Thus | = 0 refers to error-free transmission (i = 1\u2019); | = 1 refers \nto an error only in the most significant bit; | = 5 refers to errors in \nthe first and third most significant bits; etc.\n\nWith u the quantizer input and / the binary error pattern, we denote \nthe received sample in Fig. 7, uj. It is helpful to separate the complete \nerror u, \u2014 u into quantization-noise and transmission-noise compo- \nnents as follows,\n\n5.3 Conditional expectations of transmission-error effects \n5.3.1 The general approach\n\nOur goal is to evaluate the mean-square of (44) over the joint \ndistribution of input statistics and binary-error patterns. The key to \nour analysis is the definition of A factors, which are conditional mean- \nsquare errors, each related to a specific binary-error pattern, |. By \nanalyzing these conditional errors, we separate the effects of source \ncharacteristics from the effects of transmission characteristics. The \nsource effects are embodied in the A factors; the transmission effects \nare embodied in probabilities of error patterns. These probabilities \ngovern the weighted addition of the A factors to produce the final \nresult.\n\nThis approach to analyzing transmission impairments was intro- \nduced by Rydbeck and Sundberg,\u2019 \u201d who were mainly concerned with \nquantizers with 6 to 8 bits/sample. This high resolution admitted \nvarious approximations that are inaccurate in the 2- to 4-bit quantizers \nof greatest interest for embedded DPCM transmission. Thus we pro- \nceed to a precise calculation of two types of A factors: conditional \nexpectations related to the isolated effects of digital transmission \nerrors and conditional expectations that include correlations between \ntransmission errors and quantizing noise. In high-resolution quantiz- \ners this correlation is negligible, and the two types of A factors are \nessentially equal.\n\nTo compute the mean-square value of (44), conditioned on error \npattern J, we will identify three important quantities: o2(B), the\n\n* This reversal of the bit ordering of / relative to the binary representation of i will \nfacilitate bookkeeping in subsequent computations.\n\ngranular-noise power of a B-bit quantizer; A;(B), the mean-square \neffect of error pattern / on the quantized signal u; and A,(B), the \noverall effect of error pattern / on the mean-square error of the analog \noutput Uj.\n\nTo derive computationally convenient expressions for o2(B), A,(B), \nand A,(B), we defined the integrals\n\nin which p; is the lower boundary and \u00bb;4; is the upper boundary of \nquantizing interval 1:\n\nThe first integral (45) is the probability of using interval 1. The second \nintegral (46) is the average quantization error in interval 1. If B is \nlarge, Ag is small, and q; ~ 0 because u; is in the center of the \nquantization interval. The third integral (47) is the mean-square signal \nwhen the quantizer is in the granular condition.\n\nNow we write the definitions followed by computational formulas \nfor the quantizing noise and the effects of error pattern /:\n\nA\\(B), the difference between the total noise and the quantizing noise, \nincludes the correlation between quantization effects and transmis- \nsion-error effects. In multibit quantizers (B > 4) this correlation is \nsmall, and A,(B) ~ A,(B), an assumption inherent in previous work on \nPCM. Because low-resolution quantizers are of interest in DPCM, we \ntake account of this correlation in our present work.\n\nTable III displays formulas for p;, qi, and o2, that apply to inputs \nwith Gaussian and exponential probability density functions. Because \nthe input, u, of the normalized quantizer is related to the input, &, of \nthe quantizer with overload point \u00a3 max by u = &/Emax, We have\n\nwhere L is the dimensionless load factor defined in (33). Because L is \na familiar quantizer design quantity, we have written the formulas in \nTable III as functions of L.\n\nWith the formulas in Table III, it is a simple matter to compute \no2(B) precisely. However, for B = 4 the approximation\n\nis very accurate (within 3 percent of the exact value for L < 5.6). For \nB = 2 and 3, Table II shows the exact values of o2(B) and the \napproximation errors\n\nfor L = 1.78, 3.16, 5.62 (L? = 10+ 5 dB). \nTo compute A,(B), A(B), it is necessary to know uj, \u2014 u;, which \ndepends on the binary number representation of u;.\n\nWe consider two representations: natural-binary and sign-magni- \ntude, both defined in Fig. 7. Although in general the A,(B) and A,(B) \ndepend on p; and q;, there are some special cases that are important \nand illuminating. For example, in the natural-binary code, the single- \nerror A factors are independent of the signal statistics and of the \nquantizer. An error in the most significant bit causes uj, \u2014 u; = +1 \nprovided it is the only error in the B-bit code word. Thus Ai(B) = 1. \nLikewise, any isolated error in the second most significant bit causes \nan output error of +1/2, and in general a single binary error in position \nb causes the mean-square error\n\nAB) = (1/4)\u00b07; =o\", (56) \nIn the sign-magnitude code, isolated binary errors in positions b = \n2, 3, ---, B have the same effects as corresponding errors in the\n\nnatural-binary code. However, an isolated error in the most significant \nposition transforms u; to u,, = \u2014u;. The mean-square effect is\n\nTable I!I\u2014Formulas for computing A factors and quantizing noise B-bit transmission\n\nReferring to (52) and Fig. 7, we have the mean-square difference \nbetween \u00a3\u2019 and &, over all possible error patterns\n\nwhere P(l) is the probability of error pattern |. The effect of the \ntransmission errors on the quantized version of & is Exax vei \nP(1)A,(B). Returning to (19), we have two expectations: the first is the \ncombined (quantizing and transmission) noise of a D-bit signal; the \nsecond expectation is the noise due to transmission errors in the M- \nbit minimal signal. Thus, we can write (19) as\n\nwhere \u00a32... = L\u2019o? is related to o2 by \n2 2 Vo \n\u2122\u2122 G[l \u2014 apL?o2(M)] \nThe s/n, which is the principal subject of this paper, is, therefore,\n\nWith the exception of the two summations in the denominator, all of \nthe quantities in (61) are properties of the input signal and the codec \ndesign parameters. These summations, \n2P_-1 \u2018 QM_1 \noo = Py P(I)A(D) + bp Py P(I)A(M) (62) \ncomprise the effects of transmission errors on the performance of \nembedded DPCM. We analyze them in Section VI.\n\nThe 2\u201d probabilities, P(l), of binary-error patterns are properties of \nthe digital transmission system, which includes a modulator, a channel, \na demodulator, possibly a codec for forward error correction, and \npossibly a means for combining different versions (diversity branches) \nof the received signal. Depending on these components, the P(J) \nexhibit properties that facilitate evaluation of the sums in (61). In the \nfollowing subsections we consider three paradigms: (1) random errors \nwith statistically independent transmission of all bits; (2) slow fading\n\nwith the bit-error probability constant over each code word, but \nindependent from word to word; and (3) channel coding that makes \nall error patterns equally likely.\n\nErrors in all bits are statistically independent of each other and \noccur with probability, P. The probability of error pattern | depends \nonly on w, the Hamming weight of J, i.e., the number of ones in the \nB-bit binary representation of /. Thus,\n\nThis expansion leads us to express the summations in (61) as poly- \nnomials in P. The coefficients of the polynomial involve the sums of \nall A factors with a fixed weight, w. Let us denote these sums S,,(B) \nwhere, for example,\n\nwhere |, is the set of all error patterns with Hamming weight, w. \nCombining (63) and (65), we can write\n\ndiscovered and proved that for any input probability distribution, \nT,,(B) = T.,(B) = 0 for w = 3. Thus the tranmission term in (61) is\n\nThis formula is valid for channels with random binary errors, and \nTable IV presents values of T,(B), T:(B), T,(B), and T,(B) for three \nquantizer load factors, B = 2-6 bits and Gaussian and exponential \ninputs.\n\nNow the binary-error probability is a random variable that is con- \nstant over each code word but varies from word to word. In this case \nthe effects of digital errors can be calculated as in (69) but with the \naverage values P\u201d replacing P\u2019. These averages are computed over the \ndistributions of channel s/n\u2019s that govern the random fluctuation of \nP from one code word to the next.\n\nTo analyze the performance of embedded DPCM protected by an \nerror-correcting channel code, we make three simplifying approxima- \ntions. The first one, which pertains to the error-correcting code, states \nthat when there is a decoding error, all error patterns are equally \nlikely. Thus we assume that if the C most significant DPCM bits are \nprotected by the code,\n\n1 \nwhere Py is the word-error probability and P, is the binary-error \nprobability of the channel code. They are related by\n\nThe other two approximations apply when C < D, so that the C \nmost significant DPCM bits are protected and the other D-C bits are \nuncoded. To simplify computations for this case, we (1) ignore simul- \ntaneous errors in the protected and unprotected parts of the D-bit \nword and (2) ignore multiple errors in the unprotected part. We \nconsider separately three different relationships among C, the number \nof coded bits; D, the length of the entire DPCM code word; and M, \nthe number of bits in the minimal quantizer.\n\nIn this case P(/) may be calculated according to (70) for all D-bit \nerror patterns, / = 1, 2, ---, 2?\"'. This value of P(l) is constant \nthroughout the first sum in (62). In the second sum we have the \nprobability of M-bit error patterns. For each M-bit error pattern there \nare 2?\u2122 J)-bit patterns. Hence P(/) in the second sum of (62) is higher \nby the factor 2?\u201d than P(l) in the first sum. Because each sum in \n(61) is a constant probability times a sum of A factors, we write\n\nTable V contains numerical values of Agum and Asum for the sets of \nconditions of interest to us here.\n\n6.3.2 Entire minimal code word, parts of the supplemental code word \nprotected (M < C < D)\n\nIn this event we assume that all of the unprotected bits have the \nbinary-error probability P and that as before the protected bits have \nbinary-error probability P.. Furthermore, we set to 0 the probability \nof simultaneous errors in the protected and unprotected parts of the \ncode word. (These errors occur with probability related to P.P.) We \nalso set to 0 the probability of multiple errors in the unprotected part \nof the code word (which occur with probability less than P?). Thus we \nbreak the first sum in (61) into two parts. The first part accounts for\n\nerrors in the first C (protected) bits when the other D-C bits are error \nfree, 1 = 1, 2, ---, 2\u00b0 \u2014 1. The second part accounts for single errors \nin the remaining D-C bits when the first C bits are error free, | = \n2\u00b0, 20+! ..., 22-1. The result is\n\nfor the probability of a single error in the unprotected part of the code \nword. A further approximation A\u00bbi(D) = Asi(D) simplifies computation \nof the second term of (74) because, for natural-binary and sign- \nmagnitude representations, (56) applies for b > 1. This allows us to \nderive\n\nJust as we decomposed the first sum in (61) into two parts in the \nprevious case, we similarly decompose the second sum when some of \nthe M minimal bits are unprotected. The result is\n\nThe useful computational formulas (61), (62), (69), (72), (77), and \n(78) are summarized in Table VI. In this section we apply these \nformulas to illustrate some of the properties of embedded DPCM and \nits relationship to conventional DPCM.\n\nAll of our numerical results pertain to a Gauss-Markov input signal \nwith adjacent-sample correlation r, = 0.85. The codec uses single \nintegration with coefficient a, = 0.85 and the load factor, L = V10. \nFor this configuration the coding gain is G = 3.6. If the embedded \ncodec has a minimal quantizer with M = 2 bits, Cgource in Table VI is \n0.31. For conventional DPCM Cyource = 0.35 with 3 bits/sample and\n\nCsaite G1 \u2014 apL?a2(M)]) \ns/n \u2014 o2(D) rae 2\u00b0 Cia tas L? \nes Notation \nGeneral Formula for Single \nSymbol Description Formula Integration \nG Coding gain (26) (1 \u2014 2ayr; + az)! \nap Predictor gain (30) a? \nL Load factor (33) \nM Minimal codec bits \nD Transmitted bits \nbp 1 + bp is integrator gain (20), (21) a?/(1 \u2014 a?) \nrn (B) Boos autocorrelation 5} \noO uantizing noise 49 -2B \not Transmission-error effects (62) arifa oe Table: Tl \n(b) Error Formulas* \nTransmission Format Error Effect o%\n\n0.36 with 4 bits/sample. Thus the quantizing-noise penalty of the \nembedded codec is 0.54 dB when 3 bits are transmitted, and 0.68 dB \nwhen 4 bits are transmitted. As indicated in Ref. 8 these penalties \nincrease for higher values of L and a,. They decrease rapidly as M \nincreases.\n\nIn our numerical examples the modulation is coherent phase shift \nkeying (CPSK) so that in a white-Gaussian-noise channel the binary \nerror probability is\n\ncodes. For all of them, we use the following truncated union bound to \ncalculate binary-error an \nd+4\n\nwhere m = 1, 2, 3 for the rate 1/2, 2/3, and 3/4 codes, respectively, \nand the coefficients w; and the free distance, d, characterize the \nconvolutional coder and decoder. The codes considered here are punc- \ntured codes!* with constraint length 5 (16 states in the decoder mem- \nory). Table VII contains their coefficients and free distances. The \ncombination of (79) and (81) with the formulas i in Table VI produces \nthe curves in Figs. 3, 4, 5, and 8.\n\nFor transmission environments other than CPSK in a white-Gaus- \nsian-noise channel, there are formulas for P and P, to be used in place \nof (79) and (81). There are many families of modulation schemes, \nchannel conditions, error-correcting codes, and reception techniques \nthat are of practical interest. This paper provides the tools for studying \ntheir effects on the performance of embedded and conventional \nDPCM. This is a subject worthy of further investigation.\n\nWithout forward-error correction, the noise due to transmission \nerrors is dominated by the effects of single errors in the most signifi- \ncant part of the transmitted code word. With the natural-binary \nrepresentation, an error in the most significant bit always causes a \nnoise impulse of half the peak-to-peak range of the quantizer (56). \nWith the sign-magnitude representation, an error in the sign bit \ninverts the polarity of the quantized signal, thereby producing a noise \nimpulse of approximately twice the magnitude of the quantizer input\n\nTable ViI\u2014Source and channel code formats, \nconvolutional code properties\n\nFormat 1 Format 2 Format 3 Format 4 \nSource code \nbits/sample 4 3 3 2 \nbits/second 32K 24K 24K 16K \nbits/sample \nprotected 0 3 2 2 \nChannel code \nrate No code 3/4 2/3 1/2 \nFree distance, d 4 5 7 \nWeight wz 22 25 4 \nWa+1 Error 0 112 12 \nWa+2 Properties 1687 357 20 \nWa43 0 1858 72 \nWat4 66964 8406 225\n\nFig. 8\u2014Performance of embedded DPCM with sign-magnitude and natural-binary \nrepresentations of quantizer outputs.\n\n(57). Consequently, quantizers employing the sign-magnitude repre- \nsentation are somewhat less affected by transmission errors than \nquantizers with the natural-binary representation when the input \nprobability distribution has its mode at zero. This is illustrated in Fig. \n8, which pertains to uncoded 32 kb/s embedded DPCM transmission. \nWhen transmission errors are the dominant distortion, signals repre- \nsented in the natural-binary format are about 2 dB noisier than signals \nrepresented by the sign-magnitude format.\n\nWith forward error correction, all error patterns are equally likely, \nand the two representations have essentially the same s/n.\n\n1. F.S. Boxall, \u201cA Digital Carrier-Concentrator System with Elastic Traffic Capacity,\u201d \nIEEE Trans. Commun., COM-22, No. 10 (October 1974), pp. 1636-42.\n\n2. J. A. Sciulli and S. J. Campanella, \u201cA Speech Predictive Encoding Communication \nSystem for Multichannel Telephony,\u201d IEEE Trans. Commun., COM-21 (July \n1973), pp. 827-35.\n\n3. T. Bially, B. Gold, and S. Seneff, \u201cA Technique for Adaptive Voice Flow Control in \nIntegrated Packet Networks,\u201d IEEE Trans. Commun., COM-28, No. 3 (March \n1980), pp. 325-33.\n\n4. D. J. Goodman and C.-E. Sundberg, \u201cCombined Source and Channel Coding for \nVariable-Bit-Rate Speech Transmission,\u201d B.S.T.J., 62, No. 7 (September 1983).\n\n5. Proceedings of the 1982 IEEE International Conference on Acoustics Speech and\n\nSignal Processing, Paris, May 1982. Session S8, pp. 954-83, contains four papers \non adaptive DPCM at 32 kb/s.\n\n. J.-M. Raulin, et al., \u201cA 60 Channel PCM-ADPCM Converter,\u201d IEEE Trans. \nCommun., COM-30, No. 4 (April 1982), pp. 567-73.\n\n. D. W. Petr, \u201c32 Kbps ADPCM-DLQ Coding for Network Applications,\u201d Record of \nthe IREE Global Telecommunications Conference, Miami, Florida, November \n1982, pp. 239-43.\n\n. D. J. Goodman, \u201cEmbedded DPCM for Variable Bit Rate Transmission,\u201d IEEE \nTrans. Commun., COM-28, No. 7 (July 1980), pp. 1040-6.\n\n. N.S. Jayant, \u201cVariable Rate ADPCM Based on Explicit Noise Coding,\u201d B.S.T.J., \n62, No. 3 (March 1983), pp. 657-77.\n\n11. C.-E. Sundberg and N. Rydbeck, \u201cPulse Code Modulation with Error-Correcting \nCodes for TDMA Satellite Communication Systems,\u201d Ericsson Technics, 32, No. \n1 (1976), pp. 1-56.\n\n12. N. Rydbeck and C.-E. Sundberg, \u201cPCM/TDMA Satellite Communication Systems \nwith Error-Correcting and Error-Detecting Codes,\u201d Ericsson Technics, 32, No. 3 \n(1976), pp. 195-247.\n\n13. D. J. Goodman, B. J. McDermott, and L. H. Nakatani, \u201cSubjective Evaluation of \nPCM Coded Speech,\u201d B.S.T.J., 55, No. 8 (October 1976), pp. 1087-109.\n\n14, B. McDermott, C. Scagliola, and D. J. Goodman, \u201cPerceptual and Objective Eval- \nuation of Speech Processed by Adaptive Differential PCM,\u201d B.S.T.J., 57, No. 5 \n(May-June 1978), pp. 1597-618.\n\n15. L. J. Greenstein, \u201cSlope Overload Noise in Linear Delta Modulators with Gaussian \nInputs,\u201d B.S.T.J., 52, No. 3 (March 1978), pp. 387-421.\n\n16. G. C. Clark and J. B. Cain, Error-Correction Coding for Digital Communications, \nNew York: Plenum Press, 1981, pp. 237-8, and p. 403.\n\nDavid J. Goodman, B.E.E., 1960, Rensselaer Polytechnic Institute; M.E.E., \n1962, New York University; Ph.D. (Electrical Engineering), 1967, Imperial \nCollege, London; Bell Laboratories, 1967\u2014-. Mr. Goodman has studied various \naspects of digital communications including analog-to-digital conversion, dig- \nital signal processing, subjective assessment of voiceband codecs, and spread \nspectrum modulation for mobile radio. He is Head, Communications Methods \nResearch Department. In 1974 and 1975 he was a Senior Research Fellow, \nand in 1983 a Visiting Professor at Imperial College, London, England. \nMember, IEEE.\n\nCarl-Erik W. Sundberg, M.S.E.E., 1966, and Dr. Techn., 1975, Lund \nInstitute of Technology, University of Lund, Sweden; Bell Laboratories, \n1981-1982. Mr. Sundberg is an Associate Professor in the Department of \nTelecommunication Theory, University of Lund, and a consultant in his field. \nHe is Director of the consulting company SUNCOM, Lund. During 1976 he \nwas with the European Space Research and Technology Centre (ESTEC), \nNoordwijk, The Netherlands, as an ESA Research Fellow. He has been a \nConsulting Scientist at LM Ericsson and SAAB-SCANIA, Sweden, and at \nBell Laboratories. His research interests include source coding, channel coding \n(especially decoding techniques), digital modulation methods, fault-tolerant \nsystems, digital mobile radio systems, spread spectrum systems, and digital \nsatellite communication systems. He has published a large number of papers \nin these areas during the last few years. Senior Member, IEEE; member, SER, \nSveriges Elektroingenjorers Riksforening.\n\nThe Comite Consultatif International Telegraphique et Telephonique \n(CCITT) has recently recommended a code for two-level (black and white) \ngraphics transmission. A large number of pictures in graphics communication \ncontain areas that cannot be represented adequately by only two shades of \ngray. We describe techniques by which a composite picture, containing an \narbitrary mixture of two- and multilevel areas, can be coded by schemes that \nare compatible with the CCITT code. First, the composite picture is segmented \nautomatically into two types of areas: one requiring only two levels (text, \ndrawings, etc.) and the other requiring multilevels (for example, photos). A \nDifferential Pulse Code Modulation (DPCM) scheme is then used to code the \nmultilevel areas. Code assignment for the outputs of the DPCM quantizer are \nbased on the local conditional statistics, and the bit stream is processed to \nchange the statistics of the run lengths so that the CCITT run-length code \nbecomes efficient. Results of computer simulations are presented in terms of \nquality of processed pictures and the required bit rate. Simulations show that \nour CCITT: compatible scheme is as efficient as an incompatible but optimum - \nDPCM coding scheme.\n\nSimultaneous developments (algorithmic as well as systems) have \ntaken place for many years in coding and transmission of two-level \n(black and white) document facsimile, and multilevel (many shades of \ngray) pictures.\u201d The former type of pictures require very high spatial \nresolution to preserve the sharpness and have been coded by one-\n\ndimensional run-length coding and two-dimensional edge difference \ncoding [Comite Consultatif International Telegraphique et Telepho- \nnique (CCITT) one- and two-dimensional codes]. On the other hand, \nmultilevel pictures contain gradual luminance transitions, and there- \nfore require lower spatial resolution. They have been coded by Differ- \nential Pulse Code Modulation (DPCM) and transform methods. Most \npictures used in business facsimile systems and audiographics confer- \nencing contain a mixture of two-level and multilevel segments or \nsubpictures. Coding such pictures using two-level techniques would \nnot be adequate from the point of view of the picture quality, and \nusing multilevel techniques would generate enormous data rates. Thus, \nit is of interest to devise schemes that automatically divide a picture \ninto segments, each segment with a specified amplitude (gray shades) \nand spatial resolution and code each segment as best suited for it. \nAnother practical requirement is that of compatibility. A coding \nscheme that handles a mixture of two-level and multilevel segments \nshould be upwardly compatible with the CCITT standard schemes for \ntwo-level pictures. System cost will be reduced if the scheme for two- \nlevel multilevel pictures uses hardware blocks that are also used by \nthe two-level picture coder. We present such a scheme below. Principal \ncharacteristics of our scheme are:\n\n2. Automatic segmentation of pictures into two-level and multilevel \nsegments\n\n3. High coding efficiency by preprocessing the multilevel segments \nto fit the CCITT codes\n\n5. Nonlossy (information preserving) coding of two-level segments, \nand lossy coding of multilevel segments.\n\nThe function of the segmentor is to classify each picture element as \none of the three:\n\n3. Gray \u2014 Multilevel \nFigure 1 shows a neighborhood of the current picture element (pel) \nused for segmentation. We assume that each pel, obtained from the \nscanner, is specified by many shades of gray (e.g., 8 bits). The size of \nthe neighborhood can be arbitrary. If it is too small, then many\n\nFig. 1\u2014Picture elements used for segmentation of the current pel. The size of the \nneighborhood is not necessarily 5 X 5 as suggested in the figure.\n\ndiscontinuous segments of gray pixels will be generated. On the other \nhand, if the neighborhood is too large, then the ability to resolve small \ngray areas is lost. We slide this window of neighborhood over the pels \nalong a scan line and classify each pel. We consider the boundary pels \nof the picture separately. Two thresholds, \u00a2,; and ty (t; < te), are \nselected. It is hypothesized that most black pels will have intensity \nless than t,, white pels will have intensity greater than t2, and gray pel \nintensities may lie anywhere. Within the neighborhood let\n\nWe define a state, S, consisting of three components: S,, S:, and S3. \nA picture is segmented on the basis of the value of S. Let\n\nS=S,+ 82+ 83, (1) \nwhere \nS, = 1, if ng > m + ne \n= 0, otherwise \nS2 = 1, if previous pel is gray \n= 0, otherwise \nand \nS3 = 1, if t,; < intensity of present pel < te \n= 0, otherwise.\n\nS = 2 => current pel gray \nS < 2 and intensity of current pel > T = > current pel white \nS < 2 and intensity of current pel < T = > current pel black.\n\nT is a threshold used to distinguish black elements from white ones, \nonce they are known to be of the two-level type. If the range (te \u2014 \u00a2) \nis decreased by increasing t; and decreasing t2, then more elements \nwill be regarded as two-level and the quality of picture may suffer, but \nthis will also decrease the bit rate.\n\nWe evaluated the performance of the segmentor, in particular its \ndependence on the block size and (t2 \u2014 t,) by computer simulation. \nSince there are no standard mixtures of two-level and multilevel \nimages, we created our own by taking a 512 X 512 gray-level image \n(shown in Fig. 2) and superimposing it on the CCITT documents four \nand five. Since this 512 X 512 original was scanned at low resolution \n(compared to 200 pels/inch used for CCITT documents), it contains \nsignificant sharp transitions that would not be present in a photograph \nscanned at 200 pels/inch. Also, because the original gray-level picture \nand the CCITT documents are rather \u201cclean\u201d, segmentor works quite\n\nwell. However, this may not be a typical situation if a nonideal scanner \nwas used. We, therefore, added random noise to the entire composite \npicture. This noise had a variance of 425 (on an 8-bit scale, 0-255). \nTable I shows the performance of the segmentor with respect to block \nsize for a composite picture made from CCITT document 4. Here t; = \n28, tg = 195. As we view such a segmented picture we realize that a \n5 X 5 block may be too small. A 9 X 9 block appears quite adequate \neven when the added noise variance reaches 758. Higher block sizes \nresult in a larger number of contiguous gray pels, thereby decreasing \nthe number of segments. Figure 3 shows a segmented picture. Due to \nequipment limitations we show only a 512 xX 512 section of the\n\nFig. 3\u2014A segmented picture. Pels classified black and white are reproduced with \nintensities 30 and 215, respectively. Gray-level pels are reproduced with 8-bit intensities.\n\ncomposite picture. The segmentor has adequately separated the two- \nlevel areas from the multilevel areas.\n\nIn most cases, areas of the picture that are segmented to be gray do \nnot need as much spatial resolution as the two-level segments. If the \ntwo-level picture is at a very high spatial resolution (e.g., 200 pels/ \ninch), then without any significant loss of quality, spatial resolution \ncan be reduced in gray areas. Following is a scheme for subsampling \nand interpolation. A subsampling pattern is shown in Fig. 4. Interpo- \nlation is performed by averaging four surrounding pels, as in Fig. 4. \nAlthough we show only 2:1 subsampling, higher subsampling ratios \nmay be used if the quality requirements are not very high. Also, two- \ndimensional subsampling may be performed, but this may increase the \ncomplexity.\n\nAfter the pels (black, white, or gray) are classified and the gray \nareas to be transmitted are determined, a DPCM coder is used for \ngray areas. The resulting bit stream from the DPCM coder is pre- \nprocessed, multiplexed with bits from the two-level segments, and then \ncoded by a CCITT one-dimensional or two-dimensional coder. Ad- \ndresses for the segments of gray pels are coded separately and multi- \nplexed with the coded data to transmit on the channel. A block diagram \nfor the transmitter portion is shown in Fig. 5. Details of the algorithm \nare given below. Only a nonsubsampled case is illustrated; a subsam- \npled case follows trivially.\n\nThe purpose of gray segment coding is to convert an 8-bit/pel signal \nrepresenting gray areas into a coded 3-bit/pel signal, which can then \nbe preprocessed and run-length coded. This procedure reduces the bit \nrate for gray pels to about 2 bits/pel.\n\nFig. 4\u2014Subsampling and interpolation pattern used in gray areas. Only one-dimen- \nsional subsampling is considered.\n\nCCITT \u2014 COMITE CONSULTATIF INTERNATIONAL \nTELEGRAPHIQUE ET TELEPHONIQUE\n\nwee 6\u2014Configuration of pels used for prediction. Only a nonsubsampled case is \nshown.\n\nIt is assumed in this figure that all elements A, B, C are gray elements. \nAppropriate modification is made if some of these are two-level ele- \nments.\n\nThe prediction error is quantized by a symmetric seven-level quan- \ntizer with the transfer characteristics given in Fig. 7. For most pictures \nwith a resolution of 100 pels/inch, this appears adequate, although in \nsome cases dynamic range may not be sufficient. Subjective studies \nare needed to optimize the characteristics for a given set of pictures.\n\nEfficiency can be improved further by adapting the prediction and \nquantization.\n\nTo reduce statistical redundancy and create a bit stream that can \nbe coded compatibly with the CCITT code, seven levels of the quan- \ntizer output are mapped into a three-bit code. First, a table of 49 states \nis constructed by looking at the seven outcomes of the quantized \nprediction values for both elements A and B (in Fig. 6). Given a state, \nthe code words for the present pel are arranged in order of conditional \nfrequency of occurrence. Such statistics are precomputed for a set of \npictures. The code word that is most frequent (for a given state) is \ngiven the code [000], the next highest is given code [001], etc. In \naddition, to decrease the probability of occurrence of isolated \u20181\u2019, if \nthe last bit of the code word for A is a \u20181\u2019, then the entire code word \nfor the present pel is complemented (i.e., \u20180\u2019 \u2014 \u20181\u2019, and \u20181\u2019 \u2014 \u20180\u2019). The \ntable of 49 states and the corresponding code words are shown in \nTable II.\n\nThe code words for various states (e.g., runs) for the CCITT scheme \nare already defined based on the statistics. The statistics of the states \nfor the gray-level segments are quite different. As an example, Fig. 8 \nshows histograms of the runs for black and white pels on which the \none-dimensional CCITT code is based. The same figure also shows \nthe histograms of the runs of the bits from gray-level picture (only \n512 x 512) with the code assignment of the previous section but \nwithout any bit complementing. It is clear that the histograms are not \nsimilar in shape, and therefore using the CCITT code for runs of bits \nfrom gray segments would not be efficient. Since our experience shows \nthat the two-dimensional CCITT code is not efficient for the gray \nsegments, we give below a method of preprocessing that makes efficient \nuse of the one-dimensional CCITT code. Let n2(i) and n\u00b0(i) be the \nhistograms of the runs of black elements for the CCITT code and the \ngray-level segments, respectively. Also let c(i) be the code assigned to \nthe ith run by the CCITT coder. Let j(i) be the sequence that is \narranged in descending order of the histogram function h2({), i.e.,\n\nSimilarly, arrange the gray-level histogram h(i) with the function \nj*(t). Then the code word for a length j*(z) of the gray segment is the \nsame as c(j(i)). Thus, we arrange the two histograms in descending \norder and choose the code to be the same for entries of both the\n\n; ; Decreasing Frequency of Occurrence \u2014> \nQuantizer Quantizer \nNo. Levelfor B LevelforA 000 001 011 111 110 100 010\n\nCCITT \u2014 COMITE CONSULTATIF INTERNATIONAL \nTELEGRAPHIQUE ET TELEPHONIQUE\n\nFig. 8\u2014Histograms of the (a) white and (b) black runs used in the 1D CCITT code \nand the processed gray-level pictures.\n\nrearranged histograms. We found that in reality much of the gain in \ncoding efficiency can be obtained by exchanging the code words for a \nfew run lengths. This leads to simple preprocessing. Figure 9 shows \nthe results of preprocessing on the histograms. It is clear that, if bit \ncomplementing is not used, after the preprocessing the code set is \nmore attached to the histograms and therefore leads to more efficient \ncode. However, if bit complementing is used, the histogram without \nany preprocessing is not too different from the CCITT histogram. \nTherefore, when bit complementing is used, the advantages of pre- \nprocessing are not large. Although this is the case for the picture we \nconsidered, more experiments are needed to evaluate statistics of \ntypical pictures and usefulness of the preprocessing for such statistics.\n\nTo encode positional information of the boundaries of a segmented \npicture, each composite line is considered as a sequence of alternating \nblack and white runs corresponding to the lengths of two-level and \ngray-level segments, respectively. This is then coded by the two-\n\nCCITT \u2014 COMITE CONSULTATIF INTERNATIONAL \nTELEGRAPHIQUE ET TELEPHONIQUE\n\nFig. 9\u2014Effect of preprocessing on the histograms of the (a) white and (b) black runs.\n\ndimensional CCITT code and is transmitted at the beginning of each \ncomposite line.\n\nThe bits resulting from the above procedure for gray areas are \nmultiplexed pel by pel with those of the two-level pels. Our experiments \nshow that while it is advantageous to encode two-level areas by two- \ndimensional code, most of the two-dimensional correlation in gray \nsegments is removed by two-dimensional prediction and code assign- \nment. Therefore, gray areas are coded by one-dimensional code. Since \nthe number of gray-level pels may vary from line to line, in order to \nmaintain proper registration of two-level pels (for two-dimensional \ncoding), a sample count is maintained and is used to initialize the two- \ndimensional coder once it comes out of the gray segment within a line.\n\nResults of computer simulations are given in Tables III and IV. \nTable III shows results for 512 X 512 gray-level picture, and Table IV \nshows results for composite pictures with CCITT standard documents \n4 and 5. It is clear from Table III that for the gray-level picture, \nwithout any preprocessing or bit complementing, coding efficiency is\n\nCoded bits \nDocu- Docu- \nNo. Coding algorithms ment 4 ment 5 \n1. One-dimensional CCITT code on noncomposite docu- 870803 547853 \nment \n2. Two-dimensional CCITT code on noncomposite docu- 577527 286911 \nment\n\n3. Two-dimensional code for two-level, one-dimensional 1169270 893288 \ncode for gray level (no complementing)\n\nrather low. This is a result of the mismatch of the run-length statistics. \nConsiderable improvement is obtained by preprocessing the run \nlengths before applying the CCITT coder. Even higher improvement \nis obtained by the bit-complementing technique. Much of the mis-\n\nmatch between the statistics is removed by the complementing tech- \nnique, and therefore additional improvement obtained by preprocess- \ning the complemented output is marginal. The use of complementing \nmakes it possible to achieve bit rates that are close to the entropy of \nthe coded output. Coding of composite pictures shows similar results. \nAnother interesting conclusion from Table IV is that the two-dimen- \nsional CCITT code is not very efficient for gray-level segments of the \ncomposite picture. This is a result of lack of line-to-line correlation \namong bits that are outputs of the quantizer. Much of the line-to-line \ncorrelation is already removed by the two-dimensional prediction and \nthe bit assignment based on conditional statistics.\n\nWe have presented an algorithm that can automatically segment \nareas of a picture that require only two shades of gray from those that \nrequire many shades of gray. Gray areas are coded in a way that \ncreates a bit stream that subsequently can be efficiently coded by a \nCCITT coder. We find that, for the gray areas, it is possible to achieve \ncoding efficiencies close to the entropy of the DPCM quantizer output. \nTherefore, we conclude that it is possible to encode documents that \ncontain an arbitrary mixture of two-level and multilevel areas using a \nCCITT coder that requires only a preprocessor at the transmitter and \na postprocessor at the receiver.\n\n1. A. N. Netravali, ed., special issue on Digital Encoding of Graphics, Proc. IEEE, 68, \nNo. 7 (July 1980).\n\n3. R. Hunter and A. H. Robinson, \u201cInternational Digital Facsimile Coding Standards,\u201d \nProc. IEEE, 68, No. 7 (July 1980), pp. 830-46.\n\nArun N. Netravali, B. Tech. (Honors), 1967, Indian Institute of Technology, \nBombay, India; M.S., 1969, Ph.D. (Electrical Engineering), 1970, Rice Uni- \nversity; Optimal Data Corporation, 1970-1972; Bell Laboratories, 1972\u2014. Mr. \nNetravali has worked on problems related to filtering, guidance, and control \nfor the space shuttle. At Bell Laboratories, he has worked on various aspects \nof digital processing and computing. He was a Visiting Professor in the \nDepartment of Electrical Engineering at Rutgers University. He is presently \nDirector of the Computer Technology Research Laboratory. Mr. Netravali \nholds over 20 patents and has had more than 60 papers published. He was the \nrecipient of the Donald Fink Prize Award for the best review paper published \nin the Proceedings of the IEEE. Editorial board, Proceedings of the IEEE; \neditor, Transactions on Communications; member, Tau Beta Pi, Sigma Xi; \nSenior Member, IENE.\n\nHamid Gharavi, B.S.E.E., 1970, Tehran Polytechnic; M.S.C. (Digital Com- \nmunication), Ph.D. (Electrical Engineering), 1979, University of Technology, \nLoughborough, England; Research fellow at University of Technology, Lough- \nborough, England, 1979-1980; Lecturer at Auckland University, New Zealand, \n1980-1981; Bell Laboratories, 1982\u2014. Mr. Gharavi has worked on problems \nrelated to digital modulations for satellite communication. He also has worked \non bandwidth compression of color television signals, source coding, graphics, \nand pattern recognition. Member, IEEE.\n\nThis paper describes the Queueing Network Analyzer (QNA), a software \npackage developed at Bell Laboratories to calculate approximate congestion \nmeasures for a network of queues. The first version of QNA analyzes open \nnetworks of multiserver nodes with the first-come, first-served discipline and \nno capacity constraints. An important feature is that the external arrival \nprocesses need not be Poisson and the service-time distributions need not be \nexponential. Treating other kinds of variability is important. For example, \nwith packet-switched communication networks we need to describe the conges- \ntion resulting from bursty traffic and the nearly constant service times of \npackets. The general approach in QNA is to approximately characterize the \narrival processes by two or three parameters and then analyze the individual \nnodes separately. The first version of QNA uses two parameters to characterize \nthe arrival processes and service times, one to describe the rate and the other \nto describe the variability. The nodes are then analyzed as standard GI/G/m \nqueues partially characterized by the first two moments of the interarrival- \ntime and service-time distributions. Congestion measures for the network as \na whole are obtained by assuming as an approximation that the nodes are \nstochastically independent given the approximate flow parameters.\n\nNetworks of queues have proven to be useful models to analyze the \nperformance of complex systems such as computers, switching ma- \nchines, communications networks, and production job shops.'\u2019 To \nfacilitate the analysis of these models, several software packages have\n\nbeen developed in recent years, e.g, BEST/1,2 CADS,? PANA- \nCEA,!\u00b0\u201d and one based on Heffes.\u2019? These software packages contain \nalgorithms for Markov models that can be solved exactly. For some \napplications, the model assumptions are at least approximately satis- \nfied, so that the analysis can be very helpful. For many other appli- \ncations, however, the model assumptions are not even approximately \nsatisfied, so that the analysis can be misleading.\n\nA natural alternative to an exact analysis of an approximate model \nis an approximate analysis of a more exact model. This paper describes \na software package called the Queueing Network Analyzer (QNA), \nwhich was recently developed at Bell Laboratories to calculate ap- \nproximate congestion measures for networks of queues. QNA goes \nbeyond existing exact methods by treating non-Markov networks: The \narrival processes need not be Poisson and the service-time distribu- \ntions need not be exponential. QNA treats other kinds of variability \nby approximately characterizing each arrival process and each service- \ntime distribution with a variability parameter. It is also possible to \nanalyze large networks quickly with QNA because the required calcu- \nlations are minimal, the most complicated part being the solution of a \nsystem of linear equations. The current version of QNA is written in \nFORTRAN.\n\nHere is a rough description of the model: There is a network of \nnodes and directed arcs. The nodes represent service facilities and the \narcs represent flows of customers, jobs, or packets. There is also one \nexternal node, which is not a service facility, representing the outside \nworld. Customers enter the network on directed arcs from the external \nnode to the internal nodes, move from node to node along the internal \ndirected arcs, and eventually leave the system on one of the directed \narcs from an internal node to the external node. The flows of customers \non the arcs are assumed to be random so that they can be represented \nas stochastic processes.\n\nIf all servers are busy at a node when a customer arrives, then the \ncustomer joins a queue and waits until a server is free. When there is \na free server, that customer begins service, which is carried out without \ninterruption. Successive service times at each node are assumed to be \nrandom variables, which may depend on the type of customer but \nwhich otherwise are independent of the history of the network and are \nmutually independent and identically distributed. After the customer \ncompletes service, he goes along some directed arc from that node to \nanother node. The customer receives service in this way from several \ninternal nodes and then eventually leaves the network. A picture of a \ntypical network (without the external node) is given in Fig. 1.\n\nAn important feature of the model is that there may be flows from \nnode j to node i, as well as flows from node i to node j. This is of\n\ncourse useful when customers can return to a node where they previ- \nously received service, but it is also useful when customers cannot \nreturn to a node where they previously received service. Then flows \nfrom node j to node i represent different customers than the customers \nthat flow from node i to node j.\n\nTo be precise about the model, we give a list of basic assumptions. \nIt is worth noting, however, that work is under way to extend QNA so \nthat it can analyze systems in which each of the following assumptions \nis replaced by obvious alternatives. The general approximation tech- \nnique is flexible, so that it is not difficult to modify and extend the \nalgorithm.\n\nAssumption 1. The network is open rather than closed. Customers \ncome from outside, receive service at one or more nodes, and eventually \nleave the system.\n\nAssumption 2. There are no capacity constraints. There is no limit \non the number of customers that can be in the entire network and \neach service facility has unlimited waiting space.\n\nAssumption 3. There can be any number of servers at each node. \nThey are identical independent servers, each serving one customer at \na time.\n\nAssumption 4. Customers are selected for service at each facility \naccording to the first-come, first-served discipline.\n\nAssumption 5. There can be any number of customer classes, but \ncustomers cannot change classes. Moreover, much of the analysis in \nQNA is done for the aggregate or typical customer (see Sections 2.3 \nand VI).\n\ne.g., an arrival can cause more than one departure (see Section 2.2). \n(Think of messages.)\n\nThe general approach is to represent all the arrival processes and \nservice-time distributions by a few parameters. The congestion at each \nfacility is then described by approximate formulas that depend only \non these parameters. The parameters for the internal flows are deter- \nmined by applying an elementary calculus that transforms the param- \neters for each of the three basic network operations: superposition \n(merging), thinning (splitting), and flow through a queue (departure). \nThese basic operations are depicted in Fig. 2. When the network is \nacyclic (e.g., queues in series), the basic transformations can be applied \nsuccessively one at a time, but in general it is necessary to solve a \nsystem of equations or use an iterative method. To summarize, there \nare four key elements in this general approach:\n\n1. Parameters characterizing the flows and nodes that will be readily \navailable in applications and that have considerable descriptive power \nin approximations of the congestion at each node.\n\n2. Approximations for multiserver queues based on the partial infor- \nmation provided by the parameters characterizing the arrival process \nand the service-time distribution at each node.\n\n3. A calculus for transforming the parameters to represent the basic \nnetwork operations: merging, splitting, and departure.\n\n4. A synthesis algorithm to solve the system of equations resulting \nfrom the basic calculus applied to the network.\n\nFig. 2\u2014Basic network operations: (a) Superposition or merging. (b) Decomposition \nor splitting. (c) Departure or flow through a queue.\n\neters are the first two moments. However, we actually work with the \nmean service time 7 and the squared coefficient of variation c2, which \nis the variance of the service time divided by the square of its mean. \nThe user has the option of working with the service rate p = 7\u2019 \ninstead of 7. For the arrival processes, the parameters are associated \nwith renewal-process approximations. The first two parameters are \nequivalent to the first two moments of the renewal interval (interval \nbetween successive points) in the approximating renewal process. The \nequivalent parameters we use are the arrival rate \\, which is the \nreciprocal of the renewal-interval mean, and the squared coefficient \nof variation c2, which is the variance of the renewal interval divided \nby the square of its mean.\n\nWe obtain the approximation of the flows by applying the general \nframework and the basic procedures for approximating point processes \nin Whitt,'* incorporating refinements such as the hybrid procedures \ndeveloped for merging by Albin.\u2019>\u2019\u00ae Of course, the general idea of \nsimple two-parameter approximations for stochastic point processes \ngoes back at least to the equivalent random method for approximating \noverflow streams (see Wilkinson,'\u2019 Cooper,!\u00ae and references there). \nRenewal-process approximations for such point processes were intro- \nduced by Kuczura\u2019\u00ae (also see Rath and Sheng\u201d). Two-parameter \napproximations for networks of queues similar to QNA have also been \ndeveloped by others, apparently first by Reiser and Kobayashi\" (also \nsee Kuehn,\u201d Sevcik et al.,?? Chandy and Sauer,\u201d* Chapter 4 of Gelenbe \nand Mitrani,* and Shanthikumar and Buzacott\u00ae). These two-parame- \nter approximations for networks of queues are also similar in spirit to \ntwo-parameter approximations for networks of blocking systems with \nalternate routing (see Katz?\u2019).\n\nSome authors have referred to these two-parameter heuristic ap- \nproximations for networks of queues as diffusion approximations,*\u201d? \nbut diffusion processes are not actually used. Diffusion approximations \nand associated heavy-traffic limit theorems have motivated some of \nthe heuristic approximations in the literature and in QNA, and they \nare closely related to the asymptotic method for approximating point \nprocesses,\u2018 but the heuristic approximations in QNA are not the same \nas the more complicated diffusion approximations for networks of \nqueues in Iglehart and Whitt,?\u00b0 Harrison and Reiman,\u201d\u2019 and Rei- \nman.7\u00ae?9\n\na generalization of the product-form solution that is valid for Marko- \nvian networks, i.e., in the Markov models the components of the vector \nrepresenting the equilibrium number of customers at each node are \nstochastically independent, so that the probability mass functions for \nthe vector is the product of the probability mass functions for the \ncomponents. While QNA can be thought of as a decomposition method \nor an extended-product-form solution, an effort is made to capture the \ndependence among the nodes. The idea is to represent this dependence \napproximately through the internal flow parameters.\n\nTo see the motivation for QNA, consider the elementary open \nnetwork containing a single node with a single server, an infinite \nwaiting room, and the first-come, first-served discipline. Suppose there \nis a single customer class with each customer being served only once \nbefore departing. The standard Markov model of this elementary \nnetwork, which is embodied in BEST/1, CADS, and PANACBEA, is \nthe classical M/M/1 queue,\u2019*\u201d\u00ae which has a Poisson arrival process \nand an exponential service-time distribution. For the M/M/1 model, \nthe expected waiting time EW (before the customer begins service) is\n\nwhere 7 is the mean service time and p is the traffic intensity, which \nis assumed to satisfy 0 < p <1.\n\nOn the other hand, QNA uses an approximation for the GI/G/1 \nmodel to represent this one-node network. The GI/G/1 model has a \nrenewal arrival process and both the interarrival-time distribution and \nthe service-time distribution are general. In QNA, the arrival process \nis represented by a renewal process partially characterized by two \nparameters: the arrival rate \\ and the variability parameter cz. The \nservice-time distribution is also partially characterized by two param- \neters: the mean service time 7 and the variability parameter c2. In \ncontrast to (1), the formula for the expected waiting time in QNA is\n\nwhere g = g(p, c2, c2) is either one (when c? = 1) or less than one \n(when c2 < 1); see (45). When g(p, c2, c2) = 1, (2) differs from (1) by \nthe factor (c2 + c2)/2. When the arrival process is Poisson, c2 = 1; \nwhen the service-time distribution is exponential, c2 = 1. Hence, if the \nGI/G/1 model is actually an M/M/1 model, (2) reduces to (1). Of \ncourse, the user of QNA can set c2 = c2 = 1 and obtain (1). In fact, \nthe values c2 = 1 and c2 = 1 are default values that the program uses \nif the user does not have variability parameters to provide. Each c? \ncan assume any nonnegative value: c? = 0 for the degenerate deter- \nministic distribution; c? = k7! for an erlang E,, the sum of k i.i.d. \nexponential random variables; and c? > 1 for mixtures of exponential\n\ndistributions. Obviously, the difference between (2) and (1) can be \nlarge, so that (2) often significantly reduces the error.\n\nTo obtain (2), we studied the GI/G/1 queue partially characterized \nby the moments of the interarrival-time and service-time distributions. \nBuilding on previous work by Holtzman,*\u00b0 Rolski,*! and Eckberg,*\u201d we \ninvestigated the set of possible values of EW given the partial infor- \nmation.**3\u201d When c2 = 1, formula (2) is always a possible value, i.e., \nthere always is a GI/G/1 system with interarrival-time and service- \ntime distributions having the specified moments in which (2) is correct. \nIn general, (2) appears to be a reasonably typical value.\n\nFor the single-node example just considered, the arrival process was \na renewal process. More generally, it is natural to think of all the non- \nPoisson arrival processes in the model as renewal processes, either \nbecause they are initially renewal processes or because the algorithm \ncan be-interpreted as approximating general arrival processes by \nrenewal processes. Hence, with one customer class, it is natural to \nthink of the model as a generalization of the open Jackson network \nM/M/m queues to an open Jackson network of GI/G/m queues. Each \nnode is approximated by a GI/G/m queue having a renewal arrival \nprocess independent of service times that are independent and iden- \ntically distributed with a general distribution. It is significant that \nQNA is consistent with the Jackson network theory: If there is a single \nclass of customers, if all the arrival processes are Poisson, and if all \nthe service-time distributions are exponential, then QNA is exact. \nHowever, for the general model few analytical results are available, so \napproximations are needed.\n\nThe software package QNA has a flexible input procedure: the model \nwill accept more than one kind of input (see Section II). For the \nstandard input, only limited information is required. Only two param- \neters are needed for each service-time distribution and each external \narrival process. Also, a routing matrix is needed, which gives the \nproportion of those customers completing service at facility 1 that go \nnext to facility j. (The algorithm is based on Markovian routing.) \nHence, for n nodes, the input consists of n? + 4n numbers.\n\nThere is also an alternate input by classes and routes. In this scheme \nthere are different classes of customers and each class enters the \nnetwork at a fixed node and passes through a specified sequence of \nnodes. For each class, there are two parameters characterizing its \nexternal arrival process and two parameters characterizing the service- \ntime distribution at each node on its route. With this input by routes, \ndifferent classes can have different service-time distributions at a \ngiven node and the same class can have different service-time distri- \nbutions during different visits to the same node. For a class with n \nnodes on its route, the input consists of 3n + 2 numbers. (This includes\n\nQNA also provides a fairly rich output. Several different congestion \nmeasures are calculated for each node: the traffic intensity (utiliza- \ntion), the expected number of busy servers (offered load), and the \nmean and variance of the equilibrium delay and number of customers \npresent. In fact, for single-server nodes the delay distribution itself is \ndescribed. Congestion measures for the entire network are also calcu- \nlated, under the approximation assumption that the nodes are sto- \nchastically independent given the approximate flow parameters. \nMeans and variances of total service times, total delays, and total \nsojourn times (response times) are given. When the input is by routes, \nthese characteristics are given for each customer class. Otherwise, \nthese characteristics are given for any route requested by the user.\n\nA desirable feature of QNA is the structure of the calculus to \ntransform the parameters to characterize the internal flows. The \ncalculus is linear for each network operation, so that the parameters \nfor the internal flows are determined simply by solving systems of \nlinear equations. For the rates, the system of linear equations is just \nthe familiar traffic rate equations occurring in the Jackson network of \nM/M/m queues. After having obtained the rates, we obtain the vari- \nability parameters of the internal flows (the squared coefficients of \nvariations) by solving another system of linear equations. As a by- \nproduct, the existence of a unique nonnegative solution for the flow \nparameters is trivially guaranteed. There is no guarantee that an \niterative scheme will converge, and if it does, there is typically no \nguarantee that a solution is unique. The linearity also guarantees that \nthe computation required is not great. Since there is only one linear \nequation per node in the network, QNA can be used to analyze large \nnetworks repeatedly at minimal cost.\n\nThe linear calculus for transforming the variability parameters \nincorporates results of recent studies to improve the accuracy of the \napproximations. The general framework for approximating point proc- \nesses in Whitt\" is used. Significant improvement over previous ap- \nproximation methods of this kind has been obtained by paying partic- \nular attention to the difficult superposition operation. For superposi- \ntion, we use a modification of the hybrid procedure developed at Bell \nLaboratories by Albin.*16\u00b089\n\nWe emphasize that QNA is spre enats In applications it is im- \nportant to validate the QNA output by comparing it with simulations \nand/or measurements. QNA is designed so that it is easy to incorporate \nimprovements and it is easy to tune QNA for particular applications.\n\nQNA also provides a useful framework for developing new approxi- \nmation procedures. Moreover, it is easy to use QNA in conjunction \nwith other special algorithms available to analyze the nodes or the \nflows.\n\nThe rest of this paper describes QNA in more detail. The paper is \norganized\u2014just as the output is\u2014according to the main steps in the \nanalysis. The input is described in Section II. Section III describes the \npreliminary analysis to eliminate immediate feedback. The procedures \nto determine the internal flow parameters are contained in Section \nIV, and the procedures to calculate approximate congestion measures \nfor the nodes are contained in Section V. Section VI contains the \nprocedures to calculate approximate congestion measures for the net- \nwork as a whole.\n\nIn this section we describe the input options currently available for \nQNA. We anticipate more input options in the future. In Section 2.1 \nwe describe the standard input, which is relatively compact. In Section \n2.2 we describe a minor modification of the standard input, which \nallows for the creation or combination of customers at the nodes. For \nexample, when a packet completes service at some node, it may cause \nseveral packets to be sent to other nodes. In Section 2.3 we describe \nan alternate input for different classes of customers having specified \nroutes. We also describe the way QNA converts this input by classes \nand routes into the standard input of Section 2.1.\n\nWith the standard input, there is a single customer class and no \ncreation of customers at the nodes. Any number of networks can be \nprocessed during a single run, so the user first specifies the number of \nnetworks. Then, for each network, the user specifies the number of \nnodes, and for each node the number of servers. For each node in the \nnetwork, there are two parameters for the service-time distribution \nand two parameters for the external arrival process. Finally, there is \na routing matrix, indicating the proportion of customers that go to \nnode j from node i. Here is a list of the input data for each network \nwith the notation we use:\n\nco; = variability parameter of the external arrival process to node j \n(squared coefficient of variation of the renewal interval in the \napproximating renewal process)\n\ncz, = squared coefficient of variation of the service-time distribution \nat node j\n\nqi; = proportion of those customers completing service at node i \nthat go next to node j.\n\n1 X n vector. The user has the option of inputting 7; or its reciprocal \npj, the service rate at node j. (The same form must be used for all \nnodes.)\n\nThe user need not specify the variability parameters cj; and c%, in \nwhich case they are set equal to the default value one, corresponding \nto the M/M/1 model having a Poisson arrival proccess and an expo- \nnential service-time distribution. (Again, this option applies to all \nnodes.) Alternatively, the user can specify only the service-time vari- \nability parameters, c2,, in which case either all the arrival-process \nvariability parameters are automatically set equal to 1, yielding an \nM/G/1 approximation for each node, or the QNA algorithm is applied.\n\nQNA has an option to allow creating or combining customers at the \nnodes following the completion of service. For example, a message \nprocessed at some node might cause messages to be sent to several \nother nodes. Alternatively, messages might be divided into packets \nafter service at one node and then later recombined into messages \nafter service at another node. In a job shop, the focus might shift back \nand forth between units and lots, e.g., at different nodes we might \nconsider bottles, six-packs, cases, and even truckloads.\n\nWith this option, the user must specify the multiplicative factor y; \nof customer creation or combination at node j for each j. There is \ncustomer creation (combination) at node j if y; > 1 (y; < 1). If \ncustomers are neither created nor combined, then y; = 1. If ); is the \noverall arrival rate to node j, then the departure rate, after this \nmodification, is \\;; and the rate of departure from the network j is\n\nMVYj ( aoe) an) \nk=1 \nWhen artificial nodes are used, the creation or combination can also \nbe placed before service. \nTo obtain our approximation formulas, we work with the following \nmodels of customer creation and combination. These models require \ninteger values, but the approximation formulas and the QNA input do\n\nnot. For customer creation, we replace each departure from node j \nwith a batch of size y;. For customer combination, we replace yj\" \nsuccessive interdeparture intervals by a single one. From these models \nit is not difficult to calculate the impact of y;, e.g., the departure rate \nat node j is simply multiplied by y;.\n\nQNA provides the option of defining different customer classes. \nEach class has its own route or itinerary that specifies the sequence \nof nodes visited. Thus, for each class the routing is deterministic. Each \nclass has an external arrival process that goes to the first node on the \nroute. As usual, the external arrival process is characterized by rate \nand variability parameters. Also, each class may have its own service- \ntime distribution at each node on its route. The service-time distri- \nbutions can be different, not only for different classes, but also for \ndifferent visits to the same node by the same class. These service-time \ndistributions are also characterized by rate and variability parameters. \n(Alternatively, the user can elect to input the service-time parameters \nfor each node. Then all classes have the same service-time distribution \n\u2018at each visit to a particular node.)\n\nAs with the standard input, the user must specify the number of \nnodes and the number of servers at each node. Now we need the \nnumber of routes too. The required data are:\n\nn = number of nodes \nm; = number of servers at node j \nr = number of routes. \nHere is a list of the input data for the kth customer class of a network: \nn, = number of nodes on route k \nA, = external arrival rate of class k \ncz, = variability parameter of the external arrival process for class \nk \nny = the jth node visited by customer class:k \nTj = the mean service time of class k at the jth node of its route\n\nQNA converts this input by classes and routes into the standard \ninput in Section 2.1. It then calculates the parameters of a typical or \naggregate customer. Later, when computing sojourn times or response \ntimes of each customer class, QNA uses the service-time parameters \nof that customer class. The first version of QNA assumes as an \napproximation that each customer sees independent versions of the \nequilibrium distribution at each node. Hence, the waiting time before \nbeginning service at each node is assumed to be the same for all classes \nand all visits.\n\nWe now indicate how QNA converts the input by classes and routes \ninto the standard input of Section 2.1. For this purpose, let 1H be the \nindicator function of the set H, i.e., 1H(x) = 1 if x \u20ac H and 1H(x) =0 \notherwise.\n\ni.e., the external arrival rate at node j, Xo, is the sum of all route \narrival rates , for which the first node on the route is j. (Here the 1H \nnotation is used for H = {k:nz, = j}.) Similarly, the flow rate from i to \njis\n\nFrom (4) and (5), we obtain the routing matrix Q. The proportion of \ncustomers that go to j from i is\n\nIf node i is an active part of the network, then the denominator will \nbe strictly positive. Otherwise, QNA gives an error message.\n\nNext, if the service-time parameters are given by routes, we obtain \nthe service-time parameters for the nodes by averaging:\n\nThe denominator in (7) will be strictly positive if node j is ever visited. \nOtherwise, as with (6), QNA supplies an error message.\n\nWe obtain the node variability parameters c2; using the property \nthat the second moment of a mixture of distributions is the mixture \nof the second moments. Therefore, we have\n\npj = Aytj/mj. (9) \nQNA uses this information to calculate the variability parameters cj; \nof the external arrival process. The hybrid approximation for super- \nposition arrival processes in Section 4.2 is also used here because the \nexternal arrival process to node j is the superposition of the external \narrival processes to node j from the different classes. If \\o; = 0, then \ncj; does not matter and QNA sets cg; = 1. Otherwise,\n\n(np, Nes Ch Nk, Thi, Coa sey Mens Thnrys Ching). (13) \nHere suppose that the r vectors are: \n(2,22) Te Weeds 8,33) \n(35:3, 25 1,2, 072.1. -1> I-21)\n\n(2,.2, 4; 2,1, 1: 1, 2. 1). (14) \nThe first route corresponds to a Poisson arrival process at rate 2 to \nnode 1, with all customers being fed back immediately for a second \nservice before departing from the network. (Of course, the arrival \nprocess need not actually be Poisson; a Poisson process always has \n= 1 but other processes could have c? = 1 too.) The second class \nalso enters at node 1, then goes to node 2 and back to node 1 before \ndeparting from the network, etc.\n\nBy (3), the external arrival rates are \\o; = 5 and doz = 2. By (4), the \ninternal flow rates are \\y, = 2, Aig = 3, Agi = 5, and Age = 0. By (5), \nthe flow rates out of the network are Ayo = 7 sad Azo = 0. By (6), the \nrouting probabilities are: qi, = 1/6, qi2 = 1/4, doi = 1, and qo2 = 0. By\n\n(7), the mean service times are 7; = 2 and ro = 1. By (8), c2, = 1.67 \nand c2, = 1.00. Note that both service times at node 2 in (14). have \nmean 1 and squared coefficient of variation 1, as with a common \nexponential distribution, so we should want rz. = c2, = 1.\n\nTo obtain the internal arrival rates, we solve the traffic rate equa- \ntions as in Section 4.1, i.e.,\n\nFinally we obtain the variability parameters c\u00e9;. First, from (12), \nvy, = 25/13 and v2, = 1.0. Then, from (11), w, = 0.629 and wz = 1, so \nthat c; = 1.38 and c%. = 4. Since there is only one \u2018external arrival \nprocess to node 2, we should have v2 = Wz = 1 and c2. = c2 = 4.\n\nImmediate feedback occurs whenever q;; > 0. Since QNA assumes \nMarkovian routing, each customer completing service at node i is \nimmediately fed back to node i to be served again with probability q;;. \nEach time the customer goes to the end of the line. With the decom- \nposition method, QNA assumes the customer finds the equilibrium \nnumber of customers at the node each time, with each visit being an \nindependent experiment.\n\nQNA eliminates immediate feedback by giving each customer, upon \narrival from another node, his or her total service time before going \nto a different node. This is equivalent to putting a customer immedi- \nately fed back at the head of the line instead of at the end of the line. \nTransitions from node i back to node i are eliminated and the new \nprobability of a transition to node j becomes the old conditional \nprobability given that the customer departs from node i. In other \nwords, each visit to node i from elsewhere plus all subsequent times \nimmediately fed back are interpreted as a single visit. The service time \nis increased to compensate.\n\nThe motivation for this procedure is easy to explain. For a multi- \nserver node with Bernoulli (Markovian) feedback and iid service times \nthat are independent of a general arrival process (not necessarily \nPoisson or renewal), the distribution of the queue length process (but \nnot the waiting times) is the same after this transformation. Hence,\n\nwe calculate the approximate values of the mean and variance of the \nequilibrium queue length for the transformed node without feedback \nand use them to derive approximate waiting time characteristics. By \nLittle\u2019s formula,*!*\u201d the expected waiting time is also exact, i.e., the \nonly error is in the approximation of the arrival process by a renewal \nprocess and the approximations for the characteristics of the GI/G/m \nqueue; there is no additional error due to the immediate feedback. \nThe first step of the reconfiguring procedure is quite simple: the \nnew service time is regarded as a geometric mixture of the n-fold \nconvolution of the old service-time distribution. The parameters 7;, \nc2,, and qj are changed to 7;, \u00a23;, and gj when qi > 0: \nTj = Tif (_1 - Gii) \n63 = gi + (1 \u2014 ques \nGit = 0 \nGy = 9/1 \u2014 qu), J Fl (16) \nAfterwards, when calculating congestion measures for node i, QNA \nmakes further adjustments. When we eliminate immediate feedback \naccording to (16), we no longer count the times a customer is fed back \nimmediately as separate visits. Hence, we need to adjust the congestion \nmeasures that are expressed per visit. For example, since the expected \nnumber of visits to node i per visit from outside is (1 \u2014 qj)~', to obtain \nthe expected waiting time per original visit to node i, we multiply the \nvalues of the expected waiting time EW; obtained from (16) by \n(1 \u2014 qj). Of course, the number of customers at each node is not \naffected by the feedback treatment. \nLet di, 7:, etc., represent the new adjusted values. In terms of the\n\nwhere N; represents the equilibrium number of customers at node i \nand the T; variables represent the sojourn time per visit at node i.\n\nWe obtain the variables 7; and \u00a22; in (17) by inverting the operation \nin (16), so we receive the original data again. The last five formulas in \n(17) involving the second-moment characteristics of W; are based on \nthe results of N; in the transformed system and heavy-traffic limit \ntheorems for networks of queues by Reiman.\u201d*? The main quantity \ndesired is Var(W;); the variable W/ is a preliminary approximation \nfor W;.\n\nIn heavy traffic, the changes in the queue length at the nodes are \nnegligible during a customer\u2019s sojourn in the network. Hence, if node \ni is visited X; times by some customer, then the total sojourn time at \nnode i, say T/, is distributed approximately (in heavy traffic) as X;Tj, \nwhere X; is independent of T/ and T/ is the sojourn time per individual \nvisit in (17). (We use 7T/ and T; instead of 7; and T; because we do \nnot use the description of T; obtained directly from (16) and T; will \ndiffer from T/.) By the independence, ET/? = EX2ET!?. Since X; is \ngeometrically distributed with mean (1 \u2014 qj)~!, c?(X;) = qu, and we \nobtain the seventh formula in (17).\n\nThe sixth and eighth formulas in (17) just express the formula for \nc\u201d in terms of the mean and variance and the fact that the sojourn \ntime is the sum of a waiting time and a service time. The final formula \nfor Var(W/) is obtained by approximating W/ by the sum of N;; iid \nservice times, using standard formulas for the variance of a random \nsum (e.g., compute EW?\u201d by first conditioning on N;). Finally, we \nobtain the fifth formula for Var(W;) by splitting the variance of T/ \ninto waiting-time and service-time components and dividing by the \nexpected number of visits to node i. As a consequence, Var(T}) seems \nmore reliable than Var(W;,). This procedure makes Var(T;), computed \nfrom Var(W,) by adding variance components as in Section VI, agree \nwith the direct formula for Var(T';) in (17).\n\nExperience indicates that eliminating immediate feedback often \nyields a better approximation (see Kuehn\u201d and Sections V and VII of \nWhitt*\u2019). It is also often desirable to reconfigure the network to \neliminate almost-immediate feedback, e.g., flows that return relatively \nquickly after passing through one or more other nodes (see Section V \nof Whitt*\u00b0). Further study is needed to understand feedback phenom- \nena and to develop improved approximations.\n\nIn this section we indicate how QNA calculates the internal flow \nparameters. In Section 4.1 we focus on the flow rates, which are \nobtained via the traffic rate equations, just as with the Markov models. \nIn Section 4.2 we display the corresponding system of linear equations \nyielding the variability parameters. The remaining subsections explain \nhow the variability parameter equations were obtained. The basic \noperations of superposition, splitting, and departure are discussed in \nSection 4.3, Section 4.4, and Section 4.5, and their synthesis in Section \n4.6.\n\nIn this step QNA calculates the total arrival rate to each node. Let \n\\; be the total arrival rate to node j, let y; be the multiplicative factor \nof customer creation at node j as specified in Section 2.2, and let 6; be \nthe departure rate (to other nodes as well as out of the network) at \nnode j. In general, 6; = djv;. If there is no customer creation, then \n+; = 1 and the rate in equals the rate out.\n\nwhere Ao = (Ao;) is the external arrival-rate vector, Q = (q,) is the \nrouting matrix, and I' = (y,;) is the diagonal matrix with y;; = y; and \ni; = 0 fori # 7. When there is no customer creation, y; = 1 and lr = \nI. Of course, (18) is just a system of linear equations. To solve them is \nequivalent to inverting the matrix (J \u2014 T'Q) in (19). When customers \ncan be created at the nodes as in Section 2.2, special care should be \ntaken to be sure that (18) has a solution. We need to have sp(T'Q) <1 \nwhere sp(TQ) is the spectral radius of T'Q.\n\nGiven the arrival rates, it is possible to solve for the traffic intensities \nor utilizations at each node, defined by\n\nIf p; 2 1, then the ith node is unstable. If any node is unstable, the \nalgorithm gives an error message, prints out the traffic intensities, and \nstops. The associated offered load at node i, which coincides with the \nexpected number of busy servers [see p. 400 of Heyman and Sobel*! \nor (4.2.3) of Franken et al.*?] is\n\nThe parameters a; and p; coincide for a single server, with a; tending \nto be more useful as the number of servers, m;, increases (obviously \nwhen m; = \u00a9).\n\nAfter the arrival rates have been calculated for the nodes, QNA \ncalculates related quantities for the arcs:\n\nAy = AViGs \u2014the arrival rate to node j from node i \nDiy = Ay/Aj \u2014the proportion of arrivals to j that \ncame from 1, 1 = 0. (22)\n\nd; = Nii ( - > a) \u2014the departure rate out of the \n: ee network from node i \nc=) d= the total departure rate out \nte of the network. (23)\n\nThe heart of the approximation is the system of equations yielding \nthe variability parameters for the internal flows, i.e., the squared \ncoefficients of variation for the arrival processes, c3;. (These are \nderived in Sections 4.3 through 4.7.) The equations are linear, of the \nform\n\nwhere x;, vj, and w; depend on the basic data determined previously, \ne.g., pi, m; and c2;, but not on the variability parameters c3; being \ncalculated. The parameter 7; is the multiplicative factor of customer \ncreation or combination, introduced in Section 2.2. The variables x; \nand \u00bb; are used to specify the departure operation; the variable w; is \nused to specify the superposition operation. The variables \u00bb; and w; \nare weights or probabilities that are used in convex combinations\n\narising in hybrid approximations for departure and superpositions, \nrespectively. The variables x;, v;, and w; are included to make modifi- \ncation of the algorithm based on (24) easy. The specific values in this \nversion of QNA are:\n\nIt is significant that it is easy to modify this system of equations. \nFor example, other hybrid procedures for departures or superpositions \ncan be introduced just by changing \u00bb;; and w;, respectively. In this way, \nit is easy to calculate and compare the variability parameters for \nseveral different approximation procedures.\n\nThe purpose of the following sections is to explain the key approx- \nimation equations (24) through (30), which yield the variability param- \neters for the internal flows. The approximations are all based on the \nbasic methods in Whitt:** the asymptotic method and the stationary- \ninterval method. We consider the basic operations\u2014superposition, \nsplitting, and departure\u2014in turn, and then their synthesis.\n\nFor superposition, the stationary-interval method is nonlinear so it \npresents difficulties.'***?* Moreover, there appears to be no natural \nmodification that makes it linear. On the other hand, the asymptotic \nmethod is linear. By the asymptotic method, the superposition squared \ncoefficient of variation c4 as a function of component squared coeffi- \ncients of variation c? and the rates ); is just the convex combination\n\nHowever, neither the asymptotic method nor the stationary-interval \nmethod alone works very well over a wide range of cases, e.g., see \nSection ITI of Whitt.*\u00b0 Albin\u2019\u201d\u201d* found that considerable improvement \ncould be obtained by using a refined composite procedure, which is \nbased on a convex combination of cX for the asymptotic method and \nc& for the stationary-interval method. Her hybrid c# is of the form\n\nUnfortunately, since c% is nonlinear, so is c?;. However, Albin found \nthat a convex combination of c&{ and the exponential c? of 1 worked \nalmost as well, having 4-percent average absolute error as opposed to \n3 percent. Hence, we use such a hybrid procedure, namely,\n\nwhere w is a function of p and the rates. Extensive simulation \nprompted Albin to suggest the weighting function\n\nNote that if there are k component processes with equal rates then \nv = k. The parameter pv can be thought of as the number of component \nstreams, with it being an equivalent number if the rates are unequal.\n\nHowever, the weighting function (58) fails to satisfy an important \nconsistency condition: We should have w = 1 when p = 1; if there is a \nsingle arrival process, the superposition operation should leave the c? \nparameter unchanged. Moreover, new theoretical results\u00ae\u00ae indicate \nthat the exponent of (1 \u2014 p) in (34) should be 2. Hence, we use formula \n(33) based on the weight function w in (29).\n\nNo approximation is needed for splitting because a renewal process \nthat is split by independent probabilities (Markovian routing) is again \na renewal process. However, approximation is of course indirectly \nassociated with this step because the real process being split is typically \nnot a renewal process and the splitting is often not according to \nMarkovian routing.\n\nSince a renewal process split according to Markovian routing is a \nrenewal process, the asymptotic method and the stationary-interval \nmethod coincide. If a stream with a parameter c\u2019? is split into k streams, \nwith each being selected independently according to probabilities p;,\n\ni= 1, 2, ---, Rk, then the ith process obtained from the splitting has \nsquared coefficient of variation c? given by \nc} = pic? + 1 \u2014 pi, (36)\n\nrenewal-interval distribution in the split stream is a geometrically \ndistributed random sum of the original renewal intervals.\n\nFor the stationary-interval method with single-server nodes, we \napply Marshall\u2019s formula for the squared coefficient of variation of an \ninterdeparture time, say ci, in a GI/G/1 queue:***\n\nwhere EW is the mean waiting time. Since EW appears in (387), the \ncongestion at the node affects the variability of the departure process. \nA stationary-interval method approximation for ci is obtained by \ninserting an approximation for EW in a GI/G/1 queue. Our analy- \nsis***\u201d suggests that it suffices to use the linear approximation (2) \nwith g set equal to one. When this is combined with (37), we obtain\n\nA simple extension of (38) for GI/G/m queues that is being used in \nthe current version of QNA is\n\ncqg@=1+(1\u2014 p*)(e3 \u20141) + Un (cS; \u2014 1). (39) \nNote that (39) agrees with (38) when m = 1 and (39) yields cj = 1 as \nit should for M/M/m and M/G/ systems for which the stationary \ndeparture process is known to be Poisson. The third term in (39) \napproaches 0 as m increases, reflecting the way multiple servers tend \nto act as a superposition operation. A basis for further refinements of \n(39) is the asymptotic analysis of departure processes in Whitt.\u201d This \nasymptotic analysis shows that in some cases the variability of the \ndeparture process depends on the arrival and service processes in a \nmore complicated way.\n\nAs with superposition, the asymptotic method yields a more elemen- \ntary approximation than the stationary-interval method. In fact, the \nasymptotic-method approximation for the departure process is just \nthe arrival process itself, i.e., the asymptotic-method approximation \nfor c4 is just c2.4* The number of departures in a long interval of time \nis just the number of arrivals minus the number in queue, and the \nnumber in queue fluctuates around its steady-state distribution, \nwhereas the number of arrivals goes to infinity.\n\nIt remains to combine the basic methods to form a refined hybrid \nprocedure. However, limited experience indicates that this refinement \nis not as critical as for superposition. The stationary-interval method\n\nalone seems to perform better for departure processes than for super- \nposition processes.**\n\nThe most appropriate view for the departure process\u2014the station- \nary-interval method or the asymptotic method\u2014depends on the traffic \nintensities at the next nodes where the departures are arrivals. As the \ntraffic intensity of the next node increases, the asymptotic-method \napproximation for the departure process becomes more relevant. For \nexample, consider the case of two queues in series with parameters \nNols Cdr; M1, C21, Me, and \u00a22). If 42 \u2014> A while uw; remains unchanged, then \np2 \u2014 1 and the second queue is in heavy traffic. Under such heavy \ntraffic conditions, it has been shown\u201d that the congestion measures \nat the second node are asymptotically the same as if the first facility \nwere removed, i.e., as if the arrival process to the second node were \njust the arrival process to the first node. More generally, for any arrival \nprocess it has been proved that the asymptotic method is an asymp- \ntotically correct approximation for a queue in heavy traffic.\u201d\u00b0\n\nHence, it is natural to tune the departure approximation by using \nthe traffic intensities in the following nodes. Since the departure \nprocess typically will be split and sent to different nodes with different \ntraffic intensities, it is appropriate to do the tuning after splitting. Let \nc3; be the departure c\u201d at node i. Then\n\nis the c\u201d for the portion of the departures going to node j. We let c?; be \na weighted combination of the approximations obtained by the asymp- \ntotic method and the stationary-interval method [using (39)]:\n\nwhere vj; is chosen to satisfy 0 < yj; < 1 and be increasing in p; with \nvi; > 1 as p; \u2014 1. However, we have not yet found that positive v; \nhelps,\u201c so the current version of the QNA uses (28).\n\nFrom (38) it is clear that the departure process variability, as \ndepicted by QNA, is an appropriate weighted average of the arrival- \nprocess variability and the service-time variability. Hence, when the \nservice time is deterministic, so that cz, = 0, the departure process is \nless variable than the arrival process. However, the actual reduction \nof variability in a network caused by deterministic service times often \nis not as great as predicted by (38) or (39). Hence, we have replaced \n(39) by\n\nWe treat customer creation or combination as a modification of the \ndeparture process. When there is customer creation at node i, we \nreplace each departure by a batch of size y;. When there is combination \nat node i, we replace each interdeparture interval by the sum of yj; \nsuch intervals. These make more sense for integer values, but we do \nnot require it. Hence, as described in Section 2.2, the departure rate \nfrom node i is y;A; when the arrival rate is \\;. We use the asymptotic \nmethod to obtain the variability parameter. Since the number of \ndepartures from node i in a large time interval is y; times the number \nof arrivals, the asymptotic-method approximation of the variability \nparameter for customer creation or combination is just to multiply \nc3; by y;. (By the asymptotic method, c? = lim,_,.Var N(t)/EN(t); see \nSection 2 of Whitt.\u2019) This is done before splitting.\n\nWe obtain the basic system of equations (24) through (30) by \ncombining Sections 4.3 through 4.6 as follows:\n\nThe first line is based on superposition, Section 4.3, and the second \nline is based on departure, splitting and customer creation, Sections \n4.4 through 4.6.\n\nHaving calculated the rate and variability parameters associated \nwith each internal arrival process, we are ready to calculate the \napproximate congestion measures for each node. At this point we have \ndecomposed the network into separate service facilities that are ana- \nlyzed in isolation. Each facility is a standard GI/G/m queue partially \ncharacterized by five parameters: the number of servers plus the first\n\ntwo moments of the interarrival time and the first two moments of \nthe service time. Instead of the moments we use the arrival rate \\, the \nmean service time 7, and the squared coefficients of variation c? and \nc2, Since we are focusing on a single node, we omit the subscript \nindexing the node throughout this section.\n\nThere are many procedures that could be applied at this point. \nWe could fit complete distributions to the parameters, and then \napply any existing algorithm for solving a GI/G/m queue or a special \ncase. Among the attractive options are procedures for analyzing the \nGI/G/1 queue,\u201c* the M/PH/m queue with phase-type service-time \ndistributions,*\u201d>\u201d the GI/H;,/m queue with hyperexponential service- \ntime distributions\u2122* and the GI/E,/m queue with Erlang service- \ntime distributions.\u201d Also available are approximations based on \nheavy-traffic and light-traffic limiting behavior.***\u2019 The actual proce- \ndures used in this version of QNA, however, are quite elementary. Our \nstudy of the GI/G/1 queue***\u2019 indicates that these elementary proce- \ndures are consistent with the limited information available. Since the \narrival process is usually not a renewal process, and since only two \nmoments are known for each distribution, there is little to be gained \nfrom more elaborate procedures. In fact, a user of QNA should be \ncautioned not to rely too heavily on detailed descriptions such as the \ntail of the waiting-time distribution. Such detailed descriptions may \nprove to be reasonably accurate, but they should certainly be checked \nby simulation.\n\nWe begin with the steady-state waiting time (before beginning \nservice), here denoted by W. The main congestion measure is the \nmean EW, but we also generate an entire probability distribution for \nW. First, the approximation formula for the mean is as in (2):\n\nWhen c?2 < 1, (44) is the Kraemer and Langenbach-Belz approxima- \ntion,*\u00ae which is known to perform well.*?\u00b0\"\u00b0? When c?2 > 1, the original \nKraemer and Langenbach-Belz refinement does not seem to help, so\n\nLet the number of customers in the facility, including the one in \nservice, be denoted by N. The probability that the server is busy at an \narbitrary time, P(N > 0), and the mean EN can be obtained from \nLittle\u2019s formula (see Section 11.3 of Heyman and Sobel\"*\u2019):\n\nFormula (46) is exact even for stationary nonrenewal arrival processes \nand (47) is exact given EW.\n\nFor the probability of delay, P(W > 0), denoted here by oc, we use \nthe Kraemer and Langenbach-Belz approximation:*\u00ae\n\nFormula (48) also yields the correct value for M/G/1 systems, namely, \np. Additional supporting evidence for (48) is contained in Whitt.\n\nWe next focus on the conditional delay given that the server is busy, \ndenoted by D. Obviously, ED = EW/c. We first give an approximation \nformula for the squared coefficient of variation of D, ch. This formula \nis the exact formula for the M/G/1 queue, with the service-time \ndistribution being H? when c? = 1 and E, when c2 = k\u2122\u2019, where H? is \nthe hyperexponential distribution with balanced means and E, is the \nErlang distribution (see p. 256 of Cohen\u00ae and Section 3 of Whitt\"). \nThe idea underlying this approximation is that the conditional delay \nD in a GI/G/1 queue (rather than the total delay W) depends more \non the service-time distribution than on the interarrival-time distri- \nbution. Hence, the M/G/1 formula for cj is used as an approximation \nfor all GI/G/1 systems. The M/G/1 formula for cj is:\n\nwhere d? = E(v\")/(Ev)? with v being a service-time random variable. \nEven E(v*) is available, it can be used in (50), but since E(v?) is not \navailable with two parameters, we use approximations for d?. The \napproximations are based on the H8 and E* distributions.\n\nWe obtain formula (52) by considering an Erlang EF, variable, which \ncan be represented as the sum of k iid exponential random variables \nX; with mean 7/k, where 7 is the mean of the E; variable. In this case\n\nwhich reduces to (52) because c2 = k\u2122 for an E, variable. Note that \n(51) and (52) agree at the boundary when c? = 1.\n\nVar(D) = (ED)*c}, = (EW)?cd/o? \nE(D\u2019) = Var(D) + (ED). (53) \nFrom D we then obtain second-moment characteristics for W: \nE(W*) _ , oH) _ , cb tle \n(EW)? (cED)? oO i \nVar(W) = (EW)*c and E(W?\u2019) = Var(W) + (EW)?. (54)\n\nWe now indicate how QNA calculates an approximate probability \ndistribution for W. The distribution has an atom at zero as given in \n(48) and a density above zero. The density is chosen so that W and D\n\nhave the first two moments already determined for them. (This is the \ngeneral rule, but it is not quite followed in Cases 2 and 4 below.)\n\nCase 3: 0.501 < c?% < 0.99. Let the distribution of D be the convo- \nlution of two exponential distributions with parameters y; and y2 \n(v1 > \u00a52), i.e., let D have density\n\nCase 4: ch < 0.501. Let D have an E, (Erlang) distribution with \nmean ED, which has c\u201d = 0.5. Its density is\n\nFor deterministic service times, d3 = 1, so that the smallest possible \nch via (50) is (1 + 2p)/3. Hence, Case 4 above will not occur often.\n\nFinally, we come to the second moment and variance of N, the \nnumber in system. For the M/G/1 queue, it is not difficult to compute \nE(N?). Since the steady-state number in the facility is equal to the \nnumber of arrivals during a customer\u2019s time in the facility, it is easy \nto compute the moments of N from the moments of W; for example,\n\nand o is the probability of delay in (48). The maximum is used in (65) \nto avoid dividing by zero. For the M/G/1 queue, Y2 = 1; for GI/M/1 \nqueues, Y2 in (65) provides just the right correction, so that (64) is \nexact given the true value of o, EW and Var(W). The correction Y2 \nin (65) makes (64) too small for D/D/1 queues by a factor of (1 + p)7?, \nbut (64) is asymptotically correct in heavy traffic: ci) ~ 1 as p > 1 if \neither c2 > 0 or c2>0. \nFrom (47) and (64) we immediately obtain\n\nWhen there is immediate feedback at the node and it is eliminated, \nadjustments are necessary in the formulas of this section, as indicated \nin Section III.\n\nThe first congestion measures for multiserver nodes provided by \nQNA are exact. Even for nonrenewal stationary-arrival processes, the \nexpected number of busy servers is just the offered load [see p. 400 of \nHeyman and Sobel*! and (4.2.3) of Franken et al.*\u201d}:\n\nE min{N, m} = a = At (67) \nand the traffic intensity or utilization is \np = a/m. (68) \nBy Little\u2019s formula, as in (47), \nEN =a+dEW. (69)\n\nQNA currently provides only a few simple approximate congestion \nmeasures for multiserver nodes. These are obtained by modifying the \nexact formulas for the M/M/m model.\u2019\u00ae Let characteristics such as \nEW<(c?2, c2, m) represent the characteristic as a function of the param- \neters c2, c2, and m, and let characteristics such as EW(M/M/m) be \nthe exact value for M/M/m system. A simple approximation for EW\n\nEW(c?, c2, m) = ( EW(M/M/m). (70) \nFormula (70) has frequently been used for M/G/m queues and is \nknown to perform quite well in that case. \u00a9\u2019 By virtue of heavy-traffic \nlimit theorems, we know that (70) is also asymptotically correct for \nGI/G/m systems as p \u2014 1 for fixed m. Limited additional study \nindicates that (70) is also reasonable for moderate values of p when \nc2 > 0.9 and c2 = 0.9, or when c2 < 1.1 and c? < 1.1. The actual \nvalue may be significantly smaller (larger) when c2 < 0.9 and c2> 1.1 \n(c2 > 1.1 and c2 < 0.9).\n\nThe simple approximation (70) is also supplemented by simple \napproximations for the second moments of W and N. They are \nobtained from:\n\nMore detailed and sophisticated approximations for multi- \nserver nodes are being studied. As we indicated before, a variety of \nmethods and algorithms can be applied given the parameters of the \narrival process.*\u201d7*\u201d\n\nIn this section we describe the approximate congestion measures \ncalculated by QNA for the network as a whole. In Section 6.1 we \ndiscuss congestion measures representing the system view, e.g., \nthroughput and number of customers in the network; in Sections 6.2 \nand 6.3 we discuss congestion measures representing the customer \nview, e.g., number of nodes visited and response times. In fact, there \nare actually two different customer views. In Section 6.2 we discuss \nthe view of an arbitrary, typical, or aggregate customer; in Section 6.3 \nwe discuss the view of a particular customer with a specified route \nthrough the network.\n\nA basic total network performance measure is the throughput, which \nwe define as the total external arrival rate Xo,\n\nWhen no customers are created at the nodes, the total external arrival \nrate equals the total departure rate from the network, so that there is \nlittle ambiguity about what we mean by throughput. However, when \ncustomers are created or combined at the nodes, as in Section 2.2, \nthere is more than one possible interpretation. We might be interested \nin the rate at which arrivals are processed, i.e., (72). For example, the \ncustomers created at the nodes might be regarded only as extra work \nthat must be done to serve the arrivals. On the other hand, we might \nbe interested in the rate at which customers leave the network or in \nthe rate of service completions. The departure rate from the network \nis\n\nA description of the overall congestion is provided by the mean and \nvariance of the number N of customers in the entire network. In \ngeneral,\n\nand, as an approximation based on assuming that the nodes are \nindependent, we have\n\nFormula (76) is valid for the Markovian models as a consequence of \nthe product-form solution, but is an approximation in general.\n\nWhen we turn to the congestion experienced by individual cus- \ntomers, there are two very different approaches. The first approach \nkeeps strict adherence to the model assumptions with the standard \ninput in Section 2.1, and is based on interpreting the routing matrix \nas independent probabilities (Markovian routing). This means that \neach time any customer completes service at node i, that customer \nproceeds to node j with probability q,, independent of the current state\n\nand history of the network. If the network is cyclic, this means that \nevery customer has positive probability of visting some nodes more \nthan once. This is the perspective of an aggregate customer. It might \nbe that no individual customer actually ever visits the same node more \nthan once.\n\nIf the aggregate view is desired, then the customer experience can \nbe described by employing the basic theory of absorbing Markov chains \nas in Chapter III of Kemeny and Snell. We can regard the external \nnode as a single absorbing state to which all customers go when they \nleave the network or we can have more absorbing states, to distinguish \nbetween network departures from different nodes or different subsets \nof nodes. For this interpretation, the routing matrix Q is the transient \nsubchain associated with the absorbing Markov chain and the inverse \n(I \u2014 Q)7 in (19) with I = J is the fundamental matrix of the absorbing \nchain (see p. 45 of Kemeny and Snell\u00ae). Solving the traffic-rate \nequations is tantamount to solving for this fundamental matrix.\n\nFrom the fundamental matrix it is easy to calculate the moments of \nthe number n,; of visits to any state j starting from any state i (on an \nexternal arrival process). For example, En, is just the (i, j)-th entry \nof (I \u2014 Q)~\". It is also easy to calculate the probability of absorption \ninto each of the absorbing states starting from any initial distribution. \nThese various congestion measures are easily obtained working with \nn X n matrices.\n\nSuppose that we focus on an arbitrary, typical, or aggregate customer \narriving on an external arrival process. Then that customer enters \nnode i with probability Xoi/Ao, where Ao is defined in (72) and the \nexpected number of visits to node i for each customer is\n\n(We have used the fundamental matrix to get ),.) The mean of the \ntime, 7;, that an arbitrary customer spends in node i during his or her \ntime in the network is thus\n\nET; a (EV;) (7; + EW;) (78) \nand the expected total sojourn time (time spent in the network from \nfirst arrival to final departure) for an arbitrary customer is thus\n\nET = >; ET; = > EV;(7; + EW;). (79) \ni=1 i=1 \nThe variance of 7; is thus \nVar(7\u2019;) = EV\u00abVar(W;) + ric2;) + Var(V;)(EW; + 7;)*. (80) \nThe term Var(V;) in (80) as well as EV; is easily obtained from the \nfundamental matrix. In particular, Var(V;) = EV? \u2014 (EV;)? and\n\nwhere F is the fundamental matrix (I \u2014 Q)\u2019, Fag is the n X n matrix \nwith all off-diagonal entries 0 and diagonal entries the same as F.\n\nTo obtain an approximation for the variance of the total sojourn \ntime in the network, we assume that the sojourn times at the different \nnodes are conditionally independent, given any particular routing. \n(This is not valid even for all acyclic networks of M/M/1 nodes,\u00ae? but \nis often approximately true.\u201d\u201d\u00b0) In particular, for a customer entering \nsome specified node and making V; visits to node j, 1 <j <n, before \neventually leaving the network,\n\nnV; \nT= (3 > Ty), (82) \nj=l k=1 \nwhere T}, is the sojourn time for the kth visit to node j. Our approxi- \nmation assumption is that the variables T;; are mutually independent \ngiven the vector (Vi, Vo, ---, Vn). \nHence, \n2\n\nAnother approach is to decouple the macroscopic and microscopic \ninterpretations. This view is common in statistical mechanics. The \ntotal network may exhibit statistical regularity not evidenced in any \nsingle particle (customer). In this view, we think of the total system \nevolving as if customers were routed according to independent proba- \nbilities, even though individual customers may have very different\n\nrouting probabilities, perhaps nonrandom routing or acyclic routing. \nFor example, we may consider the cyclic network entirely appropriate \nfor the macroscopic view even though no individual customer ever \nvisits any node more than once. In order for this view to be realistic, \neach individual customer should have a relatively negligible effect on \nthe total network.\n\nThe procedure here is to solve for the equilibrium or macroscopic \nbehavior of the network first and then afterwards consider particular \ncustomers. The particular customers will have their own routes \nthrough the network and perhaps their own service times at the nodes \nalong the way. There are two cases, depending on whether the input \nis by classes and routes, as in Section 2.3 or the standard input as in \nSection 2.1.\n\nFirst, suppose that we are using the input by classes and routes in \nSection 2.3. Then the particular customers correspond to the customer \nclasses specified in the input. Hence, each customer has a deterministic \nroute through the network and possibly special service times at the \nnodes on the route. In this case, as described in Section 2.3, QNA first \nconverts the input by classes and routes into the standard input in \nSection 2.1. Then QNA solves for the equilibrium behavior. Finally, \ncongestion measures are calculated for the different classes under the \nassumption that they follow their originally specified special routes \nand that upon arrival at the nodes on the route they see independent \nversions of the equilibrium state of the network. Hence, in the notation \nof Section 2.3, for a customer in class k, the expected total service \ntime is\n\nand the expected total sojourn time or response time is the sum of \n(85) and (86). Similarly, for a customer in class k the variance of the \ntotal service time is\n\nWith the standard input in Section 2.1, the user must specify the \nparticular customers to be analyzed. In this case, the user specifies \nclasses with routes and possibly service times (rate and variability \nparameters), but these data are not used in calculating the equilibrium \nbehavior. The decoupling principle is used with greater force here; \nthere need not be any consistency between the microscopic and mac- \nroscopic views: This additional input does not affect the equilibrium \nbehavior of the total network.\n\nIn the current version of QNA the individual customer routes are \ndeterministic, so that the additional input required is just as in Section \n2.3 and the congestion measures are just as in (85) through (88) in \nSection 6.3.1. However, it is possible to modify QNA to allow random \nroutes. Then the additional input would be just as in Section 2.1; for \neach class it would consist of a routing matrix plus parameters for the \narrivals process and service times.\n\nThe software package QNA was written by Anne Seery. She used a \nsubroutine for analyzing the M/M/m queue written by Shlomo Halfin. \nIt has been a pleasure collaborating with Anne Seery on this venture. \nI also appreciate the help from many other colleagues and the contin- \nued support of my management: W. A. Cornell, C. S. Dawson, J. C. \nLawson, C. J. McCallum, Jr., M. Segal, R. E. Thomas, and E. Wolman.\n\n. L. Kleinrock, Queueing Systems, Volume 2: Computer Applications, New York: John \nWiley and. Sons, 1976.\n\n. M. Schwartz, Computer- Communications Network Design and Analysis, Englewood \nCliffs: Pentice- Hall, 1977.\n\n. E. Gelenbe and I. Mitrani, Analysis and Synthesis of Computer Systems, New York: \nAcademic Press, 1980.\n\nC. H. Sauer and K. M. Chandy, ie asl Systems Performance Modeling, Engle- \nwood Cliffs: Prentice-Hall, 1981\n\nJ. G. Shanthikumar and J. A. Buzacott, \u201cOpen Queueing Network Models of \nDynamic Job Shops,\u201d Int. J. Prod. Res., 19, No. 3 (1981), pp. 255-66.\n\nR. L. Disney, Queueing Networks and Applications, Baltimore: The John Hopkins \nUniversity Lectures, 1982; to be published by the Johns Hopkins University\n\nPress. \n\u201cUser\u2019s Guide for BEST/1,\u201d BGS Systems, Inc., Waltham, Massachusetts, 1980. \n. \u201cUser\u2019s Manual for CADS,\u201d Austin, TX: Information Research Associates, 1978. \n. J. McKenna, D. Mitra, and K. G. Ramakrishnan, \u201cA Class of Closed Markovian \nQueueing Networks: Integral Representations, Asymptotic Expansions, and Gen- \neralizations,\u201d B.S.T.J., 60, No. 5 (May-June 1981), pp. 599-641.\n\n11. J. McKenna and D. Mitra, \u201cIntergral Representations and Asymptotic Expansions \nfor Closed Markovian Queueing Networks: Normal Usage,\u201d B.S.T.J., 61, No. 5 \n(May-June 1982), pp. 661-83.\n\n12. K. G. Ramakrishnan and D. Mitra, \u201cAn Overview of PANACEA, A Software \nPackage for Analyzing Markovian Queueing Networks,\u201d B.S.T.J., 61, No. 10, Part \n1 (December 1982), pp. 2849-72.\n\n13. H. Heffes, \u201cMoment Formulae for a Class of Mixed Multi-Job-Type Queueing \nNetworks,\u201d B.S.T.J., 61, No. 5 (May-June 1982), pp. 709-45.\n\n14. W. Whitt, \u201cApproximating a Point Process by a Renewal Process, I: Two Basic \nMethods,\u201d Oper. Res., 30, No. 1 (January-February 1982), pp. 125-47.\n\n15. S. L. Albin, Approximating Queues with Superposition Arrival Processes, Ph.D \ndissertation, Department of Industrial Engineering and Operations Research, \nColumbia University, 1981.\n\n16. S. L. Albin, \u201cApproximating a Point Process by a Renewal Process, II: Superposition \nArrival Processes to Queues,\u201d Department of Industrial Engineering, Rutgers \nUniversity, 1982.\n\n17. R. I. Wilkinson, \u201cTheories for Toll Traffic Engineering in the U.S.A.,\u201d B.S.T.J., 35, \nNo. 2 (March 1956), pp. 421-514.\n\n18. R. B. Cooper, Introduction to Queueing Theory, Second Edition, New York: North \nHolland, 1981.\n\n19. A. Kuczura, \u201cThe Interrupted Poisson Process as an Overflow Process,\u201d B.S.T.J., \n52, No. 3 (March 1973), pp. 437-48.\n\n20. J. H. Rath and D. D. Sheng, \u201cApproximations for Overflows from Queues with a \nEe pasting Room,\u201d Oper. Res., 27, No. 6 (November\u2014December 1979), pp. \n1208-16.\n\n21. M. Reiser and H. Kobayashi, \u201cAccuracy of the Diffusion Approximation for Some \nQueueing Systems,\u201d IBM J. Res. Dev., 18 (March 1974), pp. 110-24.\n\n22. P. J. Kuehn, \u201cApproximate Analysis of General Queueing Networks by Decompo- \nsition,\u201d IEEE Trans. Commun., COM-27, No. 1 (January 1979), pp. 113-26.\n\n23. K. C. Sevcik, A. I. Levy, S. K. Tripathi, and J. L. Zahorjan, \u201cImproving Approxi- \nmations of Aggregated Queueing Network Subsystems,\u201d in Computer Perform- \ned K. M. Chandy and M. Reiser (eds.), Amsterdam: North Holland, 1977, pp. \n1-22.\n\n24. K. M. Chandy and C. H. Sauer, \u201cApproximate Methods for Analyzing Queueing \nNetwork Models of Computing Systems,\u201d ACM Computing Surveys, 10, No. 3 \n(September 1978), pp. 281-317.\n\n25. S. Katz, \u201cStatistical Performance Analysis of a Switched Communications Net- \nvere: Fifth Int. Teletraffic Cong., Rockefeller University, New York, 1967, pp. \n566-75.\n\n26. D. L. Iglehart and W. Whitt, \u201cMultiple Channel Queues in Heavy Traffic, II: \nSequences, Networks and Batches,\u201d Adv. Appl. Prob., 2, No. 2 (Autumn 1970), \npp. 355-69.\n\n27. J. M. Harrison and M. I. Reiman, \u201cOn the Distribution of Multidimensional \nReflected Brownian Motion,\u201d SIAM J. Appl. Math., 41, No. 2 (October 1981), pp. \n345-61.\n\n29. M. I. Reiman, \u201cThe Heavy Traffic Diffusion Approximation for Sojourn Times in \nJackson Networks,\u201d Applied Probability and Computer Science\u2014The Interface, \nVolume 2, R. L. Disney and T. J. Ott (eds.), Boston: Birkhauser, 1982, pp. 409-\n\n30. J. M. Holtzman, \u201cThe Accuracy of the Equivalent Random Method with Renewal \nInputs,\u201d B.S.T.J., 52, No. 9 (November 1973), pp. 1673-9.\n\n32. A. E. Eckberg, Jr., \u201cSharp Bounds on Laplace-Stieltjes Transforms, with Applica- \nioe e Various Queueing Problems,\u201d Math. Oper. Res., 2, No. 2 (May 1977), pp.\n\n33. W. Whitt, \u201cOn Approximations for Queues, I: Extremal Distributions,\u201d B.S.TWJ., \n63, No. 1, Part 1 (January 1984), to be published.\n\n34. J. G. Klincewicz and W. Whitt, \u201cOn Approximations for Queues, II: Shape Con- \nstraints,\u201d B.S.T.J., 63, No. 1, Part 1 (January 1984).\n\n35. W. Whitt, \u201cOn Approximations for Queues, III: Mixtures of Exponential Distribu- \ntions,\u201d B.S.T.J., 63, No. 1, Part 1 (January 1984).\n\n36. W. Whitt, \u201cThe Marshall and Stoyan Bounds for IMRL/G/1 Queues are Tight,\u201d \nOper. Res. Letters, 1, No. 6 (December 1982), pp. 209-13.\n\nW. Whitt, \u201cRefining Diffusion Approximations for Queues,\u201d Oper. Res. Letters, /, \nNo. 5 (November 1982), pp. 165-9.\n\nS. L. Albin, \u201cOn Poisson Approximations for Superposition Arrival Processes in \nQueues,\u201d Management Sci., 28, No. 2 (February 1982), 126-37.\n\nW. Whitt, \u201cQueues with Superposition Arrival Processes in Heavy Traffic,\u201d unpub- \nlished work, 1982.\n\nD. P. Heyman and M. J. Sobel, Stochastic Models in Operations Research, Vol. I, \nNew York: McGraw-Hill, 1982.\n\nP. Franken, D. Konig, U. Arndt, and V. Schmidt, Queues and Point Processes, \nBerlin: Akademie-Verlag, 1981.\n\nW. Whitt, \u201cApproximations for Departure Processes and Queues in Seri\u00e9s,\u201d Nav. \nRes. Log Qtr., to be published.\n\nA. A. Fredericks, \u201cA Class of Approximations for the Waiting Time Distribution in \na GI/G/1 Queueing System,\u201d B.S.T.J., 61, No. 3 (March 1982), pp. 295-325.\n\nY. Takahashi and Y. Takami, \u201cA Numerical Method for the Steady- State Proba- \nbilities of a GI/G/c Queueing System in a General Class,\u201d J. Oper. Res. Soc. \nJapan, 19 (1976), pp. 147-57.\n\nH. C. Tijms, M. H. van Hoorn, and A. Federgruen, \u201cApproximations for the Steady- \nState Probabilities in the M/G/c Queue,\u201d Adv. Appl. Prob., 73, No. 1 (March \n1981), pp. 186-206.\n\nH. Groenevelt, M. H. van Hoorn, and H. C. Tijms, \u201cTables for M/G/c Queueing \nSystems with Phase-Type Service,\u201d Report No. 85, Department of Actuarial \noe and Econometrics, The Free University, Amsterdam, The Netherlands, \n1982.\n\nM. F. Neuts, Matrix-Geometric Solutions in Stochastic Models\u2014An Algorithmic \nApproach, Baltimore: The Johns Hopkins University Press, 1981.\n\nM. F. Neuts, \u201cA Program for Analyzing the M/PH/c Queue,\u201d Department of \nMathematics, University of Delaware, 1981.\n\nP. Hokstad, \u201cSome Numerical Results and Approximations for the Many Server \nQueue with Nonexponential Service Time,\u201d Department of Mathematics, Uni- \nversity of Trondheim, Norway, 1981.\n\n. A. de Smit, \u201cA Program for Analyzing the GI/H,/s Queue,\u201d Department of \nApplied Mathematics, Twente University of Technology, Enschede, The Neth- \nerlands, 1982.\n\nA. Ishikawa, \u201cOn the Equilibrium Solution for the Queueing System: GI/E,/m,\u201d \nT.R.U. Mathematics, 15, No. 1 (1979), pp. 47-66.\n\nS. Halfin and W. Whitt, \u201cHeavy-Traffic Limits for Queues with Many Exponential \nServers,\u201d Oper. Res., 29, No. 3 (May-June 1981), pp. 567-88.\n\nD. Y. Burman and D. R. Smith,\u201d A Light-Traffic Theorem for Multiserver Queues,\u201d \nMath. Oper. Res., 8, No. 1 (February 1983), pp. 15-25.\n\nW. Kraemer and M. Langenbach-Belz, \u201cApproximate Formulae for the Delay in \nthe Queueing System GI/G/1,\u201d Congressbook, Eighth. Int. Teletraffic Cong., \nMelbourne, 1976, pp. 235-1/8.\n\nJ. G. Shanthikumar and J. A. Buzacott, \u201cOn the Approximations to the Single \nServer Queue,\u201d Int. J. Prod. Res. 18, No. 6 (1980), pp. 761-73.\n\nJ. Kollerstr\u00e9m, \u201cHeavy Traffic Theory for Queues with Several Servers. I,\u201d J. Appl. \nProb., 11, No. 3 (September 1974), pp. 544-52.\n\nJ. Kollerstrom, \u201cHeavy Traffic Theory for Queues with Several Servers. II,\u201d J. Appl. \nProb., 16, No. 2 (June 1979), pp. 393-401.\n\nA. M. Lee and P. A. Longton, \u201cQueueing Processes Associated with Airline Passen- \nger Check-In,\u201d Oper. Res. Quart., 10, No. 1 (March 1959), pp. 56-71.\n\nS. A. Nozaki and S. M. Ross, \u201cApproximations in Finite Capacity Multi-Server \nQueues with Poisson Arrivals,\u201d J. Appl. Prob., 15 (1978), pp. 826-34.\n\n69. B. Simon and R. D. Foley, \u201cSome Results on Sojourn Times in Acyclic Jackson \nNetworks,\u201d Management Sci., 25, No. 10 (October 1979), pp. 1027-34.\n\n70. P. C. Kiessler, \u201cA Simulation Analysis of Sojourn Times in a Jackson Network,\u201d \nReport VTR 8016, Department of Industrial Engineering and Operations Re- \nsearch, Virginia Polytechnic Institute and State University, 1980.\n\nThis paper describes the performance of the Queueing Network Analyzer \n(QNA), a software package developed at Bell Laboratories to calculate ap- \nproximate congestion measures for networks of queues. QNA is compared with \nsimulations and other approximations of several open networks of single- \nserver queues. This paper illustrates how to apply QNA and indicates the \nquality that can be expected from the approximations. The examples here \ndemonstrate the importance of the variability parameters used in QNA to \ndescribe non-Poisson arrival processes and nonexponential service-time dis- \ntributions. For these examples, QNA performs much better than the standard \nMarkovian algorithm, which does not use variability parameters. The accuracy \nof the QNA results (e.g., the expected delays) in these examples is satisfactory \nfor engineering purposes.\n\nThis paper is a sequel to Whitt,\u2019 which described the software \npackage called the Queueing Network Analyzer (QNA). QNA calcu- \nlates approximate congestion measures for networks of queues. The \nfirst version of QNA treats open networks of multiserver queues with \nthe first-come, first-served discipline and no capacity constraints. \nQNA is designed to treat non-Markovian models: The arrival processes \nneed not be Poisson and the service-time distributions need not be \nexponential. QNA approximately characterizes other kinds of varia- \nbility through variability parameters assigned to each arrival process \nand each service-time distribution. The first step in the algorithm is\n\nto solve for the flow rates and the variability parameters of the internal \narrival processes. The second step is to compute approximate conges- \ntion measures for each queue separately by regarding it as a standard \nGI/G/m queue in which the renewal arrival process and the service- \ntime distribution are each partially characterized by their first two \nmoments or, equivalently, the rate and variability parameters. The \nthird and final step is to calculate congestion measures for the network \nas a whole.\n\nThis paper describes the performance of QNA by comparing it with \nsimulations and other approximations of networks of queues. Even \nthough QNA can analyze multiserver queues, only single-server queues \nare considered here. Among the other approximations in each case are \nthe M/M/1 and M/G/1 approximations, which can be obtained from \nQNA by using default options. The M/M/1 approximation, which is \nembodied in the Markovian algorithms, is obtained by setting all \nvariability parameters equal to 1. With the M/M/1 approximation, \nthe nodes are treated as independent M/M/1 queues with the correct \nrates. The M/M/1 approximation yields the exact equilibrium distri- \nbution of queue lengths for the Markov model with Poisson external \narrival processes, exponential service-time distributions and one cus- \ntomer class. The M/G/1 approximation is obtained by setting the \nvariability parameter of each arrival process equal to 1 and using the \nspecified service-time variability parameter c2; then the expected wait- \ning time at each node is (1 + c3)/2 times the M/M/1 value.\n\nThe congestion measures we consider in the examples here are the \nexpected waiting time (before beginning service) and the expected \nsojourn time (waiting time plus service time) at a node or in the entire \nnetwork. Of course, QNA produces other congestion measures, but we \nare comparing with previously published simulation results, which are \nmostly limited to expected waiting times and sojourn times.\n\nWe begin in Section II with a single GI/G/1 queue and discuss the \nimplications of previous work on approximations for the GI/G/1 \nqueue.\u201d\u201d In Section III we consider a single queue with a superposition \narrival process and compare QNA with simulations by Albin.\u00ae\u201d\u00b0 In \nSection IV we consider a network of eight queues in series analyzed \nby Fraker,'\u2019 and in Sections V and VI we consider two networks \nanalyzed by Kuehn:\u2019\u201d Section V treats a tightly coupled two-node \nnetwork and Section VI treats a nine-node network. In Section VII \nwe treat a five-node network used to model a Bell Laboratories \n\u2018computer system. Finally, in Section VIII we consider a model from \nGelenbe and Mitrani?\u00ae for a packet-switched communication network. \nThe examples in Sections VII and VIII have input by classes and \nroutes as in Section 2.3 of Whitt.\u2019\n\nexpected in applications of QNA. They also demonstrate the impor- \ntance of the variability parameters when the external arrival processes \nare not nearly Poisson or the service-time distributions are not nearly \nexponential. These examples also illustrate how to apply QNA, e.g., \nto model superposition arrival processes (Section III), to eliminate \nalmost immediate feedback (Section V), and to conduct sensitivity \nanalyses for the variability (Section VII).\n\nWe begin by considering the special network containing a single \nservice facility, in particular, the GI/G/1 queue with service times and \ninterarrival times each partially characterized by their first two mo- \nments or, equivalently, by the four parameters 7 (the mean service \ntime), c2 (the squared coefficient of variation of the service time), \n(the arrival rate), and c?2 (the squared coefficient of variation of the \ninterarrival time).! The subscript indexing the node is suppressed \nsince there is only one node.\n\nIt is useful to consider this model because it has been extensively \nstudied and is relatively well understood. In many cases we can \nanalytically determine the quality of the approximations for the \nGI/G/1 queue. Hence, we can get an idea about the quality of the \napproximations for more general networks. Of course, approximations \nfor a node in a general network might be worse because the internal \narrival processes usually are not actually renewal processes. On the \nother hand, the network operations of superposition and splitting tend \nto make stochastic point processes more like Poisson processes, so \nthat larger networks may actually be better behaved.\n\nThe model specification above determines the approximate conges- \ntion measures produced by QNA, but the model specification is not \ncomplete since there are many service-time and interarrival-time \ndistributions with the given parameters. The formulas produced by \nthe QNA are approximations for all these systems, so it is natural to \nask how the approximate congestion measures compare to the set of \nall possible values that are consistent with the partial specification. \nFortunately, it is often possible to identify the set of all possible \nvalues.?\u00b0 Moreover, it is often possible to locate the more likely values \nby identifying the set of all possible values under various natural \nconstraints on the distribution. When this cannot be done exactly, it \ncan often be done approximately using bounds.\u00ae\n\nWe now give a brief summary of results evaluating approximations \nfor the expected waiting time or, equivalently (by Little\u2019s formula\u201d), \nthe expected queue length in the GI/G/1 queue based on the four \nparameters X, c+, 7, and c2. First, recall that for the M/G/1 queue, \nwith Poisson arrival process (c2 = 1), the expected waiting time\n\nactually depends on the service-time distribution only through the two \nparameters 7 and c2. Nonexponential interarrival-time distributions \ntend to be more difficult, however. For the GI/M/1 queue with an \nexponential service-time distribution (c2 = 1), the expected waiting \ntime depends on the interarrival-time distribution beyond the param- \neters \\ and c2. Given ) and c?2 in the GI/M/1 queue, the maximum \nrelative error (upper bound minus lower bound divided by lower bound) \nin the mean queue length (number in system including anyone in \nservice) is exactly c2 (see Ref. 2). A similar, but somewhat less concise, \nresult holds for the expected waiting time by virtue of Little\u2019s formula. \nThe maximum relative error for the expected sojourn time (waiting \ntime plus service time) is also c2. This result suggests more generally \nthat the reliability of the approximations might decrease when c? \nincreases, which is consistent with numerical experience.\n\nIf we assume that the interarrival-time distribution is not too \nirregular, then the maximum relative error becomes much less. In the \nH,/M/1 queue with a hyperexponential interarrival-time distribution \n(mixture of exponential distributions having c2 > 1), the maximum \nrelative error in the mean queue length is (c2 \u2014 1)/2 (see Ref. 4). \nIt turns out that the extremal interarrival-time distributions for \nH,/M/1 queues also are extremal for all interarrival-time distributions \nthat have Increasing Mean Residual Life (IMRL) (also c2? > 1) and all \nservice-time distributions,\u2019 so that the maximum relative error in the \nmean queue length for IMRL/G/1 queues is (c2 \u2014 1)/[2 + p(c2 \u2014 1)].\n\nOther kinds of shape constraints for the GI/M/1 queue have been \ninvestigated by means of nonlinear programming.\u2019 In general, we \nconclude that if the distributions are not irregular, then the maximum \nrelative error in the GI/G/1 queue might be about 0.05 c2, e.g., about \n10 percent when c2 = 2.0.\n\nFrom heavy-traffic limit theorems that describe the queue as p > \n1,\u00b0 where p = dz is the traffic intensity, we know that asymptotically \nthe queue length and waiting-time distributions depend on the inter- \narrival-time and service-time distributions only through the four pa- \nrameters X, c2, 7, and c2. This suggests that more generally the quality \nof the approximations might improve as p increases. This is certainly \nconsistent with experience for the GI/G/1 queue, but not necessarily \nfor more complex networks, e.g., the tightly coupled two-node network \nhere in Section V.\n\nThe heavy-traffic limit theorems are closely related to diffusion \napproximations because diffusion processes emerge as limits in the \nheavy-traffic limit theorems. We have recently compared various \ndiffusion approximations for the expected waiting time in a GI/G/1 \nqueue to known bounds.\u00ae We now show how QNA and other related \napproximations fit into this framework. Table I here compares four\n\nTable |\u2014Bounds and approximations for the expected waiting time, \nEW, in a GI/G/1 queue: Three cases\n\nNotes: 1. In each case the mean service time is 7 = 1. \n2. \u201cH\u201d indicates high (greater than or equal to) and \u201cL\u201d indicates low (less than \nor equal to) in comparison with the bounds.\n\napproximations for the expected waiting time, EW, with various upper \nand lower bounds in the three cases in Table 1 of Ref. 6. The four \napproximations are the M/M/1, M/G/1, Kraemer and Langenbach- \nBelz,\u2019 and QNA.\n\nThe M/M/1 approximation is obtained by replacing both variability \nparameters c% and c2 by 1. The M/M/1 approximation is produced by \na direct application of the Markovian software packages. The M/G/1 \napproximation is obtained by replacing c2 by 1 and using the specified \nvalue of c2. The M/G/1 approximation EW is the exact value for the \napproximating M/G/1 system since EW depends on the service-time \ndistribution only through its first two moments. Both the M/M/1 and \nM/G/1 approximations are produced by QNA using default options.\n\nThe QNA approximation is the Kraemer and Langenbach-Belz \napproximation when c3 < 1 and is slightly greater when c2 > 1.' The \none case in which c2 > 1 in Table I shows that the difference between \nthe two approximations is small compared to the distance between the \nupper and lower Monotone Failure Rate (MFR) bounds.\u00ae> The MFR \nbounds are for interarrival-time distributions with decreasing failure \nrate when c2 = 1 and increasing failure rate when c2 < 1. The MFR \nbounds are tight when c2 > 1 but not when c2 < 1 (see Ref. 5). Since \nthe interarrival-time distribution need not have monotone failure rate, \nthe range of all possible values is greater, but the MFR bounds indicate \nthe more likely values.\n\nFor the three cases in Table J, the M/M/1 approximation performs \nvery poorly, falling way outside the bounds. The M/G/1 approximation \nalways coincides with one of the MFR bounds\u2014the upper bound when \nc2 < 1 and the lower bound when c2 = 1\u2014but it would be better to \nhave an approximation somewhere in the middle between the bounds. \nWhen c2 = 1, the QNA approximation is a convex combination of the\n\ntwo MFR bounds.\u00b0 Since the MFR bounds are tight when c2 = 1, the \nQNA approximation always yields the exact value of EW for some \nGI/G/1 system with the given parameters.\u2019 When c2 < 1, the QNA \napproximation is slightly less than the convex combination of the \nMFR bounds. The convex combination of the MFR bounds is known \nto be an upper bound for E,/G/1 systems,\u201d so that it is appropriate to \nuse a smaller value. We conjecture that the QNA approximation always \nyields an exact value of EW for some GI/G/1 system with the given \nparameters when c2 < 1 too.\n\nTable I provides a sample of the comparisons possible using the \nprevious studies.\u201d \u00a9 Since QNA coincides with the Kraemer and Lan- \ngenbach-Belz approximation when c2 < 1 and the Sakasegawa-Yu \napproximation when c2 = 1, previous comparisons such as Tables 13 \nand 14 in Klincewicz and Whitt? also apply to QNA.\n\nIn this section we consider one single-server queue with a superpo- \nsition arrival process. Such a system with two component arrival \nprocesses is depicted in Fig. 1a. Since only one external arrival process \nat each node is allowed in QNA, this model cannot be analyzed directly, \nbut it is easy to modify the model so that QNA does apply. We added \ndummy nodes with very low traffic intensity on each component arrival \nprocess, as shown in Fig. 1b. Since the new dummy nodes have low\n\nFig. 1\u2014(a) The original queue with a superposition arrival process. (b) The equivalent \nnetwork with one external arrival process at each node.\n\ntraffic intensity, the rate and variability parameters of the departure \nprocesses from the dummy nodes will be almost identical to the \ncorresponding parameters of the external arrival processes.\n\nThis model has recently been studied quite extensively by Albin\u00ae\u201d \nand is now relatively well understood. As in Section I, here we consider \none illustrative example, which suggests the accuracy to expect more \ngenerally and shows the importance of the variability parameters.\n\nThe specific model we analyze is the 2GI;/M/1 system with an \nexponential service-time distribution and n iid stationary renewal \nprocesses as component arrival processes. The total arrival rate is 1, \nso the rate of each component process is n~'. We consider six cases \ninvolving two values of n, n = 2 and 16, and three values of the traffic \nintensity p, p = 0.3, 0.7, and 0.9. In each case the component renewal- \ninterval distribution is H2, i.e., the mixture of two exponentials with \nbalanced means: one with mean m, realized with probability p and the \nother with mean mz, realized with probability 1 \u2014 p, where pm, = \n(1 \u2014 p)mg. In each case the squared coefficient of variation of the \ncomponent renewal process interval is c? = 6. This is quite high \nvariability, so that the component processes are not nearly Poisson. \nThe three parameters of an Hp, distribution are determined by speci- \nfying the mean as n, c? = 6 and balanced means.\n\nThis model is taken from Chapter 3 and Appendix 6 of Albin.\u00ae \nAlbin\u2019s approximations are based on the rate and variability parame- \nters just as in QNA. In fact, the superposition approximation in QNA \nis a modification of Albin\u2019s procedure.' With Albin\u2019s procedure, the \nvariability parameter of the superposition process is a convex combi- \nnation of the variability parameters obtained from the stationary- \ninterval method and asymptotic method described in Whitt. The \nspecific implementation of the stationary-interval method in Whitt!\u201d \nand Albin\u00ae is part of Kuehn\u2019s\u201d algorithm for approximating networks \nof queues. Since Kuehn\u2019s implementation of the stationary-interval \nmethod is nonlinear, a different procedure is used in QNA. In QNA \nthe variability parameter of the superposition process is a convex \ncombination of the variability parameters obtained from the asymp- \ntotic method and a Poisson process.\u2019 Extensive experimentation has \nshown, however, that the approximation in QNA is very close to \nAlbin\u2019s hybrid approximation, which performed very well in many \nexperiments (about 3-percent average absolute relative percent error).\n\nSeveral different approximations for the expected waiting time are \ncompared with simulation in Table II. The simulation results were \nobtained in Albin.*\u00ae The sample standard deviation is given in paren- \ntheses below the simulation estimate in Table II to indicate the \nstatistical reliability of the estimate. The simulation program was \nwritten in FORTRAN using the \u201cSuper-Duper\u201d subprogram in Mar-\n\nTable {I\u2014A comparison of approximations and simulation of the \nexpected waiting time, EW, in a 2GI/M/1 queue with a superposition \narrival process\n\nNotes: 1. The total arrival rate is 1 in each case. \nThe component renewal processes have H\u2019 (hyperexponential) renewal-inter- \nval distributions with mean n, c? = 6, and balanced means.\n\n. The sample standard deviations appear below the simulation estimates in \nparentheses.\n\n: he relative percent error appears below the approximation values in paren- \ntheses.\n\nsaglia et al. to generate uniform random numbers.\u2019* A different random \nnumber seed was used for each simulation. The simulations began \nwith an empty system, but the first 1000 customers were not counted \nto allow the system to approach steady-state. Each simulation con- \nsisted of 20 batches, with the number of customers per batch depending \non the traffic intensity: 3,000 per batch for p = 0.3, 15,000 per batch \nfor p = 0.7, and 50,000 per batch for p = 0.9. Even though much more \nsimulation time was spent on the cases with higher traffic intensities, \nthe statistical reliability was slightly less.\n\nIn Table II, in parentheses below the approximation values are the \nrelative percent errors (RE), which are defined as\n\nAt the bottom of Table II are the average absolute relative percent \nerrors (ARE), which are defined as\n\n6 \nARE = YY |RE;\\/6. (2) \ni=l \nDividing by the simulation value perhaps inflates the errors when p =\n\nThe stationary-interval method and the M/M/1 approximation are \nnot bad for large n and small p because the superposition process \nconverages to a Poisson process as n \u2014 \u00a9 and the queue reflects this \nif p is not too big, in particular, if n(1 \u2014 p)? is sufficiently large.\" \nHowever, for p = 0.9, these two methods perform poorly. On the other \nhand, the asymptotic method performs reasonably well for p = 0.9, \nbut not well in other cases. In particular, the asymptotic method does \nnot reflect the convergence to a Poisson process as n \u2014 %; it gives the \nsame answers for n = 2 and 16. So, for fixed p, the asymptotic method \ngets worse as n increases.\n\nAs Albin determined in extensive experiments,\u00b0*\u201d her hybrid approx- \nimation is much better than either basic method alone. This example \nalso illustrates how close QNA is to Albin\u2019s hybrid procedure. For \nqueues with superposition arrival processes, we conclude that QNA \nusually gives reasonable results and strongly dominates the two basic \nmethods.\n\nIn closing this section, we add a caveat. The component processes \nin the simulation for Table II, in Albin\u2019s hybrid procedure and in \nQNA, are all based on the case of balanced means. However, as \ndiscussed in Whitt,*\u00b0 given the first two moments, the one-parameter \nfamily of renewal processes with hyperexponential renewal-interval \ndistributions range from a Poisson process to a batch Poisson process \nwith geometrically distributed batch size. For a single renewal arrival \nprocess, the expected waiting time, EW, in an H2/G/1 queue also \nranges between these same extremes, i.e., the M/G/1 and the M8/G/1 \nsystems. Unfortunately, no related theory yet exists for superposition \narrival processes. However, we can easily describe what happens in \nthe two extremes. The superposition of independent Poisson processes \nis Poisson and the superposition of independent batch Poisson proc- \nesses with geometrically distributed batches having a common mean \nis batch Poisson with geometrically distributed batches. Thus, we \nconjecture that the maximum and minimum values for EW with a \nsuperposition of iid Hj-renewal processes correspond to the M\u00ae/G/1 \nand M/G/1 systems, respectively. If the conjecture is true, then we \nwould have the same range of possible values for H\u00bb-superposition \narrival processes as for H2-renewal processes. However, if the compo- \nnent processes are not too batchy, then the superposition process will \nbecome more Poisson as n increases. We should usually expect the \nsuperposition process to be more nearly Poisson as n increases. To \nsummarize, this heuristic analysis suggests that the range of possible \nvalues for / W in the 2GI,;/G/1 queue given the basic parameters }, \nc2, T, c2 may be about the same as for the GI/G/1 queue. In fact, the\n\nsuperposition operation may actually make the arrival process better \nbehaved.\n\nIn this section we apply QNA to a network of eight single-server \nqueues in series previously analyzed by Fraker.\u2019! The external arrival \nprocess is Poisson and all service-time distributions are Erlang. Fraker \nconsidered eight cases involving four traffic intensities (9 = 0.3, 0.5, \n0.7, and 0.9) and four Erlang service-time distributions (M = F,, Eu, \nEs, and D = E..). Each of the traffic intensities and each of the service- \ntime distributions are assigned randomly to two of the eight nodes. \nFraker developed an approximation for these systems and compared \nit with simulations.\n\nTables III and IV describe Fraker\u2019s first two cases and the approx- \nimations for the expected waiting time at each node. The service-time \nsquared coefficient of variation specifies the Erlang distribution since \nc? = k7! for E,. Fraker made three simulation runs of 2500 customers, \ndiscarding the first 500 in each case to damp out the transient effects \nof starting the simulation. Statistics were collected for six blocks of \n1000 customers each. Unfortunately, this is not enough to produce \nvery good accuracy, especially for the nodes with higher traffic inten- \nsities. (Compare with the simulation length in Section III.) The \nstatistical reliability can be seen from the results of the six runs \ndisplayed in Fraker.'! (These also appear in Appendix 1 of Whitt.'\u00ae) \nAn idea of the variability can also be seen from node 1 because all the \napproximations except the M/M/1 approximation are exact for node \n1. When p = 0.9 the length of a 95-percent confidence interval \napproximately equals the estimated value; when p = 0.7 the length of\n\nTable II\u2014A comparison of approximations and simulation of the \nexpected waiting time at each node in Fraker\u2019s model of eight \nqueues in series: Case 1\n\nApproximation Methods \nSquared ee ee \nCoeffi- (Marko- (Asymp-  (Lag-1 QNA \nTraffic cient of vian Net- _ totic Corre- \u2014\u2014\u2014\u2014\u2014\u2014 \nNode Intensity, Variation, Simulated work) Method) lations) \nNo. pj eG Value M/M/1 M/G/1 Fraker EW; Cay\n\n8 0.3 1/4 0.00 0.13 0.08 0.01 0.01 0.33 \nNote: The arrival process is Poisson with rate 1 and the service-time distributions are\n\nTable IV\u2014A comparison of approximations and simulation of the \nexpected waiting time at each node in Fraker\u2019s model of eight \nqueues in series: Case 2\n\nTable V compares the approximations with simulation for the nodes \nwith traffic intensity p = 0.7 in all eight cases. Since the approxima- \ntions are exact for the first node, the first node is not included for the \ncases in which p,; = 0.7 (Cases 1, 5, and 6). For the approximations, \nthe difference between the approximation value and the simulation \nvalue is displayed.\n\nTables III through V show that QNA performs about the same as \nFraker\u2019s approximation, which is based on lag-1 correlations and is \nespecially designed for queues with Erlang service times. Both these \napproximations performed significantly better than the M/G/1 ap- \nproximation, which in turn performs significantly better than the \nM/M/1 approximation.\n\nAdditional analysis of Fraker\u2019s models plus other queues in series is \ncontained in Whitt.\u2019 The performance of QNA in these other cases \nis consistent with the description here.\n\nIn this section we consider a two-node network analyzed by Kuehn?\u201d \nand Gelenbe and Mitrani.\u2019\u00ae This network is depicted in Fig. 2. It has \none external arrival process, which comes to node 1. Customers com- \npleting service at node 1 leave the system with probability 1/2; other- \nwise they go to node 2 and then back to node 1 to be served again. At \nnode 2 customers are immediately fed back to node 2 for another \nservice with probability q22, but in most cases qz2 = 0.\n\nTable V\u2014The expected waiting time at the nodes with p; = 0.7 in \nFraker\u2019s eight cases of eight single-server queues in series\n\nFig. 2\u2014Kuehn\u2019s first example: A network of two queues with one external arrival \nprocess.\n\nthree values of the external arrival rate for each case: 9, = 0.15, 0.30, \nand 0.45. In each case the mean service time at node 1 is 7; = 1 and \nthe transition probability from node 1 to node 2 is giz = 1/2, so that \nthe traffic intensity at node 1 is p: = 2A. For node 1 these three \nexternal arrival rates correspond roughly to light traffic (p; = 0.3), \nmoderate traffic (p; = 0.6), and heavy traffic (p; = 0.9). In Cases 2 and \n3 the traffic intensity at node 2 is the same as at node 1, 1.e., po = 2Xo1, \nbut in all other cases it is pe = Ao. In these other cases node 2 is \nalways in relatively light traffic. The external arrival process is always \na renewal process. Each interarrival-time distribution and service- \ntime distribution is one of four distributions: deterministic (D with c?\n\nThe results are described in Tables VII and VIII. The simulation \nresults and Kuehn\u2019s approximation are taken from Kuehn.\u201d Two \ndifferent approximation results are given for QNA in Table VII. The \nfirst column is the standard application of QNA with the network \nreconfigured to eliminate immediate feedback in the one case it occurs, \nat node 2 in Case 3. (See Section 3 of Ref. 1.) The final column of \nTable VII is an adjusted version of QNA to eliminate almost immediate \nfeedback, which we discuss below.\n\nFor this network the quality of the standard QNA approximation is \nabout the same as Kuehn\u2019s approximation. They both work well for \nlow and moderate traffic intensities, e.g., about 10-percent average \nabsolute relative percent error when p,; = 0.6 (Table VIII), but not so \nwell in heavy traffic. This two-node network presents an obvious \ndifficulty for QNA. The network is tightly coupled so that many \ndepartures from node 1 rapidly return to node 1 for additional service. \nHowever, since these returning customers first pass through node 2, \nthere is no immediate feedback, so that QNA does not reconfigure the \nnetwork to eliminate the feedback. Nevertheless, it is evident that this \nalmost immediate feedback for node 1 is very similar to immediate \nfeedback and the potential exists for better results by reconfiguring \nthe network to eliminate this feedback too.\n\nIn all cases except 2 and 3, almost immediate feedback is eliminated \nby applying the standard version of QNA with immediate feedback \nelimination twice. The first time we apply QNA to the full network \nand the second time we apply QNA with node 2 removed. When node\n\nNotes: 1. In each case 7; = 1 and qy2 = 1/2. \n2. In each case the arrival rate assumes one of three values: Ao = 0.15, 0.30, and \n0.45,\n\nTable VII\u2014A comparison of approximations and simulation of the \nexpected total sojourn time (waiting time plus service time) in \nKuehn\u2019s two-node network\n\nSimulation \n(with 95-Per- Approximation Methods \nExternal cent Confi- \nSystem _ Arrival dence Inter- QNA \nNo. Rate, dor vals) M/M/1 M/G/1 Kuehn QNA _ Adjusted\n\nNotes: 1. In each case the traffic intensity at node 1 is p; = 2Ao1. \n2. In Cases 2 and 8 the traffic intensity at node 2 is pe = 2o1; otherwise it is \np2 = Xo.\n\n2 is removed, the feedback to node 1 becomes immediate and the \nnetwork is reconfigured by QNA to eliminate it. We use the second \nrun with node 2 removed to determine the expected waiting time per \nvisit at node 1. We use the first run to determine the expected number \nof visits to node 1 and the expected total sojourn time at node 2.\n\nWe do not treat nodes 1 and 2 symmetrically in Cases 1 and 4 \nthrough 8 because p; = 22 so that pe is relatively small compared to \npi. Customers that return to node 1 via node 2 will not be delayed long \nat node 2 before coming back, but customers returning to node 2 via \nnode 1 will be delayed relatively longer before coming back. If we had \nf1 < pz, we would remove node 1 in the second run of the QNA and \nfocus instead on node 2.\n\nIn Cases 2 and 3 the traffic intensities at nodes 1 and 2 are equal, \nso the motivation for eliminating almost immediate feedback is less. \nWhat we have done for Table IV is first calculate the congestion\n\nTable VIII\u2014A comparison of approximation methods in Kuehn\u2019s \ntwo-node network with Ao; = 0.3 (p; = 0.6): The average absolute \nrelative percent error in the expected total sojourn time compared\n\nmeasures for node 1 via the second run of QNA with node 2 removed \nas before. Then we use the results for node 1 to approximate the \nvariability parameter of the arrival process to node 2. Finally, we \nanalyze node 2 in isolation with the correct rates and this approximate \narrival variability parameter. This works slightly better than the first \nprocedure, but neither works well.\n\nThe results demonstrate that the standard version of QNA performs \nrelatively well in Cases 2 and 3 when p; = po. The adjustment to \neliminate almost immediate feedback yields a significant improvement \nwhen p; > pe, but the results after adjustment to eliminate almost \nimmediate feedback are much worse when p; = pz.\n\nAs a refined procedure for this two-node network, we suggest elim- \ninating almost immediate feedback at the node with higher traffic \nintensity when the traffic intensities differ significantly, and using the \nstandard QNA algorithm otherwise. The refined procedure in Table \nVIII is standard QNA in Cases 2 and 3 and the adjusted QNA in all \nother cases.\n\nTable VIII displays the relative percentage errors for all the approx- \nimations for the eight cases with Ao: = 0.3 (p: = 0.6). The refined \nprocedure in the last column yields very good results. Table VIII also \ndemonstrates that the standard version of QNA is significantly better \nthan the M/M/1 approximation, but not uniformly better. In some \ncases, e.g., in Case 8, errors in opposite directions can cancel for the \nM/M/1 approximation.\n\nThe improvement from eliminating almost immediate feedback \u201cby \nhand\u201d suggests that it would be desirable to develop an automatic \nprocedure for eliminating almost immediate feedback to incorporate\n\nin QNA, and this is being investigated. It also indicates the potential \nfor \u201ctuning\u201d QNA for particular applications.\n\nWe now consider Gelenbe and Mitrani\u2019s\u2019\u00ae experiment, which con- \nsists of five cases. The network is as depicted in Fig. 2 except the \nrouting probability qi2 is not exactly 1/2. The parameter values are \ngiven in Table [X and the results in Table X (pp. 137, 138 of Gelenbe \nand Mitrani\u2019*). We only include the best of three approximation \nschemes discussed by Gelenbe and Mitrani. Unfortunately, Gelenbe \nand Mitrani provided no information about the statistical reliability \nof the simulation estimates. Since p; and p2 are nearly equal in each \ncase, we did not try to eliminate almost immediate feedback.\n\nThe M/M/1 approximation values are obviously much too large \nbecause the M/M/1 approximation does not reflect the low variability \nof the service times. The M/G/1 approximation is much too low at \nnode 2 because it does not benefit from the feedback elimination \nprocedure. The Gelenbe-Pujolle procedure is better than the M/G/1 \nprocedure, but not uniformly so. The QNA approximation is clearly\n\n_ Table IX\u2014The parameter values for Gelenbe and Mitrani\u2019s \nexperiment with the two-node network in Fig. 2\n\nTable X\u2014A comparison of approximations with simulations: The \nexpected number of customers at each node in the two-node \nnetwork in Gelenbe and Mitrani'?\n\nthe best, but it may underestimate the congestion for high traffic \nintensities.\n\nWe now consider a nine-node network analyzed by Kuehn,\u2019 which \nis depicted in Fig. 3. The mean service time at node j is 7; = 1 for each \nj. There are three external arrival processes with \\o; = 0.5 for each j; \nthese come to nodes 1, 2, and 3. Kuehn let the three external arrival \nprocesses be Poisson processes. As in Section V, all service-time \ndistributions are D, E,, M, or H\u2019. Kuehn considered two cases: \nhomogeneous servers, in which all the service-time distributions are \nidentical, and heterogeneous servers, in which nodes 1 through 3 have \none service-time distribution and nodes 4 through 9 have another.\n\nKuehn compared his approximation with the M/M/1 and M/G/1 \napproximations and simulation for service-time variability parameters \nranging from c2, = 0 to 4. He focused on the expected total sojourn \ntime (waiting time plus service time) in the network and the expected \nsojourn time per visit in node 4. For this network Kuehn found, first, \nthat the service-time variability parameters are significant (the so- \njourn-time measures increase significantly with c2,); second, that his \napproximation tracks the simulation well; and, third, that the M/G/1 \napproximation also works well, but not quite as well as his approxi- \nmation.\n\nWe obtained similar results applying QNA. The QNA approxima- \ntion values are indistinguishable from the approximation values dis- \nplayed graphically by Kuehn, which are consistently within the sim-\n\nFig. 3\u2014Kuehn\u2019s second example: A network of nine queues with three Poisson \nexternal arrival processes.\n\nulation confidence intervals. In Table XI we display some of our \nresults. Here we consider only the case of homogeneous servers in \nwhich c?; = 0, 1, or 4. However, we let the variability parameter of all \nexternal arrival processes by cj; = 0.25, 1.0, or 4.0. We thus obtain \nnine cases.\n\nKuehn did simulations only in the case c\u00e9; = 1. The different values \nobtained by QNA when cj; # 1 suggest that the congestion measures \nin this model are more sensitive to the variability of the service times \nthan the variability of the interarrival times. Of course, we should \nexpect that the M/G/1 approximation might perform well when the \nvariability parameters of the external arrival processes are 1 or close \nto 1, but for relatively large and interconnected networks the M/G/1 \napproximation may perform well for other external arrival processes. \nIt performs reasonably well here when cj; = 0.25 or 4.0.\n\nIn this section we apply QNA to a network with input by customer \nclasses and routes as in Section 2.3 of Ref. 1. We compare QNA to a \nsimulation model used in the development of a computer system at \nBell Laboratories. In this model there are five nodes and two customer \nclasses.\n\nThe customer classes correspond to typical functions performed by \nthe system. The route for each class represents a typical sequence of \noperations performed by the system to process one of these functions.\n\nTable XI\u2014Approximations of expected sojourn times \nin Kuehn\u2019s nine-node network in Fig. 3: The case of \nhomogeneous servers, cz identical for all j\n\nIn the simulation model, both the routes and the service times at the \nnodes are deterministic for each class. Hence, the input for QNA is \njust as specified in Section 2.3 of Ref. 1 with the service-time variability \nparameters set equal to 0, i.e., c2,; = 0 for each class k and j on class \nk\u2019s route. The two routes we consider are given in Tables XII and\n\nXIII. Note that node 1 frequently appears several times in succession, \nso that there is immediate feedback at node 1. Also note that the \nservice times differ at different visits to the same node.\n\nThe customer classes arrive according to independent Poisson proc- \nesses. In the case we consider the arrival rates of Classes 1 and 2 are \n0.00015278 and 0.00030555, respectively.\n\nThe QNA, M/G/1, and M/M/1 approximations are compared with \nsimulation in Table XIV. The simulation values are the average of \nthree separate runs. The values from these separate runs are displayed \nto give an idea of the statistical reliability. The congestion measures \ncompared are the expected waiting times at the nodes and the expected \ntotal waiting time (excluding service time) on three route segments. \nThe first segment is the first 25 nodes of the second route; the second \nsegment is the first 21 nodes on the first route; and the third segment \nis eight nodes from node 23 to node 30 on the first route. In Table \nXIV the waiting times at the nodes are measured in milliseconds while \nthe waiting times on the route segments are measured in seconds.\n\nFrom Table XIV, it is apparent that QNA with immediate-feedback \nelimination performs reasonably well, significantly better than the \nM/M/1 and M/G/1 approximations. Since the approximating varia- \nbility parameters of the arrival processes are very close to 1, QNA \nwithout immediate-feedback elimination is very similar to the M/G/1 \napproximation. Hence, again we see that eliminating immediate feed-\n\nSection VII \nMean Mean Mean \nService Service Service \nNumber Node Time Number Node Time Number Node Time\n\nTable XIV\u2014A comparison of approximations and simulation of the \nexpected waiting times for the model of Section VII\n\nExpected Waiting Time at the Total Expected Waiting \nNodes Time on a Route Segment\n\nNotes: 1. The relative percent errors appear below the approximation in parentheses. \n2. The value ci, = 0.47 at node 1 is before adjustment for feedback; after\n\nadjustment it is 0.79. \n3. The units of measurement are milliseconds for the nodes and seconds for the \nroute segments.\n\nback helps. The M/M/1 approximations at nodes 3 and 4 evidently \nare too large because the service times are nearly constant. However, \nat node 1 the M/M/1 approximation does pretty well, apparently \nbecause two different errors cancel. (We have not displayed the relative \npercentage errors at node 2 since it seems of little consequence because \nof the low traffic intensity.)\n\nIt is interesting to know how the system would perform if the \nexternal arrival processes are not Poisson and if the service times at \nthe nodes on the routes are not deterministic. With QNA we can easily \nperform such sensitivity analyses. We can simply change the variabil- \nity parameters of the external arrival processes and the service times \non the routes. The results of such a study are given in Tables XV and \nXVI. Table XV gives the approximating variability parameter of the\n\narrival process at each node as a function of the external arrival \nprocess c? and the service time c\u2019. It is assumed that both external \narrival processes have the same c?\u201d and that all service times on both \nroutes have the same c*. From Table XV, it is evident that the \nvariability of the external arrival process hardly has any effect. Thus, \nQNA predicts that this model, and perhaps the system itself, will be \nrobust to changes in the variability of the arriving traffic. The varia- \nbility from outside is evidently dissipated on the long routes through \nthe network.\n\nTable XV\u2014The approximate variability parameter of the arrival \nprocess at each node determined by the QNA, as a function of the \ngiven variability parameters: The model of Section VII\n\nTable XVI\u2014The approximate service-time variability parameter cj \nand mean delay EW; at node j determined by the QNA, as a function \nof the variability of each service time on the route: The model of \nSection VII\n\nIn this section we consider a model of a packet-switched communi- \ncation network analyzed in Section 4.3.1 of Gelenbe and Mitrani.* \nThe basic model has 5 switching nodes and 12 one-way data links, as \ndepicted in Fig. 4. However, in this model each data link is a server \nand the packets waiting for transmission on the link form the queue. \nPackets are assumed to arrive at the switching nodes according to \nindependent Poisson processes. Each packet arriving from outside at \nnode i has final destination j with probability d;. Each packet with \ndestination j goes next to node rj from node i. Hence, there is a fixed \nroute for each origin-destination pair.\n\nWe analyze this network using the input by classes and routes in \nSection 2.3 of Ref. 1. However, unlike Section VII, here the service- \ntime parameters are associated with the nodes rather than the routes \n(which is an input option in QNA). The network of queues has 12 \nnodes with one server at each node and 20 routes. As specified by \nGelenbe and Mitrani, the service rate at nodes 1, 2, 7, 8, 11, and 12 is \n4,8 (in thousands of bits per second) and the service rate at the other \nnodes is 48. Since packet lengths are assumed constant, c2; = 0 for all \nj.\n\nFor this example the matrices D = (d,;) of destination probabilities \nand R = (r;) of next-node routes are:\n\nFig. 4\u2014Gelenbe and Mitrani\u2019s model of a packet-switched network: The 12 links are \nthe nodes in the network of queues.\n\nThe input data for the routes are given in Table XVII. These data \nare obtained from the matrices D and R plus the external arrival rates \nof 6.00, 8.25, 7.50, 6.75, and 1.50 at the five switching nodes (p. 141 of \nGelenbe and Mitrani\u201d*).\n\nThe results are compared with Gelenbe and Mitrani\u2019s approximation \nand simulation in Table XVIII. Since only 6000 packets reached their \ndestination in the simulation, the statistical reliability of the simula- \ntion estimates cannot be very good, cf. Section III. Our analysis is \nrevealing. First, Gelenbe and Mitrani describe their results as average \nbuffer queue lengths, which might be thought to exclude the customer \n(packet) being served (transmitted). However, the results obviously \ninclude the customer in service. Second, the two M/M/1 approxima- \ntions should agree, but they do not. Evidently, the arrival rates at \nswitches 1 and 2 were not actually as cited in the text.\u2019\u2019 Hence, the \nnumbers for the one heavily loaded link, link 2, cannot be meaningfully \ncompared.\n\nIt is useful to consider the expected number waiting excluding the \ncustomer in service. This is easily obtained because the probability \nthat the server is busy is exactly the traffic intensity, p (see Section \n11.3 of Heyman and Sobel?*). We thus obtain estimates of the expected \nnumber waiting by subtracting p from the numbers displayed in Table\n\nTable XVII\u2014The input data by routes for Gelenbe and \nMitrani\u2019s model of a packet-switched communication\n\nnetwork \nExternal Arrival Number \nOrigin-Des- Process Param- of Nodes \nRoute _tination _ eters on the Node \nNumber Pair Ar Ck Route Sequence\n\nXVIII. When this is done, some of the simulation estimates become \nnegative, demonstrating that the input parameters are incorrect or the \nstatistical reliability of the simulation is not very good. In this case, \nprobably both problems exist.\n\nWe also observe that the traffic intensities at all link queues but \nthe second are very small, so the numbers displayed in Table XVIII \nare mostly estimates of the traffic intensities themselves. Moreover, \nbecause the traffic intensities are small, the variability parameter of \nthe departure process produced by QNA will be very close to the \nvariability parameter of the arrival process (see Section 4.5 of Ref. 1). \nThis would not be true for the second link with traffic intensity 0.835, \nbut note that all departures from the second link leave the system. \nSince the external arrival processes are all Poisson, QNA should and \ndoes perform virtually the same as an M/G/1 approximation. In fact, \nsince the service times are all constant (cz; = 0), the approximation \nreduces to the M/D/1 system. Moreover, we predict that a proper \nsimulation of this model with the specified parameters will yield values \nvery close to the M/D/1 approximation.\n\nWe also display in Table XIX the point-to-point (origin-destination) \naverage total service times, delays, and sojourn times (service times \nplus delays) produced by QNA. No simulation values were available \nfor comparison, however. The output is useful to indicate unacceptably \nhigh or low values. It is also useful to determine the separate contri- \nbutions of service times and delays to sojourn times. Of course, in\n\nTable XVIII\u2014A comparison of approximations and simulation: The \nexpected number waiting and being served on each link in the \nGelenbe-and-Mitrani model of a packet-switched communication \nnetwork depicted in Fig. 4\n\nApproximation Methods \nM/M/1 \nGelenbe \nTraffic In- Simulation and via Gelenbe-\n\nTable XIX\u2014Average point-to-point service times, delays, and sojourn \ntimes for the Gelenbe-and-Mitrani model of a packet-switched \ncommunication network\n\nPoisson Arrivals c? = 1.0 Bursty Arrivals c? = 4.0 \nMean Total Mean Total \nMean Total Mean Total Sojourn Mean Total Sojourn \nRoute Service Time Delay on Time on Delay on Time on\n\nTable XIX delays play a significant role only for routes using the \nsecond link.\n\nWe conclude by remarking that the assumption of Poisson arrivals \nfor packets at each switch made by Gelenbe and Mitrani\u2019\u00ae often is not \nrealistic. Often messages containing many packets arrive according to \na Poisson process, but the packets arrive in a much more bursty \nmanner. Hence, it is appropriate to use QNA with arrival-process \nvariability parameters much larger than 1. The last two columns of \nTable XIX give the mean delays and sojourn times when the variability \nparameters of the external arrival processes are changed from c\u201d = 1.0 \nto c? = 4.0. When this is done here, the large delays on routes 4 and 8 \nincrease significantly. To a large extent, QNA was motivated by the \nneed to be able to systematically study the effect of such variability.\n\nI am grateful to Anne Seery for using QNA to generate much of the \ndata here, as well as for writing the QNA program. I am grateful to E. \nB. Zucker for the simulation data in Section VII.\n\n2. W. Whitt, \u201cOn Approximations for Queues, I: Extremal Distributions,\u201d B.S.TWJ., \n63, No. 1, Part 1 (January 1984).\n\n3. J. G. Klincewicz and W. Whitt, \u201cOn Approximations for Queues, II: Shape Con- \nstraints,\u201d B.S.T.J., 63, No. 1, Part 1 (January 1984).\n\n4, W. Whitt, \u201cOn Approximations for Queues, III: Mixtures of Exponential Distribu- \ntions,\u201d B.S.T.J., 63, No. 1, Part 1 (January 1984).\n\n5. W. Whitt, \u201cThe Marshall/Marshall and Stoyan Bounds for IMRL/G/1 Queues are \nTight,\u201d Oper. Res. Letters, 1, No. 6 (December 1982), pp. 209-13.\n\n6. W. Whitt, \u201cRefining Diffusion Approximations for Queues,\u201d Oper. Res. Letters, /, \nNo. 5 (November 1982), pp. 165-9.\n\n7. W. Kraemer and M. Langenbach-Belz, \u201cApproximate Formulae for the Delay in \nthe Queueing System GI/G/1,\u201d Congressbook, Eighth International Teletraffic \nCongress, Melbourne, Australia, 1976, pp. 235, 1-8.\n\n8. S. L. Albin, Approximating Queues with Superposition Arrival Processes, Ph.D \ndissertation, Department of Industrial Engineering and Operations Research, \nColumbia University, 1981.\n\n9. S. L. Albin, \u201cApproximating a Point Process by a Renewal Process, II: Superposition\n\nArrival Processes to Queues,\u201d Department of Industrial Engineering Rutgers \nUniversity, 1982.\n\n10. S. L. Albin, \u201cOn Poisson Approximations for Superposition Arrival Processes in \nQueues,\u201d Management Sci., 28, No. 2 (February 1982), pp. 126-37.\n\n11. J. R. Fraker, Approximate Techniques for the Analysis of Tandem Queueing Systems, \nPh.D. dissertation, Department of Industrial Engineering, Clemson University, \n1971.\n\n12. P. J. Kuehn, \u201cApproximate Analysis of General Queueing Networks by Decompo- \nsition,\u201d IEEE Trans. Commun., COM-27, No. 1 (January 1979), pp. 113-26.\n\n13. E. Gelenbe and I. Mitrani, Analysis and Synthesis of Computer Systems, New York: \nAcademic Press, 1980.\n\n14. D. P. Heyman and M. J. Sobel, Stochastic Models in Operations Research, Volume \nI, New York: McGraw-Hill, 1982.\n\n15. W. Whitt, \u201cApproximating a Point Process by a Renewal Process, I: Two Basic \nMethods,\u201d Oper. Res., 30, No. 1 (January-February 1982), pp. 125-47.\n\n16. G. Marsaglia, K. Ananthanarayanan, and N. Paul, \u201cRandom Number Generator \nPackage-Super Duper,\u201d School of Computer Science, McGill University, 1973.\n\n17. W. Whitt, \u201cQueues with Superposition Arrival Processes in Heavy Traffic,\u201d unpub- \nlished work, 1982.\n\n18. W. Whitt, \u201cApproximations for Departure Processes and Queues in Series,\u201d Navy \nRes. Log Qtrly., to be published.\n\nWard Whitt, A. B. (Mathematics), 1964, Dartmouth College; Ph.D. (Opera- \ntions Research), 1968, Cornell University; Stanford University, 1968-1969; \nYale University, 1969-1977; Bell Laboratories, 1977\u2014. At Yale University, \nfrom 1973-1977, Mr. Whitt was Associate Professor in the departments of \nAdministrative Sciences and Statistics. At Bell Laboratories he is in the \nOperations Research Department in the Network Analysis Center.\n\nStein A. H., An C. M., A Mean-Value Theorem for Zeta-Functions Associated \nWith Positive Definite Integral Forms. Mich Math J 30(1):3-8, 1983.\n\nAmitay N., Saleh A. A. M., Broad-Band Wide-Angle Quasi-Optical Polarization \nRotators. IEEE Antenn 31(1):73-76, 1983.\n\nBartoli P. D., CCITT Message Handling Facilities. Aust Tele R 16(3):53-62, 1982. \nBorsuk J. A., Light-Intensity Profiles of Surface-Emitting InGaAsP LEDs\u2014 \nImpact on Coupling to Optical Fibers. IEEE Device 30(4):296-303, 1983. \nCampbell J. C., Qua G. J., Dentai A. G., Optical Comparator\u2014A New Application \nfor Avalanche Phototransistors. JEEE Device 30(4):408-411, 1983.\n\nCapasso F., Tsang W. T., Williams G. F., Staircase Solid-State Photomultipliers \nand Avalanche Photo-Diodes With Enhanced [Ionization Rates Ratio. IEEE \nDevice 30(4):381-390, 1983.\n\nChin A. K., Zipfel C. L., Ermanis F., Marchut L., Camlibel I., Di Giuseppe M. A., Chin \nB. H., The Migration of Gold From the P-Contact as a Source of Dark Spot \nDefects in InP/InGaAsP LEDs. IEEE Device 30(4):304-310, 1983.\n\nDutta N. K., Current Injection in Multiquantum Well Lasers (Letter). IEEE J \nQ El 19(5):794-797, 1983.\n\nDutta N. K., Nelson R. J., Wright P. D., Besomi P., Wilson R. B., Optical-Properties \nof a 1.3-~4m InGaAsP Superluminescent Diode. IEEE Device 30(4):360-363, 1983. \nHeller A., High-Efficiency of Photo-Electrochemical Cells\u2014Reply (Let- \nter). Chem Eng N 61(22):4, 1983.\n\nJohannes V. I., Performance Parameters for Digital-Communications (Let- \nter). P IEEE 71(4):539, 1983.\n\nLee T. P., Special Issue on Light-Emitting Diodes and Long-Wavelength Pho- \ntodetectors\u2014Foreword (Editorial). IEEE Device 30(4):257-258, 1983.\n\nLevy U., Gordon E. I., Logan R. A., Laser Cathode-Ray Tube With a Semiconduc- \ntor Double-Heterostructure Screen. IEEE Elec D 4(5):155-156, 1983.\n\nMa K. T., Kusic G. L., Tie Line Contingency Studies Based Upon Partial \nInformation. IEEE Power 102(6):1838-1842, 1983.\n\nSaul R. H., Recent Advances in the Performance and Reliability of InGaAsP \nLEDs for Lightwave Communication-Systems. IEEE Device 30(4):285-295, \n1983.\n\nZipfel C. L., Chin A. K., Di Giuseppe M. A., Reliability of InGaAsP Light-Emitting \nDiodes at High-Current Density. [IEEE Device 30(4):310-316, 1983.\n\nAcuna M. H. et al., Physics of the Jovian and Saturnian Magnetospheres\u2014 \nHighlights of a Conference Held at the Applied Physics Laboratory, The Johns- \nHopkins-University, October 22\u201424, 1981 (Review). Space Sci R 35(3):269- \n292, 1983.\n\nAllara D. L., Hebard A. F., Padden F. J., Nuzzo R. G., Falcone D. R., Chemically- \nInduced Enhancement of Nucleation in Noble-Metal Deposition. J Vac Sci A \n1(2):376-382, 1983.\n\nBean J. C., Recent Developments in Silicon Molecular-Beam Epitaxy. J Vac \nSci A 1(2):540-545, 1983.\n\nChabal Y. J.. Hydrogen Vibration on Si(111)7X7\u2014Evidence for a Unique \nChemisorption Site. Phys Rev L 50(23):1850-1853, 1983.\n\nChang J. J., Julesz B., Displacement Limits, Directional Anisotropy and Direc- \ntion Versus Form Discrimination in Random-Dot Cinematograms. Vision Res \n23(6):639+, 1983.\n\nDonnelly V. M., Flamm D. L., Ibbotson D. E., Plasma-Etching of ITI-V-Compound \nSemiconductors. J Vac Sci A 1(2):626-628, 1983.\n\nDubois L. H., Rowe J. E., Surface Phonons on Clean-Covered and Adsorbate- \nCovered Nickel Disilicide (NiSiz) Thin-Films. J Vac Sci A 1(2):1232-1235, 1983. \nEvans D., Celli V., Benedek G., Toennies J. P., Doak R. B., Resonance-Enhanced \nAtom Scattering From Surface Phonons. Phys Rev L 50(23):1854-1857, 1983. \nFocht M. W., Schwartz B., High-Resistivity in P-Type InP by Deuteron Bom- \nbardment. Appl Phys L 42(11):970-972, 1983.\n\nHannay N. B., The Information Society\u2014Perkin Medal Address (Edito- \nrial). Chem Ind L 1983(11):433-437, 1983.\n\nHart A. C., Krause J. T., Coating Technique for High-Strength Lightguide Fusion \nSplices. Appl Optics 22(11):1731-1733, 1983.\n\nJackel J. L., Hackwood S., Veselka J. J., Beni G., Electrowetting Switch for \nMultimode Optical Fibers. Appl Optics 22(11):1765-1770, 1983.\n\nLyon S. A., Chen Y. H., Lin J. F., Worlock J. M., Optical-Transmission at 3.39 um \nDuring Pulsed Laser Annealing of Silicon. App! Phys L 42(11):978-980, 1983. \nMiller D. A. B., Chemla D. S., Eilenberger D. J., Smith P. W., Gossard A. C., Wiegmann \nW., Degenerate 4-Wave Mixing in Room-Temperature GaAs/GaAlAs Multiple \nQuantum Well Structures. Appl Phys L 42(11):925-927, 1983.\n\nOsinskiJ.S., Manzione L. T., Characterization and Moldability Analysis of Epoxy \nReaction Injection-Molding Resins (Review). ACS Symp S 1983(221):263-282, \n1983.\n\nPatel J. R., Golovchenko J. A., X-Ray-Standing-Wave Atom Location in Hetero- \npolar Crystals and the Problem of Extinction. Phys Rev L 50(23):1858-1861, \n1983.\n\nReichmanis E., Wilkins C. W., Price D. A., Chandross E. A., The Effect of Substit- \nuents on the Photosensitivity of 2-Nitrobenzyl Ester Deep UV Resists. J \nElchem So 1380(6):1433-1437, 1983.\n\nRousseau D. L., Ondrias M. R., Resonance Raman-Scattering Studies of the \nQuaternary Structure Transition in Hemoglobin (Review). Ann R Bioph \n12:357-380, 1983.\n\nShapira Y., Brillson L. J., Heller A., Investigation of InP Surface and Metal \nInterfaces by Surface Photo-Voltage and Auger-Electron Spectroscopies. J \nVac Sci A 1(2):766-770, 1983.\n\nSharma S. P., Sproles E. S., Reaction of Palladium With Chlorine and Hydrogen- \nChloride. J Elchem So 130(6):1242-1247, 1983.\n\nTemkin H., Logan R. A., Van Der Ziel J. P., Integrated Arrays of 1.3-u4m Buried- \nCrescent Lasers. App! Phys L 42(11):934-936, 1983.\n\nThiel F. A., The Phase-Relations in the Cu-In-S-System and the Growth of \nCulnSz Crystals From the Melt\u2014Reply (Discussion). J Elchem So 130(6):1445, \n1983.\n\nTrapp K. D. C., Ermanis F., Origin and Elimination of Crescent-Shaped Growth \nDefects in LPE Layers of InGaAs/InP Alloys. J Elchem So 130(6):1381-1383, \n1983.\n\nAnderson P. W., Suggested Model for Prebiotic Evolution\u2014The Use of \nChaos. P Nas Biol 80(11):3386-3390, 1983.\n\nTime-Compression Multiplexing (TCM) of Three Broadcast-Quality \nTV Signals on a Satellite Transponder \nK. Y. Eng, B. G. Haskell, and R. L. Schmidt\n\nSynchronization of Noncolocated TV Signals in a Satellite Time- \nCompression Multiplexing System \nK. Y. Eng and B. G. Haskell\n\nEquivalent Queueing Networks and Their Use in Approximate Equi- \nlibrium Analysis \nA. Kumar\n\nTELBECC\u2014A Computational Method and Computer Program for \nAnalyzing Telephone Building Energy Consumption and Control \nP. B. Grimado\n\nOn the Average Product of Gauss-Markov Variables \nB. F. Logan, J. E. Mazo, A. M. Odlyzko, and L. A. Shepp\n\nBandwidth-Conserving Independent Amplitude and Phase Modula- \ntion \nB. F. Logan\n\nParallel Fault Simulation Using Distributed Processing \nY. H. Levendel, P. R. Menon, and S. H. Patel\n\nGeneration of Syntax-Directed Editors With Text-Oriented Features \nB. A. Bottos and C. M. R. Kintala\n\nPerformance Analysis of a Preemptive Priority Queue With Applica- \ntions to Packet Communications Systems \nM. G. Hluchyj, C. D. Tsao, and R. R. Boorstyn\n\nPart 3 \nTHE AR6A SINGLE-SIDEBAND MICROWAVE RADIO SYSTEM: \nPrologue \nR. E. Markle\n\nSystem Design and Performance \nJ. Gammie, J. P. Moffatt, R. H. Moseley, and W. A. Robinson\n\nRadio Transmitter-Receiver Units \nR. C. Heidt, E. F. Cook, R. P. Hecken, R. W. Judkins, J. M. \nKiker, Jr., F. J. Provenzano, Jr., and H. C. Wang\n\nEqualization for Multipath Fading \nN. O. Burgess, R. C. MacLean, G. J. Mandeville, D. I. McLean, \nM. E. Sands, and R. P. Snicer\n\nThe Traveling-Wave-Tube Amplifier \nJ. F. Balicki, E. F. Cook, R. C. Heidt, and V. E. Rutter\n\nPredistortion for the Traveling-Wave-Tube Amplifier \nR. P. Hecken, R. C. Heidt, and D. E. Sanford\n\nSystem Networks \nR. L. Adams, J. L. Donoghue, A. N. Georgiades, J. R. Sundquist, \nR. E. Sheehey, and C. F. Walker", "title": "magazine :: Bell System Technical Journal :: BSTJ V62N09 198311 Part 1", "trim_reasons": [], "year": 1983}
{"archive_ref": "bitsavers_BellSystemJV64N10198512_16628827", "canonical_url": "https://archive.org/details/bitsavers_BellSystemJV64N10198512_16628827", "char_count": 593731, "collection": "archive-org-bell-labs", "doc_id": 738, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc738", "record_count": 1024, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bitsavers_BellSystemJV64N10198512_16628827", "split": "test", "text": "W. F. BRINKMAN? P. A. GANNON\u2018 J. S. NOWAK! \nH. O. BURTON? T. J. HERR* L. C. SEIFERT\u00ae \n). CHERNAK! D. M. HILL? W. E. STRICH\u201d \nM. F. COCCA! D. HIRSCH? J. W. TIMKO! \nB. R. DARNALL' S. HORING' Vv. A. VYSSOTSKY!| \nA. FEINER? N. W. NILSON? J. H. WEBER\u00ae\n\n\u2018AT&T Bell Laboratories =? AT&T Information Systems \u2014 * Sandia National Laboratories \n* AT&T Network Systems >*AT&T Technology Systems \u00b0\u00ae AT&T Technologies \n7 AT&T Communications 8 AT&T\n\nCoherent Lightwave Communications 2153 \nJ. Salz \nCross-Polarization Cancellation and Equalization in Digital 2211\n\nTransmission Over Dually Polarized Multipath Fading \nChannels \nM. Kavehrad and J. Salz\n\nCross-Polarization Interference Cancellation and Nonminimum 2247 \nPhase Fades \nM. Kavehrad\n\nAnalysis/Simulation Study of Cross-Polarization Cancellation 2261 \nin Dual-Polarization Digital Radio \nL. J. Greenstein\n\nA Laboratory Simulation Facility for Multipath Fading 2281 \nMicrowave Radio Channels\n\nSingle-Frame Vowel Recognition Using Vector Quantization 2319 \nWith Several Distance Measures \nL. R. Rabiner and F. K. Soong\n\nTraffic Capabilities of Two Rearrangeably Nonblocking 2331 \nPhotonic Switching Modules | \nR. A. Thompson\n\nUnion Bounds on Viterbi Algorithm Performance 2375 \nW. Turin \nA New File Transfer Protocol 2387\n\nA VCR-Based Access System for Large Pictorial Databases \nK. Y. Eng, O. Yue, B. G. Haskell, and C. Grimes\n\nOn Binary Differential Detection for Coherent Lightwave \nCommunication \nJ. E. Mazo\n\nPerformance Signatures for Dual-Polarized Transmission of \nM-QAM Signals Over Fading Multipath Channels \nM. Kavehrad and C. A. Siller, Jr.\n\nStatistical Model for Amplitude and Delay of Selective Fading \nP. Balaban\n\nThe chief objective of this paper is to develop a fundamental understanding \nof the effects of laser phase noise on the performance of coherent lightwave \ncommunication systems. A comprehensive treatment applicable to a wide \nvariety of coherent receiver designs under a broad range of conditions is \nprovided. Our models and analytical tools are developed in sufficient detail to \nencompass a broad range of applications. Formulas are derived for the bit \nerror rate in homodyne and heterodyne Phase Shift Keying (PSK), Differen- \ntial Phase Shift Keying (DPSK), Frequency Shift Keying (FSK) and on-off \nkeying. Estimates are provided of the penalties accrued due to phase noise. \nBased on detailed mathematical analysis and estimates, we made several \nfindings. Near quantum-limited receiver sensitivity can be achieved with PSK \nusing homodyne detection only at signaling rates 3000 times greater than the \nlaser linewidth. A receiver sensitivity 3 to 6 decibels poorer than the quantum \nlimit can be achieved with heterodyne rather than homodyne detection. DPSK, \nfor example, can operate at rates only 300 times greater than the laser \nlinewidth. At lower rates, FSK is an attractive candidate. It can be designed \nto be extremely tolerant of phase noise by using wide frequency deviations.\n\nDespite the rapid advance of lightwave technology over the past \ndecade, the basic operation of an optical fiber communications link \nhas remained essentially unchanged. Direct modulation of the source \n(on-off keying) and direct detection at the receiver using a pin diode \nor Avalanche Photodiode (APD) have been the mainstays of lightwave\n\nsystems since their infancy. Recent advances in lightwave components, \nhowever, now permit significant improvements on this time-honored \napproach. For example, external modulation of the laser with an \nelectro-optic device has recently produced better long-distance per- \nformance at very high bit rates (24 Gb/s) than direct modulation of \nthe laser itself. Another promising technique now being pursued in \nresearch laboratories around the world is the use of coherent light- \nwave\u2014the optical analog of superheterodyne radio reception. Here we \nprovide a comprehensive analytical treatment applicable to a wide \nvariety of coherent receiver designs under a broad range of conditions. \nRecognizing that not all contingencies can be covered explicitly, we \nhave endeavored to develop our model and analytical tools in sufficient \ndetail so they can be applied to other cases of interest.\n\nUnlike direct detection, where the optical signal is converted directly \ninto a demodulated electrical output, the coherent receiver first adds \nto the signal a locally generated optical wave and then detects the \nsum. The resulting photocurrent is a replica of the original signal, \ntranslated down in frequency from the optical domain (~10\u00b0 GHz) to \nthe radio domain (sS few GHz), where conventional electronic tech- \nniques can be used for further signal processing, such as filtering and \ndemodulation. This method offers significant improvements in re- \nceiver sensitivity and wavelength selectivity compared with direct \ndetection. In the 1.3- to 1.6-um lightwave band, for example, an ideal \ncoherent receiver requires a signal energy of only 10 to 20 photons per \nbit to achieve a bit error rate of 10-\u00b0\u2014far less than the roughly 1000 \nphotons required by today\u2019s APDs. And because of its improved \nselectivity, a coherent receiver might permit wavelength-division- \nmultiplexed systems with channel spacings of only, say 100 MHz, \ninstead of the 100 GHz required with conventional optical multiplexing \ntechnology. A further advantage of coherent reception, not often cited \nbut potentially very important, is that it allows the use of electronic \nequalization to compensate for the effects of optical pulse dispersion \nin the fiber. a\n\nThe possible advantages of coherent optical communications have \nbeen explored for numerous applications. Much of the earlier work \nemphasized space communications, where highly collimated laser \nbeams could be used to span enormous distances.\u201d More recently, the \nuse of coherent techniques in optical fiber systems has received con- \nsiderable attention. Especially at bit rates above 2 Gb/s, where APD \nperformance begins to deteriorate, the high sensitivity and potentially \nbroad bandwidth of coherent receivers is a powerful stimulus to further \nresearch. As part of this effort, theroretical analyses of several types \nof coherent receivers have been published in the literature.*\u00b0 (Since \nthese investigations are generally based on fundamental equations and\n\nhardware designs closely resembling those encountered in the micro- \nwave domain, it is perhaps not too surprising that their general \nconclusions are also similar. In both cases, for example, the most \nenergy-efficient binary system uses phase shift keying and coherent \ndemodulation.) The performance of several experimental receivers has \nbeen compared with theory, and the agreement, is generally good, \nespecially when HeNe or YAG lasers are used for the optical sources. \nIn practical coherent lightwave systems, however, itis expected that \nsemiconductor injection lasers will be used; when these have been \nemployed in coherent receiver experiments, measured sensitivity is \nalmost always degraded. The effect is most pronounced in angle- \nmodulation experiments, where receiver performance is often so poor \nthat low error rates (<10~\u00b0) cannot be achieved at all.\u00b0\u2019 The cause of \nthis degradation has been identified as laser phase noise, an impair- \nment that is particularly severe in semiconductor devices. The effect \nof this noise mechanism is to impress random phase modulation on \nthe otherwise monochromatic output of the laser, thereby impairing \nits performance in angle-modulation experiments. A fundamental \nunderstanding of this impairment and its effects on performance is \nthe primary objective of this paper.\n\nLaser phase noise is usually characterized in terms of the linewidth \nof the laser emission spectrum, (a readily measurable quantity that is \ndirectly proportional to the spectral density of the underlying phase \nnoise process.) As was implied above, the linewidths available with \ntoday\u2019s distributed feedback semiconductor lasers, typically 5 to 50 \nMHz, are too broad to take full advantage of coherent techniques. \nConsequently the realization of a stable, reliable narrow-linewidth \nsource is an extremely high priority in lightwave research. Several \npromising techniques have been demonstrated in the laboratory, but \ntheir usefulness under actual field conditions has yet to be established. \nSince reducing laser linewidth appears to be a difficult task, it is \nimportant to understand the effects of this impairment in order to \nestablish precisely how much reduction is required.\n\nThe paper begins with an executive summary. Section II provides a \nbrief review of direct detection methods and fundamental limits. \nProperties of phase noise in lasers are reviewed in Section III. Analysis \nof phase-lock technique are presented in Section IV. Frequency Shift \nKeying (FSK) is treated in Section V while Differential Phase Shift \nKeying (DPSK) and on-off-keying are treated in Sections VI and VII.\n\nA coherent lightwave receiver is the optical analog of a superheter- \nodyne radio set. Instead of detecting photons directly, the coherent \nreceiver first converts the incoming signal from the optical regime\n\ndown to the radio regime, and then uses conventional electronic \ncircuitry to perform various signal processing operations, such as \namplification and demodulation. In principle, this technique can yield \nlarge increases (\u201c20 dB) in receiver sensitivity compared with direct \ndetection using today\u2019s avalanche photodiodes. Indeed, in the 1.3- to \n1.6-um lightwave band, coherent receivers offer the only realistic hope \nof approaching the so-called \u201cquantum limit\u201d of receiver sensitivity: \n~10 photons/bit at 10~\u00b0 error rate.'To date, however, the performance \nof experimental coherent receivers (especially those employing semi- \nconductor lasers) has fallen short of the idealized. theoretical predic- \ntions. One of the prime causes of this degradation has been identified \nas laser phase noise, a phenomenon that is known to be particularly \nserious in semiconductor devices. And since semiconductor lasers are, \nat present, the preferred candidates for coherent systems applications, \nit is imperative to develop an understanding of their noise properties. \nOur goal in this paper is to present a comprehensive treatment of the \ndeleterious effects of laser phase noise in digital lightwave systems, so \nthat the resulting degradation can be understood and predicted.\n\nLaser phase noise is a random process driven by spontaneous \nemissions within the laser cavity, which cause the phase of the optical \noutput wave to execute a random walk away from the value it would \nhave had in the absence of spontaneous emission. This random phase \nprocess manifests itself as a broadening of the laser emission spectrum; \nit is the cause of the broad linewidth (typically 5 to 100 MHz) of \ntoday\u2019s InGaAsP Distributed Feedback (DFB) lasers. In communica- \ntion systems, phase noise degrades performance because unwanted \nphase fluctuations in the received wave impair the demodulation \nprocess, especially when Phase Shift Keying (PSK) is used. At low \nsignaling rates, the accumulated phase \u201cwander\u201d during a signaling \ninterval might be so great that PSK cannot be used at all. In general, \nhowever, as the bit rate is increased, the impairment due to phase \nnoise can be made negligibly small.\n\nThe central question addressed in this paper is, How high must the \nsignaling rate be in order to ensure tolerable system degradation due \nto phase noise? The answer, not surprisingly, depends on system \ndesign constraints. For example, if one requires quantum-limited \nreceiver sensitivity, then PSK with homodyne detection must be used. \nBased on detailed mathematical analysis and estimates we conclude \nthat the degradation or penalty due to phase noise can be kept small \n(<1 dB) only if the ratio of signaling rate to laser linewidth, R/B_z, is \ngreater than 3000. For a laser with B, = 10 MHz, this condition \nimplies a signaling rate of 30 Gb/s\u2014well outside the range of current\n\ntechnology. To operate at lower rates, one might use any of several \n- techniques available for narrowing laser linewidth, but only at the \nprice of a substantial increase in complexity. With the theoretical \nguidelines presented in this paper, the design engineer can strike a \nbalance between the cost of linewidth reduction and-the value of \n_ improved system performance. |\n\nIf a system design can tolerate a receiver sensitivity 3 to 6 dB poorer \nthan the quantum limit, then considerable robustness against phase \nnoise can be achieved by using heterodyne, rather than homodyne, \ndetection. With heterodyne differential detection of PSK, for example, \nthe phase-noise penalty is less than 1 dB for R/B, 2 300, an order-of- \nmagnitude improvement over the homodyne case. Finally, we consider \nthe intriguing case of FSK, which can be made extremely tolerant of \nphase noise by using very wide transmitter frequency deviation. At \nmoderate bit rates (S500 Mb/s), where direct-deviation laser FSK \ntransmitters operate fairly well, this modulation technique appears to \nbe most attractive.\n\nIl. REVIEW OF DIRECT DETECTION AND FUNDAMENTAL LIMITS \n2.1 Direct detection\n\nBefore commencing our principal investigation of coherent tech- \nniques, we briefly review some results related to direct detection of \nlight signals.\n\nDirect detection of light pulses implies a photodetector that converts \nlight energy to electrical signals. The detection mechanism is based \nupon photon counting, which is subject to statistical fluctuations. \n- More specifically, the photon counting process is a time-varying Pois- \nson process whose intensity function A(t) is directly proportional to \nthe information-bearing data wave.\n\nIn the case of binary transmission, the choice between a one or a \nzero is translated into the presence or absence of a burst of optical \nenergy. As an illustration, consider the passage of a single pulse \nthrough an ideal transmission model depicted in Fig. 1. In the case of \na one being transmitted, a square electrical signal turns on the laser \nor LED and energy is sent into the fiber. In the photodetector, light \nwill be detected due to the electromagnetic energy present. Exactly \nwhen in time the photons register on the detector is random. The \nactual electrical current at the output of this device caused by a photon \nis a wideband pulse g-w(t) (which is very narrow compared with the \nsignal duration T'), where g (gain) is an integer-valued random variable \nor g = 1, depending on whether an (APD) or a pin diode is used. In \npractical systems where. amplification of weak signals is required, \nAPD\u2019s are invariably used.\n\nELECTRICAL \nPULSE \nELECTRICAL SIGNAL >) 9x W(t-t,) \n| k \n(t,) ARE RANDOM ARRIVAL TIMES, POISSON DISTRIBUTED \nWITH A MEAN ARRIVAL RATE A(t) PHOTONS/s : \n(9): AVALANCHE GAIN OF PHOTODETECTOR \nE(n) = AT = EXPECTED NUMBER OF PHOTONS/BIT \n-AT \nPe = > - > e~2P p = OPTICAL ENERGY\n\nAssuming that superposition holds for optical fiber transmission, \nthe single-pulse description may be extended to an entire data wave. \nIf one transmits a sequence of on or off pulses, then the received signal, \ndefined as the electrical output of the photodetector on which proc- \nessing is performed, is written as\n\nwhere the time points {t,} form a Poisson process having intensity \nfunction A(t), with\n\nand h(t) is a square pulse, {a,} = 0 or 1 are the data levels, {g,} is \navalanche gain, 7' = signaling interval, and w(t) = output pulse of the \nphotodetector.\n\nIn this simple model, to detect the j th bit, one integrates the output \nof the photodetector over the jth T-second interval and compares the \nrandom variable with a threshold. If the output is greater than the \nthreshold, a one is declared; if it is less, a zero is declared.\n\nIn the ideal situation, when a pin diode is used, g, = 1 and when the \nthreshold is set at zero, the average output of the integrator will yield \nf\u00e9 A(t)dt = \\T when a one is sent and zero output when a zero is sent. \nSince the number of counts n with intensity AT' is Poisson distributed\n\n1 \nPe = 5 eM. | (4) \nThe average optical energy, photons per bit, is just P = 1/2 (AT) + \n1/2(0) and so (4) is written\n\nThis is a fundamental limit on the bit error rate and is commonly \nreferred to as the \u201cquantum limit.\u201d\n\nEquation 5 implies that in order to obtain an error rate of 107\u00b0, \nabout 10 photons/bit are required. This of course is the error rate \nachieved in the absence of coding. It has been shown recently\u00ae that by \nemploying coding the number of required photons per bit is on the \norder of 2 to 3 provided the information rate is less than a character- \nistic rate called channel capacity.\n\nWhen an avalanche detector is used to gain optical amplification, \nthe average value of {g,} may be large but the fluctuations are also \nlarge causing amplitude jitter. The penalties incurred by avalanche \ndetectors have been extensively studied.? Depending on the type of \navalanche detectors used, the loss can be anywhere from 10 to 20 dBs \nfrom the quantum limit (see, for example, Ref. 10). Thus one of the \nchief motivations for turning to coherent techniques is to minimize \nthis tremendous loss in detector sensitivity.\n\nWe begin the discussion of coherent techniques and their possible \nmerits by assuming that the electromagnetic wave at the output of a \nlaser can be represented as\n\nwhere A? is proportional to the optical power. Now suppose that this \nwave is phase modulated so that a one results in A cos wot and a zero \nresults in \u2014A cos wot. An ideal homodyne detector adds to the received \nwave a local carrier wave of amplitude equal to exactly A. So, the sum \nis\n\nWhen the sum is detected by a photodetector (pin diode) and the \noutput integrated for JT\u2019 seconds, one obtains for the average number | \nof counts dT, either 4A ?T or 0. The average transmitted optical energy | \nin this case is P = A\u2019T and so the probability of a bit error is\n\nThis result indicates a 3-dB improvement over the quantum limit, and \nit is often referred to as the \u201csuper quantum\u201d limit.\u2019! Reviewing briefly, \nto achieve the super quantum limit, the local laser had to know exactly \nthe frequency, phase, and the magnitude of the transmitter laser\u2014a \nrather ambitious requirement. This detector is depicted in Fig. 2 with \nalternative No. 1 used as the input to the photodetector.\n\nNow suppose that we relax the requirements on the local laser and \npermit its intensity to be any value B, but still requiring knowledge of \nthe transmitted carrier frequency and phase. Now the combined waves \nbecome | |\n\nAgain (8) is detected by a photodetector and consequently the average \nnumber of counts at the output after integration is now (B + A)\u2019T, \nwhere B is the amplitude of the local laser and it is assumed that \nB>A.\n\nTo estimate the resulting bit error rate in this situation we invoke \na limit theorem. The theorem has to do with the conditions under \nwhich a \u201cshot noise\u201d process\u2014the output current from the photode- \ntector\u2014is well approximated by a \u201cwhite gaussian\u201d noise process. The \nmain requirement is that the rate of photon arrivals be large. Since B \nin (8) can be made as large as one desires, the average number of \nphotons is proportional to \\T' = (B? + A? + 2AB)T. If the common \nbias term (B* + A*)T is subtracted from XT, there is left an antipodal \nsignal pair +2ABT for the net average counts corresponding to recep- \ntion of binary ones and zeroes. The variance of the resulting Poisson \nprocess also equals AT and since B > A by hypothesis, the variance \nis essentially TB*. Now in the limit of large number of counts due to \nthe addition of the local laser to the incoming signal, the output \nelectrical signal can be modeled by\n\nwhere n(t) is a white Gaussian noise process with double-sided spectral \ndensity equal to B*. Integrating (9) from 0 to T, results in a Gaussian \nrandom variable. The resulting bit error rate is then |\n\nwhich is asymptotically (large P) the quantum limit. We have thus \ndemonstrated that an ideal homodyne detector using a pin photodiode \nachieves the quantum limit. This is made possible by the availability \nof large \u201clocal\u201d optical power that provides indirect amplification of \nthe incoming weak optical signal. While providing amplification, the \nprocedure also produces additive noise. This mode of detection is \ndepicted in Fig. 2 with alternative No. 2 for the input to the photode- \ntector.\n\nFinally we consider a detection technique where, instead of trans- \nlating the incoming optical wave directly to baseband, it might be \nadvantageous in some cases to make a frequency translation to an \nIntermediate Frequency (IF). This procedure is called heterodyne \nreception and it is depicted in Fig. 2 with alternative No. 3 for the \ninput to the photodetector.\n\nTo understand the consequences of this approach we proceed as \nfollows. Let the local laser frequency be denoted by wy, and the \nincoming optical frequency by wo such that the IF frequency is w; = \nwo \u2014 w,. The addition of the two waves now results in\n\nExpressing s(t) in terms of the envelope and phase about w, results \nin the representation\n\nThe response of the photodiode to the wave (12) is again a shot-noise \nprocess with intensity function, Ao, equal to the envelope squared.\n\nUsing the same limit arguments as in the previous section, we first \nsubtract B? + A? from X(t), which retains the antipodal signal pair\n\nBecause B > A the fluctuating noise is white Gaussian with double- \nsided spectral density = B*. Denoting the additive noise by n(t), the \nequivalent signal-in-noise problem after heterodyning becomes\n\nThis is a standard elementary detection problem and deciding whether \na plus or a minus was sent is accomplished by multiplying (17) by \ncos w;t, integrating for T seconds, and comparing the result with a \nthreshold set to zero. The decision statistic is\n\n_ where we have neglected the double frequency term. Since the random \nvariable {\u00a7 n(t) cos(w;t) dt has variance equal to B?T/2, the bit error \nrate in this case is asymptotically\n\nThe exponent is seen to be a factor of 2 smaller than in (10) and \nbecause of this heterodyne detection is 3 dB inferior to the quantum \nlimit. The asymptotic performance of the ideal frequency translation \nmethods just discussed are summarized in Table I. |\n\nWith these preliminaries we are now in a position to discuss more \nrealistic detection systems, where laser phase noise must be taken into \naccount. Before doing this however, we briefly review the origins of \nthis noise.\n\nPhase or frequency noise in lasers is a well-known and documented \nphenomenon that sets fundamental limitations on the performance of\n\ncoherent optical communications.'!*\" It has been observed that the \nspectral density of this frequency noise has a 1/f to 1/f* characteristic \nup to around 1 MHz, and is flat for frequencies above 1 MHz,\" \nshown in Fig. 3. The flat, or white, component is associated with \nquantum fluctuations and is the principal cause for line broadening. \nFrom a communications theory point of view, the relatively low- \nfrequency components can be easily tracked, and so we shall not dwell \non this part of the noise.\u2019\u00ae Our main focus here will be on the white \ncomponent.\n\nLaser phase noise is caused by randomly occurring spontaneous \nemission events, which are an inevitable aspect of laser operation. \nEach event causes a sudden jump (of random magnitude and sign) in \nthe phase of the electromagnetic field generated by the device. As time \nevolves the phase executes a random walk away from the value it \nwould have had in the absence of spontaneous emission. The mean- \nsquared phase deviation grows approximately linearly with time, and \nsince the average time between steps in the random walk becomes \nvanishingly small, the random phase 6(t) becomes in the limit a \nWiener process characterized by a zero-mean, white Gaussian fre- \nquency noise p(t) with two-sided spectral eee No.\u2019 Thus, the \nphase process is represented as\n\nTo determine the parameter No (which is a function of both the \nlaser structure and the operating conditions), one can measure the \nspectral density of the frequency fluctuations in the emitted light and \nhence determine Np directly. Experiments of this sort have shown that \nthe representation in (20) using the white-noise approximation for \nu(t) is reasonably accurate for 0.1 ns S 7 S 1 us, which is adequate \nfor our present purposes. Another technique for measuring No makes \nuse of the fact that phase noise causes an observable broadening of \nthe laser emission spectrum. In effect, the accumulated phase error \ngiven by (21) limits the duration of temporal coherence of the laser \nradiation to an interval of roughly 1/(27)*No; the corresponding line- \nwidth is therefore proportional to the noise density No. The following \ndiscussion makes this relationship more precise.\n\nwhere the innocuous inclusion of the uniform phase \u00a2 renders s(t) a \n\u2018stationary process with correlation function,\n\nA simple calculation reveals that the Fourier transform of (23), the \npower spectrum, is\n\nS A? \n[ G(f)df = x (25) \nas it should. A sketch of the baseband spectrum, commonly referred \nto as Lorentzian is shown in Fig. 4. The parameter characterizing \nG(f), No, can be experimentally determined by measuring the 3-dB \nbandwidth of the spectrum around fp. Denoting the total (two-sided) \n3-dB bandwidth by B;, it is seen from (24) that \nB,\n\nwhen B; < fo. | \nWe will see later that seemingly modest amounts of phase noise can \nseriously degrade coherent system performance; thus it is imperative\n\nto make lasers with the narrowest possible linewidth. Unfortunately, \nthe semiconductor injection laser designs likely to be used in the 1.3- \nto 1.6-um lightwave band typically have linewidths in the range 5- to \u00bb \n50-MHz, which is too broad for many potentially important coherent \nlightwave applications.\u2019*\u201d\u00b0 (For comparison, the reader should note \nthat microwave oscillators, which are widely used in coherent radio \napplications, have linewidths on the order of 1 Hz.) To reduce laser \nlinewidth, experimenters have exploited the fact that the noise density \nNo is inversely proportional to PoQ\u2019, where Po is the laser output \npower and Q is the quality factor of the \u201ccold\u201d laser cavity resonance; \nthus high-power, high-Q lasers tend to have narrow linewidths. The \nmost impressive line-narrowing experiments have been performed \nusing a mirror or diffraction grating external to the laser chip to \nproduce a composite cavity of very high Q.\u201d\" Under relatively benign \nlaboratory conditions, linewidths of tens of kilohertz have been ob- \ntained: an improvement of three orders of magnitude! Whether this \napproach will prove practical under harsh field conditions remains to \nbe seen. In any case, it appears that phase noise will be an important \nconsideration for the foreseeable future, so we turn now to developing \na clear understanding of its consequences.\n\nWe saw in Section III that homodyne detection of an optical PSK \ndata wave makes it possible to achieve the quantum limit. However, \nto gain the full benefit of this approach the local laser must have \nperfect knowledge of the transmitted optical center frequency and \nphase. In this section we explore the possibility of deriving these\n\ncrucial parameters from either the optical data wave directly or from \n_a heterodyned version.\u201d\u201d At microwave frequencies, carrier recovery \ntechniques are well established, while at optical frequencies, the avail- \nable methodologies are still limited. For example, it is very difficult to \ndirectly multiply two optical signals or square an optical wave, which \nmakes it difficult to wipe off the binary modulation. In attempting to \nderive carrier in the optical frequencies, it is, therefore, necessary to \nresort to data-aided techniques.\u201d\n\nAt the input to the detector, the power of the incoming optical data \nsignal is split so that a fraction, A7k\u2019, is devoted to the estimation of \nthe phase process 6(t) by a phase-locked loop, and the remaining \nportion of the power, A?(1 \u2014 k?), is used for demodulation. The power \ndivision that is determined by the choice of the constant 0 < k <= 1 is \na parameter that must be optimized. In the phase-locked loop we \npresume that the local Voltage Controlled Oscillator (VCO) can be\n\nmodulated by the estimated data m(t) using decision directed tech- \nniques, and consequently the modulation can be locally wiped off and \na modulation-free carrier thus made available in the lower portion of \nthe figure to perform the homodyne demodulation function. We also \nassume that the optical center frequency can be identically matched \nby the local laser. In any event, if this is not the case, an additional \nphase-locked loop may be needed to track this mismatch.'\u00ae Making \nthese assumptions, it is possible to analyze the performance of this \ndetector and to obtain tight bounds on the degradation from ideal \nhomodyne performance.\n\nFor the subsequent analysis we refer to Fig. 5. Adding a fraction c \nof the VCO output to a fraction of the incoming optical data wave we \nobtain\n\nwhere m(t) at the output of the VCO is the reconstructed data wave \nfrom past decisions, and it is assumed to be devoid of errors. The \nsquared envelope of V(t), which is the average value of the Poisson \ncounting process n(\u00a2), at the output of the photodetector is\n\nAfter de elimination, the \u201csignal\u201d portion of the \u201cshot noise\u201d process, \nwhich in the limit becomes a Gaussian process since B can be made \nlarge is |\n\nand the resulting zero-mean white Gaussian noise process is denoted \nby v(t) with spectral density equal to B?c\u2019.\n\nThus the equivalent signal-plus-noise process, which controls the \nfrequency of the local laser VCO is\n\nu(t) = 2ABkc sin Y(t) + v(t), (31) \nand so, because of feedback, we must satisfy the equation \nt \n\u00a7(t) = o(t) +K i u(t\u2019)dt\u2019, (32) \nwhere \u00a2(t) is the phase noise process of the local laser, and K is a \nproportionality constant to be determined later. \nSubtracting from both sides of (32), 0(t), the phase of the incoming\n\nwave, and differentiating, we get the stochastic differential equation\n\nand yo(t) is now a white Gaussian process with double-sided spectral \ndensity,\n\nwhere the Lorentzian bandwidth of the transmitter laser is Bz, and \nthe local laser VCO bandwidth is By\u00bb.\n\nIt is well known**\u201d that (33) obeys a Fokker-Plank equation yielding \nthe steady-state probability density function (mod27) for the phase- \nerror process y(t),\n\nThe probability density (36) is sharply peaked at y = 0 when a is \nlarge and becomes flat, or uniform when a is small. One strives \ntherefore to design the phase-locked loop so that a is as large as needed \nto obtain minimum degradation from ideal (y = 0). For fixed A and k, \n(37) reveals that a cannot be made arbitrarily large because of the \nfinite Lorentzian bandwidth B,. However, there exist a maximum \nvalue of a (when K2 = 27B_) given by,\n\nwhere P = A?T\u2014the transmitted optical energy in the received signal. \nEquation (39) reveals that for fixed optical energy and k, ao can be \nmade large only by increasing the ratio R/B,.\n\nWe now examine the performance of this optical homodyne detector \nusing the phase-locked loop output as the reference carrier wave. \nIn the lower portion of Fig. 5 the sum signal W(t) is\n\nW(t) = AV1 \u2014 k? m(t)cos(wot + 0(t)) \n+ Bv1 \u2014 c? cos(wot + O(t)). (40) \nThe squared envelope of W(t)\u2014the response of the photodetector\u2014 \nis therefore \nE*(t) = A*(1 \u2014 k*) + B41 \u2014 cc\u2019) | \n+ 2ABV1 \u2014 k? V1 - \u00a2? m(t)cos y(t). (41) \nThe resulting shot-noise process at the output of the photodetector\n\nagain becomes in the limit a Gaussian process with average value \n(after dc elimination) equal to\n\n2ABV1 \u2014 k? V1 \u2014 c? m(t)cos Y(t), (42) \nand zero-mean white Gaussian noise v(t) with a spectrum equal to \nB*(1 \u2014 c?). \nThus the equivalent signal plus noise prior to integration is \nS(t) = 2AV1 \u2014 k? m(t)cos W(t) + volt), (43)\n\nwhere we have divided signal plus noise by BV1 \u2014 c\u2019 thus normalizing \nthe spectrum of vo({t) to unity.\n\nIntegrating (43) over a T-second interval results in the decision \nstatistic |\n\nwhere yp iS now a zero-mean Gaussian random variable with unit \nvariance, the random variable\n\nically since it requires knowledge of the nth-order probability distri- \nbution of the phase error process y(t). The most we have, however, is \nthe first-order distribution, and therefore we must resort to upper \nbounds using the only information we have.\n\nIn Appendix A, an exponential upper bound on (46) is developed \nwith the result\n\n2 \nglaje *, a/p? >1 \nPes kee a/p? <1, (47) \nwhere the coefficient, g(a), is defined in Appendix A, eq. (178). \nWhen a/p? = 1, (47) is reduced to a single bound, \nas \nPe <= g(aje *. (48) \nRecall that \np* = 4P(1 \u2014 k\u2019), (49) \nand\n\nwhere vy is the ratio of the signaling rate to the average Lorentzian \nlaser bandwidth\n\n2R \ni= B, (52) \nIt is seen from (47) that the error exponent assumes two crucially \ndifferent forms depending on whether r = 0 or r < 0. For fixed y and \nk, the exponent is linear in P when r = 1, while it behaves as the \nsquare root of P for r< 1. Thus the probability of error decays much \nmore slowly with P when P > constant y [eq. (51)], indicating a \nthreshold for Pe versus P. For fixed y and P, however, r is a monoton- \nically increasing function of k, 0 < k < 1, and so there exists a value \nk = k,, such that r(k,) = 1, given by\n\nThis value of\u2019k is not, however, the optimum value that makes the \nnegative exponent the greatest, or the probability of error the smallest. \nThe optimum value of k, or the power division, is found by setting the \nderivative of the exponent in (47) for r < 1 to zero. From (47) the \nnegative exponent is\n\nindicating that there exist a k,\u00bb < k which increases the exponent. \nSetting (55) to zero, we get the formula for the optimum R,\n\nOne can now proceed to solve (58) numerically and use this optimum \nvalue to plot the upper bound on the probability of error versus P for \ndifferent values of y. A more incisive way to exhibit the behavior of \nthe error rate, however, is to use the suboptimum value of k, > Rop \ngiven in (53), which renders r = 1, and then define a penalty, which is \nthe reduction of the exponent relative to ideal performances. Since k,, \n< k,, this still provides an upper bound on the error rate. It can be \nseen from (58) that when the term 1/(1 \u2014 k3,) is neglected, the resulting \nequation is identical to (53) and so, to this degree of approximation, \nRoy ~ k, (this is a good approximation when k,, turns out to be small, \nwhich should be the case when the penalty is small). Following this - \napproach, we solve (53) for x and write the negative exponent (54) as\n\nThe penalty incurred due to the finite value of Y may now be defined \nas follows. Equating (59) to 2P,, where P, is chosen to achieve a \ndesired error rate, say 107\u00b0, in which case P, ~ 10, one obtains a \nfunction of P,/P versus y, and for each value of y one can then \ncalculate, \u201410 log P,/P), which defines the penalty. |\n\n2P, = 4Pxvx7 +1 \u2014- x, . (60) \nand when \n1 \nx= \u2014 >= V2/P \n8Vr / \nis substituted into (60), one gets the penalty function \nee \n: P +\u00bbvy+16xP, ey\n\nFig. 6\u2014Penalty versus signaling rate divided by linewidth in homodyne PSK detec- \ntion at two different ideal error rates for identical laser linewidths.\n\ncorresponding to Pe ~ 107\u00b0 and the other to Pe ~ 107\u00b0. It can be seen \nthat negligible penalty is incurred when y = 3000 for Pe ~ 10~? and 3 \ndB is given up at y ~ 500. We will see in the next section that a \nheterodyne Phase-Locked Loop (PLL) also suffers a 3-dB penalty at \naround y ~ 500. This approach is analyzed in the next section.\n\nHeterodyning the optical data wave down to an IF frequency, and \nthen deriving the carrier from the resulting microwave signal using \nstandard well-known techniques may make it easier to wipe off the \nmodulation, and consequently ease the signal processing burden of the \nphase-locked loops.\u201d\u00ae\u201d\u201d In the subsequent analysis we presume that \nthe modulation has been eliminated, and as such the phase-locked \nloop can operate directly on the IF carrier wave. So, after heterodyning \nthe optical signal\n\nto an IF frequency f; and wiping off the modulation, we obtain the \nmicrowave signal plus noise\n\nwhere A? equals optical power, n(t) is again a white Gaussian noise \nprocess with unit double sided spectral density, and @(t) is now the \ndifference between the transmitting laser\u2019s phase noise and the local \nlaser\u2019s phase noise. Consequently, the variance of 6(\u00a2) now is\n\nThe signal (63) is\u2018now the input to a conventional PLL depicted in \nFig. 7. The analysis of the PLL is straightforward. Denote the output \nof the PLL by\n\ndouble-frequency terms from its output. Yet, the cut-off frequency is \nplaced high enough so that the output noise can still be regarded as \nwhite. Thus, because of feedback, the following equation must be \nsatisfied\n\nwhere. now the double-sided spectral peneily of v(t) is equal to K?/2. \nThe phase error therefore is\n\nt \ny(t) = 0(t) \u2014 Ke { e(s)ds, (69) \n0 : F \nand when this is differentiated one obtains | \ndy _ d\u00e9 \na a OF 7 \n= Qa(ui(t) \u2014 wo(t)) \u2014 Kee(t), (70)\n\nwhere p; and po are the frequency noises of the transmitter laser and \nthe laser involved in the local heterodyner, respectively.\n\nWhen (70) is substituted into (68), one again obtains the well-known \nstochastic differential equation governing the evolution of the phase- \nerror process,\n\nae = \u2014K, KA sin y(t), (71) \nwhere \nu(t) = \u201422(ui \u2014 pe) + v(t) Ke (72) \n- is a white Gaussian noise process with double-sided spectrum equal to > \nK?K3 | \nD= 27Bz + 9 (73)\n\nLetting K,K2 = K, we observe that (71) is identical in structure to \n(33), which again results in a Fokker-Plank equation yielding the \nsteady-state probability density function (mod 27) for the phase error\n\n_ e*cosy _ . \np(y) = Onley\u2019 nsyesr. (74) \nThe important system parameter is now given by \n2AK 2AK \nnr a a \n\u2014\u2014 27B,\n\nThe algebraic form of the PLL parameter a departs from the \nconventional form where, by decreasing the loop bandwidth, a can be \nmade as large as possible. Here there is a minimum band resulting \nfrom the presence of phase noise. The only way that a can be increased \nis by increasing the input optical energy, or by decreasing phase noise \nrelative to the signaling rate. Consequently there exists a maximum \nvalue of a (when K? = 47B,z) piven by\n\n4.4 Performance | \nNow consider the modulated heterodyned wave | \nS(t) = +2A cos(2Qrfit + O(t)) + n(t), Ost<T, (77) \nand the estimated carrier wave from the PLL \ncos(2afit + 6(t)). (78) \nMultiplying (77) by (78), integrating from 0 to T, and eliminating the\n\nwhere y(t) is the phase-error process obeying at any instant of time \nthe probability density (74). Rewriting (79) as before, \nS = +ATE + \u00bbp, (80)\n\nwhere p = V2TA and vp now has unit variance. \nThe probability of error is as before,\n\nThis probability is structurally identical to (46), and we therefore use \nthe same bound from Appendix A.\n\nWith the definition of a, in eq. (76) we can compute the threshold \nparameter in this case,\n\nWe note that the behavior of this bound is slightly different from \nthe one applicable in the homodyne case. For one, here ideal perform- \nance is 3 dB inferior to the quantum limit, which is accounted for by \nthe heterodyning operation, as we have already noted. Next, ideal | \nperformance with negligible degradation is attained when\n\nB, > 8rP, \nand for ratios less than this, the degradation is increased gracefully. \nAs an example, when P = 20, yielding an ideal error rate ~107~\u00b0, the \nthreshold parameter y = 500. For ratios greater than this, no penalty \nin optical power (from the ideal P = 20) is incurred. To assess the \npenalty at operations at less than this threshold, we use the same \ndefinition as before. The probability of error exponent when r, < 1 is\n\nFig. 8\u2014Penalty versus signaling rate divided by linewidth in heterodyne PSK detec- \ntion at two different ideal error rates for identical laser linewidths.\n\nb= = or, rs 1. (88) \nSubstituting the definition of r, [eq. (84)] and solving for y, yields \n1 \u2014 v1 \u2014 5)? \ny = 8rP, eae (89)\n\nIn Fig. 8 we plot b in decibels versus y for two different values of \nP,. We see that around y ~ 400, the asymptotic degradation of 3 dB \nis approached. Interestingly, at around this ratio the optical PLL also \ndegrades by 3 dB, as we have already mentioned.\n\nIn concluding this section we remark that our estimates are only \nupper bounds, albeit, we feel, tight bounds. The exact evaluation of \nthe error rate is not feasible because of the nonlinear functionals that \nare involved. If one chooses to ignore the time integrals, than the \nprobability of error can be evaluated numerically as has been done in \nRefs. 28 and 29. |\n\nIn the next section we will see that noncoherent techniques such as \nfrequency modulation and differential phase modulation, where knowl- \nedge of carrier phase is not essential, yields performance very near to \nwhat can be attained with heterodyne phase-lock technique at reason- \nable signaling rates.\n\nV. BINARY FREQUENCY SHIFT KEYING \nHere we discuss and analyze the performance of binary Frequency\n\nShift Keying (FSK) as one of the modulation options.*\u00b0*? In this \nmodulation method information is conveyed by switching the fre- \nquency of the laser between two different values. Thus, during a fixed \ninterval, T = 1/R, where R is the signaling rate, the receiver has to \ndecide whether\n\nwas transmitted. In (90) A? is again proportional to the received optical \npower and 6(t) is the phase noise process associated with the laser.\n\nIt appears that this modulation method is impervious to the effects \nof phase noise, since the shifted frequencies w,; and we, can be suffi- \nciently separated and if enough bandwidth is available, crosstalk due \nto the fluctuating phase noise can be minimized. These are the chief \nreasons, then, for considering FSK. We point out that this is not what \nis commonly referred to as continuous-phase narrow-band FM. The \nlatter yields the same performance as differential phase modulation \ntreated in the next section.\n\nThe first step in the processing of the FSK optical signal is to \nheterodyne (90) to an IF frequency w;. As was already noted, this can \nbe accomplished by adding to (90) a locally generated optical signal \nand then direct detecting the sum by a photodetector. The sum signal \nthen is\n\nwhere the IF frequencies are w; = w; \u2014 w;, and \u00a2(t) is the phase noise \nassociated with the local laser. The output of the photodetector is \nagain a \u201cshot noise\u201d process |\n\nWhen the local laser intensity B >> A (66) approaches a white \nGaussian process with average value equal to \\(\u00a2) and standard devia- \ntion also equal to A(t). Thus the dc part of the average value of (93) is\n\nwhile the resultant zero-mean Gaussian noise, n(t), has a double-sided \nspectral density equal to B\u201d. \nWith these preliminaries, we now confront a classical detection\n\nproblem. Given the IF signal (94), plus white Gaussian noise of unit \nspectral density\n\nhow does one process V(t) so as to attain the least probability of \nerror? While the problem is classical the solution is not tractable in \ngeneral because of the presence of the phase noise process A(t).\n\nFor calibration purposes, let us first review briefly the performance \nunder the assumption that the phase noise is effectively a constant. \nIn the case when A(t) is slowly varying with respect to the rate R = \n1/T, or when the symbol rate R is much greater than the bandwidth \nof the laser signal, the optimum detector has a well-known structure \ndepicted schematically in Fig. 9. The results in this case will serve as \na benchmark to which the later more general results will be compared. \nAlso assume that the frequency shifted signals are orthogonal, i.e., the \ntwo frequencies w; and w, are chosen such that\n\nTo proceed with the error rate analysis, assume that w; was sent. In \nthis case, we expect the x output in Fig. 8 to be greater than y, and an \nerror is made when x \u2014 y < 0. Because of symmetry, when wz is sent, \ny is expected to be greater than x and a mistake is now made when \ny \u2014 x = 0. So, the probability of error is just\n\nIn order to evaluate this probability, we express the random varia- \nbles x and y as indicated by the mathematical operations in Fig. 7. It\n\nFig. 9\u2014Structure of the optimum FM detector when linewidth approximates zero.\n\nThe desired probability is just the probability that the difference in \nthe lengths of two 2-dimensional Gaussian vectors is less than zero. \nAs can be verified, the x\u2019s are independent Gaussian random variables \nwith identical variances, o\u201d = T/2. For these random variables, (97) \ncan be expressed exactly\u2122 as\n\nComparing this with the performance of direct detection, we observe \nthat the optical signal (90), after direct detection and integration for \nT seconds, yields an average photon count equal to A?7. In this direct \ndetection case, the chance of making an error would be the chance \nof detecting zero photons in T seconds. From the Poisson distribu- \ntion, for the number of photons detected, this probability is just \n1/2 exp[\u2014A\u2019T], which is 3 dB worse than the quantum limit. Com- \nparing this with (98), however, reveals an additional 3-dB loss due to \nheterodyning. So, heterodyne FSK detection is 6 dB inferior to\n\nReturning now to the more realistic situation where phase noise is \npresent and must be included in the performance analysis, we recall \nthat the crucial assumption for the previous analysis was that the \nsymbol rate baud R = 1/T' be much greater than the linewidth of the \nlaser, so that the phase process A(t) could be regarded as a constant \nduring the integration period. The inclusion of the phase noise process \nimmediately raises fundamental problems concerning the detector \nstructure.\n\nAs is well known,\u201d\u2019 the optimal processor in the presence of phase \nnoise first estimates the phase process, and then uses this estimate to \ncoherently demodulate or detect the IF signal. This is precisely what \na PLL does, but we have seen that coherent PLL detection becomes\n\nfeasible only at very high data rates. This situation leads us to \npostulate a detector that, while not optimum, is reasonable and does \nnot require a phase-locked loop.\n\nThe proposed frequency detector structure is shown in Fig. 10. It is \nessentially an energy detector. It consists of two ideal bandpass filters \nof total bandwidth equal to 2W. The purpose of these filters is to limit \nthe added white noise bandwidth as much as possible, while at the \nsame time to retain most of the energy in the information carrying \nsignal. A precise number for the bandwidth that satisfies these two \nseemingly contradictory requirements is hard to derive because, \nstrictly, the received sine wave with phase noise has infinite band- \nwidth. The front-end bandwidth must remain as a parameter in our \nsubsequent analysis, and engineering estimates will be attempted later.\n\nFollowing these bandlimiting filters with square-envelope detectors \nand an integrator essentially provides an estimate of the energy in \neach frequency band, and this quantity should be independent of phase \nnoise provided that the front-end bandwidth is sufficiently large. We \nnow proceed to analyze the performance of this structure.\n\nIn the representation of signal plus noise, eq. (95), assume that a \nwas sent. Regarding w as the center frequency, we represent the signal \nand noise in terms of in-phase and quadrature components as\n\nS(t) = 2A cos w,t cos A(t) \u2014 2A sin wt sin A(t) \n+ nj(t) cos wt + ne(t) sin ait \n= x(t) cos a;t + y(t) sin art, | (99) \nwhere\n\nAt the output of the bandpass filters x(t) and y(t) are bandlimited \nversions of the input. As we have already stated the signals x(t) and \ny(t) remain essentially undistorted at the output, and the only effect \nof the bandpass filters is to limit the noise band. Clearly, as the band \nW increases, this approximation becomes better.\n\nThe output baseband noises with unit input noise intensity now. \nhave mean-square values |\n\nSquaring the envelope and integrating as indicated in Fig. 8 yields the \nquadratic decision statistic q,,\n\nq = t [x*(t) + y*(t)]d\u00e9, (101) \nwhere for a given A(t), x(t) and y(t) are independent Gaussian \nprocesses with\n\nThe covariance functions of the ideally bandlimited baseband noise \nprocesses n,(t) and n2(t) are |\n\nAs is well known,\u00ae associated with this covariance kernel are an \ninfinite set of orthonormal eigenfunctions {y,(t)} and a set of nonneg- \native eigenvalues {),}. Using these eigenfunctions, we represent the \nprocesses x(t) and y(t) as\n\nIn the bottom leg of Fig. 10 we assume that only the \u201cnoise\u201d goes \nthrough (by previous hypothesis), and so the resulting quadratic form \ng2 is now comprised only of noise. Of course, this is a mild assumption \nsince w, can be separated from wz as much as one wishes to provide \nminimum leakage. Thus,\n\nwhere v,(t) and v.(t) are quadrature and in-phase bandlimited noises \nin the lower leg, and are independent of the noises in the upper leg \nsince the spectra occupy nonoverlapping frequency bands. For this \nreason we denote these noises by \u00bb:(t) and v(t) to distinguish them \nfrom n,; and nz in the upper leg. The decision statistic is the difference \nof the quadratic forms (105) and (106) and consequently the probabil- \nity of error is\n\nPe = Pr[q = qi \u2014 gz = Ol. | (107) \nThis can be expressed in terms of the characteristic function of q, \n\u2014\u2014 C(w) = Ee, (108) \nas the integral** | \nPe = mo [ en dw. (109)\n\n~ The ie, \u00ab > 0 in the denominator denotes the fact that in the complex \nw plane the contour of integration goes above the singularity at w = 0. \nUsing (105) and (106) it is straightforward to calculate (108)\n\nand Eaiy(-) denotes the expectation with respect to the phase process \nA(t). To proceed further, we invoke an excellent approximation\u00ae \nregarding the behavior of the eigenvalues {\\,,} in this application. Since \nthese are the eigenvalues of the Prolate-Spheroidal wave functions, it \nis shown in Ref. 35 that\n\nWe note the nth order pole at z = \u20141 and the essential singularity \nat z = 1. When the time-bandwidth product n = 1, the contour can be \nclosed in the left-hand plane, and the value of the integral is just the \nresidue of\n\nThis result is identical to (98) and a moments reflection will reveal \n_ the reason for the consistency. Note that roughly\n\nwhere R + kB; = 2W\u2014the bandpass bandwidth of the incoming signal. \nThis band must equal to or be greater than the signaling rate R, plus \nkB,\u2014the bandwidth required to pass the sine-wave signal with the \nphase noise undistorted (k is a positive integer), and again B, is the \nlaser linewidth. Clearly, when B,/R <1, no postintegration is required \nand consequently n = 1. Therefore in this case we get the previous \nresult where we regarded the phase noise as a constant\u2014precisely the \ncase when kB;/R \u00ab 1. When this condition is not necessarily satisfied \nand son # 1, we can still evaluate the contour integral. In this general \ncase, closing the contour in the left-hand plane enclosing the nth order \npole gives for (113) the residue and hence the probability of bit error\n\nAs can be seen from (119), the degradation due to excess noise and \npostintegration manifests itself only in an algebraic coefficient in the \nbit-error-rate expression and not in the exponent.\n\nFrom the foregoing analysis we present the curves of Fig. 11 that \nare plots of Pe versus P in decibels for different values of , which is \ntwice the ratio of symbol rate to the sum of laser linewidths. The \nfront-end bandwidths around frequencies f/, and f. in Fig. 9 were \nselected to be\n\nof 10 times the laser linewidths is judged to be adequate to pass the \nincoming FM signal without appreciable distortion.\n\nWhen y \u2014 ~, Pe \u2014 exp{\u2014P/2}, which is the ideal binary FM \nperformance. As already mentioned, this ideal performance is still 6 \ndB worse than the quantum limit, since 3 dB is lost from heterodyning \nand 3 additional decibels is lost due to the fact that the heterodyned \nFM signals are orthogonal rather than antipodal.\n\nFig. 11 can be used to determine the minimum data rate for efficient \nperformance. Suppose the lasers have identical linewidths of 10 MHz, \nand one wishes the degradation not to exceed 1 dB. What is the \nminimum admissible data rate R? From Fig. 11, less than 1-dB \ndegradation yields y ~ 10, implying R = 100 Mb/s.\n\nWhat is the degradation if one desires to transmit at 20 Mb/s? For \nthe same laser linewidths as before, eq. (98) yields y = 2, and from the \nfigure we see that this value of y yields a degradation of 2 dB.\n\nAs is well known, FSK can accommodate more than two frequencies \nto convey digital information without appreciably altering the form of \nthe error rate expression provided again that bandwidth expansion is \nnot an obstacle. For example, consider using 2\u201d frequences m = 2. \n\u2018Generalizing the structure of Fig. 10 to 2\u201d legs yields a probability of\n\nerror that is 2\u201d times the binary error rate but requires only P/2m \nphotons per bit to deliver the same amount of information as in the \nbinary case. To cite an example, using four frequencies, m = 2, the \ninformation rate is 2/T\u2019 and so T can be doubled to obtain the same \nbits per second as when m = 2. This comes at a moderate increase of \nbandwidth and a factor-of-4 increase in the error rate.\n\nLet us consider the previous example where the binary rate R = 100 \nMb/s. This can be achieved with ~40 photons per bit resulting in an \nerror rate ~10~-*. Suppose one had only 20 photons per bit to expand, \nhow can this data rate be accommodated without increasing the error \n\u2018rate? Suppose we half the signaling rate so that the new T equals \ntwice the old T. To maintain the same rate in bits per second we must \nuse four frequencies rather than two. The new signaling rate has now \nbeen halved and so we must recalculate y from eq. (98) corresponding \nto R = 50. We find it to be five. From Fig. 11 we estimate that this \nvalue of y results in a ~1-dB loss. The upshot is that FSK with four \nfrequencies signaling at a data rate of 100 Mb/s can be achieved at a \n4-dB increase of optical power over the quantum limit.\n\nIn the presence of additive white Gaussian noise, Differential Phase \nShift Keying (DPSK) is known to be very efficient in terms of the \ns/n required to achieve an acceptable error rate.*** It is only a fraction \nof a decibel less efficient than coherent phase shift keying\u2014the most \nefficient known method. Our objective here is to investigate the \nperformance of this modulation method in optical communications \nand to assess the incurred penalty due to phase noise.\n\nWe begin our treatment by first considering detection at optical \nfrequencies and then analyze the heterodyned version. Processing at \noptical frequencies may be practically inhibited because of the present \nlack of efficient (noise free) amplifiers, and therefore we also analyze \nthe heterodyned version and compare performance of these two dif- \nferent approaches.\n\nIn DPSK, information is conveyed by the phase differences in two \nconsecutive signaling intervals. Thus when the optical signal in a \nparticular signaling interval is\n\nSo(t) = aoA cos(wot + O(t)), 0O<it<T, (121) \nand in the previous interval it was \nS_i(t) _ aA cos(wot + A(t)), \u2014-T<t< 0, (122)\n\nwhere do, a-; are +1 and 6@(t) is again the phase noise process, \ninformation then is conveyed by the product aoa-_1.\n\nIdeally one would measure the time average* of the phase difference \nin (121) and (122) during the interval [0, 7] and decide that a_,a) = 1 \nif it lies between \u20147/2 and 7/2 and a_,a) = \u20141 if it lies outside of this \nphase range. Consequently, in a practically implemented system a \nlower bound on the bit error rate would be\n\nFrom this lower bound on bit error rate the ratio B,/R is determined, \nsetting a floor for the minimum admissible rate in terms of B,. For \nexample, if one desires an error rate ~10~\u00b0, one equates\n\nThe problem, however, is that in the Gorical processing repertory \nthe phase differential cannot yet be obtained directly nor can two \noptical waves be multiplied directly. Therefore, one must consider a \ndetection method that provides information about the phase difference \nin an indirect manner. Thus consider the following approach: first, \nthe incoming signal is delayed by 7' seconds, then half of the power of \nthe delayed signal plus and minus half of the power of the undelayed \nversions are passed through separate photodetectors. A schematic \nrepresentation of this phase detector is shown in Fig. 12.\n\n* I am indebted to Leonid G. Kazovsky for pointing out the time average of the phase \nnoise differential provides a better lower bound than just the instantaneous value that \nappeared in my original manuscript.\n\nwhere we set \nA(t) = O(t) \u2014 (t \u2014 T). \nThe output pulse count from the \u201csum\u201d photodetector is a doubly \nstochastic Poisson process n, with conditional intensity \\4 equal to \nthe squared envelope of (124),\n\nwhile the \u201cdifference\u201d photodetector output is also a Poisson process \nny independent of n, but having intensity\n\nIntegrating the outputs of the two photodetectors for T seconds yields \ntwo average random counts |\n\nBecause of phase noise, the exact evaluation of the bit error rate is \nnot mathematically tractable, but an exponential upper bound can be \nobtained from the moment generating function of the differential \ncount, n. The moment generating function of n, conditioned on 6 and \naoa_;, is readily calculated from the Poisson distribution\n\nM,.(s| 6, doa-1) = eA \u201cDHA, (129) \nand will be used to upper bound the probability of error. \nWe begin by writing the probability of error\n\nIn the. last two inequalities we used (129) in a Chernoff bound and \nmade use of the convexity of the exponential function as well as the \nfact that A@(t) is stationary.\n\nThe tightest upper bound is obtained by selecting an optimum s for \na given pv and u. This value of s can be obtained by setting the derivative \nof the exponent to zero. The set of random s\u2019s optimizing the exponent \nis then found to be\n\nand since s, has to be positive, we require that v > u, which from (133) \nimplies that cos A@(t) must be positive. For values of A@(t) such that \ncos A\u00e9 < 0, the optimum value of s, is seen to be zero. Thus, in order \nthat the bound (131) be reasonably tight, the average with respect to \nA6(t) indicated in (131) must be carried out over two sets of A@\u00e9\n\nThe first term above can be upper bounded by setting s = 0 and \nfurther upperbounding the probability that cos A\u00e9 <= 0, yields for this \nterm Pr[A@ = 7/2]. So after some calculations and substitutions, (136) \nbecomes\n\nand again P = AT. \nThe penalty incurred due to phase noise depends on the behavior of\n\nthe expectation in (137). It does not appear feasible to evaluate this \nexpectation exactly and therefore we must resort to asymptotic anal- \nysis valid for large P. We defer this analysis to after the discussion of \nheterodyne DPSK, since the penalty evaluation there involves a sim- \nilar calculation.\n\nBefore embarking an analysis of heterodyne DPSK, however, we \nremark that even when A\u00e9 = 0, the probability of error in optically \nprocessed DPSK is 3 dB inferior to the quantum limit. This is seen \nfrom (17) since when A\u00e9 = 0,\n\n= 1 \u2014-A\u00b0T __ 1 \u2014P \nPe=<se =a (139) \nThe reason for this loss is inherent in the demodulation process. As \ncan be seen, the optical detector turned the optical signal into an \u201con- \noff\u201d signal and the explanation for the inefficiency is that twice as \nmuch average optical power has been transmitted to detect an on-off \nsignal.\n\nIt should also be pointed out that the chief motivation for using \ndifferential phase modulation in the microwave region is that it yields \nan error probability close to coherent demodulation without having to \ntransmit and recover carrier phase.*\u201d We therefore turn our attention \nto heterodyne DPSK detection next.\n\nAs has already been pointed out, heterodyning an optical signal \nresults in additive white noise and therefore it is imperative as in the \n_FSK case, to bandlimit the heterodyned signal plus noise to a band- \nwidth just sufficient to pass the received signal with the impressed \nphase noise undisturbed. In DPSK, a bandlimited signal is multiplied \nby a delayed version and the product is post integrated in order to \neliminate residual noise. This is a standard comparison detector\u2019? and \nis depicted in Fig. 13.\n\nAs has been done before, the heterodyned signal is represented in \nthe interval 0 <t < Tas, .\n\nwhere the subscript indicates a delay of T seconds, w; is the IF \nfrequency, and 6(t) = 6(t) \u2014 &(t), (A(t) is the transmitter laser phase \nnoise and \u00ae(t) is the local laser\u2019s phase noise.) Again, n(t) is a white \ngaussian noise process with unit double sided spectral density.\n\nV(t) = 2aA cos 6(t)cos wt + n,(t)cos wt + neo(t)sin wt \n\u2014 2aoA sin 6(t)sin w;t \n= [2a,A cos 6(t) + n,(t)]cos w;t \n\u2014 [2aA sin 4(t) + no(t)]sin wrt, | (142) \nVua(t) = 2a_,A cos ba(t)cos w;t + nyg(t)cos wt \n+ Nogsin wt \u2014 2a_,A cos 6g(t)sin w;t \n= [2a_,A cos \u00e9g(t) + nma(t)|cos w;t \n+ [2a_,A sin 6g(t) + neog(t)|sin w;t (143)\n\nThe baseband noise processes in (142) and (143), mi (t), mig(t), ne(t) \nand nog(t) are mutually independent and bandlimited to W hertz. \nSince the total noise power at the output of the band-pass filter is 4W, \nthe identical variances of the baseband noises must also equal to 4W. \nNow, multiplying V(t) by V(t), eliminating double frequency com- \nponents and integrating, results in the decision statistic qo,\n\nA detection error is made whenever aja_; = 1 and q S 0 or when \najoa_; = \u20141 and q < 0. So, the bit error rate then is just\n\nPe = Pr{g s Oj. (146) \nTo facilitate the calculation of (146) let, \nx(t) = 2A cos 6(t) + n(t) \nXxa(t) = 2A cos 6a(t) + me(t) \ny(t) = 2A sin 6(t) + neo(t) \nya(t) = 2A sin 6a(t) + neoa(t),\n\nwhere the u,; and v; are again the coefficients in the expansions of y(t) \nand u(t) in the eigenfunctions {y,,(t)}.\n\nStructurally (150) is identical to the quadratic forms obtained in the \nFSK case and therefore, we can express the bit error rate as\n\nThe eigenvalues as before are approximately 1 fork <n =2WT and. \nzero beyond and consequently (152) is to a good approximation,\n\nUnfortunately this integral cannot be expressed more explicitly as \nin FSK because of the essential singularity at z = \u20141. When \u00a2 = 0 (no \nphase noise), the essential singularity disappears and (154) is identical \nto the previously encountered integral associated with the error rate \nin FSK. So in this case we obtain exactly\n\nwhere P,,-; is again the (n \u2014 1)th order polynomial defined in Appendix \nB and P= A\u00b0T. \nIn the important case when 1/T = R > B_;, we obtain from (154), \nsetting n = I, | \npar BoP\n\nThis result, again as in the optical case, can be seen to be 3 dB inferior \nto the \u201cquantum limit\u201d that is solely attributed to the 3-dB loss in the \nheterodyning process.\n\nWe now return to the central problem of assessing the penalty \nincurred by DPSK due to the presence of phase noise. An exact \nevaluation of the penalty is not mathematically tractable in general, \nand, as in the previous case, we must resort to upper bounds using the \nmoment generating function of the quadratic form (153) or (154).\n\nThe inequality in (157) is valid because of convexity and the station- \narity of \u00a2(t). While the form of this moment generating function is \ndifferent from (129), similar techniques can be used to bound the \nprobability of error. We proceed as follows:\n\nSplitting the range of \u00a2 into two parts, R; such that \u00a2eR,, cos \u00a2 > 0 \nand R, such that \u00a2eR2, cos \u00a2 < 0, we write (159) as\n\nBy setting the derivative of f(z) to zero reveals that there exists an \noptimizing z > 0 namely,\n\n(Bri + Bro \nR | \nand where B,;, and Bz\u00bb are again the linewidths of the two lasers. \nFrom (162) and (137) we see that the degradation from ideal per-\n\nformance, exp(\u2014P), is essentially determined by the average value \nover the range \u00a2ceR, of\n\nWe now evaluate (167) asymptotically valid for P \u2014 \u00a9 and fixed C. \nSetting the derivative of (166) to zero, we solve the transcendental \nequation for the saddle point, 0 < \u00a2o = 7/2.\n\nUsing (170) we can summarize the error rate estimates developed in \nthe foregoing analysis.\n\n1. Optical DPSK. Substituting (170) into (137) with the definition \nof G(@) in (167), the probability of error is\n\nIgnoring unimportant coefficients, the probability of error is seen \nto be dominated by the maximum of the two terms in either (171) or \n(172). These are seen to be identical expressions barring the coeffi- \ncients. It is observed from (171) and (172) that the exponents in the \nsecond term approach the exponents of the first term as P > ~, This \ncan be verified from (168) since when P \u2014 \u00bb, C \u2014 o for fixed o% and \nso the solution, \u00a29 \u2014 2/2. As a consequence of this limit, our esti- \nmate of the error rate versus P has a threshold, or floor, at\n\nand according to our prediction the floor for optical DPSK is R/Br ~ \n100 while in heterodyne DPSK it is R/B, ~ 200, for identical laser \nlinewidths. These floor predictions are slightly pessimistic. The lower \nbounds in (123) predict a floor of R/B, = 67 for optical DPSK and \nR/B,z = 134 for heterotype DPSK. The discrepancy has to do with our \nbounding techniques. |\n\nIt is now possible to define the exponential degradation or penalty, \nfrom ideal performance (above the respective floors) in either case by\n\nThis is seen to be a function of the single parameter C defined in (167) \nand @p the solution to (168). |\n\nIt is important to observe that (171) and/or both (162) and (137) \nexhibit an exponential degradation due to phase noise unlike in the \ncase of FSK. In both systems widening the front-end bandwidths of \nthe respective detectors ensures minimal distortion suffered by the \nreceived heterodyned signals. However, in DPSK the static phase \nnoise differential manifests itself in an exponential degradation, while \nin FSK no such degradation occurs.\n\nIn Fig. 14 we exhibit the penalty function, (173), as a function of \nR/B, for both optical and heterodyne DPSK. We note that the \nexponential degradation is infinite below these respective floors. The \narrows shown on the figures at R/B, = 100 and 200 are aimed to \nemphasize that according to our estimates, the degradation is infinite \nat rates less than these values. In other words, no amount of additional \noptical energy can drive the error rate below these respective floors. \nSimilar curves can be drawn for different Pes and hence different Ps. \nA striking feature of these curves is that the 3-dB degradation is\n\nFig. 14\u2014Penalty in decibels due to phase noise in optical and heterodyne DPSK, \nPe ~ 10\u00b0, D = 20.\n\napproached very rapidly at rates greater than 400 times the laser \nlinewidths (less than 4 dB is given up in either system). Evidently \nDPSK is very sensitive to phase noise for R < 300B,.\n\nWe saw that on-off keying of an optical wave using direct detection \nachieves the quantum limit. Here we wish to analyze the performance \nof this modulation method when the on-off optical signal is first \nheterodyned to an IF frequency and then direct detected. Since the \nmicrowave version of on-off modulation has the same signal distance \nproperties as FSK, we expect similar performance. There are however \nsome differences and for the sake of completeness we include the \nfollowing analysis.\n\nLetting the IF frequency be w;, the differential phase noise be 6(t) \nand the resultant Gaussian noise with unit spectral density be n(t), \nthe observed IF microwave signal is then represented as\n\nA reasonable way to process (121) is to first bandlimit it, then \nenvelope detect the bandlimited version and finally postintegrate the \nsquare-envelope to obtain the decision statistic. This processor is \ndepicted in Fig. 15 and the output statistic is\n\nwhere n; and nz are baseband gaussian noise processes with double- \nsided spectral densities equal to 2 and are bandlimited to W hertz. \nThis must be so since the bandpass noise process\n\nIn writing (122) we assumed as before that W is sufficiently wide to \npass the in-phase and quadrature signals in the presence of phase \nnoise undistorted. \nNow let \nx(t) = 2A cos 6(t) + n,(t) \nand \ny(t) = 2A sin 6(t) + n.(t), (177)\n\nand make an expansion of x and y in terms of the prolate-spheroidal \northonormal set of functions\n\nWe note that {xn}, {yn}, {min} and {n2,} are mutually independent \nconditional Gaussian random variables (conditioned on 6(t)) with the \nfollowing parameters:\n\nwhere {A,} are the eigenvalues associated with the eigenfunctions y(t). \nOur method of estimating the error rate will be based on a Chernof\n\nTo proceed further with the analysis, we invoke again the excellent \napproximation regarding the behavior of the eigenvalues \\,. Here it \ncan be verified that\n\nand when this approximation is used in (180) and (181) we obtain for \nthe moment generating functions\n\nwhere 7 is a threshold that will be optimized later. The probabilities \nin (185) cannot be evaluated exactly; however, tight exponential upper \nbounds can be obtained as follows.\n\nUsing (183) in (186) it can be verified that there exists a 6 = 6, which \nmakes the bound tightest. By differentiation we find\n\nFollowing the same procedure for tightening the bound in (187) we \nfind an optimum 9p given by,\n\nEquating the exponent in (189) to the one in (191) we determine the \nbest threshold 7 given by\n\nwhich is identical to the FSK performance since here the average \noptical energy is half that of FSK. The exponential degradation from \nideal is seen to be\n\nwhere n = 2WT is the total band required to pass the modulated \nsignal with phase noise undistorted. A conservative example might \nbe as follows. P = 80 will yield an ideal error rate of ~107\u00b0, n = \n1/R(1 + 10B,) (10B; ~ 400 MHz should be sufficient to pass the \nsignal with phase noise undistorted.) For these parameters a simple \ncalculation reveals that when R = 0.5 GB/s the penalty is about 0.2 \ndB. At rates below this the degradation starts to be substantial. Note \nhowever, that this performance is still 3 dB poorer than DPSK and 6 \ndB poorer than the quantum limit.\n\nI wish to thank many of my colleagues at AT&T Bell Laboratories \nfor the many fruitful technical interactions during the course of this \nresearch. The contributions of Paul S. Henry have been invaluable. \nHe acquainted me with this subject matter, provided continuous \nguidance, and made significant and substantial contributions to the \nevolution of the main results. I would have been honored to have him \nas a coauthor but he modestly refused.\n\nI also had valuable discussions with the following: A. S. Acampora, \nN. Amitay, B. Glance, D. J. Goodman, L. J. Greenstein, G. J. Foschini, \nT. Li, L. A. Linke, B. F. Logan, R. W. Lucky, J. E. Mazo, A. A. M. \nSaleh, G. Vannucci, and A. D. Wyner. Thanks are also due to Vincent \nW.S. Chan from MIT Lincoln Laboratory for stimulating discussions.\n\n3. Y. Yamamoto and T. Kimura, \u201cCoherent Optical Fiber Transmission Systems,\u201d \nIEKE J. Quant. Electron., QE-17 (June 1981), pp. 919-35.\n\n4. T. Okoshi et al., \u201cComputation of Bit-Error Rate of Various Heterodyne and \nCoherent-Type Optical Communication Schemes,\u201d J. Opt. Commun., 2 (Septem- \nber 1981), pp. 89-96.\n\n5. T. Okoshi, \u201cHeterodyne and Coherent Optical Fiber Communications: Recent \nProgress,\u201d IEEE Trans. Micr. Th. and Tech., MTT-30 (August 1982), pp. 1138-\n\n.6. F. Faure and D. LeGuen, \u201cEffect of Semiconductor Laser Phase Noise on BER \nPerformance in an Optical DPSK Heterodyne-Type Experiment,\u201d Electron. Lett., \n18 (October 28, 1982), pp. 964-5.\n\n. S. Saito, Y. Yamamoto, and T. Kimura, \u201cS/N and Error Rate Evaluation for an \nOptical FSK-Heterodyne Detection System Using Semiconductor Lasers,\u201d IEEE \nJ. Quant. Electron., QE-19 (February 1983), pp. 180-93.\n\n. J. E. Mazo and J. Salz, \u201cOn Optical Data Communication via Direct Detection of\n\nJ. C. Campbell et al., \u201cHigh Performance Avalanche Photodiode with Separate \nAbsorption, Grading and Multiplication Regions,\u201d Electron. Lett., 19 (September \n29, 1983), pp. 818-9.\n\nC. H. Henry, \u201cTheory of the Linewidth of Semiconductor Lasers,\u201d IEEE J. Quant. \nElectron., QE-18 (February 1982), pp. 259-64.\n\nM. W. Fleming and A. Mooradian, \u201cFundamental Line Broadening of Single-Model \nGaAlAs Diode Lasers,\u201d Appl. Phys. Lett., 38 (April 1, 1981), pp. 511-3.\n\nC. Harder, K. Vahala, and A. Yariv, \u201cMeasurement of the Linewidth Enhancement \nNeda a of Semiconductor Lasers,\u201d Appl. Phys. Lett., 42 (February 15, 1983), pp. \n328-30.\n\nF. G. Walther and J. E. Kaufmann, \u201cCharacterization of GaAlAs Laser Diode \nFrequency Noise,\u201d Sixth Top. Mtg. Opt. Fib. Commun., New Orleans, February\u2014 \nMarch 1983, Paper TUJ5.\n\nJ. E. Kaufmann, \u201cPhase and Frequency Tracking Considerations for Heterodyne \nOptical Communications,\u201d Proc. Int. Telemetering Conf., San Diego, Calif., \nSeptember 1982.\n\nS. O. Rice, \u201cMathematical Analysis of Random Noise, W. Nelson, ed., Selected \nPapers on Noise and Stochastic Processes: Dover Publications Inc., 1954, pp. 145- \n62\n\nM. Shikada et al., \u201c100 Mb/s ASK Heterodyne Detection Experiment Using 1.3 pm \nDFB Laser Diodes,\u201d Conf. Opt. Fiber Commun., New Orleans, January 1984, \nPaper TUK6.\n\nK. Kikuchi, T. Okoshi, and R. Arata, \u201cMeasurements of Linewidth and FM-Noise \nSpectrum of 1.52 um InGaAsP Lasers,\u201d Electron. Lett., 20 (June 21, 1984), pp. \n535-6. |\n\nT. P. Lee et al., \u201cMeasured Spectral Linewidth of Single-Frequency 1.3 and 1.5 um \nInjection Lasers,\u201d Electon. Lett:, 20 (November 22, 1984), pp. 1011-2.\n\nR. Wyatt and W. J. Devlin, \u201c10 kHz Linewidth 1.5 wm InGaAsP External Cavity \nLaser with 55 nm Tuning Range,\u201d Electron Lett., 19 (February 3, 1983), pp. 110-\n\nA. L. Scholtz et al., \u201cInfra-red Homodyne Receiver with Acousto-optically Con- \ntrolled Local Oscillator,\u201d El. Lett., 19 (March 17, 1983), pp. 234-5.\n\nD. D. Falconer and J. Salz, \u201cOptimal Reception of Digital Data Over the Gaussian \nChannel with Unknown Delay and Phase Jitter,\u201d IEEE Trans. Inf. Th., /T-23, \nNo. 1 (January 1977), pp. 117-26.\n\nE. Wong, Stochastic Processes in Information and Dynamical Systems, New York: \nMcGraw-Hill, 1971.\n\nR. M. Gagliardi and S. Karp, Optical Communications, New York: Wiley and Sons \n1976, Chapter 6.\n\nR. Wyatt, T. G. Hodgkinson, and D. W. Smith, \u201c1.52 um PSK Heterodyne Exper- \niment Featuring an External Cavity Diode Laser Local Oscillator,\u201d Electron. \nLett., 19 (July 7, 1983), pp. 550-2.\n\nV. K. Prabhu, \u201cPSK Performance with Imperfect Carrier Phase Recovery,\u201d IEEE \nTrans. Aerospace and Elec. Syst., AES-12, No. 2 (March 1976), pp. 275-86.\n\nK. Kikuchi et al., \u201cBit-Error Rate of PSK Heterodyne Optical Communication \nSystem and its Degradation Due to Spectral Spread of Transmitter and Local \nOscillator,\u201d Electron. Lett., 19, No. 11 (May 26, 1983), pp. 417-8.\n\nV. M.S. Chan, L. L. Jeromin, and J. E. Kaufmann, \u201cHeterodyne Lasercom Systems \nusing GaAs Lasers for ISL Applications,\u201d IEEE Int. Conf. Commun., Boston, \nJune 1983, Paper E1.5.\n\nL. L. Jeromin and V. W. S. Chan, \u201cPerformance Estimates for a Coherent Optical \nCommunication System,\u201d Conf. Opt. Fiber Commun., New Orleans, January \n1984, Paper TUK2.\n\nR. Wyatt et al., \u201c140 Mb/s Optical FSK Fibre Heterodyne Experiment at 1.54 yum,\u201d \nElectron. Lett., 20 (October 25, 1984), pp. 912-3.\n\nK. Emura et al., \u201cNovel Optical FSK Heterodyne Single Filter Detection System \nUsing a Directly Modulated DFB Laser Diode,\u201d Electron. Lett., 20 (November \n22, 1984), pp. 1022-3.\n\nJ. E. Mazo and J. Salz, \u201cProbability of Error for Quadratic Detectors,\u201d B.S.T.J., 44 \n(November 1965), pp. 2165-86.\n\n35. D. Slepian and H. Pollak, \u201cProlate Sphyroidal Wave Functions\u2014Fourier Analysis \nand Uncertainty\u2014I,\u201d B.S.T.J., 40 (January 1961), pp. 43-63.\n\ndyne Differential Detection Simulation Experiment,\u201d Fourth Int. Conf. Int. Opt. \nand Opt. Fiber Commun., Tokyo, June 1983, Paper 30C3\u20144.\n\n38. G. Nicholson, \u201cProbability of Error for Optical Heterodyne DPSK System with \nQuantum Phase Noise,\u201d Electron. Lett., 20 (November 22, 1984), pp. 1005-7.\n\n39. R. W. Lucky, J. Salz, and E. J. Weldon, Jr., Principles of Data Communication, New \nYork: McGraw-Hill, 1968.\n\nAPPENDIX A \nAn Estimate of Probability of Error in Optical Homodyne Detection\n\nHere we derive an upper bound for the bit error rate which will be \nused in evaluating performance of optical homodyne as well as hetero- \ndyne reception. We need to estimate the following probability (eq. \n192), | |\n\nwhere \u00e9 and i are independent random variables with joint probability \ndensity p(\u00a3)p(v%). By definition, (194) is written,\n\noO x2 \nPes { ep(x)dx = Ee* = Eee = e2Ee* (196) \nTo make progress with (196), we note that since e~ is convex\n\n= \u2014p. (202) \nWe are thus assured that a \\ = Xo exists such that E(Xo) < 0. To find \nthis optimum value of 4, we examine,\n\nand conclude that when a/p = p, \\o = p. On the other hand when a/p \n< p, the optimum value of \\p is on the boundary \\o = a/p and so the \noptimum exponent in (201) is |\n\nEvaluation of Residues \nI am indebted to B. F. Logan, Jr. of Department 11219 for supplying\n\nmaterial included in this Appendix. \nWe require the residue of f,(z)/(\u2014z) at z = \u20141, where\n\nSetting P/2 = \\, z =t \u2014 1, the probability of error Pe(P, n) in Section \nV (eq. 118) is the coefficient of t\u201d in the Taylor series expansion of \nthe function\n\n\u201c5a 2 : ped, n)x*. | (208) \nThen, \ne \nPe(n, A) = panei Dri, n). \nThe generator function for the generalized Laguere polynomials, de-\n\nCross-Polarization Cancellation and Equalization \nin Digital Transmission Over Dually Polarized \nMultipath Fading Channels\n\nA theory for data-aided equalization and cancellation in digital data trans- \nmission over dually polarized fading radio channels is presented. The present \ntheory generalizes and extends previous work by admitting decision feedback \nstructures with finite-tap transversal filter implementations. Subject to the \nassumption that some past and/or future data symbols are correctly detected, \nformulas and algorithms for evaluating the least mean-square error for differ- \nent structures are presented. In a sequence of curves we evaluate and compare \nthe performance of various structures for a particular propagation model and \nseveral fading events. We find improvement in performance for decision \nfeedback over linear equalization. More importantly, we discovered that in \nthis application, as in the single-channel transmission case, decision feedback/ \ncanceler structures are much less sensitive to timing phase than linear equal- \nizers.\n\nOne of the purposes of this article is to call attention to mounting \nresearch results pointing the way toward effective methods for com- \nbating the deleterious effects of various impairments arising in digital \ndata transmission over dually polarized fading radio channels.\n\nTransmission of M-state Quadrature Amplitude-Modulated (QAM) \nsignals via orthogonally polarized carriers is an effective method for\n\nreusing existing bandwidth with obvious economic advantages. The \nmain obstacle in the way of realizing these advantages is the unavoid- \nable presence of Cross-Polarization Interference (CPI) between the \ndually polarized signals that arise due to multipath fading, antenna \nmisalignments, and imperfect waveguide feeds. The chief purpose of \nour current work is to obtain a fundamental understanding and a \nsolution to this problem.\n\nThere is a well established theory of linear and decision psaback \nequalization/cancellation to mitigate the effects of intersymbol inter- \nference (ISI) and noise in the transmission of a single digital signal.\u2019 \nHowever, consideration of data-aided CPI cancellation in addition to \nISI equalization in the presence of noise has not been treated before. \nThe work of Amitay and Salz\u2019 establishes a theoretical base for optimal \nlinear compensation of CPI and ISI in the presence of noise; however, \ntheir work is limited strictly to linear techniques and considers only \nideal infinite-tap transversal structures. |\n\nIn this article we generalize previous treatments of this subject in \ntwo major respects. Our first contribution is to cast the problem of \nCPI cancellation and ISI equalization in a general theoretical frame- \nwork that admits data-aided decision feedback techniques. Secondly, \nand most importantly, we admit finite-tap transversal structures that \nin practice can be implemented adaptively.\n\nThe receiver configuration is based on a matrix structure suggested \nby the theory of optimal detection and is shown in Figs. 1 and 2. The \noptimal structure is comprised of a linear matrix equalizer/canceler \nand an ISI and CPI estimator, which is used to subtract: some of the \ninterference from the received signals. An architecture, previously \nproposed by Kavehrad,\u00b0 is a special case of this generalized structure.\n\nThe dually polarized channel is modeled by a particular 4 x 4 real \nmatrix impulse response or its Fourier transform followed by additive \nnoise. The 2 X 2 block-diagonal elements of this matrix represent the \ncopolarized (in line) responses, while the off-diagonal 2 x 2 block \nentries represent cross-coupled and cross-polarized interfering re- \nsponses. Each matrix channel characterizes a snapshot of a multipath \nfading event, which in the presence of noise limits the achievable error \nrate of the receiver for a given data rate. We use a propagation model \nproposed in Ref. 2.\n\nIn comparing the performance of various equalizer/cancelers, the \nMean-Square Error (MSE) is used. The justification for using this \ncriterion has been amply discussed in the literature.\u2019\u201d But the chief \nmotivation for its use is due to its mathematical tractability. It turns \nout that it also leads to an exponentially tight upper bound on error \nrate. In practice, it lends itself to easy estimation and thereby is used \nto update transversal filter-tap coefficients recursively.\n\nSection II contains the system model and theoretical developments. \nComputational algorithms are provided in Section II], and our numer- \nical results and associated discussions are given in Section IV. Finally, \na summary is presented in the last section.\n\nConsider a dually polarized digital radio communications channel \nsupporting two independent QAM data signals. This type of commu- \nnication channel with an ideal QAM modulator and demodulator is \nshown in Fig. 1. The four independent synchronous data signals S,,(\u00a2), \nS(t), /= 1, 2, with the generic representation\n\namplitude modulate two linearly polarized carrier waves in quadrature. \nThe modulated signal,\n\nSi(t) = Sy (t)cos wot + So, (t)sin wot, (2a) \nis transmitted over the vertically polarized channel, while \nSr(t) = Sin(t)cos wot + Son(t)sin wot (2b)\n\nis transmitted over the horizontal channel. The carrier frequency is \nwo and the real data symbols\n\nare assumed to be independently drawn from a lattice of points with \nodd integer coordinates. The QAM constellations associated with eq. \n(2) are, therefore, rectangular. The scalar shaping pulse, g(t), is \nselected by the designer to satisfy limitations on transmitted power \nand bandwidth.\n\nThe individual transmission channels are characterized by bandpass \nimpulse responses or by their respective Fourier transforms,\n\nThe resolution of h,(t) and h;(t) into their respective baseband in- \nphase and quadrature components turns out to be convenient in our \napplication. :\n\nTo accommodate coupling between the polarized channels, two pairs \nof impulse responses, one associated with the cochannel and the other\n\nassociated with the cross-channel, are used to completely characterize \nthe medium.\n\nAt the output, two errr noises are added and the signal plus \nnoise is then coherently demodulated. The end-to-end system includ- \ning the modulators and the demodulators is shown in Fig. 1. It is \nconvenient to view this linear system as a four-input port four-output \nport network and characterize it by a 4 X 4 matrix impulse response \nor its Fourier transform, which is the overall system frequency re- \nsponse. 7 \nIt is now easy to verify that the I/O PeNaHONSHIpS can be expressed\n\nIn our application, however, it turns out to be more convenient to \nwork with the real matrix in eq. (6).\n\nWe now return to the J/O relationship in eq. (5) and substitute eq. \n(1) to obtain in more detail\n\nD(t) =); | a \u2014 nT)H(t \u2014 r)dr-A, + v(t), (9) \nwhere the real data symbol vector A, is given by \nQivun \nQoun . \nAn = : 10 \nQihn ( \nQehn\n\nIn an ideal system, eq. (11) would yield D(0) = Ao X constant. This \nresult is obtained when\n\n1. Ho = constant X J (J is the identity matrix), which implies that \nthe flat, or nondispersive, CPI vanishes;\n\nClearly, these requirements cannot be achieved in practice, and the \ndesigner of data communications systems must deal with these im- \npairments and find methods that minimize their effects on system \nperformance.\n\nA well-known approach? is the use of linear equalization. Our \nobjective here is to investigate a general cancellation technique in \nconjunction with linear equalization, which could potentially yield \nbetter performance than with just the linear equalizer alone. To this \nend, we begin our analysis by placing a linear matrix filter in cascade \nwith the channel prior to sampling, and we choose its characteristics \nso as to minimize the total MSE between the actual output sample \nand the desired output after canceling some CPI and ISI.\n\nDenote the matrix filter impulse response by W(t) and evaluate its \noutput at t = 0. This yields the column vector for the overall system \nresponse |\n\nTo describe our approach, we first discuss the following statistical \nproblem. Suppose that one observes the vector Do(0), eq. (13), and \nwishes to design the best processing strategy that estimates Ay in a \nsense of minimizing the probability of error. The precise solution to \nthis problem remains intractable because of the non-Gaussian nature \nof ISI and CPI. While the precise mathematical solution is unknown, \nsome qualitative aspects of the solution have been discussed.*\u201d It is \neasy to argue that the optimal detector structure consists of a matched \nfilter followed by a least-mean-square estimator of the interference, \nwhich is then subtracted from the matched filter output. After sub- \ntracting the estimate of the interference, the problem reduces to \ndetecting a known signal in additive Gaussian noise, which has a well- \nknown solution. The difficulty with this formulation, while physically \nappealing, is that the least-mean-square estimator of interference is \njust as difficult and intractable to evaluate as the detection problem \noriginally posed. One redeeming feature of this approach, however, is \nthat if one does not insist on least-mean-square estimation of inter-\n\nference, a reasonable detector structure can be determined. We argue \nthat constructing reasonable estimates of CPI and ISI, which are not \nnecessarily optimum, subtracting them from the incoming signal, and \nthen constructing an optimum detector essentially satisfies the spirit \nof the suggested optimal procedure.\n\nWe now formulate our approach more precisely. To start, assume \nthat over a finite set of sampling instants, S, vector data symbols, A,, \nn \u20ac S, are available at the receiver and before we make a final decision\n\nis subtracted from D,(0). Actually, this is feasible since prior to n = 0, \nsymbols have been decoded all along and what is presumed in our \nproposal is that we use the already-decoded symbols to improve on \nthe current estimate of Ap. Since practical systems are not realizable \nrelative to a large delay, there is a problem in using symbols that have \nnot yet occurred. This can be overcome by introducing a delay, making ~ \ntentative decisions, and then returning to modify the Ao decision.\n\nHow realistic is this assumption? The answer depends on the system \nerror rate prior to cancellation. For example, when the error rate is \n10\u00b0* and the cancellation window size is small relative to 10\u2018, the \nprobability that almost all of the symbols in this window have been \ncorrectly detected is fairly large. Thus, after cancellation, the error \nrate may be much improved. On the other hand, if the error rate prior \nto cancellation is high, no improvement after cancellation can be \nexpected since the estimation of the interference is not reliable. \nEvidently, decision-directed cancellation as proposed here is a boot- \nstrapping technique. It is very successful over a certain range of error \nrates and fails when the error rate is high. Unfortunately, these \nqualitative statements are extremely difficult to make precise, and it \nis necessary to rely on simulation results.\u00ae The assumption that A,, is \nknown in the canceler window will clearly result in optimistic perform- \nance predictions, and whether the predicted benefits can be realized \nmust be ascertained experimentally.\n\nWe now proceed to include this \u201cgenie\u201d in our mathematical anal- \nysis. As already stated, the performance criterion we use throughout \nthis work is the least MSE normalized to the transmitted symbols \nvariance, denoted o%. This is a mathematically tractable criterion to \nwork with, and by minimizing MSE, one also minimizes an exponen- \ntially tight upper bound on the error rate. Its use is also practically \nmotivated because it lends itself to easy estimation, and it can be used \nto update transversal filter-tap coefficients in practical adaptive sys- \ntems.\n\nReturning to the mathematical problem at hand, we define the error \nvector \u00ab as the difference between Do(0) minus the canceler output \nvector, and the desired vector data symbol, Ao,\n\n\u00a2 = Updo + X Undn\u2014 X CaAn + vo \u2014 Ao, (16) \na. n\u00e9s | \nwhere C,, represents canceler-tap values. Total MSE can be expressed \nas | |\n\nwhere \u201ctr\u201d stands for trace of a matrix, E{-} denotes mathematical \nexpectation with respect to all random variables, and * represents \ncomplex conjugate transpose.\n\nThe set of canceler matrices, C,, n \u00a9 S, can immediately be deter- \nmined. If they are not identically set to U,, they can only increase the \nvalue of MSE. Consequently, we set C,, = Un, n \u20ac S, and the residual \nMSE results in a functional of the matrix impulse response, wid), \nand the size of the cancellation window.\n\nThe minimization of MSE with respect to the matrix W(t) is \naccomplished by the use of the calculus of variations. After substituting \nfor U,,, defined in eq. (14), we get _\n\nwhere the matrices, nj, i,j = 1, --- , 4 have the entry \u201cI\u201d in the (/;)th \nposition and zero everywhere else. By computing the trace of eq. (21), \nwe obtain\n\nThe structure of Wo(r) is practically interesting. It consists of a \nmatched filter followed by a matrix-tapped delay line where the matrix \ntaps are zero for n \u20ac S. In other words, the linear transversal filter or \nequalizer specified in eq. (23) operates over a range of matrix-tap \ncoefficients where the canceler is not operative. This is to avoid \ninteraction between collocated taps and possible unstability problems. \nThe structure is shown schematically in Fig. 2. In practice, this \nstructure can be approximated and implemented by a finite transversal \nfilter whose taps can be adaptively updated.\n\nAfter post-multiplying eq. (23) by W'(\u2014r), integrating, and then \ncomparing the result with eq. (19), we get an explicit formula for the \noptimum MSE,\n\nwhere Up is obtained by solving a set of infinite linear equations \nobtained by post-multiplying eq. (23) by H(z \u2014 kT) and then inte- \ngrating. Thus,\n\nFig. 2\u2014Cross-polarization interference and intersymbol interference canceler block \ndiagram.\n\nTo evaluate the merits of our. system, we must have a solution for \nU,. The task of solving eq. (25) is rather complicated. It is made \ndifficult by the fact that the matrix equations are not specified over \nthe finite set, S. While the number of unknowns is infinite, the values \nat the gap window are not specified. A way around this dilemma was \nfound in the scalar case,* and with care applied to matrix manipula- \ntions, it is possible to adopt the same techniques here.\n\nWe proceed by first separating eq. (25) into two equations, one for \nk = 0 and the other for k \u00a5 0. Thus,\n\nwhere do; is the Kronecker delta function. The solution of eq. (28) is \nfacilitated by introducing a set of matrix variables {V,,}\u201d.. and a set of \nunknown matrices [An}=s : Pe these matrices, we write eq. (28) as\n\nFor these aati infinite sets of matrix sauation: to identically coin- \n| cide with eq. (28), the following constraints must hold: |\n\nIf these can be satisfied, the solution to eq. (81) will be identical to \nthe solution of eq. (28) with V, = U,, n \u20ac J, and this is the sole \npurpose for introducing new variables. Evidently, eq. (31) is easy to \n- solve since it is in a form of a convolutional equation. To this end \ndefine the inverse matrix sequences, {M\u2018}\u00ae.., as\n\nNow, insert this into eq. (31) to obtain explicitly the desired saliicion, \nV, = UI \u2014 Up) Y (R,\u2014 A)ME?, all n. (34)\n\nFrom this we can obtain a finite set of equations in the unknown | \nmatrices A,, since V, = 0 forn Ed,\n\n_ By substituting the definition of M, from eq. (30) into the left-hand \nside of eq. (85) and making use of eq. (33), we obtain the desired \nequations for the unknown constraint matrices Az, k \u20ac J, |\n\nwhere the last equality derives from the fact that V, = 0,n \u20ac J; V, = \nU,,n \u20acdJ; and R, = M,, n E Jd. Finally, by substituting eq. (27) into \neq. (37), we can write\n\nSubstituting this into eq. (24) provides an explicit expression for MSEp \nin terms of Ao only,\n\nOur effort in the following will be centered on determining Ao as a \nfunction of the cancellation window size, or the size of set J. | \n2.3 The matched filter bound\n\nWhen the canceler window is doubly infinite in extent, one obtains \nthe very best possible result. In other words, the genie has eliminated \nall ISI and CPI. In this special case, N, = \u2014o and No = \u00a9, and eq. \n(36) is now easy to solve since it reads ,\n\nTing \u2014 PMGSY = Yo ARMED, | (41) \nk=\u20140 \nBy evaluating the Fourier series of both sides of eq. (41), we obtain \nIT = 6\u2019? M~(6) + A(O)M (6), (42) -\n\nQa \u2014T \nSince M(6) = o7J + R(6) and M\u201c (6) is in fact the inverse, M-'(6), \nwe determine from eq. (42) that A(@) = R(@). Consequently, the zeroth\n\ncoefficient of A(0) is Ro = 1/(27) f=, R(6)d0, and when this is substi- \ntuted into eq. (40) we get the desired matched filter bound,\n\nThis will serve as a lower bound to attainable performance to which \nwe will compare all other results.\n\nIn this application it is assumed that all the causal terms, which \ndepend only on past decisions, are canceled in addition to a finite \nnumber of noncausal terms. This implies that Nz = \u00a9 and JN, is finite. \nWhen N, = 0, the canceler becomes a decision feedback equalizer\u2019 \nsince causal interference can be canceled by a feedback circuit. Here, \n\u2018we will determine MSE for the more general case when JN, is not \nnecessarily zero.\n\nTo treat this case it is more convenient to solve for Up directly from \neq. (28) rather than through eq. (36). Thus, we rewrite eq. (28) as\n\n-N, \n UrMn-n = (I - Uo)Rm, ms \u2014N,, (47) \nk=\u201400 . \nwhich is recognized to be a matrix Wiener-Hopf equation, and its \nsolution depends on being able to factor positive definite Hermitian \nmatrices.\u00ae | |\n\nThe validity of this expression and the existence of Mj; and M;, were \nfirst proved by Wiener and Akutowicz.?\n\nThe procedure for solving these is to first solve for Y(6) from eq. \n(49) in terms of M (0), an easy task in terms of the Fourier transforms \nof {M7} and {y,}. Having obtained Y(@), one proceeds to solve eq. (50) \nfor U(@) in terms of M*(6). Note that eq. (48) implies\n\nM(6) = M (6)M*(6), \nand since M(6) is Hermitian, M(#) = M\u2018(6), implying [M*(8)]' = \nM~(6), [M-(6)]* = M7*(6), and the factorization problem is reduced to \nfinding a matrix M*(6) such that\n\nwhere the entries in M*(6), [M*(6)]i; are such that [M*(@)], has a \nFourier series with only positive frequency coefficients. We shall later \ndiscuss algorithms for determining M*(0) from M(#)\u2014a rather com- \nplicated task.\u2019\u00b0\n\nWe now proceed to determine the sequence y,,. Multiply both sides \nof eq. (50) by M=,, and sum m from \u2014~ to \u2014N,. This gives the for- \nmula |\n\n-N, \n\u00bb VmMim = 2 U,M_x. (51) \nNow, recall that M, = R, + o76;ol, and, therefore, eq. (49) can be put \ninto the form\n\n1 0 \u20141 \n= oftr (2 Ma M)\") (58a) \nC n=-N, \nNotice that when N, = 0, that is, no anticausal cancellation, \nM2|\" \nMSE) = o2tr ; | , 7 (58b) \noO 7 3\n\nwhere My = Mo = (M@)\". This is the formula for decision feedback \nequalization derived by Falconer and Foschini for QAM transmission \nover a single channel, which they cast in a matrix formulation.\u201d \nEvidently, the form of the answer generalizes to arbitrary dimensions.\n\nThe theoretical results we have derived so far apply to an ideal \ncanceler of any window size and an infinite-tap linear equalizer whose \nmatrix taps vanish inside the cancellation window. To assess the \npenalties incurred by a finite-tap linear equalizer outside the cancel- \nlation window, we derive least MSE formulas applicable to this case. \nWith these formulas we will be in a position to evaluate the merits of \nequalization/cancellation using only a finite number of matrix taps \nand to gain insight as to how best to deploy the total number of \navailable taps. Also inherent in the theory derived so far is the \nindependence of MSE on sampling phase. This is so since the trans- \nversal equalizer/canceler is preceded by a matched filter whose struc- \nture presumes knowledge of sampling phase. Here, we shall relax this \ncondition and derive the MSE for a front-end filter matched to the \ntransmitter filter only rather than to the overall channel response and, \nthereby, bring out the dependence of MSE on timing phase.\n\nneF \nwhere the two sets F and S are disjoint and F now is a finite set, \n[F:n& F,n=\u2014-N, \u2014 M,, ---,-N, \u2014 1,0, No+1, ---,No+ Mo}. \nIn eq. (59), g(t), as before, is a scaler pulse shape, while {Q,},er is a\n\n4 xX 4 matrix sequence. The objective now is to select the Q,\u2019s that \nminimize the total MSE, eq. (19),\n\nSetting the derivatives of eq. (61) with respect to the elements of \nthe matrices {Q,}ner to zero, we get a set of linear matrix equations \nfor the unknowns, {Q,}ner, namely,\n\nThe solution of eq. (63) is straightforward and is discussed in a later \nsection.\n\nFor now, label the solution of eq. (63) by Q?\u2014the optimal Qs. \nPremultiply by Q\u00b0, sum over n \u20ac F, and substitute the result into eq. \n(63). This yields the desired formula for the least-mean-square error,\n\nThe next section will present computation algorithms for numeri- \ncally evaluating the formulas developed here.\n\nAn examination of Section III demonstrates that the theoretical \nanalysis of M-QAM signal transmission over dually polarized channels \nin the presence of multipath fading is a numerically intensive activity. \nIn this section we provide an overview of the major computational \nissues related to our investigation.\n\nWhen the linear equalizer in Fig. 2 has a finite-tap window size, the \noptimum receiver structure comprises a matched filter followed by a \nmatrix transversal filter and a matrix canceler. The most general case \nunder this assumption is when the matrix canceler has a finite number \nof causal and anticausal taps and the solution of eq. (36) for Ao provides \na means of calculating minimum mean-square error by use of eq. (40). \nTo solve for Ao, block matrices M\u2018~\u201d\u2019s defined in eq. (33) have to be \ndetermined first. One way to determine the M\u2018-?\u201ds is to solve eq. (33) \nby a Levinson-type algorithm\u2019? where the entries are block matrices. \nThus, matrix convolution eq. (33) is then represented as\n\nequation can be solved for M\u2018\u201ds, with the M;,\u2019s given in eq. (30). \nHaving the M\u2018-\u201d\u201ds and expressing eq. (36) in the form\n\nWhen the matrix canceler has knowledge of infinite past data \nsymbols, it becomes a decision feedback equalizer. In addition, it may \nalso employ a finite number of anticausal taps to operate on the future \nsymbols, in which case it becomes a finite window canceler. This can \nbe accomplished by a finite delay. As shown in eq. (47), to determine \nMSE\u00bb, a matrix Wiener-Hopf equation has to be solved. This involves \ndetermination of anticausal factors of the M(@) matrix as explained \nin Section 2.5. |\n\nThere are at least two computational algorithms available for solving \na matrix Wiener-Hopf equation. One method as introduced in Ref. 13 \nconverts the matrix that has to be factored directly into a nonlinear \ndifference equation of a Ricatti type, which converges to a stable \nsolution. Another method, which we adopt in our present work, is a \nBauer-type factorization of positive definite polynomial matrices.\u201c \nThis algorithm is suited to sampled data applications and takes \nadvantage of the periodic and positive nature of the channel covariance \nmatrix, M(@), as in this work. It performs the factorization in the \nfollowing steps. Suppose one desires to factor the n X n matrix M(@) \nas follows:\n\napproach 0, as |m| becomes infinitely large. One now follows the \nsteps: \nStep 1: Form the following variable-size Toeplitz matrix\n\nof respective size (m + 1)n X (m +1)n,m=0,---,thatis Hermitian \u2014 \nandreal. | \nStep 2: For every m = 0, perform the Cholesky\u2019s factorization\u201d\n\nwhere Lis the transpose of L,,, and L\u00bb is a square, real, and lower \ntriangular matrix with positive diagonal scalar entries,\n\nLe Le \nAa os ee 0 | \nIm =) 5 3 ae - (72) \nLe\u201d Lb\u201d er ee ee ee \nAll blocks in eq. (72) are real n X n, and all L\u00ae\u201d,r=0, ---, m, are\n\nThe fetdnsation of eq. (71) and the evaluation of M7_, in eq. (73) are \nperformed numerically. To find an upper bound on m for stopping the \ncalculations, a convergence point must be established. This can be \n- done by checking the trace of M;_, in each iteration to determine \nwhether it has reached a level of constancy and, if so, for what value \nof m. This completes the factorization. Indeed, in Ref. 14 it is shown \n- that a constant trace as a function of m corresponds to a minimum of \na quadratic functional:\n\nwhere T',, was defined in eq. (70) and x,\u2019s represent the elements of X. \nHence, there is a theoretical base for establishing the convergence \npoint. |\n\nFinally; we consider the case where the matrix linear equalizer \noperates on a finite set of taps that do not overlap with those of the \nfinite-tap matrix canceler. This is a case of great. practical interest. \nHere, the receive filter is assumed to have a square-root-Nyquist \ntransfer function\u2019\u00ae matching the transmit filter. Since it no longer \nmatches the overall channel and transmitter characteristics, MSEo is \na function of timing phase. Therefore, an optimum timing reference \nhas to be established before the optimum nonstationary covariance \nmatrix can be determined. This is accomplished here by minimizing \nthe mean-square eye closure (MS-EC), which is a measure of the \namount of received level perturbation caused by CPI and ISI.'\u00b0 In our \npresent work it is assumed that the demodulator removes the channel \nphase at the optimum sampling time reference.!\u00ae Once an optimal set \nof samples is found, the covariance matrix, Grm, of eq. (62) is formed \nas\n\nIn terms of the H,,\u2019s defined in eq. (12) the covariance matrix can be \nexpressed as\n\nHence, by adding a\u2019 to the diagonal elements of Gin, the matrix Ram \nis formed, as expressed in eq. (64). The Q,\u2019s, that is, the coefficients \nof the finite window equalizer, can be computed as follows:\n\nIn this section, the minimum mean-square error (MSEp) is evaluated \nfor the various techniques covered in the previous sections. We will \nfirst discuss a channel model, and then we will exhibit and discuss the \nbehavior of MSEp as a function of the number of equalizer/canceler \ntaps (M,, Mo, Ni, Ne) (see Fig. 2). |\n\nThe cross-polarization fading propagation model employed is the \none that is proposed in Ref. 2 and is briefly reviewed here. The \nfrequency characteristics of the propagation model are presented by \nthe complex matrix | |\n\nwhere the functional form of C,;(w) and C2o(w) is that of a single (in \nline) fading channel model documented by Rummler\u201d\u2019 with the generic \nrepresentation\n\nwhere a and p are real variables representing flat and dispersive fading \nlevels, @ is related to the fade notch offset, and 7 is the delay between\n\ndirect and reflected paths assumed to be 6.3 ns in this study. Also in \nthe model,\n\nwhich is in the same form as C;(w), except for an additional variable \nAw that allows noncollocated fade notches to occur on the two polar- \nization signal transfer characteristics. From Ref. 2, cross-polarized \npaths are assumed to behave as\n\nwhere K,, Kz, Ks, and Ks are constants that incorporate the nonideal \nproperties of antennas and waveguide feeds at both ends of the \nchannel, typically taking on values varying from one hop to another \nin the \u201435 to \u201420 dB range. The last term in eqs. (83) and (84) \nrepresents a nondispersive cross-polarization response contributed by \nan independent ray. In the present work, R3, Re, and Aw are assumed \nto be zero and the K;\u2019s are assumed to be \u201420 dB.\n\nComputation of the channel covariance matrix is the initial neces- \nsary step behind all the MSE, calculations. In the case of the infinite \nwindow-size equalizer discussed in Sections 3.3 through 3.5, the receive \nfilter is assumed to be a matched filter; hence, no reference timing \nestablishment is necessary. The peak of the correlation function serves \nas a timing reference. By computing the sampled correlation matrix \nof eq. (26), we can proceed with the normalized MSE, calculations as \nexplained in previous sections. |\n\nBy applying the finite window equalizer, as discussed in Section 3.6, \na set of optimum samples of overall impulse response is found by \nestablishing a timing reference, to, for which the MS-EC of the received \nin-line signal is a minimum, and at this reference, the channel phase \nis removed.\u2019\u00ae This has to be done for the two polarized signals inde- \npendently.\n\nwhere C(w) is the propagation transfer matrix and P(w) is the diagonal \nNyquist-shaping filter transfer matrix. Now, for instance, if the im- \npulse response of the vertical in-line signal is\n\nand the upper row block matrices of the overall impulse response \nmatrix have to be multiplied by exp(\u2014j @(t)) < J (J being the unity \nmatrix) in order to remove the channel phase at t). With this back- \nground, we now present the numerical results in the following subsec- \ntion.\n\nTo provide a single set of curves for MSEo, independent of the \nnumber of transmit states in M-QAM signal space, we normalize \nMSE, as defined in eqs. (40), (43), (46), (58a), (58b), and (65) by \ndividing the formulas by o4, that is, the transmitted symbols variance. \nIn addition, we only compute the normalized MSE\u00bb for one of the M- \nQAM signals that comprise the dually polarized signals, namely, S,(t).\n\nIf one defines the unfaded signal-to-noise ratio (s/n) by I\u2019, it can be \nverified that in the case of a matched filter receiver, the normalized \nMSE\u00bb in the absence of any cross-polarization interference (K,, Ko, \nKy, Ks, R3, Re = 0) is simply\n\nOd \nand, consequently, for a large unfaded s/n, it becomes I~\u2019. Hence, eq. \n(88) establishes an ultimate performance bound that can only be \nachieved in a utopian environment. This reference will be our baseline \nin the following evaluations. In a dually polarized system with a finite \namount of nondispersive coupling (Ky, Ko, K4, Ks > 0), the matched \nfilter bound is degraded somewhat. For K, = Ky = K, = Ks = \u201420 dB \nwe found a small amount of degradation in the ideal MSEo, which is \nnot a function of the dispersive fade depth and only diminishes when \nthere is no cross-coupling, that is, in a completely orthogonal system.\n\nIn all that follows it is assumed that the transmit filter is square- \nroot Nyquist shaped,\u2019\u00ae and the receive filter either matches the overall \ntransmitter and channel or the transmitter only. A Nyquist roll-off of \n45 percent, both a 40- and a 22-MHz channel bandwidth, and an s/n \nof 63 dB are used in our numerical evaluations.\n\nFigure 3 depicts the normalized MSE\u00bb as a function of the number \nof canceler taps, Q, when a 40-dB centered fade over a 22-MHz channel \nband is applied to both polarized signals. The linear equalizer in this \ncase possesses an infinite number of taps. The case of pure linear \nequalization (N, = Nz = 0), no cancellation, exhibits the largest MSEo\n\nFig. 3\u2014 Optimum normalized MSE versus number of canceler taps for a 40-dB \ncentered fade over a 22-MHz channel.\n\ndegradation relative to the asymptotic matched filter bound. This is \ndue to the noise enhancement experienced by the linear equalizer \nduring deep fades. When both causal and anticausal canceler taps are \npresent, all the curves rapidly approach the matched filter bound for \na finite constant coupling (K; = \u201420 dB, i = 1, 2, 4, 5). The curve for \na decision feedback type canceler starts at an ideal decision feedback \nequalizer normalized MSE\u00bb and approaches the asymptotic value with \ntwo anticausal taps. The finite window size canceler curve starts at \nthe linear equalizer case (Ni; = Nz = 0) and reaches the matched filter \nbound asymptotic value with a total of four causal/anticausal taps. \nFinally, when no anticausal taps are employed the curve asymptoti- \ncally approaches the ideal decision feedback case with only two causal \ntaps.\n\nIn Fig. 4, we depict results similar to Fig. 3 for the case when the \ncentered fade notch depth is reduced to 20 dB over a 22-MHz channel.\n\nFig.4\u2014Optimum normalized MSE versus number of canceler taps for a 20-dB \ncentered fade over a 22-MHz channel.\n\nAs can be observed, the linear equalizer (N; = No = 0) performance is \nimproved. In both figures the fade notch is located at the band center; \nhowever, since in both cases the receive filter matches the overall \nchannel and transmitter, an offset fade notch does not have a serious \nimpact on the results for the same fade notch depth.\u2019\u00ae\n\nIn Figs. 5 and 6 we depict the achievable MSE\u00bb when the linear \nequalizer has a finite number of taps. The fade notch in Fig. 5 is \ncentered, but in Fig. 6 it is offset from the band center. For ease of \npresenting the results in our work, fade notch offset from the band \ncenter is expressed in terms of the ratio of the fade notch distance \nfrom the band center to the channel equivalent baseband bandwidth \nin percentage. In Fig. 6, the fade notch is offset by 69 percent over a \n22-MHz channel, that is, an offset of 7.6 MHz from the band center. \nAs observed from Fig. 5, a total of nine taps (including the center tap) \nare required to achieve the asymptotic matched filter bound when \ndecision feedback taps are present. It is interesting to note that the \nsame asymptotic performance can be achieved no matter how the nine \nsynchronously spaced taps are deployed between the linear equalizer\n\nFig.5\u2014Optimum normalized MSE versus number of canceler taps for a 40-dB \ncentered fade over a 22-MHz channel.\n\nand the canceler as long as the canceler operates in a decision feedback \nmode. This is because the equalizer and canceler-tap windows comple- \nment one another; therefore, since the taps do not overlap, for the \nsame number of taps, the performance remains almost the same in \nthe decision feedback cases. An important configuration is when the \nlinear equalizer operates only on the main lobe of CPI by means of its \ncenter matrix taps (M,; = Mz = 0). This is a single tap linear equalizer \nstructure as opposed to the single tap decision feedback CPI canceler \nproposed by Kavehrad.? It is clear that as long as the canceler window \nis sufficiently wide, a main lobe CPI canceler can achieve the asymp- \ntotic matched filter bound. The curves again indicate that deep fades \ndegrade the linear equalizer (N; = No = 0) performance significantly.\n\nFig. 6\u2014Optimum normalized MSE versus number of canceler taps for a 40-dB, 69- \npercent offset fade over a 22-MHz channel. |\n\nPrevious studies\u2019? showed that every 3-dB degradation in MSEp \ntranslates into a loss of 1 bit/s/Hz of data rate efficiency. Hence, \nlinear equalization may not prove\u2019 adequate rate efficiency in deep \nfades. \u00b0 |\n\nIn Fig. 6 we depict curves similar to Fig. 5, except for a 40-dB offset \nfade with the notch frequency offset by 69 percent. Improved perform- \nance turns out to be due to the particular notch position as is brought \nout in the discussion of Fig. 8.\n\nFigure 7 illustrates a similar set of curves for a 20-dB centered \nfading of dually polarized signals over both a 22- and a MHz \nchannel. As can be observed, the linear equalizer (N; = =) \nperformance improves because of the decreased fade depth; ic\n\n_ is more, as expected. This is due to the wider channel band over which \nthe same fade notch depth causes more dispersion. The degradation \namounts to 2.2-dB loss of MSE\u00bb comparing to a matched filter bound, \nthat is, roughly 1 bit/s/Hz loss of data rate efficiency, and the loss can \neven be more for offset fades, as will be seen in Fig. 8. Hence, even\n\nFig. 7\u2014Optimum normalized MSE versus number of canceler taps for a 20-dB \ncentered fade over a 22-MHz and a 40-MHz channel.\n\nwith more typical fades the use of the linear equalizer can be trouble- \nsome over a 40-MHz channel.\n\nM, = M,=4 | \nN, = Nz = 0 (no cancellation). \n2. Center tap only linear equalizer/finite window canceler with \nM, = M2. =0 \nN, = No = 4.\n\nnounced. The center tap equalizer with a finite window canceler \nexhibits a very small sensitivity to offset fades. The degradation of \nMSE\u00bb) for some offset fades can be explained considering the fact that \nthese fades cause cross-coupling of the imaginary part of a complex \nQAM signal into its real part, and for a particular notch offset \nfrequency within the band, the coupling reaches its maximum. There- \nfore, the MSE\u00bb versus fade notch offset curves exhibit this phenome- \nnon. In dually polarized systems, as in the case of the problem at \nhand, this is even more pronounced than in single signal transmission, \nbecause in the 4 X 4 system under offset fading there is coupling of \nthree interfering data streams into the fourth one. A decision-feed- \nback-type canceler structure, by canceling the major contributors to \nCPI and ISI and with a lesser noise enhancement, exhibits an improved \nperformance compared with the linear equalizer. Note that all the \ncurves in Fig. 8 have been obtained under optimum timing conditions.\n\nTo investigate the sensitivity of the two structures to timing phase, \nwe plot in Fig. 9 the normalized MSE\u00bb and superimpose the normalized \nMS-EC of the received signal before equalization/cancellation as a \nfunction of sample timing offset from the optimum timing reference.\n\nFig. 9\u2014Sensitivity to timing phase for a 40-dB fade notch, offset by 34.5 percent, \nover a 22-MHz channel.\n\nThis is done for a severe fade, namely, a 40-dB fade with a notch \nfrequency offset by 34.5 percent over a 22-MHz channel. It is clear \nthat the finite linear equalizer is much more sensitive to timing phase \nthan the decision feedback type. This was previously shown in a paper \nby J. Salz\u2019\u00ae for infinite window structures in single-signal transmis- \nsion. We demonstrated the concept here for dual-polarization trans- \nmission and finite window architectures. Notice that in Fig. 9 the \noptimum timing reference is established based on minimizing the MS- \nEC of the received signal in presence of fading, before CPI and ISI \ncancellation; hence, after cancellation occurs this timing reference \nmay not be the one that minimizes the canceler output MSE, and\n\nindeed the linear equalizer curve on Fig. 9 indicates this fact. As seen, \nthe MS-EC curve has a minimum at the optimum timing reference. \nThe sensitivity of the matrix linear equalizer to timing phase can be \nreduced by applying half-a-baud spaced taps, that is, by deploying \nfractionally spaced taps.\u201d\u00b0 Also, a decision feedback timing method \nmay prove more robust\u201d than the minimum MS-EC timing that has \nbeen adopted in our work. Needless to say, that decision feedback \ntiming method is more complex in terms of implementation than the \nminimum MS-EC timing, which can essentially be implemented at \nintermediate frequency. The degradation in MSHp seen in Fig. 9 is \npartly attributed to the asymmetric amplitude and delay responses of \nthe fading channel that in the presence of a nonzero roll-off shaping \nfilter cause a destructive addition of aliases.\u2019\u00ae\n\nCurrent work generalizes and extends previous results\u201d* in the \nfollowing respects. Data-aided decision feedback and canceler struc- \ntures, known to be effective in single-channel data transmission, are \n_adopted and included in our class of receiver structures. As a practical \n_ feature, we admit transversal filter realizations with a finite number\n\nBecause of the departure from ideal linear infinite structures con- \nsidered previously, we encountered extremely difficult numerical prob- \nlems, which we addressed and solved.\n\nThe dually polarized digital radio channel is modeled as a four-input \nport, four-output port linear. network followed by additive noise. We \n\u2018determine the optimum admissible receiver structures when the trans- \nmitted signals are two independent M-state QAM digital data signals.\n\nThe mathematically tractable criterion, the MSE, is used through- \nout our work. This figure of merit has several redeeming features in \naddition to its being mathematically tractable. For one, it can be used \nto determine a sharp upper bound on error rate. More importantly, it \nis the quantity which is estimated in practice to provide information \nfor updating tap coefficients in adaptive systems.\n\nThe receiver structure that minimizes the MSE consists of a matrix \nmatched filter in cascade with a transversal filter combined with an \nintersymbol interference as well as a cross-polarization interference \ncanceler. The canceler uses the detected data symbols to estimate the \ninterference to be canceled. This is a major assumption on which our \nresults rest. Since data-aided operations presume correct knowledge \nof detected data symbols and since wrong decisions will be occasionally \naccepted, our proposal is necessarily a boot-strapping approach. Thus, \ncancellation is only feasible when tentative decisions are correct most\n\nof the time, and yet the error rate is not sufficiently low to meet \nsystem specifications. Our approach makes possible the reduction of \nthe final error rate to an acceptable level. In circumstances where the \ninitial error rate is very high (=1077), our proposal will not work, and \nin order to ensure availability of reliable data symbols, one option is \n_to dedicate a small fraction of the main data frame to a sequence a \npriori known to both transmitter and receiver for proper acquisition \nof data.\n\nWe use the assumption of the availability of correct data symbols \nto derive our main results. These are expressions for minimum attain- \nable MSE as a function of various system parameters and numerical \nalgorithms for evaluating the mathematical formulas\u2014a rather inten- \nsive activity because of the large number of matrix equations that has \nto be solved. Inclusion of the effects of errors in the feedback/canceler \nloops has proved so far to be mathematically intractable.\n\nFrom our extensive numerical work, which is exhibited in a sequence \nof graphs, we draw these major conclusions:\n\n1. For a reasonable copolarized and cross-polarized propagation \nmodel? and a severe centered fade\u201440-dB notch depth with a second- \nary ray delay of 6.3 ns over an approximately 22-MHz channel band- \nwidth\u2014the performance of transversal filters with a finite number of \ntaps deployed in a decision feedback/canceler structure is substantially \n(6 dB) better than linear equalization, and the difference can be up to \n10 dB for offset fades: It can be shown that a 3-dB increase in MSE \ntranslates into about 1 bit/s/Hz decrease in data rate at a fixed error \nrate or an order of magnitude increase in outage probability. Hence, \nlinear equalization may not be adequate in deep fades. Whether this \ngain can be realized in practice depends on the degree of error propa- \ngation. This is difficult to assess mathematically and must be studied \nby computer simulation and/or by experimentation.\n\n2. Decision feedback/canceler structures achieve the ultimate \nmatched filter bound with only nine matrix taps provided that error \npropagation is neglected.\n\n3. Nine linear equalizer taps essentially achieve the performance of \nthe infinite-tap linear equalizer. This method is, of course, free of error \npropagation. |\n\n4. For milder centered fades \u201420- dB depth with a secondary ray \ndelay of 6.3 ns\u2014the linear equalizer configuration with nine taps is \nonly 1-dB inferior to the decision feedback structure over a 22-MHz \nchannel. However, if the channel bandwidth is increased to 40 MHz, \nthe performance of the linear equalizer is worse than that of the \ndecision feedback structure by 2.2 dB, and the difference can be up to \n- 8 dB for offset fades.\n\n5. Decision feedback/canceler configurations are less sensitive to \ntiming phase than linear structures.\n\nDiscussions with N. Amitay; G. J. Foschini; L. J. Greenstein; D. J. \nGoodman; T. Kailath and Hanoch Lev-Ari of Stanford University; N. \nKazanjian; A. M. Saleh; and D. C. Youla of Polytechnic Institute of \nNew York were extremely valuable during the course of this work.\n\n1. G. J. Foschini and J. Salz, \u201cDigital Communication Over Fading Radio Channels,\u201d \nB.S.T.J., 62, No. 2, Part I (February 1983), pp. 429-59.\n\n2. N. Amitay and J. Salz, \u201cLinear Equalization Theory in Digital Data Transmission \nOver Dually Polarized Fading Radio Channels,\u201d AT&T Bell Lab. Tech. J., 63, \nNo. 10 (December 1984), pp. 2215-59.\n\nM. Kavehrad, \u201cBaseband Cross-Polarization Interference Cancellation for M-Quad- \nrature Amplitude- Modulated Signals Over Multipath Fading Radio Channels,\u201d \nAT&T Tech. J., 64, No. 8 (October 1985), pp. 1913-26.\n\nM. S. Mueller and J. Salz, \u201cA Unified Theory of Data-Aided Equalization,\u201d B.S.TWJ., \n60, No. 11 (November 1981), pp. 2023-38.\n\nT. Kailath, \u201cA General Likelihood-Ratio Formula for Random Signals in Gaussian \nNoise,\u201d IEEE Trans. Inform. Theory, [T-15, No. 3 (May 1969), pp. 350-61.\n\nA. Gersho and T. L. Lim, \u201cAdaptive Cancellation of Intersymbol Interference for \nData Transmission,\u201d B.S.T.J., 60, No. 11 (November 1981), pp. 1997-2020.\n\nJ. Salz, \u201cOptimum Mean- Square Decision Feedback Equalization,\u201d B.S.T.J., 52, \nNo. 8 (October 1973), pp. 1341-73.\n\nD. C. Youla, \u201cOn the Factorization of Rational Matrices,\u201d IRE Trans. Inform. \nTheory, J T-7 (July 1961), pp. 172-89.\n\nN. Wiener and KE. J. Akutowicz, \u201cA Factorization of Positive Hermitian Matrices,\u201d \nJ. Math. Mech., 8 (August 1959), pp. 111-20.\n\n11. D. D. Falconer and G. J. Foschini, \u201cTheory of Minimum Mean-Square-Error QAM \nSystems Employing Decision Feedback Equalization,\u201d B.S.T.J., 52, No. 10 (De- \ncember 1973), pp. 1821-49.\n\n12. N. Levinson, \u201cThe Wiener RMS (Root Mean Square) Error Criterion in Filter \nDesign and Prediction,\u201d J. Math. Phys., 25 (February 1946), pp. 261-78.\n\n13. W. G. Tuel, \u201cComputer Algorithm for Spectral Factorization of Rational Matrices,\u201d \nIBM J. Res. Develop., 12 (March 1968), pp. 163-70.\n\n14. D. C. Youla and N. Kazanjian, \u201cBauer-Type Factorization of Positive Matrices and \nthe Theory of Matrix Polynomials Orthogonal on the Unit Circle,\u201d IEEE Trans. \nCircuits Syst., CAS-25, No. 2 (February 1978), pp. 57-69.\n\n15. G. H. Golub and J. H. Welsch, \u201cCalculations of Gauss Quadrature Rules,\u201d Math. \nComp., 26, No. 106 (April 1969), pp. 221-30.\n\n16. R. W. Lucky, J. Salz, and E. J. Weldon, Principles of Data Communication, New \nYork: McGraw-Hill, 1968.\n\n17. W. D. Rummler, \u201cA New Selective Fading Model: Application to Propagation Data,\u201d \nB.S.T.J., 58, No. 7 (May-June 1979), pp. 1037-71.\n\n18. M. Kavehrad, \u201cAdaptive Decision Feedback Cancellation of Intersymbol Interfer- \nie Over Multipath Fading Radio Channels,\u201d Proc. ICC (June 1983), pp. 869-\n\n19. J. Sale, \u201cOn Mean-Square Decision Feedback Equalization and Timing Phase,\u201d \nIEEE Trans. Commun., COM-25, No. 12, pp. 1471-76, December 1977.\n\n20. G. Ungerboeck, \u201cFractional Tap-Spacing Equalizer and Consequences for Clock \nRecovery in Data Modems,\u201d IEEE Trans. Commun., COM-24, No. 8 (August \n1976), pp. 856-64.\n\n21. G. R. McMillen, M. Shafi, and D. P. Taylor, \u201cSimultaneous Adaptive Estimation \nof Carrier Phase, Symbol Timing, and Data for a 49-QPRS DFE Radio Receiver,\u201d \nIEEE Trans. Commun., COM-32, No..4 (April 1984), pp. 429-43.\n\nCross-Polarization Interference Cancellation and \nNonminimum Phase Fades\n\nThis paper reports on a continuation of the work presented in Ref. \n1, that is, cancellation of cross-polarization interference by a trans- \n. versal structure at baseband when two M-state quadrature amplitude- \nmodulated (M-QAM) signals are transmitted over the same channel \nusing orthogonal field polarizations. In this paper we extend the scope \nof the previous study, which only dealt with minimum phase fades, to \ninclude evaluation of nonminimum phase fade effects encountered in \ntransmission over multipath fading channels such as those experienced \nin line-of-sight terrestrial radio applications. _\n\nRecent studies** have examined IF and baseband equalizers in the \ntransmission of a single QAM signal under nonminimum phase fades. \nFor example, Ref. 4 examines transition distortions when the fade \nphase changes from minimum to nonminimum. In particular, Ref. 2 \nconcludes that, in mitigating intersymbol interference, baseband \nequalizers outperform IF structures used for this purpose in the\n\npresence of nonminimum phase fades. Since, as pointed out in Ref. 2, \non the average 50 percent of deep selective fades take on a nonmini- \nmum phase, we examine the proposed baseband interference canceler \nperformance\u2019 in nonminimum phase fades. We use the static modeling \nof the dual-polarized channels introduced in Ref. 5 to emulate snap- \nshots of fadings of the main (reference) and the cross-coupled paths. \nThis simple model follows the results of recent measurements cited in \nRef. 5. | |\n\nIn this study, we first show that some of the nonminimum phase \nfade channel models available in the literature that were meant to be \nused in single-signal transmission may not be appropriate for appli- \ncation in multicarrier transmission systems such as dual-polarized or \nspace-diversity systems. Then, employing what seems to be the proper \nstatic model for nonminimum phase fades, we evaluate dual-polarized \nsystem performance with and without the proposed canceler.\u2019 It is \ndemonstrated that, for a proper model, the type of fading has no \nimpact on the performance of the baseband cross-polarized interfer- \nence canceler, as long as the fadings of the reference (main) copolarized \npath signal and the cross-coupled path signal are of the same type, \ni.e., both minimum or both nonminimum phase. As in Ref. 1, dual- \npolarized system performance signature (M-curve) is used as a measure \nin the performance evaluation.\n\nIn this section we compare three commonly used static models of \nnonminimum phase fades and demonstrate that only one of them \nseems to be appropriate for application in dual-polarized systems. The \nfirst model, which has been used in Ref. 4 and some other studies, \nstates that if for minimum phase fades we employ a two-ray transfer \n_ function shown by | |\n\n| H(w) = a{l \u2014 6 exp(\u2014j (w \u2014 wo)7)], 0<b<1, (1) \nin which b represents the fade notch depth and w is its displacement \nfrom the midband frequency, 7 is the delay between the arrival times\n\nof the two rays, and a is the flat fade level, then for nonminimum \nphase fades we have to use\n\nin which 0\u2019 is the fade notch depth. The second model suggests using \nH(w) = alb \u2014 exp(\u2014] (w \u2014 wo)r)], O<sb<1 (3).\n\nComparing equations (1) through (4), it is obvious that the magnitudes \nof the transfer functions are the same at the fade notch frequency in \nall three different models and are all equal to\n\nFor the magnitude to be the same at the fade notch frequency in eq. \n(2), the value of b\u2019 has to be set to (2 \u2014 b) so one can compare \nminimum and nonminimum phase fades of equal magnitude. This \nstems from the fact that the minimum phase fade depth here is defined \nas\n\nand for the nonminimum phase fades of eq. (2) to have comparable \nmagnitudes with minimum phase fades, b\u2019 has to be equal to (2 \u2014 b). \nAlthough the fade magnitudes are all the same, the envelope delay \nresponses of the different transfer functions cited above are not. That \nis exactly why one has to be careful in choosing a model to form static \nnonminimum phase fades in multicarrier channels. The envelope delay \nresponse as a function of frequency for the minimum phase transfer \ncharacteristic in eq. (1) can be found by using\n\nTuer(w) = \nwhere the subscript MPF stands for Minimum Phase Fade. In the \nfollowing, for simplicity we use midband fades to prove our point, i.e., \nwhen w in eq. (6) is equal to wo. The conclusions arrived at, however, \nare felt to be general. Hence, the envelope delay of a midband minimum \nphase fade from eq. (1) is\n\nFor the first, second, and third nonminimum phase fade transfer \nfunctions the envelope delay response is\n\n_ respectively, where the subscript NMPF stands for nonminimum \nphase fade. Notice that the first and the second models introduce \nconstant delays of 27 and 7 in the envelope delay response, respec- \ntively, and these delays have nothing to do with the nonminimum \nphase fade nature. To prove the point, in Fig. 1 we show the overall \nimpulse response of a single minimum phase faded channel using a \n7.5-dB midband fade notch. As seen, the optimum timing reference, \ntopt, is to the left of the unfaded signal optimum timing reference, i.e., \nthe time origin. This is of course due to the negative delay character- \nistic. Using a 7.5-dB midband nonminimum phase faded transfer \nfunction of the third model, the overall impulse response of the system \nfor 45-percent roll-off Nyquist-shaped transmit/receive filters is \nshown in Fig. 2. Note that the optimum time reference is now to the - \nright of the time origin owing to the positive delay character, and the \nimpulse response is time reversed about the timing reference when \ncompared to the corresponding one for the minimum phase fade of \nFig. 1. The first or the second model will translate to the right the\n\nFig. 1\u2014In-phase received impulse response of a single-carrier system for a minimum \nphase fade.\n\nFig. 2\u2014In-phase received impulse response of a single-carrier system for a nonmini- \nmum phase fade.\n\nposition of the optimum timing reference of the nonminimum phase \nfaded impulse response of Fig. 2 by 27 or 7, respectively. Such a \ntranslation is harmless in single-carrier systems because once the \noptimum timing reference is established, taking samples of the impulse \nresponse at every baud period results in the same set of samples for \nminimum and nonminimum phase fades regardless of the actual | \nposition of the timing reference. However, in dual-polarized, or gen- \nerally in multicarrier systems, the position of the timing reference can \nbe as important as the sampling points in time. To demonstrate this, \nwe use the dual-polarized channel model shown in Fig. 1 of Ref. 1, and \nillustrate the received in-phase impulse responses formed by the main \ncopolarized and the cross-coupled paths in Fig. 3. These responses are \ncomposed of the real part of the main-path signal U;1, cross-coupling \nof the main-path quadrature phase part U,1, cross-coupled in-phase \npart of the cross-coupled path U;1, and cross-coupled quadrature \nphase part of the cross-coupled path U,1. Using notations of Refs. 1 \nand 5, the in-phase received waveforms shown in Fig. 3 correspond to \n7.5-dB midband minimum phase fading of the main-polarization path \nand a cross-coupled path fading of (\u201420., 0., 0.),* that is, an interfering \ncross-coupled signal with a flat fading level of 20 dB below the main- \npolarized signal level. Hence, the main-path impulse response timing\n\nFig. 3\u2014In-phase received impulse responses of dual-polarized system for a midband \nminimum phase fade on the main path and a flat fade on the cross-coupled path.\n\nFig. 4\u2014In-phase received impulse responses of dual-polarized system for a midband \nnonminimum phase fade on the main path and a flat fade on the cross-coupled path.\n\nreference is to the left of the time origin because of its minimum phase \nfading, and the cross-polarized path impulse response is an attenuated \nNyquist pulse centered at the origin because of its flat fading, as \nexpected. The corresponding shapes for an equivalent nonminimum \nphase fade of the main path, using the third model, are shown in Fig. \n4. As seen, the interferer impulse-response shape does not change; \nhowever, the main-path impulse-response timing reference moves to \nthe right of the time origin, and the impulse response shape itself is\n\ntime reversed. Now in the dual-polarized system, once the receiver \ntiming recovery circuit locates the optimum time reference by mini- \nmum-peak-distortion or minimum-mean-square error criterion, the \nmain and the interfering impulse responses are sampled using the \nsame reference timing, as explained in detail in Ref. 5; hence, the \nunnecessary time shifts inherent to the first and the second nonmini- \nmum phase fade models will result in establishing an incorrect refer- \nence point and a false set of samples of the cross-coupled interfering \nimpulse responses. For further illustration, the displacement of the \noptimum timing reference from the origin as a function of the fade \nnotch offset for different fade notch depths is shown in Fig. 5. Clearly, \nthe upper half-plane represents nonminimum phase and the lower half \ncorresponds to minimum phase fades. We employed the second non- \nminimum phase fade transfer function model in this case. As seen, the \ntwo sets of curves are vertically symmetrical except for a constant \ntime shift of 6.3 ns which happens to be the value of 7 used in this \nexample. The constant time shift is what the transfer function model\n\nFig. 5\u2014Constant time shift in optimum timing reference by the second nonminimum \nphase fade model.\n\nThe system. model and cross- eeoupled interference canceler are the \nsame as those introduced in Figs. 1 and 6 of Ref. 1, respectively. The \nchannel model allows for a single Frequenicyiae lective fade notch, both \nin the reference copolarized and the cross-coupled paths; therefore, it \nrequires the two-ray model parameters of both paths. 7\n\nThe canceler structure shown in Fig. 6 of Ref. 1 operates on the \nbaseband received digitized signals. The main-lobe estimators and \ndecision circuits together make preliminary estimates of the main lobe \nof the impulse response of the reference copolarized path. These \n_ estimators can be made of two to three tap transversal equalizers, and \ndecisions made by detection circuits are provided to the cross-polari- \nzation canceler transversal filter to form and then eliminate the \ninterfering main lobe from the opposite polarization received signal \nthat has been properly delayed. Then, the cross-coupled interference \ncanceled signals are equalized for mitigating intersymbol interference \nand cross-rail interference. The tap coefficients of the main-lobe \nestimators can either be derived from the preliminary decision circuit \nerror signals or, to get a better performance, they can be set by using \nthe error signals of the final decision circuits, as shown in Fig. 6 of \nRef. 1. The slow channel time variations allow the use of the final\n\nIll. NUMERICAL CALCULATIONS OF CANCELER PERFORMANCE IN \nNONMINIMUM PHASE FADES |\n\nIn this section we assume two 16-QAM signals are dual-polarized \nand transmitted over the channel model shown in Fig. 1 of Ref. 1 \nassuming nonminimum phase fading of the main path. The cross- \ncoupled path is assumed to be flat or minimum phase faded. For \nnonminimum phase fades we use the third model described in Section \nII. The transmit/receive filters are the Nyquist type of 45-percent roll- \noff. The symbol rate is chosen to be 22.5 Mbaud and a signal-to-noise \nratio of 60 dB is assumed. Performance signature (M-curve) for a 107\u00b0 \nsymbol error rate is used in the numerical evaluation, and the Gauss \nquadrature method\u2019 is used to derive the probability of error, accu- \u2014 \nrately.\n\nconsider the following two examples. In the first example the main \npath is nonminimum phase faded with a fade notch depth of 7.5 dB \nand a notch offset of 11.2 MHz from the band center. The cross- \ncoupled path in this case is assumed to be flat faded by 20 dB. Clearly, \nthe cross-coupled path signal has a perfect Nyquist shape and is \ncentered at the time origin. The main-path signal timing reference is \nto the left or the right of the origin depending on whether the main- \npath fade is minimum or nonminimum phase. However, since the \namount by which the timing reference is translated has the same \nabsolute value, as seen in Figs. 6 and 7, once the timing reference is \nfound, the main and the cross-coupled signals are sampled using the \nsame reference; hence, the same set of interferer samples results, \nregardless of the type of the fade. Therefore, the resulting end per- \nformance should not change. However, this is no longer true when the \ncross-coupled path is faded dispersively because the interfering im- \npulse response is no longer symmetrical about the vertical axis going \nthrough the origin. This is shown in Figs. 8 and 9 where, in addition \nto a 7.5-dB depth and an 11.2-MHz offset fading of the main path, \nthe cross-coupled path is faded by a 5-dB midband fade. Hence,. \ndepending on whether the main-path fading is a minimum or nonmin- \nimum phase, different sets of samples of the interfering impulse \nresponse are involved and the end performance may not necessarily \nremain the same.\n\nThe system performance signatures for two different fading eonde \ntions of the cross-coupled path with and without the canceler of Ref. \n1 are shown in Figs. 10 and 11. In Fig. 10 it is assumed that the cross-\n\n_ Fig. 6\u2014In-phase received impulse responses of dual-polarized system for an offset \nminimum phase fade on the main path and a flat fade on the cross-coupled path.\n\nFig. 7\u2014In-phase impulse responses of dual-polarized system for an offset nonmini- \nmum phase fade on the main path and a flat fade on the cross-coupled path.\n\nFig. 8\u2014In-phase impulse responses of dual-polarized system for an offset minimum \nphase fade on the main path and a midband minimum phase fade on the cross-coupled \npath.\n\ncoupled path is only flat faded; hence, the nonminimum phase per- \n\u2018formance signatures with and without the canceler are identical to the \nminimum phase faded signatures of Fig. 7 in Ref. 1. As seen in Fig. \n10, the single-tap canceler\u2019 improves the dual-pol system performance \nto nearly that of a single-pol system shown by a dotted M-curve. For \nmidband fades, as observed in Fig. 10, the timing reference of the \nmain-path impulse response is offset from the peak of the main lobe\n\nFig. 9\u2014In-phase impulse responses of dual-polarized system for an offset nonmini- \nmum phase fade on the main path and a midband minimum phase fade on the cross- \ncoupled path.\n\nFig. 10\u2014Canceler performance in dual-polarized 16-QAM radio for a flat fade on the \ncross-coupled path.\n\nFig. 11\u2014Canceler eetionnanens in dual- polarized 16- QAM radio for a midband dis- \npersive fade on the cross-coupled path.\n\nof the interfering impulse response shape; hence, the canceler does \nnot perform as well for midband fades as it would for offset fades of \nthe main path because as the main-path fade notch moves toward the \nband edge, the timing reference of the overall impulse response moves \ntoward the origin; hence, the two peaks tend to align. Of course, the \nremedy for this situation is to increase the number of canceler trans- \nversal filter taps. In Fig. 11, we consider the case where the cross- \ncoupled path is dispersively faded by a minimum phase midband fade \nof 5-dB notch depth. In this case, since by asymmetry two different \nsets of samples of the interfering impulse responses are involved, the \nresulting signatures under minimum or nonminimum phase fades are \nnot the same, and the nonminimum phase fade increases the outage ~ \nslightly for the same cross-coupled path fading conditions. As the fade \nnotch moves away from the band center, and since the reference \ntiming of the main-path impulse response moves toward the time \norigin for both minimum and nonminimum phase fades, the two sets \nof samples taken of the interfering impulse responses become similar; \nhence, the minimum and nonminimum phase fade signatures approach \none another and become approximately the same around the band \n~ edge..\n\nA comparison of the signatures for minimum and nonminimum \nphase fadings of the main path indicates that the baseband canceler\n\nperformance is relatively insensitive to the type of the fade as antici- \npated in Ref. 1. |\n\nNotice that the difference in the signatures for minimum and non- \nminimum phase fading of the copolarized channel is because we have \nassumed a minimum phase fade for the cross-coupled channel in both \ncases. Given that the copolarized and the cross-coupled channel fading \n_ have both minimum or both nonminimum phase, the pplenanuaee are \nindeed identical in both cases.\n\nIn this paper performance of a previously proposed canceler\u2019 is \nevaluated under nonminimum phase fades, and it is shown that the \ncanceler performance is practically transparent to the type of the fade. \nAn investigation of different static models for nonminimum phase \nfades is also performed, and it is shown that some of the known fading \nmodels for single signal transmission may not be appropriate for \nmulticarrier systems such as dual-polarized channels.\n\nThe writer is deeply indebted to W. C. Jakes, Jr. for discussions \nduring this work.\n\n1. M. Kavehrad, \u201cBaseband Cross-Polarization Interference Cancellation for M-Quad- \nrature Amplitude-Modulated Signals Over Multipath Fading Radio Channels,\u201d \nAT&T Tech. J., 64, No. 8 (November 1985), pp. 1913-26.\n\n2. L. Leclert and P. Vandamme, \u201cNon-Minimum Phase Fadings Effects on Equaliza- \nears Techniques in Digital Radio Systems,\u201d GLOBECOM Conf., November 1983,\n\n3. H. Sari, \u201cBaseband Equalizer Performance in the Presence of Selective Fading,\u201d \nGLOBECOM Conf., November 1983, San Diego.\n\n4, G. Niezgoda et al., \u201cEffects of Non-Minimum Phase Fades on the Performance of \na 49 QPRS 90 \u2018Mb/s Digital Radio Using Decision Feedback Equalizations,\u201d \nGLOBECOM Conf., November 1983, San Diego.\n\n5. M. Kavehrad and C. A. Siller, Jr., \u201cPerformance Signatures for Dual-Polarized \nTransmission of M-QAM Signals Over Fading Multipath Channels,\u201d AT&T \nTech. J., this issue.\n\n6. W.D. Rummler, \u2018 \u2018A New Selective Fading Model: Application to Propagation Data,\u201d \nB.S.T.J., 58, No. 5 (May-June 1979), pp. 1037-71.\n\nAnalysis/Simulation Study of Cross-Polarization \nCancellation in Dual-Polarization Digital Radio\n\nThis paper analyzes cross-polarization cancellation in dual-polarization \ndigital radio links transmitting M-ary quadrature amplitude-modulation \n(M-QAM) signals. We consider the use of three options in the receiver: (1) no \ncancellation ; (2) ideal (i.e., total) cancellation; and (3) optimal nondispersive \ncancellation. For every option, we assume the canceler to be followed, in each \npolarization branch, by an ideal minimum mean-square error equalizer. By \npostulating a statistical model for the co-polarization and cross-polarization \nresponses of a digital radio channel, and then simulating thousands of sets of \nthese responses, we obtain curves that relate outage probability to the number \nof modulation levels. We show graphically that the no-canceler case is unthink- \nable; that total cancellation permits results close to those for single-polariza- \ntion transmission; and that optimal nondispersive cancellation can have a \nlimited range of application. We also examine the effects of key system \nparameters and the various modeling assumptions.\n\nThis paper reports on a theoretical study of cross-polarization \ncancellation in dual-polarization microwave digital radio receivers. In \nanother recent theoretical study, Amitay and Salz derived and evalu- \nated the optimal linear receiver response to cross-pol coupling and \nmultipath fading in combination.\u2019 By contrast, this paper analyzes \nsuboptimal structures wherein the cross-pol canceler and multipath\n\nequalizer stages are in cascade. We do this because, in the design of \npractical adaptive receivers, it may be desirable to deal with cross-pol \ninterference and co-pol dispersion separately. For example, cross-pol \ncancellation may be simpler to perform at IF, while multipath equali- \nzation is best done at baseband. More important, modularity in pro- __ \nviding the cross-pol cancellation and multipath equalization functions . \n- may permit more economy and flexibility in designing and using the \nradio system.\n\nOur primary aim is to show that, in using cross-pol cancellation \nfollowed by multipath equalization, overall outage performance can be \nmade nearly as good as that for an equalized single-pol radio link. Our \nsecondary aim is to quantify the differences in attainable outage \nperformance for three distinct approaches to cancellation, namely, (1) \nno cancellation; (2) ideal (i.e., total) cancellation; and (3) optimal \nnondispersive cancellation. Finally, we aim to show how these results \nare influenced by key parameters of the system and various assump- \ntions regarding the radio propagation.\n\nTo obtain these quantitative results, we use a combination of re- \nceiver analysis and Monte Carlo simulation of the co-pol and cross- \npol responses of the propagation channel. The simulations require \nspecifying a model for the channel responses. We have done so, despite \nthe sparsity of published work in this area, by using data, tentative \ntheories and speculations by the author and others. Model uncertain- \nties are dealt with, in part, by means of sensitivity studies.\n\nWe consider digital radio links transmitting independent random \ndata streams on two nominally identical cofrequency channels using \nnominally orthogonal polarizations. For reasons of both practicality \nand convenience, we assume the polarizations to be vertical (V-pol) \nand horizontal (H-pol). \u00a9 |\n\nThe cofrequency channels are in the microwave common carrier \nbands at 4, 6, and 11 GHz, corresponding to channel bandwidths, \nrespectively, of 20, 30, and 40 MHz. The modulation is M-ary Quad- \nrature Amplitude Modulation (M-QAM), with M = 4, 16, 64, 256, etc. \nThe end-to-end spectral shaping (excluding channel dispersion and\n\nTRANSMITTER \nFig. 1\u2014Schematic representation of a dual-pol M-QAM transmitter with root-cosine-\n\nroll-off spectral shaping and a microwave channel with co-pol dispersion and cross-pol \ncoupling. The H-functions are slowly time varying during periods of multipath activity.\n\nadaptive filtering) is cosine roll-off, with roll-off factor a. This shaping \nis divided evenly between transmitter and receiver. In the receiver, \nthe H-pol and V-pol signals are applied to a cross-pol canceler that \nlinearly combines the two input branches, followed by fixed filtering \nand adaptive equalization in each output branch.\n\nThe above description is reflected in Figs. 1 and 2.* Figure 1 depicts \nthe transmitter and channel, the latter being characterizable by a pair \nof co-pol responses [H,,(f) for the V-pol signal and H>.(f) for the H- \npol signal] and a pair of cross-coupling responses [Hj.(f) and Ho,(f)]. \nIdeally (and in fact under normal conditions), H,,(f) and Ho2(f) are \nidentical nondispersive responses. During multipath fading, however, \nthey become dispersive and possibly smaller as well, as recounted in \nnumerous papers on statistical models.?\u00b0 \n_ The cross-pol. responses, on the other hand, are ideally zero. In fact, \nhowever, they are generally nonzero, even under normal conditions. \nDuring such times, they are typically nondispersive and small com- \npared to the co-pol responses; but, during multipath fading, they tend \nto be dispersive and can become relatively large as well,\u00ae giving rise to \n_ potentially serious impairments in detection.\n\nFigure 2 shows the structure of the receiver to be analyzed. All the \nG-functions are adaptive, and we assume they can be controlled in \npractice to satisfy the various criteria specified below. For convenience\n\n* Note that f is referred to the center frequency of the radio channel, and that C(f) \nrepresents the raised cosine function (dimensionless).\n\nFig. 2\u2014Schematic representation of dual-pol receiver processing consisting of cross- \npol cancellation, root-cosine-roll-off filtering, and adaptive equalization. The G-func- \ntions are assumed to be adapted to the prevailing channel response functions in Fig. 1.\n\nIn principle, the location of the fixed filtering is immaterial, and each of the various \nlinear stages can be at RF,IF or baseband, as appropriate.\n\nonly, all the processing stages are shown as passband (rather than \nbaseband) circuits. |\n\nFinally, we assume in this study that the RF carriers and the symbol \ntimings are nominally identical but asynchronous. This assumption \nenables the cross-pol interference to be treated as a noise-like process, \nwhich simplifies the analysis in some respects. The results should \ndiffer little from those for synchronized dual-pol transmissions.\n\nTo evaluate the canceler/equalizer structure, we need some analyt- \nical relationships. Let Sy(f) be the Fourier transform of a long random \nsequence of (complex) data into the V-pol transmit filtering (Fig. 1), \nand let Sx(f) be the same for the H-pol branch. Because of cross- \ncoupling in both the propagation medium and the receiver, each branch \nwill show a mixture of V-pol and H-pol data streams at its equalizer \noutput (Fig. 2). Because of symmetry, however, we need only consider \nthe output of the V-pol branch, and will follow that custom throughout.\n\nThe Fourier transform of the equivalent baseband signal at the V- \npol output is :\n\n_'The second term clearly represents the cross-pol interference (hereafter\n\nabbreviated CPI); the bracketed quantity of the first term represents \nthe changes in the desired signal due to the propagation medium, \ncanceler, and equalizer. If H,,(f) = Hoo(f) = 1 and Hi2(f) = Half) = \n0, Sv(f) reduces to the case of an undistorted single-pol transmission. \nWe now state the three conditions on G,;(f) and Gj2(f) that will be \nanalyzed here: \n1. No cancellation, 1.e.,\n\nwhere gop is that complex value for which the mean square CPI at the \ncanceler output is minimized. There are other possible criteria for \ndefining opt (e.g., that value that minimizes the mean-square CPI at \nthe equalizer output), but the simple criterion used here is good enough \nfor our purposes.\n\nFinally, we assume that the equalizer associated with each of the \nabove three conditions is an ideal Minimum Mean-Square Error \n(MMSE) equalizer. That is, it maximizes the ratio of the signal half- \ndistance to the total root-mean-square (rms) distortion (thermal noise, \nCPI, and intersymbol interference) as sampled at the baseband detec- \ntor. Previous studies for single-pol channels have shown that the \nperformance of an ideal MMSE equalizer can be closely approached \nusing fractionally spaced tapped-delay-line filters with just a few taps.\u2019\n\nObtaining an empirical statistical model for the four channel re- \nsponse functions\u2014Aj1(f), Hie(f), Hoi(f) and Hoo(f)\u2014is a very difficult \nbusiness. The model to be used here is based only partly on measured \ndata, the other parts being theories on the underlying physics of the \npropagation medium, and speculations on what mathematical artifacts \nto include so that important issues are not overlooked. The model \ndraws on the published data, models or ideas of N. Amitay,\u2019 W. D. \nRummler,\u201d\u00ae K. T. Wu,\u00b0 S. Lin,\u00ae? M. L. Steinberger? and M. Liniger,?\u00b0 \nas well as on private discussions with these investigators and others. \nThe author\u2019s own ideas are included in the mixture, and he accepts \nsole responsibility for flaws in the speculative model presented next.\n\nThe co-pol responses can be represented using the three-path func- \ntion introduced by Rummler.\u2019 For most of the year, H1,(f) = Hoo(f) = \n_ 1; but, during 7 seconds per hop per year, the two functions become \nHilf) = ai{1 \u2014 bye\u2019*'e~*\"]; minimum phase response 5)\n\nwhere 7\u00bb is a function of the radio path and band; 7 is a fixed parameter \nof 6.3 ns; (ay), G22, b11, b22611, 622) are slowly varying \u201cfitting\u201d parameters \nthat collectively provide accurate approximations to the true re- \nsponses; 0;; < 1 and bg < 1 at all times; and these representations \napply over bandwidths of 40 MHz or less.\n\nRummler\u00ae has published a widely used joint probability density \nfunction (pdf) for the a-, b-, and \u00a2-parameters of a co-pol function \nsuch as (5). We will use that pdf to characterize the statistics of both \nHy, (f) and H2o(f). Moreover, we assume each function to be minimum \nphase over a fraction p of all fade events, with p a parameter lying \nbetween 0 and 1. |\n\nIn evaluating the dual-pol receiver, it will be necessary to assume \nsomething about the similarity in time between H,,(f) and Ho2(f). \nAccording to a limited body of data, these functions tend to track \nquite closely, i.e., Hio(f) ~ Hi(f) during most fade events. On the \nother hand, significant impairments could accrue during those events \nwhere they are dissimilar. To bracket the range of possibilities, we \nhave obtained simulation results under two distinct assumptions: \nAoo(f) = Hii(f) in every fade event; and Hi,(f) and Hoo(f) are statis- \ntically identical but mutually independent over the ensemble of fade \nevents.\n\nHi; (f) = {0.5 ky[Hi(f) + Ao2(f)] + ejeVi} eet (6) \nLx] \nwhere k;; is a fixed complex constant related to the residual cross-pol \ncoupling in the radio antennas and waveguide runs (typically, 20 \nlog| ky | < \u201425 dB); Yj is a uniformly random phase over the ensemble \nof fade events; \u00ab; is a Rayleigh variate over that ensemble (typically, \n10 log| \u00ab; |? < \u201435 dB); and 6T is a fixed time \u201cmisalignment\u201d between \nthe co-pol and cross-pol paths to the receiver. We further assume that\n\nThis model for the cross-pol responses is more or less consistent | \nwith data and theories reported previously.\u00ae* \u00a9 The delay parameter \ndT is new and is included to add richness to the model; we will see if \n6T has any important effects on performance as it ranges from 0 to 4 \nns. This way of including delay effects may, in fact, be too mild. \nAmitay\u2019 has suggested that 67' might be larger than 4 ns, and that the \nfactor e\u2019*? might more properly be attached to just \u00a2,e\u2019*%. Such an > \napproach would make the cross-pol responses more dispersive, con- \nsistent with some reported measurements. For now, however, we will \nuse the model for cross-pol responses described above.\n\nTo begin, we note that an M- L-QAM signal contains two /M-level | \nAM signals in phase quadrature, and that the possible data values in \neach quadrature rail are +1, +3, --- + (VM \u2014 1). We also note that, \nafter cross-pol cancellation, fixed filtering, adaptive equalization, and \ncoherent demodulation (in whatever order), two baseband pulse \nstreams are sampled every T seconds at each of two data detectors in \nthe V-pol receiver branch, and similarly for the H-pol receiver branch. \nOur analytical goal is to derive a signal-to-distortion ratio that char- \nacterizes performance at either baseband detector of either receiver \nbranch. To the maximum extent possible, we invoke known relation- \nships and any simplifying assumptions that do not compromise the \ngenerality of the results.\n\nOne simplifying assumption, noted and explained in Section 2.1, is \nthat the cross-pol interference can be treated as a noise-like process. \nAnother simplification accrues by assuming a zero-percent roll-off \nfactor (a = 0) for the end-to-end spectral shaping, in which case C(f) \nis a unit rectangle on the f-interval [\u20141/2T, 1/27]. This simplifying \npee an is justified by earlier findings that, over the practical range\n\nFinally, for convenience we ere a carrier-to-noise ratio as follows: \nlet Po be the average received power in an M-QAM signal in the \nabsence of fading; and let No be the power spectrum density of the\n\n* The effects of the phases of ki2 and ky, and of any dissimilarities between these \ntwo constants, should be small. We thus remove them from the study to simplify \nmatters.\n\nnoise input to the V-pol (or H-pol) receiver, including the contribution \nof the receiver noise figure. Then\n\nCNR 4 PoT/No. (7) \nThis is the unfaded Carrier-to-Noise Ratio (CNR) in the Nyquist \nbandwidth, and is typically on the order of 10\u00b0 (60 dB). \n3.2 Signal and distortions at the canceler output \u2014\n\nFrom previous discussions and Figs. 1 and 2, we can show that the \ndata pulse at the canceler output for a data value of unity would have \na Fourier transform\n\nSimilarly, the thermal noise at that output has a power spectrum \ndensity\n\nwhere (see Fig. 2) | \nB,(f) = [| Gulf) |? + | Gie(f) | 71. (11) \nThe cross-pol interference has a power spectrum density \nX(f) = PoTC(f)B.Af), (12)\n\nThe composite noise power spectrum density* can then be written as \nN'(f) =N(f) + X(f) | \n= Nol B,(f) + CNR C(f)B.(f)I- (14)\n\nHenceforth, we will make use of our assumption that a = 0, ie., that \nC(f) is a unit rectangle on [\u20141/2T, 1/2T)].\n\n* This spectrum density is for the sum of CPI and thermal noise, which can be treated \nas noise-like but is clearly not Gaussian except when the CPI vanishes.\n\nbaseband detector, the ratio of sampled-signal half-distance to rms \ndistortion, the latter consisting of thermal noise, CPI, and Intersymbol \nInterference (ISI). We need not concern ourselves with how such an \nequalizer is realized, as this is a well-developed art; we merely need to \nderive and analyze its frequency response.\n\nThe derivation is accomplished in two steps. First, we assume a \nnoise-whitening input filter, 1.e., one having a real transfer function \nproportional to 1/VN\u2019(f), and then we invoke the result in Section \n5.1 of Ref. 12 for MMSE equalizers with white input noise. The \nequalizer response for a = 0 turns out to be\n\nThe data pulse following an seus lick with this response has a real, \nnonnegative Fourier transform. Assuming optimal carrier and timing \nrecovery, we can show that the squared half distance between adjacent \nsignal samples (constellation points) at the detector is therefore\n\nwhere the integration limits, both here and in subsequent expressions, \nare +1/2T. |\n\nThe mean-square intersymbol interference at the sample times can \nlikewise be shown to be\n\nThe signal-to-distortion ratio (p) at either baseband detector is now \ndefined to be\n\nThe significance of P is a the Bit Error Rate (BER) can be tightly \nupperbounded by!\u00ae\n\nBy using the various relationships in Sections 3.1 through 3.3, we can \nexpress p in the form\n\nThe focus of our investigation is the signal-to-distortion parameter \u2014 \nT introduced by (21). To \u201ccalibrate\u201d this quantity, let Io denote the \nvalue of I\u2019 that must be exceeded if BER is to lie below some threshold \nvalue BERo. We can find a tight upperbound for Ip given BER\u00bb and \nM by invoking the equality in (20) and combining it with (21). The \nresult is :\n\nSome values of Io, expressed in decibles, are given in Table I for. \nvarious BER\u00bb and M of interest.\n\nAlthough the above results are applicable to any assumptions we \nmake about the canceler, two special cases are worthy of note. The\n\n_ first is the case of total cancellation [see (3)]. Under this condition, \nX(f) in (12) vanishes entirely and so N\u2019(f) reduces to N(f). Also, as \nthe mathematics of this situation makes clear, the choices of Gi;(f) \nand G,.(f) individually are immaterial, just so their ratio satisfies (3). \n\u2018This is a consequence of the (presumed) fact that the dominant \nthermal noise is introduced before the canceler. |\n\nThe second special case is that of nondispersive cancellation [see \n(4)]. We assume that g.,: is chosen so as to minimize X(f) [see (12) \nand (13)]. For a zero-percent roll-off factor, it is easy to show that\n\n\u2018For convenience in what follows, we now redefine two ypereinelers \nof the channel model, namely, |\n\nWe can say that the channel is statistically specified once we (1) \nassign numerical values to p, 6T, K and E; and (2) declare H,,(f) and \nH2.(f) to be either identical or statistically independent. Similarly, we \ncan say that the radio system is design-specified once we (1) assign \nnumerical values to a, CNR, and 1/T; and (2) declare the type of \ncross-pol cancellation to be used.\n\nA computer program has been written that apeiae for any joint \nspecification of the channel and system, the yearly probability distri- \nbution of I\u2019. To accomplish this, the program combines Monte Carlo \nsimulation methods with the channel model of Section 2.3 to generate \na large statistical \u201censemble\u201d of channel response functions. That is, \neach member of the ensemble is a set of functions, {Hii(f), Hio(f), \nHalf), Hoo(f)}, generated by deriving the various parameter values \n(Qy1, 011, Ge2, bee, \u20ac12, \u20ac21, etc.) in accordance with the statistics of the \nmodel. For each set of H-functions thus generated, the program \ncomputes I\u2019 using the formulas of Section 3.4. After generating and \nevaluating the prescribed number of sets (20,000 in our study), the \nprogram computes a cumulative probability function, P(T), for the \npopulation of I'-values thus obtained.\n\nWe present results for the following channel/system parameter \nvalues or conditions, where those in boldface are the ones used the \nmost extensively:\n\nStatistical dependence between co-pol functions: Totally depend- \nent (Hi:(f) = Hee(f)) and totally independent. |\n\nType of cancellation: No cancellation, total cancellation, and \nnondispersive cancellation. |\n\nSymbol rate, 1/T: 15 Mbaud, 22.5 Mbaud, and 30 Mbaud (typical \nvalues for digital radio systems in the 4-, 6-, and 11-GHz radio \nbands, respectively).\n\nThroughout our simulations and computations, we have assumed a \n0-percent roll-off factor (a = 0) and a 63-dB unfaded carrier-to-noise \nratio (CNR = 2 X 10\u00b0). However, we will also discuss how the computed \nresults would vary with these parameters.\n\nThe results of this study are given by Figs. 3 through 6. Each figure \nshows curves of P(I) versus I for a number of different channel/ \nsystem specifications. These curves can be interpreted as conditional \noutage probabilities and can be used to estimate yearly outage seconds \non a radio hop, as follows: Given a threshold bit error rate (BERo) \nand the number of modulation levels (M), the minimum acceptable \nvalue of I can be obtained using (27) or Table I. (Assuming BERo = \n107*, To is roughly 20 dB for M = 16 and 26 dB for M = 64.) The \nresulting P(IT) is the probability of outage on a given hop conditioned \non the occurrence of fading. To estimate yearly outage seconds on the \nhop, one would need to estimate, measure, or obtain from available \nmodels the expected number of yearly fading seconds, 7\u00bb. (A repre- \nsentative value is 16,000 seconds.) Multiplying Ty) by P(I'o) would \nyield the expected number of outage seconds per hop per year. This \nutilitarian aspect of the curves in Figs. 3 through 6 should be kept in \nmind as we make some relative comparisons.\n\nFigure 3 shows P(T) for each of three symbol rates and each of \nthree cancellation options. The \u201ccommon conditions\u201d listed on the \nfigure are assumed to be the most representative for an actual dual- \npol channel.\n\nThe top curve shows what can be expected if no cancellation \nwhatsoever is employed. The heavy line used here contains the results \nfor all symbol rates from 15 Mbaud to 30 Mbaud. It is clear that the \nabsence of some sort of cancellation is unthinkable for the dual-pol\n\nFig. 3\u2014Cumulative probability functions for I. Results are shown for each of the \nthree canceller options considered and for symbol rates of 15, 22.5, and 30 Mbaud. In \naddition to the \u201ccommon conditions\u201d shown, a = 0 and CNR = 2 X 10\u00b0 (63 dB).\n\nsystems of interest, and we dismiss this option from further consid- \neration. |\n\nThe results for nondispersive cancellation reveal an order-of-mag- \nnitude improvement (for the lower values of I\u2019) and a fairly strong\n\ndependence on symbol rate. The latter is not surprising since larger \nsymbol rates involve larger bandwidths and, hence, greater dispersion \nin the channel H-functions. In microwave common carrier channels \nwith narrow bandwidths (e.g., 3.6 MHz in the 2-GHz band), nondis- \npersive cancellation might thus be quite adequate. This would depend, \nof course, on the number of modulation levels and the MOutage proba- \nbility requirements.\n\nThe narrow band at the bottom, for the case of total cancellation, \ncontains results for all symbol rates between 15 Mbaud and 30 Mbaud. \nThis approach provides, at the lower values of I, another order-of- \nmagnitude improvement beyond nondispersive cancellation. More- \nover, the dependence on symbol rate (or bandwidth) is seen to be \nsmall. Later, we will compare these results with those for single-pol \ntransmission using ideal MMSE equalization.\n\nFigure 4 shows results for both nondispersive and total cancellation \nas p and 67\u2019 range over the sets of values we have specified for them. \nThe \u201ccommon conditions\u201d assumed here are typical ones; alternative \nrealistic assumptions would not alter the trends revealed by these \ncurves. |\n\nThe main conclusion we can draw here is that outage performance \nfor nondispersive cancellation is sensitive to the details of the channel \nmodel, while that for total cancellation is not. This is reflected in the \nwideness and narrowness, respectively, of the bands for these two \ncases. Since total cancellation is the obvious design choice for a high- \nquality system, this is good news. It implies that certain hard-to- \ndetermine fine details of the channel model need not be accurately \nspecified to obtain reliable performance estimates.\n\nFigure 5 illustrates, for both nondispersive and total cancellation, | \nhow performance is affected by the statistical dependence between \nAy, (f) and Hoo(f). The performance variation over the range between \ntotal dependence [Hi;(f) = Hoo(f)] and total independence is seen to \nbe fairly small. Nevertheless, it would be clearly beneficial if the co- \npol responses were more independent than they apparently are. |\n\nA simple explanation can be given for the improvement shown when \nHy,(f) and H2(f) are independent. Let us consider the case of total \ncancellation and assume, for simplicity, that all H-functions are essen- \ntially flat with frequency. Combining (10) and (11) with (3) and (1) \nunder these circumstances, we can show that the receiver output\n\nFig. 4\u2014Cumulative probability functions for I. Results are shown for nondispersive \nand total cancellation, with the symbol rate fixed at 22.6 Mbaud. The parameters here \nare the minimum phase probability (p) and the delay parameter (67). All else is the \nsame as in Fig. 3.\n\nsignal-to-noise ratio would be proportional to | Hy,;He2 \u2014 Hy.H2, |7/ \n[| Hoo|\u00b0 + | Hie|?]. The potential benefit of statistical independence \narises when | H;,| is weak; at such times, the numerator can be quite \nsmall if | H22| is similarly weak, via near cancellation of Hy,H2. by\n\n20 25 30 35 40 \nTIN DECIBELS \nFig. 5\u2014Cumulative probability functions for [. Same conditions as in Fig. 4, except\n\nthat p and 6T are fixed at 0.5 and 2 ns, and results are shown for both totally dependent \nand totally independent co-pol response functions.\n\nHy2H>,. If Hi, and He are independent, however, there is a chance \nthat Hoo will be strong enough at such times to avoid this near- \ncancellation. The statistical effect of all this is reflected in the com- \nparative results of Fig. 5. |\n\nFigure 6 shows results for both nondispersive and total cancellation \nas K and E range over the sets of values we have specified for them. \nNote that these particular results are for 1/7 = 22.5 Mbaud and H,(f) \n\u2014 Hof ).\n\nFor nondispersive cancellation, each heavy line contains results for \na particular E and all K below \u201425 dB. The negligible influence of K \nover this range is apparent. However, FE is another matter. As this \nquantity decreases in 5-dB steps from \u201435 dB, the probability curves \nshift to the right by nearly 5 dB. These features reflect the fact that \nnondispersive cancellation virtually eliminates the effect of the cross- \ncoupling gain component kHi;(f) in (6) [remember that H.(f) = \nH,,(f) in these calculations] but is less effective against the second \ncomponent, whose mean-square value in decibels is E. This reveals \nyet another way in which precision in the channel model is needed to \nestimate nondispersive canceller performance.\n\nFor the case of total cancellation, on the other hand, the lower band \nin Fig. 6 shows how much that need is attenuated. Even so, this band \nwidens measurably as F decreases below \u201445 dB. The point is made \nclear by the dotted curve, for K = E = \u2014o, which is equivalent to the \ncase of single-pol transmission (i.e., no input CPI) and ideal MMSE \nequalization.\n\nWe now see how closely dual-pol outage performance comes to that \nfor single-pol operation when total cross-pol cancellation is used. We \ncan also make a limited comparison between the cascaded approach \n(i.e., canceller and equalizer in tandem) and the optimum linear \nreceiver (i.e., jointly optimizing {Gi,(f), Gie(f), Golf), Goo(f)} against \nboth CPI and multipath dispersion, thereby eliminating the separate \nequalizer). The latter case is treated comprehensively by Amitay and \nSalz in Ref. 1. In that paper, the probability functions are plotted \nagainst spectral efficiency, in b/s/Hz, and so the following correspond- \nences apply: the abscissa values of 4 and 6 b/s/Hz in Ref. 1 correspond \nclosely to I = 20 and 26 dB, respectively, in the present paper. Now \nassuming that E = \u201435 dB and 1/T = 30 Mbaud, Fig. 3 of Ref. 1 \nshows the outage probability for the optimal linear receiver to be \nroughly four times higher than for single-pol transmission when the \nspectral efficiency is 4 b/s/Hz, and roughly two times higher when the \nspectral efficiency is 6 b/s/Hz. The corresponding results in the \npresent study for ! = 20 and 26 dB are quite similar and, for higher \nabscissa values, the similarities are even greater. While recognizing \nthat the two studies used somewhat different models and methods of \nanalysis, and entirely different random numbers in their Monte Carlo\n\n20 25 30 , \u201c80:34 40 \nTIN DECIBELS \nFig. 6\u2014Cumulative probability functions for T. Same conditions as in Fig. 4, except\n\nthat p and 6T are fixed at 0.5 and 2:ns, and K and E are parameters. Note the dotted - \ncurve \u201cNo CPI,\u201d which corresponds to an ideally equalized single-pol channel.\n\nsimulations, we perceive a general truth in this comparison: specifi- \ncally, the cascaded approach, in which cancellation and equalization \nare functionally separate, leads to outage statistics very close to those \nfor optimal linear reception. | |\n\nAll of the results in Figs. 3 through 6 are for a = 0 and CNR = 2 X \n10\u00b0 (63 dB). Nevertheless, we can say something about the influences \nof these parameters. From previous studies,!! for example, we know \nthat a has a very small effect on outage probability over the practical \nrange 0 < a < 0.5. Also, the curves for total cancellation would \nessentially shift X dB to the right (or left) for every X-dB increase (or \ndecrease) in CNR. This is because the residual distortion in the case \nof total cancellation followed by MMSK equalization is almost entirely \nthermal noise. In the case of nondispersive cancellation, wherein the \ndominant residual distortion is uncancelled CPI, this sensitivity of the \nresults to CNR would be sharply reduced.\n\nWe have used analysis and Monte Carlo simulation to estimate \nconditional outage probabilities in dual-pol digital radio as functions \nof a detection measure (I) that can be related to the number of \nmodulation levels and the bit error rate. In performing the simulations, \nwe have resorted to a statistical model for the dual-pol channel that \nlacks a firm empirical basis. This results unavoidably from the current \nincomplete status of dual-pol channel measurement and modeling. We \nhave dealt with this limitation, in part, by treating various uncertain \naspects of the model parametrically.\n\ne Obtaining reliable estimates of outage probability for the case of \nnondispersive cancellation requires accurate, detailed descriptions \nof the underlying channel model. The outage performance of this \ncancellation approach is also quite sensitive to bandwidth (or \nsymbol rate).\n\ne In the case of total cancellation, by contrast, outage performance \nis insensitive to many details of the channel model as well as to \nsymbol rate. |\n\ne While far superior to no cancellation, nondispersive cancellation \nleads to outage probabilities an order-of-magnitude greater than \ndoes total cancellation. Nonetheless, it may find applications \nwhere symbol rates are low (less than 5 Mbaud) and the outage \nrequirements are liberal.\n\ne The outage statistics for total cancellation followed by ideal equal- \nization are fairly close to those for single-pol transmission, \nwherein there is no cross-pol interference to cancel.\n\ne More significantly, total cancellation in cascade with ideal equal- \nization appears to produce outage statistics very close to those for \noptimal linear reception, wherein the effects of cross-pol interfer-\n\nence and multipath dispersion are jointly minimized in the same \nreceiver stage. Use of the cascade approach, therefore, may permit \nsuch benefits as design simplicity, manufacturing economy, and \noperational flexibility with no serious loss in performance.\n\nI am grateful to Lisa J. (Domenico) Case for her help i in executing \nthe computer simulation/analysis programs.\n\n1. N. Amitay and J. Salz, \u201cLinear Equalization Theory in Digital Data Transmission \nOver Dually Polarized Fading Radio Channels,\u201d AT&T Bell Lab. Tech. J., 63, \nNo. 10, Part 1 (December 1984), pp. 2215-59.\n\nW. D. Rummler, \u201cA New Selective Fading Model: Application to Propagation Data,\u201d \nB.S.T.J., 58, No. 7 (May-June 1979), pp. 1037-71.\n\nW. D. Rummler, \u201cMore on the Multipath Fading Channel Model,\u201d IEEE Trans. \nCommun., COM-29, No. 3 (March 1981), pp. 346-52.\n\nL. J. Greenstein and B. A. Czekaj, \u201cA Polynomial Model for Multipath Fading \nChannel Responses,\u201d B.S.TW., 59, No. 7 (September 1980), pp. 1197-225.\n\nJ. C. Campbell and R. P. Coutts, \u2018 \u2018Outage Prediction of Digital Radio Systems,\u201d \nElectron. Lett., 18, No. 25/26 (December 1982), pp. 1071-2.\n\nK. T. Wu, \u201cMeasured Statistics of Multipath Dispersion of Cross Polarization \nInterference,\u201d Paper 46.3, Int. Conf. Commun., May 14-17, 1984, Amsterdam. \nN. Amitay and L. J. Greenstein, \u201cMultipath Outage Performance of Digital Radio\n\nReceivers Using Finite-Tap Adaptive Equalizers,\u201d IEEE Trans. Commun., COM- \n32, No. 5 (May 1984), pp. 597-608. \n8. S. H. Lin, \u201cImpact of Microwave Depolarization During Multipath Fading on Digital \nRadio Performance, \u201d B.S.T.J., 56, No. 5 (May 1977), pp. 645-74. \n9. M. L. Steinberger, \u201cDesign of a Terrestrial Cross Pol Canceller, \u201d Paper 2B.6, Int. \nConf. Commun., June 13-17, 1982, Philadelphia, Pa.\n\n10. M. Liniger, \u201cSweep Measurements of Multipath Effects on. Cross-Polarized RF- \nChannels Including Space Diversity,\u201d Paper 45.7, GLOBECOM \u201984, November \n26-29, 1984, Atlanta, Ga.\n\n11. W. C. Wong and L. J. Greenstein, \u201cMultipath Fading Models and Adoptive Equal- \nizers in Microwave Digital Radio,\u201d IKEE Trans. Commun., COM-32, No. 8 \n(August 1984), pp. 928-34.\n\n12. R. W. Lucky, J. Salz, and E. J. Weldon, Jr., Principles of Data Communication, New \nYork: McGraw- Hill, 1968.\n\n13. G. J. Foschini and J. Salz, \u201cDigital Communications Over Fading Radio Channels,\u201d \nB.S.T.J., 62, No. 2, Part 1 (February 1983), pp. 429-56.\n\nA Laboratory Simulation Facility for Multipath \nFading Microwave Radio Channels\n\nThis paper describes a laboratory facility capable of simulating time-varying \nradio multipath channel responses in real time under computer control. Four \nindependent fading channels are available that can be used for single-polari- \nzation nondiversity, combined in pairs for single-polarization dual diversity, \nor cross-coupled to simulate the two outputs of a dual-polarization nondivers- \nity channel. Each channel contains a controllable variable network capable of \nproducing a narrowband intermediate frequency response that resembles the \n\u201cthree-path\u201d function of Rummler. A wide range of models can be accommo- \ndated by altering the computer-stored sequences used to control each variable \nchannel network. The only assumption implicit in the choice of model is that \nthe channel response can be fitted to the generic function over bandwidths up \nto 40 MHz. The channel responses are controlled by either entering fixed \nparameters from a keyboard, or by reading time-varying parameters stored in \ndisk memory. This description includes the architecture, hardware design, \nsoftware implementation, and performance of the simulation facility.\n\nThe primary impediment to the operation of digital radio on micro- \nwave line-of-sight paths is multipath fading. In dual-polarization\n\n(dual-pol) systems, this problem is augmented by cross-polarization \n(cross-pol) coupling. Numerous radio measurements and data analyses \nover the past several years have served to characterize multipath and \ncross-pol responses, and to translate these characterizations into sta- \ntistical models.\u2019!\u2019 Many of these models have been eagerly engaged \nby radio systems analysts to estimate, using analysis and/or Monte \nCarlo simulation, the expected link outage times for different modu- \nlations, link parameters and receiver techniques.\u2019***? Over the same \nperiod, a number of outage measurements have been reported for \nspecific hardware designs, based on either field trials conducted over \nradio paths or laboratory \u201csignature\u201d measurements coupled with \nassumed radio channel models.??*!\n\nIf one stands back from this myriad of activities, important limita- \ntions in all of the above approaches become apparent. The analysis/ \nsimulation of statistical models applied to specified designs permits \nrapid comparisons among contending radio schemes, but relies on \nidealized models of the hardware behavior. Moreover, such study \nmethods do not readily take account of the time dynamics of the \nchannel responses. The coupling of statistical models with lab-mea- \nsured hardware \u201csignatures\u201d likewise omits channel time dynamics \nand their effects on system techniques. This deficiency is absent only \nin the case of field measurements, but these are both costly and time- \nconsuming. More important, system qualities inferred from this ap- \nproach are subject to the particular responses that nature provides \nduring the test interval. Meaningful comparisons between systems, \ntherefore, require measurements conducted in parallel under identical \nconditions of path and time.\n\nThe Channel Simulation Facility (CSF) reported here is intended \nto fill the gaps between these various study approaches. It simulates \ntime-varying radio channel responses in the laboratory in real time, \nand it plays the channel responses into actual hardware realizations \nrather than idealized system models. The channel responses are dic- \ntated by computer-generated control signals, and so the twin benefits \nof model selectability (software-controllable changes in the history of \nthe channel response functions) and repeatability (ability to replicate \nthe channel response history for different systems at different times) \nare realized. And, finally, the ability to simulate channels in a labo- \nratory can sharply contract the time needed to cover a \u201cfading year\u201d \nand permit considerable reductions in cost.\n\nThe heart of the CSF is a group of four identical, variable channel \nnetworks, each of which produces, under computer control, a narrow-\n\nband Intermediate Frequency (IF) response that resembles the \u201cthree- \npath\u201d function of Rummler.\u2019 That function is commonly expressed in \nthe literature as\n\nwhere a, 6b and \u00a2 are slowly varying random \u201cfade parameters,\u201d and \nw(= 2rf) is measured from the selected intermediate frequency. For \nlater convenience, the following parameters are defined:\n\nThus each variable network has a response about the IF (either 70 \nMHz or 1.070 GHz, as selected) given by\n\nwhere a and c are computer-controlled attenuations and @ is a com-. \nputer-controlled phase shift. In terms of (1), the hardware variables c \nand @ are ab and \u00a2 + 7a, respectively. For later convenience, we define\n\nThe four variable channel networks are physically paired, i.e., two \nsuch networks with a common input but separate outputs are con- \ntained in each of two identical Dual Channel Units (DCUs). The four \nnetworks can be used in three distinct ways: (1) A single network (any \nof the four) can be used to produce the output of a single-pol nondi- \nversity channel, e.g., see Fig. 1a; (2) two networks within the same \nDCU can be used to produce the two outputs of a single-pol dual \ndiversity channel, e.g., see Fig. 1b; or (3) all four networks can be \ncombined, using a Cross-Coupling Unit (CCU), to produce the two \noutputs of a dual-pol nondiversity channel, e.g., see Fig. 2.\n\nWe thus see that a total of three units comprise the CSF, permitting \neither of three modes of use at either of two IF's. These and other \nfeatures of the CSF are summarized in Table I.\n\nTwo important facts about the CSF are important to emphasize. \nOne is that the responses of the four networks are completely and \nseparately controllable, and can be driven either via specified hardware \nparameters typed into a computer terminal (dial-up responses, which \nare fixed until the input parameters are changed); or via software \ncontrol, whereby the electronically adjustable hardware parameters \nare time varied by reading stored sequences from a disk and applying \nDigital-To-Analog (D/A) conversion and low-pass filtering. In the \nlatter case, the sequences are produced by software routines tailored \nto a specified statistical channel model. A wide range of models can\n\nFig. 1\u2014Simulation modes of (a) a single-pol channel nondiversity and (b) a dual \nspace diversity.\n\nthus be accommodated by suitably altering the software routines. This \nfeature is the key to the flexibility and simplicity of use of the CSF.\n\nThe second important fact is that designing the variable networks \nabout the \u201cthree-path\u201d function is not restrictive with respect to \npermissible channel models. The only assumption implicit in this \napproach is that all channel response functions of interest can be well \nfitted by this generic function. For the vast majority of radio paths \nthat have been measured, the evidence supports this assumption for \nchannel bandwidths up to 40 MHz. This means that if a given user of \nthe CSF wishes to consider a model with a different fitting function \n(e.g., the first-degree polynomial\u00ae or the \u201ctwo-path\u201d function\u2019), soft- \nware can be written to interface this model with the simulator hard- \nware.\n\nSection II describes the architecture of the CSF, showing block \ndiagrams of the three main units and discussing how they are used to \nprovide the various possible modes of operation. Section III gives a \nmore detailed description of the electronic circuits and components, \nwhile Section IV gives details on the construction. Section V discusses \nthe software associated with the CSF. Specific topics include the \ngeneration of the sequences that drive the variable hardware parame- \nters, and the calibration and measurement of the variable networks. \nThe existing software for generating the sequences is tailored to a \nspecific dual-diversity channel model. Section VI presents perform-\n\nTable I\u2014Definitions and some features of the channel \nsimulation facility\n\nCapabilities Simulates either a single-pol nondiversity channel \n(1 input, 1 output); a single-pol diversity channel \n(1 input, 2 outputs); or a dual-pol nondiversity chan- \nnel (2 inputs, 2 outputs).\n\nPhysical configuration Consists of two identical dual channel units (DCUs) \nand one cross-connecting unit (CCU), plus connect- \ning cables.\n\nDCU (Multipath fade simula- Contains two parallel networks (common input, sep- \ntor arate outputs), each producing a separate, variable \n\u201cthree-path\u201d frequency response.\n\nCCU (\u201cCross-pol coupler\u201d) Connects the four outputs of two DCUs so as to \nsimulate a dual-pol nondiversity channel. \nVariability of the network Computer-controlled; either keyboard entry fixed re- \nresponses sponses or program-generated time-varying re- \nsponses. \nInput IF 70 MHz \nInput power level 0 dBm \nOutput IF 70 MHz or 1.070 GHz, as selected. \nOutput power 0 dBm for each variable channel network at 70 MHz. \n~15 dBm for each variable channel network at 1.070 \nGHz. \nBandwidth Each network provides a \u201cthree-path\u201d response over\n\nance results. This includes specific data on bandwidths, power levels \nand calibration stability; assessments of how well the computer-gen- \nerated sequences satisfy the underlying statistical model; and assess- \nments of how well the hardware response variations match the in- \ntended ones, i.e., those dictated by the computer-generated outputs.\n\nFigure 3 shows, in simplified form, the block diagram of a DCU. A \nDCU contains two parallel networks with a common IF input, the top \nnetwork being labeled Horn Channel and the bottom are being labeled \nDish Channel. These labels are particularly apt when the two networks \nsimulate the two paths of a dual space diversity link. For convenience, \nwe will use these designations throughout our discussions to distin- \nguish the top and bottom networks. .\n\nAn actual DCU contains circuitry not depicted in Fig. 3, including \nbandlimiting IF filtering, upconversions from 70 MHz to 1.070 GHz, \nand the provision of the variable phase shifts (@) via local oscillators \nand mixers. This circuitry will be discussed in Section II.\n\nA control computer is used to control, through D/A conversion and \nlow-pass filtering, each of the circuit parameter a, c and 0. It is thus\n\nclear that each of the two networks provides a frequency response \nidentical in form to (4).\n\nFinally, Fig. 3 shows the manner in which the intermediate fre- \nquency of the network outputs is selected. When the external con- \nnecting cables (shown dashed) are absent, the two outputs are at 1.070 \nGHz. When the cables are connected as shown, each 1.070-GHz output \nis applied to a separate downconverter, with a shared 1.0-GHz Local \nOscillator (LO) providing the other input. In this case, the outputs \nwill be at 70 MHz, the IF of the DCU input.\n\nFigure 1 shows two obvious ways to use a single DCU. In Fig. 1a, \nthe control computer delivers fade parameters (A, C and @) to just the \nHorn Channel network, and the output of just that network is applied \nto the follow-on equipment. Alternatively, one could control, and use \nthe output from, just the Dish Channel. In either case, the simulator \nis used in this way to represent a single-pol nondiversity channel.\n\nIn Fig. 1b, the control computer delivers fade parameters to both \nthe Horn and Dish Channel networks, and both outputs are used. In \nthis case, the simulator can be configured as a dual space diversity\n\nchannel. On a typical space diversity link, the primary (upper) and \nsecondary (lower) receiving antennas on the radio tower would be of \nthe horn and dish type, respectively, thus leading to the designations \nused here. |\n\nFigure 4 shows how the CCU combines four inputs to produce the \nsimulated V-pol and H-pol outputs of a dual-pol channel: the co-pol \nV and cross-pol H signals are combined in a 10-dB directional coupler, \nfollowing a manually set attenuation and phase shift of the latter, to \nproduce the V-pol receiver input. Similar combining produces the \nH-pol receiver input. |\n\nThe manual attenuation adjustments in the CCU, achieved via. \npanel-mounted controls, extend the dynamic range over which the \ncross-pol coupling levels can be varied. Each attenuator can be incre- \nmented in 1-dB steps from 0 dB to 43 dB.\n\nThe manual phase shift adjustments, also achieved via panel- \nmounted controls, provide flexibility in how the co-pol and cross-pol \nsignals from the same original transmission are relatively phased. The \nphase adjustment is continuous over a range of nearly 360 degrees.\n\nEach variable channel network of a DCU contains a direct path \nwith a variable (analog voltage controlled) attenuator, and a parallel \npath with a 6.3-ns delay, a variable attenuator and a variable (analog \nvoltage controlled) phase shifter. The summed outputs of these paths \nyield a network response akin to the \u201cthree path\u201d function of Rumm- \nler. Each of the parallel paths must be nondispersive, in fact, over a \nbandwidth of 40 MHz or more. Further, it should be possible to vary \neach attenuator without introducing phase changes and to vary the \nphase shifter without introducing attenuation changes.\n\n\u2018An input frequency of 70 MHz and a channel bandwidth of 40 MHz \nwere chosen for the CSF. The variable channel networks could be \nbuilt at this or any other frequency if ideal components were available. \nHowever, given the limitations of practical components, we chose the \napproach of converting the input signal frequency to a higher fre- \nquency, where narrowband circuit techniques can be used. This ap- \nproach, moreover, simplifies the task of providing the variable phase \nshift, as we shall see.\n\nA network frequency of 1.070 GHz was chosen. This choice trades \noff the availability of \u201cphase shift free\u201d narrowband attenuators, and\n\nease of filtering of unwanted mixing products and local oscillator \nleakage through the mixer. The network output at 1.070 GHz can be \neither downconverted back to 70 MHz, used directly, or easily upcon- \nverted to, say, 4, 6, or 11 GHz for radio equipment tests.\n\nA simplified block diagram of a single variable channel network is \nshown in Fig. 5. The input signal at 70 MHz (Point 1) is power divided \u2014 \ninto two paths. One component is upconverted in a double-balanced \ndiode mixer to 1.070 GHz; the other is delayed and then similarly \nupconverted to 1.070 GHz by a second double-balanced diode mixer. \nThe output from the first mixer is controlled in amplitude by a variable \nattenuator and provides one input to a signal combiner. The output \nfrom the second mixer (in the delayed path) is controlled in amplitude \nby a second attenuator and in relative phase by the local oscillator \nphase shifter. This delayed signal sums with the direct input in the \noutput signal combiner.\n\nThe 6.38-ns delay is provided by adding additional coaxial cable in \nthe 70-MHz signal path between the input power divider and the \ndelayed path upconverter. The exact length of this cable is determined \nempirically by measuring the signal delay difference at the output \nsignal combiner (points 2 and 3). This is most easily done in the \nfrequency domain by sweeping the network input signal, setting the\n\ndirect path and delayed path attenuations to be about equal (to form \nperiodic deep fades), and then measuring the frequency spacing be- \ntween successive nulls. When the path difference is 6.3 ns, the null \nspacing is 158.73 MHz, which is the reciprocal of the delay.\n\nThe output of the variable channel network contains both the upper \nand lower mixing products at 1.070 GHz and 0.930 GHz. If the 1.070- \nGHz signal is to be mixed back to 70 MHz, the contribution from the \n0.930-GHz signal will distort the desired channel response and must \nbe effectively eliminated by filtering.\n\nA detailed block diagram of a DCU is shown in Fig. 6. This diagram \nshows the two variable channel networks driven by a common 70- \nMHz input signal and the two downconverters for mixing the simulator \noutputs back to the input frequency. A crystal-controlled phase-locked \nContinuous Wave (CW) source at 1 GHz acts as a common local \noscillator to all upconverters and downconverters, to ensure frequency \ncoherence between the input signal and both output signals. When \nboth DCUs are used in the complete CSF, they are virtually identical \nexcept for the provision to drive all four variable channel networks \nfrom a common local oscillator residing in one DCU (master). This \nhigher power source in the master DCU can drive the second DCU \n(slave). The result is four frequency-coherent outputs, assuming the \ninput signals to the two DCUs are themselves frequency coherent. (As \ndescribed in Section 2.2, the two input signals to the DCUs may or \nmay not be frequency coherent, depending on the experiment.)\n\nEach variable network uses two local oscillator signals for upcon- \nversion. One is derived through power dividers from the 1-GHz source; \nthe other passes first through a voltage-controlled phase shifter. The \nphase shifter varies the phase of the 1-GHz CW signal from 0 degrees \nto over 360 degrees, when the @ control signal varies between 0 and \n+10 volts. This phase is imparted to the signal in the delayed path of \nthe variable network by the mixing process of the upconverter.\n\nThe signals are scaled in the direct and delayed paths of each \nvariable channel network by voltage-controlled attenuators. The at- \ntenuations are variable from 0 dB to over 36 dB when the A and C \ncontrol signals vary between 0 and +10 volts. The outputs from the \nattenuators are added in a reactive signal combiner. The combined \nsignal is bandpass filtered to eliminate the lower mixing product at \n0.930 GHz. The desired output from the filter at 1.070 GHz is amplified \nto provide the final fading channel output.\n\nThe output filter, a fifth-order Butterworth filter with a 3-dB \nbandwidth of 70 MHz, is the most narrowband circuit of each variable \nnetwork. This filter adequately attenuates the unwanted mixing prod-\n\nuct while maintaining a flat amplitude response with small time delay \nvariation (+1.5 ns) over the design bandwidth of 40 MHz.\n\nto prevent reflections from mismatches that produce unwanted ripples\n\nin the channel response. Fixed coaxial attenuators and ferrite isolators\n\nAs we have seen, the four attenuators and two phase shifters within \neach DCU are controlled by time-varying analog voltages, the latter \nbeing derived from digital sequences supplied by the control computer. \nTo accomplish this, two D/A converters mounted within the computer \nprovide six control voltage sequences to each DCU. Each sequence \nconsists of 12-bit words delivered at a 1-Hz rate. Each sequence is \nconverted into a smooth analog variation by a fourth-order Butter- \nworth state variable low-pass filter with a 3-dB bandwidth of 0.5 Hz.\n\nline (.141 inch) was used to interconnect components to improve phase \nstability. Line lengths in both the direct and delay paths were closely \nmatched in each variable channel network of each DCU, so as to \nobtain nearly identical characteristics. The elongated coils in the \ncenter of the figure are the additional line lengths required to produce \nthe 6.3-ns path delays.\n\n8. The input and output signal ports are available on the panel for \ninterconnection convenience. The 70-MHz input port and the 70-MHz \ndownconverter output ports are the three BNC-type coaxial connec- \ntors in the lower center of the panel. The 1.070-GHz outputs from the \nHorn and Dish variable channel networks are brought out on N-type \ncoaxial connectors located at the upper left and upper right sides of \nthe panel. The inputs to the two downconverters associated with the \nHorn and Dish channels are located directly below these N-type \nconnectors. \n_ Screw-trimmer adjustments, paired in 12 holes in the upper center \nleft of the panel, are provided for shifting and scaling the control \nvoltages to the four variable attenuators and two phase shifters. The \nrange of these control voltages is approximately preset by adjustments \non the D/A converter boards within the control computer. The front \npanel adjustments allow for convenient trimming at the DCU. The \npanel meter can be selectively switched to monitor each of these \nvoltages.\n\nFigures 9 and 10 show a top view of the inside and the front panel \nof the cross-coupling unit. This relatively simple unit contains the two\n\ncross-pol phase shifters (large rectangular devices in the center), the \ntwo cross-pol attenuators (cylindrical devices adjacent to the phase \nshifters) and two 10-dB directional couplers.\n\nThe front view of the complete CSF is shown in Fig. 11. The cable \ninterconnections shown represent the cross-coupled case described \nin Section 2.2. The lower DCU can be arbitrarily called the \nV-pol channel and the upper DCU, the H-pol channel. The V-pol \nHorn Channel output is connected to the left cross-pol coupler Main \ninput. The H-pol Horn Channel output is connected to the left cross- \npol coupler X-Pol input, to be attenuated and phase-shifted before \nsumming to the main V-pol signal. The dish and horn outputs from \nthe H-pol and V-pol channels, respectively, are similarly combined in \nthe right cross-pol coupler. The cross-pol coupler outputs are shown \nreturned to the respective downconverters in the DCUs to provide 70- \nMHz outputs. If desired, instead, the 1.070-GHz outputs from the \ncoupler can be used directly.\n\nV. SOFTWARE DESIGN \nThis section deals with two distinct aspects of the CSF software.\n\nOne is the computer routines that generate random time sequences \nfor parameters of the specified channel models; the other is the \nsoftware that runs on the control computer to effect real-time opera- \ntion of the CSF hardware.\n\nThe sequence generation software was developed on the UNIX\u2122 \noperating system with the option to move it to the control computer. \nIn the initial arrangement, however, the fade parameter sequences \nwere generated on a UNIX system minicomputer and down loaded to \nthe control computer. As described earlier, these sequences were \nconverted into continuous analog signals and played into the simulator \nhardware to achieve time-varying channel responses. Special software \nwas written to test the accuracy of the generated sequences and of the \nresulting hardware responses. These issues are dealt with in Section \nVI.\n\nWe describe here how to generate the random time-varying param- \neters for a fade model. We will first summarize the generation proce- \ndure employed by our software to generate random variations with \nany given Probability Density Function (PDF) and Autocorrelation \nFunction (ACF). Then we will describe the techniques used to generate \nrandom variations with PDFs and ACFs specified by Rummler.\u2019 In \nthis subsection, we show how to generate the random variations A, B, \nand @, as defined in (2), for both the Horn and Dish Channels. The \ntransformation of these variations into A, C, and @ is performed by \nthe control computer and is described briefly in a later section on \ncontrol computer software.\n\nThe problem of generating a random variation with arbitrary PDF \nand ACF is difficult in general and sometimes impossible.** In our \ncase, however, we wish to match the statistics of fade models with \nwell-behaved PDFs and ACFs. Further, while accurate matching of \nthe PDF is essential, particularly when the simulator is used for outage \nmeasurements, slight variations in the ACF shape are acceptable since \nthe ACF is either unknown or only sparsely measured for many models. \nThese considerations ease the variation generation problem consider- \nably. It should be noted that the second-order statistics are not \nuniquely determined by the PDF and ACF and that we will accept \nwhatever second-order statistics are generated when matching the \nPDF and ACF.\n\nThis method is general in that it generates random variations with \nany specified PDF and with an ACF close to the specified ACF. The \nmethod starts by generating a Gaussian random process with an initial\n\nACF that we normally take as the desired ACF. From the Gaussian \nvariation, a new variation with the required PDF can be formed by a \nnonlinear transformation. It is easier to think of this transformation \nin two steps\u2014a Gaussian-to-uniform transformation followed by a \ntransformation to the required PDF. The first transformation involves \nusing the Q function, while the second transformation involves using \nthe inverse of the required distribution function. While this gives \nvariations with the correct PDF, the nonlinear transformations will \nmake the output ACF different from the initial ACF.\n\nTo make the output ACF match the desired ACF, we use the \nfollowing iterative procedure: The initial ACF is applied to the under- \nlying Gaussian variation. Based on the output ACF and an educated \nguess, we select a new ACF for the underlying Gaussian variation. \nThis procedure stops when the desired ACF is obtained. In our case, \nthe iterative procedure converges quickly. We found that the ACFs of \ntwo of our parameters remained essentially unchanged after the non- \nlinear transformations.\n\nThe nonlinear transformation required to convert the Gaussian \nvariations into variations with the desired PDF is broken up into a \ntransformation to uniform followed by a transformation to the desired \nPDF. The transformation to uniform can be accomplished through \nthe Q function, i.e.,\n\nwhere x is the original Gaussian variation and u is the uniform \nvariation. This function can also be used as a test of the quality of the \nGaussian variation by computing the new probability distribution \nfunction and comparing the result with the uniform probability dis- \ntribution function. This test was used to verify the uniform-to-Gaus- \nsian random variation conversions.\n\nWe chose to use the Rummler dual-channel space diversity fade \nmodel,\u2019 although Rummler\u2019s rationalized model\u00ae or any other dual \nchannel model could have been employed as well. Familiarity with \nRummler models is assumed in the following paragraphs.\n\nSince ACF measurements are not available for this model but are \navailable for the single-channel model,** we use the single-channel \nmodel ACFs for both channels of the diversity model. These ACFs, \nmeasured by Rummler, are not true ACFs; they were rescaled so that \nR(0) = 1. The scaling factor was determined by extrapolating the \nmeasured ACF to the ordinate and finding the intercept. The ACF for \neach of the fade parameters closely approximates an exponential. For \nthis reason, we applied an exponential ACF to each underlying Gaus- \nsian variation and adjusted its time constant to form different ACFs. \nMatching the time constants at the 1/e points of the rescaled Rummler \nACFs gives time constants of 129, 53, and 32 seconds for the respective \nA, B, and \u00a2 parameters. These ACF's were used in generating the \ndniderlying Gaussian variation discussed above.\n\nThe easiest parameters to generate are the \u00a2\u2019s because evs are \nindependent of each other and of all the other parameters. For each \n@, we begin with U(0, 1) variation, u, generated as discussed in the \nprevious subsection. The \u00a2 parameter, with a pedestal-type PDF, is \ngenerated from the u variation by the simple rescaling\n\nOk oe + a)(2u \u2014 1) + 90(1 \u2014 a)sgn(2u \u2014 1); otherwise, (7) \nwhere a is the pedestal parameter (5 for the Horn Channel and 8 for \nthe Dish Channel) and \u00a2 is in degrees.\n\nThe B parameters are generated next because, in both the Horn and \nDish Channels, the A parameter\u2019s mean value is dependent on the B \nparameter. The general technique for converting a U(0, 1) random \nvariation to a particular distribution can be accomplished by passing \neach sample through the inverse of the distribution function. For the \nB parameters, finding the inverse function algebraically is rather \ncomplicated, so this was done numerically.\n\nFinally, the A parameters are generated. For this case, the final \ndistribution is Gaussian so no transformations are need to obtain the \ncorrect PDF. However, the mean of each A is dependent on the \nassociated value of B. Further, Ajom is correlated with Agis,. To \ngenerate the A\u2019s, two independent N(0, 1) random variations (x, and \nX2) with identical ACFs are generated. Next, they are linearly combined \nto produce the desired correlation and, finally, the variances are scaled \nand appropriate mean values are added. In short,\n\nAdish = POdishX1 + Gaish V1 \u2014 p?X2 + Maish(Baisn), _ (9) \nwhere the Greek constants and the u(B) functions are defined in Ref. \nvi\n\nThe control computer software falls into two categories, namely \nsoftware to produce the fades and software to calibrate and test the \nhardware. The purpose of the fade software is (1) to translate fade \nmodel variations. (e.g., A, B, and \u00a2) into the A, C, and 6 variations \nsupported by the hardware; (2) to perform calibration table lookup \nand quantization to the closest binary output sample value; and (3) to \nplay the binary variations through the analog hardware. The remaining \nsoftware generates the calibration tables, tests the overall accuracy of \nthe fade control and calibration software, as well as the analog hard- \nware, and allows manual control over the fade simulator. These \noperations are described in more detail in the following subsections.\n\nThe analog hardware implements the transfer function given by (4), \n(see Fig. 5), where c = ab and 0 = \u00a2 + x. The voltage-controlled \nattenuators have control signals proportional to 10 log (power atten- \nuation). These final log conversions are handled by a lookup table \n(described shortly), but the software-generated parameters A and B \nare in decibel form and must first be converted to linear form before \ncomputing c. The phase offset of + is required in the hardware to \nensure continuity into the D/A low-pass filter.\n\nTo ensure accuracy, calibration lookup tables are used to map and \nquantize the analog floating point values generated from the fade \nmodel software into one of the integer output values of the D/A \nconverter. Software automatically finds the closest table entry to the \nfloating point value and outputs the corresponding integer value. \nSeparate software outputs these integer variations from RAM memory \nor disk to the D/A converters at a 1-Hz rate.\n\nThe calibration lookup tables discussed above are automatically \ngenerated under software control via hardware measurements taken \nwith a network analyzer connected to a computer. The a and c \nparameter calibrations are similar as shown in Figs. 12a and b and \nrequire the disconnection and termination of the delayed path (direct \npath for the c parameter). The control computer then measures the \nend-to-end attenuation of the connected path for each of the 4096 \nvalues and generates a table. These values are smoothed by averaging \nwithin a window centered around each value. This is required to obtain \na monotonic table, especially for large attenuations where measure- \nments become noisy.\n\nTo simplify the software, the 6 calibration was made interactive. \nWith both paths connected as in Fig. 12c, a spectral dip or notch is\n\n(C) CALIBRATION OF 0 \nFig. 12\u2014Block diagrams of parameter calibrations: (a) calibration of a, (b) calibration \nof c, and (c) calibration of 6.\n\nformed by manually adjusting the relative values of a and c. This \nnotch is placed at the upper and lower band edges as well as the center \nof the channel and the binary control values are stored. This infor- \nmation is then used to generate the lookup table via interpolation and \nextrapolation. The extrapolation is needed since notch position mea- \nsurements can only be made in the bandlimited channel.\n\nAdditional software allows for manual control over each of the \nparameters, A, C, and @, in real time. This provides for quick checking \nof the hardware and the lookup tables. This software also provides a \nuseful mode of operation for the simulator, allowing selected responses \nto be easily \u201cdialed\u201d into the hardware.\n\nFinally, there is software for playing a selected response and com- \nparing it with the actual response measured by the network analyzer \nto obtain an error measure. More details about this are contained in \nthe performance section that follows.\n\nWe will evaluate the performance of the CSF in several ways here. \nFirst, we illustrate some \u201cthree-path\u201d responses produced by the \nvariable channel networks over the 40-MHz bandwidth of interest. \nNext, we demonstrate that the joint statistics of the computer-gener- \nated sequences of fade variables (A, B and \u00a2) conform to the model \nfrom which they are derived. Finally, we demonstrate that the fre- \nquency responses of the hardware, over a computer-generated popu- \nlation of A, C, 6 values, are indeed the responses these fade variables \nare intended to produce. With these demonstrations, we affirm that \nthe CSF can produce laboratory fading environments similar to those \non actual digital radio links.\n\nThe static and dynamic transmission performance of each DCU was \nmeasured using a Hewlett-Packard Model 8505A Network Analyzer. \nHere we discuss frequency response measurements on one or both \nvariable channel networks of a DCU, and show sample results.\n\nIn a typical measurement, fixed control voltages are set by keyboard \ninputs on the control computer, and applied to the attenuators and \nphase shifters of the variable channel networks. The values of the \ncontrol voltages produce a particular transmission function in each \nnetwork. The Network Analyzer provides a frequency-swept input \nsignal, in the vicinity of 70 MHz, to the DCU. The two 70-MHz \noutputs from the DCU are returned to the analyzer, where the fre- \nquency response magnitudes of the two networks are simultaneously \nobserved. Alternatively, only one network output from the DCU is \nreturned to the analyzer, along with a reference sample of the input\n\nsignal, to permit measurement of both the transmission magnitude \nand phase (or group delay) of that network.\n\nIn our initial experiment, we set the control voltage of the delay \npath attenuator so as to give maximum attenuation (>36 dB). This \nenabled us to measure the transmission performance of the direct path \nalone, with the output filter and downconverter included. Figure 13 \nshows the corresponding frequency response for one channel of a \nDCU. The upper trace is the magnitude and the lower trace is the \ngroup delay. Over a 40-MHz bandwidth centered at 70 MHz, the \nmagnitude is flat within +0.12 dB and the group delay is flat within \n+1.5 ns. This response is primarily determined by the output Butter- \nworth filter described in Section 3.3. |\n\nIn another experiment, we set the control voltage to the delay path \nattenuator so that the signal level at the variable channel network \ncombiner was 0.92 dB weaker than that of the direct path. A minimum- \nphase fade was thus created, with a maximum fade depth 20 dB below \nthe direct path gain. The upper trace in Fig. 14 shows the magnitude \nof the resulting response. The delay path phase shifter was set to place \nthe maximum fade in the center of the channel passband. The lower \ntrace shows the channel group delay, wherein the time delay distortion \ncan be clearly seen: Frequency components at the edges of the channel \nare delayed by more than 60 ns relative to components in the center \nof the channel.\n\nWhen the signal levels in the direct and delay paths are interchanged \nso that the delayed signal to the network combiner is 0.92 dB stronger,\n\na nonminimum-phase fade occurs, with the same maximum fade depth \nof 20 dB. The upper trace in Fig. 15 shows the transmission magnitude \nof the resulting channel response, with the maximum fade placed once \nmore in the center of the channel passband. The time delay distortion \nis shown in the lower trace. Under these conditions, signal components \nat the center of the channel are delayed relative to components at the \nedges of the channel.\n\nIn general, by appropriately setting the relative signal levels in the \ndirect and delay paths of a variable channel network, the depth of fade \nin the channel response can be selected. This fade can then be \npositioned in frequency by appropriately setting the phase shift in the \ndelay path. Figure 16a shows the transmission magnitude of the\n\nFig. 16\u2014Transmission magnitude response as a fade is moved in frequency across \nchannel.\n\nchannel response as a 10-dB fade is moved in frequency across the \nchannel. The fade depth remains constant within better than +0.25 \ndB over the 40-MHz channel bandwidth. If the delay path signal level \nis increased in magnitude to produce a 20-dB fade depth, the variation \nin fade depth across the channel increases to +1.0 dB, as shown in \nFig. 16b. These variations are due to minor amplitude response im- \nbalances between the direct and delay paths. The +1.0-dB variation \nfor a 20-dB fade depth, for example, is caused by imbalances of about \n+0.1 dB.\n\nThe DCU was designed to be a general-purpose laboratory instru- \nment. A nominal input signal level of 0 dBm was chosen for the 70- \nMHz DCU input. This level is normally available from the modulator\n\nportion of typical radio modems. The unfaded 70-MHz output from \neach network of the DCU is designed to have a level of 0 dBm when \nthere is 3-dB attenuation in the direct path. This 3 dB can be removed, \nunder software control, to provide some up-fade capability. The un- \nfaded outputs at the alternative frequency of 1.070 GHz are designed \nfor a level of \u201415 dBm.\n\nFinally, test ports are provided to permit frequency response mea- \nsurements of each variable network prior to the 1.070-GHz bandpass \nfilter. The signal level at each of these ports is nominally \u201452 dBm.\n\nThe distributions were obtained from a single variation containing \n50,000 samples, which corresponds to about 1666 independent samples \nfor the ACF used. Similar results were obtained for the other fade \nparameters but are not shown for brevity.\n\nFor the pedestal-type density functions for \u00a2, accuracy was checked \ndirectly by measuring the slopes of the final distribution functions. \nThis yielded the desired values of 5 and 8 for the density function \npedestal ratios for the Horn and Dish Channels. The A parameter \ndistributions were checked using the same Gaussian-to-uniform trans- \nformation as in the generation procedures for B and \u00a2. Again, a linear \nplot should result and was obtained.\n\nThe ACF for Baish are shown in Fig. 19. The time separation among \nACFs for the underlying Gaussian, uniform, and final variations is \nabout 2 seconds for this parameter. Results for Bor, and the \u00a2 \nparameters are similar but show less variation. The underlying Gaus- \nsians variations were adjusted so that their ACF's agree with the fade \nmodel ACFs at the 1/e points.\n\nThe ACF for the Gaussian variations of Aporn and Agisn must be the \nsame so that the resulting distributions both follow the required ACF. \nTherefore, the ACF time constants for the Horn and Dish Channels \nare set equal and adjusted together. It was found that changes in ACF \nresulted when the A parameters were correlated and then each was \nbiased by a mean related to B, as specified by the model. This is shown \nin Fig. 20.\n\nand \u00a2 parameters, respectively. The agreements are quite good. \nThus, we have carried out our parameter generation method for the\n\nFig. 21\u2014Comparison of desired Rummler ACF with measured ACF for Apom and \nAdish:\n\n. Fig. 22\u2014Comparison of desired Rummler ACF with measured ACF for Bhom and \ndish:\n\naccuracy. For reasonable PDFs, our iterative procedure yields a ran- \ndom variation whose ACF is close to the desired ACF when the latter \nis applied to the underlining Gaussian variation. The B and \u00a2 param- \neters showed little variation or no variation after the nonlinear func-\n\nFig. 23\u2014Comparison of desired Rummler ACF with measured ACF for \u00a2porm and ddish-\n\ntions were applied. The A parameter (which does not use the iterative \ngeneration method) showed a modest variation due to the addition of \nthe mean which is related to the B parameter. Finally, the PDFs of \nthe generated parameters have been shown to agree with the PDFs \nspecified by the model.\n\nIn this subsection, we assess the accuracy with which the simulation \nhardware produces the responses called for by the software. We begin \nby discussing appropriate error measures and end with experimental \nresults.\n\nConsider any one of the variable channel networks, i.e., the Horn \nChannel or Dish Channel of a given DCU. The input control variables \n(A, C, and @) to this network at any instant are intended to produce a \nshort-term frequency response [see (2), (4), and (5)] given by\n\nAs the control variables change slowly, H(f) is meant to change \nslowly, too, in accordance with this equation. |\n\nNow suppose that, for given A, C, and 6, we measure the actual \nresponse of the network, Ho(/), and compute some index of its depar-\n\nture from the intended response, (10), over some bandwidth. For \nexample, we could compute\n\nwhere the overbar denotes a frequency integration from, say, -20 MHz \nto +20 MHz. Next, suppose that \u00ab were computed at fixed intervals in \ntime as A, C, and @ moved slowly through their computer-generated \nvariations. We could then obtain a population of e-values and compute \na cumulative distribution, P(e). If the joint variations of A, C and @ \nfor this experiment were representative of a particular statistical \nfading model, we could say that P(e) is the error-measure distribution, \nover the fade response ensemble of that model, of the variable network. \nFinally, a set of such distributions for all four variable networks could \nbe used to characterize\u2014for that particular model\u2014the accuracy of \nthe simulator hardware. |\n\nFirst, we chose, for the sake of simplicity, to avoid measuring the \ncomplex response Ho(/); instead, we measured its squared magnitude \n(amplitude response) only. This function can be usefully compared \nwith the squared magnitude of H(f), (10), by means of the following \nerror measure:\n\nand the frequency averaging is over the specified bandwidth. Given \nthe uncomplicated nature of the frequency-selective networks, we \nassert that little is lost by omitting phase response data from the error \nmeasures; amplitude response accuracy alone is a reliable indicator of \nhow well Ho(f) matches H(/).\n\nIt can be shown that the above error measure can be used to \napproximate quite closely the root mean square decibel error between\n\n* We show the control variables A, C and @ here to make their pertinence explicit. \nHenceforth, we shall suppress them.\n\nG(f) and Go( f ), as averaged over the specitiad bandwidth. Specifically, \nthis quantity is approximated by (4.34 ve) dB for all \u00a2 < 0.2.\n\nOur second change was to separate the error measure into a scale \nerror and a shape error.* This approach acknowledges the fact that a \nsmall scale difference between G(f) and G)(f) (i.e, a small level \ndifference in their decibel variations) would be relatively unimportant \ngiven the vagaries of transmitter power, free space path loss, receiver \nnoise figure, and other link power quantities. We estimate that pure \nlevel differences lying within +1 dB would be quite acceptable in view \nof these factors.\n\nAccordingly, we modify \u00a2 in (12) so as to permit a scaling of Go(f) \nby an \u201coptimal\u201d factor, ropt, to be defined shortly. The value of \u00ab with \nthis scaling incorporated is then a measure solely of the shape differ- \nence between the intended and measured responses. The mathematics \nfollows.\n\nThe minimizing r, which we define to be optimal and call ropt, is easily \nfound to be\n\n(Gol f)/G(f)) \nopt > oS SS 15 \not GA/GAY. = \nand the resulting \u00ab is \n[(Go( f)/G(f))I\u00b0 \n\u2014 7 16 \n(Ga fV/GAY? ve \nFor a given A, C and 6, the quantity \nk= 10 logio(Topt) (17)\n\nis the scale error of the variables network, in decibels, for these control \nvariables; and the quantity\n\nestimates the root-mean square decibel shape error (or just shape \nerror) of the variable network for these control variables. A conserv- \native accuracy requirement is that these quantities lie within +1 dB \nand below 0.5 dB, Bi over all likely combinations of the \ncontrol variables.\n\n* Ifthe power responses G(f) and Go(f) are replaced by their decibel equivalents, this \nseparation is akin to isolating the \u201cdc\u201d and \u201cac\u201d error variations with respect to f.\n\nThe error measures defined above were obtained for each of the four \nvariable channel networks of the CSF. The results can be regarded as \nmore than a check on the hardware alone. They also reflect on the \naccuracy of the software translations into electrical signals, the lookup \nroutines and tables, and the software that generates those tables.\n\nThe hardware and software test configuration is shown in Fig. 24. \nThe control computer first calculates the magnitude of the desired \nresponses G(/), from the down loaded files containing A, B, and \u00a2 \nparameters. These files are also used to generate the binary A, C, and \n6 values for the D/A converters by transformations and lookup tables \nas previously discussed. The D/A converters are loaded with the binary \nvalues and after a delay to allow the lowpass filters on the D/A control \nlines to settle, the response of the channel, Go(/), is sampled at 40 \nfrequencies spaced at 1-MHz intervals over the specified bandwidth. \nThese frequency samples are compared to the calculated samples, and \nerror measures are computed and saved on a disk. Specifically, for \neach time sample of G(f) and Go(/), we computed \u00a2 and r.,; using (16) \nand (15). In doing so, we used numerical-sum approximations to the \ncontinuous-frequency integrations called for in these equations. Given \nthe smooth frequency responses produced by the variable channel \nnetworks, we found the 1-MHz spacings between measured values of \nG and Gp to be quite adequate.\n\nThe final step in the procedure was to obtain, for Sich network, a \ncumulative distribution from the 7200 values of e, and another for Fropt. \nResults for r.\u00bb, are given in Fig. 25 for the Horn and Dish Channels\n\nFig. 25\u2014Cumulative distributions of scale error, R, for each channel of a dual channel \nunit. |\n\nFig. 26\u2014Cumulative distributions of shape error, o, for each channel of a dual channel \nunit.\n\nof one of the DCUs; the corresponding results for \u00ab [converted to o via \n(18)] are given in Fig. 26. Very similar results were obtained for the \nother DCU. There are small but discernible differences between the \ndistributions for the Horn and Dish Channels. Mostly, these are due\n\nto the slightly different multipath fading histories generated for these \ntwo channels by the software.\n\nThe important findings from Figs. 25 and 26 are that the scale error \n(R) lies well within +1 dB over all fades and that the root-mean- \nsquare decibel error (or shape error, a) lies below 0.5 dB for over 98 \npercent of all fades. This is true for all four networks of the CSF and \nis the result we wanted. The accuracy of the hardware is thus con- \nfirmed.\n\nThe authors want to thank M. F. Wazowicz for his help in hardware \nassembly and E. C. Cox, P. Knoetgen and V. R. Dillard for their help \nin the development of control software. We also want to especially \nthank L. J. Greenstein and Y. S. Yeh for their generous advice and \nsupport in both design and documentation of the simulation facility.\n\n. W. D. Rummler, \u201cA New Selective Fading Model: Application to Propagation Data,\u201d \nB.S.T.J., 58, No. 5 (May-June 1979), pp. 1037-71.\n\n. L. J. Greenstein and B. A. Czekaj, \u201cA Polynomial Model for Multipath Fading \nChannel Responses,\u201d B.S.T.J., 59, No. 9 (September 1980), pp. 1197-225.\n\n. W. D. Rummler, \u201cMore on the Multipath Fading Channel Model,\u201d IEEE Trans. \nCommun., COM-29, No. 3 (March 1981), pp. 346-52.\n\n. M. Liniger, \u201cSweep Measurements of the Transferfunction of an RF-Channel and \nTheir Representation by Polynomials,\u201d Intl. Conf. Commun., Philadelphia, Penn., \nJune 13-17, 1982, paper 7B3.\n\n. L. Martin, \u201cStatistical Results on Selective Fading,\u201d Intl. Conf. Commun., Phila- \ndelphia, Penn., June 13-7, 1982, paper 7B85.\n\n7. W. D. Rummler, \u201cA Statistical Model of Multipath Fading on a Space Diversity\n\nRadio Channel,\u201d B.S.T.J., 61, No. 9, Part 1 (November 1982), pp. 2185-219. \n8. W. D. Rummler, \u201cA Rationalized Model for Space and Frequency Diversity Line- \nof-Sight Radio Channels,\u201d Int. Conf. Commun., June 19-22, 1983, Boston, Mass., \npaper E27. \n. K. T. Wu, \u201cMeasured Statistics on Multipath Dispersion of Cross Polarization \nInterference,\u201d Intl. Conf. Commun., May 14-7, 1984, Amsterdam, paper 46.3. \n10. M. H. Meyers, \u201cMultipath Fading Characteristics of Broadband Radio Channels,\u201d \nGLOBECOM \u201984, Atlanta, Ga., November 26-29, 1984, paper 45.1.\n\n11. M. Liniger, \u201cSweep Measurements of Multipath Effects on Cross-Polarized RF- \nChannels Including Space Diversity,\u201d GLOBECOM \u201984, Atlanta, Ga., November \n26-9, 2984, paper 45.\n\n12. L. J. Greenstein and V. K. Prabhu, \u201cAnalysis of Multipath Outage With Applications\n\n13. W. C. Jakes, Jr., \u201cAn Approximate Method to Estimate an Upper Bound on the \nEffect of Multipath Delay Distortion on Digital Transmission,\u201d IEEE Trans. \nCommun., COM-27, No. 1 (January 1979), pp. 76-81.\n\n14. R. P. Coutts and J. C. Campbell, \u201cMean Square Error Analysis of QAM Digital \nRadio Systems Subject to Frequency Selective Fading,\u201d A.T.R., 16, No. 1, 1982, \npp. 23-38.\n\n15. L. J. Greenstein and B. A. Czekaj-Augun, \u201cPerformance Comparisons Among Digital \nRadio Techniques Subjected to Multipath Fading,\u201d IEEE Trans. Commun., COM- \n30, No. 5 (May 1982), p pp. 1184-97.\n\n16. G. J. Foschini and J. Salz, \u201cDigital Communications Over Fading Radio Channels,\u201d \nB.S.T.J., 62, No. 2, Part 1 (February 1983), pp. 429-56.\n\n17. D. P. Taylor and M. Shafi, \u201cFade Margin and Outage Computation of 4\u00a2-QPRS \nRadio Employing Decision Feedback Equalization,\u201d Intl. Conf. Commun., Boston, \nMass., June 19-22, 1983, paper F2.1.\n\n18. O. Andrisano, \u201cThe Combined Effects of Noise and Multipath Propagation in \nMultilevel PSK Radio Links,\u201d IEEE Trans. Commun., COM-32, No. 4 (April \n1984), pp. 411-8.\n\n19. N. Amitay and L. J. Greenstein, \u201cMultipath Outage Performance of Digital Radio \nReceivers Using Finite-Tap Adaptive Equalizers,\u201d IEEE Trans. Commun., COM- \n32, No. 5 (May 1984), pp. 597.\n\n20. W: C. Wong and L. J. Greenstein, \u201cMultipath Fading Models and Adaptive Equal- \nizers in Microwave Digital Radio,\u201d IEEE Trans. Commun., COM-32, No. 8 \n(August 1984), pp. 928-34.\n\n21. N. Amitay and J. Salz, \u201cLinear Equalization Theory in Digital Data Transmission \nOver Dually Polarized Fading Radio Channels,\u201d AT&T Bell Lab. Tech. J., 63, \nNo. 10, Part 1 (December 1984), pp. 2215-59.\n\n22. L. J. Greenstein and Y. S. Yeh, \u201cA Simulation Study of Space Diversity and \nAdaptive Equalization in Microwave Digital Radio,\u201d AT&T Tech. J., 64, No. 4 \n(April 1985), pp. 885-905.\n\n23. M. Emshwiller, \u201cCharacterization of the Performance of PSK Digital Radio Trans- \nmission in the Presence of Multipath Fading,\u201d ICC78, June 1978, Paper 47.3.\n\n24. C. W. Lundgren and W. D. Rummler, \u201cDigital Radio Outage Due to Selective \nFading-Observation vs. Prediction from Laboratory Simulation\u201d B.S.T.J., 58, No.\n\n25. Y. Y. Wang, \u201cSimulation and Measured Performance of a Space Diversity Combiner \nfor 6 GHz Digital Radio,\u201d IEEE Trans. Commun., COM-27, No. 12 (December \n1979), pp. 1896-1907.\n\n26. S. T. Matsuura, \u201cEstimated Performance of a QPR Digital Microwave Radio in the \nPresence of Frequency Selective Fading,\u201d Intl. Conf. Commun., Philadelphia, \nPenn., June 13-7, 1982, paper 7B2.\n\n27. J.C. Campbell and R. P. Coutts, \u201cOutage Prediction of Digital Radio Systems,\u201d \nElectron. Lett., 18, No. 25/26, December 1982, pp. 1071-2.\n\n28. A. L. Martin, R. \u2019P. Coutts and J. C. Campbell, \u201cResults of a 16 QAM 140 Mbit/s \nDigital Radio Field Experiment,\u201d Intl. Conf. Commun., Boston, Mass., June 19- \n22, 1983, paper F2.2.\n\n29. C. P. Bates and M. A Skinner, \u201cImpact of Technology on High Ganccigy Digital \nRadio Systems,\u201d Intl. Conf. Commun., Boston, MA, June 19-22, 1983, paper F2.3.\n\n30. S. Barber, \u201cCofrequency Cross-Polarized Operation of a 91 Mb/s Digital Radio,\u201d \nIEEE Trans. Commun., Vol. COM-32, No. 1, January 1984, pp. 87-91.\n\n31. M. H. Meyers, \u201cMultipath Fading Outage Estimates Incorporating Path and Equip- \nment fe \u201d GLOBECOM 84, Atlanta, Ga., November 26-9, 1984, \npaper\n\n32. M. M. Sondhi, \u201cRandom Processes With Specified Spectral Density and First Order \nProbability Density,\u201d B.S.T.J., 62, No. 3 (March 1983), pp. 679-702.\n\n' 33. G. Vannucci and M. C. Teich, \u201cComputer Simulation of Superposed Coherent and \nCohaotic Radiation,\u201d Applied Optics, 19, No. 4 (February 15, 1980), pp. 548-53.\n\n34. W. D. Rummler, \u201cAdvances in Multipath Channel Modeling,\u201d Intl. Conf. Commun., \nSeattle, Wash., June 8-12, 1980, 52.3.1.\n\nHarold H. Hoffman, New York University; AT&T Bell Laboratories. Mr. \nHoffman, a member of the Radio Systems Research Department, has worked\n\non microwave radio relay systems, cordless telephone, satellite orientation, \nmobile radio, millimeter wave propagation, and a 7-meter radio astronomy \nreceiver and antenna. He holds patents for IF amplifier design and signal \nprocessing. He is presently concerned with digital radio fading simulators and \nmobile radio propagation studies.\n\nR. S. Roman, AT&T Bell Laboratories, 1970\u2014. Mr. Roman, a member of \nthe Radio Systems Research Department, has been concerned with digital \u2014 \nradio system studies. He is currently a student at Brookdale Community \nCollege.\n\nA. J. Rustako, Jr., B.S.E.E. 1965, M.S.E.E. 1969, New Jersey Institute of \nTechnology; AT&T Bell Laboratories, 1957\u2014. Mr. Rustako, a member of the \nRadio Systems Research Department, has been engaged in radio system and \npropagation studies. He contributed to early work in UHF and microwave \nmobile radio communications, in particular, propagation measurements of the \nmultipath medium, and the use of diversity techniques. He has also made \u00a9 \ncontributions in the field of satellite communications. These include earth- \nspace propagation measurements of rain attenuation and depolarization at 12 \nGHz, and the use of phased array techniques for rapid scanning spot beam \nsatellite antennas. He is presently concerned with high-speed digital radio \ntransmission techniques.\n\nClark B. Woodworth, B.S.E.E., 1977, Monmouth College; M.S.E.E., 1980, \nRutgers University; AT&T Bell Laboratories, 1977\u2014. Mr. Woodworth worked \nin the Satellite Systems Research Department and the Radio Systems Re- \nsearch Department. Work in the latter department included digital radio \nstudies and multipath channel simulation. He is currently working in the \nNetwork Systems Research Department. Member, Eta Paupe Nu, Sigma Pi \nSigma, IEEE.\n\nSingle-Frame Vowel Recognition Using Vector \nQuantization With Several Distance Measures\n\nOne of the most fundamental concepts used in the standard pattern recog- \nnition model for speech recognition is that of distance between pairs of frames \nof speech. Several distance measures have been proposed and studied in the \ncontext of an overall speech recognizer. The purpose of this investigation was \nto isolate the effects of different distance measures in a recognizer from the \nother types of processing typically used in recognition. The way in which this \nisolation was achieved was to use a recognizer based on single-frame distance \nscores, using a vector quantization approach to give the single-frame reference \npatterns required by the recognizer. The vocabulary for recognition was the \nset of continuant vowels extracted from carrier words. A speaker-dependent \nvowel recognition experiment was carried out using seven talkers (four male, \nthree female) and five distance measures. Results indicated that there were \ndifferences in performance for the different distance measures when the \nnumber of code-book patterns per vowel was one or two; however, when the \nnumber of code-book patterns was four or more, these differences in perform- \nance became insignificant.\n\nIn the past several years, interest has focused on defining and \nstudying distance measures for speech recognition that reflect mean- \ningful differences between pairs of speech spectra.\u2019~\u2019 Although several \ndifferent distance measures have been proposed,'~\u201c and they have been \nstudied in a variety of recognition systems,\u2019 as yet there is little\n\nconsistency in the reported performance of different recognizers using \ndifferent distance measures. For example, although Shikano and \nSugiyama\u2019 found consistent recognition performance improvements \nusing the weighted likelihood ratio distance measure (as opposed to \nan unweighted likelihood ratio distance measure) for a Japanese \nspeech recognition system, Nocerino et al.\u2019 were not able to match \nthese results in English alpha-digit recognition experiments. Similarly, \nalthough Davis and Mermelstein\u00ae achieved the best performance \namong several distance measures with a mel-based cepstral distance \nmeasure, this result has not been confirmed in other recognition tests.\n\nThere are several possible explanations for the discrepancies in \nresults obtained in the various investigations of distance measure \nperformance cited above. One explanation is that the basic feature \nmeasurements of each of the recognizers were different in all cases, \nfor example, filter bank analysis versus Linear Predictive Coding \n(LPC), different recording conditions and bandwidths, etc. These \ndifferences in recognizer front ends could account for the differences \nin performance, but if this were the case then the robustness of the \ndistance measure would become a major issue. A second explanation \nis the difference in vocabulary, talkers, and transmission conditions \n(e.g., telephone line versus microphone input). Again these differences \ncould be important, but they should not be factors for a robust distance \nmeasure. A third explanation is that the experimental results did not \njust reflect differences in distance measures but were affected by the \ninteraction between the components of the recognizer and the distance \nmeasure. Thus, for example, improved performance for a distance \nmeasure might be overshadowed by the power of dynamic time warp- \ning, which could compensate for a distance measure with poorer \nperformance.\n\nIt is the purpose of this paper to investigate he last possibility \ndiscussed above\u2014namely to isolate the effects of different distance \nmeasures from all the other temporal alignment processing used in \nrecognition. The way in which we accomplish this goal is to design a \nrecognizer that makes its decisions based on single frames of speech. \nIn this manner any real differences in distance measures wall manifest \nthemselves as differences in recognition scores. |\n\nThe implication of using single-frame distance scores for recognition \nis that the only vocabulary that can be considered is the set of \ncontinuant (steady) vowel sounds. We have considered ten such vowel \nsounds and they are listed in Table I, along with carrier words.in \nwhich the vowels occur. One side benefit of the experiments to be \nreported here is that a range of performance scores for single-frame \nrecognition of vowel sounds will be established and can be used to \nassess future recognition algorithms in much the same way as digit\n\nand alphabet scores have become standardized for isolated word rec- \nognition.\u00ae\n\nBased on the above discussion a series of speaker-dependent recog- \nnition tests was performed in the following manner. Each of seven \ntalkers (four male, three female) spoke the carrier words in Table I \nten times each, in two separate recording sessions, over a dialed-up \ntelephone line. Each talker also created, in a separate recording - \nsession, a single robust pattern for each of the ten carrier words. For \ndiagnostic purposes, an isolated word recognition test was performed \non the 100 isolated word tokens for each talker. All words which were \nmisrecognized were manually checked to make sure that no recording \nerrors (by either the talker or the automatic recording system) were \nmade. ,\n\nThe way in which the vowel frames, of each of the ten recordings of \neach carrier word, were selected was as follows. The energy contour of \nthe word was measured, and the vowel portion was defined as the set \nof frames whose log energies were contiguous to and within 6 dB of \nthe global energy peak of the word. The first five replications of each \ncarrier word were used as a training set, and a series of LPC Vector \nQuantization (VQ) code books were designed from the vowel frames \nfor each vowel and for each talker. The second five replications were \nused as an independent test set for recognition purposes.\n\nFive distance measures were used in the evaluations, namely the \nlikelihood ratio;\u2019\u201d the weighted likelihood ratio;* the cepstral distance;\u00b0 \na weighted cepstral distance;? and a bandpass liftered, weighted cep- \nstral distance.\u201d\n\nThe overall results of the single-frame recognition tests show that \nfor speaker-trained code books with moderate size\u2014that is, either four \nor eight vectors per vowel\u2014there were no significant differences in \nperformance for the five distance measures. For code books with one \nor two vectors per vowel, the two weighted cepstral distances per-\n\nformed best; the likelihood ratio was third; the (unweighted) cepstral \ndistance was fourth; the weighted likelihood ratio was last.\n\nThe outline of this paper is as follows. In Section II we discuss the \nspeech analysis system, show how we extracted the vowel frames from \neach carrier word, review the process of creating VQ code books, and \npresent the five distance measures used in our experiments. In Section \nIII we summarize the experimental conditions and present the results \nof the word recognition tests, the code-book design, and the single- \nframe recognition tests. In Section IV we discuss the results and give \ngeneral conclusions.\n\nA block diagram of the single-frame, VQ-based recognition system \nis given in Fig. 1. For each vowel frame, an LPC analysis is performed \nto give either an LPC vector or an LPC derived cepstral vector. We \ndenote the resulting vector as a. This vector is then passed to a series \nof ten vector quantizers (VQ\u2019s), one for each of the ten vowels, and \nthe minimum VQ distance, \u00a2\u2019, from the VQ for the ith vowel is \ncomputed as |\n\nwhere we assume that the ith vowel code book consists of the set of \nM vectors bi,, 1 < m <= M. The local distance measure of eq. (1) can \nbe any of five measures, namely the likelihood ratio; the weighted \nlikelihood ratio; the (unweighted) cepstral distance; the weighted \ncepstral distance; and the bandpass liftered, weighted cepstral dis- \ntance. The recognized vowel, i*, is chosen as the one whose VQ distance \n\u00e9 is minimum, that is, \ni* = argmin [e\u2019]. (2) \n1<1=10 \nIn the following sections we briefly review the LPC analysis condi- \ntions, the method of extraction of vowel frames from carrier words, \nthe procedure for VQ code-book formation, and the five distance \nmeasures used in this study.\n\nThe speech signal, s(n), was recorded off a dialed-up, local, telephone \nline. We used a sampling rate of 6.67 kHz. The speech signal is \ndigitized and then preemphasized using a first-order digital network \nwith transfer function H(z) = 1 \u2014 0.9527\". The signal is then blocked \ninto frames of size N = 300 samples (45 ms), with consecutive frames \nspaced by L samples (15 ms). A Hamming window is applied to each \nspeech frame and an eighth-order (p = 8) autocorrelation analysis is \nperformed. The zeroth-order autocorrelation is the energy for the \nframe, and it is used as the basis for word detection\u2122\u2019 and energy \nnormalization. An eighth-order LPC analysis is done on each frame, \nusing the autocorrelation method of LPC,\u201d to give the LPC vector for \nthat frame. If a cepstral representation is required, a simple transfor- \nmation of the LPC vector is performed.\u201d\n\nThe way in which the vowel frames were extracted from the isolated \nword tokens is illustrated in Fig. 2. Basically we used the log energy \ncontour of the word to find the vowel region\u2014which was arbitrarily \ndefined as the set of frames\u2014in the vicinity of the maximum energy \nvowel frame, such that the log energy of each frame was within Epi \n(dB) of the vowel maximum energy, Emax. After some preliminary \nexperimentation, a value of Ep = 6 dB was used. Thus, for a typical \ncarrier word, as illustrated in Fig. 2a, we first located the frame of \nmaximum energy, t,, and then, by searching in the local region around \nt,, found the beginning, tg, and ending, tz, frames of the vowel. \nAlthough this procedure worked well, in general, there were some \nspecific cases in which it failed. One such example is illustrated in Fig. \n2b, in which the carrier word had a stop release at the end (e.g., boot) \nwhose energy exceeded the maximum vowel energy. The simple strat-\n\nFig. 2\u2014Illustrations of how vowel frames were extracted from the carrier word.\n\negy of finding the frame with the maximum energy across the word \nwould fail in this case. Hence a check was made to ensure that all \nstrong local maxima of the energy contour were found, and that the \ncorrect vowel maximum was located.\n\nThe code-book training set for each vowel (and for each talker) \nconsisted of all the \u201cvowel frames\u201d that occurred in five occurrences \nof each carrier word. In fact, there were between 35 and 90 training \nvectors for each vowel. From these training vectors a series of VQ code \nbooks were designed with 1, 2, 4, and 8 vectors per vowel, using a \nstandard VQ code-book design algorithm.\u2019*\u2019* The distance measure \nused in the code-book design was the same one used in the single- \nframe recognizer\u2014that is, each of the five distance metrics was used. \nThe centroid of the vectors in each cluster was chosen to represent \nthe whole cluster. In our VQ design algorithm the centroid was chosen \nto minimize the average distortion of the whole cluster.\u2019\u00ae\n\nThe five distance measures used in the recognizer included the \nfollowing: |\n\n. Bandpass liftered, weighted cepstral distance\u2014dgpcrp(a, b). \n\u2018he form for computation of the likelihood ratio is\n\n(6) \noO \nwhere a is the residual error of the LPC analysis of the frame with \nautocorrelation R,. (1).\n\nwhere R,(i) and R,,(t) refer to the signal autocorrelations of the \nframes corresponding to vectors a and b, and c,(t) and c,(z) are the \ncorresponding LPC-derived cepstral vectors. It should be noted that \nwe use q > p terms, in the summation of eq. (7), to approximate the \ninfinite summation of the true weighted likelihood ratio distance. In \nparticular we have used q = 2p (16), where the \u201cextended\u201d autocorre- \nlations and cepstra were derived from the so-called \u201cmaximum en- \ntropy\u201d extension of the first (p + 1) terms.\u201d \nThe form for the (unweighted) cepstral distance is\n\nee we have again used the cepstrum extended to q = 2p terms. The \nform for the weighted cepstral distance\u2019 is\n\nand o? is the sample variance of the ith cepstral coefficient, where the \naveraging is over the individual vowel sounds, that is,\n\n10 \nY [o?].-n, \nof = ay | (11) \nym \nv=1 \nwith [o?], being the variance of c(i) over the n, frames in the training \nset for vowel v. Typically the weighting function w; increases mono- \ntonically with the index 1. \nFinally, the form for the bandpass liftered, weighted cepstral \ndistance\u201d\u2019 is\n\nwhere g was set to 12, and w} had the form of a bandpass lifter, that \nis, a raised sinewave of the form\n\nA series of recognition tests was run in which each of seven talkers \n(four male, three female) first created robust training tokens of each \ncarrier word!\u00ae and then, in separate recording sessions, spoke each \ncarrier word ten times each. The first five such recordings were used \nas a training set for the VQ code books; the second five recordings \nwere used as an independent test set. The robust tokens were used in \nan isolated word recognition test to check the validity of the recorded \ncarrier words. The results of the isolated word recognition test are \ngiven in Table II. It can be seen that for three of the talkers (1, 4, and\n\n_ Table II\u2014Word recognition errors for carrier words for each talker \n(100 recognition trials per talker)\n\n6) there were no word errors; for talkers 2 and 7 there were 2 word \nerrors (out of 100 trials each); for talkers 3 and 5 there were 10 and 8 \nword errors. The overall isolated word recognition accuracy for the \nseven talkers is 96.9 percent. The word \u201cbit,\u201d which accounted for 8 \nof the 22 recognition errors, was confused with the word \u201cbet\u201d in all \nsuch cases.\n\nThe results given in Table II indicate that there is a lot of variability \nin the recognition performance on the isolated words across both \ntalkers and vocabulary words.\n\nThe results of the single-frame vowel recognition tests are given in \nTable III and are shown plotted in Fig. 3. The data in Table III are \nthe average vowel error rates in percent averaged over the ten vowels \nand the seven talkers as a function of VQ code-book size and distance \nmeasure for both the training and testing sets, that is, there were \nabout 4000 recognitions per set. Figure 3 shows these same data in \ngraphical form. Several observations can be made from these results, \nincluding the following:\n\n1. There are significant degradations in performance, for all dis- \ntance measures and for all code-book sizes, between the training and \ntesting sets of data. Thus for the VQ code-book size of one we see \ndegradations of 3 to 4 percent, whereas for the VQ code-book size of \neight we see degradations of from 9 to 10 percent in averaged vowel \nerror rates.\n\n2. The effects of different distance measures can be seen primarily \nfor code-book sizes of one and two vectors per vowel, in which case \nthe two weighted cepstral distances consistently outperformed the \nother three metrics, and the weighted likelihood ratio consistently \n- performed the worst of the five measures. For code-book sizes of four \nand eight vectors per vowel, there were no significant performance \ndifferences among the five distance measures.\n\nTable II|\u2014Average word error rate (%) as a function of VQ code- \nbook size and distance measure for both the training and testing sets \nResults on Training Set Results on Testing Set \n; VQ Code-Book Size VQ Code-Book Size \nTUSCAN se ae \nMeasure 1 2 4 8 1 2 4 \ndir 176 122 7.0 3.4 216 169 141 \u00b0&129 \ndwir 188 13.6 12 3.8 224 189 142 125 \ndcrp 18.6 11.8 (e' 3.0 21.6 17.4 13.7 13.4 \ndwcrp 16.5 11.0 6.7 3.8 20.0 16.5 14.2 13.3 \ndsecer 16.7 105 6.4 3.2 194 155 134 124\n\nFig. 3\u2014Average vowel error rate (%) versus code-book size for each of the four \ndistance measures and for the testing and training sets of data.\n\nrate of about 20 to 22 percent for a single code-book vector per vowel, \nand the error rate dropped to about 13 percent for eight code-book \nvectors per vowel. Thus we conclude that vowel recognition (among \nthe ten vowels in Table I) cannot be performed reliably using any of \nthe distance measures we have considered, in the framework of a \nsingle-frame VQ code book-based recognizer.\n\nThe results presented in Section III can be interpreted as follows. \nIn the case where we have a good representation of the patterns to be \nrecognized, the effects of different distance measures on. recognition \nperformance are small. Such was the case when we used four or eight \ncode-book vectors to characterize each vowel in the vocabulary. How- \never, when the representation of the patterns to be recognized becomes \nmore coarse, then the effects of different distance measures start to \nbecome important. In these cases a better characterization of speech \nsound differences, as obtained from a good distance measure, should \ngive better recognition scores. Such was the case when we used one or \ntwo code-book vectors to characterize each vowel.\n\nresults presented in Table III. We see a big difference in average vowel \nerror rates between comparable test conditions (distance measure, VQ \ncode-book size) for the training and testing sets, especially when we \nhave four or eight vectors per vowel code book. Thus, in a sense, the \neffects of different distance measures are small when the code-book \nvectors begin to characterize well the seemingly insignificant details \nof the training set, and are larger when the code-book vectors char- \nacterize mainly the gross spectral behavior of the vowels. For real- \nworld recognition systems it is most probably the latter case that is \nthe more important one in that the reference patterns would be \nexpected to characterize the gross behavior of spectral variations with \ntime. In general there is not enough training data to reach the point \nwhere we have characterized the fine spectral variations of words \nreliably.\n\nThe conclusion we reach from the above discussion is that the \nresults for small code-book sizes, in which there were significant effects \nof different distance measures, are probably more representative of \nreal recognition systems than the results for large code-book sizes. In \nthese cases\u2014as is evidenced by recent investigations by Tokhura,\u2019 \nJuang et al.,\u2019\u00b0 and Nocerino et al.,\u2019/\u2014the weighted cepstral distances \nand the likelihood ratio would be expected to give better recognition \nperformance than the unweighted cepstral distance or the weighted \nlikelihood ratio measures.\n\nWe have presented results on speaker-dependent, single-frame, VQ- \nbased, vowel recognition for five different distance measures and for \nfour different size VQ code books. Our results indicate that for small \ncode-book sizes (one or two vectors per vowel) there is improved \nrecognition performance using a weighted cepstral distance rather \nthan the likelihood ratio, the unweighted cepstral distance, or the \nweighted likelihood ratio measures. For larger code-book sizes (four or \neight vectors per vowel) the performance differences among the five \ndistance measures decrease. For practical recognizers, the weighted\n\ncepstral distances appear to have advantages for application to \n_ gpeaker-independent systems and for large vocabulary recognizers. \nThese advantages include increased efficiency of representation, re- \nduced complexity of computation, and improved performance.\n\n1. F. Itakura and S. Saito, \u201cAnalysis-Synthesis Telephony Based on the Maximum \nLikelihood Method,\u201d Proc. Int. Cong. Acoustics, Tokyo, Japan, Paper C5-6, 1968.\n\n2. F. Itakura, \u201cMinimum Prediction Residual Principle Applied to Speech Recogni- \ntion,\u201d IEEE Trans. Acoust., Speech, Signal Process., ASSP-23, No. 1, Sear \n1975), pp. 67-72.\n\n. K. Shikano and M. Sugiyama, \u201cEvaluation of LPC Spectral Matching Measures for \nSpoken Word Recognition,\u201d Trans. IECE, J65-D, No. 5 (May 1982), pp. 535-41.\n\n. D.H. Klatt, \u201cPrediction of Perceived Phonetic Distance From Critical Band Spectra: \nA First Step,\u201d Proc. ICASSP \u201982, (May 1982), pp. 1278-81.\n\n5. A. H. Gray Jr. and J. D. Markel, \u201cDistance Measures for Speech Processing,\u201d IEEE \nHina Acoust., Speech, Signal Process., ASSP-24, No. 5, (October 1976), pp. \n380-91.\n\n6. S. Davis and P. Mermelstein, \u201cComparison of Parametric Representations for \nMonosyllabic Word Recognition in Continuously Spoken Sentences,\u201d IEEE \nTrans. Acoust., Speech, Signal Process., ASSP-28, No. 4, (August 1980), pp. 357- \n66.\n\n. N. Nocerino et al., \u201cComparative Study of Several Distortion Measures for Speech \nRecognition,\u201d Speech Commun., 4, No. 4 (November 1985).\n\n8. G. R. Doddington and T. B. Schalk, \u201cSpeech Recognition-Turning Theory to \nPractice,\u201d IEEE Spectrum, 18 (September 1981), pp. 26-32.\n\n. Y. Tokhura, \u201cSpeaker Independent Recognition of Isolated Digits Using a Weighted \na a Distance,\u201d J. Acoust. Soc. Am., Suppl. 1, 77, Paper \u00a313 (Spring 1985), \np. S11.\n\n11. L. F. Lamel et al., \u201cAn Improved Endpoint Detector for Isolated Word Recognition,\u201d \nIEEE Trans. Acoust., Speech, Signal Process., ASSP-29, No. 4 (August 1981), \npp. 777-85. .\n\n13. Y. Linde, A. Buzo, and R. M. Gray, \u201cAn Algorithm for Vector Quantization,\u201d IEEE \nTrans. Comm., COM-28, No. 1 (January 1980), pp. 84-95.\n\n14. B. H. Juang, D. Wang, and A. H. Gray, Jr., \u201cDistortion Performance of Vector \nQuantization for LPC Voice Coding,\u201d IEEE Trans. Acoust., Speech, Signal \nProcess., ASSP-30, No. 2 (April 1982), pp. 294-303.\n\nLawrence R. Rabiner, S.B. and S.M., 1964, Ph.D., 1967 (Electrical Engi- \nneering), The Massachusetts Institute of Technology; AT&T Bell Laborato- \nries, 1962\u2014. From 1962 through 1964 Mr. Rabiner participated in the coop- \nerative plan in electrical engineering at AT&T Bell Laboratories, in Whippany \nand Murray Hill, New Jersey. He worked on digital circuitry, military com- \nmunications problems, and problems in binaural hearing. Presently, Mr. \nRabiner is engaged in research on speech communications and digital signal \nprocessing techniques. He is coauthor of Theory and Application of Digital \nSignal Processing (Prentice-Hall, 1975), Digital Processing of Speech Signals \n(Prentice Hall, 1978), and Multirate Digital Signal Processing (Prentice-Hall, \n1983), Member, National Academy of Engineering, Eta Kappa Nu, Sigma Xi, \nTau Beta Pi, Fellow, Acoustical Society of America, IEEE.\n\nFrank K. Soong, B.S., 1973, National Taiwan University, M.S., 1977, \nUniversity of Rhode Island, Ph.D., 1983, Stanford University, all in Electrical \nEngineering; AT&T Bell Laboratories, 1982\u2014. From 1972 to 1975 Mr. Soong \nserved as a teacher at the Chinese Naval Engineering School at Tsoying, \nTaiwan. In 1982 he joined the technical staff at AT&T Bell Laboratories, \nwhere he engaged in research in speech, coding, and speaker recognition. \nMember, IEEE.\n\nTraffic Capabilities of Two Rearrangeably \nNonblocking Photonic Switching Modules\n\nThe architectures of two small switching networks are compared as potential \nimplementations of a 4 X 4 photonic switching module. Such a module would \nbe made by interconnecting several 2 X 2 photonic directional couplers on a \nsingle LiNbO3; substrate. While both networks are rearrangeably nonblocking, \nwe investigate whether one network requires significantly more rearrange- \nments than the other. The analysis includes transient, Monte Carlo simulation, \nand Markov steady-state techniques. We conclude that the traffic capabilities \nof the two structures are not significantly different, and that selection of an \narchitecture can be based on other criteria, like loss, crosstalk, or ease of \nmanufacture.\n\nThe percentage of the world\u2019s voice and data communications that \nis carried by optical fibers increases daily. The importance of research \nin photonic switching increases with it. A promising element for the \nimplementation of photonic switching systems is the photonic direc- \ntional coupler, made from titanium-diffused lithium niobate. The \ncurrent state of this technology allows a level of integration of tens of \ndevices on a single substrate. We investigate two competing architec- \ntures for a 4 X 4 switching module, fabricated in this technology, that \ncould be useful in building photonic systems.\n\nof switching networks and describe the two candidate photonic switch- \ning modules. In Sections IV and V, we present the results of transient \nanalysis and Monte Carlo simulation, respectively, as applied to the \ntwo switching modules. In Sections VI through VIII, we present steady- \nstate Markov analysis as applied to a generalized module and to each \ncandidate switching module, respectively. In Section [X, we enumerate \nthe network configurations in each of the candidate switching modules.\n\nThe traffic-handling performance of a switching network depends \non two entities: |\n\ne The topology of the internal elementary switching stages. \ne The rule by which paths through the network are selected.\n\nOne topological classification applied to switching networks is the \nhierarchy of \u201cblockingness.\u201d Networks are classified by their ability \nto establish connections, especially sequences of connections with \nintermediate disconnections. An excellent tutorial and summary of \nthe state-of-the-art in switching network topologies is found in Ref. 1.\n\nA switching network is nonblocking if any desired connection be- \ntween two idle ports can be completed immediately without interfer- \nence from connections that may be already established in the network. \nIf this property is independent of any switching rule used to select \npaths through the network, then the topology is nonblocking in the \nstrict sense. If the property is true, provided the paths assigned to the \nestablished connections were selected under some switching rule, then \nthe topology is nonblocking in the wide sense.\n\nA switching network is rearrangeably nonblocking if any desired \nconnection between two idle ports, which might be temporarily blocked \nby connections already established in the network, can be completed \npossibly after some established connections are moved to different \npaths. Such topologies are parallel nonblocking in that any set of I/O \npairs can be connected without blocking if the network is initially idle \nand the set is known in advance.\n\nA switching network is blocking if there exist network configurations \nof established connections from which some connection between two \nidle ports can not be completed, even with rearrangement of the \nestablished connections. Both switching modules under consideration \nin this paper are rearrangeably nonblocking. \u00a9\n\nA switching network generally allows more than one path by which \nports on either end of the network may be connected. The switching \nrule is the means of selecting one path. For certain special cases, such \nas the strict-sense nonblocking networks, the performance is inde- \npendent of the switching rule. In general, however, the switching rule \nis an important factor in the performance of a switching network.\n\nThe optimal switching rule follows: Route a new connection through \nthe network in a way that least affects the routing of any future \nconnections. |\n\n.. route a call through the most heavily loaded part of the network that will still \ntake the call.\n\nSuch switching rules are called packing rules because of their analogy \nto similar rules used in the problem of packing spheres into boxes. For \none module under consideration here, the four switching rules above \nappear equivalent, and the rule is called prudent in the paper. For the \nother module under consideration here, a case will be demonstrated \nthat represents a counter example to the general adoption of switching \nrules that recommend close packing as the primary consideration.\n\nThree measures of the quality of switching networks are transmis- \nsion, topology, and traffic cepanuy:\n\nTwo measures of transmission quality are insertion loss and cross- \ntalk. Either measure may be in terms of average or worse-case value. \nUnder a uniform wiring scheme, insertion loss and crosstalk worsen\n\nas a network becomes deeper. In electrical implementations, crosstalk \nusually worsens as interchip, interboard, and especially interframe \nwiring increase. This should be much less noticeable in photonic \nsystems. Loss and crosstalk measures will not necessarily recommend \nthe same switching networks.\n\nTopological complexity is not so much a performance issue as it is \na manufacturing and economic issue. Quantitative measures that \ncorrelate with manufacturing cost are not known, nor is the impor- \ntance of this issue relative to transmission or traffic measures. It is an \nimportant open question.\n\nTraffic capability is a well-known analytical measure applied to \nblocking networks.? The probability of blocking is that weighted \nproportion of (new connection, network configurations) in which the \ngiven new connection cannot be completed through the network in \nthe given configuration.\n\nA corresponding measure for rearrangeably nonblocking networks \nis developed here. The probability of requiring rearrangement P,; is \nthat weighted proportion of (new connection, network configurations) \nin which the given new connection can be completed through the \nnetwork in the given configuration, but only after some established \nconnections in the network are first rearranged.\n\nA dichotomy appears. If the malevolence is the CPU real time used \nin effecting the rearrangement, then the demerit is that any rearrange- \nment is required, regardless of the count of switches or connections \ninvolved, because they are probably all rearranged in parallel. If the \nmalevolence is the count of established connections disturbed by the \nrearrangement, then the demerit may not be binary and one choice of \nrearrangement may cost less than another. We will show that the \ncount of established connections disturbed by a rearrangement is \u00a9 \ndifferent for the two modules under consideration, so the dichotomy \nis relevant. We will derive expressions for how many established \nconnections must be rearranged, on the average, for each of the \ncompeting architectures.\n\nIn transient analysis, a sequence of connections and disconnections \nis applied to an idle module. Over a set of such sequences, P,, is the \nproportion of those sequences in the set in which a rearrangement was \nrequired to complete a connection. The inadequacy of this analysis is\n\nthat it does not produce a single number or expression, because many \nsets of sequences must be investigated. But the advantages over the \nclassical analyses are that the results are not dependent on potentially \nunrealistic assumptions about traffic statistics, and that this analysis \nentails a level of detail (and corresponding tedium) not required in the \nother analyses. This detail uncovered an optimal switching rule for \none module that probably would have been overlooked had the analysis \nbeen confined to traditional steady-state techniques.\n\nIn Monte Carlo simulation, a sequence of randomly generated events \nis applied to an initially idle module. The inadequacies of this analysis \nare that repeated simulations with identical event statistics yield \ndifferent results and that no closed-form solution is obtained, giving \nP,, as a function of those statistics. The benefit of this analysis is that \nboth transient and steady-state behavior may be observed, and in fine \ndetail. If properly recorded, the events leading up to anomalies may \nbe studied.\n\nIn steady-state Markov analysis, the module is assumed to be in \nsome random network configuration and then a new random connec- \ntion or disconnection occurs. P,, is a weighted sum over all network \nconfigurations, and all possible new connections from those network \nconfigurations, of those cases in which the module requires rearrange- \nment to complete the given connection from the given configuration. \nThe weighting is the steady-state probability distribution over the set \nof network configurations or states of equivalence classes. Both the \nsteady-state probability distribution and the interstate transition \nprobabilities are dependent on the traffic load. This analysis yields a \nclosed-form result, but it is only valid in the steady state and it is \ndependent on potentially unrealistic statistical assumptions.\n\nWe review the technology of photonic switching and present the \ntopologies of two proposed implementations of a 4 X 4 switching \nmodule.\n\nA photonic directional coupler is a two-input two-output device \nwhose transmission state depends on the magnitude of an applied \nexternal voltage.? (See Fig. 1.) With nominal 0-volt applied, the \ntransmission is such that the signals cross over from input to output \n(called the crossed state) and with a positive voltage applied, the \ntransmission is straight through from input to output (called the bar \nstate). This device is a photonic realization of the generic 2 X 2 beta \nswitching element,* where the control signal is electronic and the \nswitched data is photonic. |\n\nUPPER UPPER UPPER \nINPUT MIDDLE OUTPUT \nSWITCH \u2014- SWITCH \nLOWER oa), LOWER \nINPUT MIDDLE OUTPUT \nSWITCH SWITCH SWITCH\n\nSwitching speeds in the tens of picoseconds have been reported,\u00b0 \nbut speeds in the nanoseconds are more common.\u00ae The attainable \nswitching speed is highly dependent on device packaging, on the \nquality of the electronic driver circuit that switches the applied voltage \nto the device, and on the magnitude of this required voltage. A 1:16 \nmultiplexer/demultiplexer has been built as an integrated circuit mod- \nule.\u2019 It is a simple tree topology of photonic directional couplers with \none unused port at the input to each photonic directional coupler. \nAnother integrated circuit is a 4 X 4 switching module built as a square \narray.\u201d\u201d These circuits show the level of integration available today in \nthis technology. Because each photonic directional coupler has length \nin the order of a centimeter, the expectation of increasing the level of \nintegration by several orders of magnitude is not high, unless some \nmajor breakthrough changes the photonic coupling length by orders \nof magnitude.\n\nTwo waveguides, made by diffusing titanium in a lithium-niobate \nsubstrate, can be made to intersect or cross over. Crosstalk increases \nas the angle of intersection decreases, and becomes unpredictable at \nvery small angles.\u2019\u00b0 Thus a configuration, like that of Fig. 2, is feasible \nif the crossovers are truly at large angles, as the figure suggests. \nHowever, the photonic directional couplers have lengths in the order \nof a centimeter and widths measured in microns, so Fig. 2 is not drawn \nto true scale. Furthermore, the loss in a waveguide increases rapidly \nas its radius of curvature decreases. So, a configuration like that of\n\nFig. 2 would require a physically large chip to allow geometries with \nhigh crossover angles and high radii of curvature.\n\nA major technological issue with today\u2019s photonic directional cou- \nplers is that they have significant insertion loss. While most of this \nloss is due to coupling between fiber and chip and would be alleviated \nwith integrated circuits, the device loss is still high enough that \nmultidevice systems, like large switching networks, will require signal \namplification. Photonic gain can be simulated, expensively and with \nimposing a bit-rate constraint, by a circuit with a photodetector, an \nelectronic amplifier, and a laser. However, direct photonic amplifica- \ntion is believed to be coming available soon.\u2019\u2019 The devices described \nin the reference are fabricated from gallium arsenide, while the pho- \ntonic directional couplers are fabricated from lithium niobate. Thus, \nintegration on the same chip is currently impossible. However, when \nsuch devices become practical, they could be integrated onto the same \nchip carrier as the chip containing the photonic directional couplers. \nIf this form of integration is truly practical in the future, it allows \nsmall radius bends and alleviates many of the problems described in \nthe preceding paragraph.\n\nWe call one candidate 4 X 4 module the 222 Module because it is a \nthree-stage module, where each stage has two switches. (See Fig. 2.) \nThis topology is in the class of networks called Clos networks, and this \n~ exact configuration is a frequently used example in the literature. The \nfollowing terminology is used to identify symbolic inputs and outputs \nin the 222 Module:\n\ne A and W represent an arbitrary input and output, respectively, each \nshown arbitrarily as an upper port on an upper switch.\n\ne B represents the other port on the same input switch as A and X \nrepresents the other port on the same output switch as W.\n\ne C represents either port on the input switch that A and B do not \nshare, and Y represents either port on the output switch that W and \nX do not share, each shown arbitrarily as an upper port on a lower \nswitch.\n\ne D represents the other port on the same input switch as C, and Z \nrepresents the other port on the same output switch as Y.\n\nBecause of the symmetry of the 222 Module, there is no topological \nrelationship between A and W.\n\nWe call the other candidate 4 X 4 module the 2121 Module because \nthe topology has four stages, where the input and middle stages have \ntwo switches and the central and output stages have one switch. (See\n\nLOWER LOWER \nSWITCH SWITCH \nFig. 3\u2014Configuration and terminology of the 2121 Module.\n\nterms middle and central when applied to switching stages. The follow- \ning terminology is used to identify symbolic inputs and outputs in the\n\ne A represents an arbitrary input and B represents the other port on \nthe same input switch as A, both shown arbitrarily on the upper \ninput switch with A above B.\n\ne W represents the only output that A or B can reach through two \nstages (necessarily on the same horizontal level as A and B).\n\ne C represents either port on the input switch that A and B do not \nshare, and D represents the other port on the same input switch as \nC, both shown arbitrarily on the lower input switch with C above \nD.\n\ne Z represents the only output that C or D can reach through two \nstages (necessarily on the same horizontal level as C and D).\n\ne X represents either port on the only switch with two outputs on it, \nand Y represents the other port on the output switch with X, shown \narbitrarily with X above Y.\n\nBecause the 2121 Module is not symmetric, there is a topological \nrelationship between the inputs and the outputs. Accordingly, W or Z \nis on the upper or lower middle switches depending on whether A and \nB or C and D are on the upper or lower input switches, respectively.\n\nAn interesting and nonintuitive switching rule was discovered for \nthe 2121 Module. Consider the following sequence of events applied \nto an idle 2121 Module. Let A to W be the first connection, and let \nthe path that avoids the central switch be selected (intuitively, the \nbest choice), setting the states of A\u2019s input switch appropriately, and \nW\u2019s middle switch to the bar state.\n\nIf the second connection is B to X, representing two of the nine \nsecond connections, the intuitively preferred path is not available \nbecause the A to W connection has already set the switches, and is \nusing that junctor that B can reach that skips the central switch. Two\n\nother paths are available and the choice is extremely interesting. One \npath shares the state of A\u2019s input switch and W\u2019s middle switch, and \nassigns the central switch to the bar state and the output switch, \nappropriately. This path requires the assignment of two switches in \naddition to the ones already assigned, and it is intuitively appealing. \nAfter A to W is disconnected, A to Y, A to Z, and C to W, representing \nfour of the nine third connections, require rearrangement, and the \nother five do not.\n\nThe other path for B to X still shares the state of A\u2019s input switch, \nbut it assigns the central switch to the crossed state, the unused middle \nswitch to the bar state, and the output switch appropriately. This path \nrequires the assignment of three switches in addition to the ones \nalready assigned, and is, therefore, intuitively less appealing than the \nprevious path. However, after A to W is disconnected, only A to Z, \nrequiring some of the junctors and switch connections used by B to \nX, requires a rearrangement of the network configuration.\n\nI have two general interpretations that cover the essence of this \nswitching rule: (1) If you must use the central switch, the crossed state \nis preferred, and (2) Minimize the count of switches that are shared \nwith any other path. It is not clear yet whether either or both of these \nstatements is applicable to a generalization of the 2121 Module to an \nn Xn architecture. Even in the 2121 Module, I was afraid that the \noverall use of this switching rule, while improving the performance \nwith some sequences of events, would degrade the performance with \nother sequences. In all the sequences investigated and all the simula- \ntions that were run, no such case arose.\n\nWe give an example of a transient sequence of events applied to the \n- 2121 Module in which the final connection in the sequence requires \nthe disturbance of two established connections. We then argue that \nthe worst-case number could not be greater than two.\n\nThe example begins, with an idle module, by connecting A to X. \nThe optimal path avoids the central switch by assigning W\u2019s middle \nswitch to the crossed state and A\u2019s input switch and the output switch, \nappropriately. Let the second connection be from B to Y. Since two \nof the paths are blocked, the remaining path must be used, by assigning \nthe central switch to the crossed state and Z\u2019s middle switch to the \nbar state. Now disconnect the first connection, the one from A to X, \nfreeing the state of W\u2019s middle switch.\n\nLet a new second connection be from C to X. Since the output \nswitch and Z\u2019s middle switch are in the wrong states to use two of the \n_ paths, the remaining path must be used, by assigning W\u2019s middle \nswitch to the bar state, and using the idle link through the central\n\nswitch. The states of all six switches are set, and two of four new \nconnections will require rearrangement.\n\nConsider a third connection from A to Z. Its only satin requires the \nreconfiguration of the input switch shared by A and B and Z\u2019s middle \nswitch, and also requires the link through the central switch now used \nby B to Y. So the B-to-Y connection must be moved to its other \n(optimal) path. But it can\u2019t be moved directly because W\u2019s middle \nswitch and the output switch are in the wrong state. So, the C-to-X \nconnection must also be moved to its other (optimal) path.\n\nHaving established, by example, that the worst-case number is at \nleast two, the question remains whether it could be higher. Since there \ncan only be four established connections, the only other number to \nconsider is three. However, in any network configuration with three \nestablished connections, the unused fourth connection is always avail- \nable. While this fact may not be immediately obvious, it can be proven \nexhaustively for the 2121 Module. A more elegant proof, applicable to \nthe general n X n module, is desirable.\n\nThe generalized Clos Network has the topology of Fig. 2, but with \nn inputs and m junctors on each of r rectangular input switches, with \nn outputs and m junctors on each of r rectangular output switches, \nand with m square middle switches each connecting to r junctors \non each side. It is known\u2019 that such a switch is nonblocking if m = \n2n \u2014 1 and is rearrangeably nonblocking if m = n. The latter is \nsatisfied in the 222 Module, in which n = m =r= 2.\n\nThe 2121 Module is also rearrangeably nonblocking. This result can \nbe proven easily by exhaustion for the 2121 Module and is obvious by \ninspection of the state diagram in Section VIII. A general prow fora \ngeneralized n X n network is being developed.\u201d |\n\nIn the 222 Module each I/O pair has exactly two connection paths. \nThat is, there are exactly two paths through the module from any \ninput to any output. The 2121 Module does not have this symmetry. \nThe I/O pairs at opposite corners of the module (A to Z, B to Z, C to \nW, and D to W) have only one path through the module, requiring the \ncentral and the appropriate middle switches in the crossed state. The \nI/O pairs straight across the module (A to W, B to W, C to Z, and D \nto Z) have two paths through the module: one avoiding the central \nswitch and the other using the central switch in the bar state. Any \nconnection to an output on the only output switch (A to X, B to X, A \nto Y, B to Y, C to X, D to X, C to Y, and D to Y), has three paths\n\nthrough the module: one avoiding the central switch, the second using \nthe central switch in the bar state, and the third using the central \nswitch in the crossed state.\n\nIn the 222 Module each path through the module passes through \nexactly three photonic directional couplers. The 2121 Module does not \nhave this symmetry, either. In the 2121 Module some paths pass \nthrough only two photonic directional couplers, some through three, \nand some through four. This variation will make deterministic gain \ndifficult. If crossovers in the 222 Module require photonic directional \ncouplers, or have similar loss and crosstalk as photonic directional \ncouplers, then this module would not be symmetric in this sense either.\n\nIf crossovers in the 222 Module are insignificant, its inputs and \noutputs are uniform. That is, all inputs and outputs have the same \nproperties of the following: count of paths per I/O pair, count of \nswitches per path, and equal access to any port on the other side of \nthe module. The last two properties do not hold true if crossovers are \nsignificant in the 222 Module, but none hold true in the 2121 Module.\n\nThe interconnection topology of the 222 Module has two crossovers: \none between the input and central stages and one between the central \nand output stages. The interconnection topology of the 2121 Module \nhas no crossovers.\n\nAn implementation is illustrated in Fig. 4 in which the two cross- \novers in the 222 Module are eliminated by off-chip fibers. Its practi- \ncality depends on the difficulty of off-chip fibering, the impracticality \nof crossovers on the photonic substrate, and the magnitude of the \nadvantage of the 222 Module over the 2121 Module (if any).\n\n3.6.4 Transmission 7 \nIt is difficult to predict the difference in the crosstalk characteristics\n\nof the two modules. The crossovers in the 222 Module suggest that it \nmay be worse than the 2121 Module, but the extra stage in the latter \nsuggests otherwise.\n\nInsertion loss is easier to predict. The worse-case count of switches \nin a network path is four in the 2121 Module and is three in the 222 \nModule. Thus, the 2121 Module would appear to have a poorer \ninsertion loss characteristic. However, if crossovers must be imple- \nmented by photonic directional couplers, or if they have equivalent \nloss, then the worse-case count of (equivalent) switches in a network \npath is five in the 222 Module and it would have worse insertion loss \nthan the 2121 Module.\n\nThe switching rule used in the 222 Module is the classical packing \nrule discussed in Section 2.2. The switching rule used in the 2121 \nModule is the unusual rule discussed in Section 3.4. Two implemen- \ntations of the switching rule are discussed in Section IX and the \nnetwork configurations for each module are enumerated.\n\nIf some network configuration of the 222 Module must be rearranged \nbefore a new connection can be completed, only one established \nconnection need ever be disturbed.\u201d It was shown in Section 3.5 that \nthere exist network configurations in the 2121 Module in which two \nestablished connections may need to be disturbed. Consequently, we \nnot only compute P,, for each module in the sections below, but we \nalso compute, N,,, how many established connections must be dis- \nturbed, on the average.\n\nBy contrast, the 2121 Module has been counter-intuitive. While its \ntransient behavior is smooth, it degrades with simple connect/discon- \nnect sequences. The switching rule and count of established connec- \ntions disturbed has been startling to those familiar with such things. \nThe plethora of network configurations is partitioned, with great \ndifficulty, into 50 equivalence classes, making Markov analysis tedious.\n\nIV. TRANSIENT ANALYSIS \nIn transient analysis, we assume the module is idle and study its\n\nbehavior under different sets of connect/disconnect sequences. The \nnetwork quality, the average probability of requiring rearrangement \nfor a specific set of sequences, is the proportion of those sequences in \nthe set that require rearrangement. Four sets are applied to each of \nthe two modules, giving eight results. Computer programs applied all \nsequences in the four sets to each module and tabulated whether \nrearrangement was required. The program\u2019s output was studied and \ngeneralized into theorems whose formal proofs are exhaustive and \ntedious and not presented in this paper.\u2019\u00ae The proofs are available to \nthe interested reader.\n\nA template is a set of fixed-length sequences of events applied to the \nswitching module, where an event is a connection or a disconnection \nof an input-output pair. The nomenclature for templates is x, ... Xn, \nwhere x; represents the ith event in a sequence; x; has value C or D, \ndepending on whether this ith event is a connection or disconnection, \nrespectively; and n is the length of a sequence.\n\nThe following rules govern the determination of significant tem- \nplates: \noe Since the module is assumed to be idle before any sequence is\n\ne Similarly, no sequence, nor initial subsequence, would have more Ds \nthan Cs. |\n\n\u00a9 Since a module can only support four connections, no sequence, nor \ninterior subsequence, would have four more Cs than Ds.\n\no No significant sequences would have, nor would contain an initial \nsubsequence that has, as many Cs as Ds (e.g., no significant sequence \nwould begin with CD) because it would have the effect of restarting \nfrom an idle module.\n\ne Since we expect no problems Hononne disconnects in these modules, \nno significant sequence would end with a D.\n\noe In no significant sequence would a D apply to an immediately \npreceding C because it would have the effect that the C never \noccurred. \u2014\n\nhaustively. Fortunately, enough templates with lengths 4 and 5 are \nsignificant. Combining these length and content constraints, the sig- \nnificant templates are CCCC, CCDC, CCDCC, and CCCDC.\n\nFor sequences in a CCCC template, beginning with all switches idle, \nthe assumed order of events is\n\n4, The last idle input connects to the last idle output. \nWith 4 x 4 = 16 cases of first connection, 3 X 3 = 9 cases of second \nconnection, 2 X 2 = 4 cases of third connection, and 1 X 1 = 1 case of \nfinal connection, the CCCC template contains 16 x 9 X 4 xX 1 = 576 \nsequences.\n\nFor sequences in a CCDC or CCDCC template, beginning with all \nswitches idle, the assumed order of events is \n. One of four inputs connects to one of four outputs. \n. One of three idle inputs connects to one of three idle outputs. \n. The first input-output pair is disconnected. \n. One of three idle inputs connects to one of three idle outputs. \n. In a CCDCC sequence only, one of two idle inputs connects to \none of two idle outputs. \nWith 4 X 4 = 16 cases of first connection, 3 X 3 = 9 cases of second \nconnection, 1 case of disconnection, 3 X 3 = 9 cases of third connection, \nand, in the CCDCC sequences only, 2 X 2 = 4 cases of fourth \nconnection, the CCDC template contains 16 x 9 x 1 X 9 = 1296 \nsequences and the CCDCC template contains 16 X9xX1xX9x4= \n5184 sequences.\n\nFor sequences in a CCCDC template, beginning with all switches \nidle, the assumed order of events is\n\n5. One of two idle inputs connects to one ot two idle outputs. \nWith 4 X 4 = 16 cases of first connection, 3 X 3 = 9 cases of second \nconnection, 2 X 2 = 4 cases of third connection, 2 cases of disconnec- \ntion (either the first or second connection), and 2 X 2 = 4 cases of \nfinal connection. The CCCDC template contains 16 x 9 xX 4 X 2 X \n4 = 4608 sequences.\n\nIn either module, with a prudent switching rule, no CCCC sequences \nrequire rearrangement. Since all combinations of idle I/O pairs can be \nsimultaneously interconnected in both modules, they are both at least \nrearrangeably nonblocking.\n\nThe 222 Module outperforms the 2121 Module under CCDC and \nCCDCC sequences. In the 222 Module, with a prudent switching rule, \nno CCDC sequences, nor CCDCC sequences, require rearrangement. \nHowever, in the 2121 Module with its unusual switching rule, 2 percent \nof the CCDC sequences require rearrangement and 6 percent of the \nCCDCC sequences require rearrangement.\n\nConversely, the 2121 Module outperforms the 222 Module under \nCCCDC sequences. In the 222 Module, with a prudent switching rule, \n8 percent of all CCCDC sequences require rearrangement. However, \nin the 2121 Module, with its unusual switching rule, only 6 percent of \nall CCCDC sequences require rearrangement. Since neither module is \ngenerally nonblocking but both are at least rearrangeably nonblocking, \nthey are identically rearrangeably nonblocking.\n\nSummarizing transient analysis, Fig. 5 illustrates module perform- \nance versus template complexity for the two modules. The scale of the \nX axis has no mathematical nor physical meaning. The 222 Module \noutperforms the 2121 Module under all tested templates up to and \nincluding CCDCC sequences, but is outperformed by the 2121 Module \nunder CCCDC sequences. I expected the 222 Module to be consistently \nbetter, so I was surprised by this turn of events. A qualitative expla- \nnation of this unusual behavior is elusive, but I propose a conjecture.\n\nThe 2121 Module is distorted from its optimal connectivity by simple \nsequences of events, because of its asymmetric junctor pattern and \nnonuniform switch-count in alternate paths. These distortions usually\n\ninvolve connections, like A to Z, that go diagonally across the module \nand have only one path. The severity of these distortions is softened \nby the presence of I/O pairs that have three paths, and by the \nopportunity to be clever in the choice of switching rule. The 222 \nModule, however, is not distorted from optimal connectivity until the \nevent sequences are more complex. When it finally becomes distorted, \nit is more severely distorted than the 2121 Module.\n\nIn a simulation, a sequence of randomly generated events is applied \nto an initially idle module. In this section, the results of simulation \nprogram are plotted in a scatter diagram and interpreted.\n\nTwo distinct programs were written\u2014one for each module. The \nprograms are, however, similar at a high level of description. Program \nvariables record switch states, paths assigned to connections, inter- \nconnected parties, and other parameters and outputs.\n\nThe program begins by initializing a random number generator and \nreading module connectivity information from files. The program then \nenters a loop on traffic intensity: varying from 1.1 to 3.8 in increments \nof 0.3. Since the double exponential model of traffic is simulated, the \naverage holding time and the average quiet time of each input are \ncomputed from the traffic intensity.\n\nRather than simulate the Poisson environment, by stepping through \nsmall intervals of time and simulating events, the program skips \nthrough time from one event to the next. Corresponding to each input \ni, is the time to the next event tne[i], associated with that input. These \nvalues are initialized to an exponentially distributed random number \nwhose mean is the quiet time. The number of rearrangements is \ninitialized to zero and an interior loop, on the number of events in the \nsimulation, is entered. |\n\nTime is advanced to.the minimum value of the four tne[i] and all | \ntne[i] are reduced by that value. The event that timed out is simulated. \nIf it was a holding time that expired, the connection associated with \nthe relevant input is disconnected and tne[i] for that input is set to a \nrandom quiet time. If it was a quiet time that expired, a connection is \nmade with the relevant input.\n\nThe connection is established by first locating a random idle input. \u00a9 \nA subroutine that implements the switching rule determines the best \navailable path through the module. If there is no available path, the \ncount of rearrangements is incremented, and established connections \nare disconnected and reconnected by their best paths. The connection\n\nassociated with the event is established and tne[i] for that input is set \nto a random holding time.\n\nThe loop on events terminates, and after printing out results as a \nfunction of traffic intensity, the loop on traffic intensity terminates.\n\nSimulations were run with sequences of 1000 and 5000 events, for \nthe 222 and 2121 Modules, under varying traffic load. These numbers \nof events should give statistically significant results, and should be \nlarge enough that the transient effects from starting idle would be \ndamped out. The independent parameter is the mean of traffic inten- \nsity, the product of global rate-of-origination by per-call holding time, \nhow many simultaneous connections exist in a module at any time. \nThe dependent variable is the percentage of new connections requiring \na rearrangement. Traffic intensity is varied from slightly over 1.0 to \nslightly under 4.0, and the result of each simulation is plotted as a \nscatter diagram in Fig. 6. Simulation results for the 222 Module are \nshown with X and for the 2121 Module with +. Results of simulations \nwith 5000 events are circled and with 1000 events are not.\n\nAt low and moderate traffic intensities, it is noted that P,, gets \nworse as the traffic intensity increases and that P,, is uniformly, but \nslightly, worse in the 2121 Module than in 222 Module. In neither case \nis P,, particularly bad, even at their maxima. That there is a maximum \nis somewhat surprising. A monotonically increasing P,, might have \nbeen expected.\n\nAt high traffic load, there will be times when a 4 X 4 module is \u00a9 \ncompletely connected. The next event would have to be the discon- \nnection of one I/O pair, and the event after that would be likely to be \nthe reconnection of the same pair. It would be likely to be a connection \nbecause the traffic load is high, and it would have to be the same pair \nbecause they are the only idle input and output, the only connection \nthat could be made. Such a connection would never require rearrange- \nment because it was just disconnected. This behavior is consistent, \nwhether the module is isolated as a simple 4 X 4 network or part of a \nlarge network.\n\nA familiar result in classical statistics comes from applying Che- \nbyshev\u2019s inequality to Bernoulli trials. This Bernoulli law of large \nnumbers is used, for example in determining sample sizes of public \nopinion polls:\n\nwhere f, is the observed frequency after n trials, p is the given or \nassumed probability, and e is an arbitrary tolerance. For simulations \nwith n = 1000 events and p = 0.08, the inequality states that the \nvariation in the outcome should be within +3 percent for 97 percent \nof the simulations. The tolerance is even less with smaller values of p. \nA casual glance at Fig. 6 shows that the variation is much greater than \nthis. A conjectured explanation is that law of large numbers is derived \nfrom the assumption that the Bernoulli trials are independent. Since \nP,, in the ith event of a network simulation is highly dependent on \nprevious events, this fundamental assumption is invalidated.\n\nThree analyses were performed. The first, presented in the remain- \nder of this section, is of a generalized nature and pertains to both \nmodules. It establishes a general model to be used as a check for the \nmodels of the modules under investigation. The second analysis, in \nSection VII, is based on a Markov model of the operation of the 222 \nModule. The third analysis, in Section VIII, is based on a Markov \nmodel of the operation of the 2121 Module.\n\nIn this oversimplified model, any network configurations in which \nthe same count of connections are established are deemed to be \nequivalent. There are five equivalence classes of network configura- \ntions, corresponding to zero to four established connections, inclusive. \nHence, there are five states in the corresponding Markov process, \nshown in Fig. 7. The model is general enough to cover both modules.\n\nThe stochastic behavior of the model is based on the classical traffic \nassumptions: Poisson-distributed service arrivals and exponentially\n\ndistributed holding times. Since the Poisson and exponential distri- \nbutions are mutual inverses, this model is equivalent to having Pois- \nson-distributed arrivals of connections and disconnections or expo- \nnentially distributed off-hook and :on-hook times (as used in the \nsimulation program). Hither way of looking at it, there are two random \nvariables and there are two underlying parameters: the busyness and \nthe true time scale. If results are considered per unit time or limited \nto steady-state behavior, then true time scale is irrelevant, and one \nrandom variable may be arbitrary and the other specifies busyness per \nunit time. We will use the double-Poisson process, for this and the \ntwo later models, and arbitrarily set the per-line disconnect rate to 1.\n\nThe state transitions are computed from the individual connect and \ndisconnect rates of the four inputs (or outputs, equivalently), which \nare assumed to be statistically identical. Let \\ be the rate at which \neach individual input requests a connection through the module, and \nlet 1.0 be the rate at which each individual established connection is \ndisconnected. In other words, ) is the ratio of the connect rate to the \ndisconnect rate of each individual input (or output). The global rate \nof new connections from state i is (4 \u2014 1)A, because there are 4 \u2014 1 idle \ninputs that could make such a request. The global rate of disconnec- \ntions from state 1 is i, because there are i established connections, any \nof which could request disconnection.\n\nThe analysis is a special case of a queue with dependence on the \nstate of the system,'* but can also be viewed as a queue with finite \ncustomer population and infinite servers.\u2019\n\nThe steady-state probabilities are calculated from a conservation \nlaw. In the steady state, the mean exit rate from state 1 is the sum of \nthe rates on all exit arcs from state 1 times the steady-state probability \nof being in state 1. The mean entry rate into state 1 is the sum of the \nproducts of a rate on each entry arc times the steady-state probability \nof being in the state from which the arc comes. Conservation of calls \ndictates that the mean exit rate must equal the mean entry rate in the \nsteady-state for each state. The resulting simultaneous equations have \nthe general solution:\"*\n\nwhere n > 0, A; is the global connect rate on the arc from state i to \nstate i + 1, uw; is the global disconnect rate on the arc from state i to \nstate i \u2014 1, and g; is the steady-state probability of being in state 1. \nSetting A; = (4 \u2014 1) and yp; = 1 gives\n\nNote that each g; is expressed as a function of go. The probability go \nthat there are no connections is found by setting the sum of all steady- \nstate probabilities to 1. This gives the neat result\n\nTraffic intensity is a random variable giving the count of established \nconnections at any time. In the classical traffic model, having Poisson \narrivals with rate r and exponential holding time with mean h, traffic \nintensity is known to be Poisson distributed with mean, 7 = rh.\n\nOne way. of looking at the model of Fig. 7 is that the system has \nfour sources, each generating one calling cycle per unit time. The \ncalling cycle consists of an exponentially distributed off-hook interval \nwith mean \\/(1 +A) and an exponentially distributed on-hook interval \nwith mean 1/(1 + A). Setting the arrival rate to 4 and the holding time \nto the mean off-hook interval, the Poisson-distributed traffic intensity \nhas mean 7 = 4 X A/(1 + A). This intuitive argument is verified by \ncomputing the mean count of established connections in the steady \nstate\n\nConsider the two extremes. If \\ = 0, the steady-state probability \nmass function is\n\nand the mean traffic intensity 7 = 0. If \\ \u2014~ \u00a9, the limit of the steady-. \nstate state probability mass function is\n\nand the limit of the mean traffic intensity 7 \u2014 4. As a better example, \nlet X = 1, which means that the individual arrival rate equals the \ndeparture rate, or that each input is on-hook for the same average \ntime as off-hook. The steady-state probability mass function is\n\nshowing a trend toward state 2, with decreasing probability in either \ndirection away from state 2. The symmetry about state 2 suggests an \naverage of two connections in the steady state and setting \\ = 1 in the \nequation for traffic intensity gives 7 = 2.\n\nAs another example, let \\ = 2.0, meaning that each input is off- \nhook twice the time that it is on-hook, that is, two-thirds of the off- \nhook/on-hook cycle. The steady-state probability mass function is\n\nshowing a trend slightly under state 3 and a lack of the symmetry \nobserved in the case where \\ = 1. If \\ = 0.5, the steady-state state \nprobabilities are reversed from the case where A = 2. The mean traffic \nintensity in these cases is 8/3 and 4/3, respectively, representing the \ncenter of mass for each distribution.\n\nThe state model of the 222 Module is illustrated in Fig. 8.1\u00b0 Each \nbubble in the figure represents a state and contains the state\u2019s name \nand a representative network configuration from the state\u2019s equiva- \nlence class. State I represents the idle network configuration and is \nequivalent to state 0 in the generalized model. State J represents the \n16 network configurations with one connection established. These 16 \nnetwork configurations are all equivalent, for purposes of determining \nP,,, and this state is equivalent to state 1 in the generalized model.\n\nStates S through V represent four equivalence classes of network \nconfigurations in which two connections are established. This set of \nstates is equivalent to state 2 in the generalized model. In state S, the \ntwo established connections terminate on the-same input switch and \non the same output switch. In states T and U, the two established \nconnections terminate on opposite input switches and on opposite \noutput switches. In state T, both middle switches are used and in state\n\nU, one middle switch is shared by both established connections and \nthe other is idle. State T is the only malevolent state in the model and \nthe only transition in which rearrangement is required is the one from \nstate T to state X. In state V, either the two established connections \nterminate on the same input switch and on opposite output switches, \nor the two established connections terminate on opposite input \nswitches and on the same output switch. These conditions are equiv- \nalent for purposes of computing P,,.\n\nStates W and X represent two equivalence classes of network \nconfigurations in which three connections are established. These two \nstates are equivalent to state 3 in the generalized model. In state W, \ntwo of the established connections terminate on the same input switch \nand on the same output switch, and the third established connection \nterminates on the other input switch and the other output switch. In \nstate X, the two established connections that terminate on the same \ninput switch terminate on opposite output switches, and the two \nestablished connections that terminate on the same output switch \nterminate on opposite input switches. These are the only distinctions \nthat need be made among all network configurations with three \nconnections established.\n\nStates Y and Z represent two equivalence classes of network config- \nurations in which four connections are established. These two states \nare equivalent to state 4 in the generalized model. In state Y, two \nestablished connections that terminate on the same input switch \nterminate on the same output switch. In state Z, two established \nconnections that terminate on the same input switch terminate on \nopposite output switches. These are the only distinctions that need be \nmade among all network configurations with four connections estab- \nlished.\n\nCorresponding to the transition from state I to J is a connection \nthrough the module in which two paths are possible. The choice of \npath is arbitrary and in no way affects the results. Corresponding to \nthe transitions from state J to S or V is a second connection that has \nonly one available path. Corresponding to the transition from state J \nto U, is a connection in which the selected path shares a middle switch \nwith the established connection. Selecting the other path would cor- \nrespond to an imprudent transition to state T.\u2019\u00b0 Corresponding to \ntransitions from states T, U, and V to states W and X are third \nconnections that have only one available path, even with the rear- \nrangement required before the connection corresponding to the tran- \nsition from state T to X. Corresponding to the transition from state \nS to W, is a third connection in which two paths are possible. The\n\nchoice is arbitrary, not affecting the results, but the choice determines \nthat established connection, which if later disconnected, would corre- \nspond to the undesired transition to state T. Corresponding to the \ntransitions from states W or X to states Y or Z, respectively, are \nfourth connections in which only one path is available. The transition \nfrom state J to U represents the only connection where the switching \nrule is relevant.\n\nThe rates on the arcs connecting the states in Fig. 8 are similar to \nthose in the generalized model, but several need further explanation. \ne The sum of the connect rates on transitions from state J to states S\n\ntransition from state 1 to 2 in the generalized model. With one \nestablished connection, there are nine possible second connections: \none that terminates on the same input and output switches as the \nestablished connection, four that terminate on the opposite input \nand output switches as the established connection, and four that \nterminate on one same switch and one opposite switch as the \nestablished connection. Thus the rates on the transitions to states\n\ne The sum of the connect rates on transitions from each of states S \nthrough V must be 2), corresponding to the connect rate on the \ntransition from state 2 to 3 in the generalized model. In states S and \nV, all third connections lead to the same next state, state W or X, \nrespectively, and so each single transition is labeled with 2A. In \nstates T and U, however, half the third connections terminate on \nthe same input and output switches as an existing connection, and \nhalf terminate on the same input switch as one connection and the \nsame output switch as the other connection. Thus, states T and U \neach transit to both states W and X, and the rates on those \ntransitions are all X. All third connections associated with transi- \ntions from state T to state X require rearrangement, and these are \nthe only connections requiring rearrangement in the entire model.\u2019\u00ae \nHither of the two established connections may be rearranged so that \none middle switch is shared by both connections and one is idle\u2014 \nan intermediate network configuration that conforms to state U.\n\ne The sum of the disconnect rates on transitions from each of states \nW and X must be 3, corresponding to the disconnect rate on the\n\n_ transition from state 3 to 2 in the generalized model. In any network \nconfiguration belonging to state X, two of the three single discon- \nnections results in a configuration belonging to state V, so that arc \nis labeled with a rate of 2. The other single disconnection results in\n\nrate of 1. In any network configuration belonging to state W, a \ntransition to a configuration belonging to state V is impossible, but \neach of the three single disconnections results in a configuration \nbelonging to state S, T, or U, respectively. Thus, each of these \ntransitions is labeled with a rate of 1. The transition from state W \nto T is the only entry into that malevolent state, and it is unavoidable \nunder any prudent switching rule. |\n\nThe rates on the state transitions and the notion of X in this devel- \nopment are slightly different from that of the reference.\u2019\u00ae In that \npaper, \\ represented the rate of origination of connections between \nspecific inlet and outlet pairs. In this development, traffic is assumed \nto originate at an inlet, and terminate on any random idle outlet.\n\n7.3 Steady-state probabilities \nThe rate equations for each of the ten states are\n\nAs is customary with such processes, only n \u2014 1 of the n equations are \nindependent. Setting the sum of the n probabilities to 1 provides the \nnth independent equation. The solution, after several hours of manual \nalgebra is\n\nAs a check, it is verified that p; above equals gp from the generalized \nmodel, py = \u00a31, Ps + Pr + Pu + Pv = &, Pw + Px = 8s, and py + pz = \ng4. Of particular interest, of course, is the steady-state probability of \nthe malevolent state, pr.\n\nP,, is the proportion of those new connections that require that \nan (one) established connection be rearranged before a new connec- \ntion may be completed. The numerator is the sum over all states of \n(the average count of new connections requiring rearrangement from \nstate i) X (the steady-state probability of state i). For this network, \nit is simply 1 X pr. The denominator is the weighted average count of \npossible new connections, similar to the calculation of 7 in the previous \nsection\n\npoints for the 222 Module, taken from the scatter diagram of Fig. 6. \nSetting the derivative of the expression above to zero yields an extre- \nmum at \\ = 7 = 3, and the value at that maximum is 1.76 percent.\n\n- Related to this figure is Ny, the average count of established con- \nnections that must be rearranged when a new connection is completed. \nSince exactly one established connection is rearranged when the 222 \nModule requires rearrangement,\n\nSince the Markov model for the 2121 Module has 50 states, a natural \nnomenclature is a mapping to a familiar set with 50 elements, also \ncalled \u201cstates.\u201d The mapping is shown in Fig. 10, where each U. S. \nstate represents a set of equivalent network configurations from the \n2121 Module. A representative configuration is shown with each state \nin Fig. 10. The states are grouped according to the number of estab- \nlished connections, or their relationship to states in the generalized \nmodel. FL represents the idle Markov state and the six New England \nstates represent six Markov states in which all four inputs connect to \nall four outputs. States in three intermediate east-to-west bands across \nthe U. S. A. represent Markov states with one, two, and three estab- \nlished connections, respectively. To avoid clutter in Fig. 10, the tran- \nsitions among the states are shown in later figures.\n\nThe upward transitions in the model for the 2121 Module, in which \nno rearrangements are required, are shown in Fig. 11. Consider the \nthree upward transitions from FL. The transition from FL to GA \nrepresents establishing, in an originally idle module, an A-to-Z (or C- \nto-W) connection diagonally across the module. This connection has \nonly one path through the module, and that path requires the central \nswitch in the crossed state. By contrast, A-to-W (or C-to-Z) connec- \ntions have two paths through the module, represented by NM and AZ. \nA direct upward transition from FL to AZ, and none from FL to NM, \ndemonstrates the preference for the path that avoids the central \nswitch. Similarly, the direct upward transition from FL to AL, and \nlack of same to LA or TX, demonstrates the choice of the path for an \nA-to-X connection that avoids the central switch over the other two \npaths. Similar logic governs the other upward transitions on the graph. \nFor simplicity, the weights on the arcs are not shown, but they are \nsimilar to those of the 222 Module. |\n\nThe upward transitions, in which one or two established connections \nrequire rearrangement, are shown in Fig. 12. The three states with \none established connection from which rearrangements -may be re- \nquired\u2014LA, TX, and NM\u2014have no direct upward transitions from \nFL, while the only three states that can be reached directly from FL\u2014 \nGA, AL, and AZ\u2014have no upward transitions requiring rearrange- \nment. Two of the states with two established connections from which \nany rearrangements may be required\u2014TN and KS\u2014have no upward \ntransitions from any state with one established connection. The other \nfive states with two established connections from which any rearrange- \nments may be required\u2014VA, IL, AR, IA, and NE\u2014have upward \ntransitions from states with one established connection, but only from \nLA, TX, and NM, the states that cannot be reached directly from FL. \nThe transitions from IA to MD, KS to SD, and KS to ND are \nparticularly interesting because both established connections require \nrearrangement before the new connection, corresponding to the tran- \nsition, can be completed. Since the states from which rearrangement \nis required are reached only through complex sequences of connection \nand disconnection, we expect P,, to be small.\n\nThe downward transitions in the model for the 2121 Module are \nshown in Fig. 13. The rates on all the arcs connecting the states, as in \nthe two previous figures, are not shown.\n\nThe rate equations for each of the 50 states are given in Appendix \nA. A closed-form solution to this set of equations has not been found, \nbut approximations were derived by a simple, and tedious, iteration.\n\nThe states are initially distributed equilikely within their equiva- \nlence class such that the sum of the state probabilities within each \nclass equals the corresponding state probability in the generalized \nmodel. Thus,\n\nThen, temporarily deleting (1 + A)* from every denominator, these \nexpressions were substituted into the right side of the rate equations \nin Appendix A. The resulting intermediate expressions for each q, are \npolynomials in X divided by the (a + bd) term on the left side of the \ncorresponding rate equation. The long division was effected and the \nquotient was truncated to a simple polynomial in ), a new expression \nfor each q,. Most of these new expressions had coefficients between \ndouble and half their analogs in the original expressions.\n\nThese new expressions were then substituted into the rate equations, \nand a similar simplification and approximation was effected. By the \nthird iteration, the expressions were surprisingly close to those of the \nsecond iteration, and rapid convergence was observed. These approx- \nimations to the steady-state probabilities, without the long division \nand quotient truncation in the final iteration, are given in Appendix \nB. Of particular interest, again, are the steady-state probabilities of \nthe ten malevolent states.\n\nP,, is the proportion of those new connections that require rear-\u2014 \nrangement of an (at least one) established connection before the new \nconnection may be completed. The numerator is the sum over all \nstates of (the average count of new connections requiring rearrange- \nment from state i) X (the steady-state probability of state 1). For the \n2121 Module, the numerator is\n\n. = (2.1d* + 4.8d\u00b0 + 1.0X? \u2014 1.1A)/(1 + 38dA)(1 + A), \nusing the approximate expressions from Appendix B. The denomi- \nnator, 4/(1 + A), is the weighted average count of possible new \nconnections, as in the calculation from the previous section. The ratio \nis then\n\nThe curve for this expression is plotted in Fig. 14 through the data \npoints for the 2121 Module, taken from the scatter diagram of Fig. 6. \nSetting the derivative of the expression above to zero yields an extre- \nmum at 7 = 2.6, and the value at that maximum is 3.1 percent.\n\nRelated to this figure is, N,, the average count of established \nconnections that require rearrangement when a new connection is \ncompleted. Since this count is two on three of the state transitions, \nthis figure is not equal to P,,:\n\nThis expression is also plotted in Fig. 14. We observe that the Markov \nanalysis of such an innocent-looking network is surprisingly complex.\n\nTwo algorithms for a switching rule are by direct calculation or by \ntable look-up. The direct calculation of the switching rule for either \nmodule would be time-consuming, with the switching rule for the 2121 \nModule significantly more complicated than that for the 222 Module. \nThe count of connection combinations in either module is also large, \nwith little difference between the two modules, so a table look-up \nimplementation of the switching rules would require a large ROM.\n\nTwo distinct realizations of such an algorithm are by a microcon- \ntroller per module or by a common controller governing all modules\n\nin a network. Direct calculation in a common controller of a large \nnetwork suggests a real-time bottleneck and table look-up in a per- \nmodule algorithm suggests considerable replication of a large ROM. \nTherefore, the logical choices are distributed control by direct calcu- \nlation in a per-module microcontroller, or centralized control by table \nlook-up. The count of network configurations in a table look-up \nalgorithm is discussed in this section.\n\nThe Markov diagram of the 222 Module is repeated in Fig. 15, except \nthat an additional number is placed in each bubble, the count of unique \nnetwork configurations that are represented by the Markov state. \nState I represents the only idle network configuration. Considering \nconfigurations with a single established connection, since any of four \ninputs can be connected to any of four outputs, and each connection \nhas two paths through the network, state J represents 4 x 4 x 2 = 32 \nnetwork configurations.\n\nStates S through V represent all configurations with two established \nconnections. A configuration in state S derives from a configuration \nin state J by connecting the only other input on the same input switch \nas the established connection to the only other output on the same \noutput switch as the established connection by the only available path. \nThus state S also represents 32 configurations. A configuration in \nstate T derives from a configuration in state J by connecting either \ninput on the opposite input switch as the established connection to \neither output on the opposite output switch as the established connec- \ntion by the path using the unused middle switch. Thus state T \nrepresents 32 X 2 X 2 = 128 configurations. A configuration in state \nU derives from a configuration in state J by connecting either input \non the opposite input switch as the established connection to either \noutput on the opposite output switch as the established connection by \nthe path sharing the used middle switch. Thus state U represents \n32 X 2 X 2 = 128 configurations. A configuration in state V derives \nfrom a configuration in state J by connecting the only other input on \nthe same input switch as the established connection to either output \non the opposite output switch as the established connection, or either \ninput on the opposite input switch as the established connection to \nthe only other output on the same output switch as the established \nconnection by the only available path. Thus state V represents 32 x \n(2 + 2) = 128 configurations.\n\nStates W and X represent all configurations with three established \nconnections. A configuration in state W derives from a configuration \nin state S by connecting either input on the unused input switch to \neither output on the unused output switch by either available path, or\n\nit derives from a configuration in state T or U by interconnecting \neither of two specific I/O pairs by the only available path. The specific \npair share both an input switch and an output switch with either \nestablished connection. Thus state W represents 32 X 2X 2X 2=128 \nxX 2 = 256 configurations. A configuration in state X derives from a \nconfiguration in state U by connecting either of two specific I/O pairs,\n\nthose that share an input switch with one established connection and \nan output switch with the other one, by the only available path. Each \nalso derives from configurations in state V by connecting either idle \ninput to either idle output by the only available path, but each derives \nfrom two different configurations in state V. Thus state X represents \n128 X 2 = 128 X 2 X 2/2 = 256 configurations.\n\nStates Y and Z represent all configurations with four established \nconnections. Each configuration in state Y derives from a configura- \ntion in state W by connecting the only idle I/O pair by the only \navailable path. Each configuration in state Z derives from a configu- \nration in state X by connecting the only idle I/O pair by the only \navailable path. Thus state Y represents 256 configurations and state \nZ represents 256 configurations.\n\nThe Markov diagram of the 2121 Module, showing downward tran- \nsitions, is repeated in Fig. 16, except that a number replaces the state \nname, the count of unique network configurations that are represented \nby the corresponding Markov state.\n\nState FL represents the only idle network configuration. States GA \nthrough AZ represent all configurations with one established connec- \ntion. In the four configurations in GA, any of the four inputs connects \nto the output on a middle switch diagonally across the module by the \nonly path. In the configurations in AL, LA, and TX, any of the four \ninputs connects to either output on the output switch. Each state \nrepresents 4 X 2 = 8 configurations and the states are distinguished \nby which of three paths is used for the connection. In the configura- \ntions in NM and AZ, any of the four inputs connects to the output on \na middle switch directly across the module. Each state represents four \nconfigurations and the states are distinguished by which of two paths \nis used for the connection. Summing it up, states GA through AZ \nrepresent 3 X 4+ 3 X 8 = 36 configurations.\n\nStates NC through HI represent all configurations with two estab- \nlished connections. In one subset of these 21 states, the inputs, one \nfrom each input switch, connect to the two outputs on the middle \nswitches. Each state in this subset\u2014SC, MS, or HI\u2014represents four \nconfigurations, and the states are distinguished by whether the con- \nnections are diagonally across the module or directly across by either . \nof two similar paths. In OK, either output on a middle switch connects \nto either input directly across the module by the best path, and the\n\nother input on that same input switch connects diagonally across the \nmodule to the other output on a middle switch. In CA, either output \non a middle switch connects to either input directly across the module \nby the best path and either input on the other input switch connects \ndirectly across the module to the other output on a middle switch by \nthe worst path. OK represents 2 x 2 = 4 configurations and CA \nrepresents 2 X 2 X 2 = 8 configurations.\n\nIn another subset of these 21 states, any of the four inputs connects \nto either output on the output switch and the other input on the same \nswitch connects to any other output by the best remaining path. Each \nstate in this subset\u2014NC, IN, MO, CO, or NV\u2014represents 4 x 2 = 8 \nconfigurations, and the states are distinguished by the path of the first \nconnection and/or the output of the second connection. In another \nsubset of these 21 states, either input on the upper input switch \nconnects to either output on the output switch, and either input on \nthe lower input switch connects to the other output on the output \nswitch, and both connections use similar paths. Each state in this \nsubset\u2014KY, TN, or [A\u2014represents 2 x 2 X 2 = 8 configurations, and \nthe states are distinguished by the three possible paths. In the final \nsubset of these 21 states, any of the four inputs connects to either \noutput on the output switch and either input on the other input switch \nconnects to any other output. Excluding the cases, covered in the \nprevious subset, where the paths are symmetric, each state in this \nsubset\u2014VA, WV, IL, AR, NE, KS, WY, and UT\u2014represents 4 x 2 x \n2 = 16 configurations, and the states are distinguished by the path of \nthe first connection and/or the output and/or the path of the second \nconnection. Summing it up, states NC through HI represent 4 x 4 + \n9X 8+8 xX 16 = 216 configurations.\n\nStates DE through AK represent all configurations with three \nestablished connections. In all 16 states, any of four inputs connects \nto either output on the output switch, and either input on the other \ninput switch connects to some other output. Each state represents \n4 x 2 X 2 = 16 configurations, and the states are distinguished by the \npath of the first connection and/or the path and/or output of the \nsecond connection and/or the input and/or output and/or path of the \nthird connection. Summing then, states DE through AK represent \n16 X 16 = 256 configurations.\n\nStates RI through VT represent all configurations with four estab- \nlished connections. In the subset containing RI, CT, ME, and VT, \neither input on the upper input switch connects to either output on \nthe output switch, either input on the lower input switch connects to \nthe other output on the output switch, and both connections use \nsimilar paths. The remaining two inputs connect to the remaining two \noutputs, by the best remaining similar paths. Each state represents\n\n2x 2 X 2 = 8 configurations, and the states are distinguished by the \npath of the first two connections and/or the parties interconnected by \nthe last two connections. In MA, both inputs on one input switch \nconnect to the outputs on the middle switches by the best paths, and \nboth inputs on the other input switch connect to the outputs on the \noutput switch. In NH, both inputs on one input switch connect, one \nto the output straight across on a middle switch and the other to one \nof the outputs on the output switch, using paths like those in VT. \nBoth inputs on the other input switch connect to similar outputs, but \nusing paths like those in CT. Both MA and NH represent 2 x 4 xX \n2 = 16 configurations. Summing them up, states RI through VT \nrepresent 4 X 8 + 2 X 16 = 64 configurations. \nSumming them up, we count a total of\n\ndistinct network configurations in the 2121 Module. The module \nsupports more configurations, but no others are reached using the \nprudent rule. We see that the 222 Module has more than 2.5 times as \nmany reachable configurations as the 2121 Module and, hence, a table- \nlook-up implementation of a control algorithm would be more complex \nfor the 222 Module than for the 2121 Module.\n\nTwo architectures for a 4 X 4 photonic switching network were \ncompared by their traffic-handling capacity. Both networks are rear- \nrangeably nonblocking. The percentage of sequences requiring rear- \nrangement was found to be tolerable for both modules. Thus, both \nmodules are judged acceptable, insofar as rearrangeably nonblocking \nmodules are acceptable, and practically indistinguishable in their \ntraffic performance. The selection of one module over the other may \nproceed based on criteria other than traffic capacity, like loss, cross- \ntalk, or cost of manufacture.\n\nAcknowledgments are extended to Vic Benes, Nick Maxemchuk, \nand Krishnan Padmanabhan for their time, encouragement, and useful \ndiscussion and to Eric Grosse for his help with SMP.\"\"\n\n1. G. Broomell and J. R. Heath, \u201cClassification Categories and Historical Development \nof Circuit Switching Topologies,\u201d Comput. Surv., 15, No. 2 (June 1983), pp. \n95-133.\n\n2. V. E. Benes, Mathematical Theory of Connecting Networks, New York: Academic \nPress, 1965.\n\n3. R. C. Alferness, R. V. Schmidt, and E. H. Turner, \u201cCharacteristics of Ti-Diffused \nLithium Niobate Optical Directional Couplers,\u201d Appl. Opt., 18, No. 23 (December\n\ntor for Time-Division Multiplexing and Data Encoding,\u201d J. Lightwave Technol.,\n\n. L. McCaughan, \u201cDesign and Performance Limitation of Integrated Electro-Optic \nCross Points,\u201d Proc. IEEE Globecom, Alanta, November 1984, pp. 878-9.\n\n. J. E. Watson, \u201cPolarization-Independent 1 xX 16 Optical Switch Using Ti:LiNbOs \nWaveguides,\u201d Conf. Optical Fiber Commun., San Diego, February 1985, p. 110.\n\n. H. S. Hinton, \u201cA Nonblocking Optical Interconnection Network Using Directional \nCouplers,\u201d Proc. of IEEE Globecom, Atlanta, November 1984, pp. 885-9.\n\n11. S. Kobayashi and T. Kimura, \u201cSemiconductor Optical Amplifiers,\u201d IEEE Spectrum, \n21, No. 5 (May 1984), pp. 26-33.\n\n14. M. Schwartz, Computer-Communication Network Design and Analysis, Englewood \nCliffs, N.J.: Prentice-Hall, 1977.\n\n16. V. E. Benes, \u201cProgramming and Control Problems Arising from Optimal Routing \nin Telephone Networks,\u201d B.S.T. J., 45, No. 9 (November 1966), pp. 1373-438.\n\n17. SMP, A Symbolic Manipulation Program, Primer and Summary, Inference Corp.,\n\nRichard A. Thompson, B. S. (Electrical Engineering), 1964, Lafayette \nCollege; M. S. (Electrical Engineering), 1966, Columbia University; Ph.D. \n(Computer Science), 1971, University of Connecticut; AT&T Bell Laborato- \nries, 1963-1968, 1977\u2014. From 1963 to 1968 Mr. Thompson was involved in \nswitching systems development at AT&T Bell Laboratories. He was a member \nof the Electrical Engineering Department at Virginia Polytechnic Institute \nfrom 1971 to 1977, achieving the rank of Associate Professor. Since 1977 Mr. \nThompson has been a member of the Digital Systems Research Department \nat AT&T Bell Laboratories, with a brief stint in ABI/ATTCP in 1983. His \nresearch interests are probabilistic formal languages, fault tolerance and \ncellular automata, terminals and the human-machine interface, communica- \ntions switching systems, and, most recently, photonic switching. Mr. Thomp- \nson is an active participant in IEEE Computer and Communications Societies. \nSenior Member, IEEE.\n\nIn the present paper we use transform methods (characteristic function \ntechniques) and contour integrals to derive a closed-form expression for the \nperformance union bound of a general discrete-time system. We show that \npreviously published results may be derived as particular cases of the general \nformulation developed in this paper. It is well known that the maximum- \nlikelihood Viterbi algorithm may be employed not only for decoding of con- \nvolutional codes but also for optimal detection in other situations. Examples \ninclude bandwidth-efficient demodulation, optimal accommodation for inter- \nsymbol interference and cross-channel coupling, text recognition, simultane- \nous carrier phase recovery and data demodulation, digital magnetic recording, \nnonlinear estimation and smoothing. The union bound is a useful measure of \nthe performance of the Viterbi algorithm. Past closed-form expressions for \nthe union bound have usually involved considerable approximation.\n\nA Maximum-Likelihood Receiver (MLR) is optimal when the input \nsignal is distorted not only by noise but also by some deterministic \nfactors. The MLR compares the received signal with all possible \nsignals distorted by the same deterministic factors but not by the \nnoise. The latter comparison signals must be available at the receiver. \nPossible deterministic impairments include intersymbol interference, \ncross-channel coupling, modem-implementation errors, channel mis- \nequalization, signal distortion by the channel nonlinearities, etc. \nGuided by some metric, the MLR searches for the comparison signal\n\nCopyright \u00a9 1985 AT&T. Photo reproduction for noncommercial use is permitted with- \nout payment of royalty provided that each reproduction is done without alteration and . \nthat the Journal reference and copyright notice are included on the first page. The title \nand abstract, but no other portions, of this paper may be copied or distributed royalty \nfree by computer-based and other information-service systems without further permis- \nsion. Permission to reproduce or republish any other portion of this paper must be \nobtained from the Editor.\n\nthat is closest to the signal actually received and asserts that this \nsignal, as it existed before being subject to the deterministic impair- \nment, was the transmitted message.\n\nThe MLR performance depends on how well we can model the \ndeterministic distortions of the signal and also on distances between \nsignals. The signal separation may be increased by coding.\n\nTechnical problems arise in MLR implementation. Strictly speak- \ning, we need to store the whole transmission history and generate all \npossible comparison sequences. However, if the system may be mod- \neled by a Markov process, then the Viterbi algorithm may be used to \nrealize a recursive MLR.\u2019\n\nIn designing the MLR it is very important to be able to accurately \nevaluate the receiver performance. Because of a very large number of \n\u2018possible comparison signals, it is difficult to find the exact formula for \nan MLR performance characteristic. The performance characteristic \nupper bound (so-called union bound) is more easily found. Viterbi \nintroduced transfer function techniques to evaluate the union bound \nfor some performance characteristics of binary convolutional codes.\u2019 \nThese methods have been extended to obtain performance bounds of \nthe general finite-state system.\u201d However, the original union bound \nwas loosened to simplify a series summation. Using the transform \nmethods developed in this paper, the original union bound is expressed \nin closed form.\n\nwhere u, is a source symbol, x; is the transmitted channel symbol, and \ns, is the corresponding system (transmitter) state. Symbols x, are \ntransmitted over a noisy memoryless channel that outputs symbols\n\nwhere n, are independent identically distributed variables. The re- \nceiver outputs symbols w, = (S;, U,) which minimize the sum\n\nwhere m(4\u00a5z, wz) is the so-called branch metric and may be treated as \na cost function for making a decision that the transmitter superstate\n\nwas w, if y, was received. This metric is usually a measure of the \nsignal degradation due to noise. For example,\n\nThe optimal solution may be found using the Viterbi algorithm. A \nsequence Wo, W1, \u00ab++ , W;-1, Wj, Which terminates at time J, is called a \nsurvivor if it minimizes the metric accumulated to this time:\n\nJ 7 j \nDY Myr, We) = Mini...) L M(Yes Wa), \nk=0 k=0 \nwhere w; = W;. It is obvious that the sequence that minimizes the total \nsum M(y, w) must begin with one of the survivors. The survivors may\n\nbe determined recursively via the Viterbi algorithm: Wo, Wi, --- , W,; \nW +1 is a survivor if and only if Wo, Wi, \u00ab++ , Wj-1, Ww; iS a Survivor and\n\nIf all the survivors have a common part, then this part also belongs \nto the sequence that minimizes the total metric M(y,w) and the \ncorresponding symbols are output by the Viterbi receiver.\n\nDepending on the application, we may want to evaluate the Viterbi \nreceiver performance using some distortion measure d = E{d(wz, wz)}, \nwhere d(w;, W,) is the distortion characteristic of the symbol w; which \nthe receiver identifies as w,. For example, if we wish to find a symbol \nerror probability, then the distortion characteristic d(w,, W,) is equal \nto zero if there are no errors in the symbol (u, = &) and is equal \nto one otherwise. If we wish to evaluate a bit error probability, then \nd(w,, W,) = m,/b, where m, is the number of bit errors in the symbol \nu, (the Hamming distance between u, and u,), and 6b is the total \nnumber of bits in the symbol.\n\nIn order to find the distortion measure union bound, consider an \nerror event of length L (see Ref. 1) that is a pair of the correct path\n\noy = {w,} and an incorrect path o, = {w,} such that s, # \u00a7, for k = 1, \n2,---,L\u20141 and s = \u00a7, otherwise. \nThen the distortion measure is upper bounded by the union bound:?\n\nwhere the average is taken over all source sequences, P;(c,/c,) is the \nconditional probability that the incorrect path o, has a larger metric \nthan the correct path oz, and 0.5 the prOnen ha of equality of metrics \nif a tie is resolved randomly:\n\nConsider first a discrete memoryless channel. Define a generating \nfunction of a variable Am(yp, Wz, We):\n\nD(z; we, We) = X Prfy,/weper Powe), (4) \nYk \nThe generating function of the sum of variables Am(y;, wz, Wz) is \nequal to the product of the generating functions (4), and the probability \nP,(\u00a2,/o,) is equal to the sum of the coefficients of the positive power \nterms of the product and one half of the zero power term. Using \ncontour integrals, we may express the sum as\n\nwhere p > 1. The total distortion along the incorrect path o, may also \nbe expressed using generating functions:\n\nCE {u:Rg N |v| > 1}, Re is the region of convergence of (8). The \nright-hand side of the inequality (7) is a new expression of the union \nbound of the average distortion d.\n\nThe generating function G(z, v) may be found from the system state \ntransition graph with branch weights\n\nor the corresponding matrix equation.\u2019\u00b0 The system symmetry sim- \nplifies the construction of the generating function.\n\nIn the case of a continuous channel we may also use a transform \n(characteristic function) technique to obtain a union bound similar to \n(7). For example, if h(x, n) = x +n, where n is a zero-mean Gaussian \nvariable with one-sided spectral density No (AWGN) and Euclidean \nmetric is used by a Viterbi algorithm, then (3) takes on the form\n\nEquation (12) is similar to (5), and therefore we may derive a new \nunion bound for the analog case that is similar to (7):\n\nwhere \u201c+\u201d denotes mod2 addition (exclusive-or). \nThe second equation of the system (1) 5,4; = g(w,) may also be \nexpressed as the system\n\nHowever, usually it is represented by the system state graph whose \nnodes correspond to the system states, and the symbols along the \ntransition lines indicate the encoder input symbols (see Fig. 2).\n\nThe next step is to define the algorithm branch metric. Suppose \nthat we received a sequence yo, V1, --: , Yn. Chen the MLR will output \nthe sequence i, t, --- , &y, which maximizes the likelihood proba- \nbility |\n\nwhere z is the number of bit errors in the sequence xo, %1, --: , Xn OF, \nin other words, the Hamming distance (HD) between the sequences: \nz2=HD{(yo, --+ , yn); (x0, \u00ab++ \u00bb Xn) }. Since 0 < (p/q) < 1, maximization \nof the likelihood is equivalent to minimization of the HD. Therefore \nwe may define the algorithm metric as m(w, w) = HD{y; x}. This \nmetric depends only on the signal difference.\n\nThe channel model may be expressed by eq. (2), which takes the \nform yz = x, + nN\u00bb, where nz = (nz, nz) has the following distribution:\n\nWe wish to evaluate the decoder output error probability. Therefore \nthe distortion characteristic is\n\nThis distortion characteristic depends only on the signal difference.\n\nWe assume that the all-zero sequence is transmitted. The generating \nfunction (4) now takes on the form |\n\nwhere w = (0, 0, 0), u(w) = HD{(0, 0); x} is the Hamming weight of x \n= f(w). The distortion measure d(w, w) = t; therefore 24\u201d) = z\" \nand\n\nIf we use these values as branch weights on the system state graph, \nwhere the all-zero state is split into an initial and final state\u2019 as shown \nin Fig. 3, we obtain\n\nas the transition from the initial state to the final state. \nThe union bound for the bit error probability is found from (7):\n\nIn this example we find the union bound on the bit error probability \nfor the same convolutional code but with soft-decision decoding. \nSuppose that symbols x, are BPSK modulated. The receive demodu- \nlator is followed by a soft-decision sampler with infinite precision.\n\nis the Euclidean distance (11) between the signals and 6? = E,/Np is \nthe signal-to-noise ratio. |\n\nIt is clear that the maximum likelihood corresponds to the minimum \nof the Euclidean distance and vice versa. Therefore we may define the \nMLR metric as m(wW, w) = ||x \u2014 x|]* = (x\u2019 \u2014 x\u201d)? + (x\u201d \u2014 x\u2019)*. This \nmetric depends only on the signal difference.\n\nUsing the same arguments as in the previous example, we may \nassume that the all-zero sequence was transmitted. According to (10) \nand (13)\n\nwhere a = 1 \u2014 oe\", (B= vin 2), and g = Qe\" (1 \u2014 e~\u201d), we may \nobtain better approximation. For n = 2 \nPp < 0.5a77[A erfc(8V5) +B erfc(BV6) + C erfe(BV7)],\n\nUsing contour integrals we have derived a closed-form expression \nof the union bound on Viterbi algorithm performance. The transform \nmethods developed in this paper may be used for some other applica- \ntions. For example, using eq. (12), a sum\n\nmay be expressed as \n_ 2B aa ren a \ndu, \n(B? + v2). \nwhere F(z) = )) a,z\u201d is the sequence generating function. \n: .\n\nThe union bound on the performance characteristic of hard- and \nsoft-decision codes was found as an illustration.\n\nThe author wishes to thank D. Leed and the referees for their \nvaluable suggestions and comments.\n\n1. A. J. Viterbi, \u201cConvolutional Codes and Their Performance in Communications \nSystems,\u201d {EEE Trans. Commun. Tech., COM-19 (October 1971), pp. 751-72.\n\n2. J. K. Omura, \u201cPerformance Bounds for Viterbi Algorithms,\u201d ICC\u2019 81 Conf. Record, \nDenver, June 1981, pp. 2.2.1-.5.\n\n3. G. D. Forney, Jr., \u201cMaximum-Likelihood Sequence Estimation of Digital Sequences \nin the Presence of Intersymbol Interference,\u201d IEEE Trans. Inform. Theory., IT- \n18 (May 1972), pp. 363-78.\n\n4. F. Oberhettinger and L. Badii, Tables of Laplace Transform, New York: Springer- \nVerlag, 1973, p. 16, eq. 2.34.\n\n5. K. A. Post, \u201cExplicit Evaluation of Viterbi\u2019s Union Bounds on Convolutional Code \nPerformance for the Binary Symmetric Channel,\u201d IEEE Trans. Inform. Theory., \nIT-23 (May 1977), pp. 403-4.\n\nWilliam Turin, M.S. (Mathematics), 1958, Odessa State University; Ph.D., \n(Mathematics), 1966, Institute for Problems of Mechanics, Academy of Sci- \nences, USSR; AT&T Bell Laboratories, 1981\u2014. Mr. Turin has worked at New \nYork University and was an Associate Professor at the Moscow Institute of \nElectrical Engineering and Telecommunications. His research activities in- \nclude modeling satellite communication channels, simulation, design and \nanalysis of error-correcting codes, and analysis of Markov processes. He is the \nauthor of three books. Senior member, IEER.\n\nWe describe a new File Transfer Protocol (FTP) that provides a simple and \nefficient way of transferring files between heterogeneous systems, such as the \nUNIX\u2122 operating system, Duplex Multi-Environment Real-Time (DMERT), \nand IBM/MVS. This FTP has been adopted as an AT&T standard. In this \nprotocol, global functions requiring close coordination are separated from local \nfunctions. Functions that require global coordination are mandatory parts of \nthe protocol and must be implemented uniformly. Local functions, such as file \nmanagement and user interface, can be adjusted to local needs and might even \nbe optional. This results in a flexible protocol that can be implemented at \nvarious levels of complexity. Thus, FTP implementations can range from very \nsimple ones that provide basic file transfer service to highly complex ones that \nprovide extensive security checking and allow a variety of file management \nservices.\n\nFile transfers typically represent a large percentage of the traffic \nvolume on data networks. In one internal AT&T network, for example, \nover 50 percent of the data volume is file transfer traffic. Thus, it is \nimportant that file transfers be implemented through an efficient, \nflexible, and reliable protocol.\n\nthe functionality of all file transfer protocols previously used within \nAT&T. This protocol, called the BX.25 File Transfer Protocol (or \nsimply the FTP), provides a simple and efficient way to transfer files \nbetween heterogeneous systems such as the UNIX operating system, \nDuplex Multi-Environment Real-Time (DMERT),\u2019 IBM/MVS, and \nUNIVAC/EXEC. This FTP has been adopted as an AT&T standard. \nSeveral groups in the company have implemented or are implementing \nthis protocol.\u2019\n\nOur approach in designing the FTP was to separate global functions \nrequiring close coordination (such as parameter negotiation) from \nfunctions that could be done locally at each individual node. Functions \nthat require global coordination are mandatory parts of the protocol \nand must be implemented uniformly. Local functions, such as file \nmanagement and user interface, can be adjusted to local needs and \nmight even be optional. This results in a flexible protocol that can be \nimplemented at various levels of complexity. One can view the FTP \nas essentially an intelligent bulk data transfer facility.\n\nThe FTP is a three-party protocol in which a user at a remote node \n(the initiator) can transfer files between a source and a destination \nnode. The FTP offers three types of services: Copy, Cancel, and Status. \nThe Copy service allows the transfer of a set of files between the \nsource and destination nodes. The Cancel and Status services, respec- \ntively, allow the user to cancel a transfer and request information \nabout a transfer. All services can be done in the background, requiring \nminimal user supervision.\n\nOur approach in specifying the FTP is consistent with the one \ncurrently recommended by the protocol community. This approach \nrecommends that one first describe the services offered by the protocol \nto the upper layer, as well as the services required by the protocol \nfrom the lower layer. Then, it defines the peer-level protocol that fills \nthe gap between the upper and lower layers. Typical protocol specifi- \ncations, such as levels 2 and 3 of X.25,* cover only Peer: -level message \nexchange rules and formats.\n\nThe application program resident at each node that performs file \ntransfers will be called a File Transfer System (FTS). Functions \nperformed by the FTS cover the application and presentation layer \nfunctions as defined in the Open System Interconnection (OSI) \nmodel.\u00b0\n\nAs shown in Fig. 1, an FTS has several interfaces: an interface with \nthe user, a data transfer facility interface, and interfaces with the file \nsystem and with the operating system. In addition, an FTS commu- \nnicates with the remote FTSs using a peer-level protocol. The user is \na person or a computer program that wants to use the file transfer \nservice. The data transfer facility is the lower level of protocol, such\n\nas the BX.25 session layer. The file system is the local storage system \nfor files, and the operating system manages the local computing \nresources.\n\nIn Section II, we describe our overall design approach in greater \ndetail. We then describe services offered by the FTP in Section III, \nand identify the services required from the data transport layer in \nSection IV. The peer-level protocol is described in Section V. Section \nVI describes the negotiation procedure during which the source FTS \nand the destination FT'S agree on the options for file transfers. Some \nexamples are given in Section VII. We also describe a formal specifi- \ncation of the FTP using the selection/resolution model in Section \nVIIL.\u00ae\u2019 This specification gives a more complete and precise descrip- \ntion of the protocol than an English language specification alone can \nprovide. In Section [X, we compare the FTP with others reported in \nthe literature. Finally, we make some concluding comments in Section \nX.\n\nIl. DESIGN APPROACH \nIn our design, we have clearly separated functions requiring global\n\ncoordination from those that are local to a node. Every implementation \nwould be required to implement the global coordination functions that \nwould form the core of the protocol. Implementors would be free to \nimplement the local functions as needed. Implementations would thus \nrange from very simple ones that provide basic file transfer services \nto highly complex ones that provide extensive security checking, allow \na variety of file management capabilities, and provide a sophisticated \nuser interface. Furthermore, the flexibility in implementing local func- \ntions results in a file transfer protocol that works in a heterogeneous \nenvironment, is adaptable to diverse local needs, and can evolve as \nrequirements change. ,\n\nThe core of the protocol implements the coordination functions to \ntransfer a byte stream efficiently. Translation between the file formats \nat the source and the destination is done outside the core of the \nprotocol by local functions called preprocessors and postprocessors.\n\nFor example, if the initial requirements are file transfers between a __ \nUNIX system and an IBM/MVS system, we need only write pre- and \npostprocessors for translation between their file formats. As additional \nsystems, such as a Honeywell computer, are added, we will then add \nnew pre- and postprocessors.\n\nThe FTP does not have a networkwide standard for file naming. A \nuser must supply all the naming information required for accessing a \nremote file. This includes the information about the device name on \nwhich the file is stored, as well as the path name of the file. This\n\nThe FTP uses checkpointing to transfer a byte stream efficiently. \nDuring a file transfer, checkpoints are set along the way. After recovery \nfrom a transmission failure, the file transfer is resumed from a check- \n\u2018point up to which the destination has received the file data correctly. \nThe file transfer does not have to be restarted from the beginning. \nThis feature is useful when transferring very large files such as those \nstored on several magnetic tapes.\n\nOur security policy makes it difficult for a user to break into the \nsecurity mechanism of a remote file system. To access a remote file, \nthe user must provide access information required at the remote file \nsystem. The user without the proper access information is not allowed \nto access a remote file. In the following sections, we give further details \nof the protocol.\n\nTo invoke a service, the user must issue a Sominand to ie local \nFTS. The syntax of the command is a local decision. However, the \ninformation contained in the command is required for global coordi- \nnation. Thus, in any implementation of the FTP, each command must \ncontain the information given below.\n\nFor the Copy service, the command must contain the following \ninformation:\n\n(File descriptor at the source) (Preprocessors) \n{Destination address) \n(File descriptor at the destination) (Postprocessors) \n(Priority)\n\nwhere \n(User id) identifies the user initiating the file transfer, \n(Source address) identifies the computer on which the files to be \ntransferred reside, \n(File descriptor at the source) is the list of file names to be \ntransferred, \n(Pre-processors) is a list of programs executed on the files before \ntransfer, \n(Destination address) identifies the computer to which the files \nare to be transferred, _ \n(File descriptor at the destination) is the list of file names at the \ndestination. The number of files in this list must be the same as \nthat in the list (File descriptor at the source), \n(Post-processors) is a list of programs executed at the destination \non the files transferred, \n(Priority) specifies the priority level assigned to the copy com- \nmand. This field specifies how the FTS should treat the present \njob compared to the other file transfers waiting to be done at the \nlocal FTS.\n\nThe initiator FTS checks the syntactic correctness of the copy \ncommand. The number of file names in the list (File descriptor at the \nsource) must be the same as that in (File descriptor at the destina- \ntion). If there is an error in the copy command, then the FTS returns\n\nan error message to the user. Otherwise, the FTS returns a job number \nto the user. The job number identifies a copy command throughout \nthe network and the command\u2019s lifetime. The format for a job number \nensures that it is unique throughout the network. The job number has \nthe following format:\n\nwhere \n(Initiator address) specifies the address of the initiator, \n(Unique number assigned by the initiator FTS) is a number \nassigned to the command; it uniquely identifies the command at \nthe initiator. \nThe local FTS must notify the user upon successful or unsuccessful \ncompletion of the file transfer.\n\nA user at the initiator FTS can inquire about the status of a file \ntransfer request using the following command: \nSTATUS (Job number).\n\ne Whether preprocessing, file transfer, and postprocessing have \nstarted, and, if they have, whether they have been completed \nsuccessfully or unsuccessfully;\n\nin progress. For such a service, a user issues the following command: \nCANCEL (Job number). |\n\nThe connection-oriented service must transfer messages of variable \nlength from a transmitter FTS* to a receiver FTS in error-free form \nand in the same sequence as originally delivered to the transmitter \nFTS. If it is unable to deliver messages in sequence because of a \nnetwork failure, it must inform the transmitter FTS. Once a connec- \ntion is set up between two FTSs, it can be used to transport the files \nfrom several file transfer requests. The following service primitives, \nor their equivalent, must be provided to the FTS:\n\n1. Connect (Destination Address, FTS). Using the Connect primi- \ntive, the FTS in the local computer establishes a connection with an \nFTS resident in the computer with the address Destination Address. \nIf the connection is successfully established, the data transport entity \nreturns to the FTS Connection id, a unique identifier. The FTS uses \nConnection id later to reference the connection just established. If the \ndata transport entity is unable to establish the connection, it notifies \nthe local FTS about the failure.\n\n2. Send (Connection id, message). This primitive is used to send a \nblock of data called message on the connection Connection id.\n\n3. Receive (FTS, message, Connection id). The local FTS uses the \nservice primitive Receive to receive a block of data message from the \nconnection Connection id.\n\n4, Disconnect (Connection id). The FTS uses the Disconnect service \nprimitive to break the connection Connection id. The data transport \nfacility delivers any messages in transit: before the breakup.\n\n5. Abort (Connection id). This primitive is used to abnormally break \nup the connection Connection id. The data transport facility does not \nensure delivery of the messages in transit before the breakup.\n\nA Connectionless service is useful for the transport of single mes- \nsages between FTSs. It ensures that such message exchanges are done \nwith minimal network resources. Establishment and termination of a \nconnection, such as a BX.25 session, require a significant amount of\n\n* A transmitter FTS is a program that establishes connection with a receiver FTS to \ntransfer a sequence of messages. A transmitter FTS can be either the initiator or the \nsource. Similarly, the receiver FTS can be either the source or the destination.\n\nnetwork resources, which is not justified for short message exchanges. \nA Connectionless service must provide the following service primitives:\n\n1. Send (Destination Address, message). This primitive is used to \nsend a block of data called message to the FTS in the computer with \nthe address Destination Address.\n\n2. Receive (message). The local FTS uses this primitive to receive a \nblock of data called message from the data transport entity.\n\nSince the peer-level protocol gives the rules for interaction between \nany pair of FTSs, it is the core of the global coordination functions. \nThe rules and the formats specified in this section must be uniformly \nimplemented in all FTSs. |\n\nAll fields in the peer-level messages have been encoded using the \nInternational Organization for Standardization (ISO) communication \nheading format standard.**\u00ae Figure 2 shows a general message format. \nThe first field, the message type field, identifies the message type; it \nis one byte long. Encodings for various message types are given in \nTable I. The first field is followed by one or more parameter fields, \neach of which carries a parameter, such as a file descriptor. The first \nbyte uniquely identifies the parameter type, such as file descriptor. \nCodes for various types of parameters are given in Ref. 2. Each \nparameter field can be of variable length, with a maximum of 255 \nbytes. The list of parameters is followed by an end of heading field, \nH\u201980\u2019.* The last field contains data for a file data message. |\n\n*H stands for hexadecimal. H \u2019xx\u2019 is used to represent a two-digit hexadecimal \nnumber.\n\nMessage \nMessage Message Type \nType Name Code \nCommand Authenticated COPY H\u201924\u2019 \ncommand, CAC \nNegotiation H\u201931\u2019 \npackage, CNEG \nCheckpoint \u2018H\u201932\u2019 \ncommand, CCM \nInquiry about H\u201933\u2019 \ncheckpoint command, CIC \nPositive confirmation H\u201934\u2019 \nmessage, CPC \nNegative confirmation H\u201935\u2019 \nmessage, CNC \nInquiry command, H\u201936\u00b0 \nCIO \nCancel command, CCC H\u201937\u2019 \nInterrupt command, CIN H\u201938\u2019 \nRestart command, CRS H\u201939\u2019 \nResponses Acknowledgment, RAC H\u201941\u2019 \nReply negotiation H\u201942\u2019 \npackage, RRN | \nCheckpoint response, H\u201943\u2019 \nRCH \nStatus Response, RSR H\u201944\u2019 \nCancel Response, RCR H\u201948\u2019 \nNotification message, RNM H\u201946\u2019 \nData File Header H\u201901\u2019 \nMessages Message, DHM | . \nFile Data H\u201903\u2019\n\nThe rules for message exchanges for the Copy and Status services \nare given in Sections 5.2.1 and 5.2.2, respectively. The rules for the \nCancel service are similar to those for the Status service and are not \ngiven here.\n\ne Transfer of an authenticated copy command from the initiator \nFTS to the source FTS (if the initiator FTS is not the same as \nthe source FT'S),\n\naccess privileges of the user initiating the file transfer and the syntactic \ncorrectness of the copy command. If the user does not have the\n\nnecessary access privileges and the syntax of the copy command is \nincorrect, the initiator FTS returns an error message to the user and \nthe file transfer is abandoned. Otherwise, the initiator ships the copy \ncommand to the source only if the initiator is different from the source \nin a peer-level message called the Authenticated Copy Command \n(CAC). After sending the CAC, the initiator FTS waits for an acknow]- \nedgment from the source FTS. If it does not receive an acknowledg- \nment within a time-out period FT1, the initiator will again send the \nmessage CAC to the initiator. The initiator makes at most Al attempts \nat sending the authenticated copy command.\n\nAfter receiving an authenticated copy command, the source FTS \nsends an acknowledgment to the initiator FTS and then enters the \nnegotiation phase. During the negotiation phase, the source and des- \ntination FTSs negotiate about options for the file transfer, such as \nselection of pre- and postprocessors, the values of the intercheckpoint \ninterval and the window size for checkpointing. Intercheckpoint inter- \nval and window size are defined later in this section. Further details \nabout the negotiation phase are given in Section VI.\n\nIf the negotiation phase is successful, the source and destination \nFTSs enter the file transfer phase. Each file is transferred completely \nbefore the transfer of the next file is initiated. The source FTS sends \nthe following sequence of peer-level messages to the destination FTS \nfor each file transmitted:\n\ne File header message: the file header message carries the name of \u00a9 \nthe file in which the file to be transferred will be stored, the file\u2019s \nattributes, such as code set type and file length,\n\ne One or more checkpoint commands (a checkpoint command is a \npeer-level message which marks a checkpoint during a file trans- \nfer).\n\nThe checkpoint commands are assigned sequence numbers. Since \nthe sequence numbers cannot be infinitely large, a cyclically reusable \nsequence numbering scheme is used. We take the sequence range to\n\nbe 0 to (2\u00b0? \u2014 1), where 2* is the size of the sequence space. The \nnumber of outstanding checkpoint commands is limited to w, the \nwindow size for checkpointing. The checkpoint command and the \ncheckpoint response carry the sequence number of the checkpoint. \nThe source\u2019s reception of a checkpoint response with the sequence \nnumber x means that the destination FT'S has received everything \ncorrectly up to the checkpoint assigned the sequence number x.\n\nEnd of File (EOF) is marked by sending a checkpoint command. \nThe source FTS must receive the checkpoint response to the check- \npoint command indicating EOF before it initiates the transfer of the \nnext file. The checkpoint command indicating EOF for the last ele- \nment in the file transfer request must also indicate whether additional \n\u2018file data will be transferred during the current session.\n\nNow, we give the rules for the Status service. A user can issue a \nstatus command at the initiator FTS to inquire about the status of a \nfile transfer identified by a job number. The initiator FTS checks if it \nhas sent out the authenticated copy command corresponding to the \njob number given in the status command. If it has not, the initiator \nFTS informs the user that preprocessing, file transfer, and post- \nprocessing have not started. Otherwise, the initiator FTS sends an \ninquiry command to the source FTS. The source FTS checks to see if \nit has completed preprocessing and file transfer. If not, it sends a \nstatus message to the initiator FTS, stating the steps it has already \ntaken and the step presently in progress. If the source F TS has already \ncompleted preprocessing and file transfer, then it sends an inquiry \ncommand to the destination FTS. On receiving the inquiry command, \nthe destination FTS sends a status message indicating whether post- \nprocessing has been completed successfully or unsuccessfully. The \nsource FTS combines this information from the destination FTS with \nthe status information it has about preprocessing and file transfer. \nThe source FT'S sends a status message to the initiator FTS. The \nstatus message contains the following information:\n\nA file transfer using the FTP has available a choice of several \noptions. For example, checkpointing may or may not be used during\n\nfile transfer; if checkpointing is used during a file transfer, then several \nparameters, such as the intercheckpoint interval and the window size, \nmust be agreed upon between the source and the destination. These \nparameters will determine the buffer requirements during the file \ntransfer at the source and the destination. Another parameter to be \nchosen is the list of postprocessors to be used at the destination.\n\nBefore the file transfer phase starts, the source FTS and the desti- \nnation FTS must negotiate regarding the parameters to be used during \nthe file transfer and come to an agreement. For this negotiation, a \nthree-way message exchange between the source FTS and the desti- \nnation FTS is used, as shown in Fig. 3. _ :\n\nIn the first step, the source FTS sends a negotiation package to the \n' destination and then waits for a reply negotiation package. The \nnegotiation package contains the following information about the \nparameters to be used during the file transfer: | |\n\ne Block size. This field specifies the size of the file data transported \nin one file data message.\n\ne Intercheckpoint interval. This field specifies in units of bytes the \namount of file data transmitted. between two checkpoints. This \nfield is required only if the checkpointing option is chosen.\n\ne Window size for checkpointing. This field contains the maximum \nnumber of outstanding checkpoint intervals at the source FTS. If \nthis field is blank, then the default value is 1.\n\nIn the second step, the destination FTS examines the options for \nfile transfer after receiving a negotiation package from the source \nFTS. The destination FT'S decides whether it can accept them. If the \ndestination FTS cannot accept some or all options, it selects alterna- \ntives that it can accept. It puts acceptances, rejections, and alternatives \nin a response negotiation package and sends them to the source FTS. \nA reply negotiation package contains the following information:\n\nIn the third step, on receiving a reply negotiation package, the \nsource FTS replies with a positive confirmation message if the desti- \nnation FTS accepted all options or if the source FTS can accept the \nalternatives given in the reply negotiation package. Otherwise, the \nsource FTS will send a negative confirmation message to the desti- \nnation FTS.\n\nIf the source FTS does not receive a reply negotiation package from \nthe destination FTS within a time-our period FT2 after sending a \nnegotiation package, it sends the negotiation package again. If the \nsource FTS does not succeed within A2 attempts, it informs the \ninitiator FTS about the failure and aborts the file transfer.\n\nOne parameter that the source and destination FTSs might negoti- \nate on is the time when the file transfer should be started. The FTP \ndescribed here does not negotiate on the starting time of the file \ntransfer, but it can be easily extended to do so. If the underlying \nnetwork only carries file transfer traffic, then.such negotiation can be \nvery useful.? Otherwise, we feel that it should not be used for the \nfollowing reason. Such negotiation requires that the FTSs should be \nallowed to schedule the future allocation of connections, but this is \nnot consistent with the OSI reference model approach.\u201d Several appli- \ncation layer protocols must share the same network resources. Sched- \nuling of the connection allocation, if any, must be done by the lower \nlayers, not by an application layer protocol, such as the FTP.\n\nFor networks that carry file transfers exclusively, negotiation about \nthe starting time can be very useful. File transfers typically require a \nheavy commitment from the network. A connection used for a file \ntransfer typically utilizes nearly 100 percent of its maximum transfer \nrate as defined by its throughput class. On the other hand, a connection \ncarrying interactive data utilizes only 1 to 2 percent of its maximum \ntransfer rate. As a result, the number of outgoing and incoming \nconnections used for file transfers is very small. Scheduling of connec- \ntions and file transfers to optimize a cost function, such as network \nutilization, or minimizing delay can be very useful.\n\nWe give two examples to illustrate the use of the FTP: one for the \nCopy service and one for the Status service. \n- In the first example (Fig. 4), a user issues a copy command to the \nfile transfer system in Computer-C to copy a file FSOU from Com- \nputer-A to the file FDEST in Computer-B. The file transfer system \nin Computer-C (henceforth, referred to as FTS-C) checks the com-\n\n@ AUTHENTICATED COPY FROM FTS-C TO FTS-A \n@ NOTIFICATION MESSAGE FROM FTS-A TO FTS-C\n\nmand for its syntactic correctness and checks the access privileges of \nthe user to find out whether the user is allowed to initiate this file \ntransfer.\n\nIf the copy command has a syntax error, then FTS-C returns an \nerror message to the user. If the user has the proper access privileges, \nthen FTS-C returns a job number to the user. The job number is a \nunique identifier assigned to the copy command. It uniquely identifies \nthe command throughout the network. The format for the job number \nis given earlier in Section III.\n\nFTS-C sends an authenticated copy command to the FTS in Com- \nputer-A (henceforth, referred to as FTS-A). Then, it waits for an \nacknowledgment from FTS-A. If FTS-C does not receive an acknowl- \nedgment within the time-out period FT1, then it retransmits the\n\nFTS-A sends a file header message to FTS-B. A file header message \ncontains the information such as the file length and the file name. \nThen, FTS-A processes the file data using the preprocessors and \ndivides it into blocks. Each block is packed into a separate file data \nmessage. Then, FTS-A sends these file data messages to FTS-B. It \nalso sends the checkpoint commands at every intercheckpoint interval. \nIf the session being used for the file transfer breaks down, then FTS- \nA can resume file transfer from an intermediate checkpoint up to \nwhich FTS-B has received the data correctly. FT'S-B unpacks the file \ndata in the messages received. Then, it processes the file data received \nusing the postprocessors agreed upon during the negotiation ees \nFTS-B.-then stores the file data in the file FDEST.\n\nAfter the successful file transfer, FTS-B sends a notification mes- \nsage to FTS-A informing it about the successful completion. FTS-A, \nin turn, sends a notification message to FTS-C informing it about the \nsuccessful completion. Finally, FTS-C informs the user about the \nsuccessful completion of the file transfer.\n\ne Similar information about file transfer and postprocessing. The \nstatus can further include additional information, such as how far \nthe file transfer has progressed.\n\nFTS-C checks the status command for its syntactic correctness. If \nthe command has a syntax error, then FTS-C returns an error message \nto the user. Then, FTS-C checks the access privileges of the user to \ndetermine whether the user can inquire about the file transfer. If the \nuser has the necessary privileges, then FTS-C checks the local records \nto find out whether the file transfer has been completed. Other- \nwise, FTS-C sends a status command to the destination FTS, FTS-B. \nFTS-B checks the local records to locate the status of the file transfer. \nIt checks whether it was ever involved in that file transfer. If it was \ninvolved in the file transfer, FTS-B checks whether the file transfer\n\nhas been completed or if it is still in progress. FTS-B sends the status \ninformation in a status response message to FTS-A. FTS-A also \ncollects the local information about the file transfer and combines it \nwith the information received from the destination FTS, FTS-B. Then, \nit sends the status information to FTS-C. Finally, FTS-C delivers the \nstatus information to the user.\n\ndescription of the protocol. In this section, we first present an informal \ndiscussion of the s/r model; the reader is referred to the references for \nmore details. We then discuss how the FTP was formally specified \nusing this approach.\n\nThe selection/resolution model is a method in which a complex \nsystem such as a protocol can be represented as a set of coordinating \nfinite state machines. Each component finite state machine, called a \nprocess, is described by an appropriately labeled directed graph. For \ninstance, suppose we wished to describe the coordination of two \nprocesses A and B that can be executing segments of code in either a \nnoncritical section or a critical section. We wish the coordination to. \nbe such that both processes are not simultaneously in their critical \nsections. Figure 6 has two labeled graphs representing the processes.\n\nIn the labeled graphs of Fig. 6, the vertices represent states of the \nprocess, and the edges represent possible single-step transitions. Thus, \nprocess A (also B) can be in the states NCS (Noncritical Section), \nTRY (trying to enter the critical section), and CS (Critical Section). \nA state encapsulates the relevant past of the process needed to deter- \nmine future behavior and is known only to that process. The processes \ncoordinate by exchanging information about their future intentions. \n~ That is, they communicate their selections (intentions) to all the other \nprocesses. For example, in state NCS, process A can only choose the \nselection ok, while in state TRY it can nondeterministically choose \nbetween head and tail. Note that a process can make only one selection \nat any time. Selections of a process are given in curly braces next to \nthe state.\n\nEach edge has a label next to it. The edge labels are Boolean \nconditions based on the selections of all the processes, and determine \nthe possible transitions of a process. In conditions, + is the same as a \nlogical or and * is the same as a logical and. For instance, the condition \nfor process A to go from CS to NCS is simply (A:nok), that is, it really \ndoesn\u2019t depend on B, and is in fact always true since the only selection \nof A in state CS is nok. However, the condition for A to enter the CS \nstate from the state TRY is [(B:ok) + (A:head)*(B:head) + \n(A:tail)*(B:tail)]. In simple words, the condition says that A can go to \nthe CS state if B selects ok, or if A and B both select head, or if A and \nB both select tail. The transition of process A from state TRY to its \nnext state clearly depends on what B is selecting. Both processes \nessentially toss coins to determine who wins if they are both trying to \nenter the critical section. The transition of a process to one of the \npossible next states (there may be more than one transition enabled) \nis called resolution.\n\nIn the above example, developed in more detail in Ref. 11, both \u00a9 \nprocesses initially start in the state NCS at time-step 0. Then, in each \nunit of time, the processes first make a selection based on the state \nthey are in, and then they resolve (transition to a new state) based on \nthe selection of all the processes. It can be seen that the processes will \nnever both be in the states CS at any time step.\n\nThe s/r model gives precise algebraic descriptions of processes that \ncan be combined to give a precise description of the entire system. \nFurthermore, many computational aspects of this model can be auto- \nmated. In fact, the formal specification can be used to test the \ncorrectness of the FTP by conducting a reachability analysis using the \nprocedure given in Refs. 6 and 7.\n\nThe formal specification of the FTP consists of descriptions of 45 \nprocesses divided for convenience into 9 clusters (see Fig. 7). A cluster\n\nis a convenient grouping of a set of processes. We have grouped \nprocesses into clusters based on their functions. Thus, there are \nclusters that correspond to the basic service functions such as Copy \nservice, Status service, and Cancel service. There are also clusters that \ncorrespond to interface functions such as interfacing with the user, \nthe data transfer facility, and the file system. Finally, there is a main \ncontroller cluster that acts as an overall coordinator, and there are \nclusters for the timers and the variables.\n\nEach process is defined in a language that essentially provides the \nsame information as the labeled graph corresponding to that process. \nFor example, we can define a process that does a syntax check on the \nuser command. We model this as the process syntc shown in Fig. 8. \nAs discussed in Section III, the exact choice of the command syntax \nis a local decision. Notice that we model only the portion of the process \nbehavior that describes global coordination with other processes. The \nprocess syntc makes the selection ok, only if the command contains \nall the required information and is syntactically correct. Otherwise, it \nselects nok. ?\n\nIn this figure we have not indicated the conditions for the self- \n_loops.* We shall assume that the self-loop condition is like an other- \nwise; that is, it is the negation of the or of the conditions on the other | \nedges. The process syntc is defined using the specification language in \nFig. 9. Notice the close correspondence with the graphical specifica- \ntion. We use $ as shorthand to signify the current state. In this\n\nIDLE {idle, nok, ok} \n> OK :((user:copy) + (user:status) + \n(user:cancel)) * (Syntc:ok)) \n=~ NOK :((user:copy) + (user:status) + \n(user:cancel))*(syntc:nok)) \n>s = \u00a2 \nOK {ok} \n> IDLE :(usit:idle) \n>Ss soe. \nNOK {nok} \n> IDLE :(usit:idle) \n>$ 33\n\nexample, the empty condition on the self-loop is equivalent to other- \nwise.\n\nProcess syntc is one of the processes in the User Interface Cluster. \nThis cluster consists of four processes: usit, syntc, secur, and jobnum. \nEach process\u2019s description is on the average about as complex as that \nof syntc. The process usit gets into one of the three states\u2014COPY, \nCANCKL, and STATUS\u2014if the user command is syntactically correct \nand if the user has the permission to execute the command. The \nprocess secur is similar to syntc; it checks the security privileges of the \nuser. It makes the selection (secur:ok) if the user can execute the \ncommand. Otherwise it makes the selection (secur:nok). The process \njobnum generates a job number for each copy command. The job \nnumber uniquely identifies a copy command throughout the network.\n\nThe full formal specification (about 50 pages long) consists of similar \ndescriptions of all the clusters. The descriptions of the processes, of \ncourse, vary in complexity. The formal specification can be used as a \nprecise guideline by FTP implementors. In fact, the high-level design\n\nof the FTP implementation may be readily derived from the formal \ndescription. A development group at AT&T Communications used the \nformal specification in their design process. They found it to be \nextremely useful in developing a high-level C-like language design of \nFTP.\n\nFor implementation, each user command can be viewed as invoking \nan independent copy of the 45 processes. If an FTS must handle \nseveral file transfer commands at the same time, then an independent \ncopy of the processes is invoked for each command. Since our FTP is \na three-party protocol, a file transfer at an FT'S can be initiated from \na remote FTS. For a file transfer initiated from a remote FTS, an \nindependent copy of the processes is invoked at the local FTS. The \nFTS can thus be implemented as a reentrant program that can be \nused independently by each command.\n\nAs discussed previously, our approach here is to separate the file \ntransfer issues from local service functions. This allows great flexibility \nin local matters and avoids proscribing rigid formats for the user \ninterface, for file management and naming, and for presentation \nfunctions such as code sets. Our FTP differs in this regard from other \nfile transfer protocols that generally are rather inflexible and less \nadaptable. For example, none of the other file transfer protocols has \na general mechanism such as pre- and post-processors that allow \nadditional capabilities to be incorporated in a standard way. In this \nsection, we compare the FTP with four file-transfer protocols: Arpanet \nFTP,\u201d Network Independent (NI) FTP,\u2019*\u201d* Autodin II FTP,\u2019\u00ae and \nthe International Organization for Standardization (ISO) File Service \nProtocol.!\u2019\n\nThe Arpanet FTP requires the user (initiator) to establish separate \ncommand and data connections with both the source and the desti- \nnation of the file transfer. The user must keep track of these connec- \ntions. The user is also involved at too detailed a level with the file \ntransfer operation; it is not possible to simply copy a set of files from \nthe source to the destination with a single macro command. It is \nexpected that the user remain on-line at each phase of the transfer, \nand the user must be aware of details such as the socket number of \nthe data connection. The protocol does provide for both copy and \nappend services, as well as status notification. File management serv- \nices such as creating and deleting files are also available. Security \nmeasures are fixed and encompass only user name, password, and \naccount number. There is no provision for adding other local security \noptions.\n\nThe NI FTP is a two-party protocol (no provision is made for third \nparty initiation of a file transfer) that divides the transfer operation \ninto three phases. The protocol handles only transfers of single se- \nquential files; multiple files must be sent as separate file transfers. \nThe NI FTP protocol provides for a high degree of flexibility in the \nactual data transfer operation. This is accomplished through the \nnegotiation of a large number of possible alternatives. The possible \nalternatives are defined in any particular implementation, and cannot \nbe readily modified. There is a very limited file management capability \nin NI FTP. The protocol assumes only minimal support from the \nlower layers, and specifies presentation services. It allows for modular \nimplementation, but does not have provisions for easily adapting to \nnew requirements. The protocol provides for only foreground file \ntransfers.\n\nThe Autodin II FTP provides high- level service primitives and \nallows background file transfers. However, it goes too deeply into \ndefining the structure of a file at the network level; this results in the \nnegotiation of a large number of parameters for the actual data \ntransfer, rather than allowing this to be handled through local proc- \nessing. Furthermore, each of the three parties involved in the file \ntransfer must keep an elaborate list of parameters that define all the \noptions that have been agreed upon. There is an extensive security \nmechanism, but it is not modifiable. Similarly, the user interface is \nvery specific, and cannot be adjusted to local preferences. Autodin II \nFTP is definitely comprehensive but all the detail is defined at the \nnetwork level. This results in a protocol that requires a a complex \nimplementation at each. node. |\n\nA committee of the ISO is working on the File Access, File Transfer, \nand File Management (FTAM) services and on a corresponding pro- \ntocol. The FTAM services allow a user to access and transfer a remote \nfile. They also allow a user to manipulate a file in a remote file system. \nAn example of a service is a command to open a file in a remote file \nsystem. These are micro services. In contrast, our FTP provides macro \nservices, such as the copy service. Each service of our FTP can be \nprovided by a long sequence of the ISO service calls. We have carried \nout a study to define these sequences of the ISO service calls.'\u00ae\n\nThe ISO file service uses the concept of the Virtual File Store. The \nVirtual File Store is a common model for describing file names and \ntheir attributes. Different file systems have a wide range of styles for \ndescribing the storage of data and the means by which it can be \naccessed. The Virtual File Store allows the differences in style and \nspecification to be absorbed in a local mapping function. The Virtual \nFile Store attempts to encompass all possible variations of file defi-\n\nnition, resulting in an excessively detailed description of a file that \ncannot be easily modified in the future. \u2014\n\nIn the FTP presented here, we separated global functions requiring \nclose coordination (such as parameter negotiation) from functions that \ncould be done locally at each individual node. In an FTP implemen- \ntation, the global functions must be implemented. Local functions can \nbe implemented depending on user needs. As a result, a minimal \nimplementation of the protocol is fairly simple. Furthermore, it re- \nquires only incremental efforts to extend the minimal implementation.\n\nWe feel that the FTP is flexible, adaptable, and powerful enough to \nmeet most file transfer requirements. Furthermore, it has been speci- \nfied precisely enough so as to make implementation a relatively \nstraightforward task.\n\n1. M. E. Grzelakowski, J. H. Campbell, and M. R. Dubman, \u201cDMERT Operating \nSystem,\u201d B.S.T.J. 62, No. 1, Part 2 (January 1983), pp. 303-22.\n\nOperations Systems Network Protocol Specification: BX.25 Issue 3 Addendum-A, \nAT&T Bell Laboratories, Holmdel, NJ, August 1983.\n\nOperations Systems Network Protocol Specification: BX.25 Issue 3, AT&T Bell \nLaboratories, Holmdel, NJ, June 1982.\n\n. A. S. Tanenbaum, Computer Networks, Englewood Cliffs, N.J.: Prentice-Hall, 1981, \nChapter 1, pages 10-21.\n\n. S. Aggarwal, R. P. Kurshan, and K. Sabnani, \u201cA Calculus for Protocol Specification \nand Validation.\u201d In Protocol Specification, \"Testing, and Verification, III, edited by \nH. Rudin and C. H. West, Amsterdam: North-Holland, 1983.\n\n10. S. Aggarwal, R. P. Kurshan, and harma, \u201cA Language for the Specification \nand Analysis of Protocols,\u201d In aise Specification, Testing, and Verification, \nIIT, edited by H. Rudin and C. H. West, Amsterdam: North-Holland, 1983.\n\n11. S. Aggarwal and C. Courcoubetis, \u201cDistributed Implementation of a Model of \nCommunication and Computation,\u201d Proc. 18th Hawaii Int. Conf. on System \nSciences, January 1985, pp 206-218.\n\n12. N. J. Neigus, \u201cFile Transfer Protocol for the ARPA Network.\u201d In ARPANET \nProtocol Handbook, edited by E. Feinler and J. Postel, Defense Communications \nAgency, 1978.\n\n13. C. J. Bennet and D. N. Frost, \u201cNetwork Independent File Transfer,\u201d Tech. Report, \nUniversity College, London.\n\n14. R. W. S. Hale, \u201cFile Transfer Protocols\u2014Comparison and Critique,\u201d NPL Report \nDNACS 48/81, Teddington, United Kingdom, 1981.\n\n15. High Level Protocols Group, \u201cA Network Independent File Transfer Protocol,\u201d \nHLP/CP(78), NPL, Teddington, United Kingdom, 1977. |\n\n16: H. C. Forsdick, \u201cAutodin II File Transfer Protocol,\u201d Bolt Beranek and Newman \nInc., Report No. 4246, Boston, 1980.\n\n17. Second Draft Proposal on File Transfer, Access, and Management, ISO/TC97/SC21 \nNo. 8571, 1985. (Available from ANSI)\n\nSudhir Aggarwal, B.S. (Mathematics), 1969, Stanford University; M.S. \n(Mathematics), 1971, and Ph.D. (Computer and Communication Sciences), \n1975, University of Michigan; Assistant Professor of Computer Science, Uni- \nversity of Oregon, 1975-1976; Research Scientist, Lawrence Livermore Labo- \nratory, 1976-1977; Assistant and Associate Professor of Mathematics, Univer- \nsity of California, Riverside, 1977-1982; AT&T Bell Laboratories, 1982\u2014. Mr. \nAggarwal is a member of the Mathematical Sciences Research Center. His \nresearch interests are computer communication protocols, local area networks, \nand modelling and simulation.\n\nB. Gopinath, M.Sc. (Mathematics), 1964, University of Bombay; Ph.D. \n(E.E.), 1968, Stanford University; Research Associate, Stanford University, \n1967-1968; Alexander von Humboldt Research Fellow, University of Gottin- \ngen, 1971-1972; Gordon McKay Professor, University of California, Berkeley, \n1980-1981; AT&T Bell Laboratories, 1968-1983; Bell Communications Re- \nsearch, 1983\u2014. Mr. Gopinath is Division Manager for Systems Principles \nResearch, and is engaged in research in communications and computer science.\n\nKrishan Sabnani, B. Tech. (E.E.), Indian Institute of Technology, New \nDelhi; Ph.D. (E.E.), Columbia University, New York, 1982; fellowship, School \nof Engineering and Applied Sciences, Columbia University, 1977; research \nassistant, Columbia University, 1978 to 1979; he was employed by RCA, \nPrinceton, 1979-1981, AT&T Bell Laboratories, 1981\u2014. Mr Sabnani\u2019s current \ninterests are computer communication protocols and fault-tolerant computing. \nMember, Sigma Xi, Eta Kappa Nu, and Epsilon Pi Upsilon.\n\nAutomated protocol validation tools are by necessity often based on some \nform of symbolic execution. The complexity of the analysis problem however \nimposes restrictions on the scope of these tools. The paper studies the nature \nof these restrictions and explicitly addresses the problem of finding errors in \ndata communication protocols of which the size precludes analysis by tradi- \ntional means. The protocol tracing method described here allows one to locate \ndesign errors in protocols relatively quickly by probing a partial state space. \nThis scatter searching method was implemented in a portable program called \nTrace. Specifications for the tracer are written in a higher-level language and \nare compiled into a minimized finite state machine model, which is then used \nto perform either partial or exhaustive symbolic executions. The user of the \ntracer can control the scope of each search. The tracer can be used as a fast \ndebugging tool but also, depending on the complexity of the protocol being \nanalyzed, as a slower and rather naive correctness prover. The specifications \ndefine the control flow of the protocol and may formalize correctness criteria \nin assertion primitives.\n\nProtocol validation by symbolic execution is inherently a time- and \nspace-consuming task. For lack of better methods, though, many \nautomated protocol validation tools do use symbolic execution algo- \nrithms,!~\u00b0 and even methods based on validation algebras such as CCS\u00b0 \nor PVA\u2019\u00ae still implicitly formalize symbolic executions.\u2019 Unfortu-\n\n* AT&T Bell Laboratories. \nT Cf. the expansion theorem in CCS and the shuffle operator in PVA.\n\nCopyright \u00a9 1985 AT&T. Photo reproduction for noncommercial use is permitted with- \nout payment of royalty provided that each reproduction is done without alteration and \nthat the Journal reference and copyright notice are included on the first page. The title \nand abstract, but no other portions, of this paper may be copied or distributed royalty \n. free by computer-based and other information-service systems without further permis- \nsion. Permission to reproduce or republish any other. portion of this paper must be \nobtained from the Editor.\n\nnately, the assumption that a computer will always be able to take \nover when the complexity of a complete analysis surpasses our ability \nto perform algebraic expansions by hand is decidedly wrong.?\u201d\u00b0\n\nA protocol of a realistic size can generate a state space of in the \norder of 10\u00b0 system states and up. As little as adding one single \nmessage type, one protocol variable, or one slot to the message queues \ncan further expand the number of reachable system states by orders \nof magnitude. For a protocol of this size a symbolic execution algorithm \ncan at best analyze in the order of 10 to 100 system states per second \nof CPU time if the state space is built in core.\"! To analyze 10\u00b0 states \nexhaustively would then take at least 115 days of computation. Fur- \nthermore, assuming that each state can be encoded in no more than \n10 to 100 bytes, storing a state space of this size would still require a \nmachine with several gigabytes of main memory.\n\nSo, if this appears to be infeasible, what is the best that can be \ndone? In the design phase one would like to have tools that can trace \nthe most glaring bugs in a protocol in no more than a few seconds of \nreal time. The completeness of an analysis is not really at issue here; \nspeed is. To find more subtle design errors of a completed protocol one \nmay be willing to spend more time, but not much more than perhaps \n10 hours or in the order of 10\u00b0 seconds of CPU time. For symbolic \nexecution algorithms, this requirement sets an upper limit to the \nnumber of states that can be searched at roughly 10\u2019 states. At 10 to \n100 bytes per state, however, we cannot expect to do anything useful \nwith a state space of more than in the order of 10\u00b0 states. Therefore, \nit is preferable that the tracer be able to perform complete analyses in \nsmall state spaces holding just a fraction of the total number of states. \nIn the remainder of this paper we concentrate on these two issues: the \neffectiveness of partial searches and the possibility of performing \ncomplete searches in partial state spaces.\n\nIn the following discussion we assume that the protocol submitted \nto a tracer is likely to contain errors and that a designer is interested \nin seeing any nonempty subset of these. A protocol tracer may, for \ninstance, scan the state space in an effort to quickly discover typical \nviolations of user-specified correctness requirements. It is important \nto note that the objective of such a partial analysis, or scatter search \nas we Shall call it, is to establish the presence rather than the absence \nof errors.\n\nWhat we are aiming for is a protocol tracing method that allows us \nto spend a small fraction of the time required by an exhaustive analysis \nto find a substantial portion of all design errors. The emphasis is on \nspeed, not on completeness. If a protocol contains an error, an ex- \nhaustive search would meticulously report every possible circumstance\n\nunder which the error could make the protocol fail. For our purposes, \ntracing a single variant of the error in a partial search will suffice.\n\nSection II explains how a general symbolic execution algorithm \nbased on depth-first search can be organized. It discusses a variant of \nsymbolic execution called scatter searching and compares its perform- \nance with exhaustive searching. Section III discusses in more detail \nheuristics that can be used to perform a partial search, and Section \nIV shows how depth-first searches can be organized in incomplete \nstate spaces. Section V shows how protocol specific correctness criteria \ncan be verified with a standard symbolic execution algorithm. Section \nVI gives a small example of the use of correctness assertions in tracing \nbugs in a protocol. A larger example is presented in the Appendix. \nSection VII summarizes the main results.\n\nIn this section we discuss some experiments with a program called \nTrace, which performs a simple depth-first search in a partial state \nspace generated by a set of interacting finite state machines, where \nthe state space is maintained as a tree of system states. To determine \nthe effectiveness of partial searches, the performance of exhaustive \nsearches and scatter searches was compared, using a search depth \nrestriction as a parameter. But, first let us consider the working of the \ntracer in a little more detail. |\n\nWith the exhaustive tracing method a state space tree is searched \nstarting from the initial system state, exploring every possible execu- \ntion path until an end state, a previous state, or an error state is \nreached, or until the search depth limit is encountered. A return to a \npreviously analyzed state terminates the search under two conditions:\n\n1. If the previously analyzed state is in the execution path that \nleads from the root of the state space tree to the current state, or\n\n2. If the previously analyzed state was encountered elsewhere in the \nstate space tree either at the same depth or closer to the root of the \ntree than the current state.\n\nIn the first case the tracer has discovered an execution loop in the \nprotocol. The loop could be checked further on liveness, but to save \ntime the program Trace simply checks that its repetition does not \nviolate the user-specified correctness criteria and continues. In the \nsecond case the subtree that would be explored by continuing the \nsearch down to the search depth restriction would be contained in the \nsubtree of the previously analyzed state, and cannot lead to new results. \nThe tracer can therefore ignore the subtree and continue exploring \nnew leaves in the tree. )\n\ndescribed in, for instance, Ref. 1, the design of the experimental tracer \n_ go far is fairly standard. The exhaustive trace method, however, can \nbe considered to be a special case of the scatter search. In a scatter \nsearch not every possible execution sequence is explored. The tracer \nmakes an estimate of the likelihood that exploring a new sequence can \nlead to the discovery of a new error, and will search only those \nsequences that optimize its chances of finding the largest set of unique \nerrors in the smallest amount of time. The tracer\u2019s estimate will be \nbased on a heuristic that should be general enough to work on any \ntype of protocol. One straightforward way to do this, for instance, is \nto restrict the amount of nondeterminism that will be taken into \naccount by the tracer. These and other techniques will be discussed in \nmore detail in Section ITI. |\n\nTo test the performance of a partial search, we want to compare its \ncoverage or \u201cscope\u201d with that of an exhaustive search. The test \nprotocol chosen for these comparisons was large enough to show the \nnecessity of partial searches and also to give some room for experi- \nmenting with different flavors of partial searches. However, the size \nof the test case precluded, by the nature of the problem, the compila- \ntion of a definitive list of \u201call\u201d errors for reference. As a measure of \nthe scope of the scatter search we will therefore take the number of \nerrors traced and compare it with the number of errors traced by an \nexhaustive search method.\n\nprotocol, generating a state space in the order of 10\u00b0 system states.\" \nThe protocol was analyzed several times, both for the exhaustive \nsearch and the scatter search, with a search depth restriction that was \nincremented in steps of 10 levels for each new analysis run.\n\nAn exhaustive search for this protocol became infeasible beyond a \ndepth of 80 levels, that is, numbers of states down from the root of \nthe state space tree. The tree scanned by the scatter search method \nhad a maximum depth of 189 steps. Setting the search depth restriction \nbeyond 189, therefore, no longer affects the scope of the analysis. To \nillustrate this, the curve for the scatter search was continued in Fig. 1 \nup to a depth of 230. The longest scatter search required less than 4 \nminutes of CPU time to complete. The run time of the exhaustive \nsearch tends to be exponential in the search depth. Using Fig. 1, it \ncan be estimated that searching the state space tree down to the same \ndepth (189 steps) with the exhaustive search would take some 3000 \nyears of CPU time. |\n\nFortunately, the test protocol analyzed contained a generous number \nof design errors. No attempt was made to classify them. In Fig. 2a the \nnumber of deadlocks reported by the tracer is shown as a function of \nthe search depth, and in Fig. 2b the number of deadlocks versus the \ntime it took to find them is plotted on logarithmic scales.\n\nNo deadlocks are found at search depths 10, 20, and 30. The first \nerror is reported with the scatter search for a search depth of 40 steps, \nrequiring 4 seconds of CPU time. For the same search depth restriction \nthe exhaustive search reports the first 3 errors in 6 minutes. By \nrepeating the analyses for intermediate levels between 30 and 40, we \nfound that the first error is reported both in the scatter and the \nexhaustive search mode at level 35, requiring 4 seconds for the former \nand 3 minutes for the latter search. The two intermediate tests were \nincluded in the results shown in Fig. 2.\n\nVery probably, no protocol designer would be interested in tracing \nthis protocol beyond the first 100 error sequences generated. For the \ngiven test case this would mean that with an exhaustive search the\n\nfirst 70 steps in state space can be searched in roughly 3 hours of CPU \ntime. Alternatively, the first 100 steps can be traced with a scatter \nsearch in only 30 seconds of CPU time. |\n\nNote that the time required to find the first error, the minimum \nsearch depth required to trace it, and the relation between search \ndepth and the number of errors reported are favorable for the scatter \nsearch method. |\n\nThe protocol used for these tests requires roughly 40 bytes in the \nstate space per system state. A total of 332,527 system states is \ngenerated in the longest exhaustive search analysis performed. As a \nresult, for every new state generated, in the exhaustive search a data \nbase of up to 15 megabytes must be probed for a state match. Even \nwith the best hashing methods, this is bound to slow down the analysis \nnoticeably. In the scatter search the largest number of states seen is \n172,402 at a depth of 189 in the tree, corresponding to a database of 8 \nmegabytes. The scatter search therefore should slow down less rapidly. \nThis effect is illustrated in Fig. 3. The time efficiency is expressed in \nthe average number of states analyzed per second for each analysis \nrun.\n\nThe steep left-hand side of the curves can be attributed to the \noverh\u00e9ad involved in the setup of a state space, which is felt more if \nthe number of states explored is small. With the current tracer, the \noptimum speed for both search methods is reached when the state \nspace contains roughly 1000 states.\n\nIt is relatively straightforward to give preference to the shortest \ncomplete execution sequences and to defer analysis for longer se- \nquences. We have already used this method in the preparation of the \nfigures above by bounding the depth of the tree explored during a \nsearch. In this section we consider some other partial search heuristics.\n\nA method for reducing the run time of an analysis effectively is to \nrestrict the amount of nondeterminism in the protocol model. In an \nexhaustive search, each node in the state space tree is root to one \nsubtree for each executable option in each finite state machine in the \nprotocol. Not all interleavings of these actions are necessarily relevant. \nConsider two executable actions: one action a local to machine M1, \nfor example, an assignment to a local variable,.and the other an \nexternal action b in machine M2, for example, a send or a receive. \nThere are two possible orders in which these two actions could be \nexecuted, corresponding to the two sequences:\n\nEach of the two sequences leads into a new state that forms the root \nof an entire subtree in the state space. The question is of whether or \nnot the two subtrees are equivalent with respect to the errors to be \ntraced. Note that the execution of internal action a will not change \n_ the environment for the remote machine M 2, so neither the executa- \nbility nor the result of b can be any different when a is executed first \nor last. Similarly, the execution of a is independent of the environment \naffected by M2 and also its executability and result is independent of \nwhether b preceded it or not. In this case, then, it suffices to search \none of the two possible interleavings and to ignore the other. Unfor- \ntunately, there are not many cases where a complete subtree can be \nignored without restricting the scope of an analysis. In some cases, \nthough, we can predict in what way the scope will be affected. It would \nbe unwise to restrict the nondeterminism that is local to a finite state \nmachine, as shown in the following Argos fragment:\u201d\n\nArgos is a CSP-like\u2019\u00ae guarded command language\u201d\u2122 defined on buffered \nmessage channels. A detailed discussion of the language itself can be \nfound in Ref. 12. In the above example a and B are channel names \n(bounded buffers declared elsewhere), one and two are message names,\n\nand P and Q are procedure names. If message one \u2018is the first message \nin A and message two is the first message in B, both input statements \nare executable and the process executing the above fragment can make \na nondeterministic choice between the two alternatives, and then \nproceed with the execution of either P( ) or Q( ). The protocol tracer \ncannot foresee which of the two alternatives may produce an error \nwithout executing them. Note that ignoring one of the two alternatives \nin an analysis implies ignoring a potentially important code fragment, \nthat is, either P( ) or Q( ), without having reason to assume that this \ncode would be error free. In this case then both alternatives will have \nto be explored. The situation is different for the nondeterminism that \nresults from concurrency, as illustrated by the following Argos frag- \nment: ,\n\nIt defines two processes P 1 and P2. Assuming that both initial actions \nare executable, it must be decided in what order they will be executed \nby the tracer. This time it may, but it will not always make a difference \nin what order these two I/O statements are executed. In an exhaustive \nsearch both orders are always analyzed. Ignoring one of the two \npossible orders, however, can halve the amount of work to be done for \nthis node in return for the chance that it will cause the tracer to miss \nerror sequences. No code fragments are ignored here, only a potentially \u2014 \nerroneous timing of executions. Fortunately, not all orderings have \nthe same probability of leading into error states. For instance, if we \nare primarily interested in finding deadlock states, that is, states in \nwhich all message channels are empty and not all processes have \nreached their end states, we may choose to explore the sequence \nstarting with a receive action and ignore the other. In practice this \nheuristic performs remarkably well, as illustrated by the results dis- \ncussed earlier. |\n\nIf at some node in the state space tree there are N concurrent \nprocesses, all executable, the tracer can decide to ignore any M <= N \nof the processes to reduce the search. In the tests reported in Figs. 1 \nto 3 we set M = 1 for the scatter search and M = N for the exhaustive \nsearch. In the case where M = 1 the search heuristic is implemented \nas a priority scheme that determines which process should be executed \nnext. Highest priority is given to internal actions. At the next level we \nplace receive actions, since these tend to bring the system closer to a \ndeadlock state with empty channels. A lower priority is given to send \nactions, and a lower priority still to channel time-outs. Time-outs are\n\ngiven lowest priority in the partial searches since they tend to create \nmany spurious error reports. In partial search mode the correct work- \ning of the time-out mechanism is assumed, that is, a time-out is only \nconsidered to be enabled when there is no other option to continue \nthe protocol. Though this definitely reduces the scope of an analysis, \nit does allow us to trace for another class of errors first and defer the \ncostly tracing of timing errors.\n\nThe capacity of a communication channel for holding messages can \nalso have an important effect on the size of a state space. In the \nspecification language Argos the channels are modeled by finite \nqueues. A channel then can be in only a finite number of states\n\nwhere N is the number of slots in the channel (i.e., the queue size), \nand S is the size of the channel sort, that is, the set of all messages \nthat can be recognized by the channel. Reducing the number of slots \nN by 1 can reduce the size of the state space and speed up the analysis \nby a factor of up to\n\nIn the scatter searches of Figs. 1 to 3 the queue sizes were restricted \nto two slots. To study the effect of a variation of the queue size the \ntests were repeated for a small range of sizes. Figure 4 shows the effect\n\nof a variation in the number of slots between one and four for both \nthe exhaustive and the scatter search.\n\nIn the Introduction we mentioned that the tracer should be able to \nperform searches in even incomplete state spaces since the size of a \ncomplete state space generally precludes its storage or even its usage \nduring the search. In this section we show how this can be accom- \nplished.\n\nFirst it should be noted that in a depth-first search, at each execution \nstep only those states that lead from initial state to the current state \nare indispensable in the state space. The presence of these states is \nnecessary for the detection of system execution loops. Not every \nsystem state, though, can be found at the start of such an execution \nloop, and therefore it is not necessary to remember each state along a \nsingle execution path. The only states that must be remembered are > \nthose in which at least one of the interacting finite state machines is \nat the start of a local execution loop. Figure 5a shows a small but \nconsistent reduction in the numbers of states if we restrict the state \nspace to such \u201cloop states.\u201d\n\nSince the analysis is performed on finite state machines we can try \nto minimize the machines in an effort to reduce time or space com- \nplexity without affecting the scope of an analysis. The machines \ncannot be reduced under the standard notion of language equivalence, \nsince that will change the behavior or the protocol. A stronger notion \nof state equivalence,\u2019 similar to that defined in CCS\u00ae can be used.\n\nFigure 5b compares the analyses of minimized state machines and \nnonminimized state machines. The protocol tested defines 4 processes, \n34 message types, 6 message channels, and 3 local variables. The state \nmachines generated for the processes contain 69, 47, 7, and 5 states, \nrespectively. The strongly equivalent minimized machines contain 35,\n\n31, 7, and 5 states. As it turns out, the number of states generated in \nthe state space is roughly the same in both cases. The connectivity of \nthe state space tree, however, is different, causing the same states to \nbe visited more frequently for the nonminimized machines, resulting \nin a small increase in run times.\n\nThe effort to minimize the amount of work to be done in the search \nalgorithm is concentrated on minimizing the theoretical maximum \nnumber of states in the product space of the individual finite state \nmachines. We can do this by reducing the number of states per state \nmachine (e.g., by masking a variable or a message queue) or, less \nstraightforwardly, by reducing the number of state machines as such. \nThe last thing we would like to do is, of course, to extend the number \nof state machines that we begin an analysis with.\n\nSomewhat paradoxically, this approach seems to conflict with the \nmore conventional structured approach to program design that tells \nus to identify functions and to separate these in a relatively large \nnumber of logical entities. For protocol design this approach was most \nrecently suggested in Ref. 15, which describes a method where each \nlogical entity is formalized in a small finite state machine that interacts \nwith the others. Dividing a single automaton of 16 states into 2 state \nmachines of 8 states each, however, quadruples the number of states \nin the product space. Similarly, dividing it into 4 even simpler state \nmachines of 4 states each expands the product state space to 16 times \nits original size. In general, increasing the number of state machines \nleads to an exponential growth of the product state space and is \ncounterproductive in analyses.\n\nWe noted above that, unlike the more commonly used breadth-first \nsearch (see Refs. 5, 9, and 12), the state space in a depth-first search \nneed only contain the states in a single execution path from the root \nto the current state. Storing other states can avoid double work, but \ndoes not affect the scope of the analysis. This property of the depth- \nfirst search method gives us greater flexibility in controlling the state \nspace size during analysis runs. If more states can be stored, though, \nthe search will be more time efficient. Figure 6 shows the effect of the \nsize of the state space on the time and space requirements of a search, \nfor a state space cache of 150,000 states that is reduced in steps of \n1,000 to a cache of 50,000 states. \u201cDouble work\u201d is measured here as \nthe total number of states created or recreated while searching. Note \nthat the number of states stored in the cache could roughly be halfed \nwithout noticeable effect on the runtime or the total number of states \ncreated. With a partial state space cache it has to be decided which \nstate will be deleted from a full cache when a new state must be\n\ncreated. A simple blind round robin selection of states was found to \noutperform a series of other, more subtle, schemes.*\u00bb\u201d\u201d It is the strategy \nused in the test of Fig. 6.\n\nBy default a protocol tracer can check a protocol for the observance \nof general correctness requirements such as absence of deadlock and \ncompleteness. The validation language Argos allows for the specifica- \ntion of assertions to check on the observance of other correctness \nrequirements. Assertions are defined as a restricted class of processes. \nThey specify global system behavior in terms of external actions: For \nexample, the specification\n\nis a requirement on the order in which messages of the type mesg are \nsent to the two channels large and smail. The assertion is that in \neach execution sequence a message on channel large must precede a \nmessage on channel sma11, and that these two actions will be executed \nrepeatedly (they are enclosed in a do loop) in precisely this order. \nThe main restriction to assertion specifications is that they can only \nrefer to external actions, that is, sends and receives, and not to \nvariables. Assignments and Boolean conditions are only allowed in \nprocess definitions, not in assertions. The control flow constructs are \nthe same as those for process specifications: concatenations, selections, \niterations, jumps, procedure calls, and macros. In other words, the \nassertions specify global constraints on the execution of the system as \na whole in terms of message exchanges only. The scope of the assertion, \nthat is the set of external actions that is traced to verify or to violate \nan assertion, is implicitly defined by the set of external actions it\n\nrefers to. If an external action occurs at least once in an assertion \nbody, all its occurrences in an execution of the protocol are required \nto comply with it. Compliance with the assertion then means that the \nexecution of these actions should match the context specified in the \nassertion.\n\nSince we define assertions as restricted processes, the assertion \nprimitives can be compiled into a restricted class of state machines | \nand minimized with the same algorithm that is used for the compila- \ntion of the protocol processes. The protocol tracer uses the assertion \nstate machines to monitor the external actions on which they are \ndefined. Alternatively, though our tracer does not exploit this possi- \nbility, it may be possible to develop a heuristic that allows the tracer \nto select those executions in a partial search that have the best chance \nof violating the correctness requirements expressed in the assertions.\n\nIf an action is within the scope of an assertion, the state of the \ncorresponding state machine will be updated as a side effect of the \nexecution of that action, as if the assertion machine itself generated \nit. Since the assertion primitives cannot access variables or channels, \nthe \u201cstate\u201d of an assertion machine is uniquely defined by its control- \nflow state. The \u201cexecution\u201d of an assertion machine then costs very \nlittle in the tracing algorithm. When the protocol system reaches an \nend state, compliance with the assertion can be established by verifying \nthat the assertion machine can reach an end state, too. If this is not \ntrue, the assertion is violated and the current execution sequence can \nbe listed as a counter example. Similarly, if the assertion machine . \ncannot be executed for an action that is within its scope, the assertion \nhas been violated and a counter example can be produced. With little \noverhead or added complexity, the finite state machine model can thus \nbe exploited to combine the depth-first search with assertion checking \ncapabilities.\n\nThe assertion states that the two messages a and b will be appended \nprecisely once to queue c when the three processes are executed, and \nthat they can be sent in that order only. Process a starts by sending \nmessage a to queue c and then waits for a response a to arrive in \nqueue A. Similarly, process b first sends b to queue c and then waits \nfor a message c. The third process waits for a message to arrive in \nqueue C, which is assumed to be either an a or a b, anything else would \nbe an error. Process c then responds by sending a c message to queue \nc and waits for a second message to arrive: a b if the first received \nmessage was an a, or an a if the first message was a b. It will complete \nby sending a c into queue B.\n\nThe protocol is compiled into four finite state machines of three \nstates each for a and b, three states for the assert primitive, and \nseven states for process c. The protocol tracer then takes over and \ncompletes an exhaustive search in 1.35 seconds, reporting the obvious \nassertion violation for the execution sequence that starts with C!b. \nThe violation is reported by the tracer in the following format:\n\nEach column corresponds to a queue and each line to a time step. \nThe first event is the sending of message b to queue C, which already \nviolates the assertion. Then message c is sent to A by process c, \nmessage a is sent to c by process a, and finally a c message is sent to \nqueue B by process c.\n\nChanging the assertion to a more reasonable statement such as \nassert {A!c:B!c} will avoid\u2018the problem. The exhaustive search\n\nfor this assertion completes in 1.32 seconds. Omitting the assertion \ncompletely will trigger a default search for deadlocks and incomplete- \nness (e.g., unspecified receptions), which completes in 1.18 seconds. \nNote that it is relatively straightforward to formalize liveness criteria \nin assert statements. In this case, as for many protocols generating up \nto 10\u00b0 system states, exhaustive analyses are quite feasible. The real \nproblems of partial searching only occur for the larger protocols \ncomparable in size to the experimental protocol used for the tests \nreported earlier in this paper.\n\nThe main assumption we make in this paper is that in a design \nphase a protocol is typically known to contain errors and there is a \nneed for a protocol tracing tool that can quickly find a representative \nsubset of these errors. The user of such a protocol tracer is not so \nmuch interested in completeness but is very much interested in speed. \nWith these assumptions important reductions in the time and space \nrequirements of a tracer become feasible.\n\nThe protocol tracer described here consumes only a small fraction \nof the time and space required by an exhaustive analysis algorithm to \nfind a relatively large fraction of the errors present. The run time of a \nstate space search is reduced by several orders of magnitude by \nrestricting the number of interleavings, by using search depth and \nqueue size restrictions, and by using compile time minimizations (Figs. \n1, 2, 4, and 5(b)). A more general method of reducing run time would \nbe the definition of equivalence classes, or state space foldings, as \ndescribed in Ref. 8. The experimental protocol debugger Trace does \nallow for the definition of such foldings, but too little experience with \nthis technique has yet been gained to report any results.\n\nThe number of states stored in a state space can be reduced by \ncarefully selecting the states that may be revisited (Fig. 5a). More \nimportantly, though, the depth-first search technique used allows one \nto perform searches with an incomplete state space cache. For the \nprotocol tested, the cache could be reduced to less than 50 percent of \nthe state space size (Fig. 6). |\n\nThe experiments with assert primitives and the state space cache \nwere inspired by discussions with Bob Kurshan and Sudhir Aggerwal. \nI am also grateful to Doug McIlroy, Lee McMahon, Rob Pike, Ed \nSitar, and Ken Thompson for discussions, support, and inspiration \nduring the development of the protocol tracer.\n\n1. T. P. Blumer and R. L. Tenney, \u201cA Formal Specification Technique and Implemen- \ntation Method for Protocols,\u201d Comput. Networks, 6, No. 3 (1982), pp. 201-19.\n\n. D. Brand and W. H. Joyner Jr., \u201cVerification of Protocols Using Symbolic Execu- \ntion,\u201d Comput. Networks, 2 (1978), pp. 351-60.\n\n. d. Hajek, \u2018 \u2018Automatically Verified Data Transfer Protocols,\u201d Proc. 4th ICCC, Kyoto, \nSeptember 1978, pp. 749-56.\n\nC. West, \u201cApplications and Limitations of Automated Protocol Validation,\u201d Proc. \n2nd IFIP WG 6.1 Int. Workshop on Protocol Specification, Testing, and Verifi- \ncation, USC/ISI, Idyllwild, Calif., May 1982, pp. 361-73.\n\nP. Zafiropulo et al., \u201cToward Analyzing and Synthesizing Protocols,\u201d IEEE Trans. \nCommun., COM-28, No. 4 (1980), pp. 651-61.\n\nR. Milner, \u201cA Calculus for Communicating Systems,\u201d Lecture Notes in Computer \nScience, 92 (1980).\n\nG. J. Holzmann, \u201cA Theory for Protocol Validation,\u201d IEEE Trans. Comput., C-31, \nNo. 8 (August 1982), pp. 730-38.\n\nG. J. Holzmann, \u201cThe Pandora System\u2014An Interactive System for the Design of \nData Communication Protocols,\u201d Comput. Networks, 8, No. 2 (1984), pp. 71-81. \nD. Brand and P. Zafiropulo, \u201cSynthesis of Protocols for an Unlimited Number of\n\n10. P. R. F. Cunha and T. S. E. Maibaum, \u201cA Synchronization Calculus for Message \nSele tris Programming,\u201d Proc. Int. Conf. on Distributed Systems, IEEE 1981, \npp. 433-45.\n\n11. G. J. Holzmann, \u201cTrace\u2014Performance Measurements,\u201d AT&T Bell Laboratories, \ninternal report, Jan. 1, 1985.\n\n12. G. J. Holzmann, \u201cAutomated Protocol Validation in \u2018Argos,\u2019 Assertion Proving and \nScatter Searching,\u201d 1984, available from the author.\n\n14. E. W. Dijkstra, \u201cGuarded Commands, Nondeterminacy and Formal Derivation of\n\n15. R. P. Kurshan, \u201cProposed Specification of BX.25 Link Layer Protocol,\u201d AT&T \nTech. J. 64, No. 2 (February 1985), pp. 559-96.\n\n16. National Bureau of Standards, Specification of a Transport Protocol for Computer \nCommunications, 4, \u201cService S ecifications,\u201d June 1984.\n\nThe following specification describes a transport protocol defined \nby the National Bureau of Standards.\u2019\u00ae The specification is based on \nthe model given in Ref. 17. Four processes are defined: a local user \nprocess AU connected to a server process A, and a remote user BU \nconnected to server process B. The control flow constructs and the \nI/O statements in Argos are based on CSP, using buffered message \nchannels instead of rendezvous. Process A, for instance, receives mes- \nsages via two channels: one is named ua and is used by the user \nprocess to request services, the other is named ca and is used here to \nreceive control messages from the remote server. Messages from server \nto user are sent through channel ua. The communication between the \ntwo servers is modeled with control messages m1 to m7, as defined in \nRef. 17. The analysis discussed here uses no assert primitives and is \nthus a general one for completeness and absence of deadlocks. The \narrow and the semicolon are syntactically equivalent statement sepa- \nrators. A double colon flags the start of an option in a repetitive\n\nconstruct (do --- od) or in an alternative construct (if --- fi). In \nthis case, the state transition diagram defining the protocol is most \nconveniently, though not most elegantly, modeled by assigning a label \nto every state and including a goto-jump for every transition. Processes \nA and B are symmetrical. Null transitions from the original protocol \nwere deleted from the model.\n\ndo \n: ca?m1 \u2014 UA! conn_ind \u2014 goto rcvd \n2: ua?conn_req \u2014 cb!m1 \u2014 gotocrsent \n:: uaPabort \u2014 cb!m4 \n:: ca?m4 > UA! d\n\n2: ca?m2 \u2014 UA! conn_conf \u2014 goto estab \n:: ca?m7 \u2014 UA! disconn \u2014 goto closed \n2: uaP?abort \u2014 cb!m4 \u2014 goto closed\n\n3:3: ua?conn_resp \u2014 cb!m2 \u2014 goto estab \n2: ca?m7 \u2014 UA! disconn \u2014 goto closed \n:: ua?abort \u2014 cb!m4 \u2014 goto closed\n\nAclose: \ndo \n2:3: ca?m3 \u2014 UA! close_ind \u2014 goto closed \n2: ca?m5 \u2014 UA! data_ind \nca?m6 \u2014 UA! expid_ind \nca?m7 \u2014 UA! disconn \u2014 goto closed \nua?abort \u2014 cb!m4 \u2014 goto closed \nca?m4 \u2014 UA! d \u2014 goto closed\n\nPclose: \ndo \n:: uaPdata_reg \u2014 cb!m5 \n2: uavexpid_req \u2014 cb!m6 \n2:3: ua?abort \u2014 cb!m4 \u2014 goto closed \n:: ca?m4 \u2014 UA! d \u2014 goto closed \n:: ua?close_req \u2014 cb!m3 \u2014 goto closed \n2: ca?m7 \u2014 UA! disconn \u2014 goto closed\n\n:: cb?m2 \u2014 UB! conn_conf \u2014 goto estab \ncb?m7 \u2014 UB! disconn \u2014 goto closed \nub?abort \u2014 ca!m4 \u2014 goto closed\n\nrevd: \nif \n2:3: ub?conn_resp \u2014 ca!m2 \u2014 goto estab \n:: cb?m7 \u2014 UB!disconn \u2014 goto closed \n:: ub?abort \u2014 ca!m4 \u2014 goto closed \n:: cb?m4 \u2014 UB!d \u2014 goto closed \nfi;\n\nub?close_req \u2014 ca!m3 \u2014 goto Aclose \ncb?m3 \u2014 UB!close_ind \u2014 goto Pclose \nub?data_req \u2014 ca!m5 \ncb?m5 \u2014 UB! data_ind \nub?expid_regq \u2014 ca!m6 \ncb?m6 \u2014 UB! expid_ind \ncb?m7 \u2014 UB! disconn \u2014 goto closed \nub?abort \u2014 ca!m4 \u2014 goto closed \ncb?m4 \u2014 UB!d \u2014 goto closed \nod; \nAclose: \ndo \n:: cb?m3 \u2014 UB!close_ind \u2014 goto closed \n:: cb?m5 \u2014 UB! data_ind \ncb?m6 \u2014 UB! expid_ind \ncb?m7 \u2014 UB! disconn \u2014 goto closed \nub?abort \u2014 ca!m4 \u2014 goto closed \n:: cb?m4 \u2014 UB! d \u2014 goto closed \nod: \nPclose: \ndo\n\nod \nThe queue sizes were arbitrarily set to 8 slots per channel. The protocol \ntested is defined by the behavior of the two server machines, as visible \nto the users. The user behavior is no part of the formal protocol. An \narbitrary set of user processes was defined specifically for the test. \nThe local user Au will open the connection by sending a conn_req \nmessage to ua and some arbitrary time later it will close it with an \nabort message. The remote user BU is considered to be passive, \nresponding only to close messages and accepting, but ignoring all \nothers. Default is a keyword for receptions that match any input \nfrom the queue specified.\n\nThe protocol as specified above is compiled\u2014in 23 seconds of CPU \ntime on a VAX-11/750*\u2014into four finite state machines of 27, 27, 10, \nand 6 states, respectively. The compiler flags a series of incompleteness \nerrors, noting for instance that control message m7 can be received \nbut is never sent. Ignoring those warnings, an exhaustive analysis with \ntrace takes just under 3 seconds of CPU time and reports 4 error \nsequences that reduce to two types of errors. The first one is an | \nunspecified reception of the message conn_resp In state closed, for \ninstance, after the following message exchange:\n\nEach column corresponds to a queue and each line to a time step. The \n_ first event recorded is the sending of a message conn_req into queue \nua, followed by an m1 into queue cb, etc. The last message sent is\n\n_ The second problem is an unspecified reception of m2 also in state \nclosed:\n\nIt is now straightforward to study the behavior of the protocol for \ndifferent user behaviors, which can reveal, for instance the possibility \nof the unspecified reception of a message conn_resp in state Pclose, \nor the more obvious deadlock after a simultaneous conn_req message \nfrom both users.\n\nGerard J. Holzmann, Kand. Ir. (B.S.), 1973, Ir. (M.S.) (Electrical Engi- \nneering), 1976, Ph.D. (Technical Sciences), 1979, Delft University of Tech- \nnology, The Netherlands; Mr. Holzmann obtained a Fullbright Fellowship in \n1979; with the University of Southern California in Los Angeles, 1979-1980; \nAT&T Bell Laboratories, 1980-1981, 1983\u2014. Before returning to AT&T Bell \nLaboratories, Mr. Holzmann was an Assistant Professor at the Delft Univer- \nsity of Technology. In 1981 he was awarded the Prof. Bahler prize of the Royal \nDutch Institute of Engineers (KIVI) for his work on telecommunication \nsystems. His current research is in distributed systems and computer graphics \nin the computing science research center.\n\nWe describe a Videocassette Recorder (VCR)-based information system \nwhereby we can distribute frequently updated large pictorial databases to \nindividual users and provide a variety of interactive video services. The four \nkey advantages of this system are: (1) economics, (2) good picture quality, (3) \ncapability to reach nationwide users, and (4) ability to update the database \nfrequently (say, daily, preferably in early morning hours when many trans- \nmission facilities are unused). An experimental home terminal consisting of a \nVCR driven by a personal computer for random-access searches was con- \nstructed to demonstrate this concept. The pictorial database used in the \ndemonstration includes real estate listings, vacation guides, autos and Sears- \ntype merchandise catalogs. We also make comparisons of this system to other \nvideo services and conclude that the present approach has potential advantages \nin many applications.\n\nThere are presently many systems under development for providing \ninteractive visual displays.! For example, videotex uses the switched \ntelephone network to send and receive digital data, which are then \nused by a microprocessor terminal to construct color graphics on a TV \nscreen. Teletext is another technique whereby digital data for the same \npurpose are imbedded in the vertical blanking period of a video signal \nbroadcasted to the end users. The graphics capability of these two\n\nmethods is limited in representing those real objects, e.g., clothing and \nfurniture, for which visual attractiveness is more important than \nfunctional appeal. -\n\nLimitations in the computer-graphic representation of real objects \ncan be overcome by storing the pictures on an interactive videodisc \n(as in the \u201celectronic book\u201d application).? In such a case, hard copies \nof the disc have to be distributed by mail or through stores to the end \nusers. If real-time access to a central database is really required, then \nthe so-called frame-grabber approach is frequently suggested. In this \nlatter system, single frames are sent to individual users, time multi- \nplexed on a dedicated video channel.? The user terminal must then \nstore the received frame so that it can be examined by the human \nviewer. Digital frame stores for this application are expensive, although \ntheir cost should decline eventually. Nevertheless, a multiuser com- \nputer system at a central location must manage the data requests and \nsend different video frames to different users. Such a system can be \noverloaded easily, and practical solutions for giving nationwide service \nto thousands of users simultaneously have yet to be worked out.\n\nHere we suggest an alternative arrangement which appears to be \nmore economical than the videotex or frame-grabber systems. We \npropose that the home terminal consist of a home Videocassette \nRecorder (VCR) connected to a personal computer, and that a \u201cframe- \nsearch\u201d capability be provided whereby the home computer can specify \nwhich frame of the videocassette is displayed (in still frame) at any \none time. The VCR must of course have good still-frame performance. \nThe overall system for distributing pictorial databases from a central \nstation to end users with such home terminals is outlined in Section \nII. Then we describe an experimental home terminal for demonstrating \nthe feasibility of this idea (Section III). Finally, we make comparisons \nwith other systems (Section IV) and conclude that our present proposal \nhas potential applications in both the business and consumer markets.\n\nWe propose the direct distribution of pictorial information to the \nend users\u2019 VCRs via a TV broadcasting channel, as illustrated in Fig. \n1. The pictorial database is assembled in one central location, where \nindividual color pictures (35 mm photographs or slides) are recorded \nonto a master videotape (one-inch type) in a frame-by-frame manner, \ni.e., a single color picture on a single video frame. A two-hour videotape \ncan then store up to 216,000 single-frame pictures. In the vertical \nblanking period of these video pictures, we insert a frame number for \nidentification as analogous to the page number in a book. This tech- \nnique of numbering the video frames can be implemented easily with \nthe conventional Vertical Interval Time Code (VITC). In addition,\n\nAs suggested in Fig. 1, the database from the central station can be \ntransmitted through a satellite broadcasting system (C-, K,-band, or \nDBS) whereby nationwide users can either receive the information \ndirectly or through a local broadcaster. In the latter case, the local\n\ndistributor simply takes the received video signal from the satellite \nand retransmits it to the users via his own broadcasting system (e.g., \ncable TV, off-the-air VHF/UHF channels, optical fiber, etc.). Attach- \ning additional information from a local database is optional. Since \nmany transmission facilities and T'V channels are idle in early morning \nhours, this operation can most conveniently be done in the middle of \nthe night with the VCRs programmed or pretimed for unattended \nrecording. Once the database is recorded on a videocassette, the users \ncan assess the information at their home terminals at their leisure. \nDirect distribution in this manner avoids the cost of recording and \nshipping thousands of tapes (or discs). Moreover, the database is \nalways up-to-date, depending on how often it is sent (once a day should \nbe adequate for most applications). No special equipment is required \nat the TV station or CATV head-end for sending out the databases. \nIndeed with satellite transmission, nationwide distribution is possible \nwith transmission systems already in place. For example, the Sears \ncatalog could be sent in 4.5 minutes, assuming 1600 pages and 5 frames \nper page. One thousand real-estate listings could be sent in 5.5 minutes, \nassuming 10 frames per listing. Custom orders for still-picture or full- \nmotion information (e.g., instruction manuals) could even be served if \nmore transmission time were available.\n\nWith the database recorded on a videocassette, we can use the \ninteractive home terminal suggested earlier to browse through the \npictures in a random-access manner. We constructed an experimental \nterminal to demonstrate this idea, which is discussed in the next \nsection. But first, we should point out that the assembly of large \ndatabases at the master station is no easy task. In fact, the production \ncost could be an important consideration in systems of this type. The \nneed for automation in database production is mandatory and indeed \nmanageable with modern production equipment.\n\nWe show in Fig. 3 the block diagram of a VCR-based interactive \nterminal suitable for examining stored video data from a videocassette. \nThe solid-line portion represents what was implemented in our labo- \nratory aS an experiment to demonstrate the proposed concept, while \nthe dashed-line part stands for an alternate approach using a remote \ncomputer (via telephone hook-up) to control the operation. They are \nfunctionally the same, but offer different kinds of user flexibility. More \nis discussed about their differences after we explain the terminal itself.\n\nThe terminal consists of three major components: the computer, \nwhich serves as a controller for the entire system; the VCR/computer\n\na a aa a 7 \ncea HEE gcenae \nPERSONAL | \nbe Ee ad es | i] \n| PHONE \nj LINE \nasec ead 1 \nOPTIONAL i \nCONNECTION FOR p CENTRAL poy \nORDER/INFORMATION | COMPUTER | \nTHROUGH A MODEM bose ge 4\n\ninterface; and the VCR* (with TV) itself. The video signal from the \nVCR is passed through the VCR/computer interface before being \ndisplayed on the TV. Inside the interface (Fig. 4), a field number \ndecoder is used to examine the video frame identification number \n(VITC) recorded in the vertical blanking interval. This information \nshould be made available to the computer at all times, i.e. when the \ntape is in still frame or in motion. But for simplicity, it is sufficient to \nprovide valid frame numbers-for tape motions from still-frame to play \nspeed (30 frames/second). For higher tape speeds, the computer can\n\n* Both the VHS and the Betamax formats would work as long as the VCR has good \nstill-frame performance. We used VHS in our terminal.\n\nuse calculated estimates based on the initial known tape location, the \ncurrent known tape speed, and the measured elapsed time plus some \nprecalibrated adjustments. As an adjunct, the status encoder monitors \nthe current operational status of the VCR (such as stop, fast forward, \nplay, etc.) and conveys it to the computer. The command decoder, on \nthe other hand, receives commands from the computer and translates \nthem into actual instructions to the recorder, i.e., emulating a human \npushing buttons to control the VCR. This was done simply by wiring \nup computer-driven electronic switches over the contact points of the \npushbuttons on the remote control unit of the recorder.\n\nAlthough the field number decoder is designed to detect VITC for \nframe identification, it is easy to extend the idea for decoding digital \ndata recorded in all or part of an active TV field. Thus, part of the \ndatabase can be devoted to digital information such as table of con- \ntents, programming or search instructions, synthesized audio, etc. \nThey can be copied directly to the computer memory for use. _\n\nA sample video database was put together for the experimental \ndemonstration of the terminal. It consists of four sections: (a) real \nestate listings; (b) automobiles; (c) a vacation package; and (d) mer- \nchandise catalogs. The interactive access to these data is done via a \ntouch-sensitive screen on which menus are printed to prompt the user. \nThe choices are all self-explanatory and are available to the user as \ntouch-sensitive buttons on the personal computer. Instructions have \nbeen kept to a minimum, and the operation is so user friendly that a \nuser manual is seldom needed. The majority of the material in the \ndemonstration database is still-frame pictures. However, the full- \nmotion video segments (with sound too) in the Hawaii Vacation Guide \nand also in a merchandise catalog on how to use the riding lawn mower \nconsume much more recording time than the still-frame pictures. Note \nthat our use of the personal computer with a touch-sensitive screen \nwas merely a choice for experimental convenience. Any other user \ninterface may be substituted to serve the same purpose in an actual \nsystem.\n\nIn addition to the user-prompting menu, 16 additional touch-sensi- \ntive buttons are always available in the bottom of the computer screen \n(see Table I). As an example, if one selects the button for real estate \nlistings, then a table of choices for different towns appears on the \ntouch-sensitive screen (Fig. 5). Touching one of these choices would \nbring up the next menu for specific price ranges. After that we have \none more menu selection for specific house types, e.g., two-story \ncolonial, log home, etc. The computer then fetches the text listing of \nthe house selected from its memory and displays it on the touch- \nsensitive screen. Meanwhile, the search routine to locate the picture \nfor the house is executed to the VCR. Therefore, the user sees the\n\nBrings up the Hawaii Vacation Guide consisting of \u201cMap of \nthe Hawaii Islands,\u201d \u201cScenes From Hawaii,\u201d \u201cThe Polyne- \nsian Culture Center,\u201d and \u201cThe Sea Life Park,\u201d all of which \n(except the map) are full-motion video with sound.\n\nBrings up the real estate listings: still pictures of the houses \nfrom the VCR and their corresponding text descriptions on \nthe touch-sensitive screen.\n\nBrings up the merchandise catalogs: still pictures of individual \nitems plus full-motion-and-sound segments of merchandise \ndemonstrations.\n\nReturns to the previous menu. \nAllows user to place an order or obtain order information. \nBrings up function definitions.\n\nSingle-frame advance in the forward direction. \nSingle-frame advance in the reverse direction.\n\nCauses video to move forward with momentary pause at each \nstill-frame picture for browsing.\n\nCauses video to move backward with momentary pause at \neach still-frame picture for browsing.\n\nPauses for still-frame viewing and brings up text description \nof current item.\n\ncolor picture of the house on the TV and has the text listing simulta- \nneously from the computer. The 16 buttons at the bottom of the screen \nprovide plenty of choices and flexibility for the next step.\n\nThe search algorithm implemented in our software is designed to \nperform random access on the VCR in the most expedient manner. It\n\nshould not be a surprise, however, that the VCR access time is \nconsiderably slow compared to that of a device truly capable of random \naccess, e.g., a videodisc player. It takes a VCR typically 3 to 5 minutes \nto rewind a standard T120 (2/4/6-hour) cassette, and this would be \nthe worst-case waiting time for getting to a picture from one end of \nthe tape to another. Such a long access time is probably unsuitable for \nmany applications. To circumvent this inconvenience, we propose that \nthe whole database be divided into segments of, say, 1000 pictures \neach. Then the search within each segment is quite fast. As an \nillustration, if a 10 times play (10X) speed is used for the search, i.e., \n300 frames/second, the maximum access time within such a segment \nis only 3.3 seconds. In our experimentation, we found that the speed \nsearch feature of the VCR was the most appropriate function to use \nfor random access (i.e., fast visual search with 10X, 5X, 2X, etc.) \nbecause the video heads did not have to be disengaged and then \nreengaged during the process, and the continuous display of fast \nmotions on the T'V helped to make the waiting time less objectionable. \nIn the event that the desired TV frame is very far away from the \ncurrent position, the normal fast forward or rewind might have to be \nused. The optimization process is based on information previously \nobtained by an automatic calibration program which sizes the start \nand stop times, the video head engage/disengage intervals, and so on \nfor each specific VCR. In short, we have developed an automatic \ncalibration process that can characterize the access performance of a \nVCR, and the random-access software is intelligent enough to use this \ninformation for optimal control.\n\nLaying out the database in segments of 1000 frames each requires \nintelligent partition of the overall source into groups of \u201chumanly \ncorrelated\u201d materials. For example, we don\u2019t want to put the listings \nof all different tires in one segment as in a conventional merchandise \ncatalog. Instead we want to group all the accessories of a specific car \nmodel together. The situation is analogous to that of a large library \nwhere the books are carefully categorized and authored in such a way \nthat most of the information desired for a subject is confined to a \nbook of 1000 pages. The user can browse through each book very \nquickly, but probably does not mind spending more time in looking \nfor another book. The assumption here is that most readers would \nspend some time reading a specific book of interest rather than reading \npages of uncorrelated materials from different books in a random \nmanner. We recognize that such an idea needs much more research \nbefore it can be put into practice. Nevertheless, recent experience \nfrom the videotex experiments seems to suggest evidence supporting \nthe validity of this concept.\n\nIt is clear from the foregoing discussion that software plays a key \nrole in this system. In our experimental setup, all the software resides \nin the personal computer (Fig. 3), and the user has complete control. \nThe dashed-line portion in the figure indicates the other alternative \nof having the controlling software in a remote computer, and the user \ninteracts with the system via a key pad connected to a modem. This \nlatter approach has the advantage that more powerful software can be \nused at the expense of less user control as well as less privacy. The \nbasic idea remains, however, that very intelligent software is needed \nin managing the enormous database made possible by a simple con- \nsumer-type VCR.\n\nFinally, let us point out that the VCR system can be used as a high- \ndensity digital storage unit, which can provide digital high-fidelity \nmusic as well as synthesized voice and text data.\n\nThe VCR-based information system has the same applications as \nother interactive video services. Some of the generic examples include \nreal estate listings, vacation/entertainment guides, merchandise cat- \nalogs, product demonstrations, and service/instruction manuals.\n\nWe summarize in Table II a comparison with videotex and the \nvideodisc-based system. It should be emphasized before we discuss \nthese results that there is no single criterion possible for judging the \nrelative merits or shortcomings of any approach. Instead, most systems \ntend to be application oriented. In other words, each individual video \nservice tends to appeal more for the application it is intended for, and \nthere is probably no single system that is universally \u201cbetter\u201d than all\n\nVCR Based Disc Based Videotex \nVideo quality Good. Good Graphic \nDatabase creation Automatic Automatic Manual photo- \nphoto-to-tape photo-to-tape to-graphic \nDistribution VHF/UHF, By mail or Telephone \u00a9 \nCATV, DBS, through stores \nfiber, etc. \nNumber of hard copies 1 \u2014 100,000+ 100+ \nFrequent updates Yes No Yes \nReal-time interaction with Limited Limited | Yes \ndata suppliers \nResponse time Slow Fast Depends on \nnumber of\n\nThe picture performance of VCRs is usually designed to be compat- \nible with the characteristics of other devices they are connected to in \nmost home use, e.g. resolution is similar to that of a popular consumer \nTV (without comb filtering) and signal-to-noise ratio is comparable \nto most cable TV systems. In any event, their picture quality in our \nsubjective viewing was found to be remarkably close to that of cable \nTV, while the videotex picture tends to be cartoon-like. The videodisc \nis potentially capable of noticeably better quality than the VCR \nalthough this is usually not so in practice.\n\nThe database creation process for the VCR and the videodisc is \nalmost the same. In both systems, the original material (e.g., photos \nor slides) is recorded on a one-inch video tape serving as the master, \nand the difference between the two cases is that the videodisc requires \nfurther processing in transferring the tape material onto a master disc \nbefore mass duplication. As for videotex, the original material has to \nbe recreated in the computer (with computer-aided tools) as a graphic \nrepresentation of the real object.\n\nThe VCR system uses the TV broadcasting for distribution. Video- \ntex uses the phone lines to connect customers to a central computer, \nwhile videodiscs have to be distributed via mailing or store sale.\n\nOnly one master copy of the database needs to be maintained \nnationwide in the VCR system. Videotex service requires a hard copy \nat each computer center, and hundreds are thus needed nationwide. \u2014 \nThe videodiscs are, of course, hard copies of the master, and a national \nmarket would require hundreds of thousands of them.\n\nBecause the database resides in the user terminal, real-time inter- \naction with the data supplier tends to be limited in both the VCR and \nvideodisc systems. On the other hand, videotex users enjoy continuous \nreal-time interactive picture transmissions with the central computer.\n\nThe VCR access time is considerably slow compared to the videodisc, \nas discussed earlier. The videotex response time depends mainly on \nthe number of simultaneous users and is probably slow (tens of \nseconds) for a fair size of simultaneous accesses (say thousands).\n\nThe VCR approach could become extremely economical if there was \na mass availability of VCRs and personal computers, both of which \nhave tremendous appeal in their own right and are gaining popularity \namong businesses and consumers. The custom interface necessary to \nconnect the VCR to the personal computer is so simple that it can \neasily be incorporated into either the VCR or the computer. In any \ncase, its cost should only constitute a small fraction of that for the \ntotal user terminal. Videodisc players capable of true random-access \nsearch are fairly expensive, and their popularity with consumers seems \nto be on the decline. Videotex terminals are also quite expensive, but \ntheir cost could decrease dramatically if more large-scale integration \nwere employed.\n\nWe have proposed a system for distributing large pictorial databases \nto home videocassette recorders (VCRs). Distribution is done by\n\nbroadcasting from a master station where the picture information has \nbeen assembled in a frame-by-frame manner, i.e., one picture per video \nframe, resulting in 30 independent pictures in a 1-second video seg- \nment. The broadcasting medium could be a combination of direct \nbroadcast satellites, cable TV, conventional VHF/UHF TV channels, \nor custom fiber systems. The main idea is that this can take place in \nthe middle of the night when many transmission facilities become \nvacant, and the home VCRs can easily be pretimed. for unattended \nrecording. In fact, distribution or updating is possible as often as \ntransmission time permits, but once a day is probably adequate for \nmost applications.\n\nWith the complete database stored in a videocassette, an end user \ncan retrieve the data of his particular interest at his leisure with the \naid of a simple home terminal. We constructed such a terminal \ncomprising a VCR capable of good still-frame performance, a personal \ncomputer serving as a controller for random access, and a custom \ninterface connecting them together. The VCR/computer interface \ntranslates digital commands from the computer into actual operational \ninstructions to the VCR, i.e., emulating a human pushing the control \nbuttons on the recorder. It also feeds back the operational status of \nthe VCR to the computer. Most important of all, it examines the video \nsignal and decodes a frame number previously recorded in the vertical \nblanking interval during data assembly. This frame number is the \n\u201cpage number\u201d of the electronic book and is supplied to the computer \nso that it knows which video frame or picture is being displayed on \nthe TV. Software on the computer was implemented to do random- \naccess search through the database, and the interface to the user is in \nthe form of a touch-sensitive screen with menu-driven selections. The \ncapability of this experimental terminal was demonstrated with a \nsample database consisting of real estate listings, new car models, \nvacation packages, and merchandise catalogs:\n\nThe main attraction of our proposal is its potential economics. That \nis, it takes advantage of other potentially low-cost and widely available \nterminal equipments, namely, the personal computer and the VCR, \nboth of which have consumer appeal in their own right. Distribution \nrequires only a single database plus transmission facilities that are \nalready in place. Thus, a service supplier need only provide a hardware \ninterface plus software, a networker need only supply a satellite or a \nTV station plus a video production unit, and purveyors of information \nneed only furnish color photos and text.\n\nThe authors wish to thank R. F. Weihs for his work on the experi- \nmental terminal.\n\n2. R. D. Gordon, \u201cAn Intelligent Electronic Book System and Publishing Facility,\u201d \nConference Record of Outlook For Optical and Video Disc Systems and Appli- \nene February 20, 1985, The Institute for Graphic Communications, Miami, \nFlorida.\n\n3. H. Ando and H. Yamine, \u201cStill-Picture Broadcasting\u2014A New Informational and \nInstructional Broadcasting System,\u201d IEEE Trans. Broadcasting, BC-19 (Septem- \nber 1973), pp. 68-76.\n\nThe IEEE 802 standard for local area network based on Carrier Sense \nMultiple-Access with Collision Detection (CSMA/CD) operates at a peak rate \nof 10 Mb/s on a cable of maximum length 2.5 km using baseband signaling. \nIn many situations, larger channel rates are required over a much larger area. \nHowever, the efficiency of the CSMA/CD access method decreases rapidly if \neither the length of the cable is increased for a fixed bit rate or if the bit rate \nis increased for a fixed cable length. In this paper, we propose a broadband \nnetwork for computer communications containing several CSMA/CD-type \nsystems, each operating in a different frequency band. In addition, in order to \nhave a wide area access, while minimizing the loss of performance associated \nwith large collision delays, terminals in a small given geographical area are \ngiven one of the frequency bands for transmission. Two access protocols are \ndeveloped. Using these schemes, it is possible to increase the channel through- \nput and the access area and to reduce the collision delay. We present a \nsimplified analysis to quantify the improvement in performance using our \nschemes.\n\nLocal Area Networks (LANs) share computing and other resources \namong many users and, if properly designed, increase the reliability \nby reducing the dependence of a user on one processing unit or a \nperipheral. Unlike long-haul networks, where channel utilization has \nto be optimized owing to high cost of communication over long\n\n* AT&T Bell Laboratories. \nt Visiting from the Department of Electrical and Electronic Engineering, University \nof Western Australia, Nedlands, Australia. |\n\nCopyright \u00a9 1985 AT&T. Photo reproduction for noncommercial use is permitted with- \nout payment of royalty provided that each reproduction is done without alteration and \nthat the Journal reference and copyright notice are included on the first page. The title \nand abstract, but no other portions, of this paper may be copied or distributed royalty \nfree by computer-based and other information-service systems without further permis- \nsion. Permission to reproduce or republish any other portion of this paper must be \nobtained from the Editor.\n\ndistance, local networks use bandwidth somewhat extravagantly to \nreduce the switching costs. Several network topologies, such as rings, \nbuses, and trees, have been proposed along with access methods such \nas carrier sense, token passing, etc.!\u201d\n\nThe IEEE 802 standard for local area networks uses CSMA/CD \n(Carrier Sense Multiple-Access with Collision Detection) as one of its \naccess methods.\u00ae It uses baseband transmission on coaxial cables \n(although other media are possible) at a peak rate of 10 Mb/s. For a \nvariety of reasons, length of the cable (and therefore length of each \nsegment of the network) is limited to 2.5 km. Within the limitations \nof the above parameters, the CSMA/CD-based access method provides \nan efficient means of computer communication for low loads on the \nchannel. However, if the channel loading is increased, or if the require- \nments dictate either higher bit rates or longer cable lengths\u2014for \nexample, to serve a metropolitan area\u2014there is considerable loss of \nefficiency. Much of this inefficiency comes from the use of the CSMA/ \nCD protocol. In CSMA/CD, a source transmits a packet when the \nchannel is sensed as idle, but this injection of the packet can be known \nto the other sources only after it has propagated throughout the length \nof the cable, during which time another source may attempt to transmit \non the channel. Thus, the number of bits wasted due to collision is \nproportional to the propagation delay and the peak bit rate. Also, the \nneed to detect collisions makes it necessary that each packet have a \nduration equal to at least the round-trip delay. With very large nets \nand high bit rates, that may represent an unreasonably large minimum \nnumber of bits.\n\nBaseband CSMA/CD has been extended to broadband CSMA/CD \nby several CSMA/CD networks, each in a different frequency band \nput on the same cable (see, for example, Ref. 4). However, each of \nthese networks operates almost independently, connected usually by \na signaling channel. Also, the cable length limitation still applies, \nmaking it difficult to use for a metropolitan area. In this paper, we \npropose schemes that extend the capabilities of both the baseband and \nthe broadband CSMA/CD networks by allowing higher bit rates on a \ncable, larger cable segments, and at the same time smaller collision \ndelays. We do this by dividing the available bandwidth of the cable \ninto several frequency bands and operating a network (or channel) in \neach frequency band. Since coaxial cables can easily carry up to 400 \nMHz, several networks can be accommodated on one cable rather \neasily. By modulating the baseband data from devices connected to \nthe network to a high-frequency band, total channel bit rates of higher \nthan 50 Mb/s can be obtained easily. However, since the bit rate of \neach of the nets is kept low, channel inefficiency due to the use of \nCSMA/CD protocol is not increased. To increase the length of the\n\ncable segment, and at the same time limit the collision delay, we divide \nthe users into communities based on their location and give each user \ncommunity a network (i.e., one band of frequencies) to transmit most \nof the time. Thus the \u201ceffective\u201d end-to-end delay is reduced although \nthe cable length is increased. The principal characteristics and advan- \ntages of our system are the following:\n\n2. Larger cable length but smaller collision delay by dividing the \ncable into several parts and operating a network in a given frequency \nband for each part to be used by a user community, while retaining \nthe listening ability on the full cable length.\n\n4, Different grade of service, depending on the complexity of net- \nwork interface.\n\n5. Restricting most of the high-speed processing to the analog \ndomain and baseband processing to digital domain. Thus, although \nthe network may have channel throughput over 50 Mb/s, individual \nnetworks may carry at much lower bit rates.\n\nIn this section, we describe one possible implementation of our \nsystem. A block diagram is shown in Fig. 1. Each terminal has a \nfrequency-agile Radio Frequency (RF) modem that can modulate \nbinary data for transmission on the cable and demodulate the signal \nfrom the cable to extract the transmitted binary data. Unlike long \ndistance transmission, since. the intent is not to maximize the data \ntransmission rate, simple inexpensive modulation schemes can be \nchosen with enough separation between the various frequency bands \nto keep the filtering simple and to reduce the crosstalk. As an example, \nif modems based on Frequency Shift Keying (FSK) with 1/4 bit/Hz \nare used, then a cable of bandwidth 300 MHz can support six CSMA/ \nCD networks of peak rate 10 Mb/s each with a guard band of 10 MHz \nto separate each of them.\n\nOur block diagram in Fig. 1 shows bidirectional transmission, that \nis, signals injected on the cable at each tap travel in both directions \nand the amplifiers are bidirectional. It is necessary for the taps to be \nbidirectional so that they can receive signals from either direction. \nAlthough this is a straightforward extension of the baseband CSMA/ \nCD network, bidirectional amplification and taps may present engi- \nneering difficulties, particularly at high frequencies. Alternative de-\n\nIn Fig. 1, there are two networks (called Homenets) and therefore \ntwo frequency bands. The transmitter can, by the agility of the modem, \ntransmit on any of the two frequency bands, and the receiver can \nreceive and demodulate data from both the frequency bands. Termi- \nnals attached to homenet 1 transmit mostly on frequency band f1 and \nthose attached to homenet 2 transmit mostly on frequency band f2. \nIf several simultaneous conversations with terminals on different \nhomenets are required (as in the case of a host computer), then a \nterminal may need multiple transmitters and receivers. Details of the \nprotocols for the access are given in the next section.\n\nBelow we give two types of access protocols; the first does not \nrequire synchronization of the different terminals, whereas the second \ndoes. Some desirable characteristics of any protocol should be noted \nfirst. The access delay should be decreased by scheduling the trans- \nmission on a net that is either least busy or has the least chance of \ncollision. The load on the different networks should be distributed \nsuch that a situation does not arise in which many terminals are trying \nto transmit on a network and are unable to do so, while the rest of the\n\nnetworks are carrying a very light load. In carrier sense multiple- \naccess systems, collisions are a result of the terminals knowing the \ntransmission by other terminals only after the propagation delay. \nTherefore, by requiring all the terminals that are close to each other \nto initiate their transmission on a particular network, the collision \ndelay can be reduced considerably. Thus, in both the protocols, each \nterminal is assigned to a particular network. This network is called \nthe Homenet of the terminal. Homenet assignment is included as a \npart of the address of the terminal. Each terminal maintains a list of \nthe homenet assignments of the other terminals. The homenet assign- \nment, although made primarily by geographical location, may also \ntake into account the desired connectivity, traffic patterns, etc.\n\nFigure 2 shows a system in which there are two homenets and three \nterminals per homenet. The transmission initiated by any of terminals \n1, 2, and 3 is mostly on homenet 1, whereas the transmission by \nterminals 4, 5, and 6 is on net 2. Since net 1 and net 2 are on two \ndifferent bands of frequency, the collisions are now localized. That is, \ndata from terminal 1 can only collide with data from terminals 2 and \n3. Since the distance between the taps on the cable of terminals 1, 2 \nand 3 is much shorter compared with distance between the taps of \nterminals 1 and 6, the probability of collision and, therefore, of data \nwasted due to collision is significantly reduced. This increases the \nchannel utilization and decreases the delay. Of course, the protocols \nmust and do allow communication between the terminals on different \nhomenets.\n\n1. Every terminal has at least two receivers and, therefore, is capable \nof listening to at least two networks. One of these receivers always \nlistens to the homenet. The other receiver is free to listen to any net.\n\nPUT S REC2 \nON HS \nB \u2014 CHANNEL BUSY SNM \u2014 SCHEDULE TRANSMISSION OF \nC \u2014 COLLISION DURING TRANSMITTING NEXT MESSAGE \nHD \u2014 HOME NET OF DESTINATION SNP \u2014 SCHEDULE TRANSMISSION OF \nHS \u2014 HOME NET OF SOURCE NEXT PACKET \nLBT \u2014 LISTEN BEFORE TRANSMITTING SRB \u2014 SCHEDULE RETRY AFTER BUSY \nLM \u2014 TEST !F TRANSMITTED MESSAGE WAS LAST CHANNEL \nLP \u2014 TEST IF TRANSMITTED PACKET WAS LAST SRC \u2014 SCHEDULE RETRY AFTER COLLISION \nLWT \u2014 LISTEN WHILE TRANSMITTING TR \u2014 TRANSMITTER \nREC1 \u2014 RECEIVER 1 WIC \u2014 WAIT FOR IDLE CHANNEL \nREC2 \u2014 RECEIVER 2 WNM \u2014 WAIT FOR NEXT MESSAGE \nS \u2014 SOURCE XMIT \u2014 TRANSMIT\n\nhomenet. The other receiver of that terminal becomes active only after \nthe first receiver starts listening to a network other than its homenet.\n\n3. Any terminal A desiring to transmit to terminal B goes through \nthe following sequence.\n\nc. If net, is idle (i.e., absence of carrier), transmit net, carrier for a \nperiod 7, the two-way propagation delay through the total network. \nThis amounts to a priority preempt on nety,.\u00ae*\n\nd. If during the second half of the period T there is collision on \nnet,, then it implies a preemptive transmission from another terminal, \nnot on net,. In that case, terminal A backs off and attempts a \ntransmission on net, with reduced probability at the next time slot of \nT. If there is no collision, then terminal A follows its preempt with a \nmessage to terminal B.\n\ne. Terminal B always has one receiver listening to net,; therefore, \nit receives information from every collision-free packet on net\u00bb.\n\nf. If packet communications is to be continued, then terminal A \nstarts listening on net,, terminal B on net,, and both terminals \ntransmit on their homenets. Thus, if a message has several packets, \nonly the first packet may be transmitted on a homenet different from \nthe homenet of the source; all the subsequent packets are transmitted \non their own homenet with standard CSMA/CD protocol with retrial \nperiod equal to round-trip delay of the homenet.\n\ng. If at step e terminal B is already in communication with some \nother terminal on a different network, then it still has a receiver on \nnet,. If terminal B\u2019s transmitter is on net, (as it normally is, except \nwhen it is trying to set up an initial connection with a terminal on a \nnetwork other than net,), even if it is in communication with some \nother terminal, it can send an acknowledgment back to terminal A on \nnet,. If, however, Terminal B\u2019s transmitter is transmitting on a \ndifferent channel, there may be delay in sending the acknowledgment.*\n\nh. If after successful connection there is no communication for \na given amount of time and if the receiver on the homenet receives \na message for communication from another source, then the \nother receiver of both the home terminals go back to their respective \nhomenets.\n\n* Alternatively, there could be a signaling channel in a different frequency band \naccessed by all the terminals, and the first packet could be transmitted on the signaling \nchannel. This alternative is attractive for a large network, since it confines collision to \na common homenet and, hence, does not constrain minimum packet length. With the \nutilization very low and the network large, the most appropriate protocol on the signaling \nchannel would be ALOHA.\n\nTIf several simultaneous conversations are required, then a terminal may need \nmultiple transmitters and receivers.\n\ni. Many different broadcast modes are possible. If the broadcast to \nonly the terminals in the particular homenet is desired, then data is \ntransmitted on that network only. However, if broadcast is required \nto all the terminals, then the transmitter has to successfully transmit \non each network.\n\nThe above protocol is reasonable in that it reduces collisions and \nworks well when the traffic is quite bursty, with many terminals \ntrying to transmit messages containing large numbers of small packets \nfrequently. However, when there are large file transfers, use of the \nhomenet by a terminal prevents other terminals with the same \nhomenet from using the channel even though the other networks may \nbe idle. Thus, a reasonable protocol is needed that will share the \nchannels more evenly in the presence of large file transfers by one of \nthe terminals. Protocol 2 attempts to accomplish this at the cost of \nslightly larger average delay in establishing a connection. In this \nprotocol, networks on which a given group of terminals begin trans- \nmitting are switched on a periodic basis. The period is of the order of \nseveral packets long (or tens of milliseconds). It thus requires a clock \nat every terminal, which may be provided from a central clock on a \ndifferent band of frequencies. Details of the protocol are\n\n1. As in protocol 1, terminals in a given geographical area are \ngrouped together. This grouping is made known to all the terminals \n(similar to homenets).\n\n2. A group has a homenet and an assigned transmission network. \nThe homenet is fixed, whereas the transmission net changes cyclically. \nA terminal may, at any time, initiate a transmission only on the \ntransmission network to which its group is then assigned. Once initi- \nated, the transmission may spill beyond this fixed interval, since the \npacket size is not fixed.* As an example, Fig. 4 shows the case of three \ngroups and three sectors of time.\n\n3. A terminal has at least two receivers. When the terminal is idle, \nboth of them listen to the homenet. After establishing a connection, \nhowever, one of the receivers switches to the network on which it has \nestablished connection with the other terminal (and, therefore, the \nnetwork to which it listens changes cyclically) and the other receiver \nremains at the homenet.\n\n*In CSMA/CD the entire packet must be received and CRC checked before the \ndestination address is verified. Thus, if the entire packet is not received before the \nperiod ends, the receiver may miss it. To overcome this problem, a separate CRC is \nprovided for the header information and a source terminal starts a transmission \nsufficiently before the end of a period such that the destination i is able to receive the \nheader information before the period ends. ,\n\nFig. 4\u2014Protocol of Section 3.2. The assignment of group j terminals to net, is made \nby rotating the inner circle at a given speed. The terminals in group j may initiate \ntransmission on net, if the group j pointer is in the sector corresponding to nety,.\n\n4, Terminal A, desiring to transmit to terminal B, goes through the \nfollowing sequence:\n\nd. If net, is idle (during the assigned time slot), then transmit to \nterminal B on net,.\n\ne. If there is no collision, a packet is assumed to have been received \nby its intended receiver. If there is a collision, terminal A ceases \ntransmission immediately and tries again, using a standard retry \nstrategy, but the additional constraint that its starting time must be \nwhen terminal A is allowed to transmit on nety.\n\nf. If communication is to be continued, then terminal B switches \none of its receivers to the net on which terminal A will be transmitting \n(this will periodically switch), and terminal A will set its receiver on \nthe net on which terminal B will be transmitting.\n\ng. If at step f terminal B is already in communication, \u2018ign its \nacknowledgment to terminal A will so indicate. |\n\nh. If after a successful connection there is no communication for a \ngiven amount of time, receivers of both the terminals go back to their\n\nNumber of stations continuously queued to transmit a \npacket; represents the total offered load.\n\nNumber of networks on the frequency-multiplexed cable. \nIl, = Probability that a message from a source contains m pack-\n\nThe average delay in sending a packet (including transmission time) \nwhen Q stations are continuously queued to transmit a packet, is given \nby\n\nThis assumes an optimum retry strategy. Since the Offered Load (OL) \nis Q packets, in terms of bits it is given by\n\nHere if a message from a source contains m packets, then the first \npacket may be transmitted on a different network, but the subsequent \n(m \u2014.1) packets will be transmitted on the homenet of the source. \nThus, the total transmission time is divided into two parts: time to \ntransmit the first packet and time to transmit the remaining packets. \nIt is assumed for simplicity that the length of each homenet is the \nsame and it is 1/N times the total cable length. The following analysis \ncan be easily modified for other configurations.\n\nIf the first packet of a source on homenet / is transmitted on homenet \nk, then\n\nTime for a packet = < + contention time. (3) \nAssume that a source on the jth homenet has probability p;, of wanting \nto communicate with a station on the kth homenet. Further, assume \nthat the number of stations is the same for all nets; distribution, II,,, \nof packets is uniform for all messages; and {p,;,} are a constant* \nindependent of j, k. Then the total traffic offered to the kth homenet \nis\n\n_Q | \nak = N\u2019 (4) \nof which the offered load from out-of-net is \nN-1 \nqk = \u2014N Qi, (5) \nand from within homenet \n\u2014 Q N-1 \nqk = N 1 N IT, }. (6) \nTherefore the time per out-of-net packet is \ntP \nDy = + TI - 1/qi)*-* \u2014 1). (7)\n\nThis transmission is on the homenet itself. Therefore, the time for \neach packet is simply obtained by \ntP T \n=\u2014 +\u2014[(1 \u2014 qi) \u2014 1]. 8 \n| CN [( qk) ] (8) \nThis neglects the traffic generated by first packets of terminals from \nother homenets. It is assumed that the first packet is a small fraction \nof the total message and does not result in any significant traffic.\n\nSince the probability that a message contains k packets is II;, the \naverage time per packet is given by\n\n* We have made no measurements of traffic on real systems to justify this assumption. \nIt is made only so that a closed-form expression can be derived for the delay. If other \nvalues of p;, are more realistic, they can be substituted easily in the equations that \nfollow.\n\nThus the delay versus the offered load characteristic will be given by \nDz versus OL.\n\nWith Protocol 2, all packets are sent on homenet, and the delay for \noptimum strategy is given as\n\nWhen II, is small, that is, messages consist on average of many \npackets, then D3 and Dz differ very little from each other.\n\nIt is possible to compute the optimum number of nets based on the \nabove expressions for average delay per packet. This can be done for \nthe case when Q (and Q/N) is large and the messages contain a large \nnumber of packets, implying that the average delay per packet is \ndominated not by the first packet, but by the subsequent packets. \nFrom eq. (1), for single network, the delay is given by\n\nIf t = N, then the total capacity C is divided among N nets equally. \nHowever, in most cases, the individual N nets may have a capacity \nsuch that the capacities add up to more than C. Thus let t = N/s. In\n\nthis case, each net has a capacity (sC)/N and the total capacity is \ngiven by sC. The delay then becomes\n\nFor large (Q/N) (or small N/Q) we can expand Dp\u00bb as a function of \nN/@Q in Taylor\u2019s series\n\nThus with standard Ethernet* parameters, from the point of view of \naverage delay per packet, N should be 1.\n\nThis implies that as the length of the cable increases, more networks \nare required.\n\nThis implies that if each net is operated at 12.5 Mb/s, adding up to \na total capacity of 12.5 x N Mb/s, average delay is minimized when \nN= 4.\n\nThe average time per packet derived in Section 4.1 was evaluated \nfor a variety of cases and is plotted in Figs. 5 and 6. In all the cases, a\n\nFig. 5\u2014Average delay per packet versus length of the cable. Performance comparisons \nare made between two networks on a cable and a single net.\n\nFig. 6\u2014Average delay per packet versus length of cable. Comparison is made between \nfour networks on a cable and single network.\n\npacket size of 500 bits was used. The curves in Figs. 5 and 6 are for \nQ = 200, that is, 200 packets are continuously queued. It is assumed \nthat a message has 512 packets, that is,\n\nAs expected, the delay increases with length, and depending upon the \nother parameters of the network, the delay corresponding to multiple \nnetwork becomes smaller than that corresponding to single network \nif the length is increased beyond a certain value. Figure 7 shows the \nvariation of the average delay with respect to number of networks. \nThe capacity of each of the nets is equal and is such that the total \ncapacity of all the nets adds up to capacity of the single network. We \nfind that, as expected, for the parameters chosen in Fig. 7, the average \ndelay does show a minimum around N = 4. This verifies our eppron \nmations of the previous section.\n\nFig. 7\u2014Average delay per packet versus number of networks. Each network has a \neapaaly of C/N, where N is the number of nets and C is the capacity of the single \nnetwork.\n\nWe have described a broadband local area computer network. It \nconsists of several local area networks whose data is frequency multi- \nplexed on a single cable. The entire cable length is divided into parts; \neach part is assigned a network and a frequency band for transmission. \nWe have also described two protocols that overcome some of the \nlimitations of the present baseband as well as broadband CSMA/CD \nnetworks. Using our schemes, it is possible to increase the channel \nthroughput and the length of the cable network, reduce the delay due \u2014 \nto collisions, and at the same time allow complete connectivity among \nall the terminals and devices logged into any network. Approximate \nanalytical results are also presented to substantiate these claims.\n\nWe are grateful to Jerry Foschini and Nick Maxemchuck for many \nhelpful discussions.\n\n. The IEEE Project 802, Local Area Network Standards, CSMA/CD Access Method \nand Physical Layer Specification, IEEE P802-3-82/0.1 10, December 1982.\n\n. N. F. Maxemchuck and A. N. Netravali, \u201cA Multifrequency Multiaccess System for \nLocal Access,\u201d Proc. ICC 83, Boston.\n\n6. N. F. Maxemchuck, \u201cA Variation on CSMA/CD That Yields Movable TDM Slots \nin Integrated Voice/Data Local Networks,\u201d B.S.T.J., 61, No. 7 (September 1982), \npp. 1527-50.\n\n. R. M. Metcalfe and D. R. Boggs, \u201cEthernet: Distributed Packet Switching for Local \nComputer Networks,\u201d Commun. ACM, 19, No. 7 (July 1976), pp. 395-404.\n\nOn Binary Differential Detection for Coherent \nLightwave Communication\n\nMotivated by the communication problems caused by phase noise in those \nsemiconductor lasers that may be used for fiber-optic data transmission, we \nconsider heterodyned binary Differential Phase-Shift Keying (DPSK) in \nconjunction with high-rate (short time chip) redundancy as provided by \nrepetition or by more complex coding techniques. In surprising contrast to \nrepetitive coherent phase-shift keying where only a loss of a 2/x (2 db) in \npower is incurred in the limit of infinitely many infinitesimal time chips, we \nshow that DPSK requires, in this limit, an infinite number of photons per bit. \nThis is true regardless of the coding scheme used with the DPSK modulation. \nNext we find the bandwidth expansion that minimizes the number of received \nphotons per bit required to hold the error rate at 10~\u00b0 for two situations: first \nfor a simple repetition code, and then for a repeated (24, 12) Golay code with \nmaximum likelihood detection. The performance of the latter is assumed to \nbe representative of other optimally detected codes of the same rate, such as \nconvolutional codes with Viterbi decoding. Explicit curves relating required \nphotons per bit to the bandwidth expansion are given for B/R ratios of 0.01 \nto 10, where B is the laser linewidth and R is the data rate. An example of the \nresults is that for B/R = 0.1 and a bandwidth expansion of 10, about 23 \nphotons per bit are required for the repeated Golay code to perform as well as \nuncoded DPSK without phase noise (which requires 20 photons per bit for Pe \n= 10-\u00b0). If B/R = 0.01 the bandwidth expansion is reduced to 2, and 12 photons \nper bit are required, thus outperforming the phase-stable, but uncoded, situa- \ntion. :\n\nSemiconductor lasers that may be used for coherent data transmis- \nsion over optical \u2018fibers* have severe phase instabilities. Attempts to \ntransmit a sine wave of frequency f, result in outputs that are modeled \nas\n\nwhere w(t) (measured in radians) is a random process representing \nthe phase instability. The process w(t) is usually taken to be a Wiener \nprocess and that then implies a Lorentzian [see eq. (4)] line shape for \nthe power spectrum of (1). Such spectra are indeed observed and 3-db \nbandwidths as large as 10 to 20 MHz have been measured.\u2019 These \nbandwidths imply that the standard deviation of the change in w(t) \nover a ws can be as large as 47. Severe problems would be encountered \nwith any conventional coherent detection scheme if one is transmitting \ndata at ten-megabit rates (or lower) rather than gigabit rates. Never- \n_ theless, one may wish to do precisely that, and our purpose here is to \nmathematically explore one very natural approach, Differentially co- \nherent Phase-Shift Keying (DPSK) in combination with code symbols \nthat have short transmission time. The purpose of using code chips of \nshort duration is to mitigate the effects of phase wander between \nadjacent chips.\n\nWe emphasize that the scheme we are about to investigate is not \nthe only possible one. One could use on-off keying of the optical carrier \n(1) with photon counting for detection. Theoretically this outperforms \nDPSK by 3 db, even with a stable transmitting carrier assumed for \nthe latter. However, photon counting is not easy to implement, and \npractical avalanche photodiodes can introduce 20 db of loss. Thus \nother techniques, which involve heterodyning, are of interest, in hopes \nthat their implementations can be closer to their own theoretical \nideals. For a general survey of lightwave communications we recom- \nmend Ref. 1. In Ref. 2 a large number of modulation schemes for \ncoherent optics are evaluated with the main purpose being that of \ndetermining the range of B/R values for which coding is not required. \nWe choose here to examine DPSK in detail, but the general behavior \nof its performance with coding is expected to be representative of \nmodulation formats that do not involve tracking the phase w(t) with \na phase-locked loop. The latter was one of the methods considered in \nRef. 2 and was shown to be feasible only if B/R < 0.003.\n\nReturning to the repetition-DPSK scheme, we note that it might be \nexpected that in the limit of an infinite number of infinitely rapid\n\n* In optical-fiber work, coherent transmission refers to any modulation format where \nan optical oscillator is required at the receiver. \nt The carrier wavelength of interest is 1.55 um, or f, is roughly 2 x 10'* Hz.\n\ncode chips, the performance would approach something like that of \nfull interval DPSK with a stable oscillator. One of the surprises that \nwe have uncovered (and perhaps the major point, of interest of this \nwork) is that this is far from the truth. We show that in this limit an \ninfinite number of incident photons per bit are required for fixed error \nrate.\n\nDecreasing the number of repetitions while holding the error rate \nfixed also will ultimately require the photons per bit to become \nunbounded. This occurs when one approaches (from above) the num- \nber of repetitions required to achieve the given error rate, with phase \nnoise being the sole impairment. Consequently, one expects that there \nwill be an optimum bandwidth expansion. This is in fact true, and it \nis discussed in Section IV, while the similar problem for more sophis- \nticated coding is treated in V.\n\nIn Section II we begin by presenting the mathematical details of the \nmodel, while the two limiting cases of large and minimum bandwidth \nexpansion are investigated for the repetition code in Section III.\n\nand set the (two-sided) spectral density of the white noise n(t) to be \nN/2. The variance o2,(t) of w(t) at time t is then :\n\nThe power spectrum of (1) can be calculated in terms of these \nquantities and is given by\u00ae\n\nOften in coherent optics one converts (1) to microwave frequencies \n(GHz) where conventional signal processing techniques are available. \nThis heterodyning is accomplished by mixing (1) with a locally gen- \nerated optical wave. The local oscillator also has phase instabilities \nthat add to those of the received signal, and thus in this paper the \neffective bandwidth at-microwave is taken to be double that at optical \nfrequencies. Furthermore, shot noise fluctuations in photon counts \nduring the heterodyning causes a white noise background to be added \nto the microwave signal. In our model we assume heterodyning to be \ndone, and thus our received unmodulated carrier is modeled as\n\nAssume, momentarily, that \u00a2 = 0 and \u00a2(t) = 0 over a bit interval T, \nand we wish to coherently detect the modulation +A. We simply \nmultiply by cos w,t, integrate the result for T\u2019 seconds, and observe the \nsign of the output. The chance of making an error, Pe, is then\n\nIn (9), E, = A?T/2 is the energy per bit in the transmitting signal. \nWhen (7) arises, as it does in our case of interest, from heterodyning \nof an optical wave, F,/No is not an arbitrary parameter but is (see \nRef. 1) numerically equal to the average number of photons per bit in \nthe optical wave at the receiver. An E,/Npo corresponding to 18 photons \nper bit yields an error rate of 10~\u00b0 for coherent detection.\n\nThe quantity E,/Npo is also numerically. equal to the signal-to-noise \nratio (s/n) if the noise power N is measured in a bandwidth equal to \n1/T.\n\nTo motivate a later discussion, consider the coherent case further \nand instead of integrating the received signal over (0, J\u2019) and making \na decision (what we might unconventionally call soft-decision decod- \ning), we make n = 2m + 1 hard decisions based on time chips of length \nT/n, and then use a majority vote to decide the sign of the transmitted \nbit. If p, is the chip error rate, the bit error rate, Pe(n), would be\n\nSince we want to keep E, constant, the energy in each chip decreases \nas H,/n. For n large, then, we have from (8) and (9)\n\nIn (13) we see the ubiquitous 2/7 penalty in s/n for using hard \ndecisions. Equation (13) was derived for coherent detection. When we \nlater consider repetition-DPSK to overcome phase noise, we will be \nconcerned with a corresponding limit for differential detection. Then \n_ the fortunate limiting behavior we have just observed will not occur, \nbecause, with DPSK, the chip error rate approaches 1/2 more rapidly \nwith n than it does in the coherent case exemplified by (11).\n\nTo make a tractable model for DPSK, we assume that the received \nwaveform is\n\nThe pulse g(t) is assumed brick-wall Nyquist with g(0) = 1, and energy \nT.. The Gaussian noise processes n,(t) and n,(t) are independent, flat \nspectrum, and of equal variance o\u201d = N,/T,. As stated earlier, the chip \ntime, T., for the repetition code is related to the bit time T via T, =\n\nFig. 1\u2014Geometry of variables for differentially coherent phase-shift keying.\n\nT/(2m + 1), and (2m + 1) repetitions* of the bit are differently \nencoded into the chips a, = +1. | \nWe imagine recovering the differentially encoded chips a, by de- \nmodulating (14) with cos w,\u00a2 and \u2014sin w,t and sampling the demodu- \nlation outputs at the appropriate T, second intervals. At each sampling \ninstant, then, we obtain a pair of real numbers, which we regard as a \npoint in the plane. In the absence of phase and additive noise, this \npoint would simply be (+A, 0), relative to a coordinate system fixed \nby the unknown phase @. With noise included, the geometry for two \n\u2018consecutive samples is equivalent to that shown in Fig. 1, drawn for \nthe case of different consecutive a,. All the components of the two \nvectors (A, 0) and (\u2014A, 0) are perturbed by additive independent and \nidentically distributed zero-mean Gaussian noise of variance o\u201d. Owing \nto the phase noise, one of the perturbed vectors is also rotated by an \nangle \u00a2, where \u00a2 itself is a zero-mean Gaussian variable of variance\n\nwhere B is the laser linewidth. If the angle y between the resulting \nvectors is less than 2/2, we would declare that the two consecutive \nchips were the same and, in the present case, an error would be made. \nOf course, for y > 2/2 we decide the chips were different.\n\nFrom (15) we see that as the chip interval decreases, o% approaches \nzero and phase noise will make a negligible contribution to the chip \nerror rate p,. Note also that the additive noise variance o\u201d increases \nas T, decreases.\n\nBefore we proceed, a few comments about the model are in order. \nAn equivalent detection procedure is to delay the signal represented\n\n* More generally, our results for the repetition code are unchanged if the bits are \nmapped into any two complementary patterns of n chips.\n\nin (14) by a chip time, T,, and sample the product of the delayed and \nundelayed signals. The resulting decision statistic is identical to the \nangle that we consider, being a representation of it as a quadratic form \nequal to the inner product of two noisy vectors. Secondly, a more \nrealistic description of the signal uses a pulse g(t) that has unit value \nover the interval T, and is zero elsewhere. The bandwidth of the signal \nthen is not precisely defined, but if one estimates the bandwidth of a \nflat front-end filter required for noise filtering to be 1/T, (and ignores \nany intersymbol interference), then the numerical results are un- \nchanged. Finally, we note that Salz\u201d integrates the product of the \nsignal and the delayed signal, rather than simply sampling. This \ncomplicates the analysis considerably. We have not attempted to \ninvestigate the difference in detail over the full range of B/R values, \nbut we note the following. Salz estimates the B/R value required in \norder that DPSK detection suffers only 1-db degradation compared \nwith B/R = 0 and finds B/R < 0.003 is sufficient. We calculate this \nprecisely for our model and find B/R < 0.002. This suggests that the \npostdetection processing provided by the integrator might not be \nsignificant.\n\nWe use (10) later to calculate the bit error rate for repetitive DPSK \n(with an appropriate p,). It may be objected that the use of this \nformula for the repetitive code is not rigorously justified, since samples \nused for detecting consecutive chips have one noise sample in common \nand thus the chip detection probabilities are not independent. This \nobjection is easily overcome by assuming that the chips for two \nsuccessive bits are interleaved in the manner abab --- .\n\nIn this section we treat two limiting cases of repetitive DPSK. One \ncase is concerned with a very large number of rapid repetitions. Here \nsince the chip interval T, becomes small, the phase noise is neglected, \nand for fixed E, the chip s/n is small. The other case examines the \nminimum number of repetitions required to achieve a fixed error rate \nif phase noise were the only impairment. To approach this limit would \nrequire a large E, provided that more than one repetition is required.\n\nWe begin with the high repetition rate limit. It is well known that \nthe error probability for DPSK (see Ref. 4) is\n\nwhere EF, = E,/n is the energy per chip. For small E, (large n) (16) \nyields for the chip error probability, p,,\n\nNote the difference in behavior for the differentially coherent problem \n[see (18)] versus the coherent one [see (11)]. In (18) the chip error \nrate approaches 1/2 as 1/n, whereas in (11) the behavior is as 1/Vn. \nThus, for DPSK the lower limit of (10) is O(1/Vn) standard deviations \nabove the mean, and for large n we have that the bit error probability \nPe(n) obeys\n\nIn essence, then, we have that if, in Fig. 1, only one of the vectors \nis noisy and \u00a2 is set to zero (coherent phase-shift keying), then Pe(n) \nis small in the limit of many repetitions with constant E,, but if both \nare noisy (DPSK), Pe(n) limits to 1/2.\n\nFor the second limiting case, when the only perturbing influence to \nthe transmission is the Gaussian phase noise variable \u00a2, the chip error \nrate Is\n\nIn (20), p. is explicitly a decreasing function of the number of \nrepetitions via the phase noise variance o4. Further, if Pe(n) is fixed \nin (10), that expression implicitly determines p, as an increasing \nfunction of n. Requiring that both (10) and (20) determine the same \nvalue of p. fixes the number of repetitions required to achieve the bit \nerror probability Pe. Including additive noise in the calculations can \nonly increase the required repetition rate for fixed Pe. Setting Pe = \n10~\u00b0, we have computed the number of repetitions, n, required when \nphase noise is the sole impairment. The probability p, is the chip error \nprobability (20) for n repetitions. Both are displayed in Table I for \nseveral values of B/R. Although our main focus will not be on numer- \nical values of p,, it is worth emphasizing that throughout this paper \nvalues of p, above 0.1, and even approaching 1/2, are possible for the \nlarger values of B/R.\n\nTable I\u2014Minimum number of \nrepetitions A required so that bit \nerror rate does not exceed 107?\n\nThe ratio signal-bandwidth/laser-linewidth equals nR/B. From Ta- \nble I we see that this number is sufficiently high so that the implicit \nneglect of the wideband filtering on the phase noise that was made \nwhen writing the model [see (14)] seems justified for the parameters \nof interest.\n\nIn this section we do a general investigation of DPSK detection \nwith repetitions, including both phase noise and additive noise as \nimpairments. Our main interest will be to determine the optimum \nrepetition rate and the corresponding bits per photon required to \nachieve a bit error probability of 107\u00b0.\n\nThere are probably several useful general expressions for the chip \nerror rate p, when both phase noise and Gaussian noise are present. \nWe shall work with the one given in (22), namely,\n\nwhere p is the chip (s/n) [see (17)], o3(n) is the phase noise variance \nwith n repetitions [see (21)], and J,(p/2) are the modified Bessel\n\nAn expression similar to (22) for constant phase error was first \nderived by Blachman\u2019 and is given in eq. (62) of Ref. 4. Performing a \nsimple average when this angle has Gaussian statistics yields (22). An \nindependent derivation is given in the Appendix.\n\nThe successive terms of the sum in (22) decrease in magnitude, and \nhence, as for any such alternating series, the first neglected term\n\nbounds the error. We find that 15 terms in double precision (single \nprecision on a Cray I) is enough to duplicate (16) when o3(n) = 0, and \np = 20 (p, = 107\u00b0).\n\nWe first use (22) to calculate the deterioration in p, as B/R increases. \nTable II shows some results for p = 20, n = 1. We note that B/R = \n0.002 yields about a 1-db degradation. |\n\nNext in Figs. 2 through 5 we plot as a function of the number of \nrepetitions, n, the number of photons per bit, y, which are required at \nthe receiver to maintain a bit error rate of 10~\u00b0 for B/R values of 0.01 \nto 10. This is done using (10) and (22) in the following way. First n is \nchosen and the required p, is determined from (10). For fixed B/R and \nknown n, og in (21) is known, and (22) is then used to compute the \nchip s/n p that will achieve that p,. Finally, y = np. These figures \nshow quantitatively how the optimum value of n (and the correspond-\n\ning value of y) increase with B/R. From these figures, we derive Table \nIII, which lists n*, the optimum repetition rate, and y*, the minimum \nnumber of photons per bit required to hold the error rate at 10\u00b0\u00b0..The \nlast column in Table III compares ;* with 20, the number of photons \nper bit required at this error rate for DPSK with no repetitions and a \nstable phase.\n\nWe note that the curves are often fairly flat around the minimum, \nand hence less bandwidth may be used without significantly increasing \nthe required number of photons per bit.\n\nThe repetition (or complementary) code for transmitting at rate \nR = 1/T is but one way to use the discrete time Binary Symmetric\n\nTable IV\u2014Minimum and optimum bandwidth expansion for capacity \nand repeated Golay code\n\nChannel (BSC) that we have to work with.* Therefore, to see what is \ntheoretically possible, we next consider the required number of pho- \ntons per bit that would be required to transmit at channel capacity. If \nC is capacity in bits per second, we will fix B/C and plot y versus n \nwhere the chip time is JT, = 1/(nC). That is, the binary code using the \nchips should have a capacity C = 1/n bits per chip. Here n is the \nbandwidth expansion and need not be an integer.\n\nThe required chip error probability is found by equating the channel \ncapacity for the BSC to 1/n, that is,\n\nThe chip s/n p required to yield this p, is then computed from (22), \nand y = np. | \nCurves for B/C = 0.01 to 10 are presented in Figs. 2 through 5, and \nsummarized in the first half of Table IV. Once again, a feature is the \nexistence of an optimum chip rate. The necessity of this is easily \nargued. The lowest value of n possible is determined by the phase \nnoise alone, which causes p, to increase as n decreases. Eventually, \nthe capacity C drops below 1/n, and this fixes the minimum n = fi. \nBut to approach the pure phase noise situation, yy must increase \nindefinitely as n is approached from above. To see why yy must increase \nas n is very large, consider plotting C versus n with y fixed. Then, as \nwe have seen [see (18)], p. = (1/2) \u2014 O(1/n) owing to Gaussian (shot) \nnoise. Setting p, = (1/2) \u2014 \u00ab and expanding the left side of (24) in\n\nQe... ; \nC = \u2014 bits/chip. (25) \nIn2 \n* In fact, interleaving (explained at the end of Section II) creates two parallel BSCs.\n\nThe total capacity is the sum of the capacity of each and is the same as the capacity \ncalculated here as if all chip errors were independent.\n\nSince \u00ab = O(1/n) and we have O(n) chips per second, the capacity \nmeasured in bits per second vanishes like 1/n. To avoid this, y must \nincrease.\n\nFinally, we consider specific coding schemes that are more involved \nthan repetition. We explicitly consider the extended binary (24, 12) \nGolay code. Thus for block length 24, 12 information bits are specified, \nyielding a rate 1/2 code. This is a linear code and any code word has \n759 nearest neighbors at minimum distance dmin = 8. The coding \nscheme that we consider is simply to construct low-rate code words by \nrepeating a given Golay code word J times, make hard decisions on \nthe chips at the receiver (assuming appropriately interleaved DPSK), \nand then use maximum likelihood decoding on the resulting binary \ncode word of length 24J. Note that the new code has dyin, = 8 and \nthat the code rate has decreased to 1/(2/).\n\nAssuming that a word error results, on the average, in six bit errors, \nwe set the word error rate, P,,, to be (1/6) 107\u201d. If p, is the chip error \nrate, then the union bound yields\n\nP, = 759 \u00bb ( ; ) pia peje (hich ) (26) \nwhere \u201chigher terms\u201d represents the probability of decoding into words \nfurther away than the minimum distance. These terms are neglected \nfor the low word error probability of interest here.\n\nWe have in mind that actual attempts at coding would use repeated \nconvolutional codes and Viterbi decoding. Our introduction of Golay \ncodes is simply to make the analysis easier, but overall performance \ngains for the same repeated code rate are expected to be the same. In \nfact, Chase\u00ae finds that a (properly chosen) repeated 16-state convolu- \ntional code performs slightly better than a repeated Golay code. The \neffectiveness of repeating an appropriately chosen code to obtain a \ngood low-rate code was, in fact, proposed by Chase,\u00b0 who was concerned \nwith codes when p, is large, as is often the case for the present \nproblem.* Perhaps better rate 1/(2J) convolutional codes exist than \ncan be generated by repeating the symbols of a given one, but Chase \nshows that, at least from a minimum distance point of view, a repeated \nrate 1/2 convolutional code (suitably chosen) is close to optimum for \ncode rates at least as small as 1/128.\n\nReturning to the Golay code, note that J repeats (of any rate 1/2 \ncode, in fact) corresponds to a bandwidth expansion of n = 2d.\n\n* A point emphasized in Ref. 6 is that for p, > 0.25 and asymptotically large block \nlength, bounded distance algebraic decoders cannot operate (Pe \u2014 1) and maximum \nlikelihood decoders must be used.\n\nare done in a similar manner, as earlier. The bandwidth expansion \nn= 2 is picked, p, is found from (26) with P,, = 1.67 x 107'\u00b0, and \nthen (22) is used to solve for p. The improvements obtained over \nsimple repetition are displayed in Figs. 2 through 5, and essential \nfeatures of the results are given in the second half of Table IV. In \ngeneral, we see that the optimum bandwidth expansion is less than \nwith repetition, and, of course, so is the required number of photons \nper bit. An important feature is that for B/R = 0.01 and 0.1, the \nrequired number of photons per bit is less, or comparable to, that \nrequired when no phase noise is present and no coding is used. In \nthese cases the bandwidth expansion is relatively modest as well.\n\nTo reduce the harmful effects of phase noise, coding schemes that \nuse DPSK detection of code symbols having short duration were \nexamined. We first considered in detail a simple repetition code and \ndetermined the optimum bandwidth expansion that minimized the \nnumber of received photons per bit required for an error probability \nof 10-\u00b0. If B/R = 0.01, we found n* = 5 and y* = 37. The corresponding \nnumbers for a repeated Golay code were n* = 2 and y* = 12.5. By \ncontrast, 20 photons per bit are required for phase stable but uncoded \nDPSK transmission.\n\nThe performance of the repeated Golay code is intended to be typical \nof that obtained with other moderate coding efforts using maximum \nlikelihood detection. In particular, it should be comparable to a re- \npeated 16-state convolutional code (of the same overall rate) with \nViterbi decoding. It is our understanding that the fastest commercially \navailable Viterbi decoders operate at about 20 Mb/s (with 64 states).\n\nMy interest in, and exposure to, this subset of communication \nproblems in fiber optics is due to several enjoyable and informative \ndiscussions with Paul Henry and Jack Salz.\n\n2. J. Salz, \u201cCoherent Lightwave Communications,\u201d AT&T Bell Lab. Tech. J., this \nissue.\n\n3. J. E. Mazo and J. Salz, \u201cSpectra of Frequency ee With Random Wave- \nforms,\u201d Inform. Contr., 9, No. 4 (August 1966), p. 4\n\n4. R. F. Pawula, S. O. Rice, and J. H. Roberts, \u2018 faa ney of the Phase Angle \nBetween Two Vectors Perturbed by Gaussian Noise,\u201d IEEE Trans. Commun., \nCOM-30 (August 1982), pp. 1828-41.\n\n5. N. M. Blachman, \u201cThe Effect of Phase Error on DPSK Error Probability,\u201d IEEE \nTrans. Commun., COM-29 (March 1981), pp. 364-5. |\n\n7. I. S. Gradshteyn and I. M. Ryzhik, Tables of Integrals, Series, and Products, Fourth \ned. New York: Academic Press, 1965.\n\nWe begin by deriving a Fourier series for the angular distribution \ng(01) of a vector (A, 0) perturbed in each component by additive, zero- \nmean, independent and identically distributed Gaussian noise of var- \niance o\u201d. In what follows,\n\npolar coordinates (r, 6,) and integrating over the r variable after \napplying (29), we obtain, after a variable change,\n\nIn our problem, we have a second noisy vector that is also perturbed \nby rotation through \u00a2, \u00a2 being zero-mean Gaussian. The modulo 27 \nangle 6. of this vector has density h(@2), where\n\nE, being expectation with respect to \u00a2. In (31), g(@2 \u2014 @) is evaluated \nby the periodic extension of g(@). We are assuming here that the two \nconsecutive chips are identical, but the error probability is the same \nwhen two consecutive chips have opposite signs.\n\nSince 6, and 6, are independent, the difference angle y = (02 \u2014 0;) \nmod 27 has density p(w) [see eq. [6] of Ref. 4], where\n\nJames E. Mazo, B.S. (Physics), 1958, The Massachusetts Institute of Tech- \nnology; M.S. (Physics), 1960, and Ph.D. (Physics), 1963, Syracuse University; \nResearch Associate, Department of Physics, University of Indiana, 1963-1964; \nAT&T Bell Laboratories, 1964-1985; AT&T Information Systems, 1985\u2014. At \nAT&T Bell Laboratories, Mr. Mazo had been concerned with theoretical \nproblems in data transmission. He now supervises the Data Theory group at \nAT&T Information Systems.\n\nPerformance Signatures for Dual-Polarized \nTransmission of M-QAM Signals Over Fading \nMultipath Channels\n\nPerformance signatures for dual-polarized transmission of M-state quad- \nrature amplitude-modulated signals over dispersive multipath digital radio \nchannels are theoretically derived in this work. We report on two major \nfindings. Firstly, we show that for the assumed propagation model, a cross- \ncoupled interferer exhibits noiselike behavior and impacts on digital radio \noutage time in direct relation to its power level. Secondly, our theoretical \nfinding is based on a new application of performance signature curves for two \ncross-coupled multipath channels. This treatment permits the prediction of \nmultipath-induced digital radio outage for specified ratios of power in the \ncopolarized and cross-coupled signals. Theoretical findings are qualitatively \nsupported by measured performance signatures obtained from a laboratory \nsimulation of the model. |\n\nThe last decade has witnessed a surge of interest in terrestrial digital \nradio transmission, with the newer high-capacity systems relying \nalmost exclusively on single-polarization microwave transmission of \nM-state Quadrature Amplitude-Modulation (M-QAM) signals. In \nthese digital radio systems, the performance degradation associated \nwith multipath propagation has been of paramount importance and\n\nthe subject of considerable prior investigations. For single-polariza- \ntion systems, the effects of multipath are well understood;)\u201d suitable \ncountermeasures\u2014particularly adaptive equalization\u2014have been \nstudied;?* and these countermeasures are now widely deployed in a \nvariety of digital radio systems.*\u00ae\n\nWhen compared with single-polarization systems, dual-polarized \noperation obviously engenders economic and efficient spectrum-utili- \nzation advantages. Unlike single-polarization transmission, however, \nan understanding of the effects of multipath propagation on this latter \nmode of operation is still in its infancy. For single-polarization oper- \nation, countermeasures to frequency-selective fading mitigate inter- \nsymbol interference (ISI) in the presence of Gaussian noise; the \ntransmission of orthogonally polarized signals over the same bandlim- \nited facility is similarly vulnerable to the effects of ISI and noise, but \nnow Cross-Polarization Interference (CPI), normally suppressed by \nthe polarization selectivity of the receiving antenna, is also present. \nConsequently, an optimal receiver must recover the transmitted signal \nin the presence of ISI, CPI, and noise.\n\nKavehrad\u2019 has previously studied dual-polarized M- QAM transmis- \nsion over nondispersive media. He concluded that satisfactory trans- \nmission is not feasible without some form of cross-polarization inter- \nference cancellation. Furthermore, the work showed that in an optimal \ndetection process, the total noise and CPI power must be adaptively \nminimized. In this paper we extended the scope of the previous work \nby considering dual-polarized transmission of M-QAM signals over \ndispersive fading channels like those experienced in line-of-sight ap- \nplications. |\n\nThe dual-polarized channel is modeled using Rummler-like\u00ae multi- \npath transfer functions to describe both the copolarized and cross- \ncoupled paths. The transfer functions emulate snapshots of independ- \nent multipath fading events on the copolarized and cross-coupled \npaths, which, in the presence of noise, limit the achievable error rate \nat the receiver. Our results are predicted on the major assumption that \n\u2018the two simultaneous fading events are statistically independent. \nPerformance \u201csignatures,\u201d defined by a locus of fade notch depths and \nfrequency positions corresponding to a 10\u2122 error rate,\u2019 are used as a \nsystem DeTORMAnce measure since they are conveniently related to \ndigital radio outage.\u2019\u00b0\n\nFor the assumed propagation model, we find that the cross- eae \ninterfering signal exhibits a noiselike behavior. That is, the power loss \nassociated with a cross-coupled signal subject to flat or dispersive \nfading brings about an actual reduction in system outage time. Appar- \nently, the deleterious effect of ISI contributed by dispersive fading of \nthe cross-polarized interference is more than offset by the power loss\n\nof the same associated with the fading event, consequently improving \nnet system outage. Furthermore, this finding is supported by labora- \ntory measurements in which the fading of two independent 16-QAM \nsignals is simulated.\n\nThe theoretical and experimental finding cited above is based on a \nnovel application of the aforementioned system performance signature \ncurves for the propagation model adopted in this work. In particular, \nit permits an immediate comparison of multipath-induced dual-polar- \nized digital radio outage for specified ratios of power in the copolarized \nsignal and cross-coupled interferer.\n\nIn the following section we begin by providing a complete model of \nthe dually polarized communication channel, including the influence \nof frequency-selective fading. In Section HI the computational meth- \nods and performance measurers are discussed. Numerical results and \nlaboratory measurements that support our finding are presented in \nSection IV. Our conclusions are summarized in Section V.\n\nThe system model for dual-polarized operation is illustrated in Fig. \n1. Two independent data sources (one for each of the dual-polarized \nchannels) are assumed to generate complex-valued symbols at the \nbaud rate, 1/T,, where T, is a symbol period. We denote these complex\n\nat consecutive instants kT,, k = 1, 2, 3 ---. The index 7 denotes \nsymbol sets transmitted on the main (i = I) and cross-coupled (i = II) \nchannels. The real and imaginary parts {5}, 61,} take on discrete levels \n+0, +380, ---, +(L \u2014 1)Q with equal probability. Parameter L = JM \nis selected in accord with the number of states in the M-ary signal; \nthe symbols 6 and 6 are independent and identically distributed; and \nQ is a measure of signal-point distance from the nearest decision \nboundary in signal space.\n\nWe assume each of the Pulse-Amplitude-Modulated (PAM) signals \nhas a raised-cosine (Nyquist) spectral shaping with a roll-off factor I. \nThe corresponding time-domain impulse shape is therefore!\n\nand the frequency-domain transfer function is designated P(w). On \neach of the orthogonally polarized channels, the shaped signals are \nmodulated by quadrature carrier signals of frequency f,. Furthermore, \nthe two independent carrier local oscillators for i = I and 1 = II at the \ntransmitter are out of phase by 6@,,, where this phase difference is a \nuniformly. distributed random variable over the range 0 < 6,, < 27. \nThe complex baseband symbol sequences are assumed misaligned by \nTm, Where Tm is also a uniformly distributed random variable over the \nrange 0 <7, = T;.\n\nAs previously stated, we follow Rummler\u2019s prescription for modeling \nmultipath fading. The model is applied to both the copolarized (1 = I) \nand the cross-coupled (1 = II) signal paths and assumes the presence \nof a single inband notch in each of the main and cross-coupled path \ntransfer functions. This latter assumption of notched fading in the \ncross-coupled signal band and notched fading in the main polarized \nsignal band agrees with recent measurements that indicate the possi- \nbility of a shallow fade notch in the interfering cross-coupled path.\u201d\n\nFor our analysis we use \u201cstatic\u201d fading models to emulate \u201csnap- \nshots\u201d of multipath fading events.\u2019* The passband transfer function \nfor the two-ray propagation model can be expressed as\n\nwhere a represents the flat fade level and all other parameters are \nrelated to dispersive fading as follows: the fade notch depth is \u201420 log \n|1 \u2014 p|, where p is the relative amplitude of the secondary ray;* the\n\nrelative delay between rays is 7 and wo is the fade notch frequency. \nFor a fade notch centered in the passband,\n\nIn the absence of a cross-coupled interfering signal, the main polar- \nization baseband waveform after demodulation at the receiver, as \nshown in Fig. 1, can be expressed as\n\nis the low-pass equivalent of the channel impulse response. The index \nI in eqs. (5) and (6) explicitly denotes reference to the main polariza- \n- tion channel; n;(t) is filtered Gaussian noise with variance o7,; and \u00a2y \nrepresents the relative phase difference between the modulator and \ndemodulator oscillators. In terms of a specific sampling epoch, to, \ndistortion in the overall channel impulse response is minimized by an \noptimum relative carrier phase\u00ae\n\nExtending the above discussion to dual-polarized operation, we give \nthe received baseband signal in the main polarization channel as\n\nwhere h; and hy represent the low-pass equivalents of the main and \ncross-coupled paths impulse responses, respectively. As long as the \nmain signal [the first term in eq. (8)] is much stronger than the cross- \ncoupled interferer, the carrier phase tracked by the demodulator of \nthe main polarization receiver is the relative phase difference between \nlocal oscillators of the main polarization modulator and demodulator, \nPtopt(to). In its expanded form, eq. (8), representing the demodulated \u2014 \ncomposite signal at the main polarization receiver, can be expressed \nas\n\nNotice that the parameters of the two independent fading events can \nbe varied independently.\n\nTo establish the relationship of eq. (9) to probability of error, and \nultimately derive the performance signature (M curve)\u2019 that describes \noutage performance for a 107\u00b0 symbol error rate, we can focus on the \nin-phase or quadrature rail signal on the received main polarization. \nBecause the signal constellations are symmetrical, the associated \nprobability of error for each of those two rails is identical and the total \nsymbol error rate can be determined. From eq. (9) the in-phase \ncomponent of the received signal is clearly Re[r;(t)] = r;3(t):\n\npreceding remarks are germane to modeling the channel. In the \nfollowing subsection we relate the received in-phase baseband signal,\n\ngiven by eq. (10), to the associated probability-of-error performance \nof the dual-polarized QAM radio system.\n\nDenoting the coefficient of the desired detected in-phase state as \nZo, we have\n\nwhere fp, the sampling time, is optimized by mimizing the distortion \ncontributed by the second and third terms in eq. (10). The set of \nslicing levels at the in-phase detector of the main polarization receiver \ncan be set to\n\n{\u2014 (L \u2014 2)QZo, Oe gem 20QZo, 0, 20QZo, eeu (L 7 2)QZo}, \nand, with no loss of generality, we set 2 = 1. \nDesignating the sum of ISI contributed by dispersion in the main\n\npolarization and cross-coupled interference at the optimum sampling \ntime to by x(dk, 5k, Bk, Br), eq. (10) reduces to\n\nConsidering the automatic gain control operation at the receiver, the \nassociated probability of error is then\n\nL-1 \nPat, =2 A= PA(niz + x) > Zo}. (13) \nAlso, note that the noise variance o7,, is equal to the total filtered \nnoise variance, subsequently denoted by o2. Additionally, \nPA(x + nit) > Zo} = Ey fPrlnit > (Zo -\u2014 x) 1x = x]}, (14)\n\nwhere E{-} denotes statistical averaging and x is the conditioned value \nof random variable x. First, taking the average over the Gaussian \nnoise, we obtain\n\n1 Zo \u2014- Xx \nP,A(x + nis) > Zo} = 1 Bert c = )p. (15) \nwhere the complementary error function, erfc(e), is defined by \na \nerfc(e) = \u2014= i) e \"dn. 16 \n(e) can 1 (16)\n\nBy symmetry of the constellation, the total symbol error rate can \ntherefore be expressed as\n\nwhere F(x) is the cumulative distribution function for the random \nvariable x.\n\nThe integration in eq. (17) can be evaluated using Gauss Quadrature \nRules (GQR).'4 Equation (17) is thus expressed as\n\nIn this equation, N is the number of terms in the finite series, and w; \nand &; are weights and nodes in the GQR approximation. At this point, \nP. in eq. (18) is calculated from the No = 2N + 1 moments of yx. \nBecause x is functionally dependent upon the independent transmitted \nsymbol states, we have adopted Prabhu\u2019s algorithm\u201d to determine the \nmoments. Note that the moments obtained in this manner are condi- \ntioned on the values of the two random variables 7,, and 6,,. Hence, \nthe resulting moments must be averaged over {tm, 6m} before they can \nbe used in the GQR method.\n\nTo derive the probability of symbol error as a function of signal-to- \nnoise ratio (s/n), we define\n\nwhere we have normalized Q to one. In the next section we expand \nfurther on the computational aspects of the GQR method.\n\nAs previously stated, the relationships above are all dependent upon \nthe timing phase. This parameter can be selected a posteriori to \nminimize P,, thus making probability-of-error computations a formi- \ndable, numerically intensive task. In this work we choose a priori \nsampling epoch by minimizing the peak distortion of the received \nsignal prior to equalization. The timing phase is thus dependent on \nboth the in-phase and quadrature rails of the main polarization M- \nQAM signal.\n\nIn addition to performance signature, we have also evaluated at \nselect points along the M curve a corresponding signal-to-interference \nratio (SIR). This ratio is defined as the relative power in the main \npolarization to that of the cross-coupled interfering signal. Using \nparameters previously defined, this ratio is simply\n\nAn examination of Section IJ demonstrates that the theoretical \nanalysis of M-QAM signal transmission over dual-polarized facilities \nin the presence of multipath fading is a computationally exhaustive \nactivity. In this section we provide a brief overview of numerical issues \nrelated to our investigation. |\n\nDistortion contributed by u;1 and u,n is specifically excluded from eq. \n(23) since the complex data sources I and II are not coherent; these \nlatter signals appear noiselike to the timing recovery circuit in the \nmain polarization receiver.\n\nFrom a computational standpoint, the operations described above \nare carried out by first computing @yopt(to) and D, (to) for all t in the \nrange {[\u2014T,, T,], and then selecting the single sampling epoch, topt, \nthat minimizes D,. With top identified, the symbol-period spaced \nsamples of u;1, Ug, Ui, and Uz are used for the subsequent probability- \nof-error computations. An a priori selection of to,.p, obviously expedites \nthe computer time necessary to perform the probability-of-error com- \nputations. The use of a timing phase that minimizes peak distortion \nis one such choice. Another choice could be minimization of mean- \nsquare eye closure. This can be shown to be equivalent to an IF timing \nrecovery.\n\nFor illustrative purposes, we present plots of u;;, Ug1, Uz, aNd Ug \nin Fig. 2 for pI = 0.80, Pu = 0.75, qi = 1.0, ay = 0.5, for = 1.63 MHz, \nfon = 0 MHz, I = 0.45, tm = 0, Om = 0, and T; = 1/(15 Mbaud). For \nthis illustration example, top, = \u20140.47;, and $y opt = 5.13\u00b0. In particular, \nnote that this carrier phase nulls the u,1 response at t = toopt, as it \nshould. However, even though the cross-coupled interferer corresponds\n\nFig. 2\u2014Dispersive effects on impulse responses of a dual-polarized digital radio \nsystem.\n\nto a notch-centered fade and 0, = 0, Ug # 0 since the carrier phase \noptimization is independent of channel II.\n\nThe signal-power-to-interference-power ratio was evaluated from \neq. (21). The presence of | P(w) |? in this expression derives from the \nfact that at the receiver, the main and interference signals have P(w) \nspectral shaping and thus a |P(w)|* spectral power density. The \na?{1 + p?\u2014 2p;cos[(w \u2014 wo;)7i]}, 1 = I, Il, terms correspond to spec- \ntral power density associated with the dispersive fade and the undis- \ntorted signal power. The integral is conveniently evaluated using \nRomberg-quadrature methods.\n\nAs explained in Section 3.1, the in-phase component of the received \nsignal on the main polarization has basically four parts. The desired \nsymbol to be detected is the k = 0 term of u;;(t) in eq. (22a). Each of \nthe four aforementioned terms consists of a sum of weighted, inde- \npendent, identically distributed random variables that together were \ndesignated as x(6}, Bk, 62, 824) and discussed in Section 3.2. Since the \nrandom variable x is in the form of a sum of sums of independent \nrandom variables, it can be considered a long sum of weighted, inde- \npendent random variables. Thus, to determine a finite number of \nmoments of x, Prabhu\u2019s\u201d recursive algorithm can be used: If\n\nBased on this recursive formula, a computer program was written to \ncalculate No moments of the random variable x.\n\nSince all zero-mean random variables involved in the sum are evenly \ndistributed, the odd moments of the sum become zero. In all our \ncomputations, No = 31 moments were found to be adequate for the \nprobability-of-error calculations.\n\nFollowing the computation of conditional moments of x by averaging \nover uniformly distributed variables 7,, and 6, the absolute moments \nof x were found and subsequently used with the GQR method for \ncomputing the error probabilities. (For a brief summary of GQRs see \nAppendix C of Ref. 17.) For additional detail, interested readers are\n\nreferred to Ref. 14. Observe that by averaging over ai,(i = I, ID), tn, \nand 6,, when computing the No moments wy,, a great deal of compu- \ntation time is saved because these calculations are made once for all \ns/n values.\n\nIn this section we present theoretical and measured performance \nsignature curves for dual-polarized digital radio. It is well known that \nsignature curves provide a locus of fade notch depths (in decibels) and \nrelative fade notch positions (Hz) for a 107\u00b0 probability of error. \nHowever, unlike their more customary presentation, we must also \ninclude parameters that define the character of the interfering cross- \ncoupled multipath channel.\n\nAs a reference, we have computed the signature of a singularly \npolarized 16-QAM system, that is, ay = 0. The data are presented in \nFig. 3 and labeled \u201c1.\u201d Along this contour we also designate the SIR \n(in decibels) computed using methods previously described. For the \ncase of no cross-coupled interferer, the SIR is obviously infinite. For \nall other cases, the SIR value at each point on the signature curve is \nfunctionally dependent upon specific fading parameters for the main \nand cross-coupled signals. Moreover, we associate with each curve a \ntriplet representing the dispersive fading status of the cross-coupled \ninterferer. This triplet is\n\nIn addition to curve 1 in Fig. 3, we show three other cases corre- \nsponding to (\u201430, 10, 0), (\u201430, 5, 0), and (\u201430, 5, 11) and denoted 2, \n3, and 4, respectively. A comparison of curves 2 and 3, with the same \n\u201430 dB flat power levels and 0-MHz notch offsets, reveals that the \nfade of curve 2, with a 10-dB inband notch, results in less outage time \nthan the fade of curve 3, with 5-dB inband notch. Hence, the greater \npower loss associated with curve 2 leads to reduced outage, even though \nthe intersymbol interference for curve 2 exceeds that of curve 3. Now\n\nFig. 3\u2014Performance signature curves for dual-polarized 16-QAM radio; I = 0.45, \nT, = 1/(22.5 Mbaud), s/n = 60 dB. Curves 1 through 4 show relationship of SIR to \nrelative digital radio outage.\n\nconsider curves 3 and 4. This data corresponds to identical flat power \nlevels and fade notch depths, with the notch position moving from 0 \nMHz (notch centered) to 11 MHz (near the band edge). The notch- \ncentered fade apparently causes less outage than the notch offset fade. \nWhen we remember that the unfaded signal spectral energy at 0 MHz \nis much more than that near 11 MHz, it should be clear that the \nrelationship of curves 3 and 4 is again that of diminished net signal \npower in the interferer that equates to a reduced outage.\n\nSIR values are listed at certain points along curves 2, 3, and 4. It \nwill be noted that moving toward higher notch offset values on each \nperformance signature results in greater SIRs, specifically for the \nreason cited above. A fade positioned in a region of normally high \nspectral energy will pull out more power than the same fade in a region \nof reduced spectral energy. Moreover, at any main polarization notch \nfrequency position, the SIR increases as the performance signature \ndrops, because interference signal power, rather than intersymbol\n\ninterference, is the dominant factor for outage due to an independent \ninterferer.\n\nTo further verify that interference power plays a major role in \noutage performance, Fig. 4 presents two relevant cases. In that figure \nwe consider (\u201430, 0, 0) and (\u201425, 5, 0) dispersive fading of 16-QAM \ncross-coupled signals. In both cases the total interference power loss \nis approximately the same. The former case corresponds to a 30-dB \nflat fade of the interferer, while in the latter case the flat fade is 25 \ndB with a 5-dB shaped fade positioned in the center of the passband \n(a region of high spectral energy). Observe that the outage performance \nand SIR are virtually identical along the entire signature curves. We \ntherefore conclude that whether fading of the cross-coupled signal is \nmildly dispersive (i.e., generating intersymbol interference) or flat, \noutage performance improvement is accompanied by interference \npower reduction provided the flat level of the cross-coupled signal is \nconsiderably below that of the main polarization signal (this is nor- \nmally the case because of antenna/waveguide polarization isolation) \nand the dispersive fading of the cross-coupled signal is shallow.\n\nTo confirm that the aforementioned property holds at lower isola- \ntion levels as well, we have repeated the performance signature cal-\n\nT, = 1/(22.5 Mbaud), s/n = 60 dB. The total interference power loss is approximately \nthe same, as is the relative digital radio outage.\n\nculations for an interfering signal flat fade level of \u201420 dB. The \nresults, presented in Fig. 5, confirm those in Figs. 3 and 4.\n\nUp to now 6, and 7, were taken to be random variables. To gain \nfurther insight, we now impose 6,, = 0 and 7,, = 0 conditions on the \ninterfering signal, thereby illustrating a case for which timing and \ncarrier phase of the two signals are aligned. We have repeated com- \nputations for the case presented in Fig. 5. These computations are \npresented in Fig. 6. Observe that the same qualitative properties hold. \nIn the case of a (\u201420, 0, 0) interfering signal, the synchronous system \nexhibits a lower outage time than the general system for the same \ninterferer (see Fig. 5).\n\nData in Figs. 3 through 6 were all computed for a 60-dB s/n, 22.5- \nMbaud symbol rate, [ = 0.45 roll-off, 16-QAM dually polarized radio \nsystem. We have repeated these performance signature computations \nfor a 66-dB s/n, 15-Mbaud, I = 0.45, 64-QAM dually polarized radio \nsystem. The resulting curves, shown in Fig. 7, are expectedly much \nwider than the 16-QAM case. However, the general phenomenon \nexhibited by the data of Figs. 3 through 6 remains appropriate to Fig. \n7, as well.\n\nT, = 1/(22.5 Mbaud), s/n = 60 dB. These cases correspond to lower isolation levels than \nthose in Figs. 3 and 4.\n\nFig. 6\u2014Performance signature curves for dual-polarized 16-QAM radio; IT = 0.45, \nT; = 1/(22.5 Mbaud), s/n = 60 dB, 6, = 0, and 7, = 0\n\nFig. 7\u2014Performance signature curves for dual-polarized 64-QAM radio; I = 0.45, \nT, = 1/(15 Mbaud), s/n = 66 dB.\n\nCHANNEL I \nAl \nDATA | DR-6 IF FADE (+) (+) DR-6 \nGENERATOR TRANSMITTER SIMULATOR RECEIVER\n\nAll \nDATA DR-6 IF FADE ERROR-RATE \nGENERATOR TRANSMITTER SIMULATOR TEST SET\n\n\u2018Fig. 8\u2014Experimental facility for the measurement of dual-polarized digital radio \nperformance signatures.\n\nUsing the experimental arrangements functionally illustrated in Fig. \n8, we have confirmed the theoretical conclusions discussed in the \nprevious section. As noted in the figure, two independent data sources \nprovide inputs to DR-6* transmitters for channels I and II. The flat \npower levels of the two outputs are separately adjusted using atten- \nuators A; and Ay, and dispersive fading is provided by IF fade simu- \nlators. The channel I and II signals are added together with low-level \nnoise (60-dB s/n) from an IF noise load set. The composite signal, \nsimulating independent fading of a main polarization and a cross- \ncoupled interferer, is then applied to a DR-6 receiver and error-rate \ntest set.\n\nAttenuators A; and Ay were used to adjust the ratio ay/a;. Power \nwas measured at the output of the IF fade simulators to establish the \nSIR and again at the input to the radio receiver to assure the demod- \nulator had the appropriate signal level.\n\nError-rate measurements were made for a number of different fad- \ning events. The first set of measurements correspond to the triplets \n(\u20140, 0, 0) and (\u201425, 0, 0), the latter triplet representing a flat, \nnondispersive fade of the interferer. The performance signatures are \npresented in Fig. 9 and show the same trend as the theoretical data of \nFigs. 3 and 7. The quantitative differences are to be expected owing \nto equipment imperfections, such as nonideal Nyquist and bandpass \nfilters, and imperfect carrier and timing recovery circuit operations. \nWe next examined the influence of a dispersive fade characterized by \n(\u201425, 5, 0). As expected, the performance signature improved, but not \nto the point of reaching the (\u2014~, 0, 0) case. This observation agrees \nwith the data of Figs. 3, 5, and 7.\n\nFig. 9\u2014Measured performance signatures for dual-polarized 16-QAM radio; I = 0.45, \nT, = 1/(22.5 Mbaud), s/n = 60 dB.\n\nThe next phase of the experimental study was to increase the flat \nfade level from \u201425 to \u201420 dB and to introduce a notch-centered 5- \ndB fade, thus the triplet (\u201420, 5, 0). The resulting performance \nsignature curve is also shown in Fig. 9 and supports the theoretical \ndata of Fig. 4. In general, conclusions drawn from the experimental \ndata support our theoretical findings. Note that in all of the experi- \nmental tests, the two transmitters were completely independent of one \nanother; hence, the conditions 0 <7, = T, and 0 <= 6, S 27 actually \nheld.\n\nIn this paper we present computed performance signatures for dual-_ \npolarized transmission of M-QAM signals over dispersive multipath \ndigital radio channels. For the assumed model, we show that a cross- \ncoupled interferer exhibits noiselike behavior, and its power loss, \nwhether flat or midly dispersive, brings about an improvement (reduc- \ntion) in dual-polarized system outage time. The theoretical findings\n\nare supported by measured performance signatures obtained from a \nlaboratory simulation of the analytic model.\n\n1. A. J. Giger and W. T. Barnett, \u201cEffects of Multipath Propagation on Digital Radio,\u201d \nIEEE Trans. Commun., COM-29, No. 2 (December 1979), pp. 1870-5.\n\n2. G. J. Foschini and J. Salz, \u201cDigital Communications Over Fading Radio Channels,\u201d \nB.S.T.J., 62, No. 2, Part I (February 1983), pp. 429-59.\n\n3. H. Sari, \u201cA Comparison of Equalization Techniques on 16-QAM Digital Radio \nSystems During Selective Fading,\u201d Globecom \u201982 Conf. Rec. (November\u2014Decem- \nber 1982), pp. F3.5.1-6.\n\n4, N. Amitay and L. J. Greenstein, \u201cMultipath Outage Performance of Digital Radio\n\nReceivers Using Finite-Tap Adaptive Equalizers,\u201d IEEE Trans. Commun., COM- \n32, No. 5 (May 1984), pp. 597-608.\n\n. G. L. Fenderson et al., \u201cAdaptive Equalization of Multipath Propagation for 16- \nQAM 90-Mb/s Digital Radio,\u201d AT&T Bell Lab. Tech. J., 63, No. 8 (October \n1984), pp. 1447-63.\n\n6. C. A. Siller, Jr., \u201cMultipath Propagation: Its Associated Countermeasures in Digital\n\n7. M. Kavehrad, \u201cPerformance of Cross-Polarized M-ary QAM Signals Over Nondis- \npersive Fading Channels,\u201d AT&T Bell Lab. Tech. J., 63, No. 3 (March 1984), pp. \n499-521.\n\n. W. D. Rummler, \u201cA New Selective Fading Model: Application to Propagation Data,\u201d \nB.S.T.J., 58, No. 5 (May-June 1979), pp. 1037-71.\n\n9. M. Emshwiller, \u201cCharacterization of the Performance of PSK Digital Radio Trans- \nmission in the Presence of Multipath Fading,\u201d ICC \u201978 (June 1978), pp. 47.3.1-6. \n10. C. W. Lundgren and W. D. Rummler, \u201cDigital Radio Outage Due to Selective \nFading\u2014Observation Versus Prediction From Laboratory Simulation,\u201d B.S.T.J., \n5 (May-June 1979), pp. 1073-100.\n\n11. R. W. Lucky, J. Salz, and E. J. Weldon, Jr., Principles of Data Communications, \nNew York: McGraw-Hill, 1968.\n\n14. G. H. Golub, and J. H. Welsch, \u201cCalculations of Gauss Quadrature Rules,\u201d Math. \nComp., 26, No. 106 (April 1969), pp. 221-30. |\n\n15. V. K. Prabhu, \u201cSome Considerations of Error Bounds in Digital Systems,\u201d B.S.T.J., \n50 (December 1971), pp. 3127-52.\n\n16. J. J. Kenny, \u201cDigital Radio for 90-Mb/s, 16-QAM Transmission at 6 and 11 GHz,\u201d \nMicrow. J., 25, No. 8 (August 1982), pp. 71-80. |\n\n17. M. Kavehrad, \u201cPerformance of Nondiversity Receivers for Spread Spectrum in \nIndoor Wireless Communications,\u201d AT&T Tech. J., 64, No. 6 (July-August 1985), \npp. 1181-210.\n\nAT&T Bell Laboratories, 1969-1978, 1979\u2014. At AT&T Bell Laboratories, \nMr. Siller has analyzed and designed reflector antennas for terrestrial micro- \nwave communications; performed exploratory investigations of digitally im- \nplemented adaptive transversal equalizers for high-speed digital radio systems; \nidentified novel approaches to digital FIR filtering and quadrature amplitude \nmodulation; and explored new algorithms for the stable control of fractionally \nspaced equalizers. Mr. Siller has authored nearly 30 papers in the aforemen- \ntioned areas and is presently involved in system engineering of future digital \ntransmission systems. He is the recipient of an AT&T Bell Laboratories \nDistinguished Technical Staff Award. Senior Member, IKEE, where he serves \non the Signal Processing and Communication Electronics Technical Commit- \ntee, and has helped plan the technical program for several international \nconferences; Member, Phi Eta Sigma, Eta Kappa Nu, Tau Beta Pi, Phi Kappa \nPhi, Sigma Xi, New York Academy of Sciences.\n\nThere have been many developments and researches in the queueing \nnetworks for modeling of computer systems in the last two decades. \nMost researchers believe, however, that it is very difficult to concep- \ntualize how to derive steady state solutions for general nonproduct \nform queueing networks, which do not have a product form solution.\u2019 \nFor instance, a steady state solution for a general queueing network \nwith first-come-first-served (FCFS) queueing discipline and general \nservice times is unavailable at present. Several recent empirical papers \nhave shown that service time distribution can have a significant effect \non performance with particular emphasis on the modeling of computer \nsystems.\u201d Recently several works describe steady state solutions for \nthe restricted cases of the nonproduct form queueing networks.*\u00ae \nNeuts investigated a single server queue with phase type distribution \nand also found the solution to the M/G/1 queue.\u00ae Carroll et al.\n\nThis paper attempts to generalize these studies. It focuses on a \nclosed network consisting of a nonexponential server and a service \nstation with two identical nonexponential servers in parallel, which is \na typical model of a computer system; a central processing unit and \ntwo input/output processors. There are a finite number of jobs, and \nthe queueing discipline is FCFS at each node. An explicit steady state \nsolution for this network is derived in a quasi matrix geometric form, \nwhere matrices are recursively defined in certain product spaces. \nObviously this form is a generalization of the matrix geometric form. \nThis result may provide insight for obtaining exact solutions for \ngeneral closed networks that do not have a product-form solution.\n\nSection II of this paper introduces notations and definitions for \ndescribing the algebraic description of the general distribution server. \nIn Section III, both external states and internal states are introduced \nto represent the states of the network. In Section IV, the global balance \nequations of the network is derived in a matrix form. Section V gives \nthe steady state solution of the global balance equation. The conclu- \nsions are summarized in Section VI.\n\nOne of the common approaches for representing general service time \ndistributions is to use the method of stages. That is, each nonexpo- \nnential service time distribution is replaced by a subnetwork of expo- \nnential stages with the constraint that the subnetwork can only be \naccessed by one job at a time. The principle on which the method of \nstages is based is the memoryless property of exponential distribution. \nThus this method is both general and compatible with the definition \nof Markov processes. All works to be studied in this paper are based \non Cox\u2019s statement that any server whose service time distribution \nfunction has a rational Laplace transform could be replaced sas \nby a subnetwork of exponential distributions.\u201d |\n\nIn this paper, we consider a closed network consisting of two service \nstations, a nonexponential server (station 0) and a service station \n(station 1) containing two identical nonexponential servers (see \nFig. 1).\n\nLet N denote the total number of jobs (there can be only one job \nactive at any time within each of the servers). When a job completes \nservice at either server of station 1, it leaves station 1 and joins the \nqueue of station 0, while another job (if any) in the queue of station 1 \ntakes its place.\n\nEach of the nonexponential servers in Fig. 1 is represented by a \nsubnetwork of exponential stages, as shown in Fig. 2. There are m; \nexponential stages, whose service rates are p;,, ..., i,m, in the server \nof station i (t = 0, 1). The server of station i can be characterized by \nvectors and matrices. These vectors are denoted by lowercase letters, \nwhereas matrices are denoted by uppercase and bold letters. This \nconvention is followed without exception. The notation that will be \nused is consistent with that of Refs. 5, 7, and 8.\n\nDefinition 1: For each server of station 1, define the following: \np; = the entrance probability vector,\n\nFor instance, (p;)z, the kth component of the vector p;, is the \nconditional probability that a job, upon entering the server of station \ni, will first go to stage k. Similarly, (q;); is the conditional probability \nthat a job, upon completing service at the stage Fk in the server of \nstation 1, will leave the server, and (P;);; is the transition probability\n\nfrom stage k to stage j within the server of station 1. One simple \nexample is provided to show how to characterize the nonexponential \nserver.\n\nExample 1: If a server of station 1 is Erlangian-2 as shown in Fig. 3, \nthen the server is characterized by the following vectors and matrices:\n\nLemma 1: For i= 0, 1, let I; be an identity matrix of order m; and e; a \nvector containing m; ones. The following relationships hold:\n\nProof: Using the law of total probability, we obtain the following \nrelationships:\n\nEquations (1) and (2) are obtained from the above relationships. This \nproves the lemma.\n\nNext, three important matrices in the paper are introduced to \nsimplify the balance equations and express the properties of the server.\n\nNote that Q; is an idempotent matrix of rank 1. We can assume \nthat I; \u2014 P;, i = 0, 1, has an inverse since this expression means that \na path exists from each stage out of the server of station 1. Carroll et \nal.\u00b0 and Neuts* have shown that traditional probability notations can \nbe expressed by vectors and matrices.\n\nLemma 2: For i = 0, 1, let b;(x) be the probability density function (pdf) \nof the server of station 1, E;(x\") its nth moment, E;(x) its mean service \ntime, and B}(s) its Laplace transform.\n\nIf there is only one job present (N = 1) in the system, queueing will \nnever take place. The system is easily solved. If there is more than \none job (N > 1), the state of the network is characterized by a number \nof jobs (called external state) and positions of active jobs (called \ninternal state) at each station. The queueing problem is thus trans- \nformed from one involving the remaining service time for a job to one \ninvolving the position of the active job in the subnetwork.\n\nIn defining the states of the network, it is important to keep the \ninternal state separate from the external state in order to make the \nbalance equation easier to solve. Each external state has a set of \ninternal states which we call the internal state space. Thus the state \nof the network can be represented by a pair, external state and internal \nstate.\n\nThe set of possible external states, which we call the external state \nspace, is determined as follows:\n\nThe cardinality of the external state space is N + 1. Next consider the \ninternal state space. Since the number of internal states at station 0\n\nis the same as the number of stages in the server, the internal state \nspace of station 0 is defined as follows:\n\nIf there is only one job at station 1, it must be active at one of the \nservers. Thus the internal state space for one job at station 1 is defined \nas follows:\n\nWhen there are two or more jobs at station 1, two of them must be \nactive. In this case, a possible internal state would be a pair of integers \n(j, k), where one job is at stage j and the other is at stage k. Since the \ntwo nonexponential servers in station 1 are identical, (Rk, ]) is the \nsame as (j, k). Thus the internal state space for two or more jobs at \nstation 1 is defined as follows:\n\nLet d(i) be the dimension (cardinality) of the state space S;. \nThere are d(0) = mo states in So, d(1) = m, states in S; and d(2) = \nm,(m, + 1)/2 such states in So.\n\nDefinition 3: For N > 2, the state of the network is represented by a \nvector [x1, x2], where x, is a vector representing an external state and \nx2 1s a vector representing the internal states of each station. That is\n\n[(N \u2014n, n), (1,j7)] = the state that N \u2014 n jobs are at station 0, with \ninternal state 1 \u20ac So, and n jobs at station 1, \nwith internal state j \u20ac So, where 2 <n S \nN -1. |\n\nNote that the cardinality of the state spaces is d(0) + d(0)d(1) + \n(N \u2014 2)d(0)d(2) + d(2). Next, the steady state probability vectors are \ndefined similarly.\n\nDefinition 4: For.N > 2, the steady state probability vectors, describing \nthe internal states, are defined as follows:\n\n[bo(N, 0)]; = the steady state probability that the network is in \n| the state [(N, 0), (7,)], where 1 \u20ac So. : \nbo(.N,0)\u00aepi, where \u00ae denotes Kronecker produc \n(see Appendix).\n\n[7i(N \u20141,1)]i; = the steady state probability that the network is \nin the state [(N \u2014 1, 1), (1,7)], where i \u20ac Sp and \nJES,\n\n[r2(N \u2014 n, n)]i,; = the steady state probability that the network is \nin the state [(N \u2014 n,n), (i, j)], where2 <n < \nN-1,1\u20ac So, andj \u20ac Sz.\n\n[b.(0, N)]; = the steady state probability that the network is in \nthe state [(0, NV), (,7)], where j \u20ac So. \nm2(0, N) = po \u00ae b,(0, N).\n\nFor k = 1, 2 and n = 0, 1, ---, N, the steady state probability \n[a1,(.N \u2014 n, n)];,; can be ordered lexicographically to create a vector \nap(N \u2014 n,n). The subscript of these probability vectors denotes the \ndimension of the objects. For instance, the subscript & is used to denote \nthe steady state probability vector of order d(0)d(k). This vector is \nessentially decomposed into d(0) subvectors, each of order d(k). The \nvectors bo( N, 0) and 6.(0, N) are of order d(0) and d(2).\n\nDefinition 5: For N > 2, the steady state probabilities, describing the \nexternal states, are defined as follows:\n\nPr(N, 0) = m(N, 0)(e)* \n= the steady state probability that there are all N jobs \nat station 0. | \nPr(N \u20141,1) = m(N-\u20141, 1)(e\u2122)? \n= the steady state probability that there is one job at \nstation 1. \nPr(N \u2014 n,n) = mo(N\u2014n, n)(e\u2122)? \n= the steady state probability that there are n (2 <n \n< N \u2014 1) jobs at station 1. \nPr(0, N) = 72(0, N)(e)* \n= the steady state probability that there are all N jobs \nat station 1,\n\nNext, objects of the state space S\u00bb are defined similarly. These are \nalso necessary to connect between S, and So.\n\nM, = the diagonal matrix whose (1, i)th element is the probability \nrate of leaving state i = (11, 12) \u00a9 Se. That is, (Me) i = wii, + \nH1,i5 \n(P.);; = the probability of transition from state (k, 1) to state (k, J), \nwhere k \u20ac So, 1 = (h, l2) \u20ac Se andj = (ji, Jo) \u20ac So.\n\nwhere 7 is the other member of the 1-pair, ] is the other member of the \nj-pair, and M means set intersection.\n\n(Rz);; = the probability that upon a job entering station 1, the state \nis changed from (k, 1) to (k,j), where k \u20ac So, i \u20ac Sy, andj =\n\n(Q2)i; = the probability that upon a job completing at station 1, the \nstate is changed from (k, 1) to (k, 7), where k \u00a9 So, 7 \u00a9 Si,\n\nA simple example is provided to show how to construct the objects of \nthe state space So.\n\nExample 2: If station 1 has two Erlangian servers (m, = 2) in parallel \nas shown in Fig. 4, then the internal state spaces and the dimension \nof the spaces are\n\nNote that R\u00bb is d(1) X d(2) dimensional rectangular matrix and Q, is \na d(2) X d(1) dimensional rectangular matrix. Thus Q2R\u00bb is the \nprobability matrix that upon a completion at station 1, a job leaves \nand another takes its place. This event can happen only if there are \nmore than two jobs at station 1.\n\nLemma 3: For the state space S; and So, let I; be an identity matrix of \norder d(t) and e; a vector containing d(t) ones. The following relation- \nships hold:\n\nProof: This lemma follows immediately from lemma 1 and the law of \ntotal probability.\n\nNext, note that we have encountered objects of different state spaces \nSo, Si, and Sg. They are of different dimensions d(0), d(1), and d(2). \nObjects of order d(0)d(1) and d(0)d(2) are said to be on the product \nspaces. 'The product space of order d(i)d(j) is represented by Foca. \nBefore any operation can be performed, all objects have to be replaced \nby their images under embedding into product spaces. For instance, \nobjects of order 1 are scalars. Since scalars constitute subspaces of\n\nspaces So, S;, and S2 and of the product spaces, and there is a 1-1 \ncorrespondence between scalars and diagonal matrices all of whose \ndiagonal elements are equal, scalars are isomorphic to scalar multiples \nof matrix I. Similarly, the spaces So, S;, and S. are subspaces of the \nproduct spaces, so that objects from these spaces have a counterpart \nin the product spaces while algebraic relationships are preserved. \nThese counterparts will be denoted by adding both superscripts and a \nhat, these counterparts are the image of an embedding (for details, see \nAppendix). For instance, Ao is a matrix in the state space So, and A; \nis a matrix in the state space S,. Aj\u2019 is a matrix in the product space \nF.a, and has characteristics inherited from matrix Ao. Also A{\u201d is a \nmatrix in the product space F'4,.4, and has characteristics from Aj. If \nthere is no possibility of confusion, the superscript (0) is omitted. That \nis, AS and A\u00ae are denoted by A; and Ao.\n\nIn order for the network to be in steady state, the probability of \nleaving a particular state must be equal to the probability of entering \nthat state. Thus the total flow into a particular internal state must be \nequal to the total flow out of that state. The steady state balance \nequations can then be derived for the internal states. All internal \nstates belonging to one external state are combined in one mathemat- \nical entity, a vector. The balance equations can then be written in a \nmatrix form, using Kronecker products and generalized embedding of \nvectorspaces (see Appendix).\n\nTheorem 1: For the closed network with a nonexponential server and a \nstation containing two identical nonexponential servers and N > 2, the \nglobal balance equations are:\n\n7m, is a vector in the product space Fa,:4,, whereas the other factors are \nobjects from space So. In order to write eq. (20) in a matrix form, these \nobjects must be augmented into objects from product space Faj.a,. By \npostmultiplying eq. (20) by p; and using generalized embedding, eq. \n(20) can be written in a matrix form where vectors and matrices are \nin product space Fa,xa,:\n\n\u2122(N, 0)M\u00ae = a(N, 0)MYPPYP) + m(N -\u2014 1, Mig? py. (21) \nIt can be simplified by substituting (i; \u2014 P;)\u00e9? for G7 and B; for \nNEI 2B: \nm(N, 0)BS? = m(N \u2014 1, 1)Bi\u00e9? p,. (22) \nBy substituting Q, for 67p,, balance equation (14) is obtained.\n\nBy simplifying eq. (23) and using generalized embedding, balance eq. \n(15) is obtained.\n\nBy simplifying eq. (24) and using generalized embedding, balance eq. \n(16) is obtained.\n\nBy simplifying eq. (25) and using generalized embedding, balance eq. \n(17) is obtained.\n\nBy simplifying eq. (26) and using generalized embedding, balance eq. \n(18) is obtained. That completes the proof.\n\nNext consider the entire Markov chain of this network. The state \nprobability vector of the entire Markov chain can be represented by\n\nThis satisfies balance eq. (10) and the law of total probability. The \ninfinitesimal generator Q* of the Markov chain can be written as \nfollows:\n\nWallace also termed such processes Quasi Birth and Death (QBD) \nprocesses.'* He has shown that if a QBD process is boundary leading, \nthe process has a matrix geometric form solution.\n\nThe global balance equations were derived by equating the flow in \nand out of a certain state. Thus the flow into an external state must \nbe equal to the flow out of the external state because the external \nstate is a collection of states.\n\nLemma 4: (Flow-balance equations between the external states.) For \nN > 2, the flow out of a certain external state equals the flow into the \nexternal state. That ts\n\nProof: The proof will be done in three parts, one part for each of eqs. \n(27) through (29).\n\n(3) By induction on n (n = 2, 3, ---, N-1), \nBasis step: By postmultiplying balance eq. (16) by (e)\u201d, it follows \nthat\n\nThe solution of the global balance equations (Theorem 1) can be \nderived now. A definition is introduced to obtain the steady state \nsolutions.\n\nSince 7;(N, 0) depends on one unknown vector bo(N, 0), we can \nexpress all state probability vector 7\u2019s in terms of it. With notations \ndefined above we are ready to derive the steady state solutions.\n\nProof: The proof will be done in three parts, one part for each of eqs. \n(37) through (39). |\n\nDefinition 8: For N > 2, define recursively the following matrix: \nR,(N \u2014 1) = (BS? + By, \u2014 R,U2(N \u2014 2)M2Q.)\" (45)\n\nThis proves the lemma. \nTo make the solution explicit, a normalization vector must be \ndetermined. The normalization vector can be derived now.\n\nbo(N, 0)BoQo = c(N)po, (48) \n1 \nwhere c(N) = \u2014\u2014\u2014\u2014\u2014 a 15 a Scalar. \ney) PofiK(N)(e\u2122)* \nProof: The normalization vector is proportional to po: \nbo N, 0)BoQo = bo(N, 0)Boes po = c(N) po, (49)\n\nwhere c(N) = bo(N, 0)Boe\u00e9 is a scalar. Thus, the proportionality is \nnormalization constant c(N). By substituting eq. (49) for eq. (47), it \nfollows that\n\nAn explicit solution is derived for a closed network consisting of a \nnonexponential server and a service station with two identical nonex- \nponential servers in parallel (e.g. CPU and I/O devices). There is a\n\nfinite number of jobs, and the queueing discipline at each node is \nFCFS. By applying an algebraic approach to the method of stages, the \nproperties of each nonexponential server are represented by vectors \nand matrices in a space identified with that server. Product spaces are \nintroduced to combine the properties of these servers and to allow \ntheir interactions to be described. Both external states and internal \nstates are introduced to represent the states of the network. That is, \nthe queueing problem is thus transformed from one involving the time \nof service remaining for a job to one involving the position of an active \njob in the subnetwork of exponential stages. In the mathematical view, \nthe problem of integral equations (continuous) is transformed to one \nof algebraic equations (discrete) over a finite dimension.\n\nwhere 7p is a normalization vector chosen to make the steady state \nprobabilities sum to 1, and each U(n) is a matrix which is recursively \ndefined by matrices involved. Obviously this form is a generalization \nof the matrix geometric form. All of techniques used in this paper are \ndrawn from linear algebra. This algorithm is easy to implement, even \nfor a person not familiar with queueing theory, because it involves \nonly matrix multiplication and inversion (we can use commercial \nmatrix packages).\n\nThe author would like to thank Phil J. Fleming and the referees for \ntheir valuable comments and suggestions.\n\n1. R. R. Muntz, \u201cQueueing Networks: A Critique of the State of the Art and Directions \nfor the Future,\u201d Computing Survey, 10, No. 3 (September 1978), pp. 353-9. \n2. M. Reiser, and H. Kobayashi, \u201cThe Effects of Service Time Distribution on System \nPerformance,\u201d Info. Proc. 74, Amsterdam: North-Holland, 1974, pp. 230-4. \n3. C. H. Sauer, and K. H. Chandy, \u201cThe Impact of Distributions and Disciplines on \nyee Processor Systems,\u201d Commun. ACM, 22, No. 1 (January 1979), pp. 25- \noo;\n\n. M. F. Neuts, Matrix-Geometric Solutions in Stochastic Models\u2014An Algorithmic \nApproach, Baltimore: John Hopkins University Press, 1981.\n\n5. J. L. Carroll, A. Van de Liefvoort, and L. Lipsky, \u201cSolution of M/G/1//N-Type \nLoops With Extensions to M/G/1 and GI/M/1 Queues,\u201d Op. Res., 30, No. 3 \n(May-June 1982), pp. 490-513.\n\n.M. F. Neuts, \u201cExplicit Steady-State Solutions to Some Elementary Queueing \nModels,\u201d Op. Res., 30, No. 3 (May-June 1982), pp. 480-9.\n\n_ A. Van de Liefvoort, \u201cAn Algebraic Approach to the Steady State Solution of G/G/ \n1/N Type Loops,\u201d Ph.D. Thesis, Univ. of Nebraska, March 1982.\n\n. S. W. Yoo, \u201cAn Explicit Steady State Solution of Closed Networks Consisting of \nTwo General Stations With Parallel Servers,\u201d Ph.D. Dissertation, University of \nKansas, June 1983.\n\n9. D. R. Cox, \u201cA Use of Complex Probabilities i us a Theory of Stochastic Processes,\u201d \nProc. Cambridge Phil. Soc., 51 (1955), pp. 3\n\n10. A. Graham, Kronecker Products and Maines Calculus With Applications, Ellis \nHorwood \u2018Limited, 1981.\n\n11. V. L. Wallace, \u201cThe Solution of Quasi Birth and Death Processes Arising from \nMultiple Access Computer Systems,\u201d Ph.D. Thesis, University of Michigan, Ann \nArbor, Michigan, March 1969.\n\nThe Kronecker product is the direct product of two disjoint operator \nspaces. In particular, if K is an m X n and L anr X s matrix, then the \nKronecker product of K and L, denoted K \u00a9 L, is the matrix of size \nmr X ns which is obtained by multiplying each element (K);; of matrix \nK by the full matrix L.\n\nIn this paper, there are equations involving matrices of different \ndimensions. Before any operation can be performed, all matrices have \nto be replaced by their images under embedding into product spaces. \nAccording to the definitions in the previous sections, d(i) denotes a \ndimension of state space S;(1 = 0, 1). Let Fu;za, be the additive group \nof d(1) X d(j) matrices.\n\nHen A.1: Define the following mappings between Fa,xa,, Fa,xa,; \n- d.xd. and F dod, xdod,\u00bb F dpdoxdodo as follows:\n\nwhere \u00ae stands for the Kronecker product and i = 1, 2,...,\u00a2 \nThese mappings are group homomorphisms and preserve all alge- \nbraic characteristics of the matrix. The proof is omitted.\n\nSeung W. Yoo, B.S., 1972, Seoul National University, Korea; M.S., 1980, \nPh.D. (Computer Science), 1983, University of Kansas; AT&T Bell Labora- \ntories, 1983-1985, AT&T Information Systems, 1985\u2014. Since joining AT&T \nBell Laboratories, Mr. Yoo has been involved in architecture study and \nperformance analysis of computer systems. He was transferred to AT&T \nInformation Systems in 1985. Member, ACM.\n\nThe transmission performance of digital radio systems is controlled by \nspectral distortion caused by multipath fading. To evaluate this performance \nfor digital systems with high-order modulation schemes, a statistical model \nfor frequency-selective fading is needed. New propagation data obtained in \nGainesville, Florida, were used to generalize Rummler\u2019s model to include group \ndelay response. The introduction of the delay response data into the model of \nthe fading channel enabled the classification of the fades as minimum phase \nand nonminimum phase. We found that 24 percent of all fades have significant \ndelay distortion, and can be characterized as being minimum phase or non- \nminimum phase. In the range of practical interest, there are as many minimum \nphase as nonminimum phase fades. The results of this work will facilitate a \nbetter understanding of the fading channel, which will be beneficial in the \nengineering of radio routes and digital radio design. The results also demon- \nstrate the need for a description of the geographical occurrence of dispersion, \nwhich will differ from that for multipath fading at a single frequency. This is \nbased on the observation, presented in this paper, that the relative amount of \ndispersive fading is significantly greater in Gainesville, Florida, than in Pal- \nmetto, Georgia. The availability of a dispersive fading map will facilitate the \naccurate engineering of digital radio routes.\n\nDigital radio communication systems are sensitive to selective fad- \ning. To evaluate their performance, a statistical model for frequency-\n\nselective fading is needed. Although the amplitude response of multi- \npath fading in a microwave communication channel has been exten- \nsively modeled previously,\u2019\u00ae statistically based results for the group \ndelay response were not available until recently,?\"' and work on \nmodeling the delay in a fading channel is just beginning.\u201d\n\nA successful model for multipath fading using amplitude data was \ndeveloped by W. D. Rummler.** The delay response in this model was \ninferred from the measured amplitude. This led to an ambiguity, since \nboth minimum phase and nonminimum phase fades are possible. In \nthis paper, we generalize the use of this model by including group \ndelay data. The model parameters are defined to characterize both \nminimum phase and nonminimum phase behavior, and statistical \ndistributions of these parameters are compiled. Rummler\u2019s model, \nalthough developed for data obtained in the Palmetto experiment for \na 6-GHz carrier with a 25.3-MHz bandwidth, has proven itself robust \nand applicable for a wide range of frequencies. Our new results are \nbased on amplitude and delay measurements at 6 GHz in a 30-MHz \nbandwidth over a 23:3-mile path between Gainesville and Sparr, \nFlorida.\n\nInclusion of the delay response into the statistical model of the \ndispersive line-of-sight channel will benefit both the engineering of \ndigital radio routes and digital radio equipment design. The description \nof the occurrence of events with minimum and nonminimum phase \nresponse is of immediate practical interest. The need for delay descrip- \ntion, such as provided by our model, will increase with the growth of \nsophistication of digital radio.\n\nSection III presents the statistical distributions of the modified \nmodel parameters for the data measured in Gainesville. Comparison \nwith Palmetto data have been made that show that the Gainesville \nfading is more dispersive than that at Palmetto.\n\ntape. The amplitude and delay were measured during alternating \nfrequency scans. Each scan duration was 0.1 second followed by a 0.1- \nsecond retrace. A measurement period was thus 0.4 second. The \namplitude scale was set up for fades between 15 to 65 decibels and the \ndelay between +50 nanoseconds.* Amplitude fades smaller than 15 \ndecibels were out of scale and recorded at 1 scan/min. Each scan was \nsampled at intervals of 2 MHz or 16 samples over the 30-MHz \nbandwidth. |\n\nThe database used for this study consists of approximately 43,000 \nscans that represent 17,000 seconds of fading activity. The data were \nrecorded in Gainesville for 11 months during the period from 1982 \nthrough 1984.\n\nA fading event recorded in Gainesville is shown in Fig. 1. As a rule, \nfading events were slow enough to be considered static during the \nfrequency scan. Although on some scans a certain amount of distortion \nthat could be attributed to the dynamics of the system was noticeable, \nit was small enough to be negligible. Many scans were incompletely \nrecorded because of the limited range of both amplitude and delay \nmeasurement equipment. Therefore, the recorded fading scans were \nfirst screened in order to eliminate fading scans that are not suitable\n\nEe a a and incomplete amplitude scans. Scans IJ and III rejected. Scan IV \nmarked.\n\nfor modeling. In addition, scans that do not need dispersive modeling, \nbut have to be included in the total fading statistics, were identified \nduring the screening process. This screening was done as follows.\n\nA. Only those responses recorded at a rate of 2.5 scans/s were \nselected for processing. Responses recorded at different rates were \nregarded as unfaded.\n\nB. Data scans were categorized on the basis of amplitude response. \nFigure 2 shows several amplitude scans as they appear in the recorded \ndata. Only complete scans of types I and IV were modeled. Type II \nwas rejected because part of the scan is out of scale at the high end \n(>65 dB). This is a rare event; about 30 such scans were recorded. \nType III was eliminated because the trace is out of scale at the low \nend (<15 dB). This is the most common type that was eliminated; \nabout 20 percent of all scans were type III. Flat fades of type IV were \nmarked to indicate that dispersive modeling is not required.\n\nC. Data scans were also categorized on the basis of delay response. \nFigure 3 shows several delay scans. Scans of types I and II were \nincluded as they are. Flat delay scans of type III that have a dispersive \namplitude response were marked to indicate that delay modeling is \nnot required. Scans of types IV and V are incomplete and were marked \nso that they can be modeled using an alternate technique.\n\nAccordingly, 51,953 scans were selected using criterion A. From \nthese, 9103 were eliminated because the amplitude was out of scale; \n16,892 were marked because the amplitude was flat; 13,580 were \nmarked because the delay was flat; and 2269 were marked because the\n\ndelay was out of scale. Consequently, 12,538 scans, or 24 percent of \nthe total, had sufficient delay information to be modeled using both \nthe delay and amplitude measurements. For the rest, only the ampli- \ntude measurements were used.\n\nThe modeling was done using Rummler\u2019s model for selective fading.? \nIn addition to wide acceptance and extensive validation this model \nhas the advantage of separating the flat component of the fade from \nthe dispersive component. The dispersive component can thus be \nstatistically characterized by one parameter. The transfer function of \nthe model is\n\nwhere we can regard 7 as the relative path delay of the second ray, b \nand wor the relative amplitude and phase of that ray, and a as the \namplitude scale factor.\n\nFig. 4\u2014Amplitude and delay characteristics of the fading model channel response: \na, = 0.1, b; = 0.9, ap = 0.09, and bz = 1.111.\n\namplitude and delay frequency responses based on eqs. (2) and (3) are \nshown in Fig. 4.\n\nAlthough the model was developed using measured amplitude data \nonly, we found it adequate for modeling of selective fading using both \nmeasured amplitude and group delay data. The statistical distribution \nof the parameters naturally differs from those obtained for amplitude \nonly, because we can distinguish between minimum and nonminimum \n- phase scans.\n\nFor convenience, the channel scan is measured in frequency incre- \nments of 2 MHz, resulting in 16 samples over the 30-MHz range\n\nChoosing M = 80, the delay is 7 = 6.25 ns. Thus the in-band frequencies \ncorrespond to n values between 1 and 16, and the model transfer \nfunction is periodic with M = 80, which corresponds to a nee \nperiod 1/7 = 160 MHz (see Fig. 4).\n\nThe first modeling step is to choose a, 8 and w so that the sequence \nof measured values { Y,,}, n = 1, 16, is closely matched by the sequence \n{Y(w,)} using eqs. (4) and (8). Betimates of a, 8 and wo were found by \nminimizing the mean square error, Ems, between these sequences. We \ndefine En5 as\n\nwhere C,, is a weighting coefficient. Since Y,, is derived from data that \nwas uniformly quantized on a logarithmic scale, we use a weighting \nthat 1s approximately logarithmic. Hence,\n\n1/2 \nEas = 53 x (Y,[dB] \u2014 Y(w,)[dB]) i ; (11) \nA plot of the distribution of this error for the total population is shown \nin Fig. 5. As seen from the plot, nearly 99.9 percent of the errors are \nbelow 2 dB.\n\nThe second step in the modeling is to find a and 6 given estimates \nfor a and \u00a3. From these estimates the parameters a and b are obtained \nby inverting eq. (5). The set [a, | has two possible solutions [a,, };] \nand [de, be]\n\nIt is clear from eq. (4) that a = 6 in every scan, since Y(w) must fit a \nsequence of nonnegative powers. Thus, both solutions for 6 are real. \nIt also follows that b,b. = 1.\n\nThe two solutions in eqs. (12a) and (12b) satisfy the amplitude \nresponse fit. The solution [a,, 6:] where b; < 1 corresponds to a \nminimum phase transfer function, since it produces a zero in the left\n\nhalf of the complex frequency plane, and the solution [a2, bz], where \nb. > 1 corresponds to a nonminimum solution, since it produces a zero \nin the right half of that plane.\n\nThe decision of whether a given scan corresponds to a minimum \nphase or nonminimum phase fade is made here by using the delay \ndata. For a substantial fraction of the available data, the delay response \nwas flat. For this population, the estimations were done by using \namplitude data only and assuming the minimum phase solution.\n\nFigure 6 shows a few typical delay scans. T\u2019o determine whether a \nscan is minimum phase or nonminimum phase, we have to examine \nthe shape of the delay response. Scan I shows clearly a nonminimum \nphase delay shape, but so does scan II, although it is negative. The \ngroup delay scans usually exhibit a constant delay offset that has to \nbe subtracted during the fitting process. Scan III is caught in the \ntransition between minimum phase and nonminimum phase; this \nsometimes happens at very deep fades.\n\nThe usual time trajectory of a deep fade is as follows: The fade \nappears as a minimum phase fade, gradually deepens, crosses to a \nnonminimum phase fade, then becomes shallower, and disappears. It \nusually does not cross back to the minimum phase shape.\n\nEach delay scan is sandwiched between two amplitude scans that \nare separated by 0.2 second. The initial parameters wo;, 61; and b2; used \nto fit each delay scan [eq. (3)] were interpolated from the two ampli- \ntude scans. Two error estimates were then set up, namely,\n\nwhere D,,(61, wo) is the minimum phase delay response, and D,,(be, wo) \nthe nonminimum phase delay response, as derived by fitting the \namplitude data. Further, D,, is the measured group delay at frequency \nwn. The group delay scans usually exhibit an unknown constant delay \noffset Do (see Fig. 6, scan JI) that has to be estimated during the fitting \nprocess. \u2014\n\nThe above errors were minimized using a quasi- Newton optimization \nalgorithm. First, E,; and HE, were minimized with respect to the delay \noffset Do. By selecting the response with the better fit, this step was \nsufficient to decide if the shape is minimum phase or nonminimum \nphase. The delay responses thus obtained usually fit within a 7-ns \nroot-mean-square error, for scans having swings smaller than 50 ns. \nThe delay response fits were further optimized, by adjusting the \nparameter b, to yield root-mean-square errors smaller than 4 ns. These \nfurther optimization did not change the values of 6 significantly. The \nEap values [eq. (11)], of the resulting amplitude fits changed by less \nthan 1 dB.\n\nDelay scans with swings larger than 50 ns were out of scale and \ncould not be fitted.* For these scans, the choice between minimum \nphase and nonminimum phase was made by minimizing the error with \nrespect to the offset Do only. The initial parameters, interpolated from \nthe adjacent amplitude scans, were used in the statistics.\n\nOne type of poor fit occurred when a delay scan was in a transition \nfrom minimum phase to nonminimum phase. These scans were elim- \ninated from the modeling.\n\nThe delay data could not be fitted more tightly to the model because, \nas seen from Fig. 7, the residuals have strong harmonic components. \nThe harmonic components indicate long delay echos in the system. \nWe have concluded from our experiments that this distortion is caused\n\n* The instrumentation was changed recently to record delays in the range of +100 \nns.\n\nby multimoding in the antenna and waveguide system of the trans- \nmitter and receiver.\n\nThe statistical distributions for the parameters of the fading model \ntransfer function in eq. (1) were characterized previously using ampli- \ntude response data from the Palmetto experiment.** We used a similar \napproach to generate parameter distributions for the present case. \nEssentially, the data scans can be divided into four distinct groups: \nMinimum-phase fades, nonminimum phase fades, flat amplitude fades \nand flat delay fades. These four groups are characterized best if the \nestimated statistical distributions of the model parameters are de- \nscribed separately. The flat amplitude and flat delay groups are degen- \nerate cases, where only limited modeling was required. A scan with a \nflat amplitude response has also a flat delay response. However, a flat \ndelay scan does not necessarily correspond to flat amplitude response. \nSince the delay notch is narrower than the amplitude notch (see Fig. \n4), many scans with a sloping amplitude response, corresponding to \nan out-of-band notch, have flat delays (see Section 3.5).\n\nThe statistical characterizations of the minimum phase fades and \nnonminimum phase fades are done first, followed by the characteri- \nzations of the flat amplitude and flat delay fades. The statistical \ndistributions of the model parameters for the whole fading population \nare estimated next, based on amplitude data only. This is done in \norder to compare the fading characteristics of the Gainesville channel \nwith those of the Palmetto channel.\n\nFor minimum phase fades the distributions are generated for the \nparameters\n\nwhere a, and b, are the minimum phase solutions of eq. (12a), wo is \nthe notch frequency, B, is the relative notch depth in decibels and A, \nis the amplitude scale factor in decibels.\n\nFor nonminimum phase fades, the parameter b. is unbounded and \nis not easily characterized. Equation (1) is then rewritten as\n\nAs shown below, the choice of these parameters leads to distributions \nvery similar to those of their minimum phase counterparts.\n\n3.3 Distributions for minimum phase and nonminimum phase fades \n3.3.1 The distributions of B, and Bz\n\nThe cumulative distributions of B,; and Bz are shown in Fig. 8. The \nabscissa is B, or By in decibels, and the ordinate shows the time T \nthat B, or B, is greater than the abscissa. It is clearly seen that, if we \nignore the fades with low dispersion (B < 8 dB) and assume that the \ndata above 30 dB have too few samples to be reliable, the two curves \ncan be fitted to within +1.75 dB by one epee distribution. This \ndistribution is\n\nB \nT = 1.37-10' e 482 = 1.87-10* L}*, for 8<B<30dB, (17) \n-B . \nwhere L = 10\u201d, and represents the minimum fading voltage relative \nto the amplitude scale factor a, (or azb2).\n\nAs seen from Fig. 8, there are overall more minimum phase than \nnonminimum phase fades (58 percent). However, if we restrict our- \nselves to the practically significant range of 8 < B < 30 dB, the \nproportion of nonminimum phase fades is 50 percent. Both the mini- \nmum and nonminimum phase fades distributions can be expressed, \nwithin the limits of the approximation, in eq. (17).\n\n3.3.2 The distribution of the parameters A, and A2 \nThe cumulative time distributions for A; and Az are shown in Fig.\n\n9. It is clearly seen that the two distributions are almost identical for \nlarger A (above 14 dB). The means and standard deviations are\n\nA; = 15 GB, o, = 3.0 dB \nA, = 15.5 GB, oo = 3.0 GB. (18) \nThe unnormalized probability density functions (pdf\u2019s) of A; and \nA, are shown in Fig. 10. (These pdf\u2019s are scaled in such a way that the \nordinate displays on a logarithmic scale the number of seconds that A \nwas within +0.5 dB of the abscissa. This same format is used in all\n\nthe pdf\u2019s shown later.) The pdf\u2019s show a definite asymmetry about \nthe mean. This deviation from a Gaussian distribution* could be\n\npartially attributed to the data gathering method used in Gainesville \n(see Section 2.2).\n\nThe means and standard deviations of A; and A2 as functions of B, \nand B, are shown in Fig. 11. These plots show clearly that the \nparameters A and B are uncorrelated for the Gainesville data.*\n\n3.3.3 Possible discrepancies in the A and B distributions caused by \ninstrumentation\n\n1. The amplitude and delay characteristics were sampled every 2 \nMHz by the recording equipment over the 30-MHz frequency scan.\n\nThe 2-MHz resolution may be too low for very deep dispersive fades, \nand cause flattening of these curves during the fitting process. This \nmay cause the distributions of B,; and Bz to appear steeper for very \nlarge B values than they really are. |\n\n2. The Microwave Link Analyzer (MLA) was set up to measure \namplitudes between 15 to 65 dB only. Therefore, some of the shallow \nfading events with small A and small B-are not recorded. This could \ncause the distribution of B for small values to appear flatter than they \nreally are, and the distributions of A to appear less symmetric.\n\nFigure 12 shows the density function of the notch frequency, for \nboth the minimum and nonminimum phase fades, as estimated by the\n\nmodeling procedure. This model estimates that approximately 70 \npercent of the scans have an in-band notch. This is to be expected \nbecause, as seen from Fig. 4, the delay shape is steeper and narrower \nthan the amplitude notch. Therefore, most of the scans that have an \nin-band slope (out-of-band notch) have a flat delay response, and are \ntherefore not characterized in this group (see Section 2.2).\n\nFitting the model to amplitude shapes that have in-band slopes is \nnot unique, since small perturbations in data can produce large vari- \nations of out-of-band notch depth and frequencies. The modeling \nalgorithm in its present form has a tendency to move the notch for \nslope shapes close to the edges of the band. We have tried, with \nvariable success, to use also the delay data in placing the out-of-band \nnotches. However, the delay data for shapes with out-of-band notches \nare limited and heavily corrupted with multimode distortion that made \nthe fitting of the model rather difficult.\n\nAmplitude scans that were flat within 1 dB were assumed to be \nnondispersive. For such scans, therefore, A was the only parameter\n\nFig. 12\u2014Unnormalized probability density function of the notch frequency Fo, mea- \nsured from midband.\n\nmodeled. The cumulative distribution of A for this case is shown in \nFig. 13. As in previous cases, the distribution exhibits a truncated \nGaussian behavior, with\n\nScans that did not have a flat amplitude but exhibit a small delay \ndistortion, D(w)peak to peak = 7 ns, were regarded as having a flat delay. \nScans of this type can be put into two categories:\n\n(a) Scans with in-band, or close to in-band, notches and small \namplitude deviations. A 7-ns delay notch corresponds to an amplitude \nnotch with B = 6.5 dB.\n\n(b) Scans with amplitude slopes that may have a notch so far out \nof band that the delay distortion is negligible.\n\nIn both of the above cases the delay information is too limited to be \nused to decide whether the scan is minimum phase or nonminimum \nphase. Since it is relatively unimportant to differentiate, in these\n\nFig. 13\u2014Unnormalized probability density function of A for flat amplitude fades.\n\nFigure 14 shows the distribution of B. As seen from the plot, 96 \npercent of the scans have B < 6.5 dB, which puts them in the category \n(a) above. Only 4 percent belong to category (b).\n\n3.6 Parameter distributions for the total fade population and comparison \nwith Palmetto\n\nThe distributions of the model parameters for the whole fade pop- \nulation, derived entirely from amplitude response data and comprising\n\nFigure 16 shows the distribution of the parameter B, as measured \nin Gainesville. Within reasonable accuracy the distribution can be \nmodeled by the exponential curve\n\nFig. 15\u2014Unnormalized probability density function of A for flat delay fades.\n\nThe fading data were recorded for 11 months, well distributed over all \nseasons during 1.5 years in 1982-1984. The curve in Fig. 16 can \ntherefore be regarded as an annual average distribution of dispersive \nfading in Gainesville. The other curve in Fig. 16 shows the annual \ndistribution of B for the Palmetto data. This distribution was extrap- \nolated from data measured in Palmetto in 1977.* The total fading for \nthe heaviest month of the year was estimated to be 8000 seconds, \nwhere the total fading was defined as the intercept of the B distribution \nwith the B = 0 axis. To approximate the average annual fading in \nPalmetto, we multiplied the seconds for the heaviest fading month by \na scale factor of 3.5. The average annual fading in Palmetto was thus \nestimated to be 28,000 seconds.\n\nFig. 16\u2014Average annual probability distributions for B for the total populations of \nGainesville and Palmetto.\n\nThe Palmetto distribution is steeper than that for Gainesville and \ncan be modeled over its central region by \nB \nT=1.6-10' e 98 =2.10' 2?% 4dBs Bs 25dB. (22) \nIt can be surmised from the shape of the Gainesville distribution that \nfading on that path is more dispersive than the Palmetto fading. By \nexamining, for illustration purposes, the total dispersive fading for \nB = 20 dB, we see from Fig. 16 that there is five times more of it in \nGainesville than in Palmetto.\n\nFigure 17 shows the probability density function of parameter A. It \nhas a truncated Gaussian shape, with mean and standard deviation\n\nFig. 17\u2014Unnormalized probability density function of A for the total population.\n\nThe corresponding distribution in Palmetto was Gaussian, with mean \nand standard deviation \nAp = 25 dB; op = 5 GB. (24) \nMoreover, the mean Ap was correlated with Bp in the Palmetto \n_ experiment. The Gainesville parameters Ag and Bg are independent \n(Fig. 18). \nThe apparent truncation of the A distribution is probably due to \nthe data gathering method in Gainesville. Specifically, fades below 15 \ndB were not recorded. This could also be a reason why A and B are\n\nuncorrelated in Gainesville. In Palmetto, A was correlated with B, for \nB< 10 dB only.\n\nThe probability density function of the notch frequency for the total \nfade population is shown in Fig. 19. Most of the scans have an in- \nband notch (55 percent). The notch frequency does not exhibit the \nbimodal distribution shown in Palmetto.\n\nThe fading data obtained in the Gainesville experiment were for \nboth amplitude and group delay. We generalized the method of \nRummler*\u201c to include group delay response data, and thus obtain both \nminimum phase and nonminimum phase model parameters.\n\nThe distributions for minimum phase and nonminimum phase fades \ncan be regarded as identical, and can be given by a single description \n[see eqs. (17) and (18)], in the significant range of interest,8 < B< \n30 dB. However, if the shallower fades are included (0 <= B <= 8 dB), \nthere are, overall, more minimum phase fades (58 percent).\n\nFig. 19\u2014Unnormalized probability density function of the notch frequency Fo for the \ntotal population, measured from midband.\n\nnel. This is seen from the estimated average annual distributions for \nthe total fade populations. For the dispersive part of the fading, the B \nparameter, the distribution is steeper in Palmetto. By examining Fig. \n16 for illustration purposes, we can deduce that there is five times \nmore fading with B = 20 dB in Gainesville. In addition, the mean of \nA is smaller in Gainesville; Ag ~ 16 dB, as compared with Ap ~ 26 dB \nin Palmetto.\n\nAT&T long-haul digital-radio performance objectives limit service \nfailure time to 0.02 percent (two way) annually on a 4000-mile route \ndue to all causes. One-half of this is allocated to causes associated \nwith equipment and maintenance. The allocation to dispersive fading, \ntherefore, is 0.01 percent annually.\u2019* The one-way annual dispersive \nfading allocation for the average 25-mile hop is thus 10 seconds.\n\nOne measure of the amount of dispersive fading in a radio channel \nis the distribution of the parameter B. As seen from Fig. 16, a 10- \nsecond annual failure limit in Gainesville requires a robust digital \nradio system designed to operate with notches B < 36 dB. The same\n\nfailure limit in Palmetto requires a radio system designed to operate \nwith notches of B < 26 dB. These comparisons demonstrate the need \nfor a description of the geographical occurrence of dispersion, which \nwill differ from that for multipath fading for a single frequency. The \navailability of such a dispersive fading map would facilitate the accu- \nrate engineering of digital radio routes.\n\nI would like to acknowledge the contributions of G. A. Axeling for \naligning and maintaining the measurement equipment in Gainesville, \nB. J. Engelmann for utilizing her computer expertise to produce the \nstatistical outputs, and G. L. Calhoun for maintaining the database. \nFor many stimulating discussions on this work and comments on this \nmanuscript, I would like to thank W. T. Barnett, L. J. Greenstein, V. \nK. Prabhu, W. D. Rummler, and A. Vigants.\n\n1. G. M. Babler, \u201cA Study of Frequency Selective Fading for a Microwave Line-of- \nSe Narrowband Radio Channel,\u201d B.S.T.J., 51, No. 3 (March 1972), pp. 731-\n\n215. 3 Greenstein, \u201cA Multipath Fading Channel Model for Terrestrial Digital Radio,\u201d \nIEEE Trans. Commun., COM-26, No 8 (August 1978), pp. 1247-50.\n\n. W. D. Rummler, \u201cA New Selective Fading Model: Application to Propagation Data,\u201d \nB.S.T.WJ., 58, No. 5 (May-June 1979), pp. 1037-71.\n\n. W. D. Rummler, \u201cMore on the Multipath Fading Channel Model,\u201d IEEE Trans. \nCommun., COM-29, No. 3 (March 1981), pp. 346-52.\n\nL. J. Greenstein and B. A. Czekaj, \u201cA Polynomial Model for Multipath Fading \nChannel Responses,\u201d B.S.T.J., 59, No. 7 (September 1980), pp. 1197-225.\n\n. M. H. Meyers, \u201cMultipath Fading Characteristics of Broadband Radio Channels,\u201d \nGlobecom\u201984 Conf. Rec., 3 (November 1984), pp. 1460-5.\n\n. W. C. Jakes, Jr., \u201cAn Approximate Method to Estimate an Upper Bound on the \nEffect of Multipath Delay Distortion on Digital Transmission,\u201d IEEE Trans. \nCommun., COM-27, No. 1 (January 1979), pp. 76-81.\n\nL. J. Greenstein and V. K. Prabhu, \u201cAnalysis of Multipath Outage With Applications \nto 90 Mbit/s PSK Systems at 6 and 11 GHz,\u201d IEEE Trans. Commun., COM-27, \nNo. 1 (January 1979), pp. 68-75.\n\n9. M. F. Gardina and A. Vigants, \u201cMeasured Multipath Dispersion of Amplitude and \nDelay at 6 GHz in a 30 MHz Bandwidth,\u201d ICC\u201984 Conf. Rec., 3 (May 1984), pp. \n1433-6.\n\n10. M. Liniger, \u201cSweep Measurements of Multipath Effects on Cross-Polarized RF- \nChannels Including Space Diversity,\u201d Globecom\u201984 Conf. Rec., 3 (November \n1984), pp. 1492-6.\n\n11. L. Martin, \u201cPhase Distortions of Multipath Transfer Functions,\u201d ICC\u201984 Conf. \nRec., 3 (May 1984), pp. 1437-41,\n\n12. M. Sylvain and J. Lavergnat, \u201cModelling the Transfer Function of a 55 MHz Wide \nRadio Channel Multipath Propagation,\u201d ICC\u201985 Conf. Rec., June 1985, 3, pp. \n1541-6.\n\nPhilip Balaban, Diplom Ingenieur (Electrical Engineering), 1950, Technical \nUniversity Munich; Ph.D. (Electrical Engineering), 1965, Polytechnic Insti-\n\ntute of Brooklyn; Scientific Department, Israel, 1950-1956; Contraves, Zurich, \nSwitzerland, 1956-1958; Computer Systems, New York, 1958-1962; Polytech- \nnic Institute of Brooklyn, 1962-1967; AT&T Bell Laboratories, 1967\u2014. Prior \nto joining AT&T Bell Laboratories, Mr. Balaban was engaged in research and \ndevelopment in the areas of circuit design, computers, and communications. \nAt AT&T Bell Laboratories he worked on statistical modeling, simulation and \nanalysis of transmission systems and circuits. Recently he has been involved \nwith problems of channel characterization, and system performance of digital \u2014 \nradio communication systems. In 1965-67 he taught at the Polytechnic Insti- \ntute of Brooklyn, and in 1970-71 he was a visiting professor at the Technion \nin Israel. Senior Member, IEEE; Member, Sigma Xi.\n\nComments on \u201cIntegration With the 5ESS \u2122 Switching System,\u201d by S. A. McRoy, \nJ. H. Miller, J. B. Truesdale, and R. W. Van Slooten*\n\nThis letter amplifies the footnote on page 2426 of the article titled, \n\u201cIntegration With the 5ESS\u2122 Switching System,\u201d in the December \n1984, issue of the AT&T Bell Laboratories Technical Journal.\n\nThe information given here has been used in the field for engineering \nRemote Terminals (RTs) of the SLC\u00ae 96 digital subscriber loop carrier \nsystem on the Digital Carrier Line Unit (DCLU) in a 5ESS switching \noffice. For engineering RT's, the telephone company engineer can \nchoose one of the three service criteria\u20141-1/2 percent blocking, 4 \npercent blocking, and 7 percent blocking. The 1-1/2 percent blocking \ncriteria is recommended for average busy season busy hour engineer- \ning, the 4 percent blocking Once-A-Month (OAM) for extreme value \nengineering, and the 7 percent blocking for high-day engineering. The \nDCLU traffic capacities for the two types (Mode I and II) of RTs \ncorresponding to the three criteria are given in Table I.\n\n1. A two-peripheral-interface-data-bus DCLU has 64 time slots. \n2. A four-peripheral-interface-data-bus DCLU has 128 time slots.\n\nBronson G., Lyon K., Twos Complement Numbers Revisited\u2014A New Tool for \nDealing With the Fundamentals of Number Storage. Byte 10(6):230-231, Jun \n1985.\n\nChung F. R. K., Tarjan R. E., Paul W. J., Reischuk R., Coding Strings by Pairs of \nStrings. SIAM J Alg 6(3):445-461, Jul 1985.\n\nCoffman E. G., Gilbert E. N., On the Expected Relative Performance of List \nScheduling. Operat Res 33(3):548-561, May-Jun 1985.\n\nEhrenreich S. L., Computer Abbreviations\u2014Evidence and Synthesis. Human \nFact 27(2):143-155, Apr 1985.\n\nJohnson C. R., Convergence of a Nonlinear Sharpening Transformation for \nDigital Images. SIAM J Alg 6(3):462-465, Jul 1985.\n\nKapadia A. S., Chiang Y. K., Kazmi M. F., Finite-Capacity Priority Queues With \nPotential Health Applications. Comput Oper 12(4):411-420, 1985.\n\nKlauder J. R., Petersen W. P., Spectrum of Certain Non-Self-Adjoint Operators \nand Solutions of Langevin-Equations With Complex Drift. J Stat Phys 39(1- \n2):53-72, Apr 1985. ,\n\nKumar A., Fine T. L., Stationary Lower Probabilities and Unstable Averages. Z \nWahrsch V 69(1):1-17, 1985.\n\nLagarias J. C., The Set of Primes Dividing the Lucas Numbers Has Density \n2/3. Pac J Math 118(2):449-461, Jun 1985.\n\nMassey W. A., Asymptotic Analysis of the Time-Dependent M/M/1 \nQueue. Math Oper R 10(2):305-327, May 1985.\n\nOddson J. K., Aggarwal S., Discrete-Event Simulation of Agricultural Pest Man- \nagement Systems. Simulation 44(6):285-293, Jun 1985.\n\nPadmanabhan K., Lawrie D. H., Performance Analysis of Redundant-Path Net- \nworks for Multiprocessor Systems. ACM T Comp 3(2):117-144, May 1985. \nSandberg I. W., Iteration and Functional Expansions. Circ Syst S 3(4):409-417, \n1984.\n\nAbramovici M., Menon P. R., A Practical Approach to Fault Simulation and Test- \nGeneration for Bridging Faults (Letter). IKEE Comput 34(7):658-663, Jul 1985. \nAlferness R. C., Veselka J. J.. Simultaneous Modulation and Wavelength Multi- \nplexing With a Tunable Ti-LiNbO3 Directional Coupler Filter. Electr Lett \n21(11):466-467, May 23 1985.\n\nAntler M., Survey of Contact Fretting in Electrical Connectors. IEEE Compon \n8(1):87-104, Mar 1985.\n\nAntreasyan A., Chen C. Y., Logan R. A., Stop-Cleaved InGaAsP Lasers for Mon- \nolithic Optoelectronic Integration. Appl Phys L 46(10):921-923, May 15 1985. \nBanu M., Tsividis Y., On-Chip Automatic Tuning for a CMOS Continuous-Time \nFilter. ISSCC Diges 28:286-287, 1985.\n\nBarber F. E., Eisenberg D. J., Ingram G. A., Strauss M. S., Wik T. R., A 2K x 9 Dual \nPort Memory. ISSCC Diges 28:44-45, 1985.\n\nBishop D. J., Licini J. C., Dolan G. J., Lithium Quench-Condensed Microstructures \nand the Aharonov-Bohm Effect. Appl Phys L 46(10):1000-1002, May 15 1985.\n\nBishop D. J., Spencer E. G., Dynes R. C., The Metal-Insulator Transition in \nAmorphous Nb-Si. Sol St Elec 28(1-2):73-79, Jan-Feb 1985.\n\nChabal Y. J., Infrared Study of the Chemisorption of Hydrogen and Water on \nVicinal Si(100) 2 x 1 Surfaces. J Vac Sci A 3(3):1448-1451, May-Jun 1985.\n\nChen C. Y., Chu S. N. G., Cho A. Y., GaAs/Gao.47Ino.53As Lattice-Mismatched \nSchottky-Barrier Gates\u2014Influence of Misfit Dislocations on Reverse Leakage \nCurrents. Appl Phys L 46(12):1145-1147, Jun 15 1985.\n\nChen C. Y., Dentai A. G., Kasper B. L., Garbinski P. A., High-Speed Junction- \nDepleted Gao.47Ino.53As Photoconductive Detectors. Appl Phys L 46(12):1164- \n1166, Jun 15 1985.\n\nCheng J., Stall R., Forrest S. R., Long J., Cheng C. L., Guth G., Wunder R., Riggs V. \nG., Self-Aligned Ino.53Gao,47As/Semi-Insulating/N+ InP Junction Field-Effect \nTransistors. JEEE Elec D 6(7):384-386, Jul 1985.\n\nChin A. K., Chin B. H., Camlibel I., Zipfel C. L., Minneci, G., Practical Dual- \nWavelength Light-Emitting Double Diode. J Appl Phys 57(12):5519-5522, Jun \n15 1985.\n\nChraplyvy A. R., Stone J., Single-Pass Mode-Locked or Q-Switched Pump Op- \neration of D2 Gas-in-Glass Fiber Raman Lasers Operating at 1.56-ym Wave- \nlength. Optics Lett 10(7):344-346, Jul 1985.\n\nColdren L. A., Boyd G. D., Burrus C. A., Dependence of Chirping on Cavity \nSeparation in 2-Section Coupled-Cavity Lasers. Electr Lett 21(12):527-528, Jun \n6 1985.\n\nDautremont-Smith W. C., Feldman L. C., Structural Damage Produced in InP(100) \nSurfaces by Plasma-Employing Deposition Techniques. J Vac Sci A 3(3):873- \n878, May-Jun 1985.\n\nDautremont-Smith W. C., Woelfer S. M., Stresses in the InP/Ti/Pt and InP/Si0,/ \nTi/Pt Multilayer Systems. App! Phys L 47(1):31-33, Jul 1 1985.\n\nDutta N. K., Wessel T., Olsson N. A., Logan R. A., Yen R., Anthony P. J., Fabrication \nand Performance Characteristics of InGaAsP Ridge-Guide Distributed-Feed- \nback Multiquantum-Well Lasers. Electr Lett 21(13):571-573, Jun 20 1985.\n\nFisanick G. J., Hopkins J. B., Gross M. E., Fennell M. D., Schnoes K. J., Laser- \nInitiated Microchemistry\u2014Dynamic Probes of Metallopolymer Thin-Film De- \ncomposition. Appl Phys L 46(12):1184-1186, Jun 15 1985.\n\nGossard A. C., Two-Dimensional Electron-Gas Systems at Semiconductor Inter- \nfaces. Surf Sci 152(Apr):1153-1166, Apr 1985.\n\nGottscho R. A., Mandich M. L., Time-Resolved Optical Diagnostics of Radio- \nFrequency Plasmas. J Vac Sci A 3(8):617-624, May-Jun 1985.\n\nGray E. W., Damage of Flexible Printed Wiring Boards Associated With Light- \nning Induced Voltage Surges. IEEE Compon 8(1):214\u2014220, Mar 1985.\n\nHaque C. A., Temperature Time Effects on Film Growth and Contact Resistance \nof a Plated Copper Tin Zinc Alloy Used as a Surface Finish on Electronic \nComponents... IEEE Compon 8(1):153-156, Mar 1985.\n\nJindal R. P., Noise Associated With Substrate Current in Fine-Line NMOS \nField-Effect Transistors. IEEE Device 32(6):1047-1052, Jun 1985.\n\nKastalsky A., Luryi S., Gossard A. C., Chan W. K., Switching in NERFET Cir- \ncuits. IEEE Elec D 6(7):347-349, Jul 1985.\n\nKershaw R. N., Bays L. E., Freyman R. L., Klinikow J. J., Miller C. R., Mondal K., \nMoscovitz H. S., Stocker W. A., Tran L. V., Hays W. P., A Programmable Digital \nSignal Processor With 32B Floating Point Arithmetic. ISSCC Diges 28:92, \n1985.\n\nKing W. C., Chin B. H., Camlibel I., Zipfel C. L., High-Speed High-Power 1.3-ym \nInGaAsP/InP Surface-Emitting LEDs for Short-Haul Wide-Bandwidth Opti- \ncal-Fiber Communications. IEEE Elec D 6(7):335-337, Jul 1985.\n\nKoch T. L., Bridges T. J., Burkhardt E. G., Corvini P. J., Coldren L. A., Linke R. A., \nTsang W. T., Logan R. A., Johnson L. F., Kazarinov R. F., 1.55-u4m InGaAsP\n\nDistributed Feedback Vapor-Phase Transported Buried Heterostructure Las- \ners. Appl Phys L 47(1):12-14, Jul 1 1985.\n\nKollaritsen P. W., Weste N. H. E., Topologizer-\u2014\u2014-An Expert System Translator \nof Transistor Connectivity to Symbolic Cell Layout. IEEE J Soli 20(3):799-804, \nJun 1985.\n\nKomal! A., Goskel K., Diodato P. W., Fields J. A., Gumaste U. V., Kung C. K., Lin K. \nY., Lega M. E., Maurer P. M., Ng T. K., An IEEE Standard Floating Point \nChip. ISSCC Diges 28:18-19, 1985.\n\nKomlos J., Greenberg A. G., An Asymptotically Nonadaptive Algorithm for \nConflict Resolution in Multiple-Access Channels. IEEE Info T 31(2):302-306, \nMar 1985.\n\nKoren U., Eisenstein G., Bowers J. E., Gnauck A. H., Tien P. K., Wide-Bandwidth \nModulation of Three-Channel Buried-Crescent Laser Diodes. Electr Lett \n21(11):500-501, May 23 1985.\n\nKull G. M., Nagel L. W., Lee S. W., Lloyd P., Prendergast E. J., Dirks H., A Unified \nCircuit Model for Bipolar Transistors Including Quasi-Saturation Ef- \nfects. IEEE Device 32(6):1103-1113, Jun 1985.\n\nLevy R. A., Vincent S. M., McGahan T. E., Evaluation of the Phosphorus Concen- \ntration and Its Effect on Viscous Flow and Reflow in Phosphosilicate Glass. J \n. Elchem So 132(6):1472-1480, Jun 1985.\n\nLin W., Stavola M., Oxygen Segregation and Microscopic Inhomogeneity in \nCzochralski Silicon. J Elchem So 132(6):1412-1416, Jun 1985.\n\nLocicero J. L., Pazarci M., Rzeszewski T. S., A Compatible High-Definition Tele- \nvision System (SLSC) With Chrominance and Aspect Ratio Improve- \nments. SMPTE J 94(5):546-558, May 1985.\n\nMottine J. J., Reagor B. T., The Effect of Lubrication on Fretting Corrosion at \nDissimilar Metal Interfaces in Socketed IC Device Applications. IEEE Com- \npon 8(1):173-181, Mar 1985.\n\nNordland W. A., Kazarinov R. F., Temkin H., Yen R., Growth of InGaAsP Distrib- \nuted Feedback Lasers by a Modified Single-Phase LPE Technique. Sol St \nComm 55(6):505-507, Aug 1985.\n\nOdajima A., Wang T. T., Takase Y., An Explanation of Switching Characteristics \nin Polymer Ferroelectrics by a Nucleation and Growth Theory. Ferroelectr \n62(1-2):39-46, 1985.\n\nPaalanen M. A., Ruckenstein A. E., Thomas G. A., Enhanced Spin-Lattice Relaxa- \ntion Near the Metal-Insulator Transition. Sol St Elec 28(1-2):121-125, Jan-Feb \n1985.\n\nPeople R., Bean J. C., Lang D. V., Modulation Doping in Ge(X)Si(1 \u2014 X)/Si \nStrained Layer Heterostructures\u2014Effects of Alloy Layer Thickness, Doping \nSetback, and Cladding Layer Dopant Concentration. J Vac Sci A 3(3):846-850, \nMay-Jun 1985.\n\nSharma R., Architecture Design of a High-Quality Speech Synthesizer Based \non the Multipulse LPC Technique. IEEE J Sel 3(2):377-383, Mar 1985. \nStillwagon L. E., Chromatography as a Tool in the Characterization and Quality \nControl of Resist Materials. Sol St Tech 28(5):113-118, May 1985. | \nStoler D., Saleh B. E. A., Teich M. C., Binomial States of the Quantized Radiation \nField. Optica ACTA 32(3):345-355, Mar 1985.\n\nStone J., Optical-Fibre Fabry-Perot-Interferometer With Finesse of \n300. Electr Lett 21(11):504\u2014-505, May 23 1985.\n\nStone J., Chraplyvy A: R., Kasper B. L., Long-Range 1.5 num OTDR in a Single- \nMode Fiber Using a D2 Gas-in-Glass Laser (100 Km) or a Semiconductor- \nLaser (60 Km). Electr Lett 21(12):541-542, Jun 6 1985.\n\nSwaminathan V. et al., Characterization of GaAs Films Grown by Metalorganic \nChemical Vapor Deposition. J Appl Phys 57(12):5349-5353, Jun 15 1985. \nSzymanski T. G., Vanwyk C. J., Goalie\u2014A Space Efficient System for VLSI \nArtwork Analysis. IEEE Des T 2(3):64\u201472, Jun 1985.\n\nTewksbury S. K., Attojoule MOSFET Logic Devices Using Low-Voltage Swings \nand Low Temperature. Sol St Elec 28(3):255-276, Mar 1985.\n\nTomita A., Cohen L. G., Leaky-Mode Loss of the Second Propagating Mode in \nSingle-Mode Fibers With Index Well Profiles. Appl Optics 24(11):1704-1707, \nJun 1 1985.\n\nTung R. T., Gibson J. M., Single-Crystal Silicide Silicon Interfaces\u2014Structures \nand Barrier Heights. J Vac Sci A 3(3):987-991, May-Jun 1985.\n\nVanderziel J. P., Mikulyak R. M., Logan R. A., 7.5-km Bidirectional Single-Mode \nOptical-Fibre Link Using Dual-Mode InGaAsP InP 1.3 um Laser Detec- \ntors. Electr Lett 21(11):511-512, May 23 1985.\n\nWaaben S., Moskowitz I., Federico J., Dyer C. K., Computer Modeling of Batteries \nFrom Nonlinear Circuit Elements. J Elchem So 132(6):1356-1362, Jun 1985. \nWertheim G. K., Kwo J., Teo B. K., Keating K. A.. XPS Study of Bonding in \nLigated Au Clusters. Sol St Comm 55(4):357-361, Jul 1985.\n\nWoody D. P., Miller R. E., Wengler M. J., 85-115-GHz Receivers for Radio \nAstronomy. [EEE Micr T 33(2):90-95, Feb 1985.\n\nBecker R. S., Golovchenko J. A., Swartzentruber B. S., Tunneling Images of Ger- \nmanium Surface Reconstructions and Phase Boundaries. Phys Rev L \n54(25):2678-2680, Jun 24 1985.\n\nBevilacqua R. M., Stark A. A., Schwartz P. R., The Variability of Carbon Monoxide \nin the Terrestrial Mesosphere as Determined From Ground-Based Observa- \ntions of the J = 1* O Emission Line. J Geo Res-A 90(ND3):5777-5782, Jun 20 \n1985.\n\nChemla D. S., Quantum Wells for Photonics. Phys Today 38(5):56+, May 1985. \nCohen R. L., West K. W., Aluminum Spot Weld Strength Determined From \nElectrical Measurements. Welding J 64(6):37-41, Jun 1985.\n\nFisher D. S., Random Fields, Random Anisotropies, Nonlinear Sigma Models, \nand Dimensional Reduction. Phys Rev B 31(11):7233-7251, Jun 1 1985.\n\nFisher D. S., Friedan D., Qui Z. G., Shenker S. J., Shenker S. H., Random Walks in \nTwo-Dimensional Random Environments With Constrained Drift Forces. \nPhys Rev A 31(6):3841-3845, Jun 1985.\n\nFraney J. P., Graedel T. E., Corrosive Effects of Mixtures of Pollutants. J Air \nPollu 35(6):644-648, Jun 1985.\n\nGardner F. F., Hoglund B., Shukre C., Stark A. A., Wilson T. L., Observations of \nOrtho-Thioformaldehyde and Parathioformaldehyde. Astron Astr 146(2):303- \n306, May 1985.\n\nHenkel C., Matthews H. E., Morris M., Terebey S., Fich M., Molecular Lines in IRC \n+ 10-Degrees-216 and CIT-6. Astron Astr 147(1):143-154, Jun 1 1985.\n\nHuse D. A., Henley C. L., Pinning and Roughening of Domain Walls in Ising \nSystems Due to Random Impurities. Phys Rev L 54(25):2708-2711, Jun 24 1985. \nKatz H. E., Chelate and Macrocycle Effects in the 2,2\u2019-Bipyridine N,N\u2019-Dioxide \nComplexation of Alkyltin Trichlorides. J Org Chem 50(12):2086-2091, Jun 14 \n1985. | \nLevy L. P., Nonlinear Spin Dynamics of Quantum Paramagnetic Fluids. Phys \nRev B 31(11):7077-7092, Jun 1 1985.\n\nLi T. G., Lightwave Telecommunication. Phys Today 38(5):24-31, May 1985. \nLovinger A. J., Polymorphic Transformations in Ferroelectric Copolymers of \nVinylidene Fluoride Induced by Electron Irradiation. Macromolec 18(5):910- \n918, May 1985.\n\nLovinger A. J., Belfiore L. A., Bowmer T. N., Crystallographic Changes in Cry- \nogenically Pulverized Polymers. J Pol Sc Pp 23(7):1449-1466, Jul 1985.\n\nMagalhaes F. M., Beccone J. P., Irvin J. C., Perelli S. J., Schlosser W. O., A Microwave \nGaAs-FET Power Module With GaAs Matching Circuits\u2014The M-FET \n(Matched Field-Effect Transistor). Microwave J 28(5):205+, May 1985.\n\nPanek M. G. et al., Thermolysis Rates and Products of the Putative Ketochlo- \nroallyl Groups in Poly(Vinyl-Chloride), as Inferred From the Behavior of \nAnalogous Model Compounds (Letter). Macromolec 18(5):1040-1041, May 1985. \nPeeters F. M., Jackson S. A., Frequency-Dependent Response of an Electron on \na Liquid-Helium Film. Phys Rev B 31(11):7098-7108, Jun 1 1985.\n\nPhillips J. C., Vibrational Thresholds Near Critical Average Coordination in \nAlloy Network Glasses. Phys Rev B. 31(12):8157-8163, Jun 15 1985.\n\nPieranski P. et al., Steps on Surfaces of Liquid-Crystal Blue Phase-I. Phys Rev \nA 31(6):3912-3923, Jun 1985.\n\nRinnert K., Lanzerotti L. J.. Dehmel G., Gliem F. O., Krider E. P., Uman M. A., \nMeasurements of the RF Characteristics of Earth Lightning With the Galileo \nProbe Lightning Experiment. J Geo Res-A 90(ND4):6239-6244, Jun 30 1985. \nRoth H. D., Hutton R. S., Geometrical Isomerism in Divalent-Carbon Interme- \ndiates. Tetrahedron 41(8):1567-1578, 1985.\n\nSermage B., Heritage J. P., Dutta N. K., Temperature Dependence of Carrier \nLifetime and Auger Recombination in 1.3-um InGaAsP. J Appl Phys \n57(12):5443-5449, Jun 15 1985.\n\nStarnes W. H., C-13 NMR-Studies of the Structures and Polymerization Mech- \nanisms of Poly(Vinyl- Chloride) and Related Copolymers. Pur A Chem \n57(7):1001-1008, Jul 1985.\n\nStoffel N. G., Kevan 8S. D., Smith N. V., Experimental Band Structure of 1T-TiSe, \nin the Normal and Charge-Density-Wave Phases. Phys Rev B. 31(12):8049- \n8055, Jun 15 1985.\n\nWebster I. A., Schwier C. E., Bates F. S., Using the Rotational Masking Concept \nto Enhance Substrate Inhibited Reaction Rates\u2014Controlled Pore Supports \nfor Enzyme Immobilization. Enzyme Micr 7(6):266-274, Jun 1985.\n\nWescott L. D. et al., Mechanistic Studies on the Role of Copper-Containing and \nMolybdenum-Containing Species as Flame and Smoke Suppressants for \nPoly(Vinylchloride). J An Ap Pyr 8(1-4):163-172, Apr 1985.\n\nWilson B. A., Kerwin T. P., Harbison J. P., Optical Studies of Thermalization \nMechanisms in A-Si-H. Phys Rev B 31(12):7953-7957, Jun 15 1985.\n\nYurke B., Squeezed-Coherent-State Generation Via 4-Wave Mixers and Detec- \ntion Via Homodyne Detectors. Phys Rev A 32(1):300-310, Jul 1985.\n\nWachter K. W., Becker R. A., Are Productive People to Be Found\u2014Robust \nAnalysis of Sparse Two-Way Tables. J Am Stat A 80(390):266-276, Jun 1985.\n\nGharavi H., Steele R., Conditional Entropy Encoding of Log-PCM \nSpeech. Electr Lett 21(11): 475-476, May 23 1985.\n\nof copies of each single issue \nissue during published \npreceding nearest to filing \n12 months date\n\nI). ree distribution by mail, carrier or other \nmeans, samples, complimentary and other", "title": "magazine :: Bell System Technical Journal :: BSTJ V64N10 198512", "trim_reasons": [], "year": 1985}
{"archive_ref": "belllaboratoriesrecord1958041", "canonical_url": "https://archive.org/details/belllaboratoriesrecord1958041", "char_count": 122501, "collection": "archive-org-bell-labs", "doc_id": 743, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc743", "record_count": 99, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/belllaboratoriesrecord1958041", "split": "test", "text": "THE CO\\ER: It. Behler (left) and J. Novotny discussing \nmaster layouts of printed circuits (see opposite page).\n\nTlie BELL LABORATORIES RECORD is published monthly by Bell Telephone \nLaboratories, Incorporated, 463 West Street, New 1 ork 14, N. 1., M. J. KELLY, \nPresident; W. C. Toole, Secretary; and G. B. Small. Treasurer. Subscriptions: \n$2.00 per \\ear*, Foreign, $2.60 per year. Cheeks should he made payable to \nBell Laboratories Record and addressed to the Circulation Manager. Printed \nin l , S. A. (\u00a7) Bell Telephone Laboratories. Incorporated. 1958.\n\nEDITORIAL BOARD \nF. J. Singer, Chairman \nJ. A. Ilornbeck \nF. A. Korn \nE. T. Mottram \nR. J. Nossaman \nW. E. Reichle \nA. H. White\n\nl util recently, the extent to which complex electronic equipment could he \ndecreased in size as a result of the new family of miniature devices \u2014 tran\u00ac \nsistors, thermistors, semiconductor diodes \u2014 was limited to a large degree \nby the physical space required for conventional wiring. This limitation has \nbeen largely overcome in many applications by the growing use of printed- \ncircuit techniques. Considerable development work has been done in this \narea at the Laboratories, particularly in fundamental studies on raw materi\u00ac \nals, and the physical and electrical characteristics of printed wiring.\n\nThe traditional method of building electronic \nequipment is being challenged today more than at \nany time in the history of the industry. The impetus \nstems from the invention of the transistor, which \nopened a new era of electronic devices, and from \nthe rising costs of fabricating and maintaining the \ncomplex electronic systems planned for telephone, \nmilitary and data-handling systems.\n\nTo cope with this situation, the designers must \nadopt more efficient concepts which will result in \nequipment that is dependable, inexpensive, simple \nto build and easy to maintain. One concept steadily \ngaining recognition in new development projects \nis the \u201cunit assembly\u2019\u2019 or \u201cmodular\u201d technique with \n\u201cprinted circuits\u201d as the basic building block. With \nthis technique, the highly repetitive low-power \nand low-voltage circuitry prevalent in most new \nelectronic systems is divided and packaged into \nlogical groups which perform specific functions, \nsuch as voltage amplification, binary counting, and \nso on. Each of these may be a module consisting of \na single printed-wiring board which contains all of \nthe components, as illustrated in Figures 1 and 2, or \nit mav consist of a printed interconnecting wiring \nboard and several submodules as illustrated in Fig\u00ac\n\nure 3. In either case, the printed wiring board is the \nbasic unit of the system.\n\nThe term \u201cprinted wiring\u201d implies that the con\u00ac \nductors which interconnect conventional electrical \ncomponents are produced on an insulating base by \nany process which does not employ conventional \nwiring. \u201cPrinted circuits,\u201d often used interchange\u00ac \nably with \u201cprinted wiring,\u201d implies that resistors, \ncapacitors and inductors are also printed.\n\nThere are at least twenty known methods of pro\u00ac \nducing printed wiring. Among these, six of the most \ncommonly used may be generally classified as the \n\u201cprinted or painted wire,\u201d the \u201cplated wire,\u201d the \n\u201cembossed wire,\u201d the \u201cstamped wire,\u201d the \u201ctransfer\u201d \nand the \u201cetched wire\u201d processes.\n\nIn the printed or painted wire processes, finely \ndivided silver or copper dispersed in a lacquer or \nink carrier is applied to an insulating board by \nspraying or brushing through a stencil or by offset \nprinting. The printed path is then made conductive \nby removing the organic carrier which separates \nthe metallic particles. In the case of ceramic boards, \nthis is done by the application of heat. Where plas\u00ac \ntics are used, volatile solvents are normally incor\u00ac \nporated in the carrier; also a small amount of\n\nFig, 1 Printed circuit boards used to intercon\u00ac \nnect com ponents of a single module as used in the \namplifier circuits of video transmission equipment .\n\nresidual resin hinder provides the necessary ad\u00ac \nhesion between the metallic particles and the base.\n\nA plated wire board is produced by depositing \na conductive pattern oil an insulating material by \nchemical reduction, electroplating or a combination \nof both. There are several variations of this process \nwhich are used primarily where it is desired to con\u00ac \ntinue the printed wiring through a hole or around \na molded contour.\n\nEmbossed \\\\ ire boards are produced bv placing \na copper foil over a laminated insulating board and \ndepressing the conductor pattern into the board \nwith a heated die. The unwanted portion of the \nfoil is removed by a mechanical surface-abrading \noperation which leaves the conductors imbedded \nin the board.\n\nStamped wiring is similar to embossed wiring ex\u00ac \ncept that a \u201cdinking die\u201d is used: this simultaneously \ncuts the wiring pattern from the foil and places it \non the surface of the insulating board. The conduc\u00ac \ntors are usually made to adhere to the board in a \nsecondary pressing operation.\n\nIn the transfer process, a mirror image of the \nprinted w iring is formed on the surface of a car\u00ac \nrier sheet by electroplating, etching or stamping. \nThis image is transferred to the insulator by a \nsubsequent pressing operation.\n\nEtched wiring, which employs many of the tech\u00ac \nniques familiar to the graphic arts industry, is by \nfar the most popular of the currently used methods. \nMost of the mechanics of this type of manufacturing \nare well established, and the copper-clad plastic \nraw materials had been under development by the \nlaminating industry for several years prior to the \nfull scale emergence of printed wiring. The process \nis adaptable to development type work where it is\n\ndesirable to obtain small quantities of a large variety \nof circuit patterns within a short time. These are \noften used for laboratory development or studv.\n\nThe photo-etch process originates with an en\u00ac \nlarged layout of the conductor pattern. This draw\u00ac \ning is called a \u201cmaster\" and is produced bv applying \nprecut strips of opaque, nonreflecting, pressure-sen\u00ac \nsitive tape onto a dimensionally stable Mylar film. \nThe reverse side of this translucent film is printed \nw ith a grid coordinate system which is used as a \nguide by the draftsman for laying out the conduc\u00ac \ntors in modular dimensions. Thus, each hole for a \ncomponent mounting lead and each conductor is \ndesignated bv cartesian coordinates.\n\nThe master is photographed and reduced to ac\u00ac \ntual size with a precision copy camera. Anv draft \ning errors on the master are thus minimized on the \nresulting negative, which may then be used to pro\u00ac \nduce the conductor pattern on the circuit board. \nThis negative is used to produce silk screens or off\u00ac \nset printing plates which in turn are used to print \nthe pattern on the board. The circuit board is a \ncopper-clad laminate, generally' one sixteenth of an \ninch thick, from which portions of copper are etched \na wav to leave a pattern of conductive lines.\n\nWith the photo-engraving technique, a photo\u00ac \nsensitive \u201cresist\u201d (acid resistant lacquer or enamel) \nis applied to each copper surface from which the \ncircuit is to be formed. The negative of the con\u00ac \nductor pattern is placed over the sensitized surface \nand exposed to ultraviolet light which renders the \nconductor pattern insoluble in the developer. The \nareas protected from the light by the opaque por\u00ac \ntions of the negative remain soluble so that the \nresist is removed in the developer. The unprotected \nareas of copper are then etched away to leave a\n\nFig. 2- Printed circuit board used to interconnect \nthe components of a single die-cast module unit.\n\nAfter the resist material is removed and the board \nis cleaned, a thin coating of solder is applied to the \nconductors to provide protection during storage \nand to enhance solderability during subsequent \nassembling operations. This coating is applied by a \n\u201croller-coating\u201d process wherein the fluxed board \nis passed through two rollers. One of these is tinned \nand partially immersed in molten solder; the other \napplies uniform pressure between the board and \nthe tinned roller to insure that there is good wetting \nof the conductors.\n\nr ig. 3 \u2014 An example of printed H iring used to interconnect \nseveral sub-modules on a single board as used in a frau\u00ac \nds! orized high-speed digital computer circuit.\n\npoint-to-point wiring. In addition, it is possible \nto incorporate small amounts of capacitance, in\u00ac \nductance and shielding directly on the board and \nthus reduce the required amount of cabling and \nshielded wires. The size reduction is compounded \nin the over-all system by reducing the need for \nheavy structures to support the many heavv metal \nchassis and cables. This is illustrated in Figure 4 \nby a printed coaxial line on a circuit board for a \ntransistor video amplifier.\n\nMaintenance of electronic svstems is improved \nby the use of printed-circuit unit assemblies which \ncan be replaced much as electron tubes are replaced, \nand can be checked as functional units. Also, the \ntime required for trouble shooting individual cir\u00ac \ncuits is reduced because the circuit path is dis- \ntinctlv outlined in the printed-wiring pattern.\n\nReliability is improved in several ways. The de\u00ac \nsigner is given greater control of the precise de\u00ac \nployment of conductors in the manufactured units. \nThis minimizes the possibilities of shorting and \nvarying stray capacitances, and results in a more\n\nFig. 4 \u2014 Coaxial line (arroics) produced by printed circuit \ntechniques on transistorized video amplifier .\n\nuniform product from which wiring errors have \nbeen virtually eliminated. In addition, the num\u00ac \nber of soldered connections can be greatly reduced, \nsince it is possible to have any number of conductors \nemanating from a single point, and three or more \ncomponents can be joined with a single continuous \nstrip of copper.\n\nThe development of printed circuits has consid\u00ac \nerably advanced the art of automatic or mechanical \nassembly of electronic equipment. Unlike the con\u00ac \nventional circuits of ten years ago with their maze \nof insulated wires, stand-offs and mounting termi\u00ac \nnals, hand wrapped and soldered joints and difficult \nassembly sequences, the printed-circuit board is \nreadily adapted to large-quantity production and is \nideally suited to mechanical insertion of compo\u00ac \nnents and mass soldering. Even in those instances \nwhere mechanical assembly is not feasible for ec\u00ac \nonomic reasons, the same design considerations re\u00ac \nsult in units which are easier to assemble manually.\n\nThere are some disadvantages; special precautions \nmust be taken, for example, to extend the field of \napplication beyond that of low-power circuits. In \nthis regard, systems designed to use printed circuits \nhave a far greater proportion of insulating material \nthan is normally encountered in conventionally \nwired equipment. This puts the burden of respon\u00ac \nsibility for satisfactory operation on the insulating \nmaterials, adhesives, conductor finishes and pro\u00ac \ntective coatings associated with the printed-circuit \nboard. Except for the metallic finish and some \nceramic bases, most of the materials used are or\u00ac \nganic plastics. High temperature is by far the most \nruinous environment for these materials \u2014 in fact,\n\nFig. 5 \u2014 Rotary switches prod it rod by printed circuit tech\u00ac \nniques; right \u201e time sharing and left , zero set.\n\nsome of the present materials arc 1 combustible. \nHence, special precautions must be taken to con\u00ac \ntrol heat dissipation.\n\nInsulation resistance is also extremely important \nin the development and use of printed circuits. \nOn a conventional plastic terminal board, the leak\u00ac \nage path is governed bv the spacing between termi\u00ac \nnals, since the effects of the strap wires are usually \nnegligible. Conductors on a printed-circuit board, \nhowever, are in direct contact with the insulating \nbase throughout their entire length. Thus, the leak\u00ac \nage between two adjacent parallel conductors may \nbe appreciable. It is, therefore, important to investi\u00ac \ngate this condition early in the design stage, and \nto employ materials which will insure proper per\u00ac \nformance under varying conditions of temperature \nand humiditv.\n\nIonic contamination of the board during etching \nor plating, and staining of the surfaces by perspira\u00ac \ntion during handling, may result in a malfunction. \nThis is due to corrosion or high leakage if the as\u00ac \nsembled unit is placed in a humid environment. \nThus, it is necessary to cleanse the board thoroughly \nand to coat the surfaces with some type 1 of insulat\u00ac \ning material. In addition, some plating solutions \nand improper soldering methods can destroy the \nbond and separate the printed conductor from the \ninsulating base. Therefore, special soldering pre\u00ac \ncautions must be exercised when replacing com\u00ac \nponents during servicing.\n\nThe Laboratories has concentrated its efforts on \nfundamental studies of raw materials, the physical \nand electrical characteristics of printed wiring \nunder various environments, and the exploration \nof methods for fabricating high-quality printed wir\u00ac \ning. This was done to learn more about the capa\u00ac \nbilities of the materials and the processes so that \nrealistic specifications could be established to in\u00ac\n\nsure a quality level consistent with the design in\u00ac \ntent of the Laboratories.\n\nThis program resulted in the development of a \nseries of specifications, one of which contains pro\u00ac \ncedures for testing the critical parameters of the \nbase material or of the finished printed circuit. \nOne test is for insulation resistance, another deter\u00ac \nmines moisture sorption, a third gives a measure \nof adherence of the copper to the insulator and a \nfourth indicates the materials\u2019 resistance to blister\u00ac \ning when subjected to soldering temperatures. Sev\u00ac \neral specimens for each test are printed on one \nlarge panel in a manner which takes into account \nvariations in properties due to orientation.\n\nContamination of the board due to processing \nhas been virtually eliminated in a process called the \nApplique Method which was invented by the late \nE. E. Franz of the Laboratories. It is a transfer \nprocess wherein the raw materials consist of a com\u00ac \nposite sheet of aluminum and copper foil bonded \ntogether with a pressure-sensitive adhesive and a \nfully cured plastic laminate sheet. The exposed \ncopper surface is treated with a partially cured \nthermosetting adhesive. Then the conductor pat\u00ac \ntern is cut from the copper foil bv a dinking process \nin such a manner that the aluminum foil is not \nsevered. Tin 1 unwanted portions are stripped from \nthe aluminum. This aluminum carrier is then placed \non the surface of an insulating laminate, and a \nlaminating press is used to bond the conductors \nfirmlv to the plastic sheet. The final operation is to \nstrip off the aluminum carrier and clean the ad\u00ac \nhesive off the surface of the conductors.\n\nThe dinking die used in this process is prepared \nbv photoengraving from a hard brass plate and \nthen chrome plating the cutting edges to extend \nthe life of the tool. Thus it is inexpensive and flexi\u00ac \nble to changes in circuitry. It is also adaptable to \ncontinuous processing which is necessary for any \nautomated production system.\n\nAllied with the problem of degradation of plastic- \nbase printed circuits due to heat and long-term \naging, the Laboratories is currently studying several \nprocesses for producing reliable inexpensive printed \ncircuits on ceramic-base materials. One of these \nhas been developed for producing a spiral inductor \nwith conductors ten thousandths of an inch wide, \nspaced at ten thousandths of an inch. The require\u00ac \nments made it necessary to develop a process which \ndid not relv on firing techniques and did not result \nin residual nonvolatile components.\n\nAlthough these processes have been used pri\u00ac \nmarily for printed wiring, thev are bv no means\n\nlimited to this application. There is an increasing \nnumber of applications in the design of switches, \ncommutators, coding discs and electrical com\u00ac \nponents. Complicated rotary and sliding switches \ncan be produced by those means at lower cost be\u00ac \ncause of the photoengraving technique. Also, the \ndevelopment of flush circuits, wherein the surface \nof the conductor and the insulator are in the same \nplane, has made possible a considerable smoother \nwiping surface for high-speed brushes. This re\u00ac \nduces the hazard of brush bounce which results \nin erroneous switching.\n\nAn example of the application of the printed \ncircuitry to components is an autotransformer w hich \ncan be placed inside a cathode-rav tube as illus\u00ac \ntrated in the photograph on page 117. The con\u00ac \nstruction, consisting of a series of flat printed in\u00ac \nductors on ceramic discs, is capable of withstand\u00ac \ning the high degassing temperatures and the cvclic \ntemperatures of the tube during normal operation. \nUse of this autotransformer would remove the high \npotential connections from the exterior of the tube \nand thus reduce electric shock hazards.\n\nThe use of multiple-layer printed cables to re\u00ac \nduce the size of a computer and to decrease its \nvunerability to shock and vibration is illustrated \nin Figure 3. Each module, representing a memorv\n\nor logic function, is an encapsulated printed-wiring \nboard with integral printed spring-contact fingers. \nThese packages are mounted on a printed cable- \nstick which consists of four layers of printed wiring \nand two printed connectors, one for test and the \nother for power and signal supply. Each stick is a \nplug-in unit which fits into a receptacle in the con\u00ac \nsole and can be tested as a functional unit.\n\nThe impact of printed circuits on automation in \nthe electronics industry is illustrated bv the in\u00ac \ncreasing number of mechanical assembly systems \nused to produce commercial and military electronic \nequipment. There will continue to be a large num\u00ac \nber of circuits which will not be economically \nfeasible to assemble automatically. The mech\u00ac \nanized assembling machines, however, will provide \nindustry with a substantial standby force of robot \nassemblers, wiring hands and inspectors which can \nbe called upon to meet sudden demands for a \nlarge amount of equipment. This would certainly \nbe true in the case of military emergency when \nthe defense services would require a tremendous \nvolume of tactical and strategic electronic systems.\n\nIt is quite evident now that printed circuits have \nemerged from the laboratory' development stage \nto become another in the series of new tools de\u00ac \nveloped to promote the advance of electronics.\n\nN. Osircinx, a native of Phillipsburg, New [erse\\, receiv ed his degree in Me\u00ac \nchanical Engineering from Rutgers University in 1951 after serving two and one- \nhalf years in the Army. He joined Bell Laboratories that year as a member of the \nNike Missile S vs tom group and was enrolled in the Laboratories Communications \nDevelopment Training Program. In 1954 he transferred to the newlv formed \nelectromechanical development department, where he was in charge of a group \nconcerned with the study and development of printed-circuit processes and ap\u00ac \nplications. At present he is in charge of a group studying mechanical design \nproblems associated with radars and missile-borne equipment. He is a recipient \nof the Lincoln Foundation Award and is a member of A.S.M.E. and Tan Beta Pi.\n\nS. J. Stockflkth, a native of Bergen, Norway, came to the United States late \nin 1924 after receiving his B.S. in E.E, in Germany. Upon completing a graduate \ntraining course at the Westinghouse Corp., he joined the Manufacturing Develop\u00ac \nment Department of the W estern Electric Company at Hawthorne, where he \ndeveloped multiple-coil winders and sealed-coil units, Mr. StockUeth transferred \nto the Apparatus Development Department of Bell Laboratories in 1929, where \nhe worked on the development of dials, multi-contact and crossbar switches on \nwhich he holds several patents. He was concerned with the mechanical design of \nmilitary equipment (hiring World War II, and after the war supervised a group \nresponsible for mechanical design of the Nike System radars. Since 1953 he has \nbeen in charge of a subdepartment responsible for the mechanical design and \ndevelopment of military radar antennas, missile equipment, printed circuits, and \nthe Whippanv Mechanical Laboratory.\n\nRelay contacts in service sometimes erode much \nfaster than laboratory life tests indicate they \nshould. Often such erosion is due to \u201cactivation\u2019* \n\u2014 a subtle effect of organic vapors present in \nthe operating environment. Activation has been \nstudied for some years at Bell Laboratories, and \nthe phenomenon is now quite well understood.\n\nIf you ask a telephone engineer, \u201cwhat is the \nmost important component in today\u2019s switching \nsystems?\u201d, he will probably reply, \u201crelays.\u201d The \n\u201cbusiness end\u201d of these dependable electromechani\u00ac \ncal devices is the contact end. Relay contacts, since \nthe advent of common-control switching systems, \nhave probably earned the right to be singled out \nas the hardest workers in the telephone plant. A \nmodern relay, for example, may have to operate \nmore than 1,000 times in an hour.\n\nIn some cases, each contact operation may short\u00ac \nen the life of the relay because of \u201cerosion\u201d phe\u00ac \nnomena at work on the contacts. The amount of con\u00ac \ntact erosion \u2014 a direct measure of the amount by \nwhich the life of the relav is shortened \u2014 mav varv\n\nby a factor greater than 100 with differing en\u00ac \nvironmental conditions. Relay designers and re\u00ac \nsearch groups at Bell Laboratories have been aware \nof and have studied the causes and effects of erosion\n\nphenomena for many years. The effects of these \nphenomena are obvious \u2014 degradation of telephone \nservice and increased relay maintenance. The \ncauses of erosion are more subtle, however, and \nare only now becoming thoroughly understood.\n\nThe basic cause of erosion is electric arcs which \noccur when relay contacts are opened or closed, \nand a major factor influencing these arcs is a \nphenomenon called \u201cactivation.\u201d Activation is the \ncontamination of the metal surface of a contact \nwhich gives rise to greater arcing (during either \nthe \u201cmaking\u201d or \u201cbreaking\u201d of an electrical circuit) \nthan would occur at clean surfaces. This phenom\u00ac \nenon is produced by repeated operation of a pair \nof current-carrying relay contacts in the presence \nof organic vapor. Basically, it is these organic va\u00ac \npors that cause increased arcing and increased con\u00ac \ntact erosion. The increase in arcing takes either of \ntwo forms: arcs when there would have been none\n\nwithout the vapor, and arcs that last longer in the \npresence of vapor than they would in pure air.*\n\nOrganic vapors that produce activation come from \nmany sources. Some of these are the plasticizer in \nthe cellulose acetate of the relay coil winding, the \nphenol fiber in the relay structure, and the paint \nused on the walls of telephone central offices. Lab\u00ac \noratory tests have shown that normally onlv noble- \nmetal surfaces become \u201cactive\u201d and that activation \ncan be produced only by those organic compounds \nwhich contain rings of carbon atoms with at least \none unsaturated bond between atoms in the ring. \nBenzene and benzene derivatives are examples of \ncompounds with this molecular structure.\n\nWhen a pair of palladium contacts is operated re\u00ac \npeatedly in the presence of such an organic vapor, \nand the voltage across the contacts is displayed on \nan oscilloscope, the trace would look, in the be\u00ac \nginning, as if no vapor were present. This non-arc\u00ac \ning condition, as it appears on the oscilloscope \nscreen, is shown in Figure 1(a). After continued op\u00ac \neration, however, arcing can be observed. The arc\u00ac \ning condition can be recognized by an irregular \ntrace similar to the one shown in Figure 1(b). Even \nbefore any electrical effects are observable, how\u00ac \never, incipient activation can be detected micro\u00ac \nscopically by the presence of black, sooty material \non the contacts. This sooty substance is carbon, and \nthe amount necessary for activation is extremely \nsmall \u2014 about 0.05 microgram.\n\nCarbon is formed by the decomposition of or\u00ac \nganic molecules adsorbed on the contact surfaces. \nArcs cause this decomposition, and the resulting \ncarbon enhances the severity of the arcing, so that \na vicious cycle of activation, arcs, carbon, increased \narcing, and so on, can easily get started. The amount\n\nFig. 1 Oscilloscope trace of (a) non-arcing con\u00ac \ntacts. and ( b) arcing at contacts after repeated op\u00ac \neration in vapor. In both traces the circuit voltage \nteas 35, indicated by the initial dots. The initial \nhorizontal line in (b) represents the potential \ndifference across the contacts during an arc lasting \nfor 2 microseconds. This teas folloued by an open- \ncircuit potential of 10 volts, lasting until final \nclosure (zero potential) at about 76 microseconds.\n\nFig . 2 \u2014 Plot of the rate of formation of carbon \non relay (ontacts in the presence of organic vapor.\n\nof carbon formed by one arc corresponds initially \nto about a single layer of organic molecules ad\u00ac \nsorbed on the portion of the contact covered by \nthe arc. The rate of carbon formation later increases \n(Figure 2) to an amount corresponding to several \nmonolayers of molecules, presumablv because the \ncarbon already present increases the active surface \narea of the relay contact.\n\nIn pure vapor, experiments have proved that the \namount of carbon formed per contact operation is \nindependent of the vapor pressure. When the va\u00ac \npor is simply part of the surrounding air, however, \nthe situation is complicated bv the presence of suffi\u00ac \ncient oxygen to burn some of the carbon already \nformed, and by the impedance offered bv air to \ndiffusion of organic molecules to the electrode sur\u00ac \nfaces. The rate of contact operation is a factor, too, \nin carbon formation. If the contacts are operated \nrapidly, less than a monolaver of activating mole\u00ac \ncules may be able to accumulate on the contact \nsurfaces in the brief time the contacts are separated. \nThese two factors \u2014 brief accumulation time and \ncarbon burning \u2014 establish a minimum pressure of \norganic vapor below which the contacts of a relay \ndo not become active.\n\nThe presence of carbon on contact surfaces \nchanges both the character and intensity of the \narcs. Consequentlv, the character and magnitude of \nthe resulting erosion are also altered. We shall \nconsider first the character of erosion with and with\u00ac \nout carbon, and then the effect of the presence of \ncarbon on the amount of erosion.\n\nTwo distinct types of arcs occur between relay \ncontacts. The type of arcing which occurs in any \nparticular case is the dominant factor in determin\u00ac \ning the character of the contact erosion. 0 \\n \u201canode \narc\u201d vaporizes metal predominantlv from the anode,\n\nFig. 3 Photomicrograph I approximately 715 di - \nameters i of marks made on cathode ( left) and \nanode surfaces by a single arc of the anode type.\n\nHere, the metal is vaporized hv the heating due to \nelectron bombardment, and a spatter of anode \nmetal is left on the cathode (Figure 3). The second \ntype, a \u201ccathode arc\u201d, vaporizes metal from mvriads \nof individual spots on the cathode. In this case, \nvaporization of the metal is caused bv resistance \nheating, since almost the entire electron current \nof the arc flows through one of these tiny spots, \nvaporizing the metal with explosive suddenness. \nCathodes arcs damage the anode verv little, in \nmany cases producing much less damage than is \nshown in Figure 4, except of course for the dis\u00ac \npersed metal spattered over from the cathode.\n\nThe type of arc \u2014 anode or cathode \u2014 is deter\u00ac \nmined bv the electrode separation at which the \narcs strike. At the beginning of an arc, a current \nmade up of field-emission electrons is drawn from \na single point on the cathode surface. If the elec\u00ac \ntrode separation is small, a spot on the anode is \nraised to the melting point bv electron bombard\u00ac \nment before the current from the cathode becomes \nlarge enough to melt the emitting point. The re\u00ac \nsulting arc is called an anode arc. If, on the other \nhand, the separation is sufficientlv large, the emit\u00ac \nting point is melted first because the field-emission \ncurrent spreads out considerablv before it reaches \nthe anode. In this case, the resulting arc is of the \ncathode type. The critical electrode separation that \ndetermines which sort of arc will occur is different \nfor different metals, increasing with the increasing \nelectrical conductivitv of the metal. For palladium, \nthe critical distance is about 0.5 micron, and for \nsilver, three to four microns.\n\nBoth anode and cathode arcs transfer metal from \none relav contact to the other when the contact \nsurfaces are clean. An anode arc transfers metal \nfrom anode to cathode, and a cathode arc trans\u00ac \nfers metal from cathode 1 to anode. After many suc\u00ac \ncess ive arcs, a mound is built up on one electrode\n\nand a pit is eroded into the other. This deposit- \nerosion situation is entirely changed, however, by \nthe presence of surface carbon. In the case of palla\u00ac \ndium contacts, surfaces that have carbon on them, \nso-called \u201cactive surfaces,\u201d transfer little or no \nmetal, because the metal vaporized from one elec\u00ac \ntrode is prevented from sticking to the other by \nloose carbon on its surface. The eroded metal can \nbe recovered in the form of a loose, sooty powder \nmixed with carbon.\n\nCarbon on electrode surfaces gives rise to another \ndifference also. When carbon is present, each arc \noccurs at a new place \u2014 where carbon has been \nformed by preceding arcs, but not at the center \nof the immediately preceding arc where the arc \nhas burned the carbon awav. This continual re\u00ac \nlocation of the arcs results in uniform erosion of \nthe electrode surfaces, with no formation of pits \nor mounds. At silver contacts, less carbon is often \nformed in the presence of organic vapors, and a \ngreat deal of the metal eroded from one contact \nmav stick to the other, even in the active condition.\n\nThe change from pit-and-mound erosion to ero\u00ac \nsion which is uniform over the contact surface is \nnot the only effect of the presence of carbon. A \nstill more important effect of activation is the pro\u00ac \nduction of many more cathode arcs than anode \narcs. Cathode arcs, as mentioned earlier, occur \nwhen an arc strikes at a contact separation greater \nthan that giving rise to anode arcs. The presence \nof surface carbon increases the tendency toward \narcing at wide separations, and this in turn results \nin a tendency for the dominant loss of metal to be \nfrom the cathode rather than from the anode.\n\nCarbon on the electrode surfaces \u2014 resulting from \nactivation \u2014 increases the contact separation at \nwhich arcing starts, because on an active surface\n\nFig. 4 -&A ppearance I 900 diameters ) of the cathode \n(left) and of the anode after a single cathode arc.\n\nan arc strikes when the electric field generates \nelectrostatic forces high enough to move the carbon \nparticles. At the same time, the motion of these \nparticles results in partial closing of the gap. In \nmany cases, the electric field at which an arc strikes \nat carbonized surfaces is of the order of 0.5 x l() w \nvolts per centimeter, which at fifty volts is a separa\u00ac \ntion of one micron. This is considerably less than \nthe critical distance of three to four microns which \ndetermines for silver electrodes whether an arc \nwill be of the anode or cathode tvpe.\n\nThus, activation does not, in most cases, alter the \ntype of arc at silver surfaces when the striking po\u00ac \ntential is fifty volts. But it mav alter the type of \narc at higher voltages as, for example, at separating \nelectrodes in an inductive circuit. For palladium \ncontacts, on the other hand, the critical distance \nof 0.5 micron is so small that activation causes all \narcs to be of the cathode type. Between clean palla\u00ac \ndium surfaces, however, most arcs at fiftv volts are \nanode arcs. At silver surfaces then, the loss of metal \nis predominantly from the anode, whether the sur\u00ac \nfaces are clean or carbonized. At palladium sur\u00ac \nfaces (at 50 volts), loss is chieflv from the anode \nwhen the surfaces are clean, but from the cathode \nwhen they are active.\n\nThis situation is of course modified at high vol\u00ac \ntages. What will happen at am given voltage is \npredictable, however, if both the critical distance \nand the electrode separation at which an arc strikes \nare known. For example, about 350 volts is the \nminimum difference in potential at which the gases \nin the air between the two electrodes \u201cbreak down\u201d \nat atmospheric pressure. The distance at which vol\u00ac \ntages just above this value cause breakdown is about \nfifteen microns. This distance is so great that for \nvoltages above 350, all arcs are cathode arcs for \nsilver as well as for palladium, whether the sur\u00ac \nfaces are clean or active. These observations re\u00ac\n\ngarding the character of erosion resulting from arcs \nare presented in the table below.\n\nTo this point, we have been concerned with the \neffect of organic vapors on the character of erosion \nat contact surfaces. As brought out earlier, the \nmagnitude of the erosion is affected as well as the \ncharacter. In most cases, activation increases ero\u00ac \nsion by a large factor. This is due to two factors: \nthe greater separation at which arcs can strike be\u00ac \ntween active surfaces, and the lowering of the min\u00ac \nimum current which can sustain an arc between \nsuch surfaces. The first of these effects \u2014 greater \ncontact separation for active arcs \u2014 has been dis\u00ac \ncussed above. This effect is also important in pro\u00ac \nlonging the length of time of arcing as contacts are \npulled apart.\n\nThe second effect \u2014 the lowering of the \u201cminimum \narc current\u201d \u2014 is most readily understood bv consid\u00ac \nering an arc of the cathode tvpe. For a cathode arc, \nthe minimum arc current is the lowest current which \nwill promptly explode a point on the cathode \nthrough which the current flows. V carbon particle \non the cathode has relatively high electrical resis\u00ac \ntivity as well as poor thermal contact to the sup\u00ac \nporting metal. The lowest current which can ex\u00ac \nplode it, therefore, is much less than the minimum \ncurrent required to explode a spot on clean cathode \nmetal. Specifically, experiments have shown that \nthe minimum arc current for a clean metal surface \nmay be as high as one ampere but for a carbonized \nsurface it mav be as low as 0.03 ampere.\n\nIt is possibly this feature of activation \u2014 the low\u00ac \nering of the minimum arc current \u2014 that makes con\u00ac \ntact activation most objectionable. At contacts with \na minimum arc current that is low, an arc can often \nlast for a much longer time than it would if the \nminimum arc current were higher. This longer \narcing time results in greath increased erosion.\n\n\u00b0 At high vapor pressure. At low vapor pressures there ruay he an anode pit, \nwith much of the eroded metal sticking to the cathode in the form of a mound.\n\npractice in telephone circuits to install a protective \nnetwork consisting of a capacitor in series with a \nresistor across the contacts, or across the inductive \nload. Such a circuit can be effective onlv when the \nminimum arc current is higher than the circuit cur\u00ac \nrent. Thus, it turns out that active contacts in ' 2 -am- \npere circuits, fairly common in the telephone plant, \ncannot be protected against arcing. Furthermore, \nactivation is damaging even in low-current circuits. \nHere, intermittent arcing can occur due to the dis\u00ac \ncharge of local capacitance. When contacts are ac\u00ac \ntive, the repeated discharge of local capacitance\n\nmay produce erosion that is almost as severe as the \nerosion from a sustained arc in a circuit that is nor\u00ac \nmally considered as carrying a high current.\n\nAlthough both the causes and effects of activation \nare now well understood, the phenomenon never\u00ac \ntheless sometimes presents problems in relay design. \nActivation can be reduced both by using stable ma\u00ac \nterials which are not sources of organic vapors, and \nby adequate ventilation. The cost and other con\u00ac \nsequences of any possible change are often an im\u00ac \nportant consideration, and what to do in any par\u00ac \nticular case is not alwavs at once clear.\n\nL. II. Gehmer, a native of Chicago, received the A.B. degree from Cornell \nUniversity in 191V, and the M.A. and Ph.D. degrees from Columbia University \nin 1922 and 1927, respectively. In 1917, he joined the Engineering Department \nof the Western Electric Company and has been at Western Electric and Bell \nLaboratories since that time, except for a period from 1917 to 1919 when he \nserved in the U. S. Air Force. Mr. Germer\u2019s work has been in various fields of \nphysical research including thermionics, electron diffraction, surface phenomena \nand the phvsies of electrical contacts. He is the author of over seventv scientific \npapers on these subjects.\n\nJ. L. Smith, a native of Morristown, N. }., joined the Laboratories in 1941 as \na messenger. After serving three years in the U. S. Navy, he returned to the Phvsi- \ncal Research Department where he worked on problems concerned with relav \ncontact erosion. From 1946 to 1956 Mr. Smith attended the evening division of \nthe Newark College of Engineering, receiving the B.S. in E.E. degree. In 1956 \nhe transferred to the Switching Development Department where he is currently \nengaged in the design of switching networks using solid-state devices. Mr. Smith \nis a member of Tan Beta Pi.\n\nA major difficulty in the development of liigli- \nlrequency transistors has heen the formation of \nreliable electrical contacts between extremely \nfine wires and minute areas of semiconductor \ncrystals. At Bell Laboratories, a promising solu\u00ac \ntion to this problem has heen found \u2014 a tech\u00ac \nnique known as tliemio-compression Ixmding.\n\nFor the past several years there has been a strong \ntrend toward miniaturized electronic devices. In \npart, this trend is attributable to the great econo\u00ac \nmies possible with small, compact equipment. The \nelectronic telephone switching systems of the future, \nfor example, are expected to be so much less bulky \nthan present equipment that they should result in \nconsiderable savings to the Bell System in floor \nspace and sizes of central offices.\n\nThere is, however, another important reason for \nminiaturization of electronic circuits, stimulated in \nlarge measure by the invention at Bell Laboratories \nof the transistor. Because of its inherently small \nsize and lower power requirements, the transistor \npermits other components in a circuit to be reduced \nproportionately in size. Further, there has been \nconsiderable emphasis on high-frequency opera\u00ac \ntion for greater bandwidth of communication sys\u00ac \ntems. Since the elements of electronic devices in \ngeneral become smaller at the higher frequencies, \nthis means that the already small transistor struc\u00ac \nture has progressively become even smaller.\n\nWhen structures are to be fabricated to virtual\u00ac \nly microscopic dimensions, many complications ap\u00ac \npear, not the least of which is the sheer mechanical\n\ndifficulty of working with tiny parts. In addition, \nsemiconductors are frequently sensitive to mechan\u00ac \nical disturbances and to minute traces of unwanted \nchemical contamination \u2014 a consideration that im\u00ac \nposes strict limitations on methods of fabrication.\n\nIn many such cases, an outstanding mechanical \ndifficulty is that of securing a bond between very \nthin electrical conductors and the semiconductor. \nThe bond must be accurately positioned in a small \narea; it must have the least possible effect on the \nelectrical characteristics of the semiconductor; and\n\nHeat and pressure produce the bond with very little \ncontamination of the semiconductor material.\n\nit must be mechanically strong and reliable. This \ndifficulty is overcome to a large degree by the \napplication of a new technique of joining solids \ncalled thermo-compression bonding. The technique \nis the result of fundamental research on the ad\u00ac \nhesion of solids* performed jointly bv O. L. Ander\u00ac \nson of the Research Department and the author.\n\nCurrent practice of attaching leads to semicon\u00ac \nductors involves alloying or soldering. These are\n\nFig. 2 \u2014 Tensile-strength curves for brittle fracture \n(horizontal lines I and plastic yield (steep lines ) \nfor germanium and silicon. Bonding region for ex\u00ac \nperiments (shaded area l is well inside these limits .\n\nobjectionable because both involve a liquid phase, \nwhich gives rise to problems of alloy depth control, \nwetting and chemical contamination. The bond\u00ac \ning process described here does not involve liquid \nphase formation and hence avoids these problems. \nAs shown in functional schematic form in Figure 1, \nit merely involves the application of heat and pres\u00ac \nsure to the members being bonded. When these \nare applied for a few seconds, adhesion occurs. The \ntwo main advantages of such a bond are its sim- \nplicitv of formation and its freedom from chemical \ncontamination.\n\nA 5000 to 10,000 psi pressure, 200 to 300\u00b0C tem\u00ac \nperature range, and a 5-second time interval are \nappropriate conditions for formation of a bond \nbetween a gold wire and a block of germanium. \nThe upper range of temperature and pressure are \nused where the wire is hardened by cold working \nor bv the addition of \u201cdoping\u201d (impurity) metals \nlike aluminum or arsenic from the third and fifth \ncolumns of the periodic table. Where metal-to- \nmetal bonds are to be made \u2014 for example in bond\u00ac \ning leads to a transistor electrode \u2014 the time for \nbonding can be reduced somewhat. Some cleaning\n\nof the surface mav be necessary prior to the bond\u00ac \ning operation. Gold wire is sometimes degreased \nand annealed, but alloyed transistor emitter and \nbase electrodes, discussed later, are satisfactorily \nclean as thev come from the alloying process.\n\nThe most important issue involved in application \nof thermo-compression bonds to semiconductors is \nthe question of possible mechanical damage to the \nsemiconductor. Figure 2 relates temperature and \npressure of bonding to brittle fracture and plastic \nstress of silicon and germanium. In this illustra\u00ac \ntion, the shaded area represents the region of suc\u00ac \ncessful bonding; it encompasses all the pressure and \ntemperature conditions used to date to bond gold, \nsilver, aluminum and various allovs of these ma\u00ac \nterials to silicon and germanium.\n\nBy noting the distance of the bonding region \nfrom the curve for germanium, it is apparent that \nsufficient margins exist for both temperature and \nbreaking strength. The curve for the onset of \nplastic flow in the crystal is at least 10()\u00b0C from \nthe bonding region, and along the vertical axis, \nbreaking stress is seen to be at least double the \nhighest stress used for bonding. It is also apparent \nin Figure 2 that silicon is even better in these re\u00ac \nspects. Actually, since no compression data for sili\u00ac \ncon and germanium are available, tensile-strength \nvalues are plotted in Figure 2, which means that the \nsafety factors are still higher. Several thermo-com\u00ac \npression bond areas have been studied, and no in\u00ac \ncrease in density of dislocations associated with\n\nFig . 3 \u2014 Extremely fine wire bonded by thermo\u00ac \ncompression technique to electrodes of high-fre\u00ac \nquency transistor using germanium semiconductor.\n\nFig. 4 Gold lends thermo-compression bonded \nto silicon in a power transistor. Bond passes 300 \nm I Hi am peres of continuous current.\n\npress ion bonding was the attachment of lead wires \nto evaporated and alloyed gold and aluminum \nelectrodes of d iff used-base transistors. Figure 3 \nshows an example of bonded leads on a high-fre\u00ac \nquency germanium transistor having verv closeh \nspaced elements. The dark stripe is the aluminum \nalloy emitter; it is about 0.006 inch by 0.001 inch. \nThe bright stripe is the alloyed gold base electrode. \nThe 0.0004-inch diameter gold lead wire is smaller \nthan can be made by direct drawing. It was formed \nby first drawing down a silver-clad gold wire and \nthen dissolving away the silver. The resistance of \nthe bonded contact, after correction is made for \nthe lead resistance, is about 0.01 ohm. This is as \nexpected for a metallic contact without intervening \nresistive skin. When the bonded lead is forcefullv \npulled away from the electrode, either the lead \nwire breaks or a bit of germanium is extracted \nwith the wire, showing that the bond is stronger \nthan the weaker of the lead or germanium.\n\nFigure 4 shows the bonded gold leads of a sili\u00ac \ncon power transistor. The wire is 0.002 inch in \ndiameter, and the bond passes up to 300 milliam- \nperes continuously without deterioration or fail\u00ac \nure. All three electrodes shown here are of a gold- \nsilicon alloy. This photograph illustrates the fact \nthat the bond is also satisfactory for large power re\u00ac \nquirements. In this case, two base stripes are used \nto reduce base resistance. In certain variations of \nthe development of this transistor, more than one \nlead could be attached to the electrode. Such fea\u00ac \ntures and many others make thermo-compression \nbonding attractive from the point of view of versa\u00ac \ntility of design.\n\nThe electrical characteristic of a metal-to-ger- \nmanium thermo-compression bond is of considerable \ninterest, both for potential application in device\n\ntechnology and also for the investigation of a metal- \nto-semiconductor interface. When carefully formed, \nbonds of this sort give either rectifying or nonrecti\u00ac \nfying (ohmic) contacts in accordance with surface \ndoping produced bv the bonding wire.\n\nFigure 5 illustrates the conditions under which \nrectification is produced in a thermo-compression \nbonded contact of an aluminum wire to a block \nof n-type germanium. For the germanium under \nthis type of bond, curve A shows theoretical values \nof spreading resistance\u2014the resistance introduced \nby the crowding of flow lines of electricity in the \ngermanium as they come to the confining area un\u00ac \nder the bond. Curve A is to be compared with \ncurve 13, yvhich represents actual data obtained \nfrom a contact bonded in air. As can be seen, this \ncurve is intermediate to curve I). Data for curve \nD were obtained after bond 13 was heated to a \ntemperature slightlv above the aluminum-germa\u00ac \nnium eutectic \u2014 the hnvest temperature that giv r es \na liquid phase when Ge and A1 are heated (425\u00b0C). \nCurve C shows the effect of using a hvdrogen at\u00ac \nmosphere for bonding instead of air. In this case, \nwe have a bond that forms a rectifier that w ill \nsaturate. The difference between curves 13 and C \nis accounted for by the contamination introduced \nby the air ambient. Trace contamination on either \nthe wire or the germanium surface probably ac\u00ac \ncounts for the difference between curves C and D \nseen in Figure 5.\n\nOne distinct possibility for application of this \ntype of bond is the process illustrated by curve D. \nAt the expense of introducing a liquid-phase eutec\u00ac \ntic into the process, yve can form both an electrode \nand a lead in one operation. A measure of control\n\nFig, 5 Theoretical \u201cspreading resistance\" (A) \ncompared to actual voltage-vs-current characteris\u00ac \ntic curves for bonds made by the various methods.\n\nof alloy depth is obtainable in this case because \nthe liquid-phase eutectic forms at one or two de\u00ac \ngrees above the eutectic temperature, and wetting \noccurs over the bonded area. Another feature of \nalloy depth control is the fact that the lead wire \ncan be of eutectic composition, so that it does not \nrequire much germanium to go into solution to \nform the liquid phase.\n\nThermo-compression bonding is thus seen to be \na significant advance in semiconductor technology.\n\nBy the application of heat and pressure, the bond \ncan be formed either with or without formation of \nthe liquid phase, as desired. The procedure may \nbe used to improve wetting and control of depth \nin the alloying process. When a bond is made to a \nsemiconductor, the contact may be either ohmic \nor rectifying, depending upon the doping of the \nbonding wire and of the semiconductor. The tech\u00ac \nnique is presently being applied experimentally in \ntransistor fabrication at Bell Laboratories.\n\nH. Christensen, a native of Manti, Utah, received the B.S. degree in physics \nfrom Brigham Young University in 1930 and the M.A. degree, also in physics, \nfrom Columbia University in 1935. He has been a member of the technical staff \nof Bell Laboratories since 1930, when his first assignment was in the Submarine \nCable Research Department. In 1937 he began semiconductor material research \nwork, which he later used in the development of the thermistor bolometer. From \nthe beginning of World War II until 1950, Mr. Christensen worked on NDRC \nand U. S. Navy contracts, and subsequently was engaged in investigation of \ngermanium material processing and surface phenomena. Since 1954 he has been \nengaged in advanced development of diffused-base transistors. He is a member \nof the American Phvsical Society and of the A.l.E.E.\n\nThe Western Electric Comp an v disclosed on \nFebruary 19 the receipt of a prime letter contract \nfrom the U. S. Air Force for an important com\u00ac \nmunications project in connection with a Ballistic \nMissile Early Warning System, and said there \nwould be \u201csubstantial participation\" by Bell Lab\u00ac \noratories in the project. The text is as follows:\n\n\u201cA prime letter contract has been received from \nthe U. S. Air Fence authorizing the Western Elec\u00ac \ntric Company to proceed with the initial phases \nof a communications project of major dimension \nin connection with a Ballistic Missile Early Warning \nSystem (BMEWS).\n\n\u201cWestern Electric's responsibilities involve im\u00ac \nproving and augmenting communications in north\u00ac \nern areas which will serve the new warning system \nand which will be coordinated with other military \nrequirements. Work on the new project is already \nunderway.\n\n\u201cThe Radio Corporation of America is prime con\u00ac \ntractor for the remote detection installations of \nBMEWS as well as certain work at terminals in the \nUnited States. Although RCA and Western Elec\u00ac \ntric have separate prime contracts, close coopera\u00ac \ntion will be required between the two companies on \nthose facilities concerned with the BMEWS project.\n\n\u201cAt Western Electric, the work will be handled \nby the Defense Projects Division where prior ex\u00ac \nperience with the DEW and White Alice systems \nis expected to be of great value. There will be \nsubstantial participation by the Laboratories in \ntechnical areas, as well as the assistance of agencies \nresponsible for interconnecting commercial circuits.\n\n\u201cThe contract covers responsibility for design, \ninstallation and testing of new or augmented routes \noutside of commercial areas and for overall com\u00ac \nmunication s vs tern tests. Responsibilitv for con\u00ac \nstruction of buildings and associated facilities has \nnot been announced. The project will require \nstandard Bell Svstem equipment as well as some \nof new and special design. Western Electric will \nmanufacture the normallv supplied Bell System \nitems as well as certain of the new designs the \nCompanv\u2019s plants may be especiallv qualified to \nfurnish. It is expected that a substantial portion will \nbe subcontracted to qualified manufacturers.\n\n\u201cThe total personnel required is not known at \nthis time but it is expected that the present \nstaff on the DEW and White Alice projects will be \nincreased bv personnel from other Bell System \ncompanies and Western Electric as the need for \nparticular skills develops.\"\n\nBell System earnings were $13 a share of Ameri\u00ac \ncan Telephone and Telegraph Company stock in \n1957, President Frederick R. Kappel said in the \ncompany\u2019s annual report, released February 18.\n\nThe earnings were based on the average of 63,- \n811,000 shares outstanding\u20146,388,000 more than in \n1956, when net income per average share equalled \n$13.16. Bell System earnings on total capital were \n6.7 per cent, compared with 6.8 per cent in 1956.\n\nThe Bell System spent more than $2.5 billion \nin 1957 to enlarge and modernize telephone plant. \nMost of the new capital obtained last year was \nthrough the sale of bonds. These totaled just over \na billion dollars. The average interest cost, Mr. \nKappel pointed out, was nearly 4% per cent \u2014 the \nhighest in many years. Part of the money to ex\u00ac \npand and improve telephone service in 1958 will \ncome from an issue of $718 million of debentures \nconvertible into stock. Share owners mav buy one \n$100 debenture for each nine shares of stock held.\n\nThe A.T.&T. report, mailed to more than 1,600,- \n000 share owners, cited many gains in service. It \nstated, for example, that in 1957 the Bell System \nadded some 2,815,000 telephones, bringing the total \noperated by the System to 52,250,000. Long dis\u00ac \ntance conversations rose more than 7 per cent \nover 1956. Long distance service \\\\ as the fastest \never. The average speed of connection was only \n72 seconds. The Bell System changed a million \nmore telephones to dial service. Today 92 per cent \nof Bell telephones are dial operated. Also, some \nfive million customers todav can use TDD\u201d \u2014 \ndirect distance dialing \u2014 to as many as 30 million \nother telephones from coast to coast. Another 10 \nmillion customers can dial nearby cities and towns.\n\nThe report also cited gains in undersea cable \nservice. Since the opening of the new telephone \ncable between the Lb S. mainland and Hawaii last \nOctober, calls to and from the island territory have \nincreased about 30 per cent. The first telephone\n\ncable across the Atlantic, opened in 1956, is now \nso heavily used that work has already begun on a \nsecond cable which is expected to be ready in 1959.\n\nUnderhing all the progress, the A.T.&T. report \nnoted, are research at Bell Telephone Laboratories \nand manufacture of equipment by Western Elec\u00ac \ntric. In addition, both organizations continue to \nwork on guided missile systems and other vital \ndefense projects.\n\nBell Telephone Laboratories, Mr. Kappel said, \u201cis \none of the largest research organizations in indus\u00ac \ntry. One reason is simph that there is so much \nwork to be done. Also, big size in itself is a distinct \nasset. To make significant scientific discovery we\n\n1C, L, Hawkins of the Chemical Research Depart\u00ac \nment conduct in ft oxidation tests on polyethylene \nto determine effectiveness of antioxidants.\n\nC. IK Thurmond and Mrs. E. A. Wood discussing \nstructure of silicon as part of basic research into \nproperties of semiconductors .\n\nneed the best scientific minds. Those minds usually \nwant the stimulus of association with others of like \ncaliber, in a well-rounded scientific community that \nhas pre-eminent standing. This we have in Bell \nLaboratories.\n\n\u201cThere for many years the attack on communi\u00ac \ncation problems has begun with basic research \u2014 \nthe search for new understanding and knowledge \nof nature. This has made possible long distance \nservice; the multiplication of talking paths over the \nsame physical conductors; the development of \nradio relay systems; the creation of modern dial \nswitching systems; and many other developments. \nIn fact, the entire telephone network of today is \nbuilt on the knowledge which basic research has \nsupplied.\n\n\u201cThe transistor, invented at Bell Laboratories \nabout ten years ago, is now finding many communi\u00ac \ncation uses: to route long distance calls automati\u00ac \ncally; to provide rural telephone circuits; to send \ndata at high speeds; to produce the \u2018ringing signal\u2019 \nin telephones; to amplify sound in other telephones \nfor people who have difficulty in hearing. . . . These \nare examples. Also, development work continues \non the new electronic switching system in which \ntransistors play a vital part. This is a major under\u00ac \ntaking. We are confident it will make possible many \nnew conveniences and improvements in service.\n\n\u201cMeanwhile basic research goes on. Bell System \nscientists today are studying the effects of nuclear \nradiation on crystals and plastics. They have found \nnew ways to join materials together\u2014to bond them \nso that thev may be used in ways previously im\u00ac \npossible. They are studying how to send huge \nquantities of information over long distances \nthrough hollow \u2018waveguides/ Thev have demon\u00ac \nstrated an entirely new microwave amplifier \u2014 a \nsmall crystal in which, at a temperature near abso\u00ac \nlute zero, tbe spin of electrons in the atoms provides \nthe amplifying action. Such a device promises to \ndetect far weaker signals than could ever be de\u00ac \ntected before \u2014 even reflections from small objects \nin outer space/'\n\nAmong the new developments now under way \nat Bell Laboratories, Mr. Kappel said, is TASI \u2014 \nshort for \u201cTime Assignment Speech Interpolation.\u201d \nTASI is expected to increase greatly tbe capacity \nof telephone circuits such as those in the oceanic \ncables.\n\n\u201cToday twin ocean cables (one for each direc\u00ac \ntion of talking) can carry 36 conversations at once,\u201d \nMr. Kappel said. \u201cWhat we look forward to is that \nTASI may as much as double this capacity. This \nit will do by taking advantage of tbe times when \npeople are listening or pausing. The voice chan\u00ac \nnels they leave temporarily unused will be auto\u00ac \nmatically assigned to other talkers; and when a \nlistener starts to talk he will instantly have the \nuse of a channel which another person has left \nidle.\"\n\nThe A.T.&T. president also commented on the \ngrowing use of microwave radio: \u201cIn fact, use of \nradio is sure to be more and more important in en\u00ac \nabling us to provide tbe best and most economical \nservice in years to come. We have therefore asked \nthe Federal Communications Commission to in\u00ac \ncrease the number of microwave frequencies allo\u00ac \ncated to communication companies.\n\n\u201cThe number of frequencies is limited, and a \npolicy assuring their orderly assignment to com\u00ac \npanies serving the general public will allow far \nmore efficient use of the radio spectrum than could \nresult from indiscriminate licensing for private \noperation. Moreover, such a policy is essential to \npermit expansion of the communication network \nin ways that will best serve the nation\u2019s defense.\u201d\n\nBecause dependable switching apparatus is of basic importance to quality \nof telephone serv ice. Bell Laboratories engineers conduct extensive reliability \nstudies. Their primary sources of data are actual performance records \u2014 \ntrouble tickets from central offices and other types of special information.\n\nModern electromagnetic switching offices con\u00ac \ntain a complex array of apparatus wired into cir\u00ac \ncuits to form a working unit. A single call estab\u00ac \nlished by this equipment may require the operation \nof a thousand or more relays. Many thousands of \ncalls are handled daily by an office without inci\u00ac \ndent, but troubles do inevitably occur.\n\nTroubles in central-office switching equipment \nare, in general, apparatus or wiring troubles rather \nthan circuit difficulties. They include items such \nas adjustment and contact troubles, defective parts, \nand wiring defects. Whatever the cause, central- \noffice' troubles are of concern to Laboratories engi\u00ac \nneers because, fundamentally, good design must \ntake into consideration the effects of service use \nand environment, and the reliability of the com\u00ac \nmercially-manufactured and installed product.\n\nKnowledge of apparatus performance is obtained \nfrom two sources: laboratory tests under simu\u00ac \nlated field conditions and actual field data. Of \nthese two, the laboratory tests are usually conducted \nfor specific reasons. They are valuable both because \nthey permit early evaluation of design and because \nthey can be so arranged that troubles occurring \nduring laboratory testing can be readily located. \nFor example, the new wire-spring relax, during\n\ndevelopment and early production, was subjected \nto manv laboratory tests of adjustment stability, \ncontact performance, and other characteristics. \nThese tests evaluated the effect on performance of \nnew features such as lowered contact pressure, \nsmaller contacts, gold-overlav contacts, and the \nindividual relay covers.\n\nAll the varied conditions of circuit usage, traffic \nload, and environment met in the field, however, \ncannot be duplicated in the laboratory. To provide \nperformance data under actual operating condi\u00ac \ntions, the Laboratories conducts frequent field \nstudies using detailed records of the troubles ex\u00ac \nperienced in central offices. These studies may be \nlimited to specific objectives or types of equipment, \nor thev may be \u201csystem studies\u201d in which complete \ntrouble records are obtained from a group of op\u00ac \nerating offices representing a particular system. An \nadvantage of systems studies is that they are not \nconfined to anv one portion of the switching office. \nThus thev may direct attention to unforeseen con\u00ac \nditions in any element of the switching equipment. \nSystems studies also provide a larger sample than \nmost laboratory tests and special studies.\n\nSystems studies are based primarily on definitely \nlocated troubles whose cause has been determined.\n\nFig. 1 \u2014 Data from the back of standard trouble \nticket are the major source of information for \nthe stitdies of telephone switching systems.\n\nThese are known as \u201cfound troubles.\u201d In common- \ncontrol switching systems, found troubles are es\u00ac \ntimated to be less than 10 per cent of total troubles \noccurring in an average central office. A recent \nstudy of No. 1 crossbar data indicated that, for \nevery trouble found, about twelve troubles are \n\u201cnot found.\u201d (A not-found trouble is one that can\u00ac \nnot be specifically identified with a cause, either \nbecause it cleared during tracing attempts, or be\u00ac \ncause tracing was unsuccessful or not attempted.) \nOf the twelve not-found troubles, an attempt at \ntrouble location is made, on the average, in only \nabout three cases.\n\nThis low percentage of troubles traced and found \ncan be better understood by considering the nature\n\not switching equipment and the trouble-location \nprocedures. Manv troubles are not service-affecting, \nsince the \u201csecond-trial\u201d feature of crossbar systems \nusualh' results in a valid connection. Both first- and \nsecond-trial troubles do, however, automatically \nleave a record of the progress of the call up to the \ntime the trouble occurred. Troubles are traced either \nfrom these records or from other indications such \nas alarms, stuck senders, routine test failures, or \nreports from customers or traffic employees. Trou\u00ac \nble indications include many cases of repeat or \nmultiple registrations from a single trouble. Also, \nmany of these indications are caused by conditions \nin the outside plant or in connecting offices.\n\nEconomically, it is undesirable to use maintenance \ntime for tracing all indications of trouble, because \nof the transient or intermittent nature of many of \nthem, and because the indications often cannot be \nsuccessfully analyzed. Troubles not traced or not \nfound probably include items such as transient \ndirty contacts, partial loss or adjustment, or inter\u00ac \nmittent opens in windings or at wiring connections. \nFound troubles, however, can be assumed to in\u00ac \nclude the more persistent troubles and those of an \nurgent nature from the standpoint of service. Ac\u00ac \ntually, these are the only source of the specific in\u00ac \nformation required.\n\nThere are several considerations in planning sys\u00ac \ntems performance studies. First, results represen\u00ac \ntative of the system are sought. To this end, all \noffices that qualify for study purposes are listed \nin groups, based mainly on geographical location \nand equal numbers of offices are selected at random \nfrom each group for participation in the study. The \ntotal number of offices selected is sufficient for a\n\nstatistically adequate sample. Offices chosen by \nthis method, known as \u201cstratification,\u201d represent a \nvariety of Operating Companies and areas, and sat\u00ac \nisfactorily provide a cross section of the svstem with \nrespect to environment, local maintenance prac\u00ac \ntices, traffic loads, and similar factors.\n\nA second consideration in planning the studies \nis the importance of obtaining, in addition to com\u00ac \nplete trouble records, certain supplementary in\u00ac \nformation that will permit proper interpretation \nand presentation of the data. Requested from each \noffice is a count of each major type of equipment \nand each type of relay. This provides a meaning\u00ac \nful basis for prorating trouble data. In addition, \ntraffic data are obtained to indicate the call load \nhandled and to prorate trouble in terms of usage. \nOther supplementary information concerns the en\u00ac \nvironment in which the switching equipment op\u00ac \nerates. It includes principally a record of any spe\u00ac \ncial activities, such as the Western Electric Com\u00ac \npany installation work occurring during the study, \nand information such as the type of ventilating \nequipment used in the switchroom and whether or \nnot the humiditv is controlled.\n\nA third consideration is the need to obtain accu\u00ac \nrate and complete data. Basically this depends on \ncooperation of the field personnel involved and on \ntheir understanding of the aims and requirements of \nthe study. Careful thought is given, therefore, to \nthe form in which the data is requested as well as \nother details of initiating the study.\n\nFound-trouble data are obtained from trouble \ntickets which are the original records made by the \nswitchmen. Because trouble tickets are made out \nfor local use only 7 , they may be incomplete or even \nmisleading from the analyist\u2019s viewpoint because \nof varying use of terms b\\ different personnel or \nareas. To insure maximum accuracy and complete\u00ac \nness of trouble reports and of other data, the Labo\u00ac \nratories prepares a folder with special instruc\u00ac \ntions and definitions for the use of the personnel \nreporting the data. The definitions apply primarily \nto the back of the standard trouble ticket (Figure \n1), which was designed mainly for study purposes \nand provides more information than is generally \nrequired for central office administration.\n\nDuring the planning stage of the study, Labora\u00ac \ntories personnel visit some of the selected offices. \nDiscussions with Operating Company personnel on \nthe proposed forms and instructions and on the type \nof data required have proved very helpful, and \nhave served to acquaint these people with the ob\u00ac \njectives of the study.\n\nIn a typical systems study, five to ten thousand \ntrouble tickets may be received in the course of a \nyear. Full use of this information requires it to be \nsummarized in several wavs, and portions to be ex\u00ac \ntracted for detailed analysis. For instance, the analy\u00ac \nsis must show the troubles experienced in different \ntypes of apparatus and circuits and the indications \nthat were effective in revealing these troubles. Also, \nit may be necessary 7 to examine details, so that fac\u00ac \ntors such as the functional use of the apparatus at \nfault may be considered in the analysis.\n\nThrough use of punched cards the data are \nput into a form that will permit this flexibility in \nsummarizing and analyzing. The information ob\u00ac \ntained from each trouble ticket is translated into \ncodes which are, in turn, punched on the cards. \nEach card represents one trouble. A coding sys\u00ac \ntem requiring a minimum of translation has been\n\ndeveloped for this purpose. Where possible it uses \nstandard apparatus codes or direct entry of such \ninformation as the date, drawing number and con- \ntact numbers. Figure 2 shows a page of coded in\u00ac \nformation for tickets received in a study 7 of No. 4A \ntoll offices. Arbitrary codes are used only for in\u00ac \ndicating the office, class of report, trouble type, and \nto some extent, the apparatus tvpe. For apparatus \ndesignated bv a number, as in the case of the 280- \ntvpe relay, that number is entered directly. Let\u00ac \ntered types, however, are coded.\n\nThe tickets, in general, can be coded quickly and \neasily In this ystem. Every study, however, in\u00ac \ncludes tickets with inconsistencies, obvious errors, \nincomplete information, or statements that require \nfurther clarification. Tickets, therefore, must be in\u00ac \nspected individually before they are coded, and \nthose raising questions that cannot be resolved are \nsent to the study office either for more informa\u00ac \ntion or for explanations.\n\nAfter the cards have been punched, business ma\u00ac \nchines sort and summarize them. The figures pro\u00ac \nvided by these summaries are used to compute \ntrouble rates \u2014 the number of troubles related to \nthe amount of equipment or number of calls han\u00ac \ndled. Table 1 shows a typical summary of trouble \nrates for different types of relays, taken from a sys\u00ac \ntems study conducted in No. 5 crossbar offices.\n\nTo understand the significance of trouble rates, \nmany factors must be considered, such as the kinds \nof functions performed, whether troubles occurred\n\nFig . 3 \u2014 Portion of chart of U and Y open-contact \ntroubles prepared to show the effect of installation \nactivity on trouble rates in three telephone offices . \nTHE AUTHOR_\n\nin normal operation or were identified by routine \ntesting, and whether the troubles occurred in cir\u00ac \ncuits having a high or low frequency of operation. \nFigure 3 is a chart prepared for study of the rela\u00ac \ntion sli ip between open-contact trouble rates and \nthe activities in progress in the offices. Such activity, \nby adding to the dust in the air, may cause an in\u00ac \ncrease in the number of open contacts.\n\nAnalysis of the data also includes statistical checks \non the consistency of the results from the individual \noffices. A control chart for \u201cdefects per unit\u201d is \nused to spot instances where the data from a par\u00ac \nticular office, because of unusual conditions, may \ndeviate too far from the average to be representa\u00ac \ntive of the system as a whole. Such data may be \nexcluded from the study results.\n\nThe final report presents summaries and includes \npertinent, detailed data for areas of particular in\u00ac \nterest. It also discusses probable factors affecting \nmalperformance and states conclusions. Engineers \nsometimes request additional details on particular \ntypes of trouble. Such data may come from sum\u00ac \nmaries already made. If desired, the trouble tickets \ninvolved can be identified from the punched cards \nand drawn from the file for examination.\n\nSignificant results, at times unpredictable by \nother methods, have been obtained from these \nstudies. In one case, results indicated the need for \nfurther study of a particular type of relay, with re\u00ac \nsulting changes in design. Frequently, of course, \ndata affirm the success of design work by showing \nconsistently low rates of trouble. Study results have \ncontributed to the search for improved reliability \nof telephone switching equipment, and provide a \nbackground of reference data for estimating expect\u00ac \ned trouble rates in new design and development \nwork. The nature of the data, combined with the \ndetails available, makes these studies a valuable \nsource of information for many uses.\n\nMrs. Charlotte H. Hamilton, a native of New York City, joined the Develop\u00ac \nment and Research Department of the A.T.&T. Company after receiving a B.S. de\u00ac \ngree from Teachers College, Columbn University, in 1924. She became a member \nof the Laboratories in 1934. Her work has been concerned primarily with central- \noffice maintenance problems, and she lias conducted performance studies in toll \nand local switching offices and in AM A centers. Her work during World War II \nincluded the writing of instructions for the maintenance and testing of com\u00ac \nmunications equipment for the armed services.\n\nIn automatic telephone switching systems, thousands of electrical connec\u00ac \ntions must he accurately established and frequently revised. A new circuit \nhas been developed for verifying the integrity of cross-connections in No. 5 \ncrossbar offices \u2014 equipment which allows the frameman to perform his \nfunctions faster and more conveniently than was formerly possible.\n\nFor the purpose of dialing calls or billing cus\u00ac \ntomers, a telephone is identified bv its number \nappearing on the telephone instrument \u2014 the di\u00ac \nrectory number. In a No. 5 crossbar office the \nsame telephone line is also identified, for switching \npurposes, as a location on a particular line-link \nframe\u2014the frame of crossbar switches that com\u00ac \nprise the first stage or last stage of automatic \nswitching in the office. This information consists \nof a series of numbers specifying (1) the line-link \nframe on which the telephone line appears, (2) \nthe horizontal group of crossbar switches, (3) the \nvertical section of the crossbar switches called \n\u201cvertical group,\u201d and (4) the \u201cvertical file.\u201d The \nvertical file number refers to a particular line in \nthe vertical group, and thus completes the identi\u00ac \nfication of the telephone line.\n\nThis series is known as the \u201cequipment number,\u201d \nand it has no fixed relationship to the directory \nnumber. In fact, the arrangements are designed\n\nVo. 5 crossbar system; author , left , and R. J. De Dona of \nthe N. Y. Telephone Company setting up line tests .\n\nfor \u201cfull flexibility,\u201d whereby any directory num\u00ac \nber can be associated with any equipment number. \nThis permits telephone traffic to be distributed \nevenly throughout the line-link frames to obtain a \nuniform grade of service to all customers.\n\nFlexibility of equipment number-versus-directory \nnumber, however, requires the initial installation \nof certain wiring on a \u201ccustom basis\u201d for each tele\u00ac \nphone. Thereafter, occasional revision of this wir\u00ac \ning is required to maintain a balance of traffic or \nto process service orders. These wires, called cross- \nconnections, are so arranged that they may be \nchanged quite easily. Checking them for correct\u00ac \nness is called line verification.\n\nPrior to development of the new line-verification \nequipment, verification tests were made by using \nthe master test control circuit, which is part of the \nmaster test frame in a No. 5 crossbar office. In \nmany offices, there are periods when this frame \nis almost continuously in use bv maintenance per\u00ac \nsonnel making routine operational tests on markers, \nregisters, senders, trunks or other units of equip\u00ac \nment. Thus, verification tests of cross-connections \nadd to the work load of the maintenance center. \nTo relieve this congestion, and at the same time \nto make the testing of cross-connections convenient \nto the frameman who actually installs them, a sepa\u00ac \nrate line-verification facility has been developed. \nThis verification facility duplicates but does not \nreplace the verification features of the master test \nframe in the office.\n\nFor the purposes of translation, cross-connections \nare located in the number-group, line-link frame, \nand AM A translator. The cross-connections of the \nnumber group are used for translating the last four \ndigits of the directory number to its corresponding \nequipment number, and also the ringing combina\u00ac \ntion and office-code group. The cross-connections of \nthe line-link frame are used for indicating the class \nof service assigned to a vertical group and vertical \nfile. The AMA translator cross-connections are\n\nFig. 2 \u2014 Relay unit and control unit of the line- \nverification equipment mounted on same frame.\n\nlays are used. Figure 2 shows the relay unit \nmounted in the upper part and the control unit \nmounted with it in the lower part of the frame.\n\nThe line-verification circuit operates on the prin\u00ac \nciple of \u201cmatching.\u201d Information requiring transla\u00ac \ntion is sent to a marker for obtaining translation \nfrom a number group or to a transverter for obtain\u00ac \ning translation from an AMA translator. At the \nsame time, however, the known translation of this \ninput information is preset on switches in the veri\u00ac \nfication or test circuit. When the returned trans\u00ac \nlation is the same as that on the preset switches \nin the test circuit \u2014 that is, when the two items \nof information match \u2014 a \u201ccheck\u201d relay is operated \nin the test circuit for that particular cross connec\u00ac \ntion. When check relays have operated for each \ncross connection associated with the particular \nline being tested, a verification match results.\n\nFor example, if the directory number 9753 is \nto be verified, this number would be set up on \nthe \u201cthousands,\u201d \u201chundreds,\u201d \u201ctens\u201d and \u201cunits\u201d \nswitches of the verification circuit. Along with \nthis information, the corresponding equipment num\u00ac \nber, which is obtained from office records or the \nservice order, is preset on switches of the verifica\u00ac \ntion-circuit panel. Such equipment numbers are \nrepresented, for example, as \u201cvg2\u201d (vertical group\n\nused for translating an equipment number to a \ndirectory number, which is just the reverse of the \nnumber-group translation.\n\nCross-connections are removed, added, or \nchanged to concur with disconnected, new, or re\u00ac \nlocated telephone service. Since an error in any \ncross-connection will result in a wrong number, \nit is important that these leads be installed cor\u00ac \nrectly, With this new verification circuit, cross- \nconnections may be checked immediately by the \nframeman after he installs them.\n\nThe apparatus of the line-verification circuit is \nmounted on two equipment units, one called a \u201ccon\u00ac \ntrol\u201d unit and the other a \u201crelay\u201d unit. The con\u00ac \ntrol unit is a panel 12 inches high on which are \nmounted switches, keys and lamps. This unit can \nbe mounted on a wall, the end of a distributing \nframe, a desk, or a relay rack. Figure I shows the \nmounting arrangement at the central office in \nTuckahoe, N. Y.\n\nThe relav unit is 32 inches high; it accommo\u00ac \ndates nine mounting plates and may be mounted \ntogether with the control unit, or it may be mounted \nremotely from the control panel. Wire-spring re\u00ac\n\nFig. 3 \u2014 The \" matching \u201d principle: simplified dia\u00ac \ngram shot vs verification of cross-connection in \nnumber group of a Ao. 5 crossbar system.\n\nA partial verification circuit is shown in Figure 3, \nwhich illustrates a case wherein it is desired to \nverify that a number-group cross-connection serv\u00ac \ning the directory number 9753 is valid. The No. 5 \ncrossbar marker takes the 9753, selects the proper \nnumber group, and receives back the proper equip\u00ac \nment number translation. Figure 3 shows circuitry \nfor only vertical group \u201c2\u201d and horizontal group \n\u201c3/' The vgn2 and hgn3 relays are merely two of \nthe many relays that operate in the marker as the \nresult of this translation. The verification circuit \nhas access, through the preset vg and iig switches \nseen in the right part of Figure 3, to the correspond\u00ac \ning vgn2 and hgn3 relay-operating paths in the \nmarker. Thus, the vg and hg relays (extreme right \nin Figure 3) will operate and lock locally in the \nverification circuit. To obtain an over-all verifica\u00ac \ntion check, all relays associated with cross-connec\u00ac \ntions must he operated in this manner in the veri \nfi cat ion circuit, and ground is therebv closed to \noperate the mlvm (marker line verification match) \nrelay. If one of the cross-connections \u2014 say hori\u00ac \nzontal group \u2014 is installed incorrectly, the hg relay- \noperating path in the verification circuit would be \nopened. The reason for this is that no other lead \nwould pass through the hg switch, since this switch \nhas been preset to the HG3 position. Thus, the iig \nrelay in the verification circuit would not operate, \nwhich prevents the mlvm relay from operating. \nIn this ease, a lamp would light on the control \npanel to indicate that the incorrectlv installed cross- \nconnection was the one associated with the hori\u00ac \nzontal-group information.\n\nThis verification circuit has access to either of \ntwo marker groups. Bv virtue of the fact that the \nmarker is used for access to the number-group and \nline-link frame cross-connections, this function is \ncalled \u201cmarker line verification\u201d or mlv. Similarly, \nsince the transverter is used for access to the trans\u00ac \nlator cross-connections, this function is called \u201ctrans\u00ac \nverter line verification\u201d or tlv.\n\nThe new verification circuit fits into the No. 5 \nsvstem as shown in Figure 4. With this block dia\u00ac \ngram and the flow numbers \u2014 the numbers on the \ndiagram indicating each step in the call \u2014 the \noperational sequence of making a line-verification \ntest may be followed.\n\nIf the verification is to be of the mlv or marker \nline verification type, the office-designation infor\u00ac \nmation and the number of the line to be verified \nare preset on switches to be transmitted to the\n\nmarker for use by the number group. At the same \ntime, the correct line location, ringing combina\u00ac \ntion and class-of-service information for that line \nare also preset on the control panel.\n\nAt the beginning of each test, a bid is made for \ntest preference in the master test frame connector \n(1). This is necessary since this test circuit uses \nthe master test frame and test relays in the markers \nand transverters in common with other test circuits. \nIf no other circuit has access to the master test \nframe connector at this time, or when other circuits \nwhich had access to the master test frame con\u00ac \nnector become idle, preference is given the line- \nverification circuit. All other circuits which would \ncause interference are locked out of the master test \nframe connector.\n\nThe line-verification circuit connects, via the \nmaster test frame connector, to a marker within the \nparticular group (2) and transfers the number and \noffice code group designation to the marker in \nthe same manner as an incoming register. The \nmarker connects to the proper number group (3)\n\nFig. 4 How the verification circuit is associated \nwith other circuits of the No. 5 crossbar system; \nthe numbers indicate steps in the test sequence .\n\nFig. 5 \u2014 Verification circuit conveniently located at right enables checking of the maze of cross-connec\u00ac \ntions in number group frames at left and AM A translator frames at the center of the photograph.\n\nand transfers the number to it. The number is \ntranslated into a line location and ringing combina\u00ac \ntion which the marker requires, and this informa\u00ac \ntion is returned to the marker (4). Using the line- \nlocation information, the marker connects to the \nline-link frame (5) and attempts to set up a con\u00ac \nnection in the normal manner.\n\nThe marker has recognized that this is a test call \nand therefore does not complete the call to the line. \nFrom the marker action, however, the cl ass-of- \nservice information is obtained from the line-link \nframe by the marker (6). Since the verification \ncircuit has access to the leads over which the \nmarker receives the information from both the num\u00ac \nber group and line link frame (7), a verification \nlamp is lighted if the settings of the corresponding \nswitches agree with this information. If one or more \ncross-connections are in error \u2014 thereby' causing \nwrong information to be returned from the number \ngroup or line-link frame \u2014 a lamp is lighted to indi-\n\nThe A\\IA translator cross-connection check mav \nbe made independently of a marker line verifica\u00ac \ntion, or the verification circuit may be arranged to \nrecycle directly to the tlv stage immediately fol\u00ac \nlowing a succesful mlv verification. The line loca\u00ac \ntion and number of the line are preset on the same \nswitches used in mlv. The office index, which con\u00ac \nsists of a single digit in lieu of the first three digits \nof the seven-digit directory number, is also set on a \nrotary switch. In a manner similar to that of the \nmlv test (1), the line-verification circuit connects \nto a transverter and transfers to it the line location \n(8). The transverter in turn transfers this line \nlocation to the selected translator (9). The trans\u00ac \nlator translates the line location into an office index \nand a four-digit directorv number, and then trans\u00ac \nfers these back to the transverter (10). If this in\u00ac \nformation matches the office index and directory \nnumber previously preset on switches, an \u201cok\n\nE. G. Crane, ]r, s a native of Trenton, N. ]., received a B.S. degree in Electrical \nEngineering at Duke University in 1942. After a period with the General Electric \nCo., he served during World War II as a Navv airborne electronics officer. Im\u00ac \nmediately thereafter he joined the N. f. Bell Telephone Company where he was \nconcerned with central office power plant equipment engineering. In 1951 he \ntransferred, on loan, to the Laboratories at Whippany, where he was involved \nwith the development of a military system for the Navy. In 1953 Mr. Crane \ntransferred to West Street where he became concerned with the development \nof the No. 5 crossbar system, particularly in the application of wire-spring relay \ncircuits. The following year he became a permanent member of the Laboratories \nand in recent years has been involved with circuit development of maintenance \nfacilities. He is a licensed professional engineer and received a M.S. degree from \nStevens Institute of Technology in 1949.\n\nmatch\u201d indication is given by a lighted lamp (11). \nIf a mismatch exists, which indicates that the instal\u00ac \nlation of a cross connection was incorrect, a lamp is \nlighted to indicate the office index or directory \nnumber that is involved in the mismatch.\n\nAlthough the end result of tests with this equip\u00ac \nment is primarily a visual display to indicate a \nparticular line verification, the circuit also has the \nability to leave behind a permanent record of a \nverification if one is desired. This is produced on \nthe office trouble recorder and is obtained by oper\u00ac \nating the record rec key at the control panel of the \nverification equipment. When the rec key is oper\u00ac \nated, upon each ok verification, a request is sent \nto the marker or transverter for a record. A record \ncard is produced by the trouble recorder with all \nthe pertinent information \u2014 line number, equip\u00ac \nment number, date and verification check. If de\u00ac \nsired, this information may be filed by the Oper\u00ac \nating Company as a permanent record of the line\n\nA field trial of this equipment has been success\u00ac \nfully conducted at the Tuckahoe office of the New \nYork Telephone Company. The control panel at \nTuckahoe is located at the end of the main dis\u00ac \ntributing frame, which is approximately midway \nbetween the number group and AMA translator \nframes. The relay panel is mounted on the miscel\u00ac \nlaneous relay rack on another floor. Since this office \nhas all number groups, line-link frames and AMA \ntranslators on one floor and the maintenance center \non another floor, and further, since there are two \nmarker groups, it provided an excellent oppor\u00ac \ntunity to observe the test circuit under conditions \nwhere it is most needed. It proved to be very reli\u00ac \nable, and its performance was most satisfactory. \nThe trial also demonstrated the desirability for a \nframeman to check his cross-connections immedi\u00ac \nately after installation without interrupting other \ntesting at the maintenance center.\n\nPatents Issued to Members of Hell Telephone \nLaboratories During January\n\nBowman, B. M., and Rieke, J. \\Y. - Polarizing Circuit for \nTelevision Signals or the Like \u2014 2,820,181.\n\nGorgas, J. \\Y., Jacobitti E., Jaeger. R. J\u201e Jr., Morrison, C. \nG., and Newsom, [. R. \u2014 Trouble Recording on Time- \nOut Circuit for Automatic Telephone \u2014 2,820,099.\n\nWhelan. J. M. \u2014 Method of Purifying Silicon Tetrachloride \nand Germanium Tetrachloride \u2014 2.821,460.\n\nIn supplying radio channels for the AN/TRC-24 \nmilitary telephone system, it was necessary to \ndesign filters for use between the antenna and \ntransmitter and between the antenna and re\u00ac \nceiver. The coaxial type of design resulted in a \nsmall, efficient package, and it was found pos\u00ac \nsible to divide the entire 50 to 600 megacyles \nrange into only twenty-two sub-hands, with each \nfilter suitable for transmission and reception.\n\nSome unknown radio engineer once observed \nthat the worth of a radio receiver is judged not by \nwhat is received but by what is rejected. He was \nprobably a transmitter designer who was plagued \nby complaints that his transmitter caused interfer\u00ac \nence in receivers that he considered insufficiently \nselective. The receiver designer similarly might \nhave said that the worth of a radio transmitter is \njudged not bv what is transmitted but by what is \nsuppressed. Such epigrammatic statements, only \npartially true and intended to impress and possibly \nto confuse, merely emphasize the importance of the \n\u201cselectivity\u201d of the device. This may be defined \nas the ability of the receiver (or transmitter) to \nreceive (or transmit) the desired radio frequency \nsignal, while preventing the reception (or trans\u00ac \nmission) of unwanted frequencies.\n\nFigure 1 is a block schematic showing the circuit \nlocation of the filters. It is a characteristic of such \nfilters that the greater the tuning range, the greater \ntheir complexity for a given amount of selectivity. \nThe tunable ranges of AN/TRC-24 are the \u201ca\" \nband, 50-100 megacycles; \u201cb\u201d band, 100-225 me; \n\u201cc\" band, 225-400 me; and \u2018V band, 400-600 me. \nFilters, tunable over each of these bands and fur\u00ac \nnishing the required selectivity, would be large, \nmechanically intricate, and expensive. \\ better \nsolution, and the one adopted, is to divide the \nbands into smaller sub-bands and to design filters \nfor each sub-band. In this wav each filter tunes and \nprovides selectivity over a relatively narrow sub\u00ac \nband, which considerablv reduces mechanical com\u00ac \nplex] tv, size and cost.\n\nThe scheme has its unattractive aspect, however. \nSince several filters are used to cover a main tuning \nband, the selection of a filter is dependent upon \nthe frequency to which the transmitter or receiver \nis tuned. The appropriate filter is plugged into the \nradio-frequency head of the transmitter or receiver, \nthe others lying idle and creating a storage prob\u00ac \nlem. Since the transmitter and receiver frequencies \nare not changed too often, howey'er, plug-in filters \ndo not impose a severe operational difficulty .\n\nThe \u201ca,** \u201ch,\u201d and \u201cc\u201d bands are covered b\\\u2019 six \nfilters each, while the \u2018V\u2019 band uses four filters. \nThe filters for the transmitter must pass 100 watts \nof rf power without undue dissipative loss or tem\u00ac \nperature rise. The receiver filters must pass only \ntransmitting^TV 1 /receiving\n\nFig . 1 \u2014Location of tunable filters in t \\ 7 /TRC-24 \nradio military telephone system .\n\nextremely small amounts of poyver; however, they \nare made identical to the transmitter filters to mini\u00ac \nmize the number of distinct designs. Actuallv, be\u00ac \ncause the transmitting and receiving frequencies \nused at any given location are different, a single \nset of filters is normally sufficient for a complete \ninstallation. Plug-in dummy filters are supplied \nyvith the radio equipment. When additional selec\u00ac \ntivity is not needed, the radio sets may be oper\u00ac \nated without filters.\n\nThe filter circuit that appeared best able to sat\u00ac \nisfy the electrical requirements is shoyvn schemati\u00ac \ncally in Figure 3. In the formula shoyvn along\n\nFig. 2 Two views of typical filter; outer cylin\u00ac \ndrical part of one coaxial section has been removed \nto shoiv features of the internal structure.\n\nyvith the diagram, R n is the impedance of the circuit \ninto yvdiich the filter yvorks, and f r is the midband \nfrequency of the filter's pass-band.\n\nIt should be realized that f r changes as the filter \nis tuned over its oyvn sub-band. For design pur\u00ac \nposes, f r is chosen as the middle of the sub-band, \nyvhich of course means that the design is theo\u00ac \nretically ideal only yvhen the filter is tuned to the \ncenter frequency. But since the sub-bands are not \nunduly yvide, the departure from the ideal is small, \neven yvhen the filter is tuned to the edge of its \nsub-band. An advantage of this type of circuit \nis that by the use of coupling coils the designer may \nchoose the impedance level of the resonant cir\u00ac \ncuits for structural convenience, and then select\n\nFig . 3\u2014Circuit for the tunable filter; coaxial-type \nunits are used for L 2 -C 2 -C, resonant circuits.\n\nthe coupling factor to yield the best over-all trans\u00ac \nmission performance.\n\nAfter appropriate theoretical designs were \nevolved, it was necessary to translate them into \nphysical elements. The frequenev range under \nconsideration, 50 to 600 megacycles, is not well \nsuited to the design of lumped-element filters \u2014 \nthat is, filters consisting of the conventional ca\u00ac \npacitor and inductor components. At these fre\u00ac \nquencies, lumped elements involve troublesome \nparasitic effects from connecting leads, and it is \ndifficult to construct them in convenient sizes pos\u00ac \nsessing adequately high \u201cq\u201d (small dissipation), \nand high power-handling capability. The 50-600 \nme frequency range, however, is well-adapted to \nthe use of filters of the coaxial type. Elements of \nthis type are comparatively free from the afore\u00ac \nmentioned objectionable features of lumped ele\u00ac \nments, and in addition are self-shielding and of \nreasonable size. The filters, accordingly, were con\u00ac \nstructed using these elements for the resonant cir\u00ac \ncuits designated L 2 ~C 2 -C ;} in Figure 3. The cou\u00ac \npling inductances, L h were made in the form of a \nsingle-turn loop, since extremely high \u201cq\u201d is not \nrequired of them. The input and output coupling\n\nFig. 4 \u2014 Loss - frequency characteristic of filter \nshoun in Figure 2; inset: expanded graph of region \nof the characteristic curve near 152 me.\n\ninductances to a cavitv are positioned to effect a \nminimum coupling between them, which permits \nthe filters to realize the maximum out-of-band \ndiscrimination.\n\nFigure 2 shows a typical filter with one of the \nouter coaxial conductors removed for display pur\u00ac \nposes. If this illustration is viewed in combination \nwith the schematic diagram, Figure 3, the com\u00ac \nponent elements may be identified and the coaxial \ntype of construction is evident. For example, the \nadjustable coaxial capacitors, C.* of Figures 3 and \n5, tune the filter to the required frequency. This \nparticular filter has a useful tuning range of 142 to \n163 megacycles per second. Figure 4 is an inser\u00ac \ntion-loss characteristic of the filter when tuned to \nthe frequency 152 megacycles per second. The \nband loss is small, since the \u201cq\u201d of the coaxial ele\u00ac \nments is high. The out-of-band discrimination, or \nselectivity, more than satisfies the design require\u00ac \nments for the system.\n\nAfter achieving an appropriate electrical design, \nit is necessary to design a suitable mechanical struc\u00ac \nture. Such a structure must be accuratelv dimen-\n\nsioned to provide accurate electrical properties and, \nin addition, must he ruggedlv constructed to with\u00ac \nstand shock and vibration of transportation, and \nsevere weather conditions. Further, the structure \nmust he of a form that can he economically manu\u00ac \nfactured in large quantities, and of a form that can \nreadily he plugged into the radio-frequency heads \nof the transmitter and receiver. All of these features \nare incorporated in the structure shown in Figure 2, \nwhich is the basic structure used in all filters.\n\nThe dial mechanism is of a special design; it has \nthe axial motion required to varv the capacitance \nof the coaxial capacitor used to tune the filter. \nSimultaneouslv. the mechanism drives a dial plate \nwhich is calibrated in terms of the various channels\n\nWhen the filters are not in use they are housed in \nconvenient transit cases. These cases are similar \nin design and construction to the ones used for \nthe transmitter and receiver. They provide protec\u00ac \ntion against adverse environmental conditions dur\u00ac \ning transportation and storage.\n\nElectrical and mechanical tests were applied to \nrepresentative models of these filters hv the U. S. \nSignal Corps. The filters performed excellently, \nand quantity production by the Western Electric \nCompany was undertaken. Although expressly \ndesigned for the AN/TRC-24 radio set, the Signal \nCorps has found them useful in a variety of other \napplications.\n\nM. I). Bhill, a native of New York Cite, joined Bell Telephone Laboratories \nin 1930 after receiving the A.B. degree and the A.M. degree in Physics from \nColumbia University. At the Laboratories he has been principally engaged in the \ndevelopment of communication networks for Bell System use, except for a period \nduring World War II when he worked mainh on radio counter-measures devices. \nAfter the war, Mr. Brill assumed responsibility for a group designing microwave \nfilters and networks for the TD-2. TH and Tf microwave radio-relay systems. \nFor the past year lie has been in charge of a group designing networks for \nmilitary projects.\n\nK. M. Jensen, whose homo town is Hurlcv, South Dakota, received the B.A. \ndegree from the University of South Dakota in 1934 and the B.S. degree from \nPurdue in 1937. He then joined the Laboratories and worked on the design of \naudio equipment, and subsequently worked in the Quality Assurance Department. \nIn 1939 he transferred to the Transmission Networks Department where, during \nWorld War II. he designed amplifiers, oscillators and filter circuits. Later, he was \nengaged in the design of networks for carrier-telephone systems and tunable co\u00ac \naxial filters tor the AN/TRC-24 radio-relav system. Mr. Jensen is currently en\u00ac \ngaged in tlie development of microwave apparatus for the TJ and TH systems, \nand of coaxial apparatus for a mobile radio system.\n\n/\\ D. Gam [left) and H . M. Gilroy examine polar - \no graph to ascertain elements in a metal solution.\n\nThe behavior of an electron tube in a circuit \ndepends to a large extent on the behavior of its \nthermionic cathode. Thus, an accurate analysis of \nthe cathode metal aids in furthering the life expect\u00ac \nancy of the tube. Long life is particularly advan\u00ac \ntageous for electron tubes that are used in such \ninaccessable equipment as the recently installed \nunderseas cables.\n\nThe nickel used in cathode material contains \nimpurities that affect the ease and degree of activity \nand the emission life. For example, certain elements \nsuch as magnesium or aluminum are believed to \nenhance the tube behavior, while other elements \nsuch as silicon or manganese may have an adverse\n\nAnalysis of aluminum. The first increase in cur\u00ac \nrent indicates the reduction of an aluminum-dye \ncomplex . The second increase indicates the reduc\u00ac \ntion of the uncomplexed dye . The closely spaced \npeaks are caused by the formation and falling auay \nof individual drops of mercury.\n\neffect. It is important to know accuratelv the con\u00ac \nstituents of cathode metal so that cathode be\u00ac \nhavior can be correlated with the amount of im\u00ac \npurities. With such information, manufacturers can \nselect proper alloy combinations to produce tubes \nwith longer life and higher thermionic activitv.\n\nAt Bell Laboratories a study has been made to \ndetermine the most accurate method of analyzing \nsamples of manufactured cathodes. These sam\u00ac \nples are analyzed after fabrication to determine \nwhat impurities remain after chemical reactions \nassociated with the manufacturing process have \ntaken effect on the original nickel alloy. These reac\u00ac \ntions include processes such as absorption and seg\u00ac \nregation, and they also include the effects of a vola\u00ac \ntile substance.\n\nThe most rapid technique for this analysis em\u00ac \nploys the spectrograph for the determination of the \npresence of extraneous substances in nickel cathodes. \nIn this method, the entire cathode is converted into \nan oxide and mixed with appropriate buffering \nagents. The resulting powder mixture is packed \ninto a crater in a graphite rod. This composite is \nbrought into contact with a second graphite rod. \nAn applied dc potential across these two \u201celec\u00ac \ntrodes,\u201d as the interval between them is opened, \nmaintains an arc. Radiation from this arc is viewed \nthrough a prism which scatters (spreads out and \nseparates) the light in a large number of character\u00ac \nistic wavelengths or colors. The scattered light is \nintercepted either by a photographic plate or by \nphotocells. By studying the spectrum, the chemist \ncan detect the presence of elements from the num\u00ac \nber and color of lines observed. He can also meas\u00ac \nure the quantity of these elements from the intensity \nof the color lines.\n\nIn searching for a more accurate technique, P. D. \nGarn and H. M. Gilroy of the Laboratories Chem\u00ac \nical Research Department have investigated an\u00ac \nother way of analyzing nickel cathodes \u2014 the \nmethod of polarography. They have found this \nmethod to be feasible, and thus have improved \nthe average accuracy from five per cent with spec- \ntrography to one or two per cent with polarography. \nA further advantage of this method is that no stand\u00ac \nards similar to the sample are required. Each im\u00ac \npurity is determined independently, and the pres\u00ac \nence or absence of any other element has no ef\u00ac \nfect on an individual determination.\n\nThe polarograph uses the characteristics of cur- \nrent-voltage curves to mark the existence and \namount of an element in a solution. To determine \nthe cathode constituents, the metal is dissolved in \nsolution. A series of precipitations, filtrations, evap\u00ac \norations, electrolyses, and additions of compounds \nseparates the solution into workable sub-solutions \nthat are studied polarographically.\n\nA pair of electrodes \u2014 a polarizable cathode and \nnon-polarizable anode \u2014 are immersed in the solu\u00ac \ntion to be studied. The electrodes are connected\n\nto an external circuit, which includes a pen-re\u00ac \ncording apparatus. A voltage is applied and is \nsteadily increased.\n\nAt a particular voltage the current flow suddenly \nincreases rapidly. This \u201cpolarographic wave\u201d is \ndue to the reduction (change in valence state), of \nthe ion at the cathode. Each element has a char\u00ac \nacteristic reduction potential and the voltage at \nwhich the current suddenly increases can be used \nto identify the ion. The current does not increase \nindefinitely because the supply of ions in the im\u00ac \nmediate vicinity of the cathode is quickly removed. \nThe current is limited, therefore, by the rate of \ndiffusion of the ions to the cathode surface. The \nrate of diffusion, and thus the current, is propor\u00ac \ntional to the concentration of the ion in the bulk \nof the solution.\n\nDetermining the presence and amount of iron \nand manganese in the nickel alloy represents a \nnovel analytical method. The addition of triethano\u00ac \nlamine to solutions containing these two ele\u00ac \nments results in an improved constant current and, \nin turn, an improved definition of the polarographic \nwave that represents the ions.\n\nThe Western Electric Company\u2019s total sales in \n1957 were $2,480,614,000, an increase of 4.5 per \ncent over 1956, according to the annual report to \nstockholders, published February 21. Sales to Bell \nCompanies reached a high of $1,821,198,000, a \n10.5 per cent increase over 1956 sales.\n\nThis is the largest volume of business ever han\u00ac \ndled by the Company in a single year, although \nchanges toward the end of the vear by the Bell \nCompanies in many of their construction programs \ncaused a leveling off in Western Electric produc\u00ac \ntion and service activities.\n\nAbout one quarter of total sales, or $575,175,000, \nwas to the Government for materials and services \nrelated mainly to national defense projects, among \nthem the Distant Early Warning Line, in the Arc\u00ac \ntic, which was completed in July; the White Alice \nSystem, in Alaska; and production of the Nike \nground-to-air missile system.\n\nDespite some downward revisions of the Com\u00ac \npany\u2019s production program in the final months of \nthe year, output of many types of telephone prod-\n\nnets exceeded 1956 peaks. By the end of the year \n6,668,000 telephone sets, 134,500 million conductor- \nfeet of exchange cable and enough dial central \noffice switching equipment to serve 3,822,000 lines \nwere shipped. Company installers worked on some \n60,000 projects in 7,200 cities and towns through\u00ac \nout the United States, setting new records for in\u00ac \nstalling, testing and readying for the Bell telephone \ncompanies\u2019 use, switching and other equipment for \n3,614,000 dial lines.\n\nThe Company made purchases directly from \nabout 33,000 companies, 90 per cent of them small \nbusinesses. In addition, transportation services were \nprovided by 3,000 other firms. The total paid by \nWestern Electric in 1957 for materials, parts, sup\u00ac \nplies and transportation was $1,224,000,000.\n\nIn connection with development of the Compa\u00ac \nny\u2019s technological resources during 1957, the annual \nreport makes special note of the establishment of \nthe Graduate Engineering Training Program and \nof the first steps toward establishing an engineer\u00ac \ning research center at Hopewell Township, N. J.\n\nAt the request of the A.T.&T. Co., Bell Labora\u00ac \ntories has developed equipment for trials of a new \ncommunications service. During February, the new \nservice\u2014called \u201cDataphone\u201d\u2014was offered on a lim\u00ac \nited basis in three Bell System areas: Illinois Bell. \nMichigan Bell and the New York Company.\n\nDataphone, which could be one of the most sig\u00ac \nnificant business developments of our times, is a \nhigh-speed transmission service for sending and \nreceiving data over regular telephone lines (sec \nRecord, April , 1957, and September , 1957), It can \nsend information over telephone facilities 10 times \nfaster than present methods\u2014about 800 words a \nminute.\n\nDesigned to send information to and from cus\u00ac \ntomers\u2019 data-processing machines in various loca\u00ac \ntions, Dataphone service makes use of special de\u00ac \nvices \u2014 subsets \u2014 which can handle various types \nof business-machine codes. They interpret and \ntransmit data read from punched cards, magnetic \nand paper tapes and other media. Thev can also \ntransmit text. Subsets are manufactured bv Western \nElectric.\n\nWherever there is a telephone there can be a \nDataphone subset \u2014 one of the advantages of the \nnew service. It can be plugged into a standard \nelectrical outlet to get its source of power.\n\nDataphone service has a promising future. \nEventually a person using the service will be able \nto transmit data to anv point on the telephone net-\n\nV. A. Vaughan, Jr., of the A.T.&T. Co. demonstrating Data - \nphone subset equipment during press showing of the neic \nsystems for transmitting data over telephone lines.\n\nwork. It is also economical: a customer will pav \nfor the service much as he would for regular tele\u00ac \nphone service 1 . His main charge is based on how \nmuch he uses it.\n\nA businessman in New York can pick up a tele\u00ac \nphone associated with a Dataphone subset and \ncall another division of his company in Chicago, \nHe tells his associate in Chicago lie\u2019s going to send \nhim some payroll data. Each presses a button on \nhis telephone, and information starts flowing over \nthe circuits from a business machine or from a \ntape prepared in advance. His charge for sending \ndata is the same as for a regular long-distance tele\u00ac \nphone call.\n\nIf the businessman has several branches of his \ncompany scattered across the land, he can trans\u00ac \nmit data at high speeds to one or to all of his \nbranches. Or people in his branch offices can take \ninformation from their electronic counterparts and \nsend it back to headquarters.\n\nDataphone service can be provided at present \nusing two kinds of subsets \u2014 the Digital Subset \nand the Recorded Carrier Subset. The Digital Sub\u00ac \nset converts electrical pulses of data received from \noriginating equipment (computers or tabulators, \nfor example) to audible tones for immediate trans\u00ac \nmission over telephone lines. At the receiving end, \nthe reverse function is performed. The equipment \ntakes the tones, reconverts them into electrical im\u00ac \npulses and delivers the data to business machines.\n\nThe Recorded Carrier Subset through a tape re\u00ac \ncorder permits storage of information received \nfrom business machines for transmission at a later \ntime. Pulses are accepted from the machines at \nrelatively slow speeds and are recorded on mag\u00ac \nnetic tape. Subsequently, the Recorded Carrier \nSubset transmits the taped information to another \nsubset at the receiving end at a speed eight times \nfaster than the information was recorded. Thus, \n24 minutes of information can bo compressed into \na three-minute call. At the receiving end, the pro\u00ac \ncedure is reversed \u2014 the information is \u201cplaved \nback\u201d to business machines.\n\nAlso available are certain optional features not \nrequiring attendance bv operating personnel. For \ninstance, information can be stored on a Recorded \nCarrier Subset throughout the day and transmitted \nduring the evening from and to a master station. \nThus, Dataphone makes it possible for machines\n\nLeft: a Digital Subset for immediate transmission \nof data; and right: a Recorded Carrier Subset \nwhich stores many types of data on magnetic tape .\n\nA wide variety of business machines can be \nadapted to Data phone equipment: card readers \nand punches, various paper and magnetic tapes, \nand teletypewriter equipment. As new commercial \nmachines are developed, thev can be designed to \nbe used with Dataphone equipment.\n\nThe Oliver E. Buckley Solid-State Phvsics Prize \nfor 1958 was presented during a recent joint meet\u00ac \ning of the American Physical Society and the Ameri\u00ac \ncan Association of Physics Teachers. The prize, \nnamed for the former President of Bell Telephone \nLaboratories, was awarded to Dr. Nicolaas Bloem- \nbergen. Professor of Applied Phvsics at Harvard.\n\nBell Telephone Laboratories and the American \nPhysical Society established the prize, which con\u00ac \nsists of an award of $1,000 to a person adjudged \nto have made a most important contribution to the \nadvancement of knowledge in solid-state physics \nwithin the five years prior to the award. Dr. \nBloemhergen was the first to conceive a solid-state \nMASER capable of continuous operation.\n\nA new 7 open-shelf type of telephone \u201cbooth\u201d has \nbeen developed in the Station Apparatus Develop\u00ac \nment Department at Bell Laboratories and is now \nundergoing consumer tests at the Michigan and \nChesapeake and Potomac Telephone Companies. \nThe units, placed on trial through the A.T.&T. Co. \nCustomer Products Planning Division, may some \nday provide supplements to regular booths in in\u00ac \ndoor locations, for example in department stores.\n\nW estern Electric Co. made the shelf units at the \nQueensboro, N. Y., Shop, and have turned out the \ntest models in three colors, red, tan and green.\n\n{Below) D. H. King of the Apparatus Development \nDepartment testing new\u201cbooth\u201dat the Laboratories.\n\nThe Audio Engineering Society has announced \nthat Homer Dudley of the Laboratories has been \nawarded a Letter of Commendation for his paper \n\u201cFundamentals of Speech Synthesis. This paper \nappeared in the Journal of the Audio Engineering \nSociety , Vol. 3, October, 1955.\n\nin a letter accompanying the award, Mr. George \nHoriuchi, Editor of the journal, cited Mr. Dudley \nfor his work in speech synthesis. \u201cIn particular,\u201d \nMr. Horiuchi stated, \u201cthe excellent paper which \nyou contributed to our journal was the subject for \na great deal of favorable comment.\u201d Mr. Dudley, \na member of the Visual and Acoustics Research \nDepartment, is the inventor of the original Bell \nLaboratories Vocoder machine for the study of \ns peecli transmission.\n\nThis component has a constant-current feature \nwhich makes it ideally suited for a current regu\u00ac \nlator in circuits where either the load or supplv \nvoltage vary over wide limits. It can also be used \nas a current limiter or pnlse shaper. Its ac im\u00ac \npedance is very high, making it useful as a coupling \nchoke or as an ac switch.\n\nThe device, closely related in principle to the \nfield-effect transistor (see Record, May, 1955), con\u00ac \ntains a single planar junction which is made by the \ndiffusion technique. Current passes parallel to this \njunction through a constricted region called the \nchannel. As the voltage across the device is in\u00ac \ncreased, current increases, and a depletion laver \nbuilds up which eventually \u201creaches through** the \nentire thickness of the channel. At this point, called \nthe \u201cpinch-off\u201d point, a further increase in voltage \ndoes not produce any increase in current. Even\u00ac \ntually an avalanche breakdown occurs, as the volt\u00ac \nage is increased still further. Between pinch-off \nand breakdown, the current is essentially constant.\n\nAt present these varistors are fabricated by cut\u00ac \nting dice from a slice of germanium or silicon con\u00ac \ntaining a single diffused junction. The dice are \nheavily plated on all surfaces, and a circular trench \nis then cut into the diffused laver to within about \n0,1 mil of the junction (see Figure 1). This cutting\n\nrequires a high degree of precision, since the char\u00ac \nacteristics of the final device depend heavily on the \nspacing between the bottom of the trench and the \njunction. In exploratory work, the trench is first cut \nwith an ultrasonic tool and finished to the desired \ndepth by etching. Leads are then attached by \nthermo-compression bonding or other means.\n\nCharacteristics of the device can be altered by \nvarying such parameters as channel depth, impurity \ngradient, length and width of channel, and selec\u00ac \ntion of semiconductor material. Using silicon, units \nhave been fabricated with a regulated current of \none milliampere, pinch-off voltage of 10 volts and \nbreakdown of 150 volts.\n\nCurrent can be held constant to within one per \ncent over a voltage range of 20 to 120 volts. Ger\u00ac \nmanium units have been made with a rating of 10 \nmilliamperes, pinch-off of 10 volts and breakdown \nof 25 volts. It appears feasible at present to produce \nvaristors which regulate current at any level be\u00ac \ntween 10 microamperes and 10 milliamperes, and \nimprovements in fabrication techniques should \nmake higher current levels possible.\n\nFor circuit applications, a parasitic shunt capaci\u00ac \ntance of the order of a few mmfd. is present and \nmust be taken into consideration. In the constant- \ncurrent region, the ratio of ac resistance to dc re\u00ac \nsistance is typically 100 and may run as high as \n1000, making it ideal as a coupling device. It differs \nfrom a conventional choke since impedance is con\u00ac \nstant over an appreciable frequency range.\n\nFig, 1 \u2014 Cross section of a field-effect varistor. \nCircular trench is cut into the diffused layer.\n\nPlans for construction of a direct submarine tele\u00ac \nphone cable system between the continental United \nStates and Puerto Rico have been announced \njointly by the Long Lines Department of American \nTelephone and Telegraph Company and Interna\u00ac \ntional Telephone and Telegraph Corporation.\n\nThe announcement coincided with the filing of \nconstruction applications with the Federal Com\u00ac \nmunications Commission seeking authority to build \nand operate the cable system.\n\nThe plans provide for work to start on the cable \nsystem in 1958, with service scheduled for early \nin I960. The cost, according to present estimates, \nwill he about $17 million and will be shared equally \nby Long Lines and Radio Corporation of Puerto \nRico, a subsidiary of I. T. & T.\n\nTelephone traffic between the mainland and \nPuerto Rico has more than doubled during the last \nfive years. At the present time, telephone service \nis handled bv radio.\n\nTlie March, 1958, issue of The Bell System \nTechnical Journal contains the following articles:\n\nA etc Developments in Militanj Switching , bv A. \nC\\ Gilmore, P. R. Gray and W. S. Irvine.\n\nBroadband Oscilloscope Tube , by D. J. Branga- \nccio, A. F. Dietrich and J. W. Sullivan.\n\nOptimum Tolerance Assignment to Yiehl Mini\u00ac \nmum Manufacturing Cost , by David H. Evans.\n\nThe Measurement of Power Spectra from the \nPoint of View of Communications Engineering \u2014 \nPart II, by R. B. Blackman and J. W. Tukey.\n\nFollowing is a list of speakers, titles, and places of presentation \nof recent talks giv en by members of the Laboratories.\n\nAmerican Institute ok Electrical Engineers Winter General Meeting New York On\n\nBall. W. C., Ground Measurement for Communication Cir\u00ac \ncuit Protection Planning.\n\nPerrv, A. 13., Edson. J. O., and Flavin, M. A.. Synchronized \nClocks for Data Transmission.\n\nEarly, |. M., Xeic and Future Solid State Devices, Their \nProperties and Apjilication,\n\nAnderson, R. R., Dynamic Design and Test Problems of \nMagnetic Tape Recorders , Am. Soc. of Mechanical Engi\u00ac \nneers, Johns Hopkins University, Baltimore, Md.\n\nBennett, \\\\ . R., Digital Transmission as a Random Process, \nElectrical Engineering Department Seminar, M. I. T., \nCambridge, Mass.\n\nBerkerv, E. A., Potter Supply Design Factors, Ridgewood \nAmateur Radio Club of \\ T . J., Hohokns, N. |.\n\nBiondi, F. J\u201e The Cleaning of Parts for Electronic Devices, \nAm. Soc. for Testing Materials, Washington, D. C.\n\nRuppel, A. E., Methods of Evaluating The Transmission \nPerformance of the W Digital Data Signaling System. \nSliair, R. C, Lead-Acid Batteries in Telephone Service. \nSmith, 13. II. A l-\\Vatt Solar Power Plant .\n\nWeboi, E. A., An b\\l Digital Subset for Data Transmission \nOver Telephone Lines.\n\nWhitman, A. L.. Votaw, C. J., and Smith, C. W. (A.TAT.), \nThe HSR I Teletypewriter Selective-Calling System.\n\nMyers, P. B.. james, 13. B., and johannesen, f. 13.. A Two- \nTransistor Cate for Time-Division Switching.\n\nBozorth, R. \\f.. Modern Magnetic Materials, Washington \nSection, A.LETT, Washington, 13. C.\n\nBrady, G. W., Structure in the Vitreous and Litfnid State, \nSigma Xi Societv. Corning Glass Works, Corning, N. Y.\n\nBrattain, W. II., Development of Concepts in Semiconductor \nResearch , Cleveland Physics Society, Cleveland, Ohio.\n\nBrattain, W. H., Surface Physics of Semiconductors, Journal \nSeminar, Physics Dept., Case Institute of Technology, \nCleveland, Ohio.\n\nFrisch, H. L., and Lloyd, S. P., One-Dimensional Impurity \nBands, Am. Physical Soc., N. Y. C.\n\nFrisch, H. L., One-Dimensional Impurity Bands , Physics Col\u00ac \nloquium, Stevens Institute of Technology, Hoboken, N. J.\n\nGeller, S., Garnets \u2014 Crystal Chemical and Magnetic Studies, \nGeneral and Physical Chemistry Seminar, Cornell Uni\u00ac \nversity, Ithaca, N. Y.\n\nGibbons, D. F., Role of Dislocations in Plasticity, Montreal \nChapter, A. S. M., McGill University, Montreal, Canada.\n\nGrossman, A. J., Design of Networks, I.R.E. New York Sec\u00ac \ntion and A.I.E.E. Basic Science Division, N. Y. C.\n\nHagstrum, II. D., Electronic Transitions Between a Solid \nand an Approaching Ion , Edison Laboratory, West Orange, \nN. J., and Catalysis Club, Chicago, 111.\n\nIlersey, R. E., Sixty Million Telephones at Y our Fingertips, \nConnecticut Valley Alumni Association, Phi Kappa Psi \nFraternity, Hartford, Conn.\n\nJensen, A. G., Efficient Use of Bandwidth, Philadelphia \nA.I.E.E.-I.R.E, Lecture Series on Modern Communica\u00ac \ntion, Philadelphia, Pa.\n\nKing, J. C., Anelasticity of Synthetic Quartz at Low Tem\u00ac \nperature, Am, Physical Soc., N. Y. C.\n\nLaudise, R. A., Hydrothermal Growth of Crystals, Am. Inst, \nof Chemical Engineers, University of Virginia, Charlottes\u00ac \nville, Va.\n\nLiehr, A. D., The Coupling of Vibrational and Electronic \nMotions in Degenerate Electronic States , University of \nMichigan, Ann Arbor, Mich.; University of Colorado, \nBoulder, Col.; and University of Chicago, Chicago, III.\n\nMacKintosh, 1. M., New Developments in Transistor Engi\u00ac \nneering, Congregational Church, Scarsdale, N. Y.\n\nMaeWilliams, W. H., The Use of Computers for Engineering \nDevelopment and Design, New York Section of the I.R.E. \nProfessional Group on Engineering Management, N. Y. C.\n\nMcMillan, B., A Theory of Communication , University of \nPennsylvania, Philadelphia, Pa.\n\nMertz, P., Information Theory Impact on Modern Comma nL \ncations , Michigan Section A.I.E.E., Communications Tech\u00ac \nnical Committee, Michigan State University, East Lansing, \nMich.\n\nOwens, C. D., A Look at Modern Magnetics for the Engi\u00ac \nneers, Merrimack Valley Subsection, A.I.E.E., Branch \nHall, Lawrence, Mass.\n\nPierce, J. R., Some Problems of Space Travel, Federal Tele\u00ac \ncommunications Laboratories, Nutlev, N. |.\n\nPrince, E., Neutron Diffraction Problems on Magnetic \nOxides, U. S. Naval Research Laboratory, Washington, \nD. C., and M. 1. T., Lexington, Mass.\n\nQuate, C. F., Microwave Tubes\u2014What is New?, I.R.E. \nSubsection, White Plains, New York.\n\nRowen, J. H., Fundamentals of Ferromagnetism, Edison Re\u00ac \nsearch Laboratories, West Orange, N. J.\n\nSlichtcr, W. P., Characteristics of Rubber-Like Materials, \nNew York Rubber Group, Course on Advanced Elastomer \nTechnology, N. Y. C.\n\nThatcher, W. II., NIKE \u2014 A Guided Missile System for \nAnti-Aircraft Defense , College Club, Linden, N. J.\n\nVacca, G. N., Comparison of Accelerated and Natural Tests \nfor Ozone Resistance of Elastomers, Am. Soc. for Testing \nMaterials, Symposium on Ozone, St. Louis, Mo.\n\nWinslow, F. PI., The Thermal Oxidation of Hydrocarbon \nPolymers, Chemistry Seminar. Rutgers University, New \nBrunswick, N. }.\n\nWinslow, F. IP, New Protectants for Hydrocarbon Poly\u00ac \nmers, Pittsburgh Section, Am. Chemical Soc., Mellon In\u00ac \nstitute, Pittsburgh, Pa.\n\nWood, E. A., Opportunities for Women in Industrial Re\u00ac \nsearch, Douglass College, Rutgers University Career \nForum, New Brunswick, N. J.\n\nWooley, M. C., Components for Submarine Telephone Cable \nRepeaters, Dayton Section I.R.E., Professional Group on \nComponents, Dayton, Ohio.\n\nFollowing is a list of the authors, titles and places of publication \nof recent papers published bv members of the Laboratories:\n\nEisinger, J., Electrical Properties of Nitrogen Adsorbed on \nTungsten, J. Chem. Phys., 28, pp. 165-166, Jan., 1958.\n\nFerrell, E. B., Control Charts for Log Normal Universe, 1958 \nMiddle Atlantic Conference Transactions, pp. 137-142, \nFeb. 28-March 1, 1958.\n\nFletcher, R. C., Physical Phenomena of the Solid State, Proc. \nof the Symposium on the Role of Solid State Phenomena \nin Electric Circuits, 7, pp. 41-62, 1957.\n\nLaw, J. T., The Interaction of Oxygen ivith Clean Silicon \nSurfaces, J, Phvs. and Chemistry of Solids, 4, p. 91, 1958.\n\nMason, W. 1\\, Ferroelectrics, Proc. of Symposium on the \nRole of Solid State Phenomena in Electric Circuits, 7, \npp. 91-108, Feb. 27, 1958.\n\nMayzner, M. S., and Tresselt, M. E. (NYU), Shifts in Con- \nnotative Meaning of Words as a Function of Previous Re\u00ac \nstrictive Experience, J. of Experimental Psychology, 55, \npp. 200-205, 1958.\n\nMcAfee, K. B., |r., Stress Enhanced Diffusion in Glass \u2014 I. \nGlass Under Tension and Compression \u2014 II. Glass Under \nShear, J. Chem. Phys., 28, pp. 218-229.\n\nMcDavitt, M. B., 6000 Mega cycles/Sec. Radio Relay System \nfor Broad-Band, Long Haul Service in the Bell System , \nCommunication and Electronics, 34, pp. 715-722, Jan., \n1958.\n\nWaite, T. R., A General Theory of Bimolecular Reaction \nRates in Solids and Liquids, J. Chem. Phvs.. 28, pp. 103- \n106, Jan., 1958.\n\nWernick, J. II., Geller, S., and Benson, K. E., X-Ray Powder \nData for Synthetic Proustite , Ag AsSn, Analytical Chemis\u00ac \ntry, 30, p. 303, Feb., 1958,", "title": "Bell Laboratories Record 1958 04 1", "trim_reasons": [], "year": 1958}
{"archive_ref": "sim_record-at-t-bell-laboratories_1944-07_22_11", "canonical_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1944-07_22_11", "char_count": 111022, "collection": "archive-org-bell-labs", "doc_id": 807, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc807", "record_count": 136, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1944-07_22_11", "split": "test", "text": "Culver City, at the western border of \nLos Angeles, is shown in white on the relief \nmap at the head of this article. As in the\n\na \n1 4 \ngeles area, its switch- 2 95260 \nAR 3 | 2345 65 go 2 \ning equipment is of AS ro \ncentral-office arrange- DISCONNECT 9 REMARKS *\u00ab278-2345-| 2071 45-03! -262-4321--23-06+\u00ab \nments are shown by Z| 36 \nCONNECT P\n\nELAPSED TIME \ndiagrams of Figure 1. \nWith both methods, \nthe calls are routed \nthrough step-by-step ATTEMPTS\n\nFig. 2\u2014One form of manual ticket at left, and the new mechanically \nprepared ticket at the right which gives all the necessary information \nas a series of numbers in a single line\n\ntimes of beginning and end of \nthe connection in hours, min- \nutes, and tenths of minutes. \nWhen these tickets reach the \naccounting department, \nelapsed time may be obtained \nby subtracting the CONNECT\n\ntime from the DISCONNECT \ntime, and the proper charges \nare written in. The automatic \nticketer, on the other hand, \npresents all necessary informa- \ntion as a series of numbers in a \nsingle line as shown at the \nright of Figure 2. What these \nvarious numbers represent is \nindicated above the ticket. \nSeven digit spaces are avail- \nable for the month, day, and \nhour: two for the month, two \nfor the day, and three for the \nhour, which is given in hours \nand tenths of hours. The par- \nticular call shown was made on \nDecember 7 at 14.5 hours, \nwhich, since the time is reck- \noned from midnight, repre- \nsents 2:30 P.M. to the nearest \nsix minutes. The two digits \nfollowing the called number, \nalthough printed together, rep- \nresent different things. The \nfirst digit indicates the calling \nsubscriber\u2019s class of service, \nand the second, the number of \nmessage units chargeable for \nthe initial interval, which may \nbe three or five minutes. The \ninformation conveyed by this \nsecond digit is used for deter- \nmining the proper charge to \nmake for the total elapsed \ntime, which is shown in whole \nminutes by the last two digits.\n\nSome of the apparatus re- \nquired to secure the informa- \ntion printed is associated with the call until \ndisconnect occurs, but other apparatus is \nrequired only for short intervals of various \nlengths. As in crossbar offices, certain units \nof common equipment are provided which \nmay be used with any of the trunks, and are \nconnected to the ticketing circuit proper \nonly for brief intervals as needed. These \nvarious units and their lines of association \nare indicated in Figure 4.\n\nSenders control the progress of the call, \nand in addition to recording and transmitting \nthe desired number, they also control the\n\nprinting of most of the information on the \nticket. As soon as the ticketing circuit is \nseized, a sender is connected to it, and begins \nrecording the digits dialed by the subscriber. \nSince at least part of the office code has al- \nready been dialed by this time, and since the \nsender must also know the number of the \ncalling line to properly control the printing \nof the ticket, it, in turn, associates itself with \nan IDENTIFIER to get the information. The \nIDENTIFIER determines the office code by \nfinding the level reached by the selector \npreceding the ticketing trunk. It determines \nthe calling subscriber\u2019s number by connect- \ning to the THOUSAND NUMBER Circuits, of \nwhich there is one for each thousand num- \nbers in the local office. With this information \navailable, the sender connects to the mes- \nsage ticketer, and at once starts printing the \nticket. At the proper time, it calls in a pay \nAND HOUR Circuit to secure the month, day, \nand hour. The 1pENTIFIER has also given the \nsender the class of service and the message \nunits for the initial interval, and thus the \nsender now has all needed information for \nprinting the ticket except the elapsed time.\n\nAs soon as the sender gets the called- \noffice code from the IDENTIFIER it begins\n\ntransmitting it over the trunk, and then \nsends the subscriber\u2019s number, which it has \nrecorded in the meantime. The IDENTIFIER \nand DAY AND HOUR Circuits are released as \nsoon as the necessary information is ob- \ntained from them, and as soon as the full \nnumber of the called subscriber has been \ntransmitted and the ticket printed, the \nsender disconnects itself. Apparatus forming \npart of the ticketing circuit starts to time the \ncall as soon as the called subscriber answers, \nand at disconnect, it automatically prints \nthe chargeable time.\n\nThe finished ticket is of the same size as \nthose used at toll boards, but the paper for it \nis supplied to the ticketing machine in a roll. \nFrom this roll the paper is passed through \ndrive rollers, thence between the lower rim \nof the type wheel and a printing hammer, \nand then through guides to the cutting knife \nand the hopper. The general relationship of \nthe major units of the ticketer is shown 1n \nFigure 5, and front and rear views of the \nticketer itself, in Figure 6. A type wheel, \ncarrying twelve characters\u2014the ten digits \nfrom zero to nine, a dash, and an asterisk\u2014 \nis mounted on one end of a shaft which also \ncarries an index wheel, a friction clutch\n\nthrough which a small motor drives it, and a \npair of interconnected brushes\u2014one riding \non a contact ring, and the other over a set \nof twelve stationary segments. Each com- \nmutator segment corresponds to one charac- \nter on the type wheel, and when the brush \njs on any particular segment, the character \ncorresponding to that segment on the type \nwheel is in position for printing. Voltage ap- \nplied to a segment by the control circuit \nenergizes the print magnet which lifts the \nprint hammer to force the paper against the \ncharacter. With the same motion a metal\n\nfinger called the sTaBBER is pressed between \ntwo teeth on the index wheel to hold the type \nwheel stationary while the impression of the \ncharacter is being made. As the magnet re- \nleases, a finger connected to the armature \nrotates a ratchet on the drive roll to step the \npaper ahead.\n\nThe slipping clutch permits continuous \nrunning of the motor in spite of the inter- \nmittent rotation of the type wheel. Contacts \noperated by the armature of the print mag- \nnet indicate when the magnet is fully oper- \nated, so that the voltage can be removed\n\nand also when the magnet is fully released, \nso that the printing of the next character can \nbe started. The type wheel rotates at about \n375 rpm and printing is at the rate of about \nseven characters per second.\n\nAutomatic ticketing was selected for the \nLos Angeles network in preference to an- \nother direct-dialing method known as zone \nregistration, which has been used to some \nextent in other dial systems. Under the latter \nplan the subscriber message register is oper- \nated more than once on answer of the called \nsubscriber in accordance with the rate to the \ncalled office, and is operated further for each \novertime interval. Several factors favor \nticketing for the Los Angeles area. The per- \ncentage of subscribers having message-rate \nservice is relatively low, particularly in the \nsuburbs, and the registers which it would \nbe necessary to buy for flat rate subscribers \nwould be used only on the zone calls. Certain \ncircuit difficulties would have to be overcome \nto make it possible to operate the registers in \nthe existing step-by-step offices more than \nonce on the same call. Finally, the individual \nrecord provided for each call is of value in \nanswering inquiries regarding bills, and this \nadvantage increases in importance as the\n\nTue Autuor: O. A. Frienp, after graduating \nfrom the University of Michigan in 1910 with \nthe degree of B.E.E., \njoined the A T & T \nand was assigned to \ndevelopment of power \nequipment. Entering \nthe Army early in \nWorld War I, he \nserved in the Re- \nsearch and Inspection \nDivision of the Signal \nCorps in France. He \nreturned tothe D&R \nat the time the ma- \nchine switching pro- \ngram of the Bell System was getting under way, \nand became associated with the development of \nthe step-by-step type of dial equipment. When \nthe D & R combined with the Laboratories in \n1934, he became Step-by-Step Engineer in what \nis now the Switching Engineering Department, \nand has been engaged in further development of \nthe step-by-step system.\n\nO* July 16, 1920, an earthquake shook \nLos Angeles, and rumors spread that \nthe \u2018Magic Isle\u201d of Catalina had sunk into \nthe sea. Earlier on this same day, Catalina \nhad been linked to the mainland by tele- \nphone, and thus to dispel this rumor, it \nwas necessary only to .\n\nthe town of Avalon on \nthe island. This new \ntelephone link was no \nordinary service ex- \ntension. The thirty- \nmile stretch of ocean \nbetween Avalon and \nLong Beach was \nbridged by radio; for \nthe first time a radio \nlink was made an in- \ntegral part of a tele- \nphone system. Al- \nthough the Arlington- \nParis experiment had \ndemonstrated the pos- \nsibility of long dis- \ntance radio telephony nearly five years \nearlier, this was the first time reliance had \nbeen placed on radio to give regular day-in \nday-out public telephone service.\n\nWhen the need arose for telephone service \nto Catalina, cable communication could not \nhave been provided without considerable \ndelay, and radio was turned to as an al- \nternative. This was shortly after World War \nI, and the development of a line of standard \nradio equipment had been undertaken by the \nEngineering Department of the Western \nElectric Company as a post-war project. Al- \nthough not ready for manufacture, designs \nwere sufficiently advanced to permit mak- \ning the equipment in the laboratory.\n\nIn establishing this link, telephone and \nradio technicians joined hands to do a job \nthat had never been done before\u2014the \nbridging by radio of a gap between sections \nof wire telephone plant to provide public \ntelephone service. The circuit established,\n\nbetween the Los Angeles toll board and the \nAvalon switchboard, was made up of twenty- \nfive miles of wire circuit on the mainland, \nthirty-two miles of radio link, and one mile \nof wire line on the island. It was handled \njust as any wire toll circuit, and could be \n_ switched at the Los \na Angeles end to connect \nwith circuits to all\n\nThe radio stations \nwere at Long Beach, \non the Pacific shore \nsouth of Los Angeles, \nand at Pebbly Beach, \njust south of Avalon \nnear the southern end \nof Catalina. The trans- \nmitters had a carrier \noutput of about 100 \nwatts, and two-way \ncommunication was \nobtained by using two \nfrequencies \u2014 638 kc \nfrom California to \nCatalina, and 750 ke \nin the opposite direction. These channels \nare in what is now the most desirable region \nof the broadcast band, but at that time \nbroadcasting was still in the future.\n\nThe Catalina radio circuit was popular \nfrom the first, and carried such a large \namount of traffic that two operators were \nnecessary at each end to handle it. So \npopular was it, as a matter of fact, that \nwithin a year or two, additional facilities \nwere urgently required. Since the trend of \ntraffic indicated that many more channels \nwould shortly be needed, submarine cables \nwere decided upon, and the radio circuit was \nremoved about three years after its opening. \nIt had conclusively demonstrated that radio \ntelephone links were eminently practicable, \nand it indicated the most promising direc- \ntions for further developments to take. It \nwas the forerunner of the present radio tele- \nphone services of the Bell System, which prior \nto the war reached some seventy countries.\n\ncate operation because the desired \ndimensions must be reached to within a \nvery few millionths of an inch. The crystal is \nbrought to approximately the required \nthickness in a lapping machine of the general \ntype already described in the Recorp.* \nThe final finishing laps are made by hand, \nhowever, since the process is short, and the \ncrystal must be tested for its proper fre- \nquency at frequent intervals. This final lap- \nping was formerly accomplished by placing \nthe crystal on a lap like those used in the \nmachine, and moving it around with the \nfinger. This method, however, was subject \nto all the irregularities of the earlier machine \nmethods. Following the design of the new \nmachine lapper, therefore, a new hand lap- \nping outfit was designed and a new tech- \n~ *REcORD, June, 1944, DP. 435-\n\nnique outlined that gave much improved \nresults, and has greatly expedited the finish- \ning process in the Western Electric plants.\n\nThe objectives of the design were to secure \nan even pressure over the entire face of the \ncrystal, to allow the crystal to rotate as it \npassed over the lap, to have the crystal over- \nride both inner and outer peripheries of the \nlaps at each passage across it so as to avoid \nthe tendency to excessive grinding on the \nouter sections of the crystal, and to secure \nessentially even wear over the entire surface \nof the lap so that it would remain flat. None \nof these objectives were very satisfactorily \nattained with the earlier hand method. It \nwas almost impossible to secure uniform \npressure by moving the crystal with the \nfinger, and rotation of the crystal was \nnegligible. In addition, over-running of the \nlap was very irregularly if at all obtained, \nand the tendency to excessive grinding over\n\nTo secure evenly distributed pressure on \nthe surface of the crystal, a glass carrier was \ndesigned. This is shown at the right of \nFigure 1. One surface of this carrier is ground \noptically flat, and the other has a small in- \ndentation at the center. In operation, a \nwetted crystal is slid onto the flat surface to \nwhich it adheres, and is then placed on the \nlap with the carrier on top of it, and the \ncrystal moved over the surface by placing \nthe point of the stylus, shown at the left of \nFigure 1, in the socket in the top of the re- \ntainer. Pressure applied at this point is dis- \ntributed evenly over the surface of the lap in \ncontact with the crystal. Moreover, as the \nretainer is pushed around by the stylus, the \neffect of the friction is to rotate the crystal \nand the retainer, thus giving rotational mo- \ntion in addition to that of translation.\n\nTo secure regulated over-shooting of the \ncrystal at both inner and outer peripheries of \nthe lap, a lap holder was designed that would \npermit the crystal retainer to move off the \nlap a fixed amount at both its edges. This\n\nTue AutHor: G. M. THurston of the Com- \nmercial Products Development Department en- \ntered the student \ncourse of the Engi- \nneering Department \nof the Western Elec- \ntric Company in \n191g. He then joined \nthe radio research \ngroup where he was \nconcerned with cir- \ncuit development. He \nWas associated with \nthe development of \ntransatlantic radio \nand in this connec- \ntion made a field survey of the transatlantic \nshort-wave project. He was also closely asso- \nciated with the development of ship-to-shore \nradio, particularly the initial installation on the \nLeviathan.\n\nFrom 1922 to 1926 he studied at Columbia Uni- \nversity and in 1927 at the Polytechnic Institute \nof Brooklyn. Since 1930 most of Mr. Thurston\u2019s \nwork has been concerned with investigations of \nquartz crystals. This has covered dimensioning \nand lapping techniques and the development \nof testing equipment.\n\nholder, shown in Figure 1, has a low round \nprojection in the center that fits inside the \nlap and holds it securely in place. Arising \nfrom the center of this projection is a cylin- \nder that is smaller than the central opening \nof the lap by just the amount of over-carry \ndesired. Around the outer periphery of the \nlap, and the same distance from it as the \ncentral cylinder is from the inner periphery, \nis a ring to limit the extreme outer position \nof the crystal retainer. The operating tech- \nnique is to move the crystal in looping arcs \nbetween the outer and inner stops, and at \nthe same time to carry it around the lap. \nThirty small loops for one round trip of the \nlap are specified.\n\nAs may be discovered by reference to the \narticle already referred to, this motion is \nalmost exactly the same as that provided in \nthe new precision machine lapper. In making \nthe finishing laps, however, hand lapping 1s \nmuch quicker because the frequent re- \nmovals of the crystal from the lap to test for \nfrequency, while quite simple with the hand \noutfit, become complicated and slow with \nthe machine. This new device and technique \ngives the speed of hand lapping and yet re- \ntains precision of the best machine lapping. \nBecause of this advantage the time required \nto train girls for this hand lapping process \nhas been drastically reduced.\n\nmatically routed to a cordless switch- \nboard where operators set up the connec- \ntion through the switching train by \u201cwriting \nup\u201d the proper code on keys. In any such \nsystem, the general problem is to arrange \nthe distributing circuits so that an idle \noperator can be promptly found and con- \nnected to the trunk regardless of the number \nof positions occupied at the time. On the \nother hand, the number of positions pro- \nvided for any particular office and the \nnumber of them occupied for any existing \nload must be kept as small as practicable in \norder to avoid excessive operating and \nequipment costs.\n\nAnalysis shows that operating efficiency \nincreases with the size of the group of \noperators, and thus the greatest operating \nefficiency is secured if all the trunks can be \ngiven access to all the positions. The amount \nof equipment necessary to give this uni- \nversal interconnection becomes excessive, \nhowever, and economy frequently makes it \nnecessary to divide the operators into \ngroups, each group serving a portion of the\n\nincoming trunks. Although maximum eff- \nciency is not secured when this is done, it \ncan be approached fairly closely since the \nincrease in efficiency tapers off as the groups \nbecome larger, while the cost of the equip- \nment increases.\n\nIn the past, call distribution has been ac- \ncomplished by using step-by-step, panel, or \nrotary switches, and the results have been \nvery satisfactory. None of these methods \ncould be readily adapted to the new toll \noffice, however, because of the large number \nof leads per trunk that must be handled. \nBecause of the exigencies of toll switching, \na simple tap connection to the trunks 1s not \nadequate; the incoming trunk for a brief \ninterval of the switching period must be \ncarried to the switchboard and then back to \nthe toll switching train, thus forming a loop. \nBecause of the use of four-wire switching, \nmoreover, and of other factors of the crossbar \ntoll system, these operator loops each re- \nquire sixteen leads, and this large number of \nleads cannot easily be obtained with any of \nthe types of switches formerly used. In the \ncrossbar toll office, therefore, crossbar link \nframes are used to distribute the calls to the\n\nthe link frames, and thus have access to all \nforty loops. Because of the splitting of the \nframe into two halves, only one half is \nneeded to illustrate the arrangement of \ntrunks, links, and loops. This arrangement \nis shown in Figure 1, where the six secondary \nswitches of the half shown are at the right, \nand the upper and lower of the six primary \nswitches are at the left. The lower three \nsecondary switches supply the sixteen con- \nductors for each of ten loops, and the three \nupper, for another ten loops, and thus each \nhalf frame supplies twenty loops.\n\nEach primary switch is split so as to form \nfive units, each with ten horizontals and two \nverticals, and thus the six primary switches \nof each half frame provide thirty such two- \nby-ten units. Again because of the sixteen \nleads required, three of the two-by-ten \nunits are used for each loop, and thus the \nthirty units provide for one hundred trunks \n\u2014ten for each group of three two-by-ten \nunits. The three two-by-five units for each \nset of ten trunks may not all be on the same \nswitch; two of them may be on one and one \non the switch above or below it.\n\nIn Figure 1, each line represents six wires, \nthe number that can be handled by each \ncrosspoint of the crossbar switches. If one \nline is allowed to represent all sixteen wires, \nthe interconnections for a complete frame \nwould be as indicated in Figure 2. Here, one \nten-by-ten secondary switch represents three \non the actual frames, and each two-by-ten\n\nyan TO \n-- OPERATOR \nNO. 99 at -- LOOP NO.I9 \n9 9 9 \u201c16 wiRES \nrss \nOo! 23 89 9 Io 9 9 \n| SWITCHES T \nUPLICATE 4 PS4 =16 WIRES \nITCHES SHOWN \n9 9 9 \n67 89 9} oO 9] oO 9 \n- \n| -- OPERATOR \nLOOP \n> \n16 WIRES 16 WIRES\n\nFig. 1\u2014One-half of operator link frame. Each line represents six wires, the number that can be \nhandled by each crosspoint of the crossbar switches\n\nPRIMARY SECONDARY three frames, all the No. 1 \n=\" loops to another and all the \nTAHT = No. 2 loops to the third. These \ni -0 w three frames are called key- \n\u00a9 frames, and are numbered o, \n' % 1, and 2 to correspond to the \nnumbers of the loops they \nmans serve. Since forty operators \nS can handle calls from more \nthan three hundred trunks, \n3 other frames are added as \n. . \nneeded. The horizontal termi- \nGROUP ATT \u2014 nals of the secondary switches \nTRUNKS @ of each added frame are multi. \npled to one of the key-frames, \n= but each of the additional \nframes serves a different group \nTEN 74 of one hundred trunks. \nTRUNKS T \nThe use of key-frames per-\n\npositions. The three loops to \neach position are marked 0, 1, \nand 2, and all the No. o loops \nare connected to one of the\n\nFig. 3\u2014Temporary multipling arrangements using the key- \nframe system are shown by the dashed lines. This multipling \nis removed when additional loops are required\n\nframes because each additional frame is per- \nmanently multipled to one of the key- \nframes. This temporary multipling is indi- \ncated by the dashed lines on Figure 3 which \nshow the general scheme of key-frames. \nShould these unoccupied positions later be \nfilled, it is necessary only to remove the \ntemporary multipling, and run the addi- \ntional loops to their assigned positions. \nForty operator positions will probably be \nsufficient for most of the toll crossbar instal- \nlations. Some of the very large cities, how- \never, may require more than forty positions, \nand to provide for them, a second arrange- \nment has been devised that will handle a \nmaximum of eighty positions. This is the ar- \nrangement provided at Philadelphia, for \nalthough there are only thirty-two positions \nat the present time, it is expected that the \nnumber will soon be greater than forty. \nThe general scheme of arrangement for the \neighty-position unit is similar to that for the \nforty, but six key-frames are used instead of \nthree. These key-frames are divided into \nthree odd and three even frames. The \nformer handle the odd switchboard positions \nand the latter, the even. Thus, the forty \nodd positions of the board\u2014positions 1, 3, 5,\n\nthrough 77, 79\u2014will be served by the three \nkey-frames Nos. 1, 3, and 5, while the forty \neven positions\u20142, 4, 6 through 78, 80\u2014will \nbe served by the three key-frames Nos. 0, 2, \nand 4. In this way, a doubling in size is pos- \nsible without any radical difference in the \ngeneral arrangement.\n\nTHe Autuor: F. A. Parsons received the \nE.E. degree from Cornell University in 1937, \nand at once joined \nthe Technical Staff of \nthe Laboratories. \nAfter spending some \ntime on relay design, \nand in the toll labora- \ntories group, he trans- \nferred to toll facil- \nities. Here he worked \non the requirements \nfor crossbar toll \nswitching systems, \nand was intimately \nassociated with the \ndesign of the No. 4 system recently installed in \nPhiladelphia. In February, 1941, Mr. Parsons \nleft the Laboratories on a military leave of \nabsence, and is now a major with the Bomb \nDisposal Group of the United States Army.\n\nFor the Fifth War Loan Drive now under way and which closes \nJuly 8, the U. S. Treasury suggested as a goal an investment of $100 \nin bonds per employee during the months of June and July. The \nLaboratories, however, through the decision of its Bond Committee, set \nitself a higher goal. It had been struggling to maintain a payroll \nallotment of 10 per cent of total pay including overtime. It had exceeded \nthat during the Fourth Drive, through the three months\u2019 allotment. \nWhen these were completed its percentage dropped to about 8% per cent \nfrom over 10% per cent. For the Fifth Drive it set a goal of 10 per cent \nplus an extra week\u2019s standard pay\u2014not including overtime\u2014dis- \ntributed through twelve weeks during July, August and September. \nThis 1s a hard goal to attain but it is easy compared to the goals the \nboys are gaining in France, Italy and the Pacific. Can\u2019t we do that \nlittle extra? The Laboratories Bond Committee knows we can; and the \nstatistics as of June 23 show that we are three-quarters of the way to \nthe goal. Give a good hard push and the goal is reached.\n\nU.S. Army\u2019s latest anti-aircraft gun\u2014the 4.7 inch, equipped with the M-10 Electrical Director designed \nby the Laboratories. Upper left, the projectile ready for its fuse to be set. Upper right, the fuse Setter \nslides down, rotates the fuse to the proper position, and returns. The tray swings down into line with \nthe breech; the lever at the left rams the powder charge and projectile and the breech closes. Center, gun \ndrill at the Weapons-of-War show at Washington. Lower left, Tracker for the M-1o Director. Lower \nright, computer and altitude converter. Normally a tarpaulin over five bows roofs the trailer\n\neth anniversary of Walter S. \nGifford\u2019s service in the Bell Sys- \ntem. As a young man about to \ngraduate from Harvard in the \nyear 1904, Mr. Gifford wrote a \nletter asking for a job with the \nGeneral Electric Company. The \nbusiness world and its connections \nwere new to him and he made the \nmistake of addressing the enve- \nlope to the Chicago address of the \nWestern Electric Company. In his \nletter of application he said, among \nother things, \u201c\u201cNow if there is only \nsome position (no matter what, \nprovided there will be a fair chance \nfor a rise if it is deserved), in the \nGeneral Electric Company, I \nshould like to try it and would \nendeavor to suit.\u201d This letter, \nthough misdirected, won Mr. Gifford an \ninterview and he was invited to file an appli- \ncation for a position with the Western Elec- \ntric Company.\n\nIn the application form Mr. Gifford stated \nthat his father was a lumber dealer in Salem, \nMassachusetts, and in reply to the question \nas to what work he would prefer and what \nspecial training he had fitting him for it, \nhe said, \u2018\u2018Good office work\u2014no special train- \ning, however.\u201d As references he gave two of \nhis professors at Harvard, the principal of \nhis high school, and the family doctor. The \nletters of recommendation written by these \nmen were all favorable, and one concluded \nwith this sentence: \u201cI judge him to be a \nman of promise.\u201d It was pointed out that \nhe had entered Harvard when but sixteen \nyears of age and had completed the four-year \ncourse in only three years, thus graduating \nat the unusually early age of nineteen. The \nPrincipal of his high school said: \u201cI can \nrecommend him in the strongest terms to\n\nanyone desiring a reliable, honest, and hard- \nworking young man.\u201d These recommenda- \ntions, made by men who were without in- \nfluence in the worldly sense, proved to be \nprophetic. Mr. Gifford\u2019s rise in the industry \nof his own choosing was as rapid and as \nbrilliant as \u2018his scholastic career had been.\n\nIn July, 1904, he started with the West- \nern Electric Company in Chicago as a clerk \nin the payroll department at a salary of $10 \na week and two years later he came to the \nNew York headquarters of that company as \na clerk in the Secretary\u2019s office at $24 a \nweek. Five years after entering the business \nhe transferred to the American Telephone \nand \u2018Telegraph Company in Boston and did \ngeneral accounting work at a salary of $275 \nper month. His rise in the Company to the \npositions of Chief Statistician, Comptroller, \nVice-President, and finally to that of Presi- \ndent, bears testimony to his industry, in- \ntegrity and ability.\n\nknown. Neither in his employment nor in \nhis subsequent rise did family influence or \nthe influence of so-called \u201c\u2018powerful inter- \nests\u201d? play any part in his development. By \nthe application of his outstanding talents to \nthe tasks which have confronted him, he has \nreached his present position of leadership in \nthe business world, and by his unfailing \ncourtesy and kindness he has won a warm \nplace in the hearts of all who know him.\n\nTelephone work is not normally haz- \nardous, but emergencies do arise, and hardly \na day passes without a striking example of \ncourage, resourcefulness, or devotion to duty \non the part of a telephone worker some- \nwhere. Those telephone people who distin- \nguish themselves through outstanding \u201c\u2018acts \nof noteworthy public service\u201d receive awards \nthat are provided by the Theodore N. Vail \nMemorial\n\nKor 1943, twenty-six awards were made to \ntelephone men and women\u2014and six of these \nawards were for especially conspicuous \nmerit. The citations accompanying the \nhigher awards were:\n\n\u201cTo Frederick Richard Hoffman, of The \nChesapeake and Potomac Telephone Com- \npany of Virginia, for initiative and prompt, \nintelligent and courageous action in the \nrescue of a man who had come in contact \nwith a high voltage wire, a silver medal and \nfive hundred dollars.\u201d\n\n\u201cTo Louis G. de Lyon, Alfred H. Gerlach, \nAlexander Mikolasy, William Mohrhoff, \nand Louis J. Rom, all of the Western Elec- \ntric Company of New York, for prompt \naction, initiative and extreme courage in \nthe face of grave personal danger during an \nemergency prior to and following an ex- \nplosion of hydrogen gas, each a silver medal \nand five hundred dollars.\u2019 Of this group, \nMr. de Lyon gave his life, and the award is \npresented posthumously to his wife, Mrs. \nElfrida J. de Lyon.\n\nIn addition, a group of Southwestern Bell \nTelephone Company employees has received \nspecial recognition. They remained on duty \nunder extremely trying conditions after an \nexplosion in the Jefferson Central Office \nbuilding at St. Louis where they worked had \ndemolished walls, shattered glass, and filled \nthe building with dust, smoke and fumes.\n\nDuring a recent visit at Murray Hill, Mr. W. B. Bell, President of the American Cyanamid \nCompany, and Dr. R. B. Barnes of his staff were accompanied by Dr. Buckley and Mr. C. P. \nCooper, Vice-President of the American Telephone and Telegraph Company. In the photograph, \nMr. Bell stands at Dr. Buckley\u2019s right and Mr. Cooper is at his left. In the back row, from left \nto right, are R. R. Williams, R. M. Burns, R. A. Haislip, W. H. Martin and Dr. Barnes\n\nLater he went to England during the \n\u201cBlitz\u201d to study at first hand the work of \nthe Bomb Disposal Group. He returned then\n\nSine to the Aberdeen Proving Ground to_be- \ncome director of research for this group \nthere. Major Parsons, for having taken \napart an unexploded bomb in England, \nwears the special Bomb-Disposal insignia\n\non his sleeve\u2014a red bomb outlined in yellow \nLieut. Commander Clarence Unnevehr on a dead-black background.\n\nCoMMANDER UNNEVEHR spent a day at All Bomb Disposal men are volunteers, \nWest Street recently while on leave. This though only officers do the actual nerve- \nwas his first return to the States in almost tingling job of taking the \u201ccranky\u201d things \ntwo years. He left the Laboratories in No- apart. A colonel in this group remarked, \nvember, 1940, and has since seen a great deal \u2018In our branch of the service you only make \nof action in the South Pacific. He partici- one mistake. You\u2019re either an expert or \npated in the battle at Rennell Island which \u2014 you\u2019re dead.\u201d \nwas a part of the Guadalcanal campaign. ~\n\nLater he also saw action in the New Georgia Lieut. Col. W. J. Galbraith \nand Bougainville campaigns. He expects to \u201cJust a line to the gang to let you know \nreturn to the Pacific area. that I can tell you I am in India, in the CBI\n\n(China-Burma-India) Theater. Had a very \nMajor Frank A. Parsons long but interesting trip and saw quite a\n\nOne of the most interesting, and probably bit of the \u2018world on the way. India is an \nthe smallest branch of the United States interesting country with its native bearers \nArmy, is the Bomb Disposal Group. Most carrying immense loads on their heads, \nof the men in this group have an engineering everyone chewing betel-nut. I suppose this \nbackground. Mayor Frank A. Parsons, a js the hottest part of the year and soon the \nmember of the Laboratories, is now in Eng- monsoon season starts, which is just about as\n\nWhen Major Parsons first came to the \nLaboratories he spent some time in the relay Lieut. Col. J. M. Hayward \ndesign and the toll laboratories groups be- \u201cThe trip down here to Australia was un-\n\nfore transferring to the Toll Facilities De- believably fast considering the distance. | \npartment where he worked on requirements Once we took off from the West Coast we \nfor toll crossbar switching systems. When were in the air almost constantly by night i \nMr. Parsons left the Laboratories for the or day. The time spent on the ground during ( \nArmy, he was sent to the Aberdeen Proving \u2014 the four stops we made was very short, with \nGround in Maryland where he was in the exception of the stop at Guadalcanal ie \ncharge of instruction on a pistol range. where we had a little more time to clean up, =\n\nWe regret that in last month\u2019s issue a photo- \ngraph of H. V. Berlin with Lieut. George \nBukur\u2019s name under it was published. Both \nare overseas, Private Berlin somewhere in the \nPacific and Lieut. Bukur in China\n\nshave and eat at a table. Without any scen- \nery to look at other than water and clouds, we \npassed the time reading, sleeping, eating or \nplaying gin-rummy. The fall weather here \nis very mild and clear. The hotel accommo- \ndations in the one I\u2019m in\u2014reserved for \ncolonels\u2014are the kind my _ grandfather \nmight have enthused about and the food just \nis not prepared properly. I doubt if I will \nbe here long enough to get used to walking \nand riding on the left side of the street, and \nthe money, in pounds\u2014shillings\u2014pence\u2014 \nwith the florin (2S) and ha\u2019penny too, are \nstill confusing me. Have yet to see a \nkangaroo or a fuzzy-wuzzy, but soldiers and \nwomen in uniform rule the town at all \ntimes. Best regards to all.\u201d\n\nLizur. GeorGeE Bukur writes from some- \nwhere in China, \u201cI finally arrived safely \nand in good shape. I\u2019m sure I\u2019ll never want \nto travel again when I get home! I used \nabout every mode of transportation known \nbefore finally reaching my destination.\n\n\u201cChina is a very interesting country and \nmuch better than India as far as food and \nclimate are concerned. | am at an American \ncamp where the food and quarters are fair. \nMy knowledge of Mandarin helps a great \ndeal and I am now an instructor in the \nlanguage. The surrounding terrain is indeed \nbeautiful. My only regret is that I have no \nphotographic equipment to record some of \nthe unusual things I have seen.\n\n\u201cChinese ten dollar bills are worth eight \ncents in our money. There is very little to \nbuy here in the way of souvenirs, \u2018and what \nthere 1 is Costs outrageous prices.\n\n\u201cOur daily program is a long one, with \nbreakfast at 5:30 a.m. and supper at 6:00 \np-m., so don\u2019t gripe about overtime in the \noffice! If you think a uniform is glamorous, \nyou should see it when one is standing in \nthree feet of mud. You haven\u2019t lived until \nyou've eaten water buffalo. It would taste \nlike leather in the States but it\u2019s heaven \nover here.\u201d\n\nBertrand H. Sommer \nCapet-MIpsHIPMAN BERTRAND H. Som- \nMER of the Merchant Marine returned to \nvisit West Street recently, after two seven- \nmonth cruises in the Mediterranean.\n\noratories have been on military leaves of \nabsence (see page 463), of which 33 leaves \nhave been completed. The 813 active leaves \nwere divided as follows:\n\nThere were also 17 members on merchant \nmarine leaves and 27 members on personal \nleaves for war work.\n\nRobert Beattie James F. Madden \nJohn J. Cozine Henry G. Petzinger \nBetty J. Dean Joan E. Schubert \nAlbert H. Diegler Frederick A. Soltow \nEdward J. Dixon Albert R. Strnad \nHoward S. Hopkins William A. Sumner\n\nDonald E. Blesse Ensign F. W. Lindberg \nEugene J. Breiding Domenick Maccia \nEdward A. Hake Peder M. Ness \nHerbert E. Henrikson T. T. O\u2019Shaughnessy \nPaul A. Hopf Seferin E. Pulis \nWilliam E. Howard William L. Rohr \nJoseph Kocan Ellsworth R. Rosen \nEdward B. Kopetz Herbert L. Smith \nDavid J. Van Slooten\n\nDuring his trips he stopped at a number of \nports in Italy, Sicily, and North Africa. The \nsurprise German raid on the Allied port of \nBari, Italy, last December took place just \nthe day before Sommer arrived there. His \nwas one of the first ships to enter the harbor \nafter the bombing.\n\nHe spent Christmas Day in Bari, where \nhe received one of his copies of the Recorp. \nNew Year\u2019s Day he spent in Augusta, \nSicily, where he found that the city had \nbeen bombed by the Germans during a raid \nthe previous day.\n\nThe cruises were a part of his preliminary \ntraining. Cadet Sommer will now take \nfurther training at the Merchant Marine \nAcademy in Long Island. When he has com- \npleted the course he is taking there, he will \nbe commissioned a Third Officer in the \nMerchant Marine or an Ensign in the \nNaval Reserve.\n\nwrites from the South Pacific, \n\u201cHave moved about on this hot and humid \njungle island quite a bit\u2014done some flying \nover tremendous mountain ranges and blue- \ngreen seas. It is a wild and unhospitable \ncountry, largely unexplored, with a most un-\n\npleasant climate. I can\u2019t tell you where I \nam now or have been, of course, but the \njungle everywhere presses in closely. Few \nwild animals are to be seen\u2014just some wild \npigs, wallabies (miniature kangaroos), and \nlots of snakes. We shot several snakes re- \ncently. Two were four feet iong, and one \nover thirteen feet by actual measurement. It \ntook five men to hold the largest one up \nfor camera shots.\n\n\u201cAt last I am now doing the work I came \nto do. It is a real pleasure to work with Bell \nLab prints, texts, and KS parts again. \nI am in charge of a base shop here where we \nmaintain and repair a highly specialized bit of \nequipment. It is not particularly strenuous, \nbut each problem is a challenge and it takes \nall my electrical experience plus the training \nthat I had received in the Laboratories to \nfind the trouble.\u201d\n\n\u201cWhen I came back from furlough I \nstarted on a five-week course of intensive \ntraining. In that short space of time I \nlearned the basic principles of jungle war- \nfare. Recently, however, I was sent to \nSilver Springs, Florida, with a detachment to \nhelp in the filming of a motion picture of\n\nthe Navy. I had the pleasure of being in the \ncompany of men who have fought in all the \nmajor engagements of the South Pacific. \nI can honestly say that I learned more, from \nmy short association with those men, than I \nhave since I\u2019ve been in the Marine Corps. \nYou yourself would thrill to hear some of \ntheir stories. The picture took only a short \ntime to film and as a result, after only three \nweeks, I found myself back with my old \noutfit. I wish I could tell you what the film \nwas about and what we Marines did, but I \nhave promised to keep all that information \nto myself as it is a military secret.\u201d\n\n\u201cThe Fighter Control Squadron in my \naddress gives as clear a picture of my work \nas is permitted. I have had a good deal of \nfield experience in this type work since the \nNorth African campaign started, working \nwith both American and British personnel. \nI had the great good fortune of being one of \ntwo officers from this theater who were \nselected to return to the States for special \nschooling. I had six weeks there, and spent \ntwo days at the Labs visiting the old gang.\u201d\n\nLieut. Thomas Pariseau, stationed at Gore \nField, Mont., recently announced the birth of \n\u201cTommy ITT\u201d\n\n\u201cSchool is darned interesting and I really \nlike it. I graduated third in the class from \nGrove City, first Marine, but two sailors \nbeat me, with a 95.7 average.\n\n\u201cT'll really be in shape when I leave this \nplace. It\u2019s a second boot camp. Parris Island \nexercises, Judo lessons, obstacle course, \naerial gunnery practice, run the _ island\n\n(three miles), drill under arms\u2014and some \nflying time. And it really tires me\u2014I sure \nam soft after being in the hospital so long. \nBut I\u2019ll toughen up.\u201d\n\nCuHarites H. Doersam, Jr., has been \ngiven a personal leave of absence to work \nwith the Naval Research Laboratories.\n\nLieut. THomas M. Pepe and Josepu A. \nFAIRBROTHER have been transferred from \nthe Navy to the Marines.\n\nLieut. Warp K. Sr. has been \nstationed for some time at New Delhi, India, \nwith the U.S.A.F. of the China-Burma-India \nCommand. A recent letter states that he 1s \nrecovering satisfactorily from a severe illness.\n\nHarry W. Doutmar is now with the \nSeabees in the Hawaiian Islands, doing elec- \ntrical work there. \u201cI have been receiving the \nRecorb quite regularly. The December and \nJanuary issues containing articles on the de- \nvelopment of the Electrical Director were \nespecially interesting, as they brought back \nmemories of the days when I was working \nwith the group that developed it.\u201d\n\nBerry Kenny from the Accounting De- \npartment has been sent to the Naval Air \nStation at Pensacola, Florida, for a course In \nAviation Free Gunnery Instruction.\n\nRECENT visitors included Ligut. Eowarp \nFiuipovits, formerly of the \u201c\u2018Model Shop,\u201d \nwho hopes to fly a P-47 pursuit plane shortly; \nand Lieut. ARTHUR PALMER of the same \ndepartment who received his wings at Craig \nField; PereR SHEARER on furlough from \nCamp Robinson.\n\nJosepH P. REppINGTON is \u201cjust about \nover the hardest hurdle of the Marine Corps \n\u2014boot camp\u2014at Parris Island, South \nCarolina.\u201d\n\nHaroip W. Rarmert writes, \u201cI have just \ncompleted five months of sea duty on board \na merchant ship acting as the Navy radio- \nman. We stopped at several interesting \nplaces including Egypt, Arabia, Ceylon, and \nIndia. The old saying, \u2018Join the Navy and \nsee the world,\u2019 fitted my experiences per- \nfectly. At present I am a candidate for the \nNavy\u2019s V12 officers training and will start \ncollege in July.\u201d\n\n\u201cMy PARTICULAR work is still in the in- \nterests of bettering the communications \nfacilities of the Naval establishments,\u201d\n\nLr. ArtHUR PALMER PETER SHEARER \nwrites Ligur. CommManperR Ratpu_ H. \nMitter from Washington, D. C.\n\n\u201cENGLAND Is a nice place in the spring,\u201d \nwrites ALFRED Wickstrom, \u201cbut I\u2019ll take \nNew York any time!\u201d\n\nRoserr T. Durrey visited the Labora- \ntories when he was in New York recently.\n\nIn thirteen months of Army life Walter \nSokolosky has been stationed in twelve different \nbases. His most recent one is Bradley Field, \nfrom which he expects to be sent to a port of \nembarkation for overseas duty\n\nWarreEN E. WILson reports, \u201cI must say, \nit is interesting training here at Camp \nWolters, Texas, in spite of the heavy work. \nGive my best to all of the gang at the Labs, \nand tell them to write.\u201d\n\nAviation Capet Tuomas R. Harpen \nwrites that he is now at the State Teachers\u2019 \nCollege in Fitchburg, Mass., taking ele- \nmentary flight training.\n\nFrom a Signal Training Regiment at \nCamp Murphy, Florida, Haro.ip H. Horr- \nMAN writes, \u201c\u201cThere\u2019s plenty of work for us \nto do but an occasional trip to West Palm \nBeach ought to break the monotony.\u201d\n\nR. J. LaFrance is in a Seabee Battalion \nsomewhere in the Southwest Pacific.\n\nVicror S1Lzer has become * * * instructor \nat the Boca Raton Field Air Base in Florida.\n\n\u201cMy PRESENT assignment as a flight in- \nstructor is very pleasant and at times inter- \nesting,\u201d writes Ensicn J. R. Boyte, who is \nnow stationed at the Naval Air Station in \nGlenview, Illinois.\n\nTHE Recorp has been notified that Lizur. \nCoronet Witiiam H. Epwarps has re- \nceived orders for active overseas duty.\n\nMcAtevey of the Wac writes, \n\u201cSpent a very enjoyable furlough in Edin- \nburgh, and from there went to Loch Lomond \nfor a day.\u201d\n\nJames J. Viccers has been transferred \nto Camp Plauche in Louisiana. He sends \n\u201cRegards to the boys in the coil room.\u201d\n\nCHARLES HeEmpeEL, stationed at Camp \nBlanding, Florida, writes, \u201cI received my \nMay issue of the Recorp. In it I saw a \npicture of Watson RicHarpson, and I \nwould like to say hello to him through the \nReEcorp.\u201d\n\nLieut. KennetH E. Waters is overseas \nwith an A.A.A. Battalion; Witiiam G. \nPimPL is now overseas with New York APO \nnumber; Rospert D. Nostranp is on active \nsea duty; Joun Marko, A.S.T.U., is at \nRandolph-Macon College, Ashland, Vir- \nginia; and Frank Sarpinua is with the \nMerchant Marine.\n\nAviation Caper James M. Hoacianp, \nJx., has completed his college training course \nat Syracuse University and is awaiting as- \nsignment for pre-flight training.\n\ncently completed training in the Merchant \nMarine, receiving his commission at Fort \nSchuyler, is now on active sea duty with \nthe Navy.\n\nOn A RECENT visit to the Laboratories \nAviation Cadet JosepH KEL ty told that he \nhas completed his primary flight training \nat the Bunker Hill Naval Air Station in \nPeru, Indiana, and now expects to be sent \nto Pensacola, Florida, for advanced training.\n\nRopertr ErcHHorn visited West Street \nbefore leaving for Corpus Christi, his last \nbasic training base, where he hopes to win \nhis wings.\n\nRecent visitors to West Street have in- \ncluded Ropert SEymMour, who trained at \nCamp Crowder in Switchboard Operation \nand Installation and is now overseas; JOHN \nA. Zweic, who is assigned to Camp Rey- \nnolds replacement center; Ligut. THomas \nB. Horton, recently commissioned pilot at \nGeorge Field; and Frank C, Wanirts.\n\nTuomas Fox, on a visit to West Street, \nhad many interesting experiences to relate \nabout students whom he instructs at Fort \nMonmouth. At one time he had a trapper\n\nfrom the Northwest who was to learn how \nto repair telephones and switchboards\u2014and \nthe man had only seen three telephones in his \nlife before joining the Army.\n\nAs cusrop1An of enlisted men\u2019s service \nrecords at New Castle Army Base, Louis \nBett has learned much about personnel \nwork, he said when he visited the Lab- \noratories recently.\n\nWituiam L. Farmer, Jr., been \nselected by the Army to study medicine. He \nis now taking a Pre-Medical Course at State \nCollege, Pennsylvania.\n\nCrit. Epwarp H. Bues from a San Fran- \ncisco A.P.O. number says, \u201cGreetings from \na far corner of the world. Although a long \ndistance away we have been getting V-mail \nletters at an average of ten days from \nNew York City.\u201d\n\nRutH RypsBerc of the Marine Corps \nWomen\u2019s Reserve sends regards to everyone \nat the Laboratories.\n\nAviation Capet Martin E. Jounson \nwrites: \u201cUpon completion of my pre-flight\n\ntraining at Chapel Hill last month, I moved \non to the Naval primary air base located at \nGrosse Ile, Michigan, where I\u2019m getting my \nfirst taste of real flying.\u201d\n\nEnsicn Rosert H. Licut, now stationed \nat Princeton, says he is learning from the \nground up the \u201cwhys\u201d of some of the equip- \nment the Labs are building.\n\nCoMMANDER Ropman D. DeKay writes \nthat he is \u201cstill paddling around in the \nPacific, looking for Japs.\u201d\n\nHenry WipMann is in a Submarine Di- \nvision of the Navy with a San Francisco \npost office address.\n\nCaptain FREDERICK B. Mone has \ncompleted courses at the Signal Corps \nSchool at Fort Monmouth, and is now over- \nseas; H. Burcess now has \na New York A.P.O. address; and WARREN \nFE. THACKER sends regards to his friends in \nthe Labs from Camp Claiborne, Louisiana.\n\nDonatp J. OaKLEy, in the Navy, writes, \n\u201cT am stationed at a Naval Air Base in the \nSouth Pacific.\u201d\n\nAFTER SPENDING seventeen months in \nNew Guinea, Ligur. MALLEry rather \nhopes his next address will be from another \nclimate, preferably New York.\n\nRECENT PROMOTIONS among members of \nthe Laboratories in the service: LIEUTENANT \nFveretr C. Watsman; MarILyNn \nPearson, Aviation Machinist\u2019s Mate 2/c in \nthe Waves; Ser. J. B. KENNeEpy, Technician \nFourth Grade rating; Lizur. CoLone \nTHomas A. McCann, now serving overseas, \nattached to General Eisenhower\u2019s Head- \nquarters; SEconp Ligur. Georce N. \nin the Air Forces; Caprain LERoy G. Ratn- \nHART; SECOND GREGORY CHABRA In \nthe Air Forces; Lieu. ComMANDER Harry \nC. Hart, U.S.N.R.; Lizur. ComManper \nJoun R. Sackman, now overseas with a \nSan Francisco Post Office address; First \nLieut. Ratpw D. Horne, Jr., pilot of an \nFighth A.A.F. Flying Fortress, now in \nEngland; Seconp Lieu. Ropert F. HEALY; \nSeconp Ligur. Gerard E. Davis; Srarr \nSERGEANT M. EHLER.\n\nLieut. Ropert F. HEALy, who was com- \nmissioned a pilot at Victoria, Texas, has \nbeen assigned to Enid Air Field as an in- \nstructor. During his leave, Lieut. Healy \nvisited the Photocopy Department where he \nwas employed before entering service.\n\nRicHarp Rarrerty is taking a specialist \ntraining course in the operation of the \nElectrical Gun Director at Aberdeen Prov- \ning Ground in Maryland.\n\nCaper. Ernest C. GRraAuNAS, now in \nAustralia after a \u2018\u201c\u2018short stay in New \nGuinea\u201d; Lizur. CoLtonet R. O. Forp, \nfrom a New York A.P.O. address; RoBERr \nS. DrypEN, now overseas; L. M. Cassano \nin the Seabees; THomMas J. O\u2019NEILL, now in\n\nEngland; Lizur. Co. A. Specur, \non temporary duty, expects to return to the \nStates some time this summer; ALFRED O, \nScHWARz, from an advanced area of the \nPacific; RaymMonp S. YERDEN, now overseas \nin the Marine Corps.\n\nHarold Jaffee, G. C. Barry, Isabelle M. Ken- \nnedy, Captain B. M. Froehly, S/Sgt. C. E. \nUnderhill, W. M. Ehler, Adele Aboutok, K. G, \nOestreicher, K. A. Josephson, Edward Schoum, \nGrace M. Connor, Kay R. Parsons, Lieut. \nG. M. Richards, James Campbell, Alfred Bertin, \nA. F. Schweizer, R. F. Logan, W. F. Edwards, \nJr., H. C. De Valve, Jr., Lt. P. W. Foy, D. W. \nWebster, P. A. Walz, J. R. Merchant, Matthew \nTomb, W. E. Springer, R. F. Rennick, Lieut. \nCommander J. B. Newsom, E. J. Moskal, J. H. \nGeiger, W. R. Frees, Sara Dolin, L. J. Antonucci, \nR. H. Funck, W. B. Sage.\n\nMEMBERS OF THE LABORATORIES on \nmilitary leaves of absence who haven\u2019t been \nreceiving the Recorp due to incorrect ad- \ndresses are given below. The editors would \nlike any information concerning them.\n\nGeorge A. Carlson \u2014_ Richard H. Koehn \nStanley W. Erickson Ellsworth A. Lichtenberger \nDavid N. Fulton Peter F. McGann\n\nLieut. and Mrs. Gregory Chabra, both of the \nLaboratories, had this picture taken during a \nvisit to West Street, shortly after Lieut. \nChabra won his wings. He is now at the Naval \nResearch Laboratory in Washington\n\nF. B. Jewett, on May 17, addressed the \nmeeting of the Esso Research Club at Eliza- \nbeth on the subject of Some Applications of \nPhysics and Engineering in the War Effort. \nOn May 22 he attended the preview of the \nSecond Edition of the Air Power Show of the \nArmy Air Forces at the New York Museum \nof Science and Industry, and, as President \nof the Museum, made the opening remarks. \nOn the evening of May 24 he spoke before \nthe Maryland Historical Society in Balti- \nmore on the subject One Hundred Years of \nElectrical Communication in the United \nStates. This meeting was in celebration of a \njoint centennial of the founding of the \nMaryland Historical Society and the sending \nof the first telegraph message by Morse from \nWashington to Baltimore.\n\nA page abstract of Dr. Jewett\u2019s address, \nThe Promise of Technology, presented in the \nsecond series of conferences on Post-War \nGoals and Economic Reconstruction held \nunder the auspices of the Institute on Post- \nWar Reconstruction of New York Uni- \nversity, was published in the April 22 issue \nof Nature (London).\n\nO. B. BLackweELt represented the Lab- \noratories at the Centennial Telegraph Din- \nner which was held in Washington.\n\nW. G. B. Tuomas, and R. J. \nHEFFNER attended the evening session of \nthe annual meeting of the National In- \ndustrial Conference Board in New York \nCity on May 18. This session was devoted \nto an exploration of the Opportunities and \nProblems of Post-War Employment. After- \nnoon conferences on related subjects were \nattended by Mr. Heffner, J. S. Epwarps \nand C. W. F. Hauner.\n\nTHE Laporarories assisted the New York \nTelephone Company in providing an exhibit \nof communications apparatus at New York \nUniversity in connection with their recent \nMorse Centennial celebration.\n\nMcLenpon, Editor of Popular \nScience Monthly, together with two of his \nassociates, R. C. Witson and Goprrey \nHammonp, visited Murray Hill.\n\nMajor N. H. Staucurer, formerly a \nmember of the Laboratories staff at which \ntime he was active in early radio develop- \nments, also visited Murray Hill. He was ac- \ncompanied by Dr. D. H. Teeter and Dr. \nH. C. Pollack of the NDRC.\n\nTHE 1944 ROSTER of officers and com- \nmittee membership in the Institute of \nRadio Engineers follows: Treasurer, R. A. \nHeising; Board of Directors, F. B. Llewellyn; \nAdmissions, Lloyd Espenschied and F. A.\n\nW. P. Fengler W. H. Scheer L. L. Lockrow o Years \nto Years Polo Grafal Frederick Schrepfer Abe Maiman 3\n\nC. F, Chapman Hast J. G. Segelken R. H. McMahon F. J. Boyle \nC. I. Cronburg J. \u2018 Heincelman  H. G. Seifried R: E. J. Poole H. C. Harrison \nBa Bois R. J. Hluboky aon Wendelen Weis \n.\" Heiss, Jr. A. B. Horgan, Jr. A. M. Skellett W. E. Whitworth C. G. Von Zastrow \nNV. A. Jakob L. C. Hosek A. H. Smalenbach L. S. Youn ling R. H. Wilson \nViggo Marcussen Clara Imler R. R. Stevens : \nJ. H. Ross J. A. Kater W. J. Thompson 4 \nMargaret Wehner James Keegan W. H. Tidd a: incited 35 Years \nE. I Yacunski Eunice Kiefer Alice Todd G. C. Berndt HHI\n\nW. J. Albershej Neil McLaughlin Charles Gittenberger H.C. Spryer \npibersheim A. McLean 20 Years M. J. Harriot \nJ. C. BI: ank \u201c\u2122 5. M. Meehan od E. Barker C. M. Hemmer \nM. EF. Campbell A. F. Pomeroy J. F. Chaney H. J. Rostkas 40 Years \nG. J. Christ W. C. Prendergast \u20184 R_ DP\u2019heedene D. H. Mann A.G. Ki \nR. A. Cushman Iru Price E. K. Eberhart A. S. Miller, Jr. . G. Kingman \nW. H. Doherty Alice Reilly \u20ac.. G. Emery vi Ronci \nA. R. Ell Marjorie Remsen E. B. Ferrell H. O. Siegmund 45 Years \nN.S. Ewing Julius Rohr L. O. Goodman R. F. Squires \n1. E. Fair P. W. Rounds R. M. C. Greenidge VV. P. Thorp C. G. Spencer\n\nPolkinghorn; 4wards, R. K. Potter; Board \nof Editors, G. W. Gilman, F. B. Llewellyn, \nFE. L. Nelson, E. C. Wente, G. W. Willard \nand William Wilson; Constitution and Laws, \nR. A. Heising, chairman; Executive, R. A. \nHeising, vice-chairman, and F. B. Llewellyn; \nInvestment, R. A. Heising, chairman; Mem- \nbership, W. H. Doherty; Nominations, \nRalph Bown, chairman, and F. B. Llewellyn; \nPapers, \u00a5. B. Llewellyn, chairman, H. A. \nAffel, Fk. W. Cunningham, E. B. Ferrell, \nD. K. Martin, G. G. Muller and S. A. \nSchelkunoff; Public Relations, P. B. Findley; \nand Sections, R. A. Heising, chairman, and \nF. A. Polkinghorn.\n\nTechnical committee memberships are as \nfollows: Annual Review, C. R. Burrows; \nElectroacoustics, G. G. Muller, chairman, \nand L. J. Sivian; Electronics, EF. 1. Chaffee, \nS. B. Ingram, A. L. Samuel and J. R. \nWilson; Facsimile, Pierre Mertz; Radio Re- \nceivers, H. B. Fischer; Radio Wave Propa- \ngation, C. R. Burrows, chairman; Standards, \nC. R. Burrows and G. G. Muller; Symbols, \nC. R. Burrows; Television, A. G. Jensen; \nand Transmitters and Antennas, J. \u00a5. Mor- \nrison and J. S. Schelleng.\n\nK. K. Darrow has been appointed a \nmember of the Committee on American \nScientific and Technical Bibliography of the \nNational Research Council to compile a \nbibliography of American scientific and tech- \nnical books. This work is intended to pro- \nvide book-sellers and libraries throughout\n\nPlease put your RECORD in the \n\u201cCorrespondence-Out\u2019\u201d box when you \nare through with it so that it can be \nsent to a Serviceman\u2019s family.\n\nthe world with an annotated catalog of the \nbest American books in those fields.\n\nK. G. Compron presented a paper, Cor- \nrosion Protection of Metals, before the Ontario \nChapter of the American Society for Metals.\n\nD. A. McLean, accompanied by R. P. \nLutz of Hawthorne, visited the Mount \nHolly, Pa., plant of the Schweitzer Paper \nCompany to discuss capacitor problems. \nG. T. Kouman and Mr. McLean later \nvisited their Elizabeth plant for the same \npurpose.\n\nFE. C. Larsen attended a symposium, \nsponsored by the Powdered Metals Associa- \ntion, on powder metallurgy, held at the \nWaldorf-Astoria.\n\nG. H. Witirams visited Hawthorne in \nconnection with plastics and adhesive mat- \nters and G. GoopMAN on problems connected \nwith the production of ceramics.\n\nR. G. Humpnrey, formerly of the Chemi- \ncal Laboratories, has been transferred to \nthe Hudson Street plant of the Western \nElectric Company as division chief of the \nVacuum Tube Shop.\n\nC. V. LunpBerc visited Point Breeze and \nKearny to confer on thermoplastic rubber \ntape developments.\n\nS. B. InGram was the author of Electronic \nCircuit Design, published in the May issue \nof Electronics.\n\nG. C. SourHwortu spoke on Recent Re- \nsearches and Post-War Radio at a meeting of \nthe Yale chapter of Sigma Xi on May 17. \nBefore the Dayton Chapter of the Institute \nof Radio Engineers, at a meeting held on \nMay 25, he discussed Microwaves and the \nHollow-Pipe Technic.\n\nR. G. McCurpy, H. O. Srecmunp, N. Y. \nPriessMAN, C. B. GREEN and J. R. FLEGAL \ndiscussed thermistors at the Naval Ord- \nnance Laboratory in Washington.\n\nAr THE Hawthorne Plant of Western Elec- \ntric Company in Chicago, W. J. Kine dis- \ncussed high-voltage cables and connectors; \nC. A. WEBBER, cables; and A. J. GRossMAN, \nspecial networks. Mr. Grossman also went\n\nto the Derry (Pa.) plant of the Westing- \nhouse Electric and Manufacturing Com- \npany on matters pertaining to ceramic \nbushings.\n\nH. H. Gienn and H. H. Sraesner con- \nferred with engineers of the Bureau of Ships \nin Washington on cord and cable specifica- \ntion requirements. Mr. Staebner was also at \nPoint Breeze on cord-design problems.\n\nR. T. Srapes, at the Boston Insulated \nWire and Cable Company, discussed special \ntypes of cables.\n\nR.R. MacGrecor visited the Camp Evans \nSignal Laboratory and the Fort Mon- \nmouth Signal Laboratory in connection with \ntests on transformers.\n\nJ. Nuner, at the Magnetic Windings \nCompany at Easton, Pa., discussed trans- \nformer problems.\n\nA. B. Haines visited the Sharon plant of \nthe Westinghouse Electric and Manufac- \nturing Company on moisture-proofing of \ntransformers.\n\nC. A. McJounstron, with H. W. Buden- \nbender of Hawthorne, visited the Line Ma- \nterial Company at Zanesville, Ohio, in re- \ngard to the manufacture of transformers.\n\nJ. E. Rances visited the Pittsburgh plant \nof the Westinghouse Electric and Manu- \nfacturing Company on matters pertaining \nto hermetically sealed terminals.\n\nJ. R. BarpsLey was at Hawthorne in con- \nnection with the approval of tool-made \nsamples of loading-coil cases.\n\nM. WuirEHEAD visited the Aerovox Cor- \nporation at New Bedford, Mass., to discuss \nelectrolytic capacitors.\n\nW. CLaypen and P. T. Hicearns visited \nthe step-by-step telephone offices at Sharon, \nPa., and Jacksonville, Fla., in connection \nwith step-by-step field studies.\n\nR. C. Jones was at Point Breeze to discuss \nsubstituting synthetic rubber in type-J \ncarrier cables.\n\nA. L. Fox, A. L. Ricuey, J. H. Gray, \nF. D. Watpron and B. Dysart visited \nPrinceton in connection with experimental \ninstallations of coaxial cable.\n\nV. B. Pike visited Chicago, Cleveland, \nand Detroit with reference to the improve- \nment in drainage switches installed to mini- \nmize damage by electrolysis of lead-covered \nunderground cables.\n\nA T & T, went to Richmond to discuss the \ninspection of poles with engineers of The \nChesapeake and Potomac Company.\n\nR. H. Coxrey presided at the Annual \nMeeting of the American Wood-Preservers\u2019 \nAssociation at Chicago. He also visited the \nForest Products Laboratory at Madison, \nWisconsin, and the Valentine Clark Corpo- \nration at St. Paul to discuss problems relat- \ning to the preservation of timber.\n\nJ. W. Van ve Water, field engineer of the \nQuality Assurance Department at Omaha, \nrecently received a letter from W. R. \nJohnson, vice-president and general man- \nager of the Northwestern Bell Telephone \nCompany, commending him for his assistance \nin restoring telephone service at the Norfolk \n(Nebr.) central office following a serious \nflood. In charge of fire protection and flood \ncontrol equipment for the Omaha Civilian \nDefense Organization, Mr. Van de Water got \npumping and other equipment to the central \noffice promptly and, working day and night \nfor several days, helped get the office back \ninto service. Following this, he moved the \nequipment to a number of other business \nbuildings where telephone company per- \nsonnel helped him pump them out.\n\nW. L. Dawson was at the Patent Office \nin Richmond relative to patent matters.\n\nTHE LABoratorigEs were represented in \ninterference proceedings at the Patent \nOffice in Richmond by G. C. Lorp before \nthe Board of Interference Examiners.\n\nI. A. McCorkenDALe appeared before \nthe Board of Appeals at the Patent Office in \nWashington relative to an application for \npatent.\n\nG. B. THomas, R. A. DELLEr, R. J. HErr- \nNER, M. B. Lone, and E. W. Warers at- \ntended the spring meeting of the Mid- \nAtlantic Section, Society for the Promotion \nof Engineering Education, held in Camden \non May 6.\n\nKk. D. Leamer has recently delivered a \nseries of four lectures to a class on labor rela- \ntions conducted under the auspices of the \nEngineering, Science and Management War\n\nMiss M. C. Brarnarp, from May I to 3, \nvisited five Pennsylvania schools\u2014Bryn \nMawr, the University of Pennsylvania, \nDrexel Institute, Temple University and \nPennsylvania State College\u2014to recruit col- \nlege women for technical positions with the \nLaboratories. On May 4 and 5 she at- \ntended a conference, War and Post-War Em- \nployment and Its Demands for Educational \nAdjustment, held in Washington by the In- \nstitute of Women\u2019s Professional Relations.\n\nDurinG THE month of May, Mrs. E. D. \nDodge visited high schools in the New York \nCity area to recruit girls for messenger and \nmail clerk positions.\n\nIn THE April issue of The Review of Scien- \ntific Instruments J. N. SHivE reviews the \nbook Electronic Physics by L. G. Hector, \nH. S. Lein and C. E. Scouter; W. A. SHEw- \nHART, Treatment of Experimental Data by \nA. G. Worthing; and K. K. Darrow, The \nLife of J. J. Thomson by Lord Rayleigh.\n\nGreat Day Spiritual \nChorus and Orchestra \nSylvia Speaks \nRide, Cowboy, Ride Guion \nOl Man River from \u2018\u2018Showboat\u201d\u2019 Kern \nWalter Cassel \nAn American in Paris Gershwin\n\nIn the Good Old Summertime Evans \nOrchestra \nLiebestraum (No. 3\u2014O Lieb) Liszt \nSpanish Dance in G Major Granados \nJos\u00e9 Iturbi \nLove Scene from \u201cRomeo and Juliet\u201d \u2014 Berlioz \nOrchestra \nConcerto No. 1, in B Flat Minor, Tschaikowsky \nOp. 23\u2014Finale \nJos\u00e9 Iturbi and Orchestra \nJULY 17, 1944 \nConcerto in E Minor\u2014First Mendelssohn\n\nBell Laboratories\u2019 Club has no more tickets for these programs \nbecause its limited supply has already been distributed to applicants.\n\nOverture to \u201cThe Marriage of Figaro\u201d Mozart \nOrchestra \nTango in D Albeniz-Kretsler \nCaprice Viennois Kreisler \nFritz Kreisler and Orchestra \nJULY 24, 1944 \nFuriant from \u201cThe Bartered Bride\u201d Smetana\n\nOrchestra \nAmour! viens aider ma faiblesse \nfrom \u2018Samson\u2019 and Delilah\u201d \nGladys Swarthout\n\nBeautiful Lady from \u201cThe Pink Lady\u201d = Caryll \nOrchestra \nFlow Gently, Sweet Afton Traditional\n\nGladys Swarthout and Male Chorus \nPiper O\u2019 Dundee Trad.-arr. Respight \nGladys Swarthout \nDubinushka Rimsky-Korsakoff \nOrchestra and Male Chorus \nL\u2019Antoueno Trad.-arr. Canteloube \nGladys Swarthout\n\nOverture to Zampa Herold \nOrchestra \nSonata No. 14 in C Sharp Minor Beethoven \n(Moonlight)\u2014First Movement \nJosef Hofmann \nConcerto in E Major\u2014Last Movement Chopin\n\nTHOUSANDS of servicemen and civilian \nemployees of the Armed Forces, stationed in \nall parts of the world, will remember Mrs. \nRoupe as their first and most personal \ncontact with the Bell Telephone Labora- \ntories School for War Training. She greets \nand registers the students on their entrance \nto the school and arranges for their release\n\non their final day, never for a moment bridg- \ning the relationship established throughout \ntheir stay at the school. Although her job \nis generally defined as associated with regis- \ntration, distribution of text material and the \nchecking in and out of classified documents, \nthis is the mere beginning of her actual work \nfor the school and for the students. As ad- \nvisor to students on housing, recreation, city \ntransportation facilities, etc., the situations \nhave been rare that she has not been able \nto cope with. She has been carrying on this \nImportant war job with infinite patience and\n\nMrs. Rohde is a native of Louisiana. She \ngraduated from the Southwestern Institute \nof Louisiana with a B.A. degree, after which \nshe taught school for a time before coming to \nNew York and joining the A T & T. There, \nin the D & R, she had charge of a group of \ngirls doing oscillogram analysis work. When \nthe D & R was merged with the Bell Lab- \noratories, she continued the same kind of \nwork until the opening of the School for \nWar Training.\n\nAt the present Mrs. Rohde\u2019s interests \nafter hours are confined to her home and \ngarden. Her leisure time, however, is ex- \nceedingly limited and, like the rest of us, she \nis looking forward to the end of the war \nwhen she will have more time for these and \nother avocations.\n\nlowing year she attended \nDrake Business School and \nthen joined the Laborato- \nries Transcription Depart- \nment. For the past three \nyears she has been doing \nsecretarial work at Mur- \nray Hill. Edith is active in \nthe Red Cross and Girl \nScouts and likes bowling \nand the less strenuous di- \nversion of needlework. Her \nsummer vacations are \nusually spent down on \nthe Jersey coast at Barne- \ngat Bay where she can \nenjoy her favorite sport, \nswimming. \n* * *\n\nHer corpiAL manner and pleasing ap- \npearance make Mrs. Maser Rocue well \nsuited for the position of receptionist at \nWest Street. Her duties include seeing visit- \nors who wish to be admitted to the build- \ning, issuing passes as required, and providing \nmessenger escort.\n\nMrs. Roche was born in \nEvanston, Illinois, but \nspent a great part of her \nchildhood traveling back \nand forth between Chi- \ncago and New York with \nher parents. At school she \nstudied art with the in- \ntention of satisfying an \nearly desire to become a \nfashion artist. However, in \nher first position as a re- \nceptionist she soon dis- \ncovered that she preferred \npublic contact work. Mrs. \nRoche then came to the \nLaboratories where she \nhas been engaged in this \nwork for two years.\n\nWith a husband and a home in Jackson \nHeights to take up much of her leisure time, \nMrs. Roche has little opportunity for relaxa- \ntion and hobbies. Nevertheless she does find \ntime to search for records for her collection \nwhich includes recordings by Enrico Caruso \u00a9 \nand Harry Lauder, her favorites.\n\nShortage of stenotypists has prompted the starting of a training class at Murray Hill. It meets\n\ndaily under the capable instruction of Muriel Cadmus (center). At the end of the third lesson\n\nall of the pupils were so enthusiastic about the course that they bought machines of their own\n\nand their progress has been most gratifying. From left to right are Corrine Marky, Charlotte \nFischer, Peggy Anderson and Muriel Bey\n\nFor JenNnrE Damiani doing precision \nmeasurement work in the laboratory, where \nmeasurements as fine as one ten-thousandths \nof an inch are made, is a striking contrast to \nher former work. A native New Yorker and \na graduate of Hunter College, she had \npreviously worked at Fort Monmouth, first \non the installation of radios in Army trucks, \njeeps and tanks and later on the building of \ntransmitters, receivers and power supplies \nfor research and development models. How- \never, she enjoys her work in the Precision \nMeasurement Laboratory far more than \nanything she has ever done. From all de- \npartments of the Laboratories war develop- \nments and communications materials are \nbrought to be measured and tested. As a \nresult she is constantly learning about new \nmaterials and new subject matter.\n\nMiss Damiani has recently become the \nbride of her childhood sweetheart. Besides \nher scientific bent, she has athletic and do- \nmestic abilities. She enjoys all outdoor \nsports and excels in swimming; and she \nadmits the ability to cook, sew, and do \nhousekeeping. Non-fiction reading is her \nhobby, particularly philosophy.\n\nIN THE SPECIFICATIONS distribution group \nof the Apparatus Development Department, \nMary McLoucuuin is one of a group of \nseven girls responsible for processing all\n\nspecifications issued by the \nDepartment. Her part is the \nfollowing through on drawings \nfor the eight or nine thousand \n\u201cspecs\u201d distributed in a year. \nYet amid the excitement and \nthe bustle of ordering, assem- \nbling and mailing them out\u2014 \nand nearly all are either rush \nor confidential\u2014Mary not only \nmanages to keep her sense of \nhumor, but also to conserve \nenough energy to spare for her \navocation, caring for the sick. \nAs a member of the Volun- \nteer Hospital Association, one \nnight a week she goes to the \nBeekman Street Hospital \nwhere, under the supervision \nof HELEN Cory of the Lab- \noratories, she helps to prepare \nthe patients for their night\u2019s \nrest. Later the harder work begins. It in- \ncludes the sterilization of instruments and \nthe cleaning up of the operating room. \nMary, upon graduation from the Girls\u2019\n\nHigh School in Brooklyn, became a member \nof the Laboratories\u2019 Blueprint Department \nwhere she worked for five years before trans- \nferring to her present position.\n\nEvery one of us has at least one relative or \nfriend in the service to whom we do or should \nwrite regularly and often. Not only the \nnumber, but also the kind of letters we write \nis important. An article in the June issue of \nVogue offers some suggestions we might con- \nsider when we write:\n\n\u201cNever in all history were more letters \nto be shuttled across the earth. Letters today \nare making the difference between hope and \ndespair. They have become the one tie be- \ntween home and the fighter, and thus letters \nhave become one of the most important \nsingle influences on the whole functioning of \nthe war.\n\n\u201cThe men who have the job of observing \nthe spirit of fighting men\u2014Army and Navy\n\n*Jack Maisel\u2014Edith Tarnover \n*Harold A. Sweet, Jr.\u2014Carol Henry \n*Lt. Richard C. Benkert, U. S. Army\u2014*Marjorie Flynn \nWilliam T. Parsons, U.S. Army\u2014*Phyllis-Jane Vibbard \nJohn H. Eick, U. S. Army\u2014*Marie Teschner \nGeorge W. Trenchard, U. S. Navy\u2014*Mildred Paul\n\n*Charles S. Graham, U. S. Army\u2014Connie Pascal \n*Roland M. Scheller, U. S. Army\u2014*Edith Jacobs \nT. D. Bergowsky, U. S. Army\u2014*Freda Schwartz \nManny Gelburd, U. S. Army\u2014*Harriet Spiro \n*Robert L. Slobod\u2014Jeanne Lamar \nFrederick Berwind, U. S. Army\u2014*Lucille Anderson\n\nments and weddings should be given to Miss Mary \nEllen Wertz, Room 1103, Extension 296.\n\nchaplains, personnel officers, Red Cross \nfield directors\u2014are much concerned about \nthe kind of letters the men are receiving. \nKrom every front they report that they see \ngood men give up under the strain of bad \nletters; see other men become heroes _be- \ncause the right letters keep on coming.\n\n\u201cThere are letters from the self-pitying \nloved ones who think of nothing but them- \nselves. They undermine a man\u2019s attempt at \nadjustments. The families shouldn\u2019t write \nto the men of things about which the men \ncan do nothing.\n\n\u201cAn entertainer on tour in Australia \nspoke of men \u2018packed twenty at a time in \nfoxholes built for four,\u2019 receiving letters \ncomplaining about crowded wartime trolleys.\n\n\u201cWhen there is really bad news, the man \nmust, of course, be told. No man wants to \nfeel he is being \u2018protected\u2019 from things he \nshould know. The art is to keep the man \nfrom being shocked, and yet keep him as- \nsured that he is not being \u2018shielded.\u2019\n\n\u201cTo someone really close, one should \nwrite every thought, every detail of daily \nlife. One should transport the life at home \nto him as clearly as words make it possible, \nfor he wants so much to live that life \nvicariously.\n\n\u201cNo man wants a lot of phoney letters \nfilled with forced good cheer. They should \nbe filled with honest hope for the future, \npride in that job he is doing, and faith in \nwhy he is doing it.\u201d\n\nThe actual campaign extended from May \n23 to June 3 and was sponsored by the \nExecutive Committee of the Bell Labora- \ntories Club with G. B. Thomas acting as \nCampaign Chairman. Much of the credit for \nthe success of the Drive was due to the \nsplendid codperation of the 260 club mem- \nbers who solicited all members of the Lab- \noratories in New York.\n\nCirarence H. Berry of the Switching \nDevelopment Department, with over forty \nyears of service, retired under the Retire- \nment Age Rule with a Class A pension on \nJune 30. After graduating from Pratt Insti- \ntute in 1899 he worked for three years with \nthe Suffolk Gas & Electric Company on \nLong Island and then joined the New York \nTelephone Company. For \nseven years he served as wire- \nman, inspector, tester and \nforeman of switchboard in- \nstallations. He left the com- \npany in 1909 and spent a year \nwith the Radio Telephone and \nTelegraph Company under Lee \nDeForest as draftsman and \ndesigner.\n\nIn tg10 Mr. Berry joined \nthe Western Electric Company \nin New York City and for \nfour years was inspecting and \ntesting central-office installa- \ntions, first in metropolitan \nNew York and later in the \nterritory of the Southwestern\n\nment, and finally became Division Inspector \nin the New York Telephone Company\u2019s \nterritory, covering New York State, Long \nIsland and part of New Jersey. In 1919 he \ncame to the Laboratories where he has since \nbeen engaged in the design, development and \nanalyzation of circuits for dial and manual \nsystems. \n* * * * *\n\nMrs. Bessie M. GIFFHORN, secretary to \nO. B. BLackwELL, Vice-President, retired at \nher own request with a Class A pension on \nJune 9. Mrs. Giffhorn was with the American \nTelephone and Telegraph Company from \n1915 to 1919 and for a short time in 1921. \nFive years later she returned to that or- \nganization as a member of the Department \nof Development and Research and came \nto the Laboratories at the time of the 1934 \nconsolidation.\n\nFive U. S. Army transport planes roared \ndown the runways of Atlanta\u2019s Candler \nField recently and winged their way west- \nward with eleven tons of packaged central- \noffice equipment. Fire and water had \nwrecked the telephone office at Camp Van \nDorn, at Centreville, Miss., and the planes\u2019 \ncargo of a 1,000-line manual switchboard \nand toll line relays, in 150 packages, was\n\nBell Telephone Company. He This \u2018Self-Service Egg Layer\u201d was presented to C. H. Berry \nthen transferred to the New by his associates in Systems Development at the time he \nYork headquarters of the Cen- retired. It was designed and built by L. E. Van Damme of\n\nPackaged central-office equipment is loaded on one of the \nfive Army transport planes which carried the emergency \nwar stock on the first leg of its journey to Camp Van Dorn \nin Mississippi, to replace a central office damaged by fire\n\nthe first consignment of the American Tele- \nphone and \u2018Telegraph Company\u2019s war \nemergency stocks to be used for replacing a \ndamaged telephone office.\n\nThat telephone service for the big Army \ncamp was fully restored in \nnine days reflects credit not \nonly on the teamwork of the \nArmy Air Forces, the Signal \nCorps, Southern Bell \u2018Tele- \nphone Company and the West- \nern Electric Company but also \non the Bell System people who \nconceived and carried out the \nplan for storing pre-engineered \nand packaged equipment at \nstrategic locations throughout \nthe country as_ insurance \nagainst enemy action or other \nwar-born emergencies.\n\nThe manual packaged unit* \nsent to Camp Van Dorn had \nbeen designed to be installed \nintact as a complete tailor- \nmade job, and as such could be \nset up and in operation within\n\nthree days. But since this cen- \ntral office combines manual, \ndial and toll equipment, the \npackaged unit had to be \ngreatly adapted. In view of \nthis, completion of the instal- \nlation in nine days stands as a \nremarkable achievement.\n\nThe transport planes took \nthe equipment from Atlanta \nto Baton Rouge, La., where it \nwas quickly transferred to a \nconvoy of Army trucks and \nsent on to Camp Van Dorn. \nSome of the toll equipment \nhad to be ordered from several \nof Western\u2019s plants and dis- \ntributing houses. The assem- \nbling, wiring and testing of \nthis equipment was done on \nthe job.\n\nManpower for the job at \nVan Dorn came from all points \nof the compass. Twenty-eight \nWestern Electric installers ar- \nrived the first day from Baton \nRouge, New Orleans and Jack- \nson, Miss. Additional Southern Bell people \ncame by air and railroad. Most dramatic, \nhowever, was the arrival of an Army sig- \nnal company, which was on a field problem \nfifty miles away at the time of the fire. The\n\nGroup of Bell System General Plant Managers viewing Spiral-4 equipment at Murray Hill\n\nsignal troops worked hard under pressure \nand made a real contribution to the job.\n\nOn the night of the fire, a call from the \nsenior switchman at Van Dorn, transmitted \nover a makeshift circuit, set the wheels in \nmotion for service restoration. Key tele- \nphone men at Jackson were called together \nat midnight; they notified others at com- \npany headquarters in Atlanta.\n\nSpeed was the keynote of the whole \noperation. Telephone people worked side by \nside with Army men. Some of them went for \ndays without sleep. During an early stage \nof the job, Army emergency service was \nmaintained by connecting Army magneto \nsets with a 16-pair telephone company \ncable. In addition, the Signal Corps set up \ntwo positions of Army magneto switch- \nboards across the street, and placed field \ntelephones at strategic locations.\n\nAt two of the camp\u2019s attended locations \nMagneto service was arranged on an emer- \ngency basis during the restoration period. \nhe trunks were connected between the \nArmy switchboards in the fire station and \nthe central office. Power was provided by a \npartial salvage of batteries from the office \nand by automobile-type batteries.\n\nSixty general plant managers of the Bell \nTelephone System visited the Laboratories \nat Murray Hill May 26 on the final day of \ntheir conference in New York. They viewed \ndisplays of war developments, including air- \nborne and tank radios, military telephones, \nquartz crystals, Spiral-4 and package toll \nand carrier and a portable information \ncenter. These were demonstrated by R. L. \nCaseg, J. A. Coy, F. E. Wottensack, W. C. \nJones, W. P. Mason and J. R. Erickson. \nA demonstration of dialing speed was given \nby L. J. Sracy. C. F. Serpe discussed auto- \nmatic ticketing developments. The visitors \nwere shown the special construction features \nof the buildings and the rubber and metal- \nlurgical laboratories by F. P. GILLILanp, \nA. R. Kemp, E. E. Scuumacuer, R. L. \nTowne and G. W. Legs, Jr.\n\nOur Service Flag \nThe blue numbers on the Laboratories\u2019 \nservice flag gave an upward jump on June 1 \nnot solely based on those left who for service \nthe preceding month. Formerly the net \nnumber on military leaves of absence was\n\nshown. This practice has been in accordance \nwith the strictly literal interpretation of the \nWar Department\u2019s regulations concerning \nservice flags, but the A T & T was advised by \nit recently that either such a net figure or \nthe total number of all employees who have \nbeen in the armed forces during the current \nwar would be entirely satisfactory.\n\nSince, therefore, continued use of the net \nfigure might soon result in a diminishing \ntotal as the number of new entrants de- \ncreases and the number of men returning \nincreases, the A 'T & T thought it desirable \nto include in the total shown on the flag all \nthose who were serving with the armed \nforces at the time of Pearl Harbor plus all \nthose who have served since that time. Upon \nbeing advised of the findings of the A T & T \nand their action, the Laboratories acted to \nchange its flag accordingly. As of May 31, \n846 members of the Laboratories have been \non military leaves of absence of which 33 \nleaves have been completed and 813 were \nstill active.\n\nIn spite of optimistic news headlines on \ngasoline, always remember that the auto- \nmobile transportation chain has more than \none link. A tank of precious gasoline may be \n\u201conly a fire hazard\u201d in a car that is out of \nservice. Tires and car repair service are at \npresent on the shortage list. Reports from\n\nmany Local Ration Boards still show tire \nrequests more than double the size of their \ntire quotas, and car repairs in many cases \nmust be scheduled days ahead.\n\nUntil the good old days return when you \ncan say \u201cfill \u2019er up,\u201d the safest way to keep \nthem rolling is to drive slowly, take care of \nyour car, and share rides when you can.\n\n\u201cGet your sun tan gradually\u2014a few min- \nutes a day to start.\u2019 We hear it every \nsummer when we start thinking about a \nnice golden tan. But with only one day a \nweek in which to soak up our quota of \nVitamin D on the beach or in our Victory \ngardens, we find it difficult to remember that \nseveral hours the first day in the sun will \nresult in several days and nights of misery.\n\nThose who take their sun at the shore \nshould remember that alternate bathing \nand baking result in even more serious re- \nactions. Headache and nausea are among \nthe symptoms. These burns are painful and \ncause sleeplessness, stiff muscles and large \nblisters. Disability may last a week or more.\n\nSo, when you are out in the open protect \nyour skin as much as possible. No matter \nhow painful, a \u201cbrilliant\u201d case of sunburn \nnever excites sympathy, but instead merely,\n\nRed Cross Campaign Results \nContributions to the New York City \ncampaign of the 1944 Red Cross War Fund \nby the various communication companies \nand their employees totaled $518,977, 34 \nper cent more than last year. Corporate \ngifts, which were $387,270, exceeded last\n\nyear\u2019s figure by 33 per cent and employee \ncontributions, which amounted to $131,707, \nrepresent an increase of 37 per cent over \nlast year. Employees contributed an addi- \ntional $50,400 which they requested be \ncredited to chapters outside New York City.\n\nThe Laboratories was third in individual \ncontributions with a total of $23,159 and \nsecond in the average amount of pledge, \n$6.31 per contributor.\n\nThe Marine communications post on \nTarawa Island brought poignant memories \nto S/Sgt. Jack Brooks, a former member of \nthe Western Electric Company from the St.\n\nA. C. Holetz* George Seidel (5) \nConstance Kaval (2) H. T. Sorensen (2) \nG. T. Loman W. D. Stratton\n\n\u201cIT set up the first switchboard in that \npillbox,\u201d he wrote from the Southwest \nPacific. \u201cIn fact, as far as I can gather, it \nwas the first board to be set up on the island. \nIt wasn\u2019t a good idea to sit out in the open \nlike those guys are doing (in the picture). \nThe sign saying there are Nips buried there is \nquite correct. The exact number was four, \n\"cause we cleaned out that pillbox and buried \nthem ourselves.\u201d\n\nWhile working on that switchboard, the \nSergeant said, his buddy was killed and his \nown tools, for which he had paid $20 in \nNew Zealand, were ruined.\n\nWhen the U. S. Marines set up business on Tarawa, they named their communications head- \nquarters as shown above\n\nManufacture by the Western Electric \nCompany of telephone sets for civilian pur- \nposes, which was interrupted due to war \nrestrictions in November, 1942, is to be \nresumed to a limited extent, with deliveries \nexpected to begin late this year. Western \nElectric, which has been authorized by the \nWar Production Board to supply approxi- \nmately eighty per cent of a projected 200,000 \ntelephones per quarter, has announced \nacquisition of a factory for that purpose in \nSt. Paul, Minn.\n\nThat resumption of manufacture of 200,- \n000 telephones every three months will not \nsolve the present set shortage was made \namply clear by Leighton H. Peebles, direc- \ntor of WPB\u2019s Communication Division, who \nwarned that authorized production will not \nfor some time approach essential demand.\n\nThe Pacific Coast, where military and \nassociated civilian requirements are urgent, \nwill receive the first of the sets. After that, \nthe newly produced instruments will be as- \nsigned to areas most needing them.\n\nObituaries \nCHARLES P. KouLer, a night watchman \nin the Building Service Department, was \nkilled in the performance of his duties as a\n\nWhe depenrtng tie, was enrbled aso volunteer \n* of the Ate Barden Service, Citijene Detense \n} Corps, City Of Rew Dork, crud versed and ly \nASUS Honcetly and Jattbtullp \u00abPost warden \nsectcr 4 \u201cZone A Precinct 6 \nuntel death on May 24th \nth, \nThe Mayor of the City of Rew Work, mesnbons \nof Go Barden Service, with prrfound the \nof Chas member of our Serve, and extend thar \npathy the memlors of family\n\nregistered Air Raid Warden during the \nstate-authorized test blackout on the night \nof May 24. Mr. Kohler was last seen entering \nthe Baker-Williams building to assume his \nassigned post as a fire-watcher on the roof. \nAt the conclusion of the blackout he did not \nreturn to his regular duties in the West\n\nStreet building and in the investigation \nwhich followed it was found that he had \nbeen killed in a fall down an elevator shaft \nin the Baker-Williams building.\n\nDr. Buckley, in his letter to Mrs. Kohler, \nhas aptly expressed the sentiments of the \nLaboratories with these words . . want \nyou to know that we feel Mr. Kohler died \nin the performance of his duty as a good \ncitizen and in the service of his country. ...\u201d\n\nMr. Kohler joined the Laboratories in \n1936 as a night cleaner and became a night \nwatchman in 1942.\n\nJ. SuLLIvAN, a staff assistant in \nthe Apparatus Development Department \nwith continuous service in the Bell System \nsince 1922, died on May 31. Mr. Sullivan \nserved in World War I from April, 1917, to \nApril, 1919, most of the time in France with \nthe 106th Machine Gun Battalion of the \n27th Division. Before joining the Installa- \ntion Department of the Western Electric \nCompany in 1922 he spent a year at New \nYork University and later took evening \ncourses at Fordham University. Mr. Sulli- \nvan transferred to what is now the budget \nand case cost group of the Apparatus Staff \nDepartment in 1924. Here his work was \nprincipally concerned with budget statistics, \nplant expenditures and general cost account- \ning on apparatus development projects.\n\nFEuma L. Way, secretary to H. M. Bas- \ncom, Director of Switching Engineering, died \non May 22. Soon \nafter graduating \nfrom the Paine \nBusiness School, \nMiss Way become \na stenographer and \nfrom then until \n1920 succes- \nsively with the \nMaryland Casualty \nCompany, Johnson \nand Higgins and \nthe Aetna Insur- \nance Company. She joined the Engineering \nDepartment of the A T & T in 1920 and \nlater that year transferred to the OW E \nand two years later to the D & R. Miss Way \ncame to West Street when the D & R was \nconsolidated with the Laboratories in 1934.\n\nSubstantial economies in time, manpower \nand money have been effected through the \nrecent standardization of radio tubes used \nby the Army, Navy and the Canadian armed \nservices. Among the more important results \nof the standardization has been easier inter- \nchangeability of tubes, com- \nmon stockpiles, elimination of \ndual inspections at the manu- \nfacturers\u2019 plants, and use of \nstandard nomenclature.\n\nBefore Pearl Harbor the \nArmy and Navy each had its \nown system of tube nomencla- \nture which was, in many cases, \nunrelated to that of the Radio \nManufacturers\u2019 Association \nand other commercial type \nnumbers. Early in 1942 work \nwas begun to prepare a Joint \nArmy-Navy Specification for \ntubes to be based on the use of \nRMA and commercial type \nnumbers. The work was done \nby a committee of Army and \nNavy representatives, assisted\n\nrepresentatives from  in- \ndustry. Among the latter were \nJ. R. Witson, director of elec- \ntronics research, and most of\n\nthe engineers of his staff. Among them were \nV. L. Rone, A. I. Crawrorp, J. O. \nMcNatty, S. B. Incram, G. T. Forp, \nM. S. Grass, A. L. Samuet, J. B. Fisk, \nH. E. MENDENHALL, and C. E. Fay. H. F. \nDopvce of Quality Assurance also partici- \npated. Many of the committee\u2019s meetings \nwere held at the Laboratories. This speci- \nfication has now been authorized by a Joint \nArmy-Navy Committee and the Canadian \narmed services have also adopted it.\n\nAfter the transition stage, during which \nexisting tube stocks will be used up, all \ntubes purchased and stocked by the organi- \nzations just named will be fully interchange- \nable. This one factor alone is of tremendous \nimportance in the field, where replacements \nare needed in a hurry. Army tubes will now \nwork in Navy equipment and vice versa, \nwithout loss of valuable time.\n\nFurthermore, by pooling their require- \nments, the Army and Navy have been able \nto improve the quality of the tubes. The \nmanufacturer no longer is obliged to make \nthe same tube meet two slightly different \nspecifications, with the result that he can \nnow concentrate on making that tube better. \nAlready reports have been received from \nfighting fronts telling of the superior char- \nacteristics and quality of these tubes.\n\nElectronic units for field use, like this BC454B radio receiver \nfor radio command in airplanes, must be designed to take \ninto account the normal tolerance variations in electron tubes. \nThese standard tolerances are set forth in Joint Army-Navy \nSpecifications JAN-1A for Radio Electron Tubes\n\nabout one million have been treated \nwith toxic compound for protection against \nattack by wood-destroying fungi and insects. \nThe one million are the surviving untreated \nchestnut and northern white cedar poles \nwhich are relatively resistant against attack. \nAt the time they were placed, it was the \ngeneral practice to set poles sufficiently over- \nsize to provide a safe margin of strength \nduring their service life and to replace them \nwith new ones when deterioration at and \njust below the ground level had so reduced \ntheir size that they were no longer strong \nenough to sustain wire and storm loads.\n\nExternal deterioration of an un- \ntreated pole in service begins in \nthe ground section with infection \nby wood-destroying fungi which \nare almost universally present in \none form or another in the soil and \nin the air. Once established in a \npole the destructive organism will \ncontinue there unless conditions \nadverse to it are established, for \nexample, by the application of a \nsuitable wood preservative.\n\nAbout eight years ago interest \nrevived in providing means of \nprolonging the life of untreated \ncedar and chestnut poles in service. \nThe Laboratories, therefore, laid \nout a program to develop a low \ncost but effective preservative \ntreatment which could be applied \nto the vulnerable section of the \npoles, that is, from a few inches \nabove to some twelve to fifteen \ninches below the ground line.\n\nThere are two general classes of \npreservatives that might be used \nfor ground line treatment, namely, \noily materials such as creosote; and water \nsolutions of toxic salts, for example, zinc \nchoride and sodium fluoride. Sodium fluoride \nappears to be particularly good for penetrat- \ning the heartwood of cedar and chestnut tim- \nbers but it is not permanent. On the other \nhand, creosote or a combination of creosote \nand coal tar, although not as penetrating as \nthe water soluble salt, is about as lasting as \nany preservative known. The characteristics \nof the two types of materials suggested \nusing both of them at the same time.\n\nBeginning in 1935, Laboratories engineers, \nwith the codperation of interested depart- \nments of several of the Associated Com- \npanies, treated experimentally a total of\n\n428 poles and posts. The preservatives used \nincluded coal-tar creosote and other coal tar \nproducts, sodium fluoride, sodium silico- \nfluoride, and proprietary pastes and solu- \ntions containing preservative compounds.\n\nPeriodic examinations and accumulated \nevidence during five years showed that treat- \nment at the ground line with sodium \nfluoride* and a mixture of creosote and coal \ntar is highly effective. This treatment is \nnow recommended Bell System practice for\n\nuntreated poles, provided they are good \nenough to remain in service until the next in- \nspection, usually three years later. The first \nstep is to distribute one pound of sodium \nfluoride against and around the pole at the \nbottom of the excavation, which has been \nopened for inspection purposes. The exca- \nvation is then filled loosely with soil to the\n\n*Sodium silicofluoride is being substituted for \nsodium fluoride because of wartime restrictions on the \nlatter material.\n\nExcavating around the butt of a telephone pole for ground line inspection and treatment with\n\npreservatives is shown at the head of this article. A pound of sodium fluoride is distributed\n\nagainst the exposed butt and in the bottom of the excavation, above left. The empty container ts\n\nburied with the preservative. The excavation is filled loosely to the ground level to cover the \nsodium fluoride as shown at the right\n\nA narrow trough is formed around the pole by pressing the loose soil away with the shovel to a \ndepth of about ten inches. From a gallon to a gallon and a half of creosote is poured against \nthe pole from a can which has a flattened spout to give a fan-shaped stream\n\nA pool of creosote surrounds the pole and saturates the soil. After a short interval the creosote- \nsaturated soil is pressed against the pole and back filling is completed\n\nground level and a trough is formed around \nthe pole by pressing the soil away from it toa \ndepth of about ten inches. Using a con- \nvenient container like an ordinary watering \ncan, with the sprinkler head removed and \nthe tip of the pouring spout flattened to \npour a fan-shaped stream, from one to one \nand a half gallons of creosote is poured \nagainst and around the pole at a height of \nabout eighteen inches above the ground \nlevel. Care is taken that the creosote enters \nany large checks in the pole surface. After a \nshort interval the creosote-saturated soil is \npressed against the pole, and back filling is \ncompleted. If the pole stands where creosote \nwould not be objectionable on the surface of \nthe soil, another shallow trough is formed \naround the pole and about one-half gallon of \ncreosote is poured against it as before. This \nlast application provides a collar of creosote- \nsaturated soil immediately surrounding the \npole from the surface of the ground down to \nthe depth of the original trough. Where \npoles stand in parked or grassed areas, the \nfinal application is omitted. This ground line \ntreatment has been adopted by several of \nthe operating companies, one of which, in \nCanada, has already treated about 258,000 \nnorthern white cedar poles out of some \n450,000 remaining in its lines.\n\nAccumulated evidence from experimental \nand operational treatments indicates that \nthe sodium fluoride and creosote treatment \neffectively reduces the rate of deterioration \nof poles in line and that their service life,\n\nTue Autuor: C. H. Amapon graduated from \nthe Biltmore Forest School in 1908 and then con- \ntinued in the practice \nof forest engineering \nuntil he joined the \nEngineering Depart- \nment of the Western \nElectric Company in \n1917. Since then he \nhas been engaged in \nestablishing standard \ninspection procedures \nand supervising 1n- \nspectors of the timber \nproducts used in out- \nside telephone plant. \nThis work was carried in on the Inspection \nEngineering Department until 1927 when it was \ntransferred to the newly organized Outside Plant \nDevelopment Department. Since that time, as 4 \nmember of the timber products group, Mr. \nAmadon has had a responsible part in setting \nstandards for timber products and in the de- \nvelopment of improved processes for the preser- \nvation of wood.\n\nas far as ground line condition is concerned, \nwill be increased by about six years. This ex- \ntension of pole life is alone considered suf- \nficient to justify the costs involved, but an \nadditional saving accrues from the possi- \nbility of placing pole line inspection on a six \ninstead of the three-year cycle, which has\n\nbeen the usual practice with untreated \npoles. The cumulative result of the ground \nline treatment is a reduction in labor and ex- \npense of pole line maintenance-inspection \nand in the need for new poles at a time when \nconservation of pole timber, labor and trans- \nportation are all important.\n\nJune 15 was a big day for a number of people in the Laboratories, \nwhen word was passed around that B-29 planes of the 20th Bomber \nCommand had raided Japan. That public announcement marked the \nculmination of an effort in Bell Telephone Laboratories which began \nin August of last year. The Laboratories was asked to undertake the \ndesign and start the production of certain electrical apparatus for use \non the new ships. Electrical design was assigned to High Frequency \nEngineering, headed by J. F. Wentz; equipment and mechanical \ndesign for manufacture to the group of equipment engineers headed by \nA. D. Knowlton. To R. H. Kreider\u2019s group in Trial Installation was \ngiven the production of a small number of units while Western Electric \nwas organizing for large-scale production. By dint of hard work, aided \nby close co\u00e9peration from the Air Force, the Laboratories quota was \nfilled around the first of the year. Part of the job was to train ten Air \nForce officers in the operation and maintenance of the equipment.\n\nHigh Frequency Engineering was formerly engaged in coaxial cable \ndevelopment. Early in 1941 it took up its first war project, and now is \ncompletely immersed in that work. It works closely with a group of \nequipment engineers who normally handled carrier and transmission \nsystems. Trial Installation, once concerned with building and installing \ntrial installations of telephone systems, made use of the skill and versa- \ntility of Western Electric installers which makes them ideal for building \nsmall numbers of new devices.\n\nOR many years cable splicers have \nJy vine lead cable-sheath by a soldering \nprocess called \u201c\u2018wiping.\u201d The name is \nan apt description of the operation. Molten \nsolder is poured over the sections to be \nunited until they are heated and thoroughly \ntinned. Then the semi-liquid mass which \nimmediately begins to form is manipulated \nby wiping with cloth pads into a well- \nrounded symmetrical joint. This requires \nconsiderable skill on the part of the splicer \nand close control of the composition of the \nsolder. Experience shows that even under \nthe best conditions fissures occasionally form \na path in the solder through which leakage \noccurs. In telephone cables, not maintained \nunder gas pressure, these leaks may permit \nthe entrance of water which, wetting the \npaper insulation, causes service interruptions. \nBy going to the extreme of wiping off all \nof the solder in excess of a fillet, many causes \nof porosity can be eliminated. The saving in \nsolder, and consequently in strategic tin, by \nthis operation is evident from Figure 2 \nwhich shows cross-sections of joints wiped \nby both methods. Over sixty per cent less \nsolder is required by the new joint and \nsplicers find it easier to make.\n\nFig. 1\u2014This conventional joint between a \ntelephone cable and sleeve requires a large \namount of solder\n\nThe success of the new method depends on \nthe characteristic behavior of lead-tin alloys \nin the process of freezing. When molten \nmetal is allowed to cool in a crucible with \nfree circulation of air it begins to freeze near \nthe walls of the vessel and, with a few ex- \nceptions, ends with a concave surface due \nto solidification shrinkage. When a lead-tin \nalloy solidifies, the first constituents to \ncrystallize form a rather porous cylinder \nwhich touches the crucible walls and ex- \ntends to a height corresponding to the \nvolume of the melt. On further cooling, tree- \nlike formations of lead-tin, called dendrites, \ngrow inward toward the center of the \ncrucible and at the same time many tiny \nnew crystals form throughout the liquid. \nThere are thus taking place simultaneously, \nshrinkage of metal as it becomes solid, \nshrinkage as it cools of the part previously \nfrozen and shrinkage of the remaining liquid \nas the temperature drops. The originally \nsolidified outer cylinder, adhering to the \ncrucible walls, remains essentially at its \noriginal height. The level of the semi-liquid \nportion nearer the center of the crucible con- \ntinuously falls until the precipitated crystal- \nlites, formed in the body of the melt, make a \nloosely piled mass extending from the bot- \ntom to the upper surface of the crucible. \nFurther shrinkage of the liquid then leaves \nthese primary crystallites at approximately \nthis level while the liquid recedes, leaving \nfissures between them.\n\nInsight into the solidification mechanism \nof wiping solder may be gained by casting \ntwo solder strips, one on a cold iron surface, \nand the other on a cloth-covered board and \nbending both to produce specimens as shown \nin Figure 3. The sample chilled by the tron \nwill exhibit fewer cracks from shrinkage \nthan the one slowly cooled on cloth. Primary \ncrystallites form throughout the slowly \nsolidifying mass and pack loosely to occupy \na volume which is greater than that of the\n\ncompletely solid solder. The sample cast on \nthe cold plate starts to freeze where it is in \ncontact with the cold iron and it continues to \nsolidify in a rapidly advancing smooth front \nuntil the last liquid at the top is solid. There \nis little opportunity for nucleation, i.e., \ncrystals spreading from point sources, and \nfor dendrite formation in the upper liquid\n\nbecause of the steep temperature gradient. \nThe surface of this melt is therefore smooth \nand free from the fissures which are caused \nin the slowly cooled sample by the shrinking \nof the liquid solder away from the dendrites. \nThis shrinking leaves a multitude of chan- \nnels which would cause leaks in a wiped \njoint if they occurred at the critical point. \n_ Another illustration of the processes tak- \ning place in joint wiping is the behavior of \nsolder which has been allowed to solidify in \na crucible until its surface is quite firm to a \nprobe. When tilted sideways to a position \nshown in Figure 4 a portion of the remaining \nliquid may be poured out leaving spongy \nregions in the solder. This loss of eutectic, \nl\u20ac., the constituent of lowest solidifying \npoint, is observed frequently while shaping \nthe former massive joints from which several \ndrops sometimes drain after the splicer has\n\ncompleted the shaping operation. It is also \nshown by the greater number of pores in the \ntop half of a joint compared with the bottom \nand by the grayer surface appearance of\n\nAlthough a solidification range in which \nquantities of liquid and solid metal may exist \nat equilibrium is an essential feature of a \nwiping solder, another factor of major im- \nportance is the nucleation rate of the alloy. \nWiping solders having high nucleation \nrates develop quickly a myriad of points or \nnuclei throughout the melt from which \nfurther crystallization proceeds. An alloy of \nlow nucleation rate, however, develops \nrelatively few of these points in the same \ntime and consequently grows fewer and \nlarger crystals. When subjected to wiping \nthe former has a texture similar to fine clay \nwhile the latter behaves like coarse sand \nand water. In the fine clay-like texture \nthere are more solid particles than in the \nwater-sand type and therefore more surface \nis available for the retention of the liquid in \nthe semi-solid mass. Drainage is thus greatly\n\nFig. 3\u2014Strips of solder bent over a rod one-half \ninch in diameter to illustrate the effect of vari- \nations in the cooling rate on the structure of \nwiping solders. The upper strip was chill cast \nand shows a sound ductile surface. The lower \none of the same solder was cooled slowly and \non bending exposed fissures between the \ncrystallites at the surface\n\nFig. 4\u2014An ingot of wiping solder which was \ntilted while in the crucible before becoming \ncompletely solidified. The lower lip is eutectic \ndrainage from the partially solidified mass\n\nretarded with the result that porosity is \nmaterially lessened. Texture determines in a \nlarge measure the ease of shaping and the \npotential porosity of a wiping solder.\n\nIn practice, the cable parts to be joined \nare cleaned and fluxed. Paper pasters are \nthen applied around the sheath and sleeve \nto restrict the spread of the solder which is \npoured from a ladle over the prepared parts. \nThe splicer catches the excess in a cloth held \nin contact with the bottom of the joint and \nrepeatedly pushes it back around the cable \nwith a wiping motion to aid \u201c\u2018tinning\u201d or \nalloying and to distribute the heat. After a \nfew such operations the prepared surfaces \nbecome thoroughly wetted by the solder.\n\nAt this stage a portion of the caught \nsolder is mixed in the ladle with more hot \nsolder and the mass, which now has a clay- \nlike consistency, is poured on the joint and \nmolded into place with cloth pads. When \nsolidification has proceeded until the solder \ncan support itself, manipulation is stopped. \nFrom this point on, loss of heat takes place \nby conduction from the joint through the \nsheath and sleeve, by radiation, and by air \nconvection currents at the surface of the\n\nsolder. As a result of these heat losses, \nsolidification occurs finally in the interior of \nthe mass near the important sheath-sleeve \njunction.\n\nThe action that causes pipes to form in \ncastings draws the eutectic from the critical \narea between the sheath and the end of the \nsleeve. If the solder has the proper charac- \nteristics there will be an outside layer which \ndoes not have interconnecting shrinkage \ncavities, drainage cavities or fissures due to \nthe wiping operation; and the finished joints \nwill be gas tight. If the solder is unduly \ncoarse, however, or has insufficient liquid \neutectic at the time the mass becomes too \nrigid to manipulate further, the joints may \nleak.\n\nThe technique of fillet wiping is similar to \nthe former procedure until the splicer \nshapes the mass. At this point he wipes the \nsolder to a small fillet similar to that shown \nin Figure 2. The joint thus formed takes \nmuch less solder and therefore has less ten- \ndency to shrink and to draw eutectic from \nthe space between the sheath and sleeve. \nAt the temperature at which wiping is dis- \ncontinued there is also insufficient solder left \nto permit drainage drops to accumulate and \nfall from the bottom of the joint. Thermal \nconduction along sheath and sleeve causes \nrapid solidification of the solder at the joint, \nthus eliminating the possibility of drainage. \nExperience has shown a consistently high \npercentage of sound joints when fillet wiping \nis rigidly practiced. During the development \nperiod of this process the few fillet joints \nthat leaked were found to have been made \nwith an excess of solder. Under the micro- \nscope they showed sponginess where the \neutectic had been drawn away from the \njunction during solidification.\n\nPhysical tests on fillet joints between \nsections of a telephone cable and its sleeve, \nwhich are similar in size to that shown in \nFigure 2b, have demonstrated that these \njoints are stronger in tensile strength and \nshow less creep and fatigue than the cable \nsheath itself. The composition of the \nsolder used was 38 per cent tin, 0.1 per cent \narsenic and the balance lead.\n\nApplication of the new technique has \ngone much further toward saving tin than \nany known permissible change in the com- \nposition of wiping solder. Using the former\n\ntechnique, a reduction of only one per cent \nin the nominal tin content of a lead-tin \nsolder caused leaky joints which indicated \nthat little tin could be saved by a simple \nchange in specification. This was expected \nbecause many studies had been conducted \nby the Laboratories with the aim of reducing \nthe tin content in wiping solders to the \nminimum consistent with the production of \nsatisfactory joints. Tin has always been \nmuch more expensive than lead and for \nlarge users of solder a reduction of even \none per cent in the tin content would result \nin substantial savings.\n\nOther avenues for conserving this stra- \ntegic metal are the substitution of ternary \nand quaternary alloys containing less tin \nthan that required by the binary lead-tin \nsolders. A satisfactory alloy of this type \nwhich contains 13 per cent tin, 23 per cent \nbismuth, 0.1 per cent arsenic and the \nbalance lead has been developed by the \nLaboratories. Though readily obtainable a\n\nshort time ago, bismuth is now unavailable \nfor solders. A new wiping solder, however, \nhas been introduced into service. Its tin \ncontent is reduced to 32 per cent and it \ncontains, in addition to lead, 2 per cent \nantimony and o.1 per cent arsenic. This \nmaterial appears suitable for fillet wiping \nalthough it requires more skill in use than \nwiping solder made of 38 per cent tin and \n0.1 per cent arsenic with the balance lead. \nOther compositions which contain less than \nthe normal amount of tin may be usable \nbut, on the whole, the savings accomplished \nby composition changes would be small com- \npared to those which have been obtained \nby the new wiping technique.\n\nThe small amount of solder required for \nthe fillet wipe not only reduces tin consump- \ntion but it produces joints less liable to \nleakage than the conventional type. This \nsuccess is based on sound metallurgical prin- \nciples and promises to survive the period of \nrestricted tin consumption.\n\nchemical Engineering at Pennsylvania State College in 1930 \nand an MLS. in Metallurgy at Columbia University in 1939. He \njoined the Laboratories in 1930. Since then he has been engaged\n\nOR the convenience of patrons at public \ntelephone stations where booths cannot \nconveniently be installed, a shelf and mount- \ning unit has been developed by the Labora-\n\nstations in locations where the usual booths \ncannot be conveniently used\n\ntories. It is made entirely of wood, to avoid \nthe use of strategic materials. Besides a coin \ncollector, it supports a suspended directory \nand a broad shelf for writing and placing \nhand bags or parcels.\n\nThe coin collector is located at the center \nof the mounting board at an angle of 30 de- \ngrees with the wall. The directory is en- \nclosed in a hard-covered binder which is \nsuspended in a niche adjacent to the coin \ncollector. Concealed behind the directory, \nthe ringer is in a compartment which has an \nopening in front to allow emission of the \nsound. A wooden card frame is provided on \nthe right side of the mounting for display \nand instruction cards.\n\nThis mounting unitis installed by attaching \nit with machine screws to nuts in a wooden \nbackboard, which is fastened to the wall. Wir- \ning the apparatus is facilitated by providing \na channel in the rear of the backboard.\n\nThese coin collector mountings may be \ninstalled singly or in groups. When in groups \nthe directory compartments provide parti- \ntions between adjacent stations.", "title": "Bell Laboratories Record  1944-07: Vol 22 Iss 11", "trim_reasons": [], "year": 1944}
{"archive_ref": "sim_record-at-t-bell-laboratories_1946-09_24_9", "canonical_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1946-09_24_9", "char_count": 92766, "collection": "archive-org-bell-labs", "doc_id": 834, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc834", "record_count": 104, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1946-09_24_9", "split": "test", "text": "PAGE \nAirborne Search Radar, Cook. . . . . . . \nTransoceanic Radio Amplifier, C. F. Rose. . . . . 826 \nProduction of Airplane Radar Speeded\n\nby New Testing Technique, F. P. Wight . . . . . . . 3830 \nRepresenting the Laboratories at the Front, C. B. Barnard, Jr. .  . 385 \nThe Ballad of the Cedar Ants of Penetang, R. H. Colley . . . . 341\n\nThe Cover\u2014\u201cCloverleaf\u201d antenna for FM broadcasting stations \noperating at the new carrier frequencies between 88 and 108 \nmegacycles and at power levels up to and including 50 kilowatts.\n\nPublished monthly by BELL TELEPHONE LABORATORIES, INCORPORATED \n463 West St., New York 14, N. Y.\n\nO. E. Buckley, President J. W. Farrell, Secretary W.C. Burger, Treasurer \nPaul B. Findley, Editor; Philip C. Jones, Science Editor; R. Linsley Shepherd, \nAssociate Editor; Helen O. McLoughlin, Assistant Editor; Leah E. Smith, \nCirculation Manager, Subscriptions, $2.00 per year. Printed in U. S. A.\n\nstage of advancement in the few \nshort years devoted to its study. At the \nwars end, a great number of different \nequipments suited to specific military needs \nwere in use. In addition to numerous land- \nbased and shipboard radar sets, many \nequipments were designed for operation in \nmilitary aircraft, where they performed \nfunctions ranging from simple location of \nenemy planes or ships to precise spotting \nof military targets for the controlled re- \nlease of bombs and other projectiles.\n\nBy early in 1942, aircraft radar equip- \nments had been developed that operated \non wavelengths as short as 9 cm _ but \nweighed 400 pounds or more. Reasonable\n\nThe photograph at the top of the page shows an \nAN/APS-4 in action. White segment indicates area \nof scanning.\n\nangular discrimination with waves of this \nlength required antenna apertures of 30 \ninches or greater. This large exposed size \nand heavy weight restricted the use of mi- \ncrowave radar to large airplanes and \nblimps, leaving the necessarily smaller air- \ncraft used on Navy carriers without the \nadvantages of the microwave system.\n\nIt was at this time that the Laboratories, \nat the request of the Navy Bureaus of Aero- \nnautics and Ships, undertook an accelerated \ndevelopment program leading to the manu- \nfacture of the ASH, later coded AN/APS-4, \nradar equipment for use on carrier-based \naircraft. The successful development of this \nlight and compact design by the Labora- \ntories, and the subsequent manufacture of \nnearly 15,000 equipments and auxiliary \nspare parts by the Western Electric Com- \npany, constituted a major contribution to \nthe war effort.\n\nThe model AN/APS-4 radar equipment \nseen in Figure 1 comprises a streamlined, \npressure-tight, bomb-shaped unit weighing \napproximately 130 pounds, and six small \nunits: junction box, control unit, and two \ncathode-ray tube indicators, each with its \nassociated amplifier. These small units, \nwhich together weigh approximately 20 \npounds, are mounted -inside the aircraft.\n\nThe large unit containing antenna, power \ncircuits, transmitter and receiver hangs on \na standard bomb rack under the wing of \nthe aircraft. It looks like a bomb, and like \na bomb it can be jettisoned by the pilot to \nprevent its capture by the enemy, or to \nlighten the plane when it is necessary to re- \ngain top-flight performance of the airplane.\n\nFig. 1-The main components of the AN/APS-4 aircraft radar equipment \nSeptember 1946\n\nwavelength of 3.2 cm, provides a radio \nbeam less than six degrees wide with an \nantenna aperture of only fourteen inches. \nIn operation, the antenna, in the radio- \ntransparent nose, or radome, of the sealed \nbomb unit, scans horizontally to cover a \nforward sector 150 degrees wide at a fre- \nquency of from one to two times a second. \nWhen searching for objects on the surface\n\nof sea or land, the elevation angle of the \nradio beam is changed on alternate scans \nto trace out a two-line pattern\u2014thus cover- \ning a band twelve degrees wide, six de- \ngrees for each line of scanning\u2014but when \nsearching for another aircraft, a four-line \npattern is used, giving a band twenty-four \ndegrees wide. At the same time the tilt of \nthe scanning radio beam may be smoothly \ncontrolled by the pilot or operator to give \na range of view extending from about \nthirty degrees above to about thirty degrees \nbelow the searching aircraft.\n\nThe equipment normally generates pulses \nof radio power at levels of 30-40 kw lasting \ntwo-thirds of a microsecond. During trans- \nmission the receiving circuits are blocked \nby a gaseous discharge switch, but within \na few microseconds after cessation of trans- \nmission, the receiver regains full sensitivity \nto radar echoes from objects in the antenna \nbeam. These echoes, detected and ampli- \nfied, are used to modulate the intensity of \nthe cathode-ray tube indicator producing a \nspot of light which persists for some seconds. \nA type \u201cB\u201d indicator which presents the \nrange or distance of an echo and its angular \nposition on a horizontal plane is used. This \nis illustrated in Figure 2. The spot-forming \nbeam in the cathode-ray tube is deflected \nvertically to depict the range of the echo, \nand moves horizontally across the face of\n\nFig. 3\u2014Left, representation of a region as it would appear in Mercator projection. Right, \nthe same region as it would appear on the scope of an AN/APS-4\n\nthe indicator screen in proportion to the \nhorizontal angular motion of the antenna to \nshow the bearing of the echo. The resulting \nscreen picture is a distorted map of the \narea being scanned. In an ordinary map, \nsuch as shown at the left of Figure 3, \ndistance north and south is represented \nvertically and distance east and west, hori- \nzontally, while the bearing of any point \nfrom the middle of the base line is meas- \nured directly by the angle between the \nbase and the line connecting the two points.\n\nThe representation of this same area on a \nB scope in a plane located at the middle \nof the base line would appear as shown at \nthe right of Figure 3. Here the distance to \nany point is measured vertically by the \nscale at the left, while the bearing to any \npoint is measured horizontally by the scale \nalong the lower edge. Although providing \na distortion of contours, such a presenta- \ntion has the advantage for military use that \nthe bearing angle discrimination does not \ndeteriorate at close range.\n\nThe AN/APS-4 may also be used to de- \ntermine the relative elevation of another \naircraft so that it can be intercepted in \nflight. For this use the indicator screen \nshows two spots of light, Figure 4, sepa- \nrated horizontally a short distance and dis- \nplaced in such a way that the angle be-\n\ntween the line connecting the two dots and \na horizontal line through the left-hand one \nindicates the angle of the target plane \nabove or below the scanning plane. The \nleft-hand dot represents the position of the \ntarget plane\u2014range and bearing being \ntaken from the two scales of the scope.\n\nThe AN/APS-4 radar equipment can \nalso provide information as to the identity \nof target aircraft or ships by presenting a \ncoded signal on the indicator screen when \nthe target is interrogated by and responds \nto associated I.F.F. (Identification\u2014Friend \nor Foe) equipment. For navigational pur- \nposes the AN/APS-4 is equipped to inter- \nrogate microwave beacon stations also and \nto display coded responses from them \nshowing their location and identity at \nranges approaching 150 miles.\n\nComplete records of the maintenance \nhistory of this equipment have been made \navailable to the Laboratories. This infor- \nmation has served as a guide for the many \nchanges that have been introduced during \nthe course of production. More than 2,700 \nchange orders were processed during the \ntwenty-three months in which approxi- \nmately 15,000 equipments were delivered.\n\nThe matter of weight, always important \nin aircraft operations because each pound \nof equipment displaces an appreciable \namount of fuel with corresponding lessen- \ning of flight range, is of particular impor- \ntance in carrier-based aircraft because of \ntheir small size and restricted take-off \nfacilities. A strenuous and continued ef- \nfort to reduce unnecessary mass in the \nAN/APS-4 resulted in a reduction of total \nweight from 180 pounds for the first pro- \nduction equipment to 150 pounds for the \nfinal equipments manufactured.\n\nThe operational record of the AN/APS-4 \nhas not yet been published, but scattered \nreports reaching the Laboratories indicate \nthat it has been eminently satisfactory. \nSeveral interesting testimonials to the use- \nfulness of the equipment are quoted below \nfrom the Airborne Cod\u00e9rdinating Group \nDigest, a Navy periodical devoted to radio \nand radar maintenance.\n\n\u201cThe ASH radar works well with the \nmany jobs we are called upon to do,\u201d Lt. \nSperry wrote, \u201cand those jobs included not \nonly searching out Japanese targets, navi-\n\ngating in strange territory and homing to \ntheir carrier, but also helping the pilots fly \n\u2018in, over and around\u2019 tropical rain squalls \nand point out targets through overcast.\u201d\n\n\u201cOur surprise attack out of the sun was \nhighly successful with many bomb and tor- \npedo hits, . . .\u201d \u201cWe retired from rendez- \nvous, leaving the target burning and dead \nin the water.\u201d\n\n\u201cThe weather between us and the CV \nwas steadily becoming worse, but by using \nthe radar scope in conjunction with my \nmaps, we steered in, over and around the \nrain squalls and unfamiliar islands to the \nouter end of the San Bernardino Straits, \nwhich we easily identified by radar.\u201d\n\nA strike group \u201cflew through heavy \nweather to Karenko on Formosa, where the \nflat area around the town was easily dis- \ntinguished from the surrounding mountains \nand identified by the strike leader\u2019s ASH \nradar. We let down through a hole in the \nclouds to sink and burn shipping in the \nharbor and blast nearby factories.\u201d\n\n\u201cThe next morning the strike group \nclimbed above the heavy weather on the \neast side of Taiwan and flew up the coast \nline on the seaward side of the large port \nof Kiirun. We used ASH radar to home in \non prominent harbor landmarks and ap- \nproached Kiirun over the solid overcast at \n10,000-foot altitude.\u201d\n\nThe division led a strike on Tarien air \nstrip on the northwest tip of Formosa. \u201cWe \napproached the island above the overcast \nand were well orientated through use of \nASH radar. The position of our target field \nwas plotted by bearing and distance from\n\nthe distinct coastline and by the nearness \nof a winding river. . . . We made a sur- \nprising and successful attack, doing much \ndamage to enemy airfield installations.\u201d\n\nDevelopment of the AN/APS-4 radar \nequipment required the well-co\u00e9rdinated \nactivity of people from many departments \nof the Laboratories. Proper recognition of \nthese efforts would require listing several \nhundred important contributors. As with \nmost of our developments, no one person \nand seldom one group carries the entire load.\n\nTue AutHor: Jackson H. Cook, after re- \nceiving an S.B. degree in Electrical Engineer- \ning from M.I.T. in \n1936, took graduate \nstudy at Harvard \nand M.I.T. during \nthe following two \nyears. He was Re- \nsearch Assistant with \nthe Electrical Engi- \nneering Department \nof M.I.T. from 1937 \nto 1940, Consultant \nto the NDRC dur- \ning 1940 and 1941, \nand Research Asso- \nciate at the Radiation Laboratory from 1940 \nto 1944 with the Modulator and Engineering \ngroups. He came on leave to the Laboratories \nin 1941, where at Whippany he engaged in \nradar development. During 1942 and 1943, on \nleave to the Navy Department, he was Senior \nEngineer (radio) in charge of all three-centi- \nmeter aircraft radar design. He returned to \nthe Laboratories in 1943 to become Project \nEngineer on the AN/APS-4 radar described \nin this issue of the REcorp.\n\non January 7, 1927, the rate for a \nthree-minute call was seventy-five dollars. \nToday the corresponding rate is twelve dol- \nlars. Continual research, development, and \nredesign, coupled with good management, \nare the chief factors which have made such \na reduction possible.\n\nThe initial transatlantic service was by \nlong waves\u201452 kc in one direction and 57 \nke in the other. While long-wave _trans- \nmission was still under development, stud- \nies of short-wave transmission were being \ncarried on at Deal, N. J., and commer- \ncial short-wave service was inaugurated \nin 1928.* By 1935, when a transpacific link \njoined telephones of the United States with \nthose of Japan, the system had expanded \nfrom transatlantic to transoceanic in scope. \nThe overseas traffic was carried by ten \nshort-wave transmitters at Lawrenceville \nand Ocean Gate, N. J., and at Dixon, Calif.\n\nExperiments were continued at Deal, \nhowever, to improve the radio transmitting \nequipment, and from these efforts came a \nsingle-sideband transmitter to replace the \ndouble-sideband transmitter. Single-side- \nband transmission offers the equivalent of \nabout eight times the useful power pro-\n\nvided by double-sideband trans- \nmission. In 1938 newly developed \nequipment was installed provid- \ning either double or single-side- \nband transmission as desired to \nmeet the alternative demands \nimposed by both types of equip- \nment at the overseas stations.\n\nIn 1939, additional newly de- \nsigned low-power radio trans- \nmitters were installed to meet \nthe continually increasing traffic \ndemands. The new design pro- \nvided twin-channel single-side- \nband transmission. This type of \ntransmission conserves space in \nthe radio spectrum by transmitting the \nintelligence of two channels in three- \nquarters of the space required by double- \nsideband transmission. The two channels \nare associated with the same carrier but \none is separated from it by a region \nwide enough to avoid inter-channel inter- \nference. This low-power equipment be- \ncame the Western Electric D-156000 Radio \nTransmitter.* It not only embodied the new \nfeatures mentioned above, but also mini- \nmized the operations required for changing \nfrequencies that were made necessary by \ndiurnal and seasonal variations in the trans- \nmission paths.\n\nThe transmitter consisted of a 2-kw unit \nthat could be used either as a short haul \ntransmitter or to drive a 60-kw two-stage \namplifier for transoceanic service. The two- \nstage amplifier did not lend itself readily \nto rapid frequency changing so a further \ndevelopment program was initiated at Deal \nto provide an improved amplifier to be \nused with the basic 2-kw unit. This pro- \ngram culminated in the Western Electric \nD-158974 Radio Amplifier, which was \nturned over to traffic at Lawrenceville in \nJune, 1942.\n\nFeaturing facilities for rapidly changing \nits frequency, this transmitter is capable ot\n\noperating at any assigned frequency within \nthe range from 4.5 to 23 me. Furthermore, \nit must stand the daily grueling test of con- \ntinuous operation for twenty-two hours out \nof every twenty-four. Safety features have \nalso been stressed in the design, and the \nsystem prevents personnel from entering \nany compartment until all voltages have \nbeen removed and circuits grounded. For \nprotecting the equipment, provision is made \nso that the various voltages can be applied \nonly in a specified sequence.\n\nIn contrast with its predecessor, the am- \nplifier itself has only a single stage. All \ncontrols are brought to the front of the \nunit, thus making it unnecessary to gain \nentrance to the unit to change frequencies. \nSimplicity has been further secured by re- \nducing as much as possible the amount of \nrotating machinery required. Another im- \nprovement is the location of much of the \nassociated power supply out of doors. A \nsingle-stage construction was made possible \nby employing four 25-kw tubes in a single \ncircuit instead of eight 10-kw tubes em- \nployed in a two-stage amplifier. In addition \nto the amplifier, there is a control unit and \na high-voltage rectifier unit. The three units \nare bolted together and all mounted on a \ncommon channel-iron base as shown in the\n\nheadpiece, where the control unit is at the \nleft and the rectifier at the right. The space \noccupied measures 4 feet by 15 feet on the \nfloor and is 7 feet high. This represents \nonly one-third of the equivalent floor space \nrequired for previous comparable equip- \nment. All electrical connections, except \nthose to the antenna transmission line, enter \nthrough the floor. The two radio-frequency \nantenna leads pass through a pyrex glass \npanel on the top of the amplifier to an \nexternal antenna selector switch. Inter- \ncabinet wiring is run in conduit through \nthe sides of the cabinets.\n\nA schematic of the amplifier is given in \nFigure 1. Four Western Electric type 340-A \nsingle-ended, 25-kw water-cooled vacuum \ntubes are connected in a push-pull bridge- \nneutralized circuit with the two tubes on \neach side of the circuit connected in paral- \nlel. The operating characteristics for these \ntubes are similar to those of the vacuum \ntubes used in the early models of the West- \nern Electric 50-kw broadcast transmitters. \nExcessive heating of the copper-glass grid \nseal, due to high-frequency losses, was \neliminated by air cooling the seal.\n\nFrequency changing through front-of- \npanel controls was accomplished by the \ndesign of the two output tuning coils illus-\n\nFig. 2\u2014View of part of the amplifier show- \ning the remote controlled tuning coil in \nthe lower center section\n\ntrated in Figure 2. The unused end of each \nvariable coil is short circuited through a \nsliding contact moving on the inner periph- \nery of the coil to prevent parasitic voltages \nfrom being developed when only a few \nturns are actively in the circuit. An addi- \ntional device for short circuiting the unused \nportion of the coils, shown in the front of \nthe photograph, is necessary to overcome \nother undesirable resonances within the\n\noperating range. This is a solenoid operated \ndevice automatically controlled by a limit- \ning switch actuated mechanically when the \nsliding contact passes a preselected point \non the coil.\n\nSince these coils have to carry 150 am- \nperes of radio-frequency current, they are \nwater cooled. The coils consist of eleven \nturns of %-inch outside diameter copper \ntubing formed to a diameter of 9.5 inches. \nAbout one gallon of water per minute is \nrequired to cool each coil. This water is by- \npassed from the cooling supply of the \nvacuum tubes, and if the inlet connection \nwere made at one end of the tuning coil \nand the outlet connection at the other, the \ncoils would be short circuited by the metal- \nlic jacket of the tubes. This is avoided by \nusing a double tube construction for the \ntuning coils with a smaller tube inside the \nlarger\u2014with the outer tube being connected \nto the inlet supply and the inner tube to \nthe outlet. In this way the tuning coil itself \nis metallically connected to the water \nsupply at only one end and is thus not \nshort circuited. By the use of special fittings \nin the two ends of the tuning coil, water \nenters the annular space between the tubes \nand after passing through the entire length \nof the coil it enters the inner tube, passes \nback up through the length of the coil and \nthen out.\n\nA radial sliding contact is rotated on an \naxial shaft and the thread action of the \nturns moves it to the desired tuning point. \nDifferential gearing permits simultaneous \nadjustment of both coils through an in- \nsulated shaft connected to a large hand- \nwheel on the front of the cabinet. When \nthe hand-wheel is pulled out, the shaft to \nthe rear coil is disengaged, and the front \ncoil may be adjusted independently. This \nadjustment allows compensation for dis- \nsymmetries which may develop within the \namplifier itself.\n\nWith the former two-stage amplifier, two \nmen required six minutes to enter the am- \nplifier units and change from one frequency \nassignment to another. With the niceties of \nthese controls, one man now can make the \nchange externally in three minutes. This \narrangement represents not only a saving \nin the cost of operation, but also a de- \ncrease in lost-circuit time.\n\nRotating machinery has been reduced to \na minimum by replacing motor generators \nwith rectifiers. Only pumps and fans for \nthe water-cooling system are required. This \nsystem is simplified since the previous \nwater-cooled type rectifier tubes are re- \nplaced by air-cooled mercury-vapor tubes. \nThe cooling system, similar in principle to \nan automobile radiator and fan, is located \nin the building, and is normally unattended. \nA predetermined water temperature must \nbe attained before the cooling fans operate. \nThereafter, thermostats automatically con- \ntrol the number of fans required in accord- \nance with the ambient temperature and the \npower dissipated.\n\nAs much of the power supply equipment \nas possible is located out of doors, and is \nillustrated in Figure 3. On the platform, \nthere are the necessary pieces of equipment \nto transform the three-phase voltage of \n4,160 to 230 volts for the low-voltage equip- \nment and to 13,000 volts for the three- \nphase high-voltage rectifier equipment. The \nlatter voltage is controlled by an automatic \nthree-phase induction voltage regulator. \nThe automatic features may also be dis- \nengaged so that the rectifier voltage may \nbe raised and lowered manually. The in- \nstallation is novel to the extent that the\n\nTHe AutTHor: CHaRLEs F. P. Rose entered \nthe Radio Research Department of the West- \nern Electric Com- \npany in December, \n1920. He graduated \nas a Student Assist- \nant in 1924, and sub- \nsequently attended \nColumbia University \nExtension School. \nSince 1925, he has \nserved as a Member \nof the Technical \nStaff of the Bell \nTelephone Labora- \ntories. Mr. Rose has \nbeen engaged in designing, developing, instal- \nling and testing short-wave transoceanic radio \ntelephone transmitters for the Bell System in \nNew Jersey, Argentina and California. During \nthe war he was engaged in developing special \nelectronic equipment used by the U. S. Gov- \nernment. In November, 1944, he was trans- \nferred from the Radio Research Department to \nthe Systems Development Department, where \nhe is continuing circuit development work for \nradio systems.\n\nvalue of voltage at which automatic regu- \nlation takes place can be changed over a \nwide range from a remote point located in \nthe operating room of the station.\n\nNE of the important steps in pre- \nparing to manufacture complex \nelectrical apparatus is the provision \nof shop testing facilities. These consist of \nspecial testing units, some of them very \ncomplicated pieces of apparatus in their \nown right, that at various points along the \nproduction lines are brought into action to \ncheck wiring and circuit connections, the \nfunctioning of the various component parts, \nand the overall operation. The design and \nconstruction of this testing apparatus is \nusually undertaken by the Western Electric \nCompany, and requires considerable time \n\u2014often to the extent of affecting the deliv- \nery of the first commercial units. When the \nLaboratories pre-production program* was \nbegun, this same procedure was at first re- \ntained, and the pre-production units were \ntested with Laboratories\u2019 equipment and \nset-ups devised temporarily to meet the \nneeds of the moment. Such a procedure, \nhowever, meant that design engineers were \nrequired to spend too much time on pre- \nproduction testing, and that because of the \nshortage of manpower, Western Electric \nproduction might be delayed in starting. \nWhen one of the most urgent radar proj- \nects came along, it seemed essential to de-\n\nvise some better method of testing. This \nwas a high-quality radar system designed \nfor pin-point bombing from B-29\u2019s, with Ja- \npan as the target. Delivery time was impor- \ntant, and since it was an elaborate precision \njob, testing was even more important than \nusual, and the requirements were more se- \nvere. As a result, the engineers responsible \nfor the development and planning of this \nequipment in the Western Electric Com- \npany and the Laboratories agreed to design \nthe shop test sets concurrently with the de- \nvelopment of the system, and to build these \ntest sets in the laboratory along with the \npre-production models. This plan was car- \nried out on this project, and the test sets \nwere proven in on the laboratory pre-pro- \nduction models, and then used in the fac- \ntory to test the first regular production lots.\n\nThe radar set is made up of a number of \nelectrical and mechanical components and \nsub-assemblies so as to meet space and \nweight requirements for plane mounting, \nto facilitate replacements and maintenance, \nand to permit manufacturing, assembling \nand testing by production line methods. \nEarly analysis of the project for factory test \npurposes indicated that testing would be \nrequired at forty-one positions, and that \ntwenty-eight distinct types of mechanical\n\nand electrical test equipment would be re- \nquired. These test facilities were as special \nand in some ways as complex as the radar \nequipment they were designed to test.\n\nIn general, these tests require the deter- \nmination of correctness of wave shapes, \namplitudes, frequencies and band widths, \nand checks of general circuit functioning, \nand of mechanical alignments and fitness. \nIn designing test sets, jigs and accessories, \nit was desirable to centralize all controls and \ntest accessories in the face of or attached \nto the sides of floor and bench-mounted \ncabinets, and to locate the apparatus con- \nveniently for operation and maintenance. \nProvisions also had to be made for rapidly \nconnecting the unit to be tested to the test \nset without using soldering operations or \notherwise attaching or detaching wires, and \nfor providing electrical and mechanical \nsafety features, easy attachment to power \nservices, mobility of all floor-mounted units, \nand overall durability of all components so \nas to stand the punishment of continuous \nuse over long periods and to avoid inter-\n\nBeginning in April, 1944, general testing \ntheory and proposed testing practices were \nworked out as rapidly as possible into spe- \ncific circuit and test requirement data. \nThen with the aid of Western test planning \nengineers, specific testing facilities were de- \nsigned, complete manufacturing drawings \nwere prepared, and one or more sets of \neach type were made available for pre- \nproduction testing at the Laboratories, and \nalso for the Western\u2019s first production. \nWestern Electric testing personnel were \nbrought in to carry on the pre-production \ntesting so as to train them for the produc- \ntion job to follow, and to relieve Labora- \ntories personnel (from repetitive testing \noperations ).\n\nFor this particular project, the Labora- \ntories built and tested forty pre-production \nsystems at New York, and eighty associated \nantennas were built by Ex-Cell-O Corpora- \ntion at Detroit on a Laboratories sub-con- \ntract, the Laboratories furnishing all test \nfacilities and being responsible for all testing\n\nThe tube assembly test set includes a reference cathode-ray tube at the upper right and the \ntube under test at the upper left. The mount for these tubes is hinged so that the tubes \nmay be viewed in their normal position, as at the right, or as at the left, tilted down \nto give safe access for adjustments while the tube is in operation. In the latter position \nthe tubes may be seen in the sloping mirror mounted between the tubes and the set\n\nOne of the tests carried out in Detroit used a test transmitter, shown at the left, which\n\nwas mounted on top of the Michigan Bell Telephone Company's building in Detroit, \nand an antenna dispersion and pattern test set at the right, mounted on the roof of the \nbuilding of the Ex-Cell-O Corporation\n\nRadar test facilities at New York were \nready for use in June, 1944, and were used \nfrom July to September in the component \nand final testing of the pre-production \nquota. System deliveries to the Signal \nCorps began late in July, and the required \nforty systems were completed early in Sep- \ntember, 1944\u2014one month ahead of schedule.\n\nAntenna test facilities at Detroit were \navailable in June, 1944, and were used from\n\nThe size and complexity of some of the testing \nunits are evident from this photograph of the\n\nJune to the middle of October in the com- \nponent and final testing of the antenna pre- \nproduction quota. Antenna deliveries to the \nSignal Corps started about the middle of \nJuly and were completed in October, 1944.\n\nIn all, a total of fifty-five test sets of spe- \ncial new types were built by the Labora- \ntories for use on pre-production quantities \nand to provide the Western with one of \neach type of set initially. Of these, twenty- \nsix were for use in New York, fifteen for \nuse in Detroit and fourteen for the \nWestern Electric Company. On \ncompletion of the pre-production \ncontract, however, all test sets were \nturned over to the Western for use \non the production contract except \nfor a small number having applica- \ntion on other current work.\n\nPerhaps the most unusual feature \nof the various test sets is the special \njig developed to facilitate testing \nsub-chassis assemblies. These jigs \nare arranged to provide instantane- \nous contact between the test set and \nunit under test by means of flexible \naction fingers aligned to engage the \nterminal posts of the sub-chassis. In \naddition, provision had to be made \nfor sufficient clearances to permit \ntest probing, and a double-hinge ar- \nrangement had to be provided to\n\nallow access to the equipment side \nof a chassis for potentiometer ad- \njustments and tube replacements \nwithout disengagement from the \ntest set circuits. As there were six \nsuch chassis per system, having an \naverage of twenty-one test termi- \naverage of twenty-one test termi- \nset connections were involved in \ntesting forty pre-production sys- \ntems, not including retests and Sig- \nnal Corps inspections. The use of \nthese jigs proved to be of real value \nin saving time. Perhaps of even \ngreater importance was the avoid- \nance of potential trouble due to \ndamaged soldered connections, \nwhich might otherwise have oc- \ncurred through use of alligator clips \nor actual soldering of the numerous \ntest leads to the terminals of the units.\n\nThe test sets required power supplies of \na range of voltage and frequency not nor- \nmally available from building sources. For \ntests of radar components at New York, \ntwenty-four power supply units of six dif- \nferent ranges were required. Testing of the \nantennas at Detroit required fifteen recti- \nfier and motor generator set units of ap- \nproximately similar range.\n\nThe general plan of designing shop test \nsets concurrently with the development of \nthe system, initiated with this project, was\n\nfollowed currently for other war projects. \nIt was felt that a forward-looking test pro- \ngram of this character would contribute \nmaterially to the overall war effort by \ngreatly speeding up both pre-production \nand production programs.\n\nThe pre-production program of forty sys- \ntems was completed a month ahead of \nschedule, due in a large measure to the \navailability of adequate testing facilities \nand a group of testing personnel trained in \nproduction methods. Of probably greater \nimportance, the Western Electric\u2019s produc-\n\nThe seven test sets below and the six shown in the photograph at the head of this \narticle are representative of the twenty-eight different types of units employed\n\nTue Autor: F. P. Wicur first joined the \nEquipment Drafting Department of the West- \nern Electric Company in 1926 \nwhere, in addition to preparing \ndrawings, he was engaged part \ntime with actual field installa- \ntions. After a leave of absence to \nattend Bowling Green State Uni- \nversity, he transferred to the Sys- \ntem\u2019s Drafting Department, where \nhe engaged in the preparation of \nequipment requirement drawings \nfor toll, radio and central-office \nfacilities. This was followed by \nsuccessive transfers to the trial \ninstallations, current analyzation\n\nand switching development groups, where his \nwork involved supervision of construction and \nsubsequent field installations of \ntype-K carrier and_ telegraph \nequipment models, the handling \nof current central-office equip- \nment and installation problems, \nand finally development of test \nfacilities for the No. 1, tandem \nand No. 4 crossbar systems. Dur- \ning the war he was engaged in the \ndesign and construction of facili- \nties for shop testing of pre-produc- \ntion models for the Armed Forces \nand took an active part in design- \ning the devices described here.\n\ntion program was helped in many ways. \nTheir test planners, who are familiar with \nproblems peculiar to assembly line produc- \ntion and with the use of comparatively un- \nskilled test personnel, assisted in test set de- \nsign, and became familiar with the product \nand the technical problems in advance of \nstart of production. Duplicate effort was \navoided, and one type of test set served\n\nboth pre-production and production. Also, \nthe test sets used, which were of a type \ndefinitely suited to production line output, \nbecame available when needed to meet \nproduction rates and, finally, a group of \ntheir own test personnel, trained in the \nproblems of pre-production testing, became \navailable as a nucleus to test the first produc- \ntion output and to train additional testers.\n\nTypical of the jig sets is this navigator\u2019s indicator chassis test set. \nThe chassis is mounted face down on a supporting frame and \nthe jig is placed on top of it. Hinges at one end of the mounting \npermit the chassis to be tipped out for adjustment or inspection\n\nconverting physicists and engineers \nof Bell Telephone Laboratories into globe- \ntrotting Messrs. Fixits deserves some con- \nsideration. Eighty of them were pried loose \nfrom the peace and security of their bell \njars and glowing vacuum tubes by the \nmodern psychological alchemy of their \ncountry\u2019s need, during the struggle and \nimmediately afterward, to become inves- \ntigators, consultants, observers, and field \nengineers for the Armed Forces.\n\nFrom Manila to Munich, from Algiers \nto Alaska, from Brisbane to Brussels, wher- \never there was a technical investigation or \na concentration of communications or of \nelectronic detection, gun-laying, or bomb- \ning equipment, their trails were likely to \nconverge. Some of them speak familiarly \n(and with distaste) of Kiska and Panama. \n\u201cI wasn\u2019t really overseas,\u201d they'll say. \u201cI \nwas in the Aleutians.\u201d\n\nThe diaries of others carry entries such \nas: . . . arrived Hollandia, New Guinea, \nMarch 9, 1945 . . . left for Manila by boat \nApril 26 . . .\u201d; or, \u201cMany of the V-bomb \nlaunching sites had been destroyed by Al- \nlied bombs or by retreating Germans . . .\n\n(but) bit by bit as we examined various \nsites, we pieced together an exact picture. \n... It rained and hailed during most of the \ntrip. Once or twice we slept on the \nground...\u201d\n\noo the most exciting experience \nwas that of P. V. Dimock, who spent \nthree hours floating around the Atlantic in \na life preserver one evening in May, 1944, \nafter an aircraft carrier was blasted out \nfrom under him by enemy torpedoes. The \ncarrier Block Island was out of Casablanca \nin the neighborhood of the Cape Verde Is- \nlands off the bulge of Africa when, for sev- \neral nights in a row, its search planes made \nradar contact with a submarine on the sur- \nface, presumably to recharge its batteries. \nThe carrier cruised to the point where the \ncontacts had been made, and one evening \nabout sunset was preparing to launch \nplanes to go after the submarine again. \nBut the enemy struck first. Two torpe- \ndoes slammed into the carrier almost si- \nmultaneously, one in the bow and one in \nthe stern. Dimock, who was on board as \na representative of the NDRC\u2019s Airborne \nInstruments Laboratory for work connected \nwith airborne submarine detecting devices,\n\nwas standing on the catwalk that runs \naround and slightly below the edge of the \nflight deck when he felt the carrier lurch. \n\u201cMy first impulse,\u201d he says, \u201cwas to grab \nsomething solid. I was surprised at the lack \nof noise when the torpedoes went off.\u201d \nThe carrier's engine room was wrecked \nand it was lying dead in the water. \nDimock hurried to the pilot\u2019s ready room \nbelow decks to get his life jacket, and was \nthere when a third torpedo exploded amid- \nships, putting out all the lights and jam- \nming the ready room door closed. It took \nabout two minutes \u201cbut it seemed a lot \nlonger waiting in the dark\u201d for someone \noutside to force the door open again. \nAbout ten minutes after the first explo- \nsion, the \u201cAbandon ship\u201d order was given. \nDimock slid down a rope into the ocean. \nThe rubber rafts were reserved for the sick \nand wounded, of which there were fortu- \nnately few, and the rest of the survivors \nhung onto the sides of the rafts to avoid \nbeing separated. Of four destroyer escorts\n\nthat were with them, one was torpedoed \nbut did not sink, and the other three began \npicking them up. The water was warm, \nDimock says, and he needed no medical \ntreatment after three hours of immersion, \nalthough getting the oil out of his hair was \nsomething of a task.\n\nDimock had taken leave of absence from \nthe Laboratories to join the NDRC group \nin March, 1942. His first trip away from \nthe States was a shakedown cruise in the \nPacific aboard the carrier Mission Bay in \nNovember, 1943. Then he shipped from the \nStates aboard the ill-fated Block Island in \nApril, and it was on the return trip that the \nsinking occurred. After returning to Casa- \nblanca on the destroyer escort that picked \nhim out of the water, Dimock took passage \non his old friend, the Mission Bay, for the \nStates, where he arrived in June, 1944.\n\nThe submarine? Dimock thinks they sank \nit. The destroyer escorts hunted it with con- \ntact mines while he was still in the water, \nand three of the mines hit something and \nexploded. Since the D-E\u2019s were not mo- \nlested as they picked up survivors, the sub- \nmarine was presumed sunk.\n\nDimock made several more trips aboard \ncarriers, including one to Newfoundland in \nSeptember and October, 1944, and a few \ncoastwise missions.\n\nOn of the most unusual trips was that \nof A. E. Bowen, who joined the Army \nand went to Trinidad with an anti-subma- \nrine air force squadron, and with an agree- \nment that he could become a civilian again \nas soon as that particular mission was fin- \nished. Bowen had been on leave from the \nLaboratories, working at Langley Field on \ndevices and techniques for anti-submarine \nwarfare, when the Army decided in Septem- \nber, 1942, that it needed a squadron of \nradar-equipped bombers in the Caribbean \nimmediately to assist in covering convoys \ncarrying supplies for the North African in- \nvasion. They needed Bowen, too, they said, \nbut he couldn't go along as a civilian. He \nwas commissioned a major and became \nradar and engineering officer for the squad- \nron which was equipped with an early Bell \nSystem search radar. He returned to the \nStates at the end of October, resigned, and \nafter a few weeks his resignation was ac-\n\ncepted. Nine months later he rejoined, \nagain as a major, and was promoted to the \nrank of lieutenant colonel, remaining in \nuniform until after the end of the war.\n\nPROJECT for the Signal Corps, still \nA under security wraps, sent Laborato- \nries men galloping all over the globe. J. M. \nBarstow and R. H. Badgely started this big \nparade, leaving for England toward the \nend of May, 1943, and C. R. Gray joined \nthem a month later. From here on the vari- \nous people, places, and dates are masked \nby secrecy. Badgely came home in August, \nand in September left for Honolulu, where \nhe stayed until the following February. \nBarstow and Gray returned from England \nin September, but they had a little more \ntime than Badgely to unpack and pack \nagain. They were off in November (still \n1943) for North Africa, eventually reaching \nCaserta, north of Naples, Italy, where they \nworked until the following March.\n\nIn the meantime, other people had come \nand gone. L. G. Schimpf and Wiley Whit- \nney preceded the other two in North Af- \nrica, arriving in Algiers in June, 1943, and \nreturning home in November. C. W. Vader- \nsen, traveling for the same project but in \nthe opposite direction, embarked for Aus- \ntralia in August, 1943, landed at Towns- \nville in the tropical northern part, and \nthumbed his way south by air to Brisbane \nwhile the ship, which had been rerouted, \nproceeded north to New Guinea with all \nhis equipment. The equipment finally ar- \nrived in Brisbane on September 28, and \ntwo days later it was set up and working, \nmeeting the deadline which had _ been \nscheduled before the ship was rerouted. \nVadersen passed part of November with a \nten-day trip to New Guinea to survey new \nsites for the project and to check on other \nBell System equipment. He returned to the \nStates in December, stopping off for a few \ndays in Hawaii to see Badgely. Thomas \nThatcher wound up the globe-girdling \ntours for the project with visits to Brisbane, \nHollandia in New Guinea, and Manila. \nThis odyssey began in October, 1944, and \nended in July, 1945. No one has yet tried \nto sum up the number of miles the seven \nof them traveled altogether, but the total \nprobably runs well into six figures.\n\noverseas on a scientific mission during \nthe war was Ralph Bown, now Director of \nResearch. Bown was in Lisbon, making \nconnections for the trip home, on the day \nPearl Harbor was attacked. He had been in \nEngland since September of that year as a \nrepresentative of Division 14, the radar di- \nvision of NDRC, to study British radar.\n\nBown left the country once more in June, \n1942, as an expert consultant for the office \nof the Secretary of War on a flying trip to \nPanama, Central and South America, dur- \ning which, among other things, he surveyed \nradar defense of the Canal Zone.\n\nT LEAST nine Laboratories communi- \ncations engineers zig-zagged franti- \ncally back and forth across France, Bel- \ngium, and western Germany, planning and \ndirecting establishment, maintenance, and \noperation of communications for the Allied \nforces during the land campaign against \nGermany. Part of this involved rehabilita- \ntion of commercial telephone systems \nwhich had been blasted by the enemy. \nFirst of them to reach Europe was D. K. \nMartin, who arrived in England May 1, \n1944, as a War Department representative \nand was assigned to the United States Stra- \ntegic Air Forces to serve as expert consult- \nant in radio communications. In September, \nMartin moved to France and the Ninth Air\n\nD. K. Martin, on his way to Miinchen- \nGladbach, poses amidst the ruins of Jiilich\n\nForce, where most of his activity was con- \ncerned with radio relay systems.\n\nL. L. Glezen and L. Pedersen, the first \nof the wire communications men, arrived \nin England late in May, 1944. The Nor- \nmandy landing was soon to come, and they \ninspected Signal Corps depots in England \nand taught schools on carrier equipment \nuntil the St. Lo breakthrough opened up \ntheir big job. They journeyed to France at \nthat time and arrived in Paris early in Sep- \ntember where they were attached to Com- \nmunications Zone headquarters.\n\nSome aspects of the task were heart- \nbreaking. There had been an extensive \ncommercial cable network throughout \nwestern Europe, but the Germans had \nblown up central offices during the retreat \nand it was necessary to replace destroyed \nexchanges with Army equipment. Pedersen \nbegan by rehabilitating the cable from \nCherbourg to Paris and erecting supple- \nmentary open-wire lines from Cherbourg \nto tie in with the Paris-Versailles cable. He \nopened a consulting office in Paris and \nGlezen had his headquarters in Rennes, \nbut most of the time they were scurrying \naround France and Belgium attending \nbreakdowns, reorganizing, rehabilitating, \nand installing communications.\n\nThey were joined at Communications \nZone in Paris by R. B. Hearn, who arrived \nin September, 1944. Hearn specialized in \nwire telegraph, and in engineering, estab- \nlishing, and clearing up troubles in teletype- \nwriter circuits. C. L. Cahill, who had already \nendured one stretch in the Pacific from April \nto September, 1944, on NDRC contract for \nradio countermeasure work, reached Paris \nin January, 1945, just in time to help repair \nthe break in the north-south cable that had \nforced transfer of northern elements of the \nFirst Army to British control. E. J. Noon, \nwho also arrived in January, swelled to five \nthe list of men attached to Communications \nZone. Noon and Cahill were originally \nscheduled to replace early birds Pedersen \nand Glezen, but there was so much work \nthat all five stayed on and eventually trans- \nferred to Twelfth Army Group headquar- \nters at Wiesbaden. Here they switched \nfrom French and Belgian equipment to re- \nhabilitation of German cable networks.\n\nJr., had been overseas since December, \n1943, as communications section head of \nthe American-British Laboratory, a branch \nof Harvard\u2019s Radio Research Laboratories, \nwhich specialized in countermeasures for \nboth radio and radar. Evans came to the \n9th Air Force on loan during the winter of \nthe Bulge to help set up radio relay sta- \ntions. Edward Praizner and P. B. Fairlamb \nreached the European Theater in March, \n1945, to join the 9th Air Force, where they \nwere assigned to forward headquarters. \nThey were concerned with wire communi- \ncations problems, including installation and \nmaintenance of wire telephone, telegraph \nand teletypewriter systems and restoration \nof German equipment. Since the Ninth was \nthe tactical air force which rendered close \nsupport to ground troops during the cam- \npaign across Germany, it was essential that \nits headquarters have quick, easy commu- \nnication with all its units and with various \nground force headquarters.\n\nPedersen began the homeward proces- \nsion of communications men in March, \n1945. May and June were busy months with \nMartin, Cahill, Hearn and Noon returning. \nGlezen came home in July after a 3,000- \nmile jeep trip around Germany, Praizner \nin August and Fairlamb stayed on until \nNovember when he, too, left the European \nTheater to its own devices.\n\n\u2014 before this group of communica- \ntions trouble-shooters went to Europe, \nthree men from the Laboratories crossed \nthe Atlantic to make special communica- \ntions studies. In September, 1942, H. S. \nBlack joined a party on its way to England \nto examine a British microwave radio relay \nset known as British Wireless Set No. 10, \nwhich had just been built on an experi- \nmental basis. Black returned in November, \nand the Laboratories subsequently de- \nsigned a new microwave radio relay sys- \ntem used extensively by our troops in \nEurope, the development of which stemmed \nin part from the studies in England.\n\nIn March, 1944, A. B. Clark and Albert \nTradup left for Europe as expert consult- \nants for the War Department to investigate \nand prepare overall reports and recommen- \ndations on problems connected with both \nfixed and mobile communications facilities.\n\nTheir travels took them through Algiers, \nTunis, Naples, Air Force headquarters in \nItaly, Fifth Army headquarters, then to \nEngland, where they visited numerous fly- \ning fields and radio installations. Along the \nway Clark had climbed Mt. Vesuvius and \nwas taken to that somewhat inaccessible \nhot spot, the Anzio beachhead, by grass- \nhopper plane. They were in communica- \ntions headquarters in England on D-Day. \nWhen they reached New York in late June, \n1944, they had flown about 15,000 miles.\n\nitems were not the only wartime \nshortages. There was often a short supply \nof materials more particularly essential to \nelectronic warfare such as mica, a natural \ninsulator which in many cases could not be \nreplaced with synthetic materials. F. J. \nGiven headed the U. S. Mica mission which \nvisited England from May 30 to July 7, \n1943, under auspices of the Combined Raw \nMaterials Board and the Harriman Mission \nto discuss allocation of available mica, of \nwhich principal sources were British mines \nin India and Brazil. It was necessary to cor- \nrelate American and British needs, to\n\nsqueeze estimates of both nations down to \nthe predictable supply, and to educate the \nBritish in uses of substitutes as well as low- \ngrade mica. These subjects had been stud- \nied extensively in America, so that enough \nhigh-grade material would be available for \nessential applications.\n\nCe. Laboratories men journeyed to \nEngland on special projects, ranging \nfrom exchange of information to study of \nnew developments. J. R. Pierce and H. D. \nHagstrum were there from late October \nthrough December, 1944, under U. S. Navy \nsponsorship, to trade notes on vacuum \ntubes for microwave radar. Pierce was par- \nticularly concerned with triodes, velocity- \nmodulated tubes, and reflex oscillators, \nwhile Hagstrum concentrated on magne- \ntrons and enclosed spark gap tubes which \nthe English prefer to call trigatrons. Among \nother things, they had tea with H. G. Wells, \nwho a year later announced his thesis that \nthere just wasn't any hope for modern man.\n\nC. R. Burrows, who at the time was on \nloan to the NDRC and has since become \nhead of the Department of Electrical Engi- \nneering at Cornell University, was in Eng-\n\nB. Fairlamb at the Sulzthal repeater station\u2014the terminal station for the 9th \nF. Hq.\u2014located at Bad Kissingen. German type repeaters and Western Elec-\n\nland during the winter of 1943-44 on an- \nother mission of particular importance in \nconnection with radio wave propagation \nstudies. Burrows was head of the NDRC \nCommittee on Propagation, which in con- \ncert with such distinguished scientists as \nSir Edward Appleton, Director of the De- \npartment of Scientific and Industrial Re- \nsearch of Great Britain, codrdinated studies \nin the United States, Great Britain and \nother nations and transmitted the informa- \ntion obtained to the Armed Forces. \nAnother trip sponsored by NDRC, and \nthe Navy also, was that of T. C. Fry, who \nexamined a number of British Army and \nNavy installations in England during the \nmonths of August and September, 1942. His \nwork had to do with target location and de- \ntection. D. E. Wooldridge also visited Eng- \nland during January and February, 1945, \nat the invitation of the Ministry of Aircraft \nProduction to exchange information on \nelectromechanical bombing devices.\n\ngated the destruction of so many air- \nplanes and buzz bombs that no one has yet \ncome up with a total, dragged a number of \nLaboratories men to diverse parts of the \nglobe. First to go was R. C. Dehmel, not \nstrictly an M-9 man, who took to England \nfor demonstration a developmental model \nof the first anti-aircraft director built for \nBritish guns. He went over at the end of \nJune, 1942, and returned in September. \nThen L. J. Kelly left home, journeying only \nas far as Camp Shilo, 120 miles north of \nWinnipeg, Canada, where, during Febru- \nary and March, 1943, he put the M-9 over \nthe jumps of winter testing. This experi- \nence was something like setting up house- \nkeeping in a cold storage vault since the \ntemperature dropped to 47 degrees below \nzero while he was there. However, both \nM-9 and Kelly survived well. By now the \nrecognized expert on how to keep one\u2019s \nM-9 alive in a blizzard, Kelly was off again \nto Alaska and the Aleutians in May of the \nsame year, staying until October.\n\nC. E. Fordham, who sits at a desk next \ndoor to Kelly\u2019s, carried matters to the other \nextreme. He doused himself in sunshine, \nheat and humidity in Panama from Octo- \nber, 1943, to January, 1944, from April,\n\n1945, to July, and again in October. His \npraise for the M-9\u2019s reaction to climate is \nas loud and well documented as Kelly\u2019s. \nDuring the first trip he inspected M-9's in \nlocations where they had been emplaced \nfor action for some time, and on the second \nand third visits he directed the M-9 part of \na formal tropical testing mission in which \nsomething like 2,000 items of Army ord- \nnance were checked under carefully con- \ntrolled conditions.\n\nK. G. Compton and J. Leutritz were also \nin Panama, courtesy of Army ordnance, \nfrom September to October, 1944, and from \nOctober to November, 1945, respectively. \nTheir studies were similar to Fordham\u2019s \nbut much broader in scope, covering the \neffects of corrosion and fungus, and of \nmoisture-proofing on many types of radio, \nradar and electromechanical equipment.\n\nWhen the M-9 began setting almost un- \nbelievable records in the European Thea- \nter, C. A. Lovell and R. R. Hough jour- \nneyed overseas to find out if it was all true \nand to consult with the British about its \nuse. After visiting large numbers of anti- \naircraft sites in England and France on a \ntour that lasted through July and August, \n1944, they were able to report that it was. \nLater on, the records were pushed even \nhigher, as D. B. Parkinson learned on an- \nother M-9 inspection tour through Western \nEurope in April and May, 1945. An all-time \nrecord was established in the last week of \nthe buzz bomb \u201cshoots\u201d at Antwerp, when \nthe combination of radar, M-9, 90-mm \nguns, and proximity fuse knocked down \neighty-nine out of ninety-one V-I\u2019s.\n\nNot an anti-aircraft device, but closely \nallied with anti-aircraft radar, was a Lab- \noratories-designed close support plotting \nboard which A. J. Borer accompanied over- \nseas in December, 1944. The board was \ntied in with ground radar and it prepared \ninformation automatically for use in direct- \ning bombers through weather and darkness \nto the target and in timing the release of \ntheir bombs. Borer was in England, i rance \nand Germany, returning in May, 1945.\n\nMr. Barnard, a former newspaperman, was \na member of the Laboratories from August 6, \n1945, to July 28, 1946, when he resigned to \nexplore his capabilities as a fiction writer.\n\nern cedar poles were demonstrated \nlast summer north of Toronto, in co\u00e9pera- \ntion with the Bell Telephone Company of \nCanada. The following ballad, somewhat \nrevised from the original, was read in due \nand ancient form in celebration of the most\n\nTheir butts and hearts by fungus plants, \nTheir bodies torn by gnawing ants,\u2019 \nAnd that\u2019s why this is written.\n\nWith shovels, axes, pipes and picks, \nAnd tapes and prods and poison mix, \nThe brave inspectors ran.\n\nAnd barristers with courtly guiles \nMade notes and records for the files, \nTo document the deeds.\n\n\u201cThe water\u2019s running out. \nWe'll use alone this magic oil, \nAll by itself, in wood and soil, \nAnd save the poles, no doubt.\u201d\n\n\"A far land, rich in \nhistory of the early ex- \nplorers, about five days\u2019 \nmarch north of the town \nof Toronto, in the Prov-\n\n*Probably the common Campo- \nnotus pennsylvanicus ligniperdus var. \nnovaboracensis, but these small crea- \ntures are no respectors of persons \nand destroy the goods of rich and\n\nsuccessful outcome of the experiments. A \nfew explanatory notes have been added and \na more complete, and shall we say more or- \nthodox account of the work, may appear in \na later issue; but the reader may or may \nnot learn more from the technical account \nthan from the ballad. Here at least for the \ntime being is the heart of the matter:\n\nAnd so the matter was resolved. \nThe summer months rolled by; \nUntil one cool October day \nThey cut the cedar\u2019s sides away \nTo see how ants would die.\n\nBy lane and wood and ditch and field \nThe happy echoes rang; \n\u201cThe cedars\u2019 broken hearts are bound, \nThe fungus dead, and we have downed \nThe ants of Penetang.\u201d\n\nforce his name was left in the line \nbecause no other attempt at versing \ndid sound so good as this; and for \nthat matter some other laird may \nhave said it.\n\nWesley Fuller of the Publication Depart- \nment is responsible for providing informa- \ntion to Bell System Companies, newspapers, \nmagazines and other publication media. \nMr. Fuller, a member of \nthe National Association of \nScience Writers, was sci- \nence editor of the Boston \nHerald for some time as \nwell as Boston Correspond- \nent of Science Service. In \nthe fall of 1941 he estab- \nlished and served as Exec- \nutive Director of the Red \nCross Blood Donor Center \nin Boston. He was also Di- \nrector of Public Relations \nfor the American Red \nCross in that city.\n\nMr. Fuller was gradu- \nated from Harvard in 1933, \nand returned to the Uni- \nversity on a Nieman Fel- \nlowship during 1938-39. \nHe holds a commission as first lieutenant in \nthe United States Marine Corps Reserve \nand had been released from duty following \nservice in the Pacific, including the Oki- \nnawa campaign, prior to joining the Lab- \noratories. Mr. Fuller reports to R. K. Hona- \nman, Director of Publication.\n\nDeaf education, speech correction, and \nother uses of visible speech, a development \nof the Bell Telephone Laboratories, will \nbe the subject of an extensive research pro- \ngram that will be started this fall at the \nUniversity of Michigan with the codpera- \ntion of the Michigan State Normal School \nof Special Education, nine miles from Ann \nArbor in Ypsilanti, Michigan.\n\nCenter of the study will be the Speech \nClinic of the University\u2019s Institute for Hu- \nman Adjustment, with Dr. Harlan Bloomer \nas the director. Uses of visible speech in \nboth speech and general education of deaf \nchildren will be carried on at the Normal \nSchool of Special Education, which is un- \nder the direction of Dr. Francis E. Lord.\n\nExperimental equipment, used by the \nBell Telephone Laboratories in early stud-\n\nies to determine the legibility of these \nspeech patterns, will be loaned to the Uni- \nversity of Michigan for the purpose. The \noriginal moving screen cathode ray trans- \nlator will be set up at the State Normal \nCollege and a sound spec- \ntrograph with other equip- \nment at the University. \nTwo temporary members \nof the Bell Telephone Lab- \noratories\u2019 technical staff, \nwho came from the educa- \ntional field to take part in \nthe early development of \nvisible speech, will con- \ntinue with the Michigan \nproject. Dr. George A. \nKopp will divide his time \nbetween an appointment \nas associate professor of \nspeech in the University \nand activities as a research\n\nGreen is to become an as- \nsistant professor at Michigan State College \nand a research assistant in the University\u2019s \nSpeech Clinic. It is anticipated that other \nunits of the University will participate in \nthe research in later phases of the project.\n\nVail Gold Medal Awarded for \nJulia Berry\u2019s Spirit of Service \nThe Bell System National Committee of \nAward for Theodore N. Vail medals rec- \nognized the supreme courage, loyalty and \ndevotion to duty demonstrated by Julia \nBerry during the La Salle Hotel fire by \nposthumously awarding her a Gold Vail \nMedal and $1,000. Both will be presented \nto her son, John, age 16, who was left an \norphan by his mother\u2019s tragic death.\n\nShe was alone on duty at the switchboard \nof the twenty-two story La Salle Hotel, Chi- \ncago, when a disastrous fire started on the \nground floor at about 12:30 a.m., June 5, \n1946. The flames swept rapidly to the seventh \nfloor before they were checked, and dense \nsmoke poured upward through the entire \nbuilding. Sixty-one persons lost their lives and \nmany more were injured or overcome by smoke.\n\nMrs. Berry\u2019s post was on the second floor. \nWhen notified of the fire, she immediately \ncalled the Fire Department. Then, despite the \noncoming smoke and flame, she set to work \nto spread the alarm throughout the hotel, call- \ning room after room to warn people of their \ndanger. Efforts were made to get her to leave. \nTwice she refused, saying there were many \nrooms that she had yet to reach. When the \nfire was over she was still at her post\u2014her \nbody slumped over the switchboard where \nshe had chosen to stay and serve others, at \nthe cost of her own life.\n\nMrs. Berry is the fifteenth to receive the \ngold medal since the establishment of the \nfund in 1920 in memory of Theodore N. \nVail, former president of the American \nTelephone and Telegraph Company. This \nis the highest honor than can be awarded \nto telephone people for acts of heroism in \ngiving noteworthy public service.\n\nNantucket Island, in the Atlantic Ocean \noff the Massachusetts coast, on July 2 \njoined Santa Catalina Island, in the Pacific \noff Southern California, in the distinction \nof being linked to the mainland by the \nfirst two commercial installations of a new \nmicrowave radio-telephone system. The \nservice to Catalina was opened on May 27.\n\nBy means of the new radio-telephone \nsystems, eight telephone circuits have been \nadded by the New England Telephone \nand Telegraph Company and the South- \nern California Telephone Company, respec- \ntively, to the twelve circuits previously \navailable to Nantucket and fifteen to Cata- \nlina through submarine cables.\n\nIn sharp contrast with the beacon fires \nand semaphore signals of Nantucket\u2019s early \ndays is this modern communication magic. \nBy it, words spoken into telephones in the\n\nancient shingled houses on the narrow, \ncobbled streets of the old whaling port are \ncarried by land telephone lines to a 42- \nfoot tower standing on high ground in the \nrolling moorland some two and a quarter \nmiles west of the town and are projected \nfrom it across Nantucket Sound to a similar \ntower about 30 miles away on a hill in \nthe western part of the town of Barnstable \non Cape Cod, there returning to land lines \nfor transmission to their ultimate destina- \ntion. The New England telephone people \nhave named this tower location Manson \nHill in honor of their late chief engineer, \nG. K. Manson.\n\nOn Santa Catalina the antenna tower \nand equipment house are located on a high\n\n1S PRESENTED THIS CERTIFICATE OF AWARD IN RECOGNITION OF \nEFFECTIVENESS OF DESIGN. EXCELLENCE OF EDITORIAL CONTENT, \nAND ACHIEVEMENT OF PURPOSE IN A CONTEST CONDUCTED BY \nTHE NATIONAL COUNCIL OF INDUSTRIAL EDITORS, U. S. A.. AND\n\nBell Laboratories Recorp was one of 84 \nmagazines out of 556 entries selected for \nan award in the First International Indus- \ntrial Publications Contest. Announcement \nof the award was made at the convention \nin Boston of the International Convention \nof Industrial Editors which was attended \nby P. B. Findley and Wesley Fuller\n\nhill about 2% miles from the island town \nof Avalon, while the other terminus of the \ninstallation is temporarily located on top \nof the telephone building at 433 South \nOlive Street, Los Angeles, 49 miles away.\n\nC. J. Christensen has been appointed \ndean of the School of Mineral Industries \nof the University of Utah. A native of Utah, \nDr. Christensen received his B.S. at Brig- \nham Young University, where he was an \ninstructor in 1923 and again in 1926. He \nreceived his master of science degree at \nthe University of Wisconsin and his doctor \nof science degree at the University of Cal- \nifornia. In 1929 he joined the Laboratories \nas a member of the Technical Staff. Since \nthen he has pioneered in the making of car- \nbon varistors and has made noteworthy \ncontributions in the fields of ceramics and \nvaristor materials and to the growing of \nsynthetic crystals. During the past year he\n\nhas been supervisor of a group engaged in \nthe research and development of magnetic \nmaterials.\n\nThe School of Mineral Industries which \nDr. Christensen will head was created by \ndivision of the School of Mines and En- \ngineering and will include mining, metal- \nlurgy, mineralogy and allied subjects, and \nthe basic earth sciences, geology, geog- \nraphy, ceramics and meteorology. The ne- \ncessity for the new school was outlined re- \ncently in press releases by Dr. A. Ray Olpin, \npresident of the University of Utah and a \nformer member of the Bell Laboratories.\n\nIn many communities across America \npeacetime telephone conversations are now \nbeing carried by wire and cable originally \nintended to link Army camps. Thousands of \ntelephone instruments, formerly installed \nin war plants and military posts, are be- \ning used. Millions of insulators, thousands \nof telephone poles, and huge amounts of \npole-line hardware, all made for wartime\n\nLIEUTENANT COLOVEL H4RPEW as Commanding \nOfficer, 71ist Region, Plant and Engineering \n4gency, Army Communications Service, from \nJanuary 109044 to September 1044, capably \ndischarged important resfonmsibislitres in \ndirecting and coordinating the construction \nand installatson of radio communication \nfacslitses and navigational aids for aircraft \nto complete communications along the routes \nof the Air Transport Command.\n\npurposes, are helping to re- \nlieve the Bell System\u2019s long \nwaiting lists of applications \nfor telephone service.\n\n\u201cWar surplus\u201d is the name \ngiven to this equipment. \nMuch of it was manufactured \nby Western Electric and its \npurchase from the Govern- \nment has become an extremely \nimportant phase of that com- \npanys job as supply unit \nfor the Bell System. Pur-\n\nLieut. Col. Harper, as Com- \nmanding Officer, 71st Re- \ngion, Plant and Engineering \nAgency, Army Communica- \ntions Service, from January \n1944 to September 1944, ca- \npably discharged important \nresponsibilities in directing \nco\u00e9rdinating the con- \nstruction and installation of \nradio communication facilities \nand navigational aids for air- \ncraft to complete communi-\n\nchases, as of August 1, \nhave totaled approxi- \nmately $7,500,000. \nThe major respon- \nsibility for buying the \nGovernment's surplus \ntelephone equipment \nhas been undertaken\n\nclosely with supplies \ninspection and_ the \ntwenty-nine distribut- \ning houses. The job \nhas become so exten- \nsive and so important \nthat a special war \nsurplus and materials \npolicy committee was \ncreated, headed by \nAssistant Purchasing \nAgent W. F. Johnson \nof Western Electric. \nMaterials being declared surplus by the \nGovernment include standard Western \nElectric telephone equipment. Other sur- \nplus items are not of Western Electric \nmake but meet Bell System specifications. \nStill other items vary from our specifica- \ntions, but are usable as substitutes. Ma- \nterials of interest are of two kinds: (1) \nequipment, including cable, telephone sets, \ncarrier apparatus, and PBX boards, and \n(2) supplies items, such as wire and \nstrand, poles and crossarms, strand and \nsteel wire, hardware, and motor vehicles.\n\nThree more inter-city highways, totaling \nover 800 miles in length, have been added \nto the two previously announced routes on \nwhich the Bell System plans to provide \nmobile radio-telephone service to vehicles. \nThe A T & T has announced that applica- \ntions have been made to the Federal Com- \nmunications Commission for authority to \nconstruct experimental stations along the \nhighways between Washington and New \nYork; Buffalo and New York, via Albany; \nand Los Angeles and San Diego. Permits \nto build transmitters and receivers for high- \nway mobile radio-telephone service be- \ntween New York and Boston and between\n\nSteel erection is practically complete on the new main building \nat Murray Hill\n\nChicago and St. Louis already have been \ngranted and construction of those stations \nis under way.\n\nOn the New York-Washington highway, \nit is planned to build transmitting and re- \nceiving stations near New Brunswick, \nPhiladelphia, Wilmington, Baltimore and \nWashington.\n\nThe New York-Albany-Buffalo route is \nthe third major highway in the thickly \npopulated eastern part of the country on \nwhich radio-telephone service for mobile \nunits is planned by the Bell System. In \naddition to the station in New York City, \ntransmitter-receivers are to be located near \nWhite Plains, Poughkeepsie, Albany, Fonda, \nUtica, Syracuse, Rochester and Buffalo.\n\nTo serve vehicles on the heavily traveled \nhighway between Los Angeles and San \nDiego, transmitter-receivers will be erected \non Mt. Wilson and Mt. Woodson, near the \nrespective cities. From those two com- \nmanding locations it will be possible to \ncover the entire 125 miles between the two \ncommunities.\n\nApplications for Bell System highway \nmobile radio-telephone installations in a \nnumber of other cities, including Cleveland, \nGrand Rapids, Knoxville and Mobile, are \npending or in preparation.\n\nThis is the Croix de \nGuerre Medal \nawarded to Col. Frank \nA. Parsons by _ the \nFrench Government \nfor the part he played \nduring World War II \nin the liberation of \nFrance. The Croix de \nGuerre is awarded in \nthree ranks and it is \nthe highest of these, \nthe award \u201cavec etoile \nde vermeil,\u201d which \nwas conferred on Col. \nParsons\n\nA cordial invitation to become members \nof the Murray Hill Chorus is extended to \nall Laboratories\u2019 men and women who live \nin the vicinity of Summit. The chorus will \nrehearse at the Summit Y.M.C.A. on Tues- \nday evenings from 8 to 10 o'clock so that \nthose who work at West Street and other \nlocations may be able to join the group.\n\nThe first fall rehearsal will be on Sep- \ntember 17, when it is hoped that all for- \nmer members and many new ones will be \non hand to start the chorus on its way to \nanother successful year.\n\nPaul Kuhn, as executive chairman of the \nchorus, has Phyllis Taylor as his assistant, \nwith Reine L. Levesque, secretary; William \nVierling, treasurer; Beatrice J. Panos, libra- \nrian; and C. Tanis in charge of membership.\n\nA NEW BOOK by R. A. HEIsinc, Quartz \nCrystals for Electrical Circuits, has been \nrecently published by D. Van Nostrand \nCompany. The book is a compendium of \nup-to-date information, both theoretical \nand practical, on quartz crystal plates, their \ndesign and manufacture. Included in the \nbook are chapters written by W. P. Mason, \nW. L. Bonp, J. ARMSTRONG, G. \nW. Wiiarp, R. A. Sykes, H. J. McSkimin, \nG. M. Tuursron, H. W. WErNuHaRT, H. G. \nWenueE, I. E. Farr, A. R. DHEEDENE, R. M. \nC. GREENIDGE and C. W. Harrison.\n\nP. S. DaRNELL visited the Allen-Bradley \nplant in Milwaukee with representatives of \nthe Western Electric Company to discuss \nrequirements and inspection practices for \nresistors purchased by Western Electric.\n\nL. S. C. NEE discussed control appara- \ntus at the Westinghouse Electric Company.\n\nK. K. Darrow spoke on Physics and the \nPublic on July 16 before the Chapter of \nSigma Xi at the Shell Development Com- \npany, Oakland. Dr. Darrow attended the \nAmerican Physical Society meeting in July \nat Berkeley, California.\n\nO. H. Dantetson, G. E. PERREAULT and \nI. V. WittraMs of the Laboratories and J. \nA. Burkart of the Western Electric Com- \npany visited the Massachusetts Institute \nof Technology concerning a Navy project.\n\nJ. R. Power and G. J. V. FALEy were in \nBurlington, N. C., on engineering problems \nassociated with hearing aid production.\n\nDuring the Four Months Ending With June, the United States Patent \nOffice Issued Patents on Applications Previously Filed by the \nFollowing Members of the Laboratories:\n\nH. H. Abbott A. G. Fox A. Majlinger W. G. Shepherd \nS. M. Babcock H. T. Friis W. P. Mason T. Slonezewski\n\nJ. Baumfalk E. W. Gent W. J. Means K. D. Swartzel, Jr. \nR. Black, Jr. A. A. Hansen R. S. Ohl (4) R. L. Taylor\n\nJ. W. Clark F. A. Hoyt A. R. Rienstra G. W. Willard (2) \nB. Dysart J. L. Hysko V. L. Ronci S. B. Williams\n\nV. T. WALLbER has been named to serve \nas a member of A.S.T.M. D20 subcommit- \ntee on Polyethylene.\n\nV. E. Lecce has received honorable men- \ntion in the field of theory and research for \nhis paper, Optimum Air Gap for Various \nMagnetic Materials in Cores of Coils Sub- \nject to Superposed Direct Current, in the \n1945 national prize paper awards of the \nA.LE.E. In 1940 Mr. Legg was awarded \nthe National Prize for Initial Paper of \nwhich F. J. Given was co-author.\n\nG. L. Pearson, P. W. Foy and W. \nSHOCKLEY visited the National Bureau of \nStandards to make measurements of con- \nductivity and Hall Effect in silicon and \nboron semi-conductors at liquid and solid \nhydrogen temperatures.\n\nStewart M. Beekman\u2014*Margaret Scharnke \nHarold Boland\u2014*Cecelia Cooney \nSal Cangemi\u2014*Frances Giambalvo \nWalter Chambers\u2014*Virginia Davis \n*Robert F. Graham\u2014June Gibson \nJohn Harrington\u2014*Genevieve Sokolosky \n*Paul T. Haury\u2014Annetta Christensen \nHarry E. Hechtman\u2014*Florence Doris \nDacel S. Lawrence\u2014*Helen Asher \n*Leo Luckner\u2014*Nellie Schofield \nLawrence L. Morrill\u2014*Catherine Denninger \nThomas J. Murphy, U. S. Navy\u2014*Dorothy Goll \nWilliam F. Ocenasek, Jr.\u2014*Daniela Sidla \nJoseph Porcelli\u2014*Dorothy Carney \nStanley M. Raines\u2014*Selma Graber \nWilliam Rigney\u2014*Margaret Heussler \nJerome Spina\u2014*Mary Sheridan \n*Thomas W. Thatcher, Jr.\u2014Lois Constance \nJ. Arthur White, Jr.\u2014*Miriam White\n\n*A. B. Anderson\u2014Ruth Johnson \n*Kenneth S$. Cadmus\u2014*Wilma Sabolsak \nStephen G. Dorosh\u2014*Virginia Voskanyan \nDavid L. Fisher\u2014*Dorothy Bebb \n*Howard C. Fleming\u2014Cecelia Toughey \n*R. Shiels Graham\u2014*Isolde Holoch \nWalter G. Henry\u2014*Isabelle Simpson \nRobert E. Huddy\u2014*Elizabeth Loch \nL. Thomas Leonard\u2014*Ruth Kampfe \n*Robert H. Meuser\u2014Alyce Benjamin \n*John C. Pfaff\u2014Muriel Jenkins \nWesley G. Rogers\u2014*Rose Slater \n*George J. Wolters\u2014Catherine Clancy\n\n*Members of the Laboratories. Notices of engage- \nments and weddings should be given to Mrs. Helen \nMcLoughlin, Room 803C, 14th St., Extension 296.\n\nNoon hour on the grounds at Murray Hill \nsuggested a picnic to a photographer who \nposed this group of girls. On the wall are \nJanet Brown and Peggy Anderson; on the \nlawn, Mary Murray, Jean Gutleber, Muriei \nBey and Georgia Plumridge, all of the \nMurray Hill Service Department\n\nR. H. Cotzey and J. Leutrirz, Jr., at- \ntended a conference at Forest Products \nLaboratory on culture technique for evalu- \nating modern wood preservatives.\n\nC. S. Futter, W. O. Baker and P. P. \nDebye attended the High Polymer Confer- \nence of the American Society for the Ad- \nvancement of Science at Gibson Island. Dr. \nBaker, who is chairman-elect of the 1948 \nHigh Polymer Conference, presented a pa- \nper on Structure of Synthetic Rubber.\n\nC. J. Froscu, R. C. PLratow, W. O. \nBakER and R. Burns attended the A.T.S.M. \nmeeting in Buffalo, at which Mr. Platow \nwas elected chairman of Committee D-14 \non Adhesives.\n\nA. C. Exva of Transmission Apparatus \nDevelopment has received an M.S. degree \nfrom Stevens Institute of Technology.\n\nOliver Heaviside\u2014Humorist, an article \nby C. M. Hesspert, was published in the \nJune \u00b046 Journal of the Franklin Institute.\n\nC. H. Amapon has been investigating \ncurrent production and treating problems \nwith suppliers for the telephone industry \nof poles and crossarms in the Mountain. \nStates and Pacific coast areas.\n\nRecent retirements from the Laboratories \ninclude J. C. Wricut with 36 years of service, \nJuly 31; and A. D. Harcan, 40 years, and C. \nM. CAMPBELL, 39 years, on August 31.\n\nMr. Campbell was graduated from Purdue \nUniversity in 1904 with a B.S. degree in E.E. \nand entered the Automatic Electric Company. \nIn March 1905 he left to join the Western \nElectric at Clinton Street where, after terms in \nthe Student Course, Inspection Department and \nEquipment Engineering, he decided to be- \ncome a patent attorney. He served with a \nChicago patent law firm and received an LL.B. \ndegree from Kent College of Law.\n\nIn November 1909 Mr. Campbell returned \nto Western Electric Company as a member of \nthe newly established patent department at \nNew York. His Bell System patent experience \nhas been very diversified, including telephony, \nboth manual and automatic, printing tele- \ngraphs, public address systems, sound pictures, \nspeech recording, phonographs, acoustics and \nmeasuring and testing. From 1918 to 1922 he \nwas in charge of the patent group at Haw- \nthorne. Since 1938 he has been concerned \nwith investigations of patents and inventions \nof outsiders of possible interest to the Bell \nSystem. He has also handled questions sub- \nmitted by A T & T relating to inventions and \ntechnical suggestions by employees of the As- \nsociated Companies. The war added to his \nduties supervision of reports to the Govern- \nment authorities on inventions originating in \nconnection with Government contracts. He is \na member of the Bar of the State of New York \nand of a number of Federal Courts.\n\nAfter graduation from Webb Institute of \nNaval Architecture and Marine Engineering, \nMr. Hargan joined the Western Electric Com-\n\npany in New York in \n1906 as a draftsman on \ncabling plans for cen- \ntral offices. After two \nyears on this work, he \ntransferred to the ap- \nparatus drafting group, \nand then was made \nchief draftsman of the \nmanufacturing organi- \nzation in New York.\n\nMr. Hargan went to \nHawthorne in 1913 to \ntake charge of techni- \ncal correspondence for \nthe Manufacturing De- \npartment. After two years in this capacity and \na year in charge of one of the drafting divi- \nsions, he returned to New York, where, in \nthe Engineering Department, he was _ con- \ncerned with the design and development of \npanel dial apparatus. He continued this work \nuntil 1928, devising improvements and refine- \nments of the apparatus and supervising a \ngroup from 1925 to 1928. He then transferred \nto the repaired apparatus group, where he \nengaged in studies of the economics and re- \nquirements for repaired apparatus, particularly \non station telephone and teletypewriter appa- \nratus. He supervised this type of work from \n1935 to 1942.\n\nMr. Hargan conducted the Laboratories Out- \nof-Hour Course in Manufacturing Methods in \n1934-35 and again in 1936-37.\n\nAt the outbreak of the recent war, he trans- \nferred to the Commercial Products Develop- \nment Department at Whippany, where he was \nconcerned with the mechanical design of elec- \ntronic computers for solving the problems \nconnected with the development of airborne \nradar navigation and bombing.\n\nMr. Wright, of Transmission Apparatus De- \nvelopment, received the M.E. degree from \nCornell University in 1909 and directly en- \ntered the student training course at Haw- \nthorne. The following year he transferred to \nthe Physical Laboratories at West Street \nwhere, during his work on dry cells and port- \nable storage batteries, he made important con- \ntributions to standardized test methods and \nautomatic test equipment. He handled the de- \nvelopment and construction of precisely con- \ntrolled air conditioning equipment which has \ncontinued in use in the Laboratories for dup- \nlicating atmospheric conditions to which tele- \nphone apparatus is subjected in service.\n\nof apparatus projects, Mr. Wright\u2019s major re- \nsponsibility was supervision of work on in- \ncandescent lamps and switchboard signaling \nthrough a time when tungsten filament switch- \nboard lamps went into large-scale use, when \nresistance lamps were first designed to meet \nspecific circuit conditions, and when testing \nof illuminating lamps was introduced. During \nWorld War II, Mr. Wright was called in on \nthe development of wire used in electronic \nequipment for the Armed Forces. In this, his \nformer experience in dealing with the behavior \nof wire and cable insulations under various test \nand service conditions enabled him to make \nvaluable contributions.\n\nMorcan Sparks discussed new types of \nbatteries at the National Carbon Company \nin Cleveland and at the Bureau of Stand- \nards in Washington.\n\nMary N. Torrey of Quality Assurance \nreceived an M.A. degree from Columbia \nUniversity in June.\n\n]. F. Morrison read a paper on Design- \ning FM Antennas at a meeting of the \nMaryland section of the I.R.E.\n\nL. Vieru has been appointed to Subcom- \nmittee F on Audiometry and Hearing Aids \nof the A.S.A. Committee on Acoustical \nMeasurements and Terminology.\n\nT. A. Durkin, in company with Western \nElectric engineers, discussed cable pro- \nduction problems at the plant of the \nSurprenant Electrical Insulation Company.\n\nJ. E. Casstpy and R. C. Terry went to \nthe Patent Office in Washington during \nJuly on patent matters.\n\nS. M. Surron and D. C. Smiru conferred \non clay conduit specifications at the Hay- \ndenville, Ohio, plant of the National Fire- \nproofing Corporation.\n\nO. E. Buckiey, R. L. Jones and R. K. \nHoNnaMAN attended the Bell System Public \nRelations Conference from July 16 to 18 \nat Absecon, N. J.\n\nDr. Buckley has been appointed a mem- \nber of a special Advisory Committee on Ord- \nnance Research and Development formed \nby the Army Ordnance Association in re- \nsponse to a request from Major General \nEverett S. Hughes, Chief of Ordnance. This \ncommittee will take a leading part in the \nbroad solution of problems in the field of \nOrdnance research and development.\n\nHarvey FLEeTcHER has been appointed \nchairman of the committee charged with \nreviewing the policy of the American Phys- \nical Society with respect to divisions.\n\nW. C. Herrinc has returned to the Lab- \noratories after a five months\u2019 leave of ab- \nsence, during which he taught at the \nUniversity of Texas.\n\nF. F. RoMANow attended a meeting of \nthe Executive Committee of the Sound \nEquipment Section of the Radio Manufac- \nturers\u2019 Association in regard to the stand- \nardization of microphone and loudspeaker \nperformance.\n\nG. W. Meszaros visited Chauncey, Ga., \nwhere modifications of rectifier-inverters \nfor power supply on coaxial systems were \nunder test.\n\nJ. F. Potnemus, H. A. Lewis and H. \nKeppicus visited the coaxial equipment at \nthe Washington, D. C., Toll Office of the \nLong Lines Department.\n\nLieut. Leon M. Gould, upon entering the Air \nCorps, became a Link trainer instructor, was later \nassigned to study at the University of California, \nand subsequently was stationed as a platoon com- \nmander giving basic training at Fort Leonard Wood.\n\nRobert E. Poirier engaged in radar maintenance \nand development at St. Simon Island, Georgia, and \nin radar beacon maintenance at bases in Cuba and \nKey West before studying at Vanderbilt University.\n\nCapt. Daniel F. Hoth spent three and a half \nyears in the Signal Corps in Arlington, Virginia. \nentering service in December, 1942, as a Second \nLieutenant, he spent most of his time in the Army \ndirecting a group of military and civilian personnel \nengaged in the development of special communica- \ntions equipment. In the summer of 1945, for two \nmonths, he investigated certain German wartime \ndevelopment activities in Germany.\n\nFrances V. Tracy was recently released from the \nNaval Reserve with the rank of Aerographer\u2019s Mate \n3/c. During her war service she was a weather ob- \nserver stationed at Banana River, Florida.\n\nWarren E. Wilson\u2019s military service was spent \nat the Infantry School, Fort Benning, where he was \nattached to the Secretary\u2019s Office as clerk and \nserved as driver on special missions for the General.\n\nEnsign Marvin Bracken served eighteen months \nin the Navy as an aviation cadet, following which \nhe was commissioned in the Coast Guard, where \nhe was engaged in air-sea rescue operations in \nthe New York area.\n\nDonald A. Loughlin participated in the invasion \nof Okinawa and in occupational landings at Korea \nand Tientsin, China, as a member of the Arcturus, \nan amphibious cargo attack ship.\n\nAndrew Olsen became team chief of a tele- \nphone installation and maintenance group which \ntook over and operated French installations at Metz\n\nand Nancy, and later did similar work on the Ger- \nman installation at Wiesbaden.\n\nDonald J. Oakley was assigned on two occa- \nsions to the Asiatic-Pacific theater of war. First he \nwas land-based on the Solomons and Admiralties \nand assigned to the Fleet Air Wing; then he re- \nturned to the States for special training and was \nassigned to duty on Okinawa and in Japan.\n\nWallace C. Hickman served for over three \nyears in the Signal Corps. After specialized training \nat Camp Crowder, he was engaged in radio tele- \ntype transmitter installation work at Algiers.\n\nGilbert Goodman has returned to the Chemical \nDepartment from military leave following approxi- \nmately two years in the Army. He served nineteen \nmonths in the Pacific theater of operations. Part \nof the time he was connected with the Manila real \nestate branch office.\n\nChief Vincent J. McCarthy returned to the De- \nvelopment Shop at Murray Hill after nineteen \nmonths in the Pacific area, where he was engaged \nas a Chief Machinist\u2019s Mate in the repair of naval \nvessels damaged in enemy action.\n\nCapt. Charles L. Semmelman, a reserve officer \nin the Signal Corps, upon entering service became \nAssistant Officer in Charge of the Performance Test \nSection, of the Signal Corps Engineering Labora- \ntories, Fort Monmouth. This Section performs any \nrequired test on all parts and materials considered \nfor use in Signal equipment except tubes, crystals \nand batteries. During the war, the Section pre- \npared nearly 10,000 written reports on the results \nof tests and assisted in the preparation of nearly all \nJAN specifications by furnishing data on parts pre- \nviously tested.\n\nBartholomew A. Stiratelli received Marine \nCorps boot training at Parris Island and Camp \nLejeune and was assigned to a corps of mainte-\n\nnance engineers in the Marine Corps Ordnance \nDivision located in Washington, D. C.\n\nMajor Robert J. Fluskey has returned to the \nLaboratories upon completion of thirty-eight \nmonths in the Army. He volunteered in April, \n1943, accepting a direct com-\n\nCapt. Helen G. Adams, after receiving her \nbasic training, was assigned to general duty in the \nNeurosurgical Department at Halloran General \nHospital, where she later became Assistant to the \nChief Nurse. Capt. Adams was transferred to Fort\n\nmission as First Lieutenant, Sig- \nnal Corps. His initial six months \nwere spent as Control Officer of \nToms River Signal Laboratory, \nSignal Corps Ground Signal \nAgency. The following twenty- \ntwo months involved assignments \nas engineering, contracting and \nproduction-expediting officer for \nMonmouth Procurement District \nof the Signal Corps Production \nand Distribution Service, for \nwhich he received two com- \nmendations. After several months \nin a staff capacity in the office of\n\ntraining for service in the Far \nEast, she was stationed with the \n122nd General Hospital. After \nV-J Day she served as Assistant \nto the Chief Nurse at Borden Gen- \neral Hospital, Chickasha, Okla- \nhoma, until her recent return to \nthe Medical Department nursing \nstaff at West Street.\n\nGeorge A. Sharpe's naval as- \nsignment was at Mechanicsburg, \nPa., where, as a Storekeeper Tech- \nnician, he was responsible for \nchecking all requisitions for elec-\n\nPentagon, Major Fluskey, subsequent to V-J Day, \nwas assigned to the Cambridge Signal Patent \nAgency, which is located at M.I.T. and Harvard, \nand of which he was executive officer when he \nwent on terminal leave.\n\nCapt. Thomas G. Woods, Jr., has been re- \nverted to inactive duty after service with the Signal \nSecurity Agency at Washington, and with the Sig- \nnal Intelligence Service, operating out of Australia \nand the Philippines during the war. He also served \nat SIS Headquarters, Kyoto, Japan, as well as in \nManila after peace was signed, and for a short \ntime in Washington upon his return.\n\nLieut. James R. Walsh, a pilot, instructed in \ntwin engine trainers before receiving specialized \ntraining in Troop Carriers. In the ETO, he flew his \nC-47 on re-supply missions and on the Rhine Drop. \nReturning to Fort Bragg, he was an instructor of \na Troop Carrier Group until he was released.\n\ntrical motors and controllers, ad- \nvising the section if material was \nin stock and ordering the items needed.\n\nChief H. John Geisler, CETM, did submarine \nwork at New London before being assigned to the \nstaff of Radio Mat\u00e9riel School, Washington, D. C., \nwhere he taught Antenna, Transmission Lines and \nWave Guides for over two years.\n\nF/O William R. Spenninger, after commission- \ning, instructed in navigation for eight months be- \nfore being assigned to train for B-29\u2019s. Overseas \nhe flew with the 20th Air Force, based on Saipan, \non bombing missions over Japan until V-J Day \nand then was assigned to flying supplies from \nSaipan and Guam.\n\nAlbert R. Strnad, while at B-29 gunnery school, \nhelped design the Hobson-Strnad Trainer, which \nhe later used at various airfields and in Detroit for \nexperiments in teaching gunners in the proper use \nand firing of the B-29\u2019s gunnery system. Patent \napplication for the trainer is being searched.\n\nLieut. John M. Woitovich received his commis- \nsion at the Midshipman\u2019s School on the Prairie \nState and attended submarine school prior to be- \ncoming Assistant Engineer and Electrical Officer on \nthe submarine Blackfin, which operated out of Aus- \ntralia and the Philippines.\n\nVictor Silzer taught radar at Boca Raton before \nhis assignment at the international military trials at \nNuremberg, where he translated original docu- \nments written by Hitler and Goering.\n\nEnsign John V. Elliott, who received a direct \ncommission in the Navy, trained at the Naval Com- \nmunications Center, Harvard University, before be- \ning assigned as Communications Officer on the \nMt. McKinley. He also served on minesweepers in \nJapan, Korea and China waters.\n\nLieut. Thomas J. Doherty was a detachment \ncommander with Hq. 4th Air Forces at Hamilton \nField and a security officer in San Francisco before \nbeing assigned in India for a year as Security In- \ntelligence Officer. He was also assigned to a tactical \nteam with the British 14th Army in Burma and in \nRangoon, and later was reassigned to India.\n\nChief Charles H. Dalm, ACETM, received spe- \ncialized training in radio and radar maintenance \nschools and at sea before being assigned to the \nWasp, which participated in battles on the Philip- \npine Sea and on which he traveled to much of the \nFar East, including Japan.\n\nHarry B. Compton served as radio operator \non the destroyer Kearny, and, after advanced \nschooling as radio technician, on the Semmes and, \nin Pacific waters, on the Altair and the Laffey.\n\nRobert A. Hauslen\u2019s varied naval career in- \ncluded aviation cadet training; ETM schooling; \nand duty on the destroyer Cowell during the inva- \nsion of Okinawa, at Pearl Harbor and at Saipan.\n\nRobert M. Eichhorn served in the Navy for over \nthree years, first as a cadet and later, after special- \nized training, as an electronic technician\u2019s mate at \nbases maintaining electronic equipment.\n\nFrank A. Chionchio has returned to work in \nBuilding T after nearly three years of naval duty, \nmost of it in the Pacific, where, after training in \nETM School, he served on the Freemont.\n\nRichard Rafferty, assigned to Frankford Arse- \nnal, traveled from there on field trips to work on \nordnance material and to instruct in fire control \nmaterial. Later he transferred to the Air Corps.\n\nLieut. William H. Christoffers earned the Air \nMedal for \u201cbombing nine enemy planes on four \nruns at low altitude.\u201d An aerial navigator, he flew \nin a patrol bomber with the VB 115 Squadron of \nthe 7th Fleet and in over 1,000 flying hours the \nplane was credited with 8 ships and 12 planes.\n\nMatthew Tomb spent twenty-six months in the \nNavy, of which he served twenty months in the \nPacific aboard the submarine tender Appollo.\n\nThomas J. Comparetta, a B-29 gunner, was \nbased on Tinian and Okinawa during three hun- \ndred hours of flying time. Later he made trips to \nthe Philippines, Korea, China and Japan, where his \nplane made pictures of the atom bomb havoc.\n\nKenneth A. Josephson was awarded the Purple \nHeart for wounds received by machine-gun fire on \nOkinawa while fighting with the 6th Marine Divi- \nsion Infantry. Later, after hospitalization on Guam, \nhe was assigned to the Military Government there \nand stationed in a native village until his release.\n\nLeaves of Absence \nAs of July 31, there had been 1,052 military \nleaves of absence granted to members of the Lab- \noratories. Of these, 827 have been completed. The \n225 active leaves were divided as follows:\n\nThere were also 10 members on merchant ma- \nrine leaves and 1 on personal leave for war work.\n\nRemember the crystal detector in the first radios\u2014 \nhunting for the right spot with a cat\u2019s whisker? \nFor years the detector lay discarded in favor of the \nvacuum tube. But when microwaves came, and \nwith them the need to convert minute energy to \namplifiable frequencies, a Bell Laboratories scien- \ntist thought back to the old crystal.\n\nSilicon of controlled composition, he discov- \nered, excelled as a microwave detector. Unlike \nthe old-style natural crystals, it was predictable \nin performance, stable in service. From 1934 to \nPearl Hlarbor, the Laboratories developed silicon \nunits to serve microwave research.\n\nThen Radar arrived. \u2018The silicon crystal came \ninto its own, and found application in long- \ndistance microwave Radar. Working with Ameri- \ncan and British colleagues, the Laboratories rap- \nidly perfected a unit which the Western Electric \nCompany produced in thousands. It became the\n\nCrystal detectors are destined to play a big role \nin electric circuits of the future. They will have \nan important part in Bell System microwave \nradio relay systems. They may reappear in radio \nsets. Here again Bell Laboratories\u2019 research has \nfurthered the communication art.\n\nEXPLORING AND INVENTING, DEVISING AND PERFECTING FOR CON- \nTINUED ECONOMIES AND IMPROVEMENTS IN TELEPHONE SERVICE", "title": "Bell Laboratories Record  1946-09: Vol 24 Iss 9", "trim_reasons": [], "year": 1946}
{"archive_ref": "sim_record-at-t-bell-laboratories_1948-01_26_1", "canonical_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1948-01_26_1", "char_count": 118830, "collection": "archive-org-bell-labs", "doc_id": 852, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc852", "record_count": 134, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1948-01_26_1", "split": "test", "text": "POWER LINE TREATMENT FOR THE MI CARRIER TELEPHONE \nSYSTEM, J. M. Dunham... 2 \nPORTABLE MICROWAVE TOWER... \nTELEPHONE SERVICE FOR TRAINS, N. Monk . . . . . \nA NEW CRYSTAL CHANNEL FILTER, E. S. Willis . 18 \nHISTORIC FIRSTS: RADIO ALTIMETER. . . |S\n\nAUTOMATIC SWITCHING FOR PRIVATE-LINE TELETYPEWRITER \nWAVEGUIDE HyBRIps, W. A. Tyrrell. . . \nMODELS 65 AND 66 HEARING Alps, J. R. Power. . . . . 80\n\nA PORTABLE TOWER HAS BEEN DEVELOPED TO SURVEY BY MICRO- \nWAVES THE ROUTE FOR RADIO RELAY SYSTEMS (SEE PAGE 6)\n\nBELL TELEPHONE LABORATORIES INCORPORATED \n463 WEST STREET, NEW YORK 14, N. Y. \n0. E. BUCKLEY, president J. W. FARRELL, secretary W. \u00a9. BURGER, treasurer\n\nPHILIP C. JONES, science editor R. LINSLEY SHEPHERD, associate editor \nHELEN McLOUGHLIN, assistant editor LEAH E. SMITH, circulation manager\n\nAs 1947 draws to its end, we can look back on it as a year \nof outstanding progress. Enthusiasm combined with hard \nwork has produced imposing results. Projects interrupted by \nthe war have been pushed to completion. Projects started \nsince the war are now coming through. The birth rate of new \nideas has been high, and the further we go the more oppor- \ntunity looms ahead.\n\nBut we are not content with what we have done, and enter \n1948 with a determination to mark the year with accomplish- \nments that will surpass anything we have done in the past.\n\nMy grateful appreciation goes to all the members of our \nLaboratories family who have worked together so harmo- \nniously and successfully, and my best wishes are for a New \nYear as happy as it is sure to be productive.\n\nPower lines over which M1 carrier sys- \ntems* are operated usually start from a \nsub-station and extend with branches and \nsub-branches through their service territory \nto supply power to the scattered customers. \nFor the most part, the lines are single \nphase with a multi-grounded neutral, and \noperate at about 7,200 volts, 60 cycles. \nTransformers near the customers\u2019 houses \nreduce the 7,200 volts to the usual 120-240 \nvoltage for lighting and power, but unfor- \ntunately, these transformers have a high \nattenuation for the M1 carrier-frequency \ncurrents. The simple expedient of plugging \na suitable carrier-telephone set into a 120- \nvolt outlet in the house thus cannot be uti- \nlized. A different method of coupling the\n\nThe coupling device employed consists \nof a high-voltage capacitor connected in \nseries with a coil which is grounded, as \nshown in Figure 1. Telephone carrier \nequipment is connected across this coil. At \npower frequencies, the impedance of the \ncapacitor is about 1.5 megohms, whereas \nthat of the coil is about 3 ohms. Thus the \nnormal 60-cycle voltage on the coil is only \na small fraction of a volt. Abnormal volt- \nages, which would otherwise appear across \nthe coil if the capacitor failed or if the coil \nwinding developed an open, are prevented \nby protector blocks across the coil and a \nhigh-voltage fuse in series with the capaci- \ntor, which insure prompt clearing of a dam- \naged capacitor or the by-passing of an \nopened coil. A switch is provided in the \ncoupler box which by-passes the coil and \nprotector blocks, so that the blocks may be \nsafely removed for inspection or the coil \nmay be replaced. The coupling capacitor \nand l-ampere fuse are shown in Figure 2, \nand a typical pole installation for a sub- \nscriber is shown on this page. The coupler \nbox is shown in Figure 3. The coils shown \nin Figure 1 are in the larger of the two \ncans immediately above the terminal strip, \nwhile most of the other apparatus of the \nunit consists of filters.\n\nThe coupling capacitor consists of an oil- \nimpregnated paper capacitor unit con-\n\nTypical pole installation at the end of a carrier \nsection showing a tap choke at upper right to \nblock the carrier from the line beyond this \npoint. A fuse and coupling capacitor are at the \nupper left, and a coupling unit below all the \npower equipment. The power transformer with \nlightning arrester mounted on it is at the level \nof the neutral wire\n\ntained in a glass tube about 16 inches long \nand 2 inches in diameter. It has a capaci- \ntance of 2,000 yyf, and a maximum rated \n60-cycle operating voltage of 8,700, but it \nmust withstand impulse tests at 95,000 \nvolts to conform with A.I.E.E. standards \nfor this type of power equipment. \nEfficient transmission of carrier over the \npower line requires proper terminations at \nthe ends of lines and choke coils at taps to \nreduce reflection losses. Early field tests in- \ndicated the approximate inductances that \nshould be used for the chokes. Three types \nof coils are specified: isolating, tap and \ntransmission chokes. Although both the iso-\n\nGROUND WIRE \nFig. 1\u2014Circuit schematic of the 60-cycle section of \na coupling unit\n\nSince all of these chokes operate at 7,200 \nvolts to ground and carry the power-line \ncurrent, they must meet the same require- \nments as similar power-line equipment for \nshort-circuit currents, temperature rise due \nto load currents, and protection against \nlightning voltages. These requirements tend\n\nFig. 2\u2014Coupling capacitor and fuse used for \nconnecting low-voltage carrier circuits to the \nhigh-voltage power lines\n\nto conflict with those for carrier frequen- \ncies. Large current-carrying and heat-dissi- \npating ability, for example, conflict with \nthe objective of low winding capacitance. \nThis imposed rather severe design prob- \nlems. The three types of chokes are de-\n\nsigned to withstand 500 amperes short- \ncircuit current. Suitable gaps are provided \non the coils for lightning protection.\n\nThe rated continuous power-line load \ncurrent for the isolating chokes is 10 am- \nperes, and they will carry up to 15 amperes \ncontinuously without undue heating. This \nchoke coil is mounted in an oil-filled steel \ntank having two high-voltage porcelain \nbushings, as shown in Figure 4. The assem- \nbled coil, which weighs sixty pounds, is \nmounted on the power pole and connected \nin series with the power line.\n\nThe tap choke and the two transmission \nchokes are not oil-filled; they are rated at 4 \namperes, but will carry up to 7, 9 and 10 \namperes, respectively. A coil of this type \nwith supporting clamp is shown in Figure \n5. It is suspended by means of its clamp \nfrom a pin-type power insulator. The coils \nare designed for hot-line mounting; that is, \nthey can be put in place with an insulated\n\nhot-line stick and connections made with- \nout interrupting the 7,200-volt circuit.\n\nWhere load currents at either end of a \ncarrier section exceed 15 amperes, two iso- \nlating chokes may be connected in parallel. \nTap chokes are not operated in parallel be- \ncause 2.5 mh is about the minimum induc- \ntance which will effectively block these \ncarrier frequencies. If tap lines have cur- \nrents greater than 7 amperes, therefore, an \nisolating choke can be installed in place of \nthe tap choke.\n\nSome rural power lines are equipped \nwith automatic reclosing circuit-breakers to \nrestore service on the line following trouble, \nand the winding of the operating solenoid \nis connected in series with the line. These \ncoils insert a large loss in the line at carrier \nfrequencies, and so, when there are circuit- \nreclosers in a carrier section, it is advisable \nto by-pass the solenoid coil with an imped- \nance that is low at carrier frequencies, but \nhigh compared with the coil impedance at \npower frequencies. The by-passing unit, de- \nsigned for mounting inside the oil tank of\n\nthe circuit recloser, has a maximum imped- \nance of 50 ohms at carrier frequencies.\n\nIn applying the M1 carrier telephone sys- \ntem to a rural power line, the equipment \nmounted on power-line poles is installed by \npower linemen, except the common termi- \nnal and cabinet, which are installed by \ntelephone linemen after power-line equip- \nment connections have been made. Since \ntelephone linemen must have access to the \ncoupler boxes on the pole, they are located \nat least 40 inches below any power-line \nequipment. Drops are run from these boxes\n\nAmong the more recent plant develop- \nments is this new UG distribution wire de- \nsigned for burying in the ground. Perfected \nby the Laboratories and now used by a \nnumber of Bell companies, the wire con- \nsists of a pair of sixteen-gauge copper con- \nductors covered with rubber insulation, \nover which is a braided steel wire armor. \nThe outside jacket is of neoprene. The in- \ncreased tensile strength of the wire results \nin fewer breaks and, with the armor act-\n\nThe choke coils were designed by S. G. \nHale and are being manufactured by the \nWestern Electric Company. The coupling \ncapacitor now being used, shown in Figure \n2, was designed by the Sprague Electric \nCompany to meet requirements set up for \nthe M1 carrier telephone system. Specifica- \ntions and acceptance test requirements for \nthe coupling capacitor were prepared by \nJ. R. Weeks, Jr., and C. C. Houtz.\n\nTHE AUTHOR: J. M. Dunnam received a B.E.E. \ndegree from Ohio State University in 1924. After \ntwo years with the Commonwealth Southern Cor- \nporation he joined the D & R, where he worked \non low-frequency induction as a member of Project \nCommittee 2J of the Joint Subcommittee on De- \nvelopment and Research, Edison Electric Institute \nand Bell System. During World War II, he was \nassociated with the development of radar equip- \nment for certain fighter planes. Subsequently he has \nbeen concerned with the power-line equipment \nand protection features involved in adapting rural \npower lines for M1 carrier and also with low- \nfrequency inductive co\u00e9rdination problems.\n\ning as a shield against lightning, no sepa- \nrate shield wire is necessary. The armor \nand jacket provide additional protection \nagainst cutting and crushing during instal- \nlation and in service. This new wire is \nused in place of open wire or cable in \nrural areas where soil conditions permit. It \nis also excellent in outdoor locations where \ntree interference exists. W. K. Oser of Out- \nside Plant Development was the engineer \non this project.\n\nSince microwaves follow a line-of-sight \npath, one of the preliminaries to setting up \na system is to assemble a set of topographic \nmaps and pick out a series of suitable hill- \ntops. But there might be a steel silo or\n\nWith the tackle fastened to the bottom of the lowest \nsection, and with a long lever on the winch handle, the \nwhole tower is lifted. The method of placing each section \nin the hoisting yoke is shown on the cover of this issue\n\nwater tank right in line; and in the plains \ncountry there are no hills to give a view \nalong the radio path. So, as a preliminary \nto laying out the Chicago-New York radio \nrelay system, the Laboratories have devel- \noped a tower for quick erection and a mi- \ncrowave transmitter and measuring receiver \nto go with the tower. With this outfit, the \npath can be surveyed by radio waves. \nNoteworthy advantage of these towers\n\nover any heretofore built is that the erect- \ning crew work entirely at, or very near, \nground level; thus professional riggers are \nnot required and the hazard of working far \nabove ground is eliminated.\n\nThe tower is lowered onto the added section and \nbolted to it; tapered bushings in the ends of the pipes \nhold the two sections in alignment. As the tower goes \nup, additional guys are added every six sections\n\nFirst step in putting up the tower is to \nlevel off a few square feet of firm soil, then \nlay down a base plate and set on it the \nerection frame seen in accompanying illus- \ntrations. Inside the frame is placed the top \nsection of the tower. It carries a commu- \nnication antenna and its lead-wire; a pair \nof pulleys carrying a loop of light steel \ncable, and four guy cables.\n\nby hand, and the second section placed \nunder it. But from then on, as shown in the \nillustration on the cover of this issue, a \nhoisting yoke is fastened to the bottom of \neach section and the whole tower is raised \nvertically upward. Guy anchors are placed, \nunless trees are available, at a distance \nfrom the tower suited to the final height, \nand a workman at each point tends a winch \nto pay out the guy cable as the top of the \ntower rises. When the tower has been lifted \nhigh enough, the erection frame is opened;\n\nWith all connections made, the receiver starts \non its dizzy voyage. Left to right, G. H. Baker, \nG. M. Phillips and H. G. Fisher\n\nand another section is put on the base \nplate. The tower is then lowered an inch or \ntwo until tapered bushings have engaged \nthe ends of the four uprights; then the two \nsections are solidly bolted together and \nthe hoisting yoke is shifted to the bottom \nin readiness for another lift.\n\nAs the tower goes up, additional guys are \nadded every six sections (48 feet). When \nthe desired height is reached\u2014which may \nbe as much as 200 feet\u2014the erection frame \nis removed; a rolling carriage is put in\n\nThe \u201cdish\u201d can be oriented in azimuth and \nelevation. Transmission is measured at various \nheights to determine a clear path\n\nMicrowave measuring equipment and antenna controls, \nleft; on the table is the v.h.f. communication set\n\nplace on one side of the tower frame- \nwork, and a hand-operated hoisting drum \nis attached. The microwave transmitter (or \nreceiver) is attached to the carriage and \ntwo electrical cables are connected to it.\n\nWith the winch, the carriage is raised while \nthe two cables are being paid out from \ntheir reels. When the desired height is \nreached, the inboard ends of the cables, \nwhich have been stowed inside the reels, \nare unrolled and led to the apparatus \nshown in the illustration at the left.\n\nElectrical equipment consists of a micro- \nwave oscillator with antenna at one end of \nthe path to be surveyed, and a receiver- \ndetector-meter set at the other end. To \nline up the transmitting and receiving an- \ntennas, each is provided with two motor \ndrives; one can swing the \u201cdish\u201d through \n180 degrees in azimuth and the other \nthrough 5 degrees in elevation. There is a \ntwo-way radio communication set similar \nto those used for mobile radio. Power for \nall equipment, including night lights for \nairplanes, comes from duplicate gasoline \nalternators at each end, which deliver 115- \nvolt, 60-cycle, a-c.\n\nThe tower was designed by R. R. Andres \nand A. H. Lince of Transmission Equip- \nment under the supervision of T. J. Grieser. \nThe electrical system was designed by \nG. M. Phillips and G. H. Baker of Trans- \nmission Engineering under the supervision \nof J. M. Barstow.\n\nHigh-frequency induction furnace in the \nMetallurgical Laboratory at Murray Hill. \nO. J. Barton is shown removing slag and \ncleaning out the crucible after pouring a \nheat in preparation for the next experi- \nmental melt\n\nOn August 15, Bell System radio tele- \nphone service became available to pas- \nsengers on certain trains of the Baltimore \n& Ohio and Pennsylvania Railroads op- \nerating between New York and Washing- \nton. This radio extension of telephone \nservice\u2014being operated only experimentally \nat the present time\u2014has been made feasible \nby the recent installation of the urban mo- \nbile radio system* in many of the larger \ncities. The same mobile receivers and trans- \nmitters are used, and all calls pass through \nthe mobile service operators in the city \nnearest the train when the call is made.\n\nOn the Baltimore & Ohio Railroad, \nthe equipment is installed in the club car \non the \u201cRoyal Blue,\u201d which makes one round \ntrip between the above points daily. On \nthe Pennsylvania Railroad, it is located in \nthe club car on both the northbound and \nsouthbound \u201cCongressional Limiteds.\u201d Since \nthese cars return northward on the \u201cPresi- \ndent\u201d and southward on the \u201cSpeaker,\u201d the \nservice will be available on these trains \nalso in the directions indicated, as well as \non the \u201cCongressional Limited\u201d in both \ndirections. In the vicinity of New York, \ncalls are completed through the Newark \nmobile service operator. After the train \npasses out of range of the Newark station, \nthe calls are handled through Philadelphia, \nBaltimore, or Washington as the train \npasses into the areas within the range of \nthese stations. Since, at the time the service \nwas inaugurated, no one radio frequency \nwas used for urban mobile service at all \nfour of these cities, each train is provided \nwith two sets of radio equipment so that \ncommunication can be established on either \nof two channels as desired.\n\nFig. 1\u2014William R. Triem, General Superintend- \nent of Telegraph of the Pennsylvania Railroad, \nplacing a call from the telephone booth on the \nlounge car \u201cJohn Adams\u201d with the recently \ninstalled mobile radio telephone apparatus\n\nwith the aid of the cabinet shown in Fig- \nure 2, which is called Control Unit B. On \nits face, this unit has a lock to prevent un- \nauthorized use of the equipment, a key to \nselect either of the two channels available, \nand a three-position switch to permit the \nattendant to talk to a mobile operator or \nto the occupant of the telephone booth, or \nto connect a call through to the booth. In \naddition, there are a number of supervisory \nlamps and buttons for turning the trans- \nmitter on and for signaling the booth. The \nrelays required for the switching operations \nare mounted in Control Unit A, which is \ninstalled with the radio equipment.\n\nIn the usual mobile system, the handset \nin the vehicle is equipped with a button \nwhich is pressed for talking and released \nfor listening. This permits a single antenna \nto be used both for transmitting and re- \nceiving; the antenna is switched to either \nthe transmitter or receiver by the action of \nthe press-to-talk switch. Since only one \nperson or a very few people use the equip- \nment in each mobile installation, they soon \nget accustomed to operating the press-to- \ntalk switch.\n\nFor train telephone service, on the other \nhand, those using the telephone will for \nthe most part not be familiar with press- \nto-talk operation, and thus the equipment \non the trains has been modified to permit \nthe handset to be used in the ordinary \nway. This is accomplished by using two \nantennas\u2014one at one end of the car for \nthe transmitter, and one at the other end\n\nFig. 3\u2014Train telephone operator, Miss Helen K. \nReese, at the control point for the mobile radio tele- \nphone \u2018service in Pennsylvania lounge car \u201cJohn \nAdams.\u201d At her right is the Control Unit B shown \nin Figure 2. The window just to the right of the \nhandset mounting opens to the telephone booth \non the other side of the wall\n\nFig. 4\u2014A. S. Hunt, Chief Engineer of Communica- \ntions and Signals of the B. & O., talks over the \nattendant\u2019s telephone in the lounge car of the \nB. & O.\u2019s \u201cRoyal Blue\u201d\n\nFig. 5\u2014Installing one of the antennas on the Pennsyl- \nvania Railroad lounge car \u201cJohn Adams.\u201d J. L. Lindner, \nBell Telephone Laboratories, at the left, and G. E. \nGoldner of The Diamond State Telephone Company\n\nand 388A radio receivers{ operating from \n12 volts d-c are employed on the trains, \nthere being two transmitters and two re- \nceivers on each car. The 12-volt source \nin each installation is supplied by the rail- \nroad from a specially installed 12-volt stor- \nage battery. This battery is charged when \nnot in use from the regular 32-volt car \nbattery through suitable voltage dropping \nand regulating equipment. The 12-volt bat- \ntery is connected to the radio equipment \nthrough a small power panel which con- \ntains a disconnect switch and fusing equip- \nment. A large part of the design work for \nthe circuits and equipment was carried \non by H. S. Winbigler, while the two con- \ntrol units together with the power panel \nand coaxial filter previously mentioned \nwere produced in the Graybar-Varick \nModel Shop under the direct supervision of \nC. P. Bartgis.\n\nA block schematic of the arrangement \nemployed in each car is shown in Figure \n6. The radio sets, filter, power panel, and \nControl Unit A are mounted together. On \nthe Baltimore & Ohio Railroad, these \nunits are contained in a built-in compart-\n\nRECEIVING TRANSMITTING \nANTENNA ANTENNA \nCOAXIAL \nFILTER \n= \nRADIO 2 > RADIO \nRECEIVERS TRANSMITTERS \n-----5 \nr \ni \nA B \nCONTROL \nUNITS \n' \nPOWER \nFig. 6\u2014Block schematic of the \ntelephone circuit on the train == ATTENDANT PASSENGE\n\nment next to the passenger booth. On the \nPennsylvania, a special equipment cabi- \nnet is mounted in the interior of the car \nnear the attendant\u2019s position, as shown in \nFigure 7.\n\nThe operation of the telephone service \non the Baltimore & Ohio Railroad is un- \nder the supervision of The Chesapeake and \nPotomac Telephone Company of Baltimore, \nwhile that on the Pennsylvania is handled \nby The Bell Telephone Company of Penn- \nsylvania. Since the trains lie over at night \nin yards outside the territories of these \ncompanies, maintenance assistance is be- \ning furnished by several other Bell System \noperating companies.\n\nFig. 7\u2014Transmitters, receivers, and power sup- \nply for the mobile telephone equipment on \nPennsylvania Railroad trains are installed in \na single cabinet\n\nTHE AUTHOR: Newron Monx interrupted his \ncollege work at Harvard to spend two years with \nthe Signal Corps during World War I. After grad-\n\nuating with an A.B. degree in 1920, he entered \nthe Harvard Engineering School, from which he \nreceived the B.S. degree in Communication Engi- \nneering in 1922. He then joined the Department \nof Development and Research of the American \nTelephone and Telegraph Company, where lis \nearly work was concerned with interference preven- \ntion investigations. Later he became interested in \ncarrier development, and continued this work after \ntransferring to the Laboratories in 1934. Just prior \nto World War II, he devoted most of his time to \napplying carrier equipment to railroad communica- \ntion systems. During World War II he was active \nin the development of voice-frequency and carrier \nsystems for the Signal Corps. Since the war, he \nhas been in charge of a group developing radio \nfor railroads and airplanes.\n\nSince their development in the years be- that could be produced in relatively large \nfore the war, broadband carrier systems quantities more economically than the orig- \nhave come steadily into more extensive use, inal design. These new 219-type channel \nand modifications have been made in their filters, such as shown in Figure 1, require \ncircuits and apparatus as advances in the less than two-thirds as much mounting \nart have made them desirable. One such space as the earlier 75-type filters. \nMore compact than the earlier design, \nthe configuration of the new filter also re- \nquires fewer coils and capacitors. This type \nof filter is critical of adjustment, and the \ndesired performance is realized only if the \nelectrical characteristics of the component \nelements are maintained close to their theo- \nretical values. Hence, it is only with the - \ndevelopment of the more precise wire-sup- q \nported crystal unit replacing the clamp- \ntype crystal unit that it has been possible \nto utilize this type of filter configuration. \nThe new filter consists of only one lattice- \ntype section in contrast to two in the earlier\n\nFig. 2\u2014The new 219-type filter, at the left, requires \nonly two-thirds the mounting space of the former 75- \ntype filter shown at the right\n\nadvance was the wire-supported, hermeti- \ncally sealed crystal* unit replacing the pre- \nvious clamp-type construction. These new \nunits are not only more economical in the \nspace they require, but also they can be \nmore precisely adjusted for effective induc- \ntance and resonant frequency. It was these \nfeatures that offered an opportunity to de- \nvelop a new channel filtert of reduced size\n\ndesign. The four coils used in the old filter \nare replaced by two less expensive coils in \nthe new design, and four of the capacitors \nare omitted. These differences, and the re- \nsulting decrease in size, are evident in Fig- \nure 2, where the new filter is at the left and \nthe old one at the right.\n\nIn the 75-type filter there was one crystal \nconnected electrically in each of the four \nbranches of each filter section, and thus \neight crystals in all. The two crystals in the \nseries branches of one section, however, \nhad the same resonant frequency. This was\n\nalso true of the diagonal branches of the \nsame section, but the series branch reso- \nnance differed from the diagonal branch \nresonance, and the resonances for the cor- \nresponding arms of the two sections dif- \nfered. Four resonant frequencies were thus \nrequired for the two sections. Since the \ncoating on each side of the crystal plate \nwas divided into two parts, and thus each \nplate became electrically two crystals vi- \nbrating in unison, only four plates were re- \nquired in each filter to provide the eight \ncrystals required in the electrical design. \nLikewise, the new filter also employs eight \ncrystals with four resonant frequencies. In \neffect, it incorporates the two lattice sec-\n\nparallel in each of the four branches, as \nshown in Figure 3. The two crystals of each \nsealed crystal unit are connected in differ- \nent lattice branches, as indicated by the \nmarkings f,, f., f, and f, of Figure 3, where\n\nFigs. 4 and 5\u2014The four glass-sealed crystal \nunits are cemented into a rectangular molded \nbase. The eight elements of these units forming \na filter are interconnected in the base to two \ninput and two output leads\n\nthe two crystals of the same unit are sep- \narated to bring them into the branches of \nthe filter into which they are connected. \nThe four glass-sealed crystal units that \nprovide the basis of the lattice structure \nare mounted at the corners of a rectangular \nmolded base, as shown in Figure 4. Four \nleads from each unit pass through the base \nand are interconnected, as shown in Figure \n5. In Figure 6, where the same frequency \nmarkings are used as in Figure 3, the inter- \nconnections are shown schematically. Two \ninput and two output leads are brought out \nfrom the interconnected points. The molded \nbase of the crystal unit is fastened, in turn, \nto one side of a metal plate forming the \nchassis of the filter, as shown in Figure 7.\n\nOn the other side of the plate are mounted \nthe capacitors and resistors associated with \nthe crystal units. As will be noticed from \nFigure 3, there are three adjustable capaci- \ntors: a double capacitor adjusting the ca- \npacitances in one series and one diagonal \nbranch differentially, and two single capaci- \ntors\u2014one at the input and the other at the \noutput of the lattice. These are the three \nelements mounted on the upright base, as \nevident in Figure 7. The fixed capacitors \nand the resistors associated with the coils \nare mounted on the bottom of the chassis.\n\nThe inductances shown at the ends of the \nfilter in Figure 3 are air-core coils mounted \nin the cylindrical copper containers, evident \nin Figure 7. The high coil-impedance called \nfor by the filter design requires that the \nparasitic capacitances between the wind- \nings, and between the windings and case, be \nkept as small as possible. This is accom- \nplished by winding the coils in four sec- \ntions, as shown in Figure 8.\n\nA reactance-frequency diagram for the \nfilter is shown in Figure 9, where the reac- \ntance of the series branches is shown solid \nand that for the diagonal branches, dashed. \nThe pass-band for the filter spans the re- \ngion where the reactances of the series and \ndiagonal branches are of opposite sign,\n\nFig. 6\u2014Interconnections for the eight elements of \nthe four crystal units incorporated in the filter. Fre- \nquency markings are the same as in Figure 3\n\nA lattice-type filter section is a \u201cbridge\u201d \ncircuit, as indicated in Figure 10, and when \nthe reactances of all branches are alike, \nboth in magnitude and sign, the bridge is \nbalanced, and no energy from the input \ngets through to the load. Over the pass- \nband this balance never exists, as evident \nfrom Figure 11, but beyond the pass-band \nit exists whenever the reactance of the se- \nries branches equals that of the diagonal \nbranches in magnitude. In the attenuating \nregion, the reactances are always alike in \nsign, as evident from Figure 9. In this latter \ndiagram, the series and diagonal branch \nreactances are shown equal except for a \nshort distance just beyond the pass-band, \nbut if the curves had been drawn to a \nlarger scale, the two reactances would be \nseen to cross and recross each other\u2014never \ndeparting very much, but exactly alike only \nat isolated points. At each crossing point \nthere is a peak of attenuation due to bal- \nance of the bridge circuit, while between \npeaks the attenuation depends on the sep- \naration of the curves of reactance for the \nseries and diagonal branches, the greater \nthe separation, the lower the attenuation.\n\nFig. 7\u2014The 219-type filter showing capacitors and re- \nsistors associated with the crystals and coils of the \nfilter. The coils for the input and output of the filter \nare in the cans at each side of the crystal units\n\nFig. 8\u2014To reduce parasitic capacitances, the air-core \ncoils are wound in four sections\n\ntances and resonant frequencies of the crys- \ntal units is required to accurately control \nthe location of these crossings and the sep- \narations of the two reactance curves. The \nrequired accuracy could not be attained. \nwith the clamp-type crystal units; hence, it \nwas not until the wire-supported crystal \nunit was developed that this compact type \nfilter design could be utilized.\n\nIn referring to the \u201cbridge\u201d balance of \nthe four reactance branches, only reactance \nwas mentioned because the arms are pre- \ndominantly reactive. Associated with each \nof these reactances, however, there is some \ndissipation, and to obtain a high peak loss \nin the filter at a definite frequency, there \nmust be a balance of the conductances of \nthese four branches as well as of their re- \nactances. Conductance unbalance of these \nbranches in the attenuating region of the \nfilter will degrade filter performance to the \nsame extent as reactance unbalances. \nWhereas the tuning of the differential con- \ndenser balances these reactances at one fre- \nquency, no means is provided for balancing \nthe conductance component. Extreme care \nmust be exercised, therefore, to keep these \nfour branches alike. Even though the crys- \ntal units are surrounded by dry air and en- \nclosed in sealed glass containers, it is still \nnecessary to protect the filter against atmos- \npheric conditions of high relative humidity. \nA trace of salt or other contamination \nacross the terminals in the molded base of\n\nFig. 10\u2014The four branches of a lattice-type filter, \nshown at the left, form a bridge circuit as indicated\n\nFig. 11\u2014Insertion-loss frequency characteristic of the \n219-type crystal channel filter\n\nthe crystal units or across the ceramic in \nthe air capacitors will make the insulation \nresistance highly sensitive to the moisture \nin the adjacent air and may result in large \nconductance unbalances in the lattice. To \nguard against this condition, only homoge- \nneous low-loss materials are used for the \ninsulating materials within the lattice struc- \nture. The insulating parts assembled in the \nfilter are kept clean, and the filters are ad- \njusted, tested and sealed hermetically when\n\nTHE AUTHOR: E. S. Wits joined the Technical \nStaff of the Laboratories in July, 1927, after re- \nceiving the degree of M.A. from the University of \nMissouri. Since then he has been engaged in the \ndevelopment of various types of transmission net- \nworks, such as electric wave filters and equalizers. \nHe had an active part in the first use of quartz \ncrystal elements as applied to various types of filter \ncircuits. In the past few years he has been chiefly \nengaged with problems associated with the crystal \nchannel filter for broadband carrier systems. Re- \ncently, this work has involved the application of \nsynthetic crystal units to filter circuits.\n\nsurrounded by an atmosphere in which the \nrelative humidity is not greater than 40 \nper cent.\n\nThe 219-type crystal channel filter is the \nfirst commercial application of lattice-type \nfilter circuits in which more than one crys- \ntal unit is used in each branch of the lattice. \nThe economies it permits in space, material \nand labor are very desirable in view of the \nincreasing use of broadband carrier sys- \ntems in the telephone plant.\n\nAt the Chester field laboratory, J. H. \nShuhart is preparing a test of various line- \nwire materials to study their fatigue prop- \nerties under vibrations induced by winds \nJanuary 1948\n\nWhile struggling with the problems of \nlong-distance telephone lines during the \nearly teen-age years of this century, Bell \nSystem engineers were considerably ham- \npered by the reflection of electrical waves \nat impedance irregularities along the line. \nThese electrical echoes proved serious ob- \nstacles, particularly to the application of \nrepeaters, and for a long time limited their \nuse in circuits. The objectionable charac-\n\nOne of the antennas used with the terrain \nclearance indicator. These antennas were at- \ntached to the under side of the plane as shown \n\u2014one to transmit and one to receive\n\nteristics of echoes on telephone lines was \nintimately associated with the distance be- \ntween the points of origin and reflection. \nIn fact, the effect of the echoes in produc- \ning a \u201chumpy\u201d impedance-frequency curve \nhad been used by telephone engineers to \nmeasure this distance, and thus locate the \nposition along the line of a defective load- \ning coil or other \u201cirregularity.\u201d\n\nThe, work that had been carried on, and \nthe precision and beauty of the reflection \nphenomena, suggested to Lloyd Espenschied \nthat electrical reflections had merit in their \nown right, and that they could be put to \nmore positive use. One of the early appli- \ncations that occurred to him was as a \nwarning and safety system for railways.\n\nHis proposal was to send an electric wave \nalong the track ahead of the locomotive. \nA broken rail or another train would form \nan impedance irregularity that would re- \nflect the wave, and the arrival of the re- \nflected wave at the locomotive would not \nonly give warning of danger ahead, but \nwould permit the distance to the danger \nto be determined and the train automat- \nically controlled. A patent on such a system \nwas applied for in 1919, and patent No. \n1,517,549 was granted in 1924. Among the \nearly ideas entertained at the time was that \nof the use of a frequency-swing oscillator \nand beat-frequency reception for indicat- \ning the distance \u2014 principles later used in \nthe radio altimeter.\n\nA number of allied and modified sys- \ntems were devised in the immediately fol- \nlowing years, but meanwhile the airplane \nwas rapidly assuming an important posi- \ntion in the transportation field, and an elec- \ntrical reflection system that would give an \nairplane a continuous indication of its \nheight above ground seemed highly desir- \nable. Barometer readings, the only indica- \ntions of distance available for this purpose \nheretofore, gave the height above sea level, \nand thus were of little use when the ele- \nvation of the land beneath the plane was \nincreasing rapidly, as when the plane was \napproaching a chain of mountains or a \ncity with tall buildings. Espenschied there- \nfore proposed an electrical reflecting sys- \ntem for use by airplanes using radio instead \nof rail transmission. At that time, however, \n1926, a really practical terrain clearance \nindicator could not be built, largely be- \ncause suitable radio instrumentalities were \nnot available. It was felt that vacuum tubes \ncapable of operating at frequencies some \nfifty times higher than existing commercial \ntubes could handle would be required be- \nfore a satisfactory system could be built.\n\nAlthough altitude determination was con- \nsidered at various times following Espen-\n\nschied\u2019s first suggestion, it was not until \n1930 that patents were applied for, and it \nwas not until June, 1936, that patents \n2,045,071 and 2,045,072 were granted to \nEspenschied for an airplane radio altim- \neter. Considerable commercial interest was \naroused, and in the following year a \ngroup of Laboratories\u2019 engineers headed \nby R. C. Newhouse undertook to develop \ncommercial apparatus for Western Electric.\n\nIn 1938, a number of demonstration \nflights were made, culminating on October \n8 and 9, 1938, in a joint demonstration* \nfor the press by the Laboratories, the West- \nern Electric Company, and the United Air-\n\nlines. During one of the early flights, it was \nnoticed that in addition to height above \nground, variations in the wave pattern on \nan oscillograph gave indications of the \ncharacter of the reflecting surface beneath. \nThis suggested a still wider field of use, \nand proposals were made that similar ap- \nparatus, designed to project the beam \nahead, could be used to give warning of \napproaching aircraft, mountains, or other \nobstructions. Experiments of this nature \nwere carried out this same year in New \nYork harbor. In the avalanche of war de- \nvelopments that shortly followed, these and \nother suggestions were used by the Lab- \noratories in their many radar projects.\n\nTo study microwave propagation through circular waveguides, it is necessary to begin with a \nwaveguide which is as nearly straight as possible. So Radio Research constructed this copper tube, \nshown above, 1,200 feet in length, on supports which permit its accurate alignment. At the lower \nleft, A. P. King closes line short-circuiting switch which disconnects the line beyond the switch. \nThe other view shows R. Bown, H. T. Friis, M. J. Kelly and A. B. Clark inspecting line\n\nIn the years before the war, a number of \nprivate-line teletypewriter systems had \ngrown to considerable size\u2014some of them \nspanning many states or even the entire \ncountry. A case in point is the Republic \nSteel Corporation. Its main office is at \nCleveland, where it also operates steel \nmills, but it has other plants and offices ex- \ntending from Boston to San Francisco and \nfrom Buffalo and Detroit to Birmingham \nand Houston. Since about 1925, these of- \nfices and mills had been linked for teletype-\n\nFig. 1\u2014The control panel for the private-line teletype- \nwriter network at the Cleveland office of the Republic \nSteel Corporation\n\nwriter communication through a system of \nprivate lines, which in more recent years \nhad been augmented by TWX service.\u201d \nSwitching was all done manually, however. \nA message from Chicago to New York, for \nexample, would first go to an operator in \nCleveland who would establish manually a \nconnection to New York. Twenty-six pri-\n\nvate-line teletypewriter circuits were in \nuse, and traffic over these circuits was ex- \ntended by TWX service to points all over \nthis country.\n\nAbout 1940, it was felt that faster and \nmore economical service could be secured \nby the use of automatic switching, and after \nan extensive analysis by the Bell System, \nthe 81B1 teletypewriter private-line switch- \ning system was decided upon. The original \ninstallation was completed in October, \n1941, just in time to handle the rush of war \nbusiness. Besides greatly increasing the \nspeed of service, this new system made it\n\nline teletypewriter circuits from twenty-six \nto thirteen. Seven of these thirteen are con- \nnected to the automatic system, and each \nof the circuits serves a number of receiving \nand sending stations.\n\nIn fundamental principles, this type of \nautomatic private-wire teletypewriter \nswitching differs basically from telephone \nswitching. With the latter, the complete \ncommunication path between the originat- \ning and called stations must be idle before \na connection can be established and the \ntransmission of intelligence begun. When \nthe paths are not idle, repeated time- \nabsorbing attempts using expensive facili- \nties must be made to reach the called sta- \ntion. In the teletypewriter system, however, \nonce a message has been perforated in a \ntape and this tape placed in a transmitter, \nno further action by the sender is required \nfor the transmission of the message. The \nsender's transmitter sends the message to \nthe switching center as soon as the path to \nthe center is available. At the switching \ncenter, facilities are provided to receive and \nsort the messages and to store them until \nthe outgoing path to the called party is \nfree. In this manner, messages are carried \nas far on their way to the ultimate receiver\n\nas idle facilities permit before they are de- \nlayed to wait for a free path.\n\nAt the switching point, Cleveland for the \nRepublic Steel Corporation, a group of \ncrossbar switches* is provided that has in- \ncoming circuits connected to their horizon-\n\ntals and outgoing circuits to their verticals, \nso that any two circuits may be connected \ntogether. Director and sequence circuits, \nconsisting chiefly of relays, control the op- \neration of the crossbar switches. Local tele- \ntypewriter transmitters and receivers are \nassociated with the crossbar switches to \nhandle messages originating or terminating \nat the switching office and for certain su- \npervisory services, but the major unit of \nthe switching system is the 14-D reperfo- \nrator transmitter, which will be described \nin more detail in a subsequent article. In \nbrief, this unit consists of a combined typ- \ning reperforator and a transmitter. The for- \nmer perforates a tape in accordance with \nthe received signals and at the same time \nprints the message on the tape, while the \nlatter receives the tape punched by the re- \nceiver and retransmits the message, includ- \ning the last character punched. Between the \nreceiver and transmitter are facilities for \nstoring tape.\n\nIncoming messages for passage through \nthe switching system arrive at one of these \nmachines, and a tape is punched as a mes- \nsage comes in. The first two characters of \nthe message will be a code indicating the \nultimate destination, and the last two char-\n\nacters indicate a disconnect. The transmit- \nter forwards this code to the control circuit \nof the crossbar switches, pauses until it has \nbeen notified that the connection has been \nmade, and then transmits the message. __\n\non the crossbar switches, and each vertical \nis connected to the receiving side of a re- \nperforator transmitter. If one of these is \nreceiving, a message from the incoming side \nof the office will be connected to the other, \nwhich will punch a tape and store it. This \nmakes that message available for transmis- \nsion as soon as the outgoing line is free, \nand permits the incoming line to send its \nfollowing messages to other outgoing lines \nthat may be ready to accept them, thus \nspeeding up trans-office transmission. Dur- \ning traffic peaks a back load of messages \nmay accumulate awaiting transmission, and \nthese will be sent out in approximately the \norder in which they were received as the \noutgoing line becomes available, thus keep- \ning the outgoing lines in use continuously. \nEach_ reperforator transmitter and _ its \nmounting arrangement can store about \nseventy-five feet of tape, which corresponds \nto about 1,500 words. Since the ordinary \nmessages and orders run from fifty to three \nhundred words, ample storage is thus pro- \nvided to carry the system over any ordi- \nnary peaks of load.\n\nBesides its reperforator transmitter, each \nincoming circuit has a receiving teletype- \nwriter that can be connected to the circuit \nin place of the reperforator transmitter\n\nFig. 3\u2014Each cabinet houses two reperforator \ntransmitters with tape storage space in the \nlower section\n\nwhen the message is for the local office. \nWhen the reperforator transmitter recog- \nnizes by the code of a message that it is for \nthat office, it operates relays to disconnect \nitself from the circuit and to connect in the \nreceiving teletypewriter.\n\nThe transmitters at the distant stations \nare under control of the switching office, \nwhich automatically starts those on any \ngiven line in turn by start patterns of tele- \ntypewriter characters sent from the switch- \ning office. Arrangements are also provided \nto start these transmitters by the manual \noperation of keys at the switching office.\n\nOne of the fundamental principles of this \ntype of teletypewriter switching is that no \nmessage shall be lost or delayed more than \na few minutes. Provisions are made, there- \nfore, for taking care of such errors as incor- \nrect or missing switching codes that other-\n\nThere are occasionally times when a par- \nticular plant or office, perhaps because of \nholidays or other such circumstances, is not \nable to receive messages. To take care of \nsuch situations, keys at the control board \nmay be operated to permit messages to any \nstation on an outgoing circuit to be directed \nto a willful intercept circuit. This includes \none or more reperforator transmitters which \nwill store all messages received until they \ncan be transmitted to their destination. \nWhen this time comes, the messages are\n\nThe control panel from which the willful \nintercept is controlled is the nerve center \nfor the entire system. It has lamps to indi- \ncate what machines are sending, what cir- \ncuits are busy or idle, and all the many \nthings that those responsible for the opera- \ntion of the system should know. From here \nalso is controlled the order in which the \nvarious stations on a single circuit are al- \nlowed to transmit. The control panel for \nthe Cleveland office of Republic Steel is \nshown in Figure 1.\n\nA general view of the teletypewriter oper- \nating room in Cleveland is shown in Figure \n2. In the left foreground are the receiving \nteletypewriters that receive messages for \nthe Cleveland office. To the right of these \nare machine cabinets containing the reper- \nforator transmitters on the receiving cir- \ncuits. As shown in Figure 3, each cabinet \nhas two reperforator transmitters with stor- \nage spaces for their tapes. Next to the right \nis the control panel and then the reperfora- \ntor transmitters on the outgoing circuits. \nIn the cabinets at the rear are the crossbar \nswitches and the various relay control cir- \ncuits. At the right are order writing and \noriginating machines where messages from\n\n== PRIVATE TWO-WAY \nPRIVATE ONE-WAY \nBELL SYSTEM \n\u00ae AUTOMATIC SWITCHING CENTER\n\nFig. 4\u2014Republic Steel Corporation\u2019s private-line teletypewriter network. It consists of six main duplex \ncircuits and a single circuit to Gadsden, Alabama\n\nthe Cleveland office are originated. The su- \npervisors desk where the intercepted mes- \nsages are handled is at the rear of this \ngroup just in front of the relay cabinets. \nThe complete teletypewriter network of \nthe Republic Steel Corporation is shown in \nFigure 4. There are six main duplex circuits,\n\nTHE AUTHOR: W. M. Bacon, after graduating \nwith the E.E. degree from Cornell in 1930, joined \nthe D & R and was concerned with teletypewriter \ndevelopment problems. This work was continued \nwith the Laboratories telegraph facilities group \nafter the 1934 consolidation, and has involved main- \ntenance, repair, and installation methods and _ pro- \ncedures as applied to teletypewriters. From 1938 \nto the present time, he has been engaged in the \ndevelopment of private-wire teletypewriter systems, \nincluding full automatic and manual switching sys- \ntems. During the war years he was engaged in \ndeveloping teletypewriter cipher arrangements for \nthe Armed Forces.\n\neach with a number of stations connected \nto it and one single circuit to Gadsden, Ala. \nAt Cleveland, Chicago, New York, and \nGadsden, the network is associated with \nthe Bell System teletypewriter service, and \nmessages to offices not on their private-line \nnetwork are transmitted over the TWX.\n\nEATTLE \n~ \n~ \n~ \n| \n| \nx. \n~ \n~. \nSAN FRANCISCO | INDIANAPOLIS MASSILLON \nKANSAS CITY 4 PCINCINNATI \nST. LOUIS \n| \nLOS ANGELES \n~~~frwxbop GADSDEN : \n~-SBIRMINGHAM \nPos \n: \nHOUSTON o% \n| \n| \n| | \n2 \n|\n\nIn the past fifteen years, the waveguide \nhas been developed from an electrical curi- \nosity to an important medium for micro- \nwave transmission. Progress was especially \nrapid during the war years, since many of \nthe most important developments in radar \nwould have been virtually impossible with- \nout waveguides. It is now possible to judge \nwith some perspective the rdle that wave- \nguides will play in communications devel- \nopments of the near future. Hollow-pipe \nguides do not assume a thoroughly practical \nform until the wavelength is shorter than \nabout thirty centimeters. At this upper \nlimit, the waveguide has supplanted coaxial \ncable in some instances. For shorter wave- \nlengths, it becomes increasingly attractive, \nso that at ten centimeters it is the preferred \nmedium for the efficient, shielded transmis- \nsion of wave power, and at three centi- \nmeters it is almost exclusively used.\n\nAlthough the earliest work emphasized \nwaveguides as a form of transmission line \nfor radio waves, it was apparent from the \noutset that waveguide structures would be \nrequired also to serve as numerous basic \nelectrical components. At low frequencies, \nthese components include such elements as \ncoils, capacitors, resistors, transformers, re- \nlays, and switches. It was soon found that \nwaveguide counterparts of these low-fre- \nquency elements could be constructed to \noperate in the microwave region. To the \nuninitiated, the electrical behavior and \nphysical form of these new components ap- \npear novel and even revolutionary.\n\nThe investigation of waveguide transmis- \nsion and of related circuit elements, as car- \nried on by Dr. G. C. Southworth and his \nassociates, opened up vistas of a radically \nnew communications art for the Bell Sys- \ntem. Expanding effort in microwave re- \nsearch and development is now bringing\n\nabout a reduction to practice. The engineer- \ning of waveguide components, for example, \nis already sufficiently far advanced to per- \nmit construction of a microwave circuit be- \ntween New York and Boston in a manner \nonly dreamed of ten years ago.\n\nOne ot the waveguide components which \nis finding a wide variety of uses is the four- \narm junction shown in Figures 1 and 2. If \nwave power is sent into this junction from \nthe arM s, Figure 2, half of the power flows \ninto each of the ARM 1 and ARM 2, with no \npower appearing in the arm p. Conversely, \nif waves are sent into the junction from the \nARM P, the power is again equally divided \nbetween ARM 1 and ARM 2, with none flow- \ning into the aRM s. Thus, ARM s and ARM P \nare balanced with respect to each other, pro- \nvided ARM 1 and ARM 2 are terminated in \na symmetrical manner. Since the junction \nprovides balance reminiscent of the familiar \nhybrid coil, it is called a hybrid junction.*\n\n*The term \u201cMagic Tee,\u201d which has also been \napplied to this construction, is felt to be a misnomer.\n\njunction by itself. An important refinement \nis realized, however, when reactive tuning, \nas exemplified by the metal post and rod in \nFigure 2, is associated with the junction. \nThese added elements improve the imped- \nance match of the junction to each of the \nfour arms, and thereby automatically bring \nabout a high degree of balance between \nARM | and ARM 2.\n\nA qualitative explanation of the hybrid \njunction can be arranged to bring out some \nof the more important aspects of modern \nwaveguide theory and practice. It will be \nrecalled* that radio waves can be freely \npropagated within any conducting tube, so \nlong as the wavelength is shorter than a \ncutoff wavelength determined by the cross- \nsectional size and shape of the pipe. There \nis, indeed, an infinitude of transmission \nmodes, corresponding to different patterns \nof lines of force, which can be established \nand maintained for wave transmission, and \nthus an infinitude of different cutoff wave- \nlengths. For reasons of simplicity, however, \nwaveguide developments have so far been \nconfined almost exclusively to the mode of \nlongest cutoff wavelength, the so-called \ndominant wave, in pipes so proportioned \nthat no other mode can be freely sustained. \nAlso, for practical reasons, only circular and \nrectangular shapes are in common use. In \nFigure 3, lines of electric intensity in the \ndominant wave are shown for tubes of cir- \ncular, square, and oblong cross-section. A \nclose resemblance to radio waves in free \nspace is evident.\n\nThere is an important difference between \nsymmetrical and nonsymmetrical pipes, \nhowever, with respect to the permissible \norientation of the plane of polarization. \nPatterns are shown in Figure 3 for both \nvertical and horizontal polarization. In the \ncircular and square pipes, the configuration \nmay have any orientation, depending upon \nthe manner in which the waves were ini- \ntially launched. In an oblong pipe, however, \nthe cutoff wavelength corresponding to the \ncross-polarized oscillation can be depressed \nbelow the operating wavelength, and only \nthe fixed orientation shown can be freely \ntransmitted.\n\nAn oblong shape is thus preferred when- \never there is any chance that the polariza-\n\nARM P \nFig. 2\u2014Perspective cut-away view of waveguide hybrid \nshowing \u201crod\u201d and \u201cpost\u201d used in impedance matching\n\nFig. 3\u2014Electric field configuration in circular, square, \nand oblong guides\n\nMETAL ROD \nFig. 4\u2014A \u201cpost\u201d at the upper left is equivalent to a \nshunt capacitor, while a \u201crod,\u201d lower left, is equivalent \nto a shunt coil\n\ntion may be rotated in an uncontrolled \nfashion. Experience has shown that un- \navoidable irregularities or deliberate distor- \ntions, such as bends, do tend to rotate the \nplane of polarization in circular waveguides \nto such an extent as to interfere seriously \nwith the normal operation of the system. \nFor the most part, therefore, waveguide \ncomponents have been developed with rec- \ntangular guide. Best results are generally \nobtained when the proportions of the cross- \nsection lie in the vicinity of 1 by 2, with \nthe larger inside dimension equal to about \nthree-quarters of a wavelength. \nWaveguide studies are greatly facilitated \nby the close analogies that can be estab- \nlished with conventional transmission line \ncircuits. Just as the waveguide itself is rep- \nresentable as a transmission line, so also can \nwaveguide components be replaced for \nanalytical purposes by equivalent networks \nof impedance elements in association with \nsections of line. For the simpler waveguide \nstructures, the equivalent networks are ob- \nvious upon inspection or can be readily \nderived from elementary consideration of \nelectric fields and currents within the con- \nfiguration. This is true, for instance, of the \npost and rod already shown in the hybrid \njunction. As indicated in Figure 4, the metal \npost concentrates the electric field between \nitself and the nearby wall of the guide, and \nis equivalent to a capacitance placed in \nshunt across a transmission line. With the\n\nmetal rod, on the other hand, the principal \neffect is to provide paths for conductive \ncurrent flow between the connected walls \nof the guide, with a resulting increase in \nmagnetic flux and flux linkages, and thus it \nis like an inductance in shunt across a line. \nProvided that reasonably good conductors \nhave been selected for the post and rod, \ntheir resistance components can be neg- \nlected except in extreme cases.\n\nFor more complex waveguide structures, \naccurate representation in terms of equiva- \nlent networks can often be reached only \nafter considerable study. This is true of the \nhybrid junction. For a first approach to this \nsubject, equivalent network analysis is not \nwell suited. Instead, use can be made of\n\nanalogies with optics. Here the principal \nconcern is with the spreading of a wave \nfront from one portion of the structure to \nthe others. In a straight section of guide, \nthe wave fronts are plane; distribution of \nthe electric field has been indicated in Fig- \nure 3. In the vicinity of a severe discontinu- \nity such as an abrupt junction, there is local \ndistortion, wave fronts become curved, and \nthe field may follow a complicated distribu- \ntion. An accurate knowledge of this distor- \ntion is not essential, however, in deriving \nimportant properties of the hybrid junction.\n\nwave front spreads into a hybrid junction \nfrom the ARM s. Lines of electric intensity \nare drawn in several successive positions of \nthe wave front. To show why no power \nflows into the ARM P, the electric intensity \nin a typical position of the front is indi- \ncated vectorially at two points a and B sym- \nmetrically disposed with respect to the cen- \nter of the junction. At a, the local field has \ncomponents a\u2019 toward the left and a\u201d down- \nward, while at B the components are b\u2019 to- \nward the left and b\u201d upward. From sym- \nmetry, the magnitudes of the respective \ncomponents at A and B are equal. Thus, the \ndownward and upward intensities a\u201d and b\u201d \ncancel, or, more precisely, they induce \nequal and opposite voltages in the ARM P\n\nas the wave front spreads into that arm. \nThe intensities a\u2019 and b\u2019 reinforce each \nother and tend to produce in the ARM P a \ncross-polarized wave. If the aRM P guide is \ncorrectly proportioned, however, the cross- \npolarized wave cannot be freely propa- \ngated. This shows that no power flows into \nthe aRM P as a result of the electromotive \nforce exerted at a and Bs. All other points \nin the wave front can be grouped in pairs, \nsymmetrically located with respect to the \njunction center, and the same conclusion \nholds for each pair of points. From the\n\nwave front as a whole, therefore, no power \nflows into the arm Pp, and the power which \nis transmitted through the junction must \nappear in ARM I and ARM 2, and from sym- \nmetry, in equal amounts. It will be noticed \nthat at points equidistant from the junction, \nthe polarities of the electric intensity are \nreversed in ARM 1 and ARM 2. With respect \nto phase, at least, the ARM s acts as a trans- \nmission line connected in series with the \nline corresponding to ARM 1 and ARM 2. \nWhat happens when a wave front spreads \ninto the junction from the ARM P is shown \nin Figure 6. With respect to flow toward \nthe arm \u00a7, the lines of electric intensity are \nseen to be directed so as to create only a \ncross-polarized wave in the arm s. As in\n\nthe previous case, there are, in planes off- \ncenter from the junction, additional com- \nponents arising from the curvature of the \nlines of force, but these are cancelled by \nthe mirror image components on the other \nside of the junction center. With properly \nproportioned guide for the ARM s, there- \nfore, no power flows into it from the aRM P. \nThis conclusion is in harmony with the reci- \nprocity theorem for electrical networks; if \nno power can be transmitted from ARM s \nand received at ARM P, then no power can \nflow the other way when the positions of\n\nFig. 8\u2014A balanced converter for the micro- \nwave system between New York and Boston\n\nFigure 6 also indicates how power from \nthe ARM P spreads into ARM 1 and ARM 2. \nWith complete symmetry prevailing in \nthese arms and their terminations, the \npower is equally divided between them. \nHere the geometry does not act to reverse \nthe polarity of lines of force, and at equal \ndistances from the junction, voltages in \nARM 1 and ARM 2 are in phase. From this \nconsideration, the ARM P is regarded as \nanalogous to a transmission line connected \nin parallel across a line corresponding to \nARM I and ARM 2.\n\nTo reveal clearly the balancing and \npower dividing properties of the hybrid \njunction, the picturization of wave propa- \ngation has been simplified by neglecting \nwaves reflected from various parts of the \njunction. On account of the sharp geometri- \ncal discontinuities involved, field patterns \nin the region of the junction are in reality \nvery complicated, corresponding to the \nsuperposition of waves bouncing back and \nforth within this region. All reflected com- \nponents are symmetrical or paired with re- \nspect to the center, so that they are inca- \npable of disturbing the balance or power \ndivision. The presence of reflections does, \nhowever, lead to a serious impedance mis- \nmatch of the device to the individual arms. \nThese mismatches can be eliminated by in- \ncorporating within the junction tuning ele-\n\nments, such as the rod and post visible in \nFigure 2. From an optical point of view, \nthey are regarded as sources of additional \nreflections whose amplitudes and phases \nare such as to bring about cancellation of \nthe original reflections. In terms of network \nanalysis, the rod and post are equivalent, \nas has been pointed out, to a coil and ca- \npacitor, and are disposed so as to tune out \nthe reactance associated with the equivalent \nnetwork of the junction. By a proper choice \nof tuning elements, effective neutralization \nof reflections can be obtained over a re- \nmarkably broad band of frequencies. Other \ntuning means, metal plates for example, can \nbe used in place of the rod and post. The \nparticular combination shown, developed \nby C. F. Edwards, is being widely used.\n\nWhen the hybrid junction has been \nmatched by reactive tuning, it exhibits the \nfurther customary property, that of balance \nbetween ARM 1 and ARM 2. If power is \ncaused to flow into the junction from ARM 1, \na hasty analysis of wave-front propagation \nleads to the conclusion that the power will \nbe divided in some fashion among the other \nthree arms. This is indeed true. The use of \nmatching reactors, however, automatically \nbrings about a balance between arm 1 and \nARM 2, so that power flows only into the \nARM Ss and ARM P and is, moreover, evenly \ndivided between them. This behavior, while \nnot obvious from an optical standpoint, is \na necessary consequence of the law of reci- \nprocity.\n\nHybrid junctions have been built from \nvarious sizes of guide for use at frequencies \nin the vicinity of 4,000, 6,000, 9,000, and \n24,000 megacycles per second. Constructed \nwith reasonable care, they provide an isola- \ntion of 35 to 40 decibels between the ARM s \nand ARM P, and satisfactory impedance \nmatching over frequency bands as wide as \ntwelve per cent. If unusual pains are taken \nto secure a geometrically symmetrical junc- \ntion, the balance is improved to 50 decibels \nor more.\n\nThe hybrid junction can be used in the \nsame way as its low frequency counter- \npart, the hybrid coil. The junction has, in \naddition, many diverse applications in \nmicrowave systems and in laboratory meas- \nurements, so that it already occupies as \nimportant a position in centimeter wave\n\ntechniques as does the hybrid coil in low- \nfrequency practice. New uses are constantly \nbeing discovered in which advantage is \ntaken of the balance, power division, and \nphasing afforded by the junction.\n\nPerhaps the most important use so far \nfound for hybrid junctions is in the bal- \nanced converter. Since this was also one of \nthe earliest applications, it has been \nbrought to an advanced stage of develop- \nment. The general principles underlying the \noperation of the balanced converter or first \ndetector are shown in the block diagram of \nFigure 7. Two detectors, in this case point- \ncontact silicon rectifiers, are used as termi- \nnations for ARM 1 and ARM 2 of a hybrid \njunction. The beating oscillator power is \nintroduced from the arm s, and the signal \ninput from the arm p. With this arrange- \nment, half of the signal power and half of \nthe beat frequency power are developed in \neach rectifier. The 180-degree phase shift \nbetween the signals in ARM 1 and ARM 2 by \nthe nature of the series connection persists \nin the intermediate-frequency voltages de- \nveloped at the two rectifiers. By combining\n\nthe two intermediate-frequency signals with \nproper phasing, however, all of the inter- \nmediate-frequency power can be made \navailable in a single output.\n\nAn important advantage afforded by such \na balanced detector is the isolation achieved \nbetween the signal line and the local beat- \ning oscillator. If care is taken to build a \ntruly symmetrical structure and to use rec- \ntifiers which are as nearly identical as pos- \nsible, this uncoupling corresponds to the \nbalance between the ARM s and ARM P of \nthe hybrid junction. Another advantage is \nthat the same intermediate-frequency phas- \ning which combines the two intermediate- \nfrequency signals additively causes noise \ncontributions from the beating oscillator to \nbe cancelled. For these and other reasons, \nthe balanced converter is considered the \nmost efficient and desirable microwave de- \ntector now available. A specific construction \ndeveloped by C. F. Edwards is illustrated \nin Figure 8. This is the converter that is \nnow being used in the microwave repeater \nsystem between New York and Boston, de- \nscribed in the last issue of the Recorp.\n\nTHE AUTHOR: Warren A. Tyrrett received a \nB.S. degree from Yale in 1935 and a Ph.D. degree \nin Physics in 1939. Upon graduation he joined the \nTechnical Staff of the Laboratories and was as- \nsigned to the waveguide research group. During \nthe war he developed a number of components for \nuse in Navy radar systems. Since the war, Dr. \nTyrrell has continued his work on waveguides at \nthe Holmdel Laboratory in New Jersey, and is \nengaged in the application of waveguide tech- \nniques to measurements on dielectric and metallic \nmaterial.\n\nAn ideal hearing aid would have the \nhighest acoustic performance that the ear \ncould appreciate and be small in size and \nweight. These two major objectives, how- \never, tend to oppose each other; the highest \nperformance, including high gain and \npower output, tends to require more appa- \nratus and thus larger size and weight. Since \nthe Laboratories has always considered that \nhigh quality was the primary factor, West-\n\nern Electric hearing aids have led the field \nin performance ever since the first 220- \npound set of 1923. Size and weight reduc- \ntions have come steadily, but they have \ndepended on the development of materials \nand techniques that would permit the two \nobjectives to be more effectively reconciled. \nAdvances since the war greatly helped in \nthis direction, and following the 64-type \nhearing aid. which came out immediately\n\nafter the war, the Laboratories has devel- \noped the Models 65 and 66 hearing aids. \nThese achieved greater convenience and a \nmuch smaller weight than previous designs \n\u2014six and eight ounces, respectively, includ- \ning batteries\u2014and the 66 equals the high \nperformance of the best previous aids.\n\nIn attaining excellent performance with \nvery small size and weight, dependability \nin service has not been sacrificed. The com- \nponents selected for the sets meet require- \nments of stability and ruggedness that \nassure they will stand up under conditions \nof use. In addition, many special measures \nhave been taken to protect the components \nagainst the adverse effects that are en- \ncountered with moisture and dirt.\n\nOne of the innovations of the new devel- \nopment is the provision of two sets rather \nthan a single set. In designing previous \nsets for the highest practical quality, the \nLaboratories has recognized that the user \nwith only a small hearing loss was wearing \nan aid that had much more gain and range \nthan he needed. The provision of two sets \nmeets this situation, since one of them is \ndesigned to provide the best possible hear- \ning for anyone with a correctable hearing \nloss, while the other, smaller and with less \ngain and power capacity, will satisfactorily \nserve the great majority of those requiring \na hearing aid.\n\nThese two new sets are shown in Figure \n1. Each consists of a plastic case housing \nthe transmitter, amplifier, battery, and con- \ntrols, to which a No. 724-type midget re- \nceiver and cord is attached. The Model 66 \ncase is in either two tones of grav or flesh \ncolor, while the Model 65 is either dark \ngray or flesh. The microphones for both \nsets are of the crvstal tvpe, and each bas a \nthree-stage vacuum-tube amplifier with a \nvolume control and also a tone control that\n\nmay be used to suppress the lower frequen- \ncies of background noise. The amplifier \ncircuit for the Model 66 set is shown in \nFigure 2. The amplifier for the Model 65 \nhas no feedback, and some of the capaci-\n\nThis protected network principle is ap- \nplied throughout the set. For the Model 66, \nall circuit elements between the first and \nsecond vacuum tubes are assembled in a \ncompact, self-supporting group molded in\n\nA cross-section of the crystal microphone \nis shown in Figure 3. It is sealed by a \nplastic-coated metal-foil membrane across \nits face, and by closing other potential \npoints of moisture leakage with moisture- \nproof plastic cement. Microphones for both \nsets are of essentially the same construction, \nbut that for the Model 65 is somewhat \nsmaller than that for the Model 66. The \ncapacitor and resistors of the tone control \nnetwork are housed in the microphone case. \nThis not only protects these high-imped- \nance elements from moisture but also \nshields them electrically.\n\na block of wax to form a moistureproof and \ntamper-proof network. Interconnections are \nmade by wires that are parts of the ele- \nments, thereby reducing the number of \njoints required. Many of the connections are \nwelded instead of soldered. The circuit ele- \nments between the second and third stages \nare treated in a like manner.\n\nIn the Model 65 set, this principle is ex- \ntended. The entire amplifier\u2014vacuum tubes, \ncapacitors, and resistors\u2014is wired together \nby the element\u2019s own leads, and is then \nmolded in wax to protect it from dirt and \nmoisture. The resultant package is 1% by \n1% by % inches. In case of failure of any of:\n\nthe elements, the entire network is replaced. \nThe complete chassis for the Models 65 and \n66 sets are shown in Figures 4 and 5.\n\nOne of the distinguishing features of both \nsets is the absence of distortion and a mini- \nmum of noise due to clothing. If adequate \nprecautions are not taken, the noise of \nclothing rubbing against the set, which \nreaches the microphone either as mechan- \nical vibration or as airborne sound, may so \nmask the useful sounds that the value of \nthe hearing aid is seriously impaired. In the \nnew hearing aids, it has been reduced at \nits source by the avoidance of any sharp \ncorners or projections on the case, and by \nmaking the case of a material that has a \nvery low coefficient of friction. In addition,\n\nthe microphone is protected by a resilient \nmounting against any mechanical vibration \nthat may be generated.\n\nBesides providing two sets\u2014one for those \nneeding considerable aid and the other for \nthose requiring only a moderate amount\u2014a \nfurther division of the range is provided by \nadditional options in the size of the battery \n(in the Model 66) and the type of the re- \nceiver used. There are three 724-type \nreceivers, the A, B, and C. Their character- \nistics are shown in Figure 6. The 724C has \na flat characteristic extending from below \n200 to over 4,000 cycles. The 724B gives \ngreater output, but its response falls off \nabove 3,000 cycles. The 724A has the same \nfrequency range as the 724B, but has a high \noutput peak at about 1,200 cycles which is \nhelpful in some extreme cases. Only the \n724B and 724A receivers are applicable to\n\nthe Model 65 set, but all three may be used \nwith the Model 66, as may also the 725 \nbone-conduction receiver.\n\nof increasing the sound intensity reaching \nthe ear by 55 db and of producing a sound \npressure in the ear of 116 db above the \nminimum pressure that can be heard by a \nnormal ear. The frequency range under \nthese conditions is from 250 to 4,000 cycles. \nWithin this range, the response is uniform \nwithin +6 db. This flat response may be \nchanged to one that rises toward the high \nfrequencies at either 5 or 10 db per octave \nby turning a three-position tone control \nswitch. Another control permits adjustment \nof amplifier gain over a 40 db range.\n\nA special feature of the design of the \nModel 66 set is that the battery compart- \nment may be detached and replaced by a \nsmall adaptor from which a cord extends to \nan external battery case. This permits the \nuse of higher voltage, or of larger, more \neconomical batteries, when desired. By tak- \ning advantage of this feature, amplification \nand sound pressure may be increased 4 db \nby using a 30-volt external B battery. These \ncharacteristics may also be increased 7 db \nby substituting the high-efficiency 724B \nreceiver for the 724C. This latter method \nentails some sacrifice of high-frequency re- \nsponse. In extreme cases, both of these sub- \nstitutions may be made with a resultant \namplification of 66 db, and the maximum \nsound pressure level of 129 db. This pres- \nsure is about the maximum that the ear can \ntolerate.\n\nWith the Model 65 aid, a 15-volt B bat- \ntery is employed, and when using a 724B- \ntype receiver, which is recommended in \nmost cases, the acoustic amplification is 50 \ndb and the maximum sound pressure in the \near is 115 db above normal threshold. The \neffective frequency range is from 700 to \n3,200 cycles, but even at 300 cycles, the \nresponse more than meets the requirements \nof the American Medical Association. With- \nin the effective range, the response is nor- \nmally quite flat, but may be tilted to slope \nupward toward the high frequencies at \neither 4 or 8 db per octave. Somewhat dif- \nferent gain and frequency characteristic is\n\npossible by using the 724A receivers. Over- \nall characteristics for the most commonly \nused combinations of the two sets are \nshown in Figure 7.\n\nThese two hearing aids thus complement \neach other. The Model 66 has the superior \nacoustic performance that is required by \nthose who have a severe hearing loss or \nwho want exceptionally good acoustic qual- \nity. On the other hand, the Model 65 is\n\nTHE AUTHOR: J. R. Power received a B.S. de- \ngree in Electrical Engineering from Carnegie Insti- \ntute of Technology in 1927. Entering the Appara- \ntus Development Department of the Laboratories \nthat same year, he engaged in the development of \nspecial motors and generators for use primarily in \nsound pictures. In 1935 he transferred to work on \nacoustical problems connected with noise condi- \ntions in the telephone plant and offices. Later he \nspent some time on the preliminary development \nof a new telephone ringer. This was interrupted by \nwork on high power auditory systems for the Gov- \nernment. Since the end of the war, he has been \nengaged in the development of hearing aids, \naudiometers, and artificial larynges.\n\nFig. 7\u2014Gain-frequency characteristics of various ar- \nrangements of the Models 65 and 66 hearing aids\n\nsmaller and therefore more convenient to \nwear. The reduced size causes no loss in \ndependability. Its acoustic characteristics \nare of good quality and are entirely ade- \nquate for the majority of hard-of-hearing \npeople. By offering the choice of either of \nthese two sets, Western Electric is carry- \ning out its long-established policy of ren- \ndering the greatest possible assistance to \nany person who is hard of hearing.\n\n\u201cCongresso Internazionale per il Cinquan- \ntenario della Scoperta Marconiana della Radic,\u201d \nwas the announcement spelled out by ban- \nners stretched across the streets of Rome as \nengineers and physicists from many countries\n\nname of Guglielmo Marconi. The tenth anni- \nversary of Marconi\u2019s death was the fiftieth an- \nniversary of the pioneering experiments of this \nbrilliant and energetic Italian in wireless \ncommunication.\n\nSponsoring the Congress was the National \nCouncil of Research, whose fine headquarters \nbuilding near the University of Rome was \nthe scene of the technical sessions. My assign- \nment was to represent the Laboratories and \nto give a paper dealing with our contributions \nto the design of modern broadcast transmitters.\n\nMouromtseft of Westinghouse, Zworykin of \nRCA, and Bolt of MIT were other Americans \nparticipating as the Congress commenced on \na Sunday afternoon with ceremonies in the \nCampidoglio, or state capitol. Among the many \nforeign scientists I met for the first time were \nSmith-Rose from England and Strutt from \nHolland, whose names are familiar to all ra- \ndio people through their many publications.\n\nThe papers presented during four days of \ntechnical sessions covered a wide field, includ- \ning acoustics, wave propagation, circuit theory, \nvacuum tubes, communication systems, aural \nbroadcasting, and television. Numerous Italian \npapers testified to a lively scientific endeavor \nin that country despite the present difficult \neconomic conditions; while the importance as- \nsumed by microwave papers was in line with \nthe technical trends of the times. Indeed, \nMarconi himself had been an enthusiast over \ncentimeter waves, and no one a generation \nago envisioned more clearly than he the vast \npossibilities for their practical use.\n\nments Rome has to offer, the delegates wel- \ncomed a program arrangement that left time \nfor sightseeing, though in a visit of ten days \none could only glimpse briefly the \u201chigh spots\u201d \nof the Eternal City. The management of the \nCongress arranged for splendid side trips to \nVilla d\u2019Este at Tivoli, with its beautiful foun- \ntains and gardens; to the picturesque ruins \nof Ostia, ancient Roman port at the mouth of \nthe Tiber; and to Castelgandolfo, papal sum- \nmer residence, for an address by Pius XII \nand a personal introduction to the pontiff for \neach member. An evening concert provided \nby the Italian broadcasting organization and \nan inspection tour of the international radio \ntransmitting station of Italeable at Torrenova \nrounded out a busy week. The Torrenova \nplant, a great communications center almost \ncompletely destroyed by the Germans, is be- \ning rebuilt and equipped with the most mod- \nern long-wave and short-wave radio telephone \nand telegraph equipment for European and \noverseas transmission.\n\nAfter the close of the Congress on October \n5, I spent a month visiting industrial and \nuniversity laboratories and radio stations in \nSwitzerland, France, Holland, Belgium, Lux- \nembourg, England, and Ireland. An outstand- \ning affair I attended in London, attracting the \ngeneral public from all over England, is Ra- \ndiolympia. Resumed this year, after a lapse \nof eight years, this exhibition of radio products \nof all kinds drew half a million spectators in \nten days, a remarkable show of interest for \nthese hard times. Improvements in radio tech- \nnique, stemming from the intensive war ef- \nfort in England and America in electronics, \nare plainly seen in these products; and _ all\n\nThe photograph at the head of the page shows \nthe author (right) and V. I. Enders of Alpine \nWestern Electric, Switzerland, returning from a \nsail on the Lake of Zurich.\n\nTo review the progress of development and \nmanufacturing of wire, cables and cords, a \nconference of engineers and executives of the \nLaboratories and the Western Electric Com- \npany was held at the Point Breeze Works in \nOctober. This was the first in a series of similar \nconferences to discuss joint development and \nmanufacturing problems. The second was held \nat the Kearny Works on December 10 and 11 \nand dealt with the development and manu- \nfacturing of transmission apparatus.\n\nUnder the general direction of M. J. Kelly, \nthe Kearny conference was planned by R. G. \nMcCurdy, H. H. Lowry and H. Rossbacher, \nWorks Engineer of Manufacture at Kearny. \nAfter an introduction by R. F. Clifford, Works \nManager, the program was outlined by Dr. \nKelly and H. C. Beal, Vice-President of the \nWestern Electric Company.\n\nPresentations were made by engineers of \nthe Laboratories and of Western Electric cov-\n\nering a number of areas of development and \nbringing out the relationships between the \nmanufacturing and development problems in \nthese fields. The subjects covered included \ncoaxial line and terminal equipment, televi- \nsion terminals and video amplifiers, networks, \nequalizers and filters, measurement problems, \ncrystals, component apparatus, the current \nsize-reduction activities, and transmission \neconomics.\n\nIt is the purpose of these conferences to \nbring up for discussion at an early date prob- \nlems of engineering design and problems of \nmanufacture relating to the developments be- \ning made by the Laboratories. As a result of \nthese discussions, the manufacturing and de- \nvelopment programs are geared together to \nachieve maximum effectiveness in translating \nnew ideas from the Laboratories into equip- \nment for the operating telephone companies \nof the Bell System.\n\nover Europe, despite severe shortages in \nmaterials and components, one finds great \nactivity in the development of new com- \nmunication equipment operating in the ultra- \nhigh-frequency and microwave bands and \nemploying FM and pulse modulation multi- \nplex methods.\n\nTelevision is popular in Britain, and video \ntransmitting stations are planned for Birming- \nham and other important centers. For linkage \nwith the present London installation, trials of \nboth radio relay and broad-band coaxial cable \nare to be made, somewhat paralleling our own \ntelevision network plans. British receivers, in- \nterestingly enough, differ from ours in being \nbuilt for one channel only and for carrier- \nand-double-sideband reception.\n\nThe diary of the American traveler in Eu- \nrope is sure to contain references to the droves \nof bicycles on the streets of Dutch and Swiss \ncities; to the immense variety of noisy motor- \nbikes and midget cars that dart about dan- \ngerously in Rome and Paris; to food ration- \ning and London fogs. My notes mention these \nand other experiences, but the really deep \nimpressions were of warm hospitality extended \nby Europeans everywhere, and of diligent and \ncourageous efforts toward recovery in coun- \ntries hard hit by the war.\n\nPolitical tension was evident in Italy and \nFrance, with daily threats of a general strike, \nbut except for a short walkout on the metro \nin Paris, there was no outward disturbance.\n\nThe subsequent outbreaks of violence make \none glad to be safely back, but concerned for \nthe new friends left on the Continent to \nweather the storm.\n\nIn a move to further accelerate the Bell \nSystem\u2019s nation-wide construction program, the \nLaboratories has developed a new type of \ntelephone cable sheath using a thin sheet of \naluminum covered with a polyethylene com- \npound, a tough, flexible plastic that looks like \nblack rubber. The new cable will supplement \nthe familiar lead-covered cable which is now\n\nat peak production. Following extensive tests \nof its suitability, the Western Electric Com- \npany has begun quantity production of the \nnew type of cable at Kearny and Hawthorne. \nThe cable, called \u201cAlpeth,\u201d is to be used \nwithin local exchange areas on pole lines and \nin underground conduit. It will be made in a\n\nvariety of sizes, ranging from the smaller ca- \nbles to those containing hundreds of pairs \nof wires.\n\nThe immediate purpose to be served by \nthe new cable will be to step up the deliv- \nery of cable to the telephone companies and \nhelp meet the continuing heavy demand for \ntelephone service.\n\nThe name \u201cAlpeth\u201d is derived from the \nsheet of aluminum which goes next to the \npaper-insulated wires, and from the outer \ncovering of polyethylene. Before carbon black \nis added, the polyethylene has a milky color.\n\nFor many years the Laboratories has car- \nried on continuous research on cable improve- \nment. Insofar as can be predicted from lab- \noratory and field tests, polyethylene promises \nto give satisfactory service as a cable sheath \nmaterial. No change has been made in the \ncable core itself.\n\nMeanwhile, with the production of lead- \ncovered cable continuing at a record pace, the \nnew composite sheathing is helping to increase \nstill further the total output of much-needed \nexchange cable.\n\nEprror\u2019s Nore: A more extensive account \nof this development will appear in an early \nissue of the REcorp.\n\nPhilatelists exhibited their prize specimens \nat an exhibition in the Lounge at Murray \nHill on November 20 and 21, signalizing the \nbeginning of the newly organized Stamp Club \nat that location under the direction of the \nBell Laboratories Club. S. C. Tallman was \nawarded first prizes for his displays in both \nthe United States and foreign classes. C. W. \nFerguson and K. H. Schunke took second \nprizes in those classes, while H. T. Webber \ntook both prizes in the novelty class and also \nwon the award \u201cbest in show.\u201d\n\nW. A. Shewhart sailed for Southampton on \nNovember 22 on the first leg of a trip by boat \nand plane to India for a four months\u2019 visit \nand lecture tour. Dr. Shewhart\u2019s invitation \nwas sponsored jointly by the Indian Standards \nInstitution; the Indian Science Congress As- \nsociation, the Indian counterpart of the Ameri- \ncan Association for the Advancement of \nScience; and the Indian Statistical Institute, \nthe largest research organization of its kind \nin the world. As Visiting Professor to the Sta- \ntistical Institute, he plans to give a series of \nlectures at Calcutta University and one or \nmore lectures elsewhere in India. He will also \nparticipate in round-table quality control\n\nconferences in Bombay and Calcutta with stat- \nisticians, industrial leaders and technical rep- \nresentatives from all over India.\n\nIn his lecture equipment, Dr. Shewhart in- \ncluded pamphlets describing the over-all or- \nganization and workings of the Bell System; \ndescriptive material and lantern slides showing \nviews of the newer Laboratories\u2019 buildings - \nand equipments; a Fastax camera; and a \ncolored sound film on quality control.\n\nThe Murray Hill Chorus presented a varied \nprogram of Christmas music in the Arnold \nAuditorium on Tuesday, December 23, with \nthe audience participating in several of the \nfamiliar carols. Daniel F. Kautzman directed \nthe Chorus and Capitola Dickerson was the\n\nM. J. Kelly receives the Presidential Certificate of \nMerit from Secretary of the Navy John L. Sullivan \n\u201cin recognition of his efforts in the field of elec- \ntronics which proved to be an invaluable contribu- \ntion to the war effort of the United States.\u201d The \npresentation ceremonies were held in Washington \non November 19\n\naccompanist. W. H. Martin, chairman of the \nMurray Hill Committee, gave the annual \nChristmas greeting.\n\nCarols were sung outdoors at Whippany, \nwhere members of the Laboratories gathered \non the steps of the Administration building at \nnoon on Christmas Eve. At about the same \ntime, the System\u2019s Christmas Chorus of sixty- \nseven voices rendered a selection of carols in \nthe drafting room of the Equipment Develop- \nment Department on the ninth floor at West \nStreet. The carolling was picked up by a \nmicrophone, transmitted into the courtyard, \nand amplified, so that many more members\n\nPresent at a conference to \nappraise technical and re- \nlated subjects submitted to \nthe Patent Department for \nconsideration are, standing \n(left to right), Loretta \nSpacek, Catherine Mc- \nQueeny, H. A. Burgess, G. \nH. Heydt, R. E. Poole, N. \nS. Ewing and E. L. Nelson. \nSeated around the confer- \nence table are (left to \nright) E. V. Griggs, J. W. \nSchmied, H. A. Blake, M. \nR. McKenney, D. A. \nQuarles, W. C. Kiesel and \nJ. R. Wilson\n\nof the Laboratories than could be accommo- \ndated in the Auditorium had the pleasure of \nhearing the program. Under the direction of \nR. P. Yeaton, the chorus sings together only \nat Christmas and has been in existence five \nyears. Originally comprised of draftsmen, it \nhas grown each year, adding, as it did in 1947, \na number of trained new voices from other \ndepartments.\n\nMany inquiries have been received regard- \ning the New York State veterans\u2019 bonus in con- \nnection with transfers to New Jersey or other \nout-of-state locations of the Laboratories and \nsubsequent changes in residence. To be eli- \ngible, the bonus law states \u201cresidence in New \nYork State at the time of entry into service \nand at least six months immediately prior, and \nresidence in the state at the time application \nfor bonus is made out\u201d is necessary. Thus, a \nmember of the Laboratories who has been a \nresident of New York State is not eligible for\n\nthe bonus if he or she changes his or her \nresidence to another state before application \nfor the bonus is made. Some veterans may be \nrequired to file special residence question- \nnaire forms after submitting their applications \nin order to clarify their domicile status.\n\nNew York State veterans who are on leaves \nof absence from the Laboratories to attend \nschool are eligible for the bonus unless there \nare indications that he or she has become a \npermanent resident of some other state.\n\nH. A. Blake, Assistant Commercial Relations \nManager, was transferred to the Patent De- \npartment on October 22 as Executive Assistant \nto M. R. McKenney, General Patent Attorney.\n\nCoincident with this appointment, C. R. \nMcConnell and H. Schmitt, Commercial Proj- \nects Supervisors for New York and Murray \nHill, respectively, now report to B. B. Webb. \nThe title, Assistant Commercial Relations \nManager, has been discontinued.\n\nDuring the Months of September, October and November the United States Patent Office \nIssued Patents on Application Filed by the Following Members of the Laboratories\n\nT. Aamodt R. C. Davis R. H. Gumley F. B. Llewellyn R. K. Potter \nW. P. Albert (2) T. L. Dimond H. A. Hilsinger, Jr. G. A. Locke R. M. Ryder \nM. C. Biskeborn R. A. Ehrhardt W.H.T. Holden (2) G.R. Lum J. H. Scaff\n\nF. B. Blake C. H. Elmendorf H. Hovland W. P. Mason (2) \u00a9. A. Shann \nE. M. Boardman E. P. Felch, Jr. R. G. \u2014\u2014* A. L. Matte F. J. Singer\n\nJ. H. Bollman (2) ~~ A. G. Fox E. M. Julic W. H. Matthies T. Slonczewski (2) \nW. L. Bond A. L. Fox A. P. King H.J.McSkimin J. Tarr\n\nA. E. Bowen (2) C. J. Frosch W. Koenig, Jr W. J. Means H S. Wertz \nD. E. Branson G. W. Gilman J. A. Krecek W. A. Mehmel H Whi ' \nH. W. Bryant H. W. Goff J. G. Kreer, Jr. E. A. Nesbitt A. H. White \nA. J. Busch W. M. Goodall A. G. Laird C. W. Norwood H. T. Wilhelm \nJ. A. Carr H. L. B. Gould A. G. Lang W. G. Pfann R. O. Wise\n\nB\u2014Graybar\u2019s committee, Rose Rovegno, Molly \nRadtke, chairman; Eve Coll, Mary Burke, \nShirley Zitzmann, and Marion Greenberger\n\nC\u2014Whippany\u2019s committee. In the rear, Dor- \nothy Clothier, Harriet Filmer, chairman; \nJoan Helm, Evelyn Selzer, and \u201cPat\u201d \nRooney; center, Bette Mocksfield, left, and \nMarion Merck, right; front row, Marilyn \nMiller, Betty Minerowitz, Elizabeth Myers, \nand Jean Force\n\nD\u2014F. D. Leamer, president, and Mrs. Anne \nBrokaw, executive secretary, Family Serv- \nice Association of Summit, accept toys from \nDorothy Thom\n\nE\u2014Peggy Grillo and three children of the \nNursery School at the Greenwich House\n\nG\u2014Dennis Cronin admiring the little dogs he \nmade from wash cloths for the display\n\nH\u2014Madeline Kiselica exhibiting story book \nand opera character dolls she dressed\n\nJ\u2014At Whippany, Mrs. Filmer (left) with \nguardians of children in boarding homes\n\nK\u2014Mary Mahairas(left)and Florence McGuire \npresent gifts to servicemen\u2019s children at a \nparty given by the Red Cross in Flushing\n\nTRANSMISSION TELEVISION \nDEVELOPMENT RESEARCH \n| | \nbe \nFIFTH FLOOR \nATTIC \nT \nEQUIPMENT \nL 3 DEVELOPMENT\n\nSTATION APPARATUS PATENT \nTRANSMISSION SPECIAL \nDEVELOPMENT RESEARCH \nr ----45 \nATTIC '\n\nPHYSICAL TRANSMISSION] , \nKPPARATUS [LIBRARY] ROOF | STATION APPARATUS OUTSIDE PLANT \n| | <= PATENT \n| \nDRAFTING \nOOR\n\nENCLOSED \nELECTRONIC TRANSMISSION v TRANSMISSION \nRESEARCH APPARATUS STATION APPARATUS cant L, RESEARCH \n= MAIN LI LOUNGE \nENTRANCE \nROOMS \n~< \nTERRACE \nELECTRONIC RESEARCH CHEMICAL \nDEVELOPMENT SECOND FLOOR DRAFTING ~\u2122\u201d LABORATORIES \n= PERSONNEL ADMINISTRATION \non \n| j MEDICAL rq \nELECTRONIC TRANSMISSION ' \n| RESEARCH APPARATUS BASEMENT \nTERMINAL \nLOUNGE : \nCAFETERIA rj CAFETERIA\n\nProgress on Project No. 2 at Murray Hill indicates that the first of its new occupants can begin to \nmove about August 1 of this year. The chart on the opposite page shows where you will find the \nvarious departments when the moves are completed some time in 1949. The shaded arrow (circle, \nupper right) indicates the direction usually referred to as \u201cNorth\u201d\n\nM. J. Ketiy addressed the engineering and \nsupervisory force of the Southwestern Bell \nTelephone Company at St. Louis. He also at- \ntended the biannual meeting at Rolla, Mis- \nsouri, of the Board of Directors of the Missouri \nSchool of Mines Alumni Association on No- \nvember 7. On the following day, Dr. Kelly de- \nlivered the main address at the Homecoming \nConvocation at the School, his alma mater \n(B.S., 1914, and Honorary Doctorate in Engi- \nneering, 1936), selecting as his subject Science, \nTechnology and the \u201cGood Society.\u201d\n\nDurinc a two-day Research Conference in \nWashington of the Navy Industrial Association, \nDr. Kelly, who is Chairman of the Research \nand Development Committee, addressed the \nConference on its opening day, November 18, \nat a dinner session. His subject was Our Coun- \ntry\u2019s Preparedness Research and Development \nProgram\u2014A Cod\u00e9perative Undertaking of Our \nMilitary, University and Industrial Laboratories.\n\nD. A. QuaRLEs went to Chicago for the Board \nof Directors meeting of the American Institute \nof Electrical Engineers. He was also in Wash- \nington on November 13 attending the meeting \nof the Committee on Electronics of the Joint \nResearch Board and on November 19 attend- \ning the Navy Research Conference of the Navy \nIndustrial Association.\n\nA. B. CLARK was a guest at the Navy Confer- \nence in Washington on November 18 and 19.\n\nOn November 25, Mr. Clark, H. H. Lowry, \nF, J. Scupper, T. C. Fry, J. G. Fercuson and \nF. A. Korn conferred with officials of The \nBell Telephone Company of Pennsylvania in \nPhiladelphia, and visited the No. 5 crossbar of- \nfice in Media.\n\nHARVEY FLETCHER has written on The Science \nof Hearing in the book The Scientists Speak. \nTwo retired members of the Laboratories have \nalso contributed articles to the book, H. E. \nIves, Physics and Art, and R. R. WiLiiaMs \nwith R. J. Williams, The Golden Age of Bio- \nchemistry. Dr. Fletcher attended meetings of \nthe National Academy of Science, November \n17 to 19, in Washington. He also visited Rut- \ngers University.\n\nA. R. Kemp, rubber technologist and insulation \nengineer of the Chemical Laboratories, has \nbeen designated one of the country\u2019s ten ablest \nchemists in the field of rubber chemistry by \nhis fellow-members in the American Chemical \nSociety.\n\nLioyp EspeNnscHiep spoke on The Dawn of \nElectrical Experimentation on November 12 \nat a supper meeting of the Institute of Radio \nEngineers, New York Section, preceding the \ntechnical meeting.\n\nC. H. Townes spoke on Microwaves at a joint \nHarvard-M.1.T. seminar at Cambridge.\n\nC. Kirre. selected Paramagnetic Resonances \nfor his topic when he spoke to the Physics \nColloquium at Rutgers University.\n\nIn 1942, Nelson E. Sowers\u2019 health would \nnot permit him to remain longer in the New \nYork area, so he was granted a Disability \nPension which allowed him to withdraw from \nRadio Research and live in the southwest. \nThree years later he went to work for the \nArmy Ground Forces Board at Fort Bliss, \nTexas, as a Mathematical Data Analyst. Re- \ncently he has received from the War Depart- \nment a Decoration of Exceptional Civilian \nService. The citation, signed by the Secretary \nof War, reads: \u201cHis outstanding service in \nthe development of instrumentation and com- \nputing equipment and procedures for antiair- \ncraft artillery service testing constituted a \ncontribution of inestimable value to the mis- \nsion of his command.\u201d\n\nMr. Sowers\u2019 friends will be gratified to learn \nthat his health has been restored. However, \nfamily reasons require his continued residence \nin the southwest, so he has now resigned from \nthe Laboratories.\n\nGeorcE B. Tuomas, until last summer Per- \nsonnel Director, is now chief of the Industrial \nRelations Department of the Los Alamos lab- \noratory of the Atomic Energy Commission.\n\nJosepH P. MAxFiELD, who retired in Sep- \ntember, is now a consulting engineer on the \nstaff of the Altec-Lansing Corporation, acous- \ntic engineers. His headquarters will be on the \nPacific Coast.\n\nCATHARINE C. MAULL, who retired in Oc- \ntober, 1945, from Personnel, is curator of the \nmuseum in the Zwaanendael House in Lewes, \nDelaware. This building is a replica of the \ntown hall in Hoorn, Holland.\n\nNews Notes \nJ. R. Haynes presented an invited paper be- \nfore the Houston meeting of the American \nPhysical Society entitled Behavior of Photo- \nelectrons in Silver Chloride Crystals Exhibited \nby Colloidal Silver Produced at Electron Traps.\n\nG. T. Kouman, A. N. W. L. Bonp \nand \u00a7. O. Morcan discussed crystal growing \nproblems at the Squire Laboratory.\n\nJ. R. Fiecat, G. T. Kouman, N. Y. \nMAN and G. K. TEAL visited the Norton Com- \npany in Worcester, where they discussed sili- \ncon carbide varistors. Mr. Kohman has been \nappointed to represent the Laboratories on the \nN.R.C. Committee on Design, Construction \nand Equipment of Laboratories.\n\nL. H. Germer and F. E. Haworru partici- \npated in a colloquium on Electrical Contacts \nat Massachusetts Institute of Technology.\n\nA. R. Kemp discussed plastic sheathing with \ncable engineers at Hawthorne during his \nweek\u2019s visit to that plant.\n\nJ. R. TownseEnp visited The Mountain States \nTelephone and Telegraph Company, Denver; \nthe Northwestern Bell Telephone Company, \nOmaha and Minneapolis; and the Southwest- \nern Bell Telephone Company, St. Louis, where \nhe spoke to telephone personnel on Materials \nDevelopments. During his trip, Mr. Townsend \nvisited the University of Colorado at Boulder, \nColorado. At Washington University, St. Louis, \nhe addressed a joint session of the American \nInstitute of Chemical Engineers and American \nSociety of Mechanical Engineers. In Los Ala- \nmos, New Mexico, he addressed the local chap- \nter of the American Society for Metals on \nMetals and Alloys for Magnetic \nand Electrical Applications.\n\nV. T. CALLAHAN and R. R. \nGay visited repeater stations \nassociated with the New \nYork-Boston radio relay proj- \nect regarding power items. \nMr. Callahan also visited the \nGeneral Motors Corporation \nin Detroit and the Duplex \nTruck Company in Lansing \nin connection with Diesel and \ngasoline engine sets.\n\nW. BaBINGTON witnessed a series of die cast- \ning experiments at the Precision Casting Com- \npany in Syracuse. He also went to Hawthorne \nfor die casting discussions.\n\nE. E. SCHUMACHER and I. V. WiLLiAMs made \nan inspection trip to the School of Mineral \nIndustries at Pennsylvania State College.\n\nH. A. Brrpsauu visited the International Busi- \nness Machines Company at Endicott, N. Y. He \nalso attended the meeting of the A.S.T.M. \nCommittee D-6 on paper and paper products \nat Albany. With D. A. McLEan he attended \nthe open house meeting of the Textile Re- \nsearch Institute at Princeton; and with K. G. \nCouTLEE, J. J. MARTIN, G. DEEc and R. Burns, \nthe meeting at Atlantic City of Committee D-9 \non electrical insulating materials.\n\nF. G. Foster received Honorable Mention in \nthe metallographic exhibit at the recent Na- \ntional Metal Congress and Exposition at Chi- \ncago for his entry of a chart with eighteen \nphotomicrographs.\n\nR. Burns\u2019 trip to Massachusetts Institute of \nTechnology at Cambridge concerned new de- \nvelopments in mechanical testing of plastics.\n\nW. O. Baker delivered an invited lecture at \nNorthwestern University, Chicago, on The Or- \nganic Structure and Physical Nature of High \nPolymers. The lecture was one of a series spon- \nsored by the university. Mr. Baker visited \nHawthorne en route for casting resin develop-\n\nment conferences. He also was in Washington \nfor an advisory panel meeting called by the \nOffice of Naval Research and the Bureau of \nOrdnance concerning polymer research.\n\nC. C. Hipxins and J. C. Osten attended the \nconvention of the Federation of Paint and Var- \nnish Production Clubs in Atlantic City. Mr. \nOsten\u2019s paper, A Comparison of Infra-red and \nConvection Oven Baking, which was presented \nat the convention, received the first award.\n\nG. N. THayer addressed the New York Sec- \ntions of the A.I.E.E. and of the Institute of \nRadio Engineers on the New York-Boston Ra- \ndio Relay System.\n\nTHE FOLLOWING MEMBERS Of the Institute of \nRadio Engineers have been elected to the \ngrade of Fellow by the Board of Directors of \nthe Institute: M. W. Batpwin, Jr., H. S. \nBiack, L. A. MEACHAM and J. R. Pierce.\n\nA. B. Haines, in Rochester on November 19, \nattended a meeting of R.M.A. 10C Committee \non power transformers (receiver section). Mr. \nHaines and L. W. Kirkwoop were at Wright \nField in connection with the development of \nhigh operating temperature transformers.\n\nJ. R. Weexs and R. K. Evenson attended \nconferences at Hawthorne on the aluminum \ncan condenser problem.\n\nW. R. NEeEIssEr observed the assembly at \nArcher Avenue of a trial lot of networks for \nthe new combined set.\n\nDuring the next few months, some members of the Laboratories may be spending their \nvacations in Florida. A number of retired members live in Florida and would be glad \nof a visit, particularly from those who have known them in their more active days.\n\nEau Callie \nBruce Freile Box 126 \nEustis \nStanley F. Nelson \nFort Pierce \nJames L. Crouch P. O. Box 405 \nJanuary 1948\n\nSt. Petersburg \n1014 Bay St., NE \nHollander Hotel \n4200 4th Ave., North \n1151-34 Ave., North \n201 6th St., South\n\nOscar E. Benson \nGeorge W. Folkner \nJoseph F. Johlfs \nCharles W. Keckler \nWm. Scharringhausen\n\nMembers of the Laboratories who retired \nrecently include E. W. Hancock on Decem- \nber 28 and S. W. SuHiteEy on December 31, \nboth with 43 years of service; ANTON LODER \non December 31, 32 years; and R. J. HEFFNER \non December 22, 29 years.\n\nRoy J. Heffner of the Personnel Department \nretired on December 22 following twenty-nine \nyears of service. After receiving the B.S. in \nE.E. degree from the University of California \nin 1916, Mr. Heffner immediately joined the \nWestern Electric Company in Chicago but was \nsoon transferred to the Engineering Department \nin New York. He served with the Air Service \nof the U. S. Army from 1917 to 1919 and \nwhen discharged had attained the rank of \nMajor. For the next two years he was first \nChairman of Engineering Extension at his \nalma mater and then Educational Director of \nthe Hawaiian Department, U. S. Army. In \nJuly, 1921, he joined The Pacific Telephone \nand Telegraph Company, where he became \nsupervisor of employment and training for \nmen and women in 1924.\n\nMr. Heffner transferred to the Laboratories \nin 1929 and after a year as Assistant Educa- \ntional Director, was made Educational Direc- \ntor in charge of the employment and training \nof members of technical staff, student assist- \nants, drafting assistants and shop apprentices.\n\nFrom 1936 to 1939, as Assistant Personnel Di- \nrector, he was in charge of employment and \ntraining. Since 1939, he has directed the per- \nsonnel analysis and planning group.\n\nSam W. Shiley of Transmission Develop- \nment retired on December 31 following forty- \nthree years of service. Mr. Shiley joined the \nClinton Street Works of the Western Electric \nCompany in 1904 where he held several super- \nvisory positions in the Drafting Department \nand was in charge of telephone circuit draft- \ning when it was moved to Hawthorne in 1907. \nThe following year he transferred to the Engi- \nneering Department to participate in the de- \nvelopment and standardization of telephone \nswitchboards, distributing frames, and racks. \nHe was Standardization Engineer when he \ncame to New York in 1915. Here he took part \nin the development of many manual and toll \nsystems. Mr. Shiley became Laboratory Engi- \nneer in the carrier telephone and telegraph \nlaboratories at the Graybar-Varick building in \n1930. Previous to his retirement, he was in \ncharge of the Transmission Development Lab- \noratory and was also responsible for handling \ndrawings and specifications of his department.\n\nEpmMunp W. Hancock \nEdmund W. Hancock of the Switching De- \nvelopment Department retired on December \n28 following forty-three years of service. His\n\nfirst work with the Bell System was in con- \nnection with installation of manual central of- \nfices in the New York area. After several years \nwith the Installation Department he _trans- \nferred to West Street, and took part in work \non the first machine switching system which \nwas tried out as a PBX in the West Street \nbuilding, and which preceded the panel sys- \ntem. He was next engaged in design and \ndevelopment of the panel semi-mechanical sys- \ntem and supervised part of the testing of this \nsystem at the Newark and Wilmington instal- \nlations. Later he was in responsible charge \nof a group designing selector circuits for the \npanel dial system. Beginning in 1926, he con- \nducted field development work in connection \nwith new features of various telephone switch- \ning systems and investigations of irregular \nconditions. In more recent years he has been \nassociated with the development of the No. 1 \ncrossbar system.\n\nAnton Loder of the Transmission Develop- \nment Department retired on December 31 fol- \nlowing thirty-two years of service. Mr. Loder \njoined the Development Department in 1914 \nand worked as an instrument maker on print- \ning telegraph equipment and then on radio \napparatus for the Signal Corps. Later, in 1917, \nhe left the Company and went with the M. I. \nInstrument Company, where he worked on ap- \nparatus they were making for the Western \nElectric Company. He returned to the Devel- \nopment Department in 1919 and for the next \nfourteen years was engaged as an instrument \nmaker with a variety of telephone apparatus, \nparticularly for transatlantic and ship-to-shore \nservice. From 1933 to 1938 he was with the \ndesign and wiring group of the Research De- \npartment on special wiring and mechanical \nwork of an experimental nature. He then was\n\ntransferred to do the same general type of \nwork for the broad-band cable group of Trans- \nmission Development. Since April, 1946, Mr. \nLoder has been with the coaxial systems group \nof the same department.\n\nC. C. Hourz inspected manufacturing facilities \nand initial production of ceramic condensers \nfor radar applications at Winston-Salem.\n\nL. W. STAMMERJOHN\u2019S visit to the New York \nTransformer Company at Alpha, N. J., had to \ndo with manufacturing problems on power \ntransformers for amplifier use.\n\nC. N. Hickman spoke on Rockets at the Deal- \nHolmdel Colloquium on December 3 at Deal. \nHe also lectured to the students of the Air War \nCollege, Maxwell Field, Ala., on December 8.\n\nR. E. PooLte, W. C. Tinus and M. H. Coox \nattended the meeting in Washington of the \nNavy Industrial Association.\n\nF. L. LANGHAMMER participated in discussions \non components of radar equipment at the \nN.R.K. Manufacturing and Engineering Com- \npany in Chicago.\n\nA. K. Bowren and B. O. BRowNE witnessed \ntests on an FM antenna tower at the Blaw- \nKnox Company, Pittsburgh.\n\nW. W. Brown visited the Western Electric \nRepair Shop at Philadelphia to discuss opera- \ntors\u2019 chair problems.\n\nR. G. Koontz and W. J. ApAms visited Win- \nston-Salem and Burlington to discuss engineer- \ning and drafting practices for Bell System \nRadio Communication equipment.\n\nM. D. Brixt and I. E. Fair attended the Na- \ntional Electronics Conference in Chicago, No- \nvember 3 to 5.\n\nR. A. Sykes attended the Piezoelectric Com- \nmittee meeting of the Institute of Radio Engi- \nneers in New York on November 3.\n\nE. B. Woop and N. INsLey visited Hawthorne \nto consult with Western Electric Company en- \ngineers on switchboard lamp problems.\n\nR. H. Ross studied aircraft radio motors at \nA. W. Haydon Company in Waterbury.\n\nF. S. Mat, H. Peters and V. T. WALLDER \nwitnessed the field splicing of \u201cAlpeth\u201d cable \nat Belleville, N. J. Mr. Wallder and J. B. \nHowarp conferred at Point Breeze on plastic \ncompounding methods.\n\nJ. P. Guerarp went to Pittsburgh for meetings \nof the A.S.T.M. Committee B5 on copper and \ncopper alloys.\n\nW. P. Smirtu has been re\u00e9lected Mayor of At- \nlantic Highlands, N. J., for another two-year \nterm. In his twelve years of civic life, Mr. \nSmith has been councilman for ten years, a \nmember of the school board for six, and mayor \nfor two.\n\nM. S. Mason, in his capacity as a member of \nthe New York Engineers Committee on Stu- \ndent Guidance, met students of the Bronx High \nSchool of Science on December 3 to discuss \npreparation for and opportunities in the engi- \nneering profession.\n\nD. E. Trucxsess and M. A. Froserc investi- \ngated noise conditions in connection with the \nuse of the new 100-ampere selenium rectifier \nat Buffalo and Erie.\n\nE. M. ToiMaAn recently received a Ph.D. de- \ngree in Chemistry from Columbia University. \nDr. Tolman presented his thesis as a paper \nbefore the September meeting of the Ameri- \ncan Chemical Society in New York City.\n\nW. C. KirkMan is treasurer of Troop Commit- \ntee No. 114 of the Boy Scouts of America \nin Rochelle Park, N. J. In the campaign just \nconcluded for funds for the Council, Mr. Kirk- \nman\u2019s committee doubled the quota set for \ntheir area.\n\nH. O. StecmMunp spoke on the subject of \nRockets at a meeting of the American Society \nof Mechanical Engineers on November 20, \nand on December 3 before the New York \nChapter of Alpha Chi Sigma.\n\nA. C. PETERSON discussed mobile radio equip- \nment problems with Motorola engineers at \nChicago.\n\nW. Strack visited Chicago in connection with \nthe multi-channel mobile telephone system.\n\nA. E. Rupprex and F. J. Skinner visited Al- \nbany to discuss problems concerning the New \nYork State Police mobile telephone system.\n\nO. J. MorzENTI went to Media and Duluth in \nconnection with equipment development prob- \nlems concerning No. 5 crossbar system.\n\nR. L. Lunsrorp conferred at the Northern \nElectric Company, Montreal, on studies of \nsmall community dial offices.\n\nJ. R. Srone participated in discussions on \nringing equipment for No. 5 crossbar system \nat the Holtzer-Cabot Electric Company in \nBoston.\n\nA. A. BurcEss and R. P. Jurson, at Philadel- \nphia, conferred with local telephone engineers \non special equipment for the No. 5 crossbar.\n\nH. H. Spencer tested electronic speed regu- \nlators for motor-generator sets used to supply \npower for the L carrier system at Dallas, Min- \neral Wells, and Sweetwater, Texas.\n\nFollowing an illustrated lec- \nture in the West Street Au- \nditorium on Night Photog- \nraphy by Stanley Rayfield \nof Life Magazine, forty- \neight members of the Pho- \ntography Club took a field \ntrip to Central Park. Putting \ninto practice the techniques \noutlined by Mr. Rayfield \nwere this group of photog- \nraphers taken by Chairman \nW. S. Suydam. Left to right, \nK. L. Warthman, A. L. Johns- \nrud, R. Olsen, R. J. Nielson, \ntwo unidentified guests, W. \nJ. Rutter and H. J. Braun\n\nP. T. Sprout discussed television switching \nproblems with engineers of the Illinois Tele- \nphone Company in Chicago.\n\nJ. L. Merrity, Jr., and J. A. WELLER con- \nducted a trial installation of the E-1 exchange \narea repeater in Chicago. G. C. ReErer ob- \nserved the installation and performance of \nthese repeaters.\n\nR. E. Crane attended the Midwest General \nMeeting of the A.I.E.E. and the National Elec- \ntronics Conference in Chicago during the pe- \nriod from November 5 to 7. He presented be- \nfore the A.I.E.E. a paper, Frequency Division \nTechniques for a Coaxial Cable Network, of \nwhich he was co-author with J. T. Dr1rxon \nand G. H. Huser.\n\nMr. Stoller, who was a member of the Tech- \nnical Staff in the Switching Apparatus Develop- \nment Department, was stricken on his way to \nwork and died shortly afterwards at a nearby \nhospital. A native of Schenectady, he had re- \nceived his E.E. degree from Union College in \n1913 before joining the Bell System. After a year \nin relay development work, he returned to his \nalma mater and received his M.S. degree in \nelectrical engineering in 1915. Returning to the \nPhysical Laboratory, he was engaged in mis- \ncellaneous research problems until World War \nI, when he designed and developed power \nequipment for radio apparatus used on air- \nplanes and submarine chasers.\n\nMr. Stoller then transferred to what is now \nthe Commercial Products Development De- \npartment, where he was responsible for the \ndevelopment of synchronizing and speed con- \ntrol devices for sound recording and reproduc- \ning systems for television and for picture trans- \nmission. Following that, he engaged in the de- \nvelopment of electromagnetic apparatus for \nswitching systems.\n\nDuring World War II he contributed to the \nresearch and development of the electrical gun \ndirector, magnetic mines and sonar. The more \nthan seventy patents that he held on voltage \nand speed regulators and on sound picture \nequipment attested to the valuable contribu- \ntions he had made in his various fields of in- \nterest. Mr. Stoller was author of numerous arti- \ncles in technical publications.\n\nMr. Kopetz, who was born in 1872, was \ntransferred from the Manufacturing Depart- \nment of the Western Electric Company to the \nModel Shop of the Engineering Department in \n1908. Prior to World War I he worked on ma-\n\nchine switching equipment and was one of a \ngroup of instrument makers assigned to install \nthe semi-mechanical switching system in the \nMarket and Waverly exchanges in Newark. \nDuring World War I he became a supervisor \nof a group of instrument makers.\n\nMr. Kopetz retired March 1, 1926, and thus \nbecame the first employee to be pensioned \nunder the benefit plan which was established \nwhen the Laboratories was incorporated in 1925.\n\nShortly after retirement, Mr. Kopetz, with his \nwife and two children, moved to a small farm \nin Greene County in the Catskills. Now both \nchildren are members of the Murray Hill Lab- \noratories; Edward is an instrument and _ tool \nmaker in the Development Shops, and _ his \ndaughter, Mrs. Elsie K. Melroy, a clerical su- \npervisor in Central Files.\n\nMrs. Force joined the Plant Department of \nthe Laboratories at Murray Hill on October 22, \n1942, as a member of the Restaurant staff. For \nthe past two years she had been making salads \nat the Cafeteria counter, where she was a fa- \nmiliar figure to the patrons.\n\n35 years E. F. Helbing W. C. Burger Dorothy Patchell CC. H. Rumpel \nM. J. Kelly H.N. Christopher H. R. Vail J. D. Tebo \nC J. Mazzi F. De Zavala C. A. Webber C. G. Wennerberg\n\n30 years R. F. Elliott 20 years 15 years \nA. O. Adam, Jr. Elizabeth Viggers J. R. Erickson E. F. Billman Theresa Marion \nPhyllis Barton Richard Haard J. H. Harding \nJ. D. Beatty 95 H.W. Heimbach \u00a9, M. Hovgaard 10 years \nM. E. Ellis \u00bb Gears I. M. Kerney A. F. Klare F. C. Cathers \nP. B. Findley A. J. Aikens E. H. Leonard M. R. Kleist C. S. Jackson \nJ. C. Gabriel Erland Anderson V. L. Lundahl Bernice Potwin \u2014_\u2018J. A. Miller \nRosario Gerardi E. T. Ball Ethel McAlevey Mildred Ralph Alma Ostar \nT. B. Grant B. G. Bjornson D. S. Myers H. G. Raupp Muriel Walter\n\n*Lillian Benjamin\u2014Earl W. Fraser \n*Marcae Bitowf\u2014James G. Carolan \nShirley Littman\u2014*Bernard Litwack \n*Janet Mysel\u2014Ralph Sinclair\n\n*Members of the Laboratories. Notices of engage - \nments and weddings should be given to Mrs. Helen \nMcLoughlin, Room 803C, 14th St., Extension 296.\n\nAt Buruincton, L. Vietu, F. S. Corso and \nR. BLAcK were concerned with production of \nthe new broadcast microphone; F. L. Crutcu- \nFIELD, matters relating to the high quality \nmoving coil receiver; T. H. CRABTREE, testing \nproblems associated with hearing aids; and \nL. Vieth, the manufacture of the new broad- \ncast microphone.\n\nH. F. Hopkins attended the meeting on Loud- \nSpeakers of the Sound Equipment Section of\n\nJ. R. Power, J. M. Rocte and F. S. Corso vis- \nited the Western Electric hearing aid dealers \nin Boston, Detroit and Washington in connec- \ntion with the field trials of hearing aid cords.\n\nF. L. CruTCHFIELD was at the Archer Avenue \nplant regarding problems on the new opera- \ntors\u2019 receiver.\n\nF. F. FARNsworTHu visited a number of plants \nin the Pacific northwest area that produce and \ntreat poles, crossarms, and other timber prod- \nucts for Bell System use. He also discussed \ncurrent outside plant development projects \nwith interested engineers of the Mountain\n\nStates Company at Denver and The Pacific \nTelephone and Telegraph Company at Port- \nland, Seattle, San Francisco and Los Angeles.\n\nJ. W. Kenarp and R. P. AsHBauGH visited \nHawthorne and the Illinois Bell Telephone \nCompany for discussions pertaining to the new \ntypes of composite sheath that is being used \nfor exchange area cable.\n\nA. H. HEARN was in Washington in connection \nwith a special committee assignment to revise \nand reissue the present Manual of Standards \nof the American Wood-Preservers\u2019 Association. \nAt Orville, Ohio, he supervised experimental \ntreatments of Douglas fir poles to prevent \nbleeding. Mr. Hearn, G. Q. LumspEN and \nR. H. Couey inspected creosoted southern \npine poles in the Harrisburg area.\n\nF. E. Warp appeared before the Board of Ap- \npeals at the Patent Office in Washington rela- \ntive to an application for patent.\n\nO. H. Coo.inceE, at Milwaukee, spoke on \nTransmission Line Theory at a six-lecture sym- \nposium on telephone and radio transmission \nunder the auspices of the Milwaukee Section \nof the Institute of Radio Engineers.\n\nW. HarTMANN attended the fall meeting of \nthe Optical Society of America in Cincinnati, \nOctober 23 to 25, at which he presented a \npaper, co-authored by B. E. Prescott, The \nQuantitative Spectrochemical Determination of \nBarium, Strontium, and Calcium.\n\nA. E. Jounson, M.D., of the West Street Medi- \ncal staff, has recently completed four articles \nin a series of ten for the Encyclopedia Ameri- \ncana. Dr. Johnson\u2019s subjects include Penicillin \nAerosol; Recent Advances in Tuberculosis; and \nPulmonary Physiology.\n\nC. B. H. FeEtpMan, H. J. FisHer, M. D. BRILL \nand I. E. Fair attended the National Electronics \nConference, November 3 to 5, in Chicago.\n\nJ. R. Power participated in conferences at the \nUnited States Naval Hospital in Philadelphia \non hearing aid problems.\n\nB. E. Stevens conferred at the Magnetic \nWindings Company in Camden on magnetic \nmaterials for power transformers manufactured \nat Winston-Salem.\n\nJ. R. Power and F. S. Corso were in Chicago \nin connection with initiating a field trial on \nhearing-aid receiver cords.\n\nWuat KIND of men are the 2,300 scientists and engineers of Bell \u2018Tcle- \nphone Laboratories?\n\nMen of many types, working in different ficlds of research, may \ncontribute to each development.\n\nBut all have certain characteristics in common: Good minds as a \nfoundation, many years of learning in the fundamentals of their science \nand the methods of research, and a co-operative attitude \u2014 for without \nco-operation of individuals these products of research could never be \nproduced.\n\nAbove all else, however, they have \u201cthe spirit to adventure, the wit \nto question, and the wisdom to accept and use.\u201d\n\nThat kind of men can produce the finest telephone equipment in the \nworld \u2014 and have done so.\n\nBELL TELEPHONE LABORATORIES eExpioriNG AND INVENTING, DEVISING \nAND PERFECTING FOR CONTINUED IMPROVEMENTS AND ECONOMIES IN TELEPHONE SERVICE", "title": "Bell Laboratories Record  1948-01: Vol 26 Iss 1", "trim_reasons": [], "year": 1948}
{"archive_ref": "sim_record-at-t-bell-laboratories_1951-03_29_3", "canonical_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1951-03_29_3", "char_count": 155141, "collection": "archive-org-bell-labs", "doc_id": 892, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc892", "record_count": 197, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1951-03_29_3", "split": "test", "text": "THE COVER: The first intelli- \ngible words over the telephone \nwere a call for help\u2014see \npage 121. From a painting by \nGeorge Rapp.\n\nAN IMPROVED TELEPHONE BATTERY, U. B. Thomas . 3... 97 \nTHE GENERAL PURPOSE ANALOG COMPUTER, A. A. Currie. . 101\n\nA MICROWAVE NOISE SOURCE, W. W. Mumford . . . . . 116 \n7o teams PATENTS ISSUED . 129 \nAT&T ANNUAL REPORT . 122 LIFE INSURANCE... . 130\n\nNEW JERSEY RAIL DISASTER 125 SERVICE ANNIVERSARIES . 133 \nJEWETT FELLOWsHIP . 126 RETIREMENTS . . . . . 134 \nIN MILITARY SERVICE . . 128 PIPE FITTER\u2014ARTIST . . 138\n\n463 WEST STREET, NEW YORK 14, N. Y. \n0. E. BUCKLEY, president J. W. FARRELL, secretary M. B. LONG, treasurer\n\nPAUL B. FINDLEY, editor \nPHILIP C. JONES, science editor \nJULIAN D. TEBO, science editor\n\nR. LINSLEY SHEPHERD, associate editor \nHELEN McLOUGHLIN, assistant editor \nTHEODORE N. POPE, circulation manager\n\nDuring 1950 Bell System Operating Com- \npanies will be receiving storage batteries \nwith an expected life at least 40 per cent \nlonger than those now being used. In addi- \ntion, the new batteries will bring about \nfurther savings, through simplified mainte- \nnance procedures. This improvement results \nfrom the use of calcium in place of anti- \nmony as a hardening agent for the lead \nalloy in the grids and other metallic parts.\n\nThe grids function as supporting struc- \ntures for a battery\u2019s active materials: the \nsponge lead of the negative plate and the \nlead peroxide of the positive plate. Under \nthe conditions encountered in telephone use \nit is now known that the eventual failure of \na storage battery is usually caused by the \ncorrosion of the positive grid. As shown in \nFigure 1, this corrosion results in an expan- \nsion or \u201cgrowth\u201d of the grid structure which \nultimately cracks the grid frame and thus \ndestroys the support for the positive active \nmaterial. The calcium alloy shows less cor- \nrosion and growth, under comparable con- \nditions than the antimony alloy which has\n\nbeen standard for many years. The story \nthat is behind this development is an \ninteresting example of the progress of a \nresearch program the first step of which \nwas the identification of a trace of an un- \nexpected gas in the atmosphere of a battery \nroom. It started in 1930, when a program of \nelectrochemical research on the lead-acid \nstorage battery was initiated under the di- \nrection of H. E. Haring. The purpose was \nto determine whether the existing battery \ntype could be improved from the standpoint \nof the telephone system\u2019s special require- \nments. Although there was little complaint \nagainst the lead-antimony storage battery, \nit had not undergone any radical change in \nfifty years. Inasmuch as recognition of fault \nis prerequisite to improvement, the initial \nobject of the investigation was to discover \nin what ways the operation of the telephone \nbattery was less than ideal. It was believed \nthat this procedure would provide a clue \nwhich would open up a profitable program \nof research. Results obtained from subse- \nquent tests supported this belief but the clue\n\nSometime previously a series of analyses \nof the atmosphere in central offices had \nbeen made in studies of relay contact tar- \nnish. Close re-examination of the results \nrevealed that the concentration of reducing \ngases, such as sulfur dioxide and hydrogen \nsulfide, was always slightly higher in the \natmosphere of a battery room when the \nbattery was being overcharged.\n\nThis mysterious excess of reducing gas \nwas traced to the batteries. Caught in a \nliquid-air trap, it was positively identified \nas stibine (SbH3) a gaseous compound of \nhydrogen and antimony.\u00b0\u00ae\n\nIdentification of this gaseous compound \nof antimony suggested the hypothesis, later \nconfirmed, that in the course of normal bat- \ntery operation, antimony a dissolves \nfrom the positive grid, passes through the \nelectrolyte, and plates out upon the sponge \nlead of the negative. During the overcharge \nperiod, the hydrogen formed at this elec- \ntrode combines with the antimony to form \nthe gas, SbHs. The migration of antimony \nfrom the positive to the negative electrode \nhas two results: the positive grid structure \nis weakened, and the negative plate is con-\n\ntaminated with antimony. Being electro- \npositive to the sponge lead which consti- \ntutes the active material of the negative \nelectrode, the antimony causes local action \nand hence self-discharge. Thus, antimony \nis objectionable in telephone batteries be- \ncause it accelerates self-discharge, and re- \nduces their life by hastening the destruction \nof the positive plates.\n\nTheory showed that the grid-hardening \nmaterial should be electronegative to lead. \nIt should also be present in minute amounts \nand well-dispersed to minimize leaching out. \nA lead-calcium alloy then being developed \nfor cable sheath approximately met the re-\n\nquirements. In 1933 experimental cells using \na lead-calcium alloy containing less than 0.1 \nper cent calcium were made in the labora- \ntory and compared with similar cells using \nthe 12 per cent antimony alloy then being \nused in commercial telephone batteries. As \npredicted, self-discharge was markedly \nslowed down in the calcium cells. They re- \nquired only one-fifth as much current as \nthe antimony cells to maintain them in a \nstate of full charge. Moreover, the new \nalloy was less susceptible to corrosion. \nNext, arrangements were made with a \nmanufacturer to produce a number of full- \nsize lead-calcium cells for field trial. Proper\n\ncontrol of the very small amount of calcium \nrequired presented a major difficulty. Some \nof the calcium originally added to the melt \nwas lost by oxidation during melting and \ncasting; determination of the calcium con- \ntent was laborious and slow. Consequently, \nthe composition of the final castings was \nvariable, with results, however, that later \nproved to be highly significant.\n\nIn November, 1936, a battery of twenty- \nfour cells with a rated capacity of 684 \nampere-hours was made and installed in a \nteletypewriter power plant of the Long \nLines Department at 32 Avenue of the \nAmericas, New York. On an adjacent rack \na comparable battery using the standard \nantimony alloy was installed and operated \nunder identical conditions. After four years \nof operation, some of the positive grid \nframes in the lead-calcium batteries began \nto develop cracks. Since this is usually a \nprecursor of failure, all the lead-calcium \nbatteries were replaced and sent to the \nLaboratories for study.\n\nThe cracked grids, it was found, had not \naffected the electrical performance of the \nbatteries, which still delivered 120 per cent \nof their rated capacity. Moreover, chemical \nanalysis showed that cracking was limited \nto grids having more than the specified 0.1 \nper cent calcium. On the basis of this in- \nformation, it was then decided to order cells \nfor additional tests, with more careful con- \ntrol of calcium content.\n\nIn 1945, after nine years of service, the \nlead-antimony battery at 32 Avenue of the \nAmericas had reached the end of its useful \nlife. A discharge test showed that some of \nits cells could deliver less than 70 per cent \nof their rated capacity, well under the 75 \nper cent minimum required for a telephone \nbattery. In contrast the remaining lead-cal- \ncium cells in simulated service at the Lab- \noratories, still delivered nearly 120 per cent \nof their rated capacity, despite the bad \nphysical condition of many of the positive \ngrids. More recent tests show no significant \ndecrease in capacity of the lead-calcium \ncells after thirteen years of life\u201440 per cent \nlonger than the useful life of standard bat- \nteries. Their total life still remains to be \ndetermined. (See Figure 3).\n\nbatch were disassembled at the end of the \nnine-year period for inspection and com- \nparison with some of the antimony cells \nremoved from 32 Avenue of the Americas. \nThe earlier findings were confirmed: grids \nwith the specified calcium content remained \nin excellent condition; they had no cracks \nand showed relatively little growth, (Figure \n2). Similar results were obtained after thir- \nteen years. The grids with more than 0.1 \nper cent calcium expanded and, in many \ncases, deteriorated badly. As shown by the \ncurve in Figure 4, the degree of growth is \nclosely correlated with the excess calcium. \nThe antimony grids from the cells which \nhad failed in service had grown three to \nfive times as much as lead-calcium grids \ncontaining 0.07 to 0.08 per cent calcium. \nThe grid-cracking in the first commercial \nlead-calcium batteries vividly demonstrated \nthe need for close control of the alloy in the \nmanufacture of these batteries for the Bell \nSystem. For the analysis of calcium in lead, \nthere was developed a completely new and \nvery simple method which virtually elimi-\n\nFig. 4\u2014Growth rate of 11-inch positive plates for \nvarious concentrations of hardening agent.\n\nnates the difficulties of calcium control in \nmanufacture. Sensitive to 0.001 per cent cal- \ncium, the new method takes less than half \nan hour, instead of the forty-eight hours \npreviously required. The first group of bat- \nteries made using the new test method \ndemonstrated that the desired calcium con- \ncentration can now be maintained with \nlittle difficulty.\n\nThe first post-war calcium cells were de- \nlivered in 1947. After a period of laboratory \ntests, twenty-four of these commercial cells \nwere installed in the PBX at the Murray \nHill Laboratories and have been a regular \npart of the power plant since January, 1948. \nThis installation is shown on the first page.\n\nAll three of the Bell System\u2019s regular bat- \ntery suppliers have now made lead-calcium \nbatteries which have operated satisfactorily \nin a broad program of field trials: in large \n48-volt batteries in the Beechmont central \noffice in Cincinnati, the Plantation central \noffice in Pittsburgh, and in the central of- \nfices at Maple Heights and Bedford, Ohio, \nand Tonawanda, N. Y. Most of the 12-volt \nfilament batteries for the TD-2 radio relay \nroute from New York to Chicago will be of \nthe lead-calcium type. Present plans call for \nthe conversion of future production of \nnearly all cells of 180-1680 ampere-hour\n\nCalcium cells will not require the equal- \nizing charges now necessary to counteract \nthe change produced by aging in lead- \nantimony cells. Their low self-discharge \nrate means a reduction of 80 per cent in the \npower consumed in maintaining a state of \nfull charge. More important, the water addi- \ntions required to maintain the electrolyte \nlevel will be reduced by 80 per cent. This \nrepresents an important saving in mainte- \nnance cost in unattended remote locations. \nReduced maintenance and a longer period \nbetween battery replacements are expected \nto result in savings which greatly exceed\n\nA word of caution is directed to those who \nmay expect to buy long-life lead-calcium \nbatteries for other uses. The new battery \nwas developed for the specific requirements \nof the Bell System where the majority of \ntelephone batteries are maintained on a \nclosely regulated floating routine with bat- \nteries held at a voltage just sufficient to \nmaintain a full state of charge. Discharge \noccurs only during the infrequent failures \nof commercial power. With continual charge \nand discharge, lead-calcium may be less \nsatisfactory than lead-antimony.\n\nTHE AUTHOR: Uprron B. Tuomas, Jr., re- \nceived the B.S. degree in chemistry at the College \nof William and Mary in 1929 and joined the Tech- \nnical Staff of the Laboratories in July of that year. \nSince that time he has been engaged in funda- \nmental studies on storage batteries and on the \ncontact resistance characteristics of tarnished metal \nsurfaces. He is active in the Electrochemical Society \nas secretary-treasurer of the Battery Division and \nas associate editor of the Society\u2019s Journal.\n\nSeated at the consol of the computer, Miss R. A. Weiss operates the controls.\n\nOne of the trends in science and tech- \nnology is that the problems requiring solu- \ntion are getting ever more complex. In \ndeveloping communication systems, exten- \nsive research and engineering effort is re- \nquired to design a system which will be \npracticable, reliable, and economical. The \nsuccessful culmination of this effort de- \npends, to a large extent, on the success with \nwhich complex systems of mathematical \nequations are solved. Many of them may \nbe solved in the conventional manner\u2014by \nuse of pencil, paper, slide rule, and pos- \nsibly a desk calculating machine. Others \nare very difficult to solve in this manner, \nand some are not susceptible to hand solu- \ntion at all. It is necessary, therefore, to \nplace more and more reliance on one or \nanother of a variety of computing machines \nwhich have been evolved over the past \ndecade or so. Such computers are of two \ngeneral classes: digital and analog. Typical \nof the former is the relay computer! devel- \noped in these laboratories a few years ago. \nTo date the only analog computers devel- \noped by the Laboratories have been for \nspecific applications,\u201d such as the control of \nanti-aircraft and seacoast artillery fire. Re-\n\ncently, however, a general purpose com- \nputer of the analog type has been con- \nstructed to assist in the work of the \nLaboratories.\n\nIn this computer, electronic circuits are \nused to perform the various mathematical \noperations of addition, subtraction, multi- \nplication, division, integration, and differ- \nentiation. Some of the circuits employed \nfor these purposes have already been de- \nscribed in the Recorp,\u00ae and to a large \nextent the new general purpose analog \ncomputer is based on these earlier circuits. \nIts basic computing unit is a three-stage \nnegative feedback amplifier, shown in sim- \nplified form in Figure 1. A number of \ninputs are provided, all feeding through \ninput resistors of the same value to the \nsignal grid of the first stage. The amplifier \noutput is normally connected through an- \nother resistor of the same value to the \nsignal grid for feedback. When a volt- \nage is applied to one or more of the inputs, \na voltage e appears on the grid on the first\n\nFig. 1 \u2014 Simplified schematic of the basic feedback amplifier of the analog computer.\n\ntube, and a voltage Ey = \u2014 pe at the output \nof the final stage. The currents through \nall but the feedback resistor are of the \nform (E, \u2014e)/R, with the direction of flow \nbeing as indicated. Since the grid of the \ntube draws essentially no current, and the \nconnection to the automatic zero set circuit \nis open except during the momentary check \nperiods, the only path for the current flow- \ning in the input resistors is through the \nfeedback resistor. Current in this latter \nresistor, in the direction indicated, is thus \n(e \u2014Eo)/R. Since this current is the sum \nof the other currents, the basic equation \nfor the amplifier is:\n\nMultiplying both sides of this equation by \nR and rearranging, the equation becomes:\n\nSince ==\u2014ye as noted above, \u2014ko/p \nmay be substituted for e in equation (la). \nThis results in the equation:\n\nIt is this characteristic\u2014that to within \nvery close limits the output voltage is equal \nto the negative of the sum of the input \nvoltages\u2014that makes this type of amplifier \nso useful in the computer. It is, of course, \nnecessary to match the input and feedback \nresistors closely and use care in the design \nof auxiliary circuits associated with the \namplifier to realize accuracies of this order.\n\nWith only slight changes in some of its \nancillary circuits, this same basic amplifier \nis used for most of the functions performed \nby the computer. That it serves as an \nadder is evident from equation 4.\n\nIt becomes a differentiator with respect \nto time if the resistor for the input voltage \nis replaced by a capacitor, as shown in \nFigure 2. With this circuit, ip \u2014 (e \u2014Eo)/R\n\nIn other words Fo is now proportional to \nthe derivative of E, with respect to time.\n\nTo convert the amplifier into an inte- \ngrator, the resistor and capacitor of Figure \n2 are interchanged, giving the arrangement\n\nshown in Figure 3. By equating the cur- \nrents as before, this circuit gives the \nexpression:\n\nFor multiplying and dividing, potenti- \nometers are associated with the amplifier. \nFigure 4, for example, indicates the circuit \nused for multiplying by Kk, where x is less \nthan 1. Here, because of the potentiometer,\n\nThe potentiometer is shaped to compensate \nfor the load imposed by the input resistor \nof the amplifier. The purpose of the ampli- \nfier in this and similar circuits is to prevent \nany variable load of succeeding elements \nof the circuit from affecting the voltage KE.\n\nFor dividing by a constant less than 1, \nthe potentiometer is put in the feedback \npath as indicated in Figure 5. As may \nreadily be calculated, the equation for this \ncircuit is:\n\n(8) \nSince multiplying by a constant greater \nthan 1 is the equivalent of dividing by the \nreciprocal of this constant, multiplication \nby a constant, L, greater than 1 may be \naccomplished by making the k of equation \n(8) equal to the reciprocal of L. Similarly \nto divide by a constant greater than 1, \nequation (7) will serve if k is made equal \nto the reciprocal of the constant.\n\nThe above two equations will take care \nof situations where a variable is to be \nmultiplied by a constant. Very often, how- \never, it is necessary to multiply two \nvariables together, such as xy, or to multiply \none variable by a function of another \nvariable, such as ycosx. For such pur- \nposes, and for a number of others that \nsometimes arise, an amplifier is used as \nthe control element of a servo that drives \ntwo or more potentiometers; the potenti- \nometers may be wound uniformly (i.e.,\n\nlinearly) or shaped to match any one of \nseveral common functions such as the sine, \ncosine, etc. The circuit employed under \nthese conditions is illustrated in Figure 6.\n\nIn this circuit the amplifier is essentially \nthe same as in the other schematics except \nfor the Rc circuit connected to the mid- \npoint of the feedback resistor. This is \ninserted to avoid oscillation or \u201csinging\u201d \nof the servo; insofar as the basic circuit \nequations are concerned, it may be ignored. \nAny output voltage, Eo, from the de ampli- \nfier interacts with one phase of a two-phase \nalternating voltage supplied to the modula- \ntor amplifier by a 200-cycle oscillator. The \nmodulated two-phase voltage is amplified \nand applied to the two-phase servo motor. \nThe motor will thus rotate clockwise or\n\nFig. 2\u2014 Block diagram of the basic amplifier \narranged for differentiation.\n\nFig. 4 \u2014 Block diagram of the basic amplifier \nused for multiplication by a constant.\n\nFig. 5 \u2014 Block diagram of the basic amplifier \nused for division by a constant.\n\ncounterclockwise at a rate dependent on \nthe polarity and magnitude of \u00a3; it will \nstop when Ey = 0.\n\nIn Figure 6 it will be noticed that two \ninputs are connected to the de amplifier\u2014 \na voltage E, representing some variable in \nthe computing system, and a voltage Ey: \nfrom the brush of one potentiometer of \nthe servo. This particular potentiometer \nis connected to the negative computing \nvoltage source, \u2014E.. The servo is so con- \nnected that when E; + Ep; does not equal 0, \nthe resultant control voltage, Eo, will drive\n\nthe motor to a position where E; + Ep: = 0. \nAt this balancing point, the angular brush \nposition will be the same fraction of the \nmaximum angular span, @m, of the potenti- \nometer as FE; is of the computing voltage, \nEc, OF\n\nNow consider a voltage, E2, representing \nanother variable, applied to another potenti- \nometer of the servo. The relative brush\n\nThe above two equations indicate how \nthe servo operates, i.e., it requires a volt- \nage input, E,, and the end product is a \nvoltage, Ep2, appearing on the brush of a \npotentiometer. These quantities may be \ninterpreted mathematically by dividing \nboth sides of the second of the above equa- \ntions by E,; it then becomes\n\nMathematically, then, the voltages in- \nvolved in the operation of the servo may be \nrepresented as decimal fractions of the com- \nputing voltage, and these ratios related to \nthe various quantities of the mathematical \nequations. This same notation is used \nthroughout the computing system in trans- \nlating mathematical quantities to operating \nvoltages. Using this notation, the ratios \nappearing in the above equation may thus \nbe replaced by single letters, as: ry2 re \n>< 1. An example will indicate the proce-\n\nm xX \n200 X 250 == 50000 \nSince each of the variables must be re- \nduced to a decimal fraction, assume that\n\nThe number 90000 is selected arbitrarily \nso that all quantities in the equation will \nbe 1.0 or less. The ratio 0.667 which repre- \nsents m is now to be associated with \u00a3;, and \nif the computing voltage, E., is 100 volts, \nwill equal 0.667 < 100 \u2014 66.7 volts, or\n\nThe ratio 0.833, which represents n, will \nbe similarly associated with F2, and thus \nE2 will be 83.3 volts. The servo will bal- \nance at a value of \u00a9/O,\u2014 0.667, since \nE,/E, \u2014 0.667, and the brush voltage Epe \nwill be Ep2 = E3(@/@m) = 83.3 X 0.667 \u2014= \n55.5 volts. But 55.5 volts is 0.555 &X Ee:\n\nThe output is thus seen to be 0.555 in the \nnotation which has been adopted, which \nagrees with the mathematical equation in \ndecimal fraction notation. Obviously, by \nretranslating this equation, the actual value \nof p is readily obtained: p/90000 = 0.555, \nOr p \u2014 0.555 X 90000 = 50000.\n\nBesides these various uses of the basic \namplifier, it is also used to obtain the basic \n+100 or +50 measuring voltages referred \nto above, to isolate two successive potenti- \nometers in a circuit, and also to multiply \nby \u20141.\n\nJACKS - SERVO POTENTIOMETERS \nPOTENTIOMETERS POTENTIOMETERS POTENTIOMETERS \n1&2 IN EACH 3&4 IN EACH 5&6 IN EACH \nSERVO SERVO SERVO + 7TO12\n\nJACKS AND CORDS \u2014 AMPLIFIER OUTPUTS \n30 ADDERS \n1 DIFFERENTIATOR \n12 INTEGRATORS\n\nJACKS \u2014 AMPLIFIER INPUTS \n30 ADDERS \n1 DIFFERENTIATOR \n12 INTEGRATORS \n9 SERVO AMPS\n\nCORDS - POTENTIOMETER BRUSHES \n(SAME ARRANGEMENT AS POTENTIOMETER JACKS ABOVE)\n\nPOTENTIOMETER KEYS \n(SAME ARRANGEMENT AS POTENTIOMETER JACKS ABOVE) \nTHESE KEYS USED TO APPLY +100 VOLTS TO \nPOTENTIOMETERS IN LIEU OF INPUT VIA JACKS\n\ntegrators, and 9 servos, for which 45 linear \npotentiometers and 15 assorted function \npotentiometers are available. In addition \nthere are 41 potentiometers which may be \nset by hand to supply the constant volt- \nages called for by various equations. To \npermit these units to be selected as required \nin setting up a problem, their inputs and \noutputs appear in one form or another on \na three-panel patch board shown in \nFigure 7.\n\nOn this board, the appropriate terminals \nof all of the adders, integrators, differ-\n\nFig. 9\u2014 A power supply unit at left, a stand carrying \ntwo of the single-motion recorders, middle, and left \nend of the control console at the right.\n\nentiator, servo amplifiers, potentiometers, \nand recording units, are brought to con- \nventional telephone type jacks and plugs. \nIn a complex problem, 100 or more con- \nnections may be required, and thus it \nis necessary to distribute the apparatus \nterminals judiciously to avoid an undue \nconcentration of cords on the board. All \namplifier terminals appear on each bay, \nwith the input terminals connected to\u2019 jacks \nand the output terminals to jacks and cord \nplugs. All potentiometer terminals are \nbrought to jacks, while potentiometer \nbrushes are connected to cord plugs. These \nelements each appear once; they are dis- \ntributed over all three bays. The general \narrangement is shown in Figure 8. By dis- \ntributing the terminals in this way, all bay- \nto-bay patching is avoided, and the patch- \ning for a given problem is distributed \ngenerally over the board as a whole. Most \nconnections are made by cords integral to \nthe board; additional connections may be \nmade as required by separate patch cords. \nLamps on the patch board are associated \nwith each potentiometer or amplifier input \njack for monitoring purposes. A lighted \nlamp informs the operator either that a \nconnection has been made to an input jack \non another bay, or that the computing \nvoltage has been applied to a_potenti- \nometer by operation of the key on the \nbottom panel. No apparatus failure will \noccur if the operator disobeys these warn- \nings, but the resulting circuit will no longer \nbe analogous to the system under study. \nFor recording the solution of problems, \nfour Leeds and Northrup single-motion \ntype-G Speedomax recorders and two Leeds \nand Northrup two-motion type-G Speed- \nomax recorders are employed. In addi- \ntion all servos are equipped with drum \ndials by means of which brush positions \nmay be read to 0.001 for problems where \nonly the end product of the computation \nis significant. The Leeds and Northrup \nsingle-motion units contain a synchronous \nmotor for the paper drive, and a servo-driven \npen. Any voltage applied to the pen servo \nwill cause the pen to assume a propor- \ntionate position between 0 and 1 (0 to 10 \ninches of pen travel) to scales of 1, 2, 5, \n10, 20, 50 or 100 volts per 10 inches as\n\ndesired. A dependent variable may there- \nfore be graphed automatically as a func- \ntion of time.\n\nThe two-motion recorders are similar to \nthe single-motion units except that the pa- \ner as well as the pen carriage is servo \ncontrolled. These units may, therefore, be \nused to plot any two variables regardless \nof how each of them varies with time. \nThese two-motion recorders have been \nmodified to enable them also to introduce \nempirical data into the computing circuits, \nand are thus referred to as function units. \nA curve representing the desired data is \nplotted on the chart of one of these \nrecorders, and then when the pen is made \nto follow the plotted curve, either by \nmanual or by automatic means, a voltage \nproportional to the value of the function \nis supplied to the computing circuits.\n\nThe four single-motion recorders are \nmounted on two movable stands, one of \nwhich is shown in the middle of Figure 9. \nThe two double-motion recorders, which \nare the ones used as function units, are \nmounted on the control console, shown in \nthe photograph at the head of this article. \nIn all, the computer includes 6 units: the \npatching panel, the control console, the \ntwo recorder stands, a power supply unit \n(shown at the left of Figure 9), and the \ncomputing unit, which is shown with its \ncovers removed in Figure 10. This latter \nunit houses the power supply regulators\n\nTHE AUTHOR: After graduating from the \nUniversity of Pittsburgh with a B.S. degree in \nElectrical Engineering in 1933, A. A. Currie spent \nsome seven years with the Westinghouse Electric \nCorporation engaged chiefly in synchronous mo- \ntor design. In January, 1941, he was ordered to \nactive duty as a First Lieutenant in the Coast \nArtillery Corps, and during the next five years \nserved as an Instructor, Officers Division, Coast \nArtillery School, Fort Monroe, Virginia, Instructor- \nin-charge, Fire Control Section, in the Enlisted \nDivision of the Antiaircraft Artillery School. Cam\n\nDavis, N. C., and as a Member of the Antiaircraft \nArtillery Board, becoming a Lieutenant Colonel be- \nfore leaving the Army. In February, 1946, he joined \nthe technical staff of these Laboratories and since \nthen with the Military Electronics Department, he\n\nTABLE I \u2014 ACCURACIES OBTAINABLE IN THE \nVARIOUS COMPONENTS OF THE COMPUTER\n\nSettable to nearest turn of \nwire on control potenti- \nometer. Potentiometer card \nratios accurate to at least one \npart in 1000.\n\nat the left, and the various computer cir- \ncuits at the right. Near the top of this \nlatter unit may be seen three of the servos, \nand in the row of apparatus immediately \nbelow them are some of the servo modu- \nlators and the 200-cycle oscillator. Be- \nneath these, about in the middle of the \ncabinet, is the automatic zero-set circuit, \nand on the left of this are 12 feedback \namplifiers and on its right are 9 amplifiers.\n\nControl of operation of the system for \nmost problems is effected by operation of\n\na single switch on the control console. In \nthe initial, or RESET, position of the switch, \nthe inputs to the integrators are opened, \nthe integrator feedback condensers are \nshort-circuited, and the timer is stopped. \nIn the second, or INTEGRATE, position, the \ninputs are closed, the feedback condensers \nreleased, and the timer started. The third \nposition, MEASURE, may be used where it \nis desired to measure quantities throughout \nthe system at particular values of the inde- \npendent variable. In this position of the \nswitch, the integrator inputs are opened \nand the timer stopped. Since there is no \ncurrent flow on the grid side of the ampli- \nfier, except the very small signal grid \ncurrent and the intermittent automatic zero- \nset grid current, there is no appreciable \nchange of voltage across the feedback con- \ndensers of the integrators, and their output \nvoltages remain fixed. This situation is \nanalogous to \u201cstopping\u201d time for a brief \ninterval.\n\nIn an analog system of this type, where \nthe circuit configuration is variable from \nproblem to problem, individual component \naccuracies must be as high as practicable \nto avoid unduly large over-all errors. In \naddition, of course, the machine must be \nused judiciously to avoid unfavorable scale \nfactors. In general, the component char- \nacteristics shown in Table I indicate the \naccuracies which are obtainable. The \naccuracy of the computer has been checked \non a limited number of problems and has \nbeen found to lie generally within the range\n\n0.1 per cent to 1 per cent. On a recent \nproblem, for instance, which utilized about \n70 per cent of the capacity of the machine, \nit was found that the solutions agreed with \nthose produced by a digital machine to well \nwithin 1 per cent on the average. On other \nproblems, spot checks made by hand com- \nputation have indicated computer accu- \nracies of the order of 0.1 per cent.\n\nThe operation of the computer is rela- \ntively simple but demands a broad knowl- \nedge of the capabilities of the components \nin order to select scale factors and circuit \nconfigurations which are optimum. Main- \ntenance, likewise, is not difficult, tech- \nnically, but may be cumbersome due to \nthe large number of components involved. \nThe components themselves are identical \nwith those developed in World War II \nfor use in fire control systems, and are \nreliable under severe operating conditions. \nThe assignment of a small portion of the \navailable operating time to routine main- \ntenance by skilled engineering personnel \nserves to maintain the instrument at peak \noperating efficiency.\n\nThe general design of the computer was \nconceived by E. Lakatos of the Math- \nematics Research Department, and_ its \ndetailed design and construction was un- \ndertaken by the Military Electronics De- \npartment under the direction of O. H. \nDanielson. J. Maas and E. Habit of this \ngroup, under the direction of J. C. Bain, \ncarried out the mechanical design, and the \nauthor, the electrical design.\n\nIn the general purpose analog computer, \ncircuits are available to perform a number \nof the basic mathematical operations, such \nas addition, subtraction, multiplication, divi- \nsion, integration, and differentiation. In \nsolving a problem, certain of these circuits, \neach corresponding to one of the operations \nrequired in solving a problem, are connected \ntogether in the proper sequence. As the \nfirst step toward a solution, therefore, the \nproblem is analyzed to determine which \ncomputer components should be employed \nand how they should be interconnected. \nAn over-all circuit is then drawn up to indi- \ncate the connections required. To simplify \nthe sketching of this over-all circuit, symbols \nhave been devised to represent the various \nelements of the computer. These are shown \nin Figure 1, where their correspondence to \nthe actual circuits described in a companion \narticle*\u00ae will be evident. All of the circuits \nshown employ the three-stage amplifier, but \nthe amplifier symbol is not shown specif- \nically in the servo symbol.\n\nHow a problem, or a part of a problem, \nmight be drawn to guide in setting it up on \nthe computer is shown in Figure 2. Here \nthe dependent variable v is expressed as a \npolynomial with time as the independent \nvariable. The first integrator, at the upper \nleft, receives a de voltage of unit amplitude. \nThe output voltage of this integrator is \ntherefore proportional to time. This voltage \nis used as the input for the next integrator \nin the chain, whose output is thus propor- \ntional to the square of time. In this manner \nthe successive powers of time may be gen- \nerated. The individual terms are taken off \nthe output terminals of the integrators, and \nthen by use of potentiometers are multiplied \nby fixed constants so as to yield the indi- \nvidual terms of the polynomial. These \nterms, including the constant term, which \nis derived from a unit negative voltage, are \nthen summed by an adder. The resulting \noutput v represents the polynomial.\n\nSuppose instead that the polynomial is \nin a variable y that is not proportional to \ntime. The first step is to convert the input \nvoltage y to a shaft motion in a servo. As\n\nshown in Figure 3 this is done through the \nuse of potentiometer Po driven by the shaft \nof the servo. This servo shaft also drives \npotentiometers P,, P2, Ps, and thus the \nbrushes of all these potentiometers will be \nmaintained at position y. The voltage of \neach brush will thus be y times the voltage \napplied to the winding of the potentiometer. \nA positive unit voltage is applied to the \nwinding of the first potentiometer, and thus \nits brush voltage is +y. An amplifier con- \nverts this to \u2014y, which is applied to the \nwinding of P2, and thus results in \u2014y? on \nthe brush of P2. This process may be con- \ntinued to secure as many powers of y as \nare required. Voltages corresponding to \neach power of y are applied to hand-set \npotentiometers set for the coefficient of \neach respective term. Amplifiers are in- \nserted when needed to change signs. The \namplifiers in the potentiometer chain are \nnot primarily for the purpose of changing \nsign, however, but to avoid the errors that \nwould arise if one potentiometer were a \ndirect load on the preceding one. The vari- \nous terms proportional to powers of y, \ntogether with the constant term from an- \nother hand-set potentiometer, are all con- \nnected to the input of an adder to obtain \nthe output voltage v.\n\nThese examples illustrate the term by \nterm correspondence existing between the \ncircuit and the mathematical expression. \nDifferential equations may be set up by an \nessentially similar process. There is one \ndifference, however. The above expressions \nfor v are not really true equations but \nmerely quantities to be computed. In a \ntrue equation, something must be solved \nfor. To accomplish this, the circuit must \nbe arranged\u2014through the use of feedback- \nto make the two sides of the equation bal- \nance each other. The equation itself should \nbe written with all its terms collected on \none side, thus making the other side zero.\n\nintegration from (2). FEEDBACK t \nThe circuit for this equation is shown in no [uct\n\narticle, the action of the basic feedback Tig. 4\u2014 Circuit set up to solve Equation 8. |\n\nthe amplifier essentially equal to the sum an increasing voltage to appear at the \nof the input voltages exclusive of feedback, output of the integrator 1, and this latter \nor, in general terms, to make: voltage, over feedback path No. 1, will affect | \n(4) = (yp the value of the output voltage y. The | \nat ee ee curve drawn by the recorder will be like | \nSince this output voltage Eo is fed back to that shown in Figure 5. | \nthe input, the sum of all the input voltages The above example is a first-order linear \nincluding the feedback is 0. This relation- differential equation with constant coeffi- | \nship may be obtained, of course, by re- cient; to solve a non-linear first-order \narranging equation (4) to the form: differential equation, a servo will be re-\n\n(3) and thus if the two to (5) dt \namplifier ai of Figure 4 be made equal to \n\u20140.5 and +5 ydt, the output voltage\u2014 \nwhich is fed back to the input\u2014will be\n\nequal to y. (6) y\u2014 ye + ff, (t + y)ydt, = 0, \nThe \u20140.5 input voltage is obtained from \na potentiometer set at 0.5 with unit negative 0.5\n\nvoltage applied to its winding. A voltage Fig. 5 \u2014 Curve \nequal to f\u00b0 ydt is obtained by connecting drawn by the re- \nthe output voltage y to the input of an corder while the \nintegrator, which thus gives \u2014f\u00b0 ydt at its computer is solving \noutput, and then connecting this output to Equation 3. \nanother amplifier to reverse the sign. The\n\noutput of this latter amplifier is thus \" \n+f, ydt, which provides the second input \nvoltage required.\n\nBefore beginning the solution on the \ncomputer, the capacitor in the integrator \nwill be discharged, and the input lead to \nthe integrator is disconnected. The solu- \ntion is started by reconnecting this lead. \nAt the instant of beginning the solution, \nthere will be no voltage on the input sup- \nplied over feedback path No. 1, but there is \na voltage \u20140.5, obtained from the hand-set \npotentiometer at the extreme left, on the \nother input. As a result the output voltage y [ all +y) dt \nwill be +-0.5, and the recorder will begin \u2014 \ndrawing its curve from this point. At once, \nhowever, this voltage will begin to cause\n\n7 as: the solving amplifier, is applied to the point \nof an adder Az whose output is thus \n\u2014(t+y). \npiney! sents To permit the multiplication of (t + y)\n\nby y, the output of the solving amplifier is \nalso connected to a servo, which thus main- \nics nected to it at position y. One of these \npotentiometers Po, is used to control the \noperation on the servo, but to the winding \nthe other, the output of az is con- \n\u00abnected. The voltage on the brush of P, is \nalways \u2014y(t + y), which is the nega- \ntive of the quantity under the integral sign \nssi of equation (6). This voltage is connected \n: to the integrator 12, which thus gives the \n+ I | ul y \n= \nFig. 7 \u2014 Curve drawn by the recorder while the tn / \nwhere yo, the constant of integration, is the | \nvalue of y when t equals 0. @=1.6375 C=-0.0625 4=1.705 \nTo solve this equation, the circuit set up\n\non the computer is as shown in Figure 6. vI vil vil \nAs before, the solving amplifier a; has three =[~ | \u2014 \ninputs: one supplied from a_hand-set \u2014t}\u2014 \npotentiometer giving the value \u2014yo, an- \nother the term y fed back from the output \nof a;, and the third from a circuit arranged | |} \nto give (t+ y)ydt. The output of a, \nwill thus always be equal to y, and is con- | ||| || \u2014 = \nnected both to the recorder and to the | \nauxiliary circuit that derives the above in- \\ \u2018 \ntegral. This auxiliary circuit is more in- \nvolved than that used for the previous | = \nproblem since it must add t and y, multiply | - \nthe sum by y, and then integrate. \nTo secure a voltage equal to t, integrator nile \n11 is employed, which has minus unity Fig. 8 \u2014 Group of graphs made by a recorder w \nvoltage applied to its input. Its output is computer was solving an equation for various \nthus \u20141dt\u2014t. This voltage, together\n\nFig. 9\u2014A non-linear differential equation, above, and the circuit set up on the computer to solve it.\n\nvoltage f\u00b0 (t + y)ydt required as the other \ninput for the solving amplifier. This equa- \ntion is of the Ricatti type, and its solution \nas actually drawn by the recorder is shown \nin Figure 7.\n\nThe ready solution of ordinary differen- \ntial equations, either linear or non-linear, is \none of the outstanding advantages of the \ngeneral purpose analog computer. Once the \nset-up has been made, solutions can be \nturned out at the rate of about one every \ntwo minutes. The highest order that can be \nsolved directly is limited by the number of \nintegrators available, namely 12. This does \nnot mean that all 12th order equations can \nbe solved, for bottlenecks may develop \nfrom other components. On the other hand, \none case of a 15th order equation, occurring \nin a PCM problem, was handled by the \nexpedient of breaking the problem into two \nparts; recording the solution for the first \npart, and using this as the input for the \nsecond part.\n\nWhile ordinary differential equations fre- \nquently occur in the work of the Labora- \ntories, they necessarily comprise only a \nmodest portion of the various mathematical \nproblems requiring solutions. However, in \nmany of these problems, inspection shows\n\nthat the most powerful technique is to de- \nliberately convert their equations into dif- \nferential form! This is a rather astonishing \nconclusion to one used to thinking in terms \nof the conventional hierarchy of math- \nematical operations. Most of us have been \nconditioned to think that integration is a \nhigher-level process than the algebraic \noperations, and the solution of differential \nequations is at a higher level than integra- \ntion. Consequently, there is a feeling that \nin some sense it is wasteful to replace \nalgebraic operations or integrations by \nmethods which require solutions of differ- \nential equation. This, of course, is a reflec- \ntion of the experience that few differential \nequations can be solved by pencil and paper \nmethods. Nevertheless, it has been our ex- \nperience that the differential equation tech- \nnique is generally the simplest and most \npowerful when backed up by a computer of \nthis type.\n\nOne of the virtues of the General Purpose \nAnalog Computer is that it permits a rapid \nand thorough exploration of a complicated \nsituation. An instance of this arose in con- \nnection with the design of delay equalizers \nfor the L-3 carrier system. The design \nengineer wished to determine that com-\n\nwhich would result in optimum perform- \nance. He wished the variable Pp to trace in \nthe complex plane a figure of given shape \nas the independent variable z traversed a \nunit circle centered on the origin. The \nproblem was run with an assumed set of \nvalues of the four constants. The design \nengineer examined the shape of the result- \ning figure, made a few slide rule computa- \ntions, chose a new set of values, and the \nproblem was then re-run, and so on. A \ngroup of these trial runs are shown in \nFigure 8. The individual runs took about \ntwo minutes. The total elapsed time from \nproposal of the problem to the time the \ndesign engineer had his solution was two \nand a half hours. To get the same result by \nany other means available to the Labora- \ntories would probably have taken some \ndays.\n\nAnother instance of such an exploration \noccurred with a problem in telephone relay \ndesign. By analysis, the engineer had re- \nduced the problem to the non-linear differ- \nential equation shown in Figure 9, and he \nwanted the solution of this equation for two\n\nTHE AUTHOR: Graduating with an M.E. de- \ngree from Stevens Institute in 1926, Emory Laka- \nTos spent a year with the New York and Queens \nElectric Company, and then joined the technical \nstaff of the Bell Telephone Laboratories. With the \nApparatus Development Department he first en- \ngaged in the design of telephone transformers, but \na few years later transferred to the Acoustical Re- \nsearch Department. From 1938 to 1941 he was \nassociated with the switching analysis group de- \nveloping switching equipment. During the war \nyears he worked on fire-control projects with the \nPhysical Research Department. During this period \nhe also acted as consultant to Section H, Division \nA, of the N.D.R.C. Since the end of the war he \nhas been with the Mathematical Research Depart- \nment.\n\nvalues of cy and for a wide range of values \nof k. The actual circuit employed is also \nshown in Figure 9.\n\nThis particular problem was complicated \nby the wide range of values of x, which \nnecessitated two changes in the machine \nset up. Nevertheless, the elapsed time from \nreceipt of the problem to completion of the \nruns was only two days. The need for a \ncomplete survey of this particular problem \nhas been felt since about 1938. At any \ntime prior to the advent of the general \npurpose analog computer, however, the \nmeans available for a direct attack on this \nvery complicated problem were so pro- \nhibitively laborious that the engineer con- \ncerned was forced to rely on somewhat un- \nsatisfactory approximations.\n\nMany other problems have been solved \nby the techniques described above. They \ninclude the equations of motion of a tele- \nphone relay for the operate case, an electron \ntrajectory problem, the effect of band lim- \niting on PCM pulse shapes, the roots of a \ntranscendental equation encountered in a \nstudy of piezo-electric crystals, harmonic \nanalysis of speech data, and a non-linear \npartial differential equation arising from the \nstudy of eddy current transients in a mag- \nnetic core. While the machine has only re- \ncently been put into use, it has already \ndemonstrated its versatility and reliability.\n\nOn March 6, 1911, the Engineering De- \npartment of A T & T issued a circular \n(T.C. 30) to the Bell operating companies \npointing out the desirability, from the point \nof view of the general public, of a uniform \nplan of designating jack-per-line party-line \nstations. At this time some companies were \nusing letters only, and some a combina- \ntion of letters and numerals. There was \nalso a considerable difference in the letters \nchosen by the various companies. In four- \nparty service, for instance, such combina- \ntions were being used in the different large \ncities as L.X.J.Y., Y.R.X.L., A.Y.M.Z., X.Y.- \nJ.M., X.Y.Z.K., and J.L.M.R.\n\nWith the thought that there must be \nsome combination of letters which would \ninvolve fewer misunderstandings than any \nother when calls were passed by subscrib- \ners to operators, extensive tests had been \nmade, the results of which were given in \nthe circular. It had been found that the let- \nters which gave the fewest misunderstand- \nings were W, R, J, and M, and the circular \nrequested comments by the operating com- \npanies regarding the standardization of \nthose letters.\n\nThe replies received in answer to this \ncircular indicated a general agreement with \nthe recommendations, and the letters W, \nR, J, and M were accordingly standardized \nfor four-party service in a circular issued \non October 28, 1912 (T.C. 52). For two- \nparty service, the letters W and J were \nchosen, it having been determined that they \noccasioned fewer misunderstandings than \nM and R.\n\nAt the time these letters were adopted, \nmanual service was predominant, and the \neasy distinguishability of the four letters \nadopted was the important factor. With the\n\nrapid expansion of dial switching, and the \nadoption of lettered dials an additional ad- \nvantage became evident, since the four let- \nters selected were sufficiently separated \nin the alphabet to bring each letter op- \nposite a different finger hole. The same \nletters could thus readily be used with the \ndial system since each corresponded to a \ndistinct digit: W \u2014 9, R == 7, J = 5, and \nM = 6.\n\nDid you ever hear the sound caused by \none molecule colliding with another mole- \ncule? Of course not, for the energy involved \nin a single collision is so small that the \nhuman ear does not respond to it. It would \ntake many simultaneous collisions to build \nup the total energy to an audible level, and \nthat is just what happens when steam \nescapes through a small orifice. The char-\n\nE. L. Chinnock holds a fluorescent lamp noise gen- \nerator in his left hand preparatory to a measurement \nof noise figure of an experimental amplifier.\n\nacteristic hissing noise of escaping steam is \nthe result of many many molecular colli- \nsions; as the agitated molecules are being \nforced through the small orifice, they col- \nlide, and the vibrations caused by these im- \npacts add up to an energy level sufficient \nto actuate the mechanism in our ears that \nimparts to us the perception of sound. It \ncan be likened to the noise of many random \ncollisions of billiard balls, although it occurs \nat a much lower sound level.\n\ncollide, but it is a different kind of energy \nand our ears do not hear it. It is electro- \nmagnetic energy, such as is light, to which \nour eyes respond, or X-rays, to which photo- \ngraphic emulsions are sensitive, or radio \nwaves which bring television and radio pro- \ngrams into our homes. As in the case of \nmolecular collisions, single electronic col- \nlisions involve minute amounts of energy \nbut when many many collisions are occur- \nring at random, the total effect may be \nstrong enough to be detected by our eyes. \nThis happens when electricity is applied to \nan incandescent lamp. The flow of current \nheats the filament, and the free electrons \nwithin the metal filament are agitated, col- \nliding as they jostle about. As the electric \ncurrent increases through the lamp, the\n\nmore vehement, and the resulting radiated \nenergy increases until the level is high \nenough to stimulate the eye.\n\nThe eye is not sensitive to all of this \nenergy, however. Some of it lies in the vis- \nible region of the spectrum to be sure, but \nmuch of it lies in other frequency ranges, \nto which the eye is insensitive. Some lies \nabove and some below the visible range, \nand the relative distribution of the energy \ndepends upon how hot the filament be- \ncomes. If it becomes very hot, the relative \nenergy in the high-frequency or ultra-voilet \nrange is enhanced. If it is cooled off, the \nrelative energy in the low-frequency range \nor red region is enhanced. The total ener \navailable in all frequency ranges also pd \npends upon the temperature, among other \nthings; the higher the temperature the more \nthe radiated energy.\n\nWhen a hot metal filament\u2014as in an in- \ncandescent lamp\u2014is operating in a circuit, \nonly that part of the collision energy at \nfrequencies that will pass through the as- \nsociated circuits will appear in the output. \nIn audio frequency circuits, only the audio\n\nwill appear in the output. If its amplitude \nis great enough, it will be heard as a hissing \nsound like that of escaping steam and is \nthus called noise. In high-frequency circuits \nthe portion of the collision energy passed \nby the associated circuits will be far above \nthe audio frequencies, but since in tele- \nphone and radio broadcasting circuits it is \nalways reduced to audio frequencies ulti- \nmately, the collision energy is called noise \neven when at its point of origin it is at fre- \nquencies very much above the audio range.\n\nHow great the noise will be, however, \ndepends on several factors. First it depends \non the intensity of the collisions, that is, on \nthe number of collisions per second and on \nthe velocity of the electrons at the time of \nthe collisions, and both of these depend in \nturn on the temperature of the filament. \nThis relationship is linear; the noise power \nis directly proportional to the absolute tem- \nperature of the filament.\n\nEven with the highest temperatures at \nwhich a metal filament is operated, how- \never, the noise is not great enough to acti- \nvate a loudspeaker. Before the noise can be \nheard, there must be amplifiers between the \nsource and the loudspeaker, and thus a sec- \nond factor in the loudness of the noise is \nthe amount of amplification. Amplification \nin turn brings in a third factor because the \namplifier is an electronic device, and thus \nintroduces noise of its own. This applies \nequally to all different types of repeating \namplifiers, whether they be operating in the \naudio frequency range or the microwave \nrange. Indeed it even applies to our home \nbroadcast and television receivers. The hiss- \ning noise that one hears in the background \nwhen the receiver is tuned to a distant sta- \ntion is caused partly by the accelerated elec- \ntrons colliding with each other within the \nvacuum tubes. The familiar \u201csnow storm\u201d \nthat one sees on a television set when it is \ntuned to a weak station is also caused by \nelectron collision energy. This type of in- \nterference has been known as noise, even \nthough no sound may be involved at all. It \nis called \u201cthermal noise\u201d if it originates in \na hot body, or \u201cshot noise\u201d if it originates \nfrom the interaction of electrons with elec-\n\ntrodes within a vacuum tube. All known \nsubstances, devices, and amplifiers possess \na noise level due to electron collisions which \nwill mask weak signals.\n\nIn any communication system, the output \nnoise is a disturbing influence. There is usu- \nally a definite level of noise, relative to the \nspeech or signal level, which will not im- \npair the quality of the transmission. If this \nrelative level is exceeded by the noise, the \ntransmission of intelligence is degraded. \nConversely, if the level of the noise is too \nfar below the tolerable value, the transmis- \nsion system will be operating uneconom- \nically. As a result, the noise level is an es- \nsential consideration in the design of prac- \ntically all telephone apparatus; its proper \ncontrol is vital in seeking the Bell System\u2019s \nobjective of \u201cthe best possible service at the \nlowest cost.\u201d\n\nOne important component of a telephone \ntransmission system is the vacuum tube am- \nplifier, and a first step in attempting to con- \ntrol its noise level is to evaluate the perform- \nance of the amplifier in terms of the elec- \ntronic noise it generates. This evaluation is \nbased upon the ratio of the actual output \nnoise power of a receiver to the output \nnoise power which would exist if the re- \nceiver contributed no noise of its own. \nThis ratio is called the \u201cNoise Figure\u201d of an \namplifier or receiver. A receiver with a poor \nnoise figure, i.e., a high noise figure, will \ngenerate most of the output noise within it- \nself; only a small fraction of the noise out- \nput will be generated in the circuit attached \nto the input terminals of the receiver. On \nthe other hand, a receiver with a good noise \nfigure, i.e., a low noise figure, will contrib- \nute very little noise power to the total out- \nput power; most of the output noise power \nwill originate in the circuit attached to the\n\ninput terminals of the receiver. From these \nconsiderations it is evident that a receiver \nwith a good noise figure will have a better \nsignal-to-noise ratio in its output circuit \nthan a receiver with a poor noise figure.\n\nTo measure the noise figure of a receiver, \nor circuit of any type, a circuit like that of \nFigure 1 is commonly used. The internal \nresistance of the generator is designated \nr1, and since the noise caused by a resis- \ntance varies as its absolute temperature, it \nis possible to change the noise produced by \nthe generator over a wide range. By making \noutput readings for two different values of \nnoise in the generator without changing the \nconditions of the circuit under test, it is \npossible to determine the noise figure of \nthe receiver.\n\nIf the noise produced in the output with \nthe generator resistance at a temperature \nof 290 degrees absolute is represented by \nngi and the noise output caused by the re- \nceiver is some constant, c, times Ngi, then \nthe total noise output is Ngi + cng. Since \nthe noise figure is defined as the ratio of \nNgi + cngi to the noise output that would \nhave been found had there been no noise \nproduced in the receiver, that is had c \nbeen zero, the noise figure, F, is thus:\n\nSuppose now that a measurement is made \nof output noise when the generator source \nis at 290 degrees absolute and thus con- \ntributes noise of value Ngi, and then that \nthe noise of the generator is increased by in- \ncreasing the temperature of the resistance \nuntil the output noise is some constant y\n\nSince the noise is proportional to the ab- \nsolute temperature, this expression may be \nwritten:\n\n(3a) \nSuch a measurement of noise figure is \ncommonly made by increasing the etfective \ntemperature of the noise generator until the \nnoise output is double what it was before. \nUnder these conditions y = 2. Thus:\n\nSince 12 may be determined, this gives a \nmethod of measuring the noise figure.\n\nNoise generators for measuring the noise \nfigure of receivers have been devised, and \nhave been used throughout most of the use- \nful radio frequency spectrum. At low fre- \nquencies, a vacuum tube diode is com- \nmonly used as a source of noise, and the \neffective temperature of this noise generator \nis readily calculable when the filament is \nsaturated. The noise diode is convenient \nin that it furnishes a source of noise that \ncan be readily adjusted by changing the \nfilament temperature, so that a convenient \nincrement in output noise power can be ob- \ntained. The noise diodes that are commer- \ncially available are not suitable for use in \nthe microwave region, however, because of \ntheir large dimensions relative to the wave- \nlength, and satisfactory measurements of \nnoise figures at microwave frequencies be- \ncome difficult to make. It has been discov- \nered, however, that ordinary fluorescent \ntubes produce a suitable level* of noise, and \nmethods have been devised in these Lab- \noratories for using them to obtain noise \nfigures at microwave frequencies.\n\nFor this purpose the fluorescent tube is \nmounted in a wave guide structure as shown \nin Figure 2. Within the guide the lamp is \nexposed, but the two ends are encased in a \nmetal shield. These shields are too small \nin diameter to permit the propagation of \n4000 mc energy, and hence noise produced \nat the cathode and anode does not enter \nthe guide; the only noise that does is that \noriginating in the positive column, where \nthe electrons are moving rapidly with \nrandom velocities in a stable manner.\n\n*Practically, this level is too low to interfere with \nordinary radio reception.\n\nThese random collisions of electrons give \nrise to microwave noise power equivalent \nto that of a resistance at a temperature of \nabout 11,400 degrees absolute. This tem- \nperature appears to be characteristic of the \nfluorescent lamp under normal conditions. \nOf 32 different lamps, including ten differ- \nent types of fluorescent coatings such as \nused in the pink, red, gold, soft white, \ndaylight, green, white, 4500-degree white, \nblack light, and blue, 31\u00b0 were all within \n+ 0.25 db of each other in excess noise, as \nwas also a germicidal lamp with no fluo- \nrescent coating. The microwave noise power \nfrom a fluorescent lamp is substantially in- \ndependent of the d-c current from 40 ma to \n140 ma, but changes slightly with ambient \ntemperature. This coefficient is only \u2014.055 \ndb per degree centigrade, however, and \ncan be ignored for some applications. The \nnoise power also appears to be indepen- \ndent of frequency, having been checked at \nseveral different frequencies, 65 mc, 3000 \nmc, 4000 mc and 9000 mc.\n\nOne thing that the fluorescent lamp lacks \nas a noise source is variability. When it is \nused to determine the noise figure, there- \nfore, a slightly different test procedure has \nto be followed. Substituting 11,400 de- \ngrees for T2 in equation 3a reduces it to:\n\nand the noise figure is readily found by in- \nserting the observed value for y. If this \nratio were 1.05, for example, the noise figure \nwould be, from Equation 5, F == 38.3/ \n(1.05 \u2014 1) = 766, or 29 db. For poorer re- \nceivers, the accuracy falls off rapidly, but \nwe are usually interested in accurate meas- \nurements only on the good receivers. The \nbest receivers at 4,000 mc have noise fig- \nures around 8 db. This corresponds to an \nincrease in output noise power of about \nseven times when the fluorescent lamp noise \nsource is connected to the input of the re- \nceiver. A power ratio of seven times can \nbe measured readily with fair accuracy on \na thermocouple type of power meter. \nGreater accuracy can be achieved, how- \never, by using a carefully calibrated attenu-\n\n*One of the 32 lamps flickered erratically. At \ntimes its excess noise was % db higher than average.\n\nator in the output of the receiver, in order \nto take advantage of the accuracy obtained \nwhen reading full scale deflection on the \npower meter. This expedient may, however, \nlead to overloading of the receiver on the \npeaks of the noise if the power handling \ncapacity of the output stage of the receiver \nis comparable with the power required to \nactuate the output power meter. In this \nevent, it is better to attenuate the noise \nelsewhere in the setup, where the noise \nlevel is low enough not to cause overload- \ning in the succeeding stages, yet high \nenough so that the noise originating in suc- \nceeding stages is insignificant. If this is not \nconvenient, as for example in an amplifier \ncontained all in one long strip, the cali- \nbrated attenuator can be inserted in the \noutput of the noise source, thus reducing \nthe excess noise to give any convenient\n\nvalue of y desired. Equation 5 no longer \napplies, since the effective temperature of \nthe combination of the fluorescent lamp \nplus the attenuator pad is no longer 11,400 \ndegrees.\n\nTo compute the actual effective tempera- \nture of the combination, one considers the \ncontributions of the two resistances in \nseries, and adds the resulting noise powers \ntogether. One of these resistances, that due \nto the hot noise generator, is at a tempera- \nture of 11,400 degrees and constitutes a \nfraction a of the total resistance, r. The \nother resistance, that due to the attenuator \npad, is at a room temperature, 290 degrees, \nand constitutes a fraction l-a@ of r. The \nmean square noise voltage in a resistance \nR is given by the relation:\n\nwhere k is Boltzman\u2019s constant, T the abso- \nlute temperature, and B the effective band- \nwidth of the system.\n\nFrom the lamp, the noise voltage is thus \n\u00a32 == 4k(11,400) ars, and from the pad it is \nE2 == 4k(290) (1-2)RB. Adding these, gives \nfor the total mean square noise voltage:\n\ntemperature of the combination, and sub- \nstituting this for T2 in Equation 3a gives:\n\nThus it is seen by adding a variable at- \ntenuator to the constant noise source, meas- \nurements of noise figure can be made just \nas accurately as they can be made with a \nvariable noise source.\n\nDuring the course of the experimental \nwork which involved the measurement of \nthe microwave noise level of the fluorescent \nlamp, an interesting tie-up between micro- \nwave noise and black body radiation de- \nveloped. Since the level of the microwave \nnoise energy from a fluorescent lamp is so \nconstant with respect to time, reproducible \nfrom tube to tube, practically independent \nof the current, and only slightly affected \nby the ambient temperature, it might be ex- \npected that it is being controlled or limited\n\nTHE AUTHOR: W. W. Moumrorp majored in \nmathematics and physics at Willamette University, \nand after graduating with an A.B. degree in 1930, \nat once joined the Technical Staff of the Labora- \ntories. With the Radio Research Department at \nHolmdel, he has worked on ultra-short-wave propa- \ngation and microwave components for radio relay \nsystems. During the war he engaged in develop- \nments for radar. Mr. Mumford\u2019s contributions in \nthe microwave field include the directional coupler, \nwide-band coaxial to waveguide transducers, helix \nto waveguide transitions as used in the traveling \nwave tube, and the gas-discharge noise generator.\n\nby some invariant physical property of the \natoms and ions within the gaseous dis- \ncharge. Suppose we ask ourselves, \u201cIf we \nshould terminate our wave guide in a re- \nsistance at 11,400 degrees, what color would \nthe resistance be?\u201d According to Wien\u2019s \ndisplacement law, the wavelength of max- \nimum radiation from such a \u201cblack body\u201d \nis given by the relation:\n\nThis is indeed an interesting result, since \nthe mercury vapor discharge in the fluores- \ncent lamp radiates most of its energy in the \nultravoilet line spectrum at = 2537 (10) \ncm. The design of the lamp was guided by \nan effort to accentuate the radiation at this \nwavelength, and the manufacturers state \nthat this has been achieved so successfully \nthat no other spectral line is excited to radi- \nate more than two per cent of the input \npower.\n\nThe striking similarity between the black \nbody and the mercury vapor discharge at \n2537 (10)-8 cm, suggests the following \nhypothesis:\n\n\u201cIn a gaseous discharge which is radiat- \ning light energy substantially monochro- \nmatically at a particular wavelength, Am, \nthe microwave noise energy is the same as \nthat available from a black body which \nradiates its maximum energy at that wave- \nlength, Am.\u201d\n\nThis hypothesis remains to be proven. It \nis a controversial matter, but in the mean- \ntime we now have a microwave noise source \nthat we have long been waiting for.\n\nHE GREAT EVENT took place in the evenin \nToe March 10, 1876. Alexander Graham Be \nand his assistant Thomas A. Watson were \nworking on Bell\u2019s experiments as they had \nbeen for months in the attic of the boarding \nhouse at 5 Exeter Place, Boston, where Bell \nhad rented two rooms for the purpose. Bell \nin the front room was at the transmitter of the \none-way line and Watson in the rear room \nhad his ear to the receiver with the connect- \ning wire strung out through the hall between \nthem when Bell accidentally spilled battery \nsolution on his trousers. This was serious be- \ncause Bell, then twenty-nine years old; had a \nsmall wardrobe and little money.\n\nTo Watson, Bell\u2019s voice seemed to leap \nfrom the receiver. He had heard sounds here- \ntofore (in June of the previous year Mr. Bell \nhad succeeded in devising and Watson in \nmaking a telephone which transmitted voice \nsounds but not words). This time he heard a \ncomplete sentence. He dropped the _instru- \nment and ran down the hall to Bell\u2019s door. \n\u201cMr. Bell,\u201d he exclaimed, \u201cI heard every word \nyou said\u2014distinctly.\u201d\n\nForgetting all about his acid-spattered \nclothes Bell dashed into Watson\u2019s room to lis- \nten while Watson talked. This was the trium- \nphant moment for which he had been ponder- \ning and laboring for years. Night after night \nhe and Watson had Jabored until the small \nhours of the morning hoping to achieve this.\n\nMan\u2019s desire to extend the sound of his \nvoice is no doubt older than history. Nobody \nknows who first discovered that if he cupped \nhis hands about his mouth and shouted, his \nvoice would travel farther than if the sound \nwere allowed to be dissipated in all directions. \nLater the megaphone, the speaking tube, and \nthe string telephone or \u201clover\u2019s telegraph\u201d were \ndevised. In the thousands of years that man had \nbeen talking, he had developed no better facili- \nties for transmitting voice sounds than these \nuntil that historic night seventy-five years ago \nwhen Bell proved that it was possible to gen- \nerate and use a current of electricity that \nwould \u201cundulate\u201d as he put it, or vary in in- \ntensity as sound waves, shrill or deep, loud \nor soft, vary as they disturb the air. He had \ncome to understand that to transmit sound \nelectrically, he had to have a current that could \nbe \u201cshaped\u201d by sound.\n\nIn Bell\u2019s newly devised variable resistance \ntransmitter on this night of March 10, 1876, \na wire attached to a voice-operated diaphragm \ndipped into a cup of acidulated water, the\n\ncup with its conducting liquid and the mov- \nable wire forming part of the electrical circuit \ncontaining also a battery and a receiving tele- \nphone. As the diaphragm vibrated, the wire \nwould rise and fall in the liquid causing the \nresistance of the circuit to change and the \ncurrent to vary in conformity to the sound \nwaves set up by the voice directed upon the \ndiaphragm.\n\nIn today\u2019s transmitter, carbon granules in a \ntiny chamber rest against the diaphragm. As \nthe vibrations of the diaphragm alternately \npress the granules closer together and allow \nthem to relax, the varying pressure on the gran- \nules causes a varying resistance to the talking \ncurrent just as in the liquid transmitter. In \nlater receivers, the simple reed became a \ndiphragm, which has undergone continued de- \nvelopment right up to the present time.\n\nReplicas of the liquid transmitter and tuned \nreed receiver used on this memorable eve- \nning are in the collection of the Bell System \nHistorical Museum at the Bell Telephone \nLaboratories. A length of the original wire \nconnecting the two rooms has been preserved.\n\nSystem served more people and handled two \nand a half billion more conversations than in \nany previous year. The national emergency \nbrought a new tide of telephone demand. Serv- \nice high in quality was well maintained. The \nmen and women of the System did a superb \njob. \n; Today more than ever, our country has the \nmost and the best telephone service in the \nworld. This is a bulwark of defense. The tele- \nphone speeds production. It is vital to the \nArmed Services. It spreads warning against \nattack. As the nation gathers its strength, the \nurgent calls of the hour are on the telephone \nlines.\n\nThe Bell System is alert to its essential task \nand is preparing further. The System will do \nits full part in helping to keep America secure. \nThat is our Number One job today.\n\nOperating revenues of the System in 1950 \nwere $3,261,528,000, an increase of $368,255,- \n000 or about 13 per cent over 1949, Operating \nexpenses of $2,334,362,000 compared with \n$2,248,833,000 in 1949. Net operating income \nin 1950 was less than 4.5 per cent on telephone \nplant and the return would be even lower if \nthe plant were valued at present costs. The rate \nof earnings on capital was 6.1 per cent, com- \npared with 4.9 per cent in 1949 and 7 per cent \nin 1940, ten years ago.\n\nWage, tax and material costs have continued \nto mount. Telephone rates and earnings must \nbe sufficient to meet these increases, to attract \nand protect equity capital, and to assure a sound \nfinancial structure. With rates adequate to \nmaintain the necessary earnings, most of the \nCompany\u2019s convertible debentures would be \nconverted into stock and the debt would be \nreduced.\n\nThough the gain in telephones in 1950 was \na little less than in 1949, the gain in the second \nhalf of the year was higher. Toll and long dis- \ntance messages increased throughout the year \nand most sharply in the second half. Demand \nwent up as the nation acted to meet the defense \nemergency. The importance of good telephone \nservice was never more clear.\n\nSix million telephones were installed to meet \nnew requests for service. The net increase of \n1,955,000 brought the total number in service\n\nto over 35,300,000. Toll and long distance calls \nwere put through at an average speed of 1.6 \nminutes and 94 per cent of them were handled \nwhile the calling party held the line.\n\nThe whole expansion and improvement of \nthe telephone system since World War II have \nmade it more sturdy, more useful, more con- \nvenient, and a greater asset to the nation. The \nlong distance network has grown from 16 to \n26 million miles. Dial service has been greatly \nexpanded for both toll and local calling. Rural \nareas served by the Bell Companies have more \nthan twice as many telephones as in 1945. Serv- \nice to automobiles, trucks, boats, trains and \nother vehicles is available in nearly 150 areas \nand more than 9,000 vehicles are being served \n\u2014a 20 per cent increase last year.\n\nRadio relay systems were extended more \nthan 2,000 miles in 1950 and construction to \nprovide transcontinental telephone service by \nradio relay is now under way. These systems, \nlike coaxial cables, provide wide communica- \ntions highways that can carry hundreds of tele- \nphone conversations, or television programs. \nFacilities for carrying television programs more \nthan doubled during the year and networks \nnow serve 82 television stations in 44 cities, or \n19 more cities than at the end of 1949. These \nstations broadcast over areas containing half the \ncountry\u2019s population.\n\nBell System research has resulted in a greater \noutput of new things in the last five years than \nin any similar period before. It has produced \na new telephone set for the customer\u2019s use; new \nkinds of wires and cables to carry his messages; \nnew ways of using the wires and cables more \nefficiently; new devices and systems for carry- \ning intelligence by microwave, without wires; \nnew dial systems to connect telephones to- \ngether; and new ways of automatically record- \ning information for the customer\u2019s bill.\n\nThese things will contribute greatly to the \nnation\u2019s defense, both in providing essential \ncommunication services and in reducing the\n\nA supply of copies of the Annual Report \nis available in the Libraries at West \nStreet and at Murray Hill.\n\nneed for scarce materials. For example, a new \nsystem is now being manufactured that permits \ntwelve conversations to be carried over each \nfour wires in short telephone cables, from 20 \nto 200 miles in length. This can be applied to \ncables already in place and hence will be of \nparticular value in saving copper. Carrier sys- \ntems, of which this is one, have been employed \nfor years and Bell Laboratories has led the \nworld in their development. Previous systems, \nhowever, have been economical only over long \ndistances. The new \u201cN Carrier,\u201d combining new \nideas and miniature apparatus, opens up a \nwhole new field to the carrier art.\n\nBell Laboratories scientists are diligently \nseeking substitutes for various scarce materials \nused in telephone equipment. Research on ma- \nterials has long been a major Laboratories \nactivity, and the experience gained will help \ngreatly in obtaining utmost economy in critical \nmaterials, with a minimum of compromise in \nthe quality of equipment.\n\nBell System research facilities are also an in- \nvaluable national resource in the development \nof military equipment. Bell Laboratories con- \ntributions in World War II, especially in radar, \nsubmarine detection, and per and bombing \ncontrol systems, as well as in military communi- \ncations by wire and radio, played a major role \non land and sea and in the air. Since the war \na substantial amount of research has been con- \ntinued for the Armed Services on projects re- \nquiring the Laboratories\u2019 skills and facilities.\n\nWestern Electric Company had another busy \nyear meeting the heavy needs of the Bell \nTelephone Companies. Production was again \nfar above the level of pre-war years although \nnot as large as in 1949. Sales amounted to \n$758,064,000. Eighty-seven per cent of the \nsales were to Bell Telephone Companies and \nmost of the remainder to the United States \nGovernment. Earnings for the year were $38,- \n647,000 or 5.1 per cent of sales.\n\nThe Company\u2019s ability to produce telephone \nequipment and cable for the Bell Telephone \nCompanies has already been affected by short- \nages of strategic materials such as copper, zinc, \nnickel, rubber and aluminum. Continuing dif- \nficulties in obtaining materials and parts may \ncurtail deliveries to the Bell System in 1951. \nWestern Electric and Bell Laboratories are \ntaking advantage of all practical substitutions \nand adaptations but these can meet the problem \nonly in part.\n\nSince the end of World War II the Company \nhas continued to produce much special equip-\n\nment for the Government defense departments. \nDeliveries to the Government in 1950 exceeded \n$53,000,000. In recent months work has started \non large additional orders for military equip- \nment, chiefly electronic in character.\n\nThrough its subsidiary, the Sandia Corpora- \ntion, Western Electric with the cooperation and \nassistance of Bell Telephone Laboratories con- \ntinues to operate the Sandia Laboratory at \nAlbuquerque, New Mexico, for the Atomic \nEnergy Commission. Sandia is concerned with \nmilitary applications of atomic energy.\n\nFor three quarters of a century the Bell Sys- \ntem has rendered service of more and more \nvalue to the American people. The telephone \nbegan in this country. Here it has been most \nwidely developed and used. Our service has \nalways been the best in the world, and its \ngreatest increase in usefulness has come in the \nlast five-years. This is a great asset in helping \nto defend the freedom of the United States.\n\nOur telephone service is also a product of \nfreedom. In the building of the Bell System, \ncountless discoveries and inventions have had \nto be achieved by the inquiring spirit of free \nmen. Opportunity has been _ to all. Com- \npetition has flourished throughout the organi- \nzation. Worthwhile incentives and reasonable \nrewards have fostered the will and capacity for \nleadership. In the rendering of service day by \nday, the responsibility to get the message \nthrough is accepted as a public trust: that too \nis the exercise of freedom.\n\nAll that has been achieved flows from the \nnation we serve. Under public regulation, the \nBell System has generally been allowed the \nfreedom it needs to perform its service well. It \nis essential that this freedom to serve be undim- \ninished; that research and invention go vigor- \nously forward; that new leaders be encouraged \nand prepared to lead; and that earnings be fully \nadequate to continue to pay good yr me to \nemployees, and a return to investors sufficient \nto attract and protect the billions of dollars of \nsavings that make the service possible.\n\nThrough the years private enterprise and \npublic policy in telephone communication have \nreturned to the nation a value beyond price. \nWe are confident they will do no less in the \nyears to come. We are determined to meet the \nresponsibilties entrusted to us, and we pledge \nour utmost efforts, always, in devotion to the \npublic service and to the lasting security and \nadvantage of the people of the United States.\n\nOn January 24, O. B. Blackwell, former \nVice President of the Laboratories, who, be- \nfore his retirement in 1947, was Assistant \nVice President of A T & T, received the Edi- \nson Medal for \u201chis pioneer contributions to \nthe art of telephone transmission.\u201d Presenta- \ntion was made at a General Session during the \nWinter General Meeting of the American In- \nstitute of Electrical Engineers at the Hotel \nStatler, New York City.\n\nO. B. Blackwell (left) receiving the Edison Medal \nfrom T. G. LeClair, President of the American Insti- \ntute of Electrical Engineers. H. S. Osborne, Chief \nEngineer, A T & T, is looking on\n\nmost important lines were built around the \nbig cities because going through the cities \nwould mean going through sections of cable, \nand that would spoil the transmission. At that \ntime, the Bell \u2014 served approximately \ntwo million telephones.\n\nMr. Blackwell was one of the outstanding \ncontributors to the developments in the tele- \nphone transmission art that brought about the \nvast extension and growth of long distance \ntelephony. Early contributions were new test- \ning routines and splicing methods which made \npracticable the application of the phantom \nprinciple to cables. He also invented and de- \nsigned the first transmission measuring set and \ninaugurated a system of transmission mainte- \nnance measurements that has been of untold \nvalue in maintaining transmission quality of \nthe telephone plant throughout the country.\n\n\u201cI can testify\u201d said Mr. Osborne, \u201cto the \ninspiration of working with Mr. Blackwell.\n\nHonorable Dan A. Kimball, Under Secretary of the Navy, and Rear Admiral C. M. Bolster, assistant \nchief for Research and Development of the Bureau of Aeronautics, visited the Murray Hill Laboratory \non Monday, January 29, to review the Laboratories\u2019 progress on research programs for the Navy. Left \nto right: R. K. Potter, J. B. Fisk, Mr. Kimball, Admiral Bolster and M. J. Kelly.\n\nHis creative imagination and penetrating \npower of analysis made him a real leader. His \nintellectual honesty was infectious. His keen \nsense of humor and dry wit, and his unfailing \nconsideration for all, made it a great personal \npleasure to work under his supervision or to \ncollaborate with him. He was not only our \nleader, but was and is our cherished friend.\u201d\n\nIn his response, Mr. Blackwell expressed \nhis appreciation to those who had placed his \nname on the roster of medalists. \u201cA medalist,\u201d \nsaid Mr. Blackwell, \u201cis allowed a short time \nduring which he is expected to say something \nimportant, or failing that, at least something \ninteresting. The most important matters I \nknow today are the appalling problems of \nhuman relations which confront us. A man \nwho has had the experience of seeing and \nfeeling the power of large scale, properly \nstaffed and organized scientific investigation, \nnaturally wonders what a similar approach \nwould do to the problems involved in human \nrelations.\n\n\u201cWe develop a great deal intellectually as \nwe become adults, but our emotional progress \nis not very marked\u2014and emotions are the \ndriving forces in our lives. The pleasantest \nuse of the intellect is in rationalizing the ac- \ntions our emotions bid us do. It is men who \nhave not grown up emotionally and do not \nknow it, and cannot be told, who are danger- \nous. Just now the world\u2019s \u2018emotional children\u2019 \nare playing with fire.\n\n\u201cSurely, everyone should welcome whatever \nthe intellect of man can find to help his emo- \ntional childishness. \u201cThe proper study of \nmankind is man\u2019 is more completely true to- \nday than when it was written; a study with \nevery technique and tool that man can devise, \nand in coverage and intensity commensurate \nwith its importance.\u201d\n\nEmergency crewmen of the New Jersey Bell \nTelephone Company doubled as rescuers and \ninstallers at the scene of the Pennsylvania \ntrain wreck at Woodbridge, New Jersey. \nEighty-four commuters died in the accident \nand over 500 were injured. Among the casual- \nties were four New Jersey Bell people killed \nand 24 injured. No members of the Labora- \ntories were involved.\n\nWithin an hour of the fatal crash, 32 tele- \nphones had been installed within a hundred \nyards of the wreck for use by police, Red Cross \nand the press. Emergency lighting equipment, \ncutting tools, blankets, bandages and other med-\n\nThree repairmen and one installer, who were \nworking on regular assignments in the vicinity \nof the crash, immediately left for the scene and \nbegan first aid treatment for those trapped in \nthe telescoped rail cars that teetered on the \nedge of a 20-foot embankment.\n\nArrangements were made to clear all lines in \nthe vicinity for emergency use. Party-line tele- \nphones in four homes nearest the wreck were \nconverted into individual lines and with the\n\nDuring this month the American Red Cross \nis making its annual appeal to all of us for \nfunds to support its work at home and with \nour Armed Forces. With so many of our men \nand women already in service, the cause of the\n\nFurther information about the campaign will \nbe sent to all members of the Laboratories call- \ning attention to the opportunity to make contrib- \nutions to the Red Cross and giving information\n\nA traffic supervisor who was a passenger on \nthe train but was uninjured left for the Wood- \nbridge central office after assisting in first aid \nrelief and there coordinated traffic operations. \nWithin the first six hours after the crash, a total \nof 150,000 calls above average were handled \nby exchanges within a 25-mile radius. Despite \nthe heavy calling, the record traffic was handled \nwith dispatch by operators, many of whom \nreported to their posts as soon as word of the \ndisaster reached them.\n\nDr. W. D. Widdowson was on leave for a \nmonth to organize an emergency medical dis- \naster plan for the Military and Civil Defense \nCommission for Pennsylvania at Harrisburg.\n\nIn a letter to Dr. Buckley, Richard K. Mel- \nlon, Commanding Officer of the Commission, \nwrote: \u201cYour cooperation in making Dr. W. W. \nWiddowson\u2019s services available to this Commis- \nsion during the recent critical planning period\n\nThere\u2019s no anachronism here; the beards are just as \nreal and even newer than the A.M.A. equipment \nwhich the wearers are watching. The group are busi- \nnessmen of Media, Pa., who engaged in a beard \ngrowing contest as a feature of the town\u2019s centennial \ncelebration. The telephone in Media has a long his- \ntory; it was introduced in 1881.\u2014The Telephone News.\n\nis one of the most valuable contributions so \nfar made to civil defense in Pennsylvania.\n\n\u201cIt goes without saying that to draft a State \nemergency plan for maximum use of medical \nresources, both of manpower and material, \nwould be an enormous task under any circum- \nstances. To do so within a month\u2019s time, with- \nout prior study either of the general problem \nor of local conditions, is an assignment that \nwould have discouraged all but the most stout- \nhearted. Dr. Widdowson not only met the \nchallenge, but brought to it such enthusiasm \nand good nature that all who worked with him \nwere sorry when his leave of absence came to \nan end.\n\n\u201cIn thanking you for this invaluable aid, I \nam speaking not only for myself personally, \nbut for everyone concerned with civil defense \nin the Commonwealth, which means more \nthan ten million men, women, and children.\u201d\n\nA seventh group of promising young scien- \ntists have been named as recipients of \nthe 1951-52 Frank B. Jewett post-doctoral \nfellowships. The awards, designed to stimu- \nlate and further the work of researchers in \nthe physical sciences, grant $3,000 to the \nrecipient and $1,500 to the institution at \nwhich he chooses to do his research.\n\nRecipients, with the subject of their re- \nsearches and the institution where they will \nwork are:\n\nMurray Gerstenhaber, higher mathematics, \nparticularly mapping problems; Harvard or \nChicago.\n\nDonald Roy Francis Cochran, light nuclei \nand their energy levels; specifically, the reac- \ntions of tritium with helium and _ beryllium; \nJohns Hopkins.\n\nIlse Lis] Novak, relation algebras; Institute \nfor Advanced Study, Princeton.\n\nDonald Robert Yennie, theory of elemen- \ntary particles; Institute for Advanced Study.\n\nAwards were made on recommendation of \na committee consisting of Ralph Bown, chair- \nman; M. B. Long secretary; C. S. Fuller, \nL. A. Wooten, Ha Nyquist, T. C. Fry, \nJ. B. Fisk, and William Shockley. Primary \ncriteria were the demonstrated research ability \nof the applicant, the fundamental importance \nof the problem proposed and the likelihood \nof growth as a scientist.\n\nA number of Laboratories People partici- \npated in the second A.I.E.E.-I.R.E.-N.B.S. \nConference on High Frequency Measurements \nheld in Washington, January 10 to 12, 1951. \nThe meeting was organized by the Joint \nA.LE.E.-I.R.E. Committee on High Frequency \nMeasurements of which E. I. Green is I.R.E. \nGroup Chairman, and E. P. Felch and E. W. \nHoughton are members.\n\nAmong the twenty-five papers in the four \ntechnical sessions were the following by Lab- \noratories authors: Measuring Techniques for \nBroad Band Long Distance Radio Relays by \nW. J. Albersheim; A Precise Sweep Frequency\n\nSome of the participants of the Introductory Survey posed for their picture at Murray Hill.\n\nMethod of Vector Impedance Measurement \nby D. A. Alsberg; Wide Band Swept Fre- \nquency Measurements Applicable to Traveling \nWave Tubes by F. E. Radcliffe; Measure- \nment of Characteristics of Crystal Units by \nL. F. Koerner; and High Frequency Crystal \nUnits for Primary Frequency Standards by \nA. W. Warner.\n\nE. P. Felch arranged and presided at the \nsession on Measurement of Transmission and \nReception and E. W. Houghton, that on Meas- \nurement of Power and Attenuation.\n\nE. I. Green presided at the evening demon- \nstration session which was also the Annual \nJoint Meeting of the Washington Sections of \nA.LE.E, and I.R.E. W. E. Kock, assisted by \nF. K. Harvey, presented a demonstration of \nthe parallel behavior of microwaves and centi- \nmeter wavelength sound waves to accompany \nhis paper, Measurement of Microwave Field \nPatterns Using Photographic Techniques.\n\nAlso attending the conference were A. W. \nClement, J. R. Flegal, A. A. Roetken, B. S. \nWoodmansee, G. R. Frantz, A. G. Fox, W. M. \nGoodall, F. F. Merriam, F. A. Polkinghorn, \nO. E. DeLange, E. D. Reed, L. H. Von Ohl- \nsen, R. C. Pomeroy, V. W. Wall, L. E. Cisne, \nC. A. Bieling and D. M. Black.\n\nAs a means of acquainting the new mem- \nbers of the technical staff with the several \nareas of technical effort and staff activities \nembraced by the Laboratories, and to intro- \nduce them to their associates\u2014particularly \nthose in management positions\u2014a series of\n\nmeetings was held during December and \nJanuary. These meetings, along with visits to \nA T & T Long Lines, the New York Telephone \nCompany, and Western Electric at Kearny, \ngave the new people an opportunity to be- \ncome better acquainted with what is going \non in the Laboratories and in other Bell Sys- \ntem Companies.\n\nTalks by representatives of management \nincluded descriptions of the functions of the \nLaboratories in the Bell System and discus- \nsions of the broad policies in those functions. \nTechnical aspects were described, such as \nSwitching Development, Transmission Devel- \nopment and Engineering, Systems Engineer- \ning, Electronic Apparatus, and Research. Visits \nto the several Departmental Laboratories gave \nthe new members a coordinated view of the \nmany Laboratories activities. Inspection trips to \nthe other Bell System locations were precede \nwith suitable talks.\n\nDuring January, five more Laboratories peo- \nple were granted leaves of absence to enter \nmilitary service.\n\nWilliam J. Benjamin, who came to the Lab- \noratories in October last year, has enlisted in \nthe Navy and is now a Chief Hospital Corps- \nman at the U. S. Naval Hospital in St. Albans, \nLong Island.\n\nThomas W. Kearkuff, an assembler of elec- \ntron tubes, who also came here last October, \nhas enlisted in the Air Force and is now in \ntraining in Texas.\n\nClarence L. Ransome, a Helper in the \nrestaurant, has been employed by the Labora-\n\ntories since May, 1950. He has been called for \nmilitary service in the Army, and is now at \nFort Dix, N. J.\n\nFrederick J. Skinner, member of the Tech- \nnical Staff, came to the Laboratories in Au- \ngust, 1938. During World War II he served \ntor 3% years, returning to the Laboratories in \nDecember 1945. As a reservist, he returns to \nduty as a Major in the Signal Corps, and is \nnow at the Coles Signal Laboratory at Fort \nMonmouth, N. J.\n\nJames E. Trogdon, Restaurant Helper, en- \nlisted in the Army and is now at Fort Dix. \nHe came to the Laboratories in October, 1950.\n\nBell Laboratories\u2019 engineers were honored by \nthe American Institute of Electrical Engineers \nduring the Winter General Meeting at the \nHotel Statler, New York, January 22 to 26. \nJohn Meszar of Switching Systems Develop- \nment received the first prize in the Communica- \ntion Division for his paper Fundamentals of the \nAutomatic Telephone Message Accounting Sys- \ntem. Second prize in this division was awarded \nto B. Ostendorf, Jr., of Telegraph Transmission \ni a for his paper A New Electronic \nTelegraph Regenerative Repeater.\n\nThe Laboratories took a very active part in \nthe program of the Meeting. Technical papers \nand talks were given at sessions held through- \nout the week. At a meeting on Magnetic Mate- \nrials at High Frequencies, January 23, with \nR. M. Bozorth presiding, K. K. Darrow gave \na talk, Electronic and Nuclear Magnetic \nResonance. On the same day, L. H. Germer\n\nspoke on Electrical Breakdown of Very Short \nGaps at a session on Electrical Breakdown in \nGases. On January 24, at a session on New \nTechniques of Network Synthesis, R. B. Black- \nman spoke on Transducer Design Based on \nStatistical Properties of the Signal. In the after- \nnoon, at a session on Advances in the Com- \nmunication Switching Art, Telephone and Tele- \ngraph, with R. C. Davis presiding, W. M. \nBacon and G. A. Locke presented a paper \nentitled A Full Automatic Private Line Tele- \ntypewriter Switching System; at the same ses- \nsion, N. A. Newell and A. Weaver presented \na paper entitled Single Frequency Signaling \nSystem for Supervision and Dialing Over Long \nDistance Telephone Trunks.\n\nOn January 25, at a meeting on A New \nCarrier System for Medium Haul Telephone \nCircuits, with P. G. Edwards presiding, R. S. \nCaruthers gave a paper entitled The Type N-1 \nCarrier Telephone System \u2014 Objectives and \nTransmission Features. This was followed by \nthe paper N-1 Carrier Telephone System \u2014 Ap-\n\nL. Espenschied spoke on The Genesis of Sub- \nmarine Cables at a meeting in the afternoon on \nElectronic Paths Under the Sea to mark the \ncentennial of cable laying. At this session, J. J. \nGilbert presented his paper A Submarine Tele- \nphone Cable with Submerged Repeaters.\n\nOn January 26, at a session on Point-to-Point \nand Mobile Radio Communication, L. A. Dorf \ngave a paper, Operational Study of a Highway \nMobile Telephone System. In the session on \nFeedback Control Systems, J. C. Lozier pre- \nsented a paper entitled Carrier Controlled \nRelay Servos.\n\nCommittee meetings were attended by a \nnumber of Laboratories representatives. R. K. \nHonaman was present at a meeting of the \nPublications Committee; at the Forum of Tech- \nnical Committee Chairmen, R. C. Davis, L. G. \nAbraham and J. D. Tebo represented their \ncommittees. D. E. Trucksess attended meet- \nings of the Electronic Power Converter Com- \nmittee, Hot Cathode Power Converter Subcom- \nmittee, and Metallic Rectifiers Committee. \nW. H. Tidd, vice-chairman of the Electronic \nInstruments Sub-Committee, attended a meet- \ning of that committee. L. G. Abraham and\n\nR. C. Davis were at the Communication Divi- \nsion Advisory Committee meeting. Mr. Abra- \nham is chairman of the Wire Communication \nSystem Committee and P. G. Edwards is \nsecretary; they both attended a meeting of \nthis committee. H. A. Affel was also present. \nR. C. Davis, chairman, and John Meszar, secre- \ntary, attended a meeting of the Communica- \ntion Switching Systems Committee.\n\nW. H. MacWilliams, vice-chairman of Com- \nputing Devices, attended a meeting of his com- \nmittee. R. L. Dietzold was present at the \nElectric Circuit Theory Subcommittee and \nBasic Science Committee meetings. J. G. Fer- \nguson attended a luncheon meeting of the \nCommittee on Instruments and Measurements; \nE. F. Watson and R. B. Shanck were at the \nTelegraph Systems meeting. E. I. Green, chair- \nman of the Science and Electronics Division \nAdvisory Committee, called a meeting at which \nJ. D. Tebo, member-at-large, was present. Mr. \nGreen also attended a luncheon meeting of the \nTechnical Advisory Committee. J. D. Tebo was \npresent at meetings of Section Delegates, the \nNominating Committee, and the Committee on \nBasic Sciences. Mr. Tebo also attended the \nluncheons that were given for medalists K. G. \nCompton and O. B. Blackwell.\n\nE.L. Alford J. B. DeCoste A. E. Hague F. W. Koller A. T. Nordsieck = W.G. Shepherd (2) \nA. E. Anderson O. E. DeLange D. A. S. Hale (2) J. G. Kreer (2) H. G. Och F. J. Singer\n\nH. L. Barne G.H.Duhnkrack _J. B. Harley B. F. Lewis E. Peterson(3) \u00a7 G.C. Southworth \n).F. Barry (2) B. Dysart H. C. Harrison W. D. Lewis (2) L. C. Peterson L. J. Stacy\n\nU.S. Berger W. B. Ellwood (3) _R. V. L. Hartley E. Ley K. W. Pfleger (2) J. C. Steinberg \nC.A. Bieling E. L. Erwin G. Hecht G. A. Locke J.R. Pierce (4) \u2014_R. R. Stevens (2) \nA.E.Bowen(2) L. Espenschied R. A. Heising J. J. Lukacs R. K. Potter(3)  W.B.\n\nW.H. Brattain (5) A. G. Fox (4) A. N. Holden A. S. Martins A. J. Rack R. L. Wallace \nH.B. Brehm M. Fritts (2) H. F. Hopkins W. P. Mason (4) O.E. Rasmussen\u2019 H. M.\n\nM.D. Brill R. R. Galbreath R. T. Jenkins C. F. Mattke (2) W. T. Rea (8) E. J. Walsh (3) \nH.W. Bryant A. S. Gano A.E. Joel J. H. McConnell J. B. Retallack J. R. Weeks\n\nH. Carmer M. S. Glass A. C. Keller K. W. Miller S. D. Robertson S. B. \u2018wetinaas (3) \nP. Caroselli M. C. Goddard S. B. Kent H. A. Miloche V. L. Ronci O. H. Williford \nL.E.Cheesman (2) W.M. Goodall L. A. Kille R. C. Miner W. L. Roth A. Wilson \n-W. Clayden J.W.Gooderham __ K. L. King (2) D. Mitchell (3) A. L. Samuel (2) I. G. Wilson \nC.F. Clement R. S. Gormley J. P. Kinzer L. F. Moose A. K. Schenck D. E. Wooldridge\n\nE. Coleman E. I. Green W. A. Klute W. A. Munson R. W. Sears G. R. Yenzer \n}.E. Corbin L. Gross W. A. Knoop R. C. Newhouse W. L. Shafer W. R.\n\nEverybody pays for life insurance. You pay \nfor it if you purchase the insurance needed; \notherwise your family pays for it by doing \nwithout things they should have. All of us can \nset up a program of insurance which, supple- \nmented by Social Security and the Employees \nBenefit Plan, will take care of most needs after \nwe are gone.\n\nSaving money is always hard work. It re- \nquires resolve and sacrifice and determination. \nbut these are the qualities necessary to carry \nany plan to completion. Without them we get \nnowhere. Since buying life insurance is saving \nmoney, it can be done with confidence.\n\nThere are certain expenses which will be \nwith us always; such as tood, rent or its equiv- \nalent, clothing, medical expenses, recreation, \ngeneral running expenses, insurance premiums, \netc, Since this is the case it is only common \nsense to understand and make proper provision \nfor them. This usually means to budget for \nthem. There are many books dealing with budg- \nets so that we can know when we are setting \naside a proper amount for most of these items.\n\nCash, to meet immediate expenses of last \nillness and funeral, to pay up debts and cur- \nrent obligations, to meet administration ex- \npenses (lawyers and other fees, etc.), to pay \nthe current income tax, to provide running \nexpenses for the family until the estate can be \nsettled. For many there will also be cash \noo to pay Federal Estate and State \nInheritance Taxes.\n\nFunds to provide permanent housing for the \nfamily. If there is a ana it is then a question \nof whether it is best to liquidate the mortgage, \nto sell and to rent elsewhere or to continue \nthe mortgage. No matter. what is decided, \nfunds are required to meet the situation.\n\nAn income is required to enable the wife to \nkeep the home together and to meet the cur- \nrent living expenses. This can be done on a \nshort time basis to give her a period of adjust- \nment to the new circumstances and time to find \na means of earning the required income her-\n\nself. Or it can be done on a basis which will \nenable her to keep the home together and man- \nage it herself until the children have acquired \ntheir secondary or college educations and are \nable to then support themselves and to help \ntheir mother. Or it can be arranged so that the \nwife will have an income for her lifetime \nthereafter and be independent of assistance \nfrom the children.\n\nEducational expenses. This is a need which \nattracts all parents today and is one which is \nmost often provided, to the exclusion of other \nneeds. Many times the wife will unselfishly \nimpoverish herself in order to set aside funds \nfor this purpose. But it is only one of the needs \nwhich a father must provide for. It should be \nconsidered as a part of the whole picture in \nsetting up a program.\n\nRetirement income. After all we either die \ntoo soon or live too long. Both situations re- \nquire financial planning. Fortunately however, \nif we plan wisely for the former, and have \nluck, those same funds will also take care of \nthe latter.\n\nA well organized and designed financial \n\u2014 is not completed overnight. It is\n\nuilt to meet the requirements discussed above. \nIt should be changed or revised as these evolve. \nFirst we examine what each now has. There \nis the Bell Laboratories Benefit and Pension \nPlan. There are the death and retirement bene- \nfits from Social Security. Then, there is the \nlife insurance already owned as well as cash \nin the bank, investments and equity in any \nproperty owned.\n\nThese assets must be examined to determine \nhow closely they meet the requirements. If \nthey do not, then more life insurance is \nneeded. This is the only way the needed \nmoney or income can be provided quickly. It \nwould take many years otherwise to provide \nthe needed funds through straight savings or \ninvestment. For one seriously desiring to \nprovide for his family, judgment requires \nthe device of the insured method instead of \ntaking the chance of living long enough to pro- \nvide otherwise.\n\n< A brief welcoming \naddress was delivered \nby H. J. Delchamps, \nVice President of the \nChapter, at the start of \nthe Pioneer party.\n\nH. G. Geetlein, at > \nthe piano and \"Evelyn \nDee (Mrs. Geetlein) \nentertain an audience \nthat responded with de- \nmands for encores.\n\nOn Wednesday evening, January 17, the \nNew Jersey Council held another regional get- \ntogether in the East Orange Hotel Suburban. \nCol. Loren B. Thompson of the New Jersey \nBell Telephone Company gave a_ first-hand \nfactual presentation on Korea and the Far East, \nwith a prolonged and interesting question and \ndiscussion period following.\n\nA short musical program was given by Harry \nGeetlein, Evelyn Dee (Mrs. Harry Geetlein), \ncourse on Koreaand the and John Dry. A pleasant evening was topped \nFar East was closely off by drawing for door prizes which were won \nfollowed. His listeners by D. G. Blattner, Mrs. R. T. Jenkins, and \nsubsequently were _  E. G. Conover, and a social hour around the \nloath to break for the buffet table where both active and life mem- \nrefreshments as wit- bers with their guests enjoyed the Pioneer fel- \nnessed by the pro- lowship. More than 180 reservations were \nlonged questioning to received for the party.\n\nthe Colonel and \nathe iron The Winnahs! A. J. Akehurst, left, is handing E. G. \nsmaller groups. Conover a prize drawn by Mrs. Geetlein who also \ngives another that she drew to D. G. Blattner. K. P. \nHansen, rear, glows because his wheel-of-fortune\n\nR. P. Wells operating the Civilian Defense Radio \nControl Station at Livingston, N. J.\n\nIncreasingly nowadays, the Recorp learns \nabout Laboratories people who are accepting \none or several responsibilities within the en- \nsarging civilian defense effort. In the amateur \nradio communications field, for example, func- \ntional organization is now geared for faster \ntempo, and activity of New Jersey radio club \nmembers, many of them from the Laboratories, \ngrows accordingly. The following news relating \nto the counties of Morris and Essex has come \nto light recently and similar items will be wel- \ncomed by the Recorp.\n\nIn Morristown, efficient amateur radio com- \nmunications emergency service for the Red \nCross, community and area civil defense per- \nsonnel, as well as for municipal authorities, is \na major objective of the Morris Radio Club. As \nthe name implies, this club is made up of radio \namateurs who are organizing the club\u2019s local \nfacilities in cooperation with the American \nRadio Relay League, the parent organization \nof radio amateurs. Radio telephone and tele- \ngraph networks, partially with independent \npower supply, have already been established \nwithin the amateur bands, These networks per- \nmit fixed or portable stations and mobile units \nowned by individual amateurs to communicate \nwith one another under a versatile centralized \ncontrol. This control, at the Morristown head- \nquarters of the Red Cross, will tie in with \nstate, inter-county, and national networks to be \nalerted as necessary to relay messages to and \nfrom National Headquarters of the Red Cross \nor other relief authorities.\n\nThe club\u2019s apparatus and procedures are \nbeing perfected and periodic drills are held to \nfamiliarize members with emergency operation \nand maintenance. Among the more active mem- \nbers who are also Laboratories employees are \nT. W. Winternitz, president of the Morristown \nRadio Club, T. A. McCann, G. H. Day, L. R. \nSchreiner, B. McWhan, A. H. Lince, F. A. \nShorkley, B. F. Orchard, J. W. Schrage, V. \nWall, W. P. Slichter, and F. B. Walters.\n\nIn Essex County 22 communities are lining \nup to link amateur, police and civil defense \ncommunications to headquarters located at the \nWest Orange armory.\n\nLivingston is a typical community enlarging \nits own communications through cooperation \nwith the Livingston Radio Club, Inc., and the \nCivil Defense Council. The township police \nradio system at present serves the communities \nof Roseland, Essex Fells and West Caldwell \nand an enlarged a is being established. A \nradio trunk line already exists between Living- \nston and Morris County headquarters. R. P. \nWells of the Laboratories is president of the \nLivingston Radio Club as well as Civil Defense \nCommunications Director, American Radio \nLeague Emergency Coordinator and Secretary \nof the Council of Essex County Emergency \nCoordinators.\n\nCivil defense activities of the Bloomfield \nRadio Club members include R. L. Tambling \nwho is American Radio League Emergency \nCoordinator for Glen Ridge, Radio Communi- \ncations (amateur) Chairman of the Glen Ridge \nCivil Defense Council and Chief Operator \nof the Bloomfield Radio Club; W. R. Neisser, \na member of the Emergency Communications \nBoard of the Bloomfield Radio Club; E. J. \nAridas, a member of the American Radio\n\nRelay League Emergency Corps for Newark; \nand J. I. Stockwell, Jr., who is American Radio \nLeague Emergency Coordinator for Montclair, \nRadio Communications Chairman of the Mont- \nclair Civil Defense Council, and of the Amer- \nican Red Cross Chapter, and member of the \nEmergency Communications Board of the \nBloomfield Radio Club.\n\nThe Bloomfield Radio Club\u2019s emergency \ncommunication plan was put into operation \nat the request of the Montclair Red Cross at the \ntime of the explosion at Perth Amboy. Radio \ncommunications were established between\n\nPerth Amboy and Montclair using three mobile \nradio units and a fixed station. This network \nhandled vital messages necessary for proper \noperations of the Montclair Red Cross Chapter \nat Perth Amboy. Recently the club has put on \ntests for the Sheriff of Essex County, the New- \nark Police, and several Civil Defense Councils \nin Essex County.\n\nAll radio amateurs in Morris and Essex \nCounties and environs are invited to participate \nin this emergency communications program \neither by operating their own or available sta- \ntions in the established networks.\n\n15 years \nElizabeth Chambers \n80 years \n85 years L. P. Bartheld \nH. A. Affel F. S. Entz \nT.C. Fry G.S. \n0. C. Hall A. J.\n\nAmong those retiring from the Laboratories \nare May Reilly and S. H. Anderson, each with \n34 years of service; and W. A. Bischoff and \nA. I. Crawford, 31 years.\n\nshe got a position in Wanamakers book depart- \nment. The work was congenial, and she built \nup a clientele which eventually included some \nof Western Electric\u2019s patent executives. Be- \ncause she could find books with but a slight \nclue to the title, she was offered a position in \nthe Patent Files and joined Western Electric \nin 1917.\n\nThere is a rumor that when Miss Reilly is \nasked to verify something that happened long \nago, she puts on a special pair of gloves with \neyes in their finger tips. Be that as it may, she \nis remarkably successful in her quests. A good \nmemory helps, but her meticulous supervision \nof her force over the last quarter-century and \nher genius for good housekeeping are princi- \npally responsible.\n\nLiving alone in upper Manhattan, Miss \nReilly looks forward to her leisure years as time \nto be filled with service to others, perhaps as \na hospital volunteer, and as time to catch up \nwith her reading.\n\nSoon after joining Western Electric in 1916 \nMr. Anderson came to the Systems Develop- \nment group at West Street, where until 1927 \nhe was engaged in the development of central \noffice systems, particularly power plant equip- \nment. He was especially active in the early \ndevelopment of large gas engines for standby \nemergency power generation in central offices. \nHe then transferred to what is now the Quality\n\nAssurance Department and continued to devote \nhis time to power apparatus and equipment. \nFor many years he has been in charge of a \ngroup handling quality surveys, investigation \nof complaints and related activities in this field.\n\nThe Andersons are home-owners in Hollis. \nTheir immediate plan is to return to their native \ntown, Brookfield, Missouri, where many of their \nkin still live. If they like it, they may stay; if \nnot, they will come back to Hollis. Mr. Ander- \nson is a graduate of University of Missouri \n(B.S. in Eng. 1916), and a member of Eta \nKappa Nu, Tau Beta Pi, and A.LE.E.\n\nAfter several years\u2019 experience elsewhere, \nBill Bischoff joined us as a draftsman early in \n1920. His first work was on the printing tele- \ngraph, at that time built by Western Electric. \nIn 1924 he was given charge of apparatus in- \nformation files, the Apparatus Card Catalogue \nand some of the departmental staff services. \nThree years later, placed in charge of appa- \nratus drafting, he reorganized the procedures \nof the group and became actively interested in \nstandardization work. One of his first activities \nin this connection involved the establishment \nof a new series of Laboratories drawings from \nwhich Western Electric manufacturing draw- \nings are produced photographically and without \nsubstantial alteration. This change has effected \nvery large savings in over-all drafting effort.\n\nDuring World War II Mr. Bischoff entered \nthe standardization problem on a national scale. \nAs chairman of an A.S.A. war committee on \ndrafting, he and his associates in Army, Navy \nand industry brought together representatives \nof many important industries and secured \nagreement with the Armed Forces as to uniform \nsizes of drawings, presentation of information \non them, and to a considerable extent the \nstandardization of details. Work of this group \nis now being carried on by the Armed Services\n\nHeard on the Telephone Hour February 5. \nAustin Bailey (left) A T & T, and Arthur Os- \nwald (right), Bell Laboratories, re-enact the \nfirst call with Tom Shirley, the announcer.\n\nAnnouncer: Let\u2019s turn back the calendar \ntwenty-five years\u2014to the night of February 6, \n1926. In a quiet room in New York City a man \nwith earphones on his head listens intently. \nThe hands of a clock move to eleven minutes \npast eleven. The man hears a crackling and \nhumming and then a voice!\n\nAnnouncer: The voices you have just heard \nare those of Austin Bailey and Arthur Oswald, \nspeaking once again the words they used in \nthe first telephone conversation ever to span \nthe Atlantic. That was twenty-five years ago, \nand Bailey and Oswald were members of the \nBell System team that for many months had \nbeen working to make two-way Trans-Atlantic \ntelephone service a reality.\n\nAnd today they are still working on re- \nsearch to extend the range of the human voice \n\u2014Austin Bailey at the headquarters of Amer- \nican Telephone and Telegraph Company, and \nArthur Oswald at Bell Telephone Laboratories.\n\nTheir work today has the same objective as it \nhad on that night twenty-five years ago\u2014to \nhelp make your telephone service ever better, \nwhether you call across town or the ocean.\n\nfor you to talk with anyone beyond this con- \ntinent. Today, you can talk from your own \ntelephone to almost every other telephone in \nthe world. A businessman picks up his tele- \nphone and orders supplies from Australia or \nSouth Africa. The far corners of the world are \nin instant touch with America by telephone.\n\nAnd today, across this land the telephone \ngives the orders, receives the orders, and helps \nmove into high gear the productive power of \nthe most productive nation on earth.\n\nAnd yet it\u2019s only twenty-five years since that \nfirst telephone conversation across an ocean. \nWhat another 25 years will bring no man can \ntell. But this we do know. Improvement in \nyour telephone service will continue, and each \nstep will come from the teamwork of telephone \npeople like Austin Bailey and Arthur Oswald, \nthe two men who were the first to talk together \nacross an ocean's span.\n\nThank you Mr. Bailey and Mr. Oswald for \nre-enacting that night twenty-five years ago.\n\nIn 1948 Mr. Bischoff moved to Murray Hill \nwhere he has most recently been in charge of \nall drafting and technical file records relating \nto transmission, station and outside plant work \nengineered at that location. In recent years he \nhas been chairman of the Laboratories\u2019 design \nstandards committee which specifies standards \nfor drawings, parts and processes.\n\nFlorida where in the vicinity of St. Petersburg \nthey are building a house. Bill expects to play \ngolf, go fishing, and make a garden, to catch up \nwith some of the leisure he has long wanted. \nThe Bischoffs have one son, a master technical \nsergeant in the Air Force.\n\nDuring Allen Crawford's Bell System career \nhe has seen electronics unfold from a few \n\u201cvacuum tubes\u201d to the extensive line of today.\n\nEntering Western Electric here at West Street \nin 1920, he had had engineering training at \nPurdue and elsewhere, and with his drafting \nexperience he first went into that work. Soon \nhe transferred to mechanical design; his sub- \nsequent work on tube bases and sockets indi: \ncated a transfer in 1928 to electronics develop- \nment. For many years he headed up a group \nwhich was responsible for the preparation, issu- \nance and distribution of manufacturing draw- \nings, specifications and technical information on \nvacuum tubes. He alsu had under his charge \nthe development of packaging for vacuum tubes \nand the coordination of the technical investi- \ngations of field complaints and the quality \nassurance matters within the Vacuum Tube \nDevelopment Department. For several years \nMr. Crawford was a member of standardization \ncommittees under the Joint Electron Tube \nEngineering Council. Thus he took part in \nthe coordination of many manufacturers who \nduring World War II made identical tubes \nfor various government agencies.\n\nWhen it was decided to station a group of \nLaboratories engineers at the Western Electric \nplant in Allentown, Mr. Crawford laid out their \nphysical facilities and supervised the installa- \ntion. At that time the specification work was \nmoved to Allentown, so he gave up that job, \nand worked on the layout of the department\u2019s \nfacilities at Murray Hill, supervising that move \nalso. Recently he has been working on labora- \ntory layouts for various developments, and do- \ning a number of special jobs.\n\nMr. and Mrs. Crawford have leased their \nhome in Manhattan and have completed a \nhouse at 500 Rafael Boulevard, Shell Isle, St.\n\nPetersburg, Florida, where they expect to spend \nthe major part of each year, returning North \nfor short visits. Mr. Crawford has been inter- \nested in numerous real estate activities in the \npast and expects to devote considerable time \nto this activity in the future. This, along with \nother hobbies such as golf, fishing, etc., will \nundoubtedly keep him busy. He has one son in \nbusiness in Dover, Delaware, and three boys \nstill in preparatory school.\n\nA sust of ALEXANDER GRAHAM BELL for in- \nstallation in the Hall of Fame on New York \nUniversity\u2019s campus will be executed by Miss \nEthel P. Hood. Among other portrait busts \nfrom her hands are those of Helen Hayes and \nBeatrice Lillie. Selection of Miss Hood was \nmade by a committee of distinguished sculp- \ntors. Dr. Bell was accorded recognition in the \nHall of Fame last year, along with Woodrow \nWilson, General Gorgas, Susan B. Anthony, \nTheodore Roosevelt and Josiah Willard Gibbs.\n\nProressor M. H. L. Pryce of the University \nof Oxford visited Murray Hill on January 10 \nand at a conference there spoke on recent de- \nvelopments in paramagnetic resonance. This \nconsists in the appearance, at certain character- \nistic frequencies, of enhanced magnetic prop- \nerties in certain metals and inorganic com- \npounds, The study of this phenomenon in- \ncreases our knowledge of the properties and \nstructure of solids in general and of magnetic \nsubstances in particular. It serves as a tool in \nthe continuing effort to improve the materials \nneeded by the Bell System.\n\nGreatness, in this hour, is not beyond the \nreach of any man, nor is it reserved for \nthose who make the headlines in world \naffairs. Its fuller measure is the willingness \nand capacity to rise to the urgency of the \ntimes, whatever the job may be.\n\nMen with these elements of greatness are \nshaping this country\u2019s destiny today. Many \nwear the proud uniforms of our country. \nMany are in the hundreds of thousands of \nfactories, mines and offices throughout the\n\nland. They are launching, once again, the \nmost powerful weapon that history has \never known\u2014the might of American \nproduction.\n\nGuiding this effort are the men of manage- \nment. They have spent their lives planning \nfor the future, inventing or developing new \nmachines, building new resources. These \nare the men who, in the last five years, \nhave lifted the industrial capacity of this \ncountry to a level never dreamed of before.\n\nIt is America\u2019s production that is the foun- \ndation of her strength and security.\n\nJ. W. Gorcas of the Switching Systems De- \nvelopment Department has devised a radically \ndifferent and very much simpler form for cir- \ncuit schematics that would replace both the \nSD and OS forms now used. He has explained \nthe construction and uses of the new schematic \nto A T & T and Western Electric engineers at \nKearny, and recently\u2014with D. H. WETHERELL \nand R. E, Cotiis\u2014went to Chicago to discuss \nthe new schematics with Hawthorne engineers. \nMore information regarding this new type of \ncircuit schematic will appear in a later issue of \nthe REcorp.\n\nType-N Carrier is a new twelve-channel sys- \ntem intended for shorter distances than have \nbeen thought economical for type-K carrier. A \nnumber of pre-production systems have been\n\nMiss Gertrude Schleifer, who has recently be- \ncome one of the three supervisors who share \nsupervision of the West Street Restaurant. She \nis a graduate of Pratt Institute (1947) in diet- \netics and institutional management. Her initial \nassignment is to the kitchen; after this tour \nis completed, in April, she will become better \nknown to restaurant patrons.\n\nbuilt at Kearny and installed in various parts \nof the country. A. J. Aikens has visited one of \nthese, at Harrisburg, to investigate noise dif- \nficulties. He has also visited Eastern Maryland \nto study the probable effect on a proposed \ntype-N system of nearby radio stations that \nuse the same part of the spectrum. He also \nmade tests on a newly-installed type-N system \nbetween Boston and Gardner.\n\nWhen men go out deliberately looking for \ntrouble, that\u2019s news: news that attracts atten- \ntion and arouses interests. Those responsible \nfor preparing script for the Telephone Hour \nfelt this way about the story of the Laboratories \nstatic studies in the September Rrcorp (page \n409), and briefly described the work for the \nTelephone Hour program of January 29 under \nthe above title. Most of the equipment used \nfor these studies is shown in the accompany- \ning interior view of one end of the Hamoka- \nhodo Test Station in Madison, Florida. The \ntape recorders on which the static was recorded \nare on the large table near the left center.\n\nA top notch noon hour program of folk songs, \nold favorites and close harmonies was presented \nby the Murray Hill Glee Club in the Arnold \nAuditorium January 25 with U. A. Matson as \naccompanist. Staging coordinator was A. J. \nAkehurst. The Glee Club, a newcomer to the \nentertainment field at Murray Hill, is the out- \ngrowth of informal gatherings of a singing \ngroup previously at Graybar Varick. Organized\n\ny R. H. Klie, with A, R. Rienstra as coach and \nconductor, the Glee Club includes the Barber \nShoppers, a twelve-man close harmony com- \nbination directed by G. B. Thomas, Jr.\n\nThose on stage for the first presentation were: \nB. H. Carmer, Jr., T. R. D. Collins, R. D. \nEhrbar, M. O. Fichter, G. T, Ford, C. T. God- \ndard, D. W. Grant, N. J. Herbert, ]. J. Jansen, \nR. H. Klie, S. Korba, H. P. Lynch, A. R. Rien- \nstra, E. F, Sartori, G. B. Thomas, Jr., R. L. \nTrent, D. C. Weller, N. C. Wittwer, W. H. \nYocom.\n\nShown below is one of Mr. Murphy's regular \njobs\u2014maintaining the sprinkler system.\n\n_ so many Laboratories people, it is hard \nto guess what Luke Murphy of the West \nStreet Building Shop does in his spare time \nwhen you see him working with a four-inch \npipe. Who would think that he could handle \na tiny artist\u2019s brush with the same facility? And \nyet Luke is an accomplished artist, having stu- \ndied in his native city of Liverpool, England \nand in Wexford, Eire. He is a member of the \nIrish Academy in New York. In fact, during the \nIrish Art Exhibit at the De Motte Galleries in \nNew York January 1948, Mr. Murphy\u2019s \u201cSea- \nscape\u201d was exhibited along with paintings and \nsculpture by such masters as Sir William Orpen, \nSir John Lavery and Andrew O'Connor,\n\nMr. Murphy came to America in 1921 as a \npassenger, but shortly afterward he was back \nat sea, working for several different shipping \nconcerns. Within two years, he had visited \nmany places on both the East and West Coasts \nof North America, Central America and South \nAmerica, Between 1923 and 1927, employed \nby the United States Lines, he made eighty \ncrossings of the Atlantic. The next fourteen \nyears he spent with the American Sugar Re- \nfining Company in Brooklyn, coming to the \nLaboratories as a steam fitter in 1941.\n\nThe Murphys live on Long Island in Kew \nGardens with their two dogs and a parrot. They \nhave no children. While out walking with his \ndogs on the evening of November 22, 1950, \nhe was startled by the crash on the Long Island \nRailroad. For the next six hours, Mr. Murphy \naided in rescue work\u2014a job, he says, that he \nhopes he will never again have to do.\n\nDuring lunch hour, cards provide recreation. In. \nthe photograph, left to right, are E. H. Johnson, \nJames Marshall, Thomas Walsh (back to cam- \nera), Mr. Murphy, Harold Shaw (partly con- \ncealed), and H. A. Rosenbohm.\n\nThe photograph, right center, shows Mr. Mur- \nphy threading a large pipe with the thread-\n\nMr. Murphy at his hobby of oil painting. \nMarine scenes are his favorite subject.\n\nMr. Siegmund (center) with Dr. Six (left) and \nProfessor Bahler. Dr. Six is with the N. V. \nPhilips Company in Eindhoven, Holland. Pro- \nfessor Bahler of Delft University is a consultant \nof the Philips Company.\n\none side of the street, in the Eastern Sector, \nwooden poles are lashed together to form the \nscaffold. At the extreme left, in the Western \nSector, a modern steel derrick and a concrete \npouring tower can be seen.\n\nIn the village of Bad Orb, a milch cow pulls a \nwagon that uses a skid to replace rear wheels.\n\nH. O. Siegmund, Switching Apparatus En- \ngineer, who visited Europe recently to study \ntelephone switching developments there, had \nan opportunity to compare East and West \nBerlin during his travels. \u201cFrom the stand- \npoint of personal reaction,\u201d says Mr. Sieg- \nmund, \u201cmy visit to Berlin was the high spot \nof the trip. In the Soviet Zone, I saw the \ncomplete destruction of the buildings formerly \noccupied by the Reich Chancellery, the Luft- \nwaffe and the Propaganda Ministry. All were \nin complete ruins and, for the most part, \nremain so, including the palace of former \nKaiser Wilhelm, the National Museum, Na- \ntional Library, the Cathedral and other public \nbuildings. There is extensive effort to rebuild \nin the Eastern Sector, particularly buildings \nfor use by the People\u2019s Government. At the \nboundary between the East and West Sectors, \nthe most casual observer cannot fail to recog- \nnize the striking differences in the conditions \nunder which life is carried on and work is \nbeing done in the two zones. In the Western \nSector, reconstruction is accomplished by \nmeans of modern power equipment, tools, \ntrucks, materials, and facilities comparable \nto what we are accustomed to in America. \nAcross the street, in the Eastern Sector, the \nmethods are different. Here one finds little \npower equipment, hand labor largely from \nwomen, derricks operated by hand-cranked \nwinches, scaffolding of tree trunks mortised \ntogether and lashed in place with rope.\n\n\u201cIn the rural sections of occupied Germany, \nthe scars of war are much less in evidence \nand life there is quaint and unchanged. The \nneeds of the people are small and their work \nis primitive. In these places, age and antiquity \nare all around.\u201d\n\nThe West Street mixed chorus, directed b \nR. P. Yeaton, will appear in concert on Marc \n20 at Grand Central terminal. The 40 voices, \naccompanied at the piano and organ by Grace \nWagner and Betty Garrow, will be heard in \na group of hymns and spirituals at 5:30 p.m.\n\nhe \u2014 \ni \n<> \nom. \nContrasti ] hods i li \nri ontrasting construction methods in Berlin. On \n=.\n\nSeveral members of the Laboratories at- \ntended the 1951 annual meeting of the Ameri- \ncan Physical Society held at Columbia Uni- \nversity in New York City, February 1 to 3. \nG. E. Moore, secretary of the Division of Elec- \ntronic Physics, took an active part in arranging \nthe meeting; preceding the annual meeting his \nDivision, along with the Panel on Electron \nTubes of the Research and Development Board \nheld a conference at the City College of New \nYork on January 30-31. At this conference, \nJ. A. BEckER gave a paper on Migration of \nW Atoms on the Surface of a W Single Crystal \nAs a Function of Temperature and Electric \nField Strength. H. D. Hacstrum spoke on \nEjection of Electrons by Positive Ion Impact: \nHe+ and Het++ on Mo; H. W. Auuison and \nG. E. Moore gave a paper on The Adsorption \nof Sr Metal on Tungsten, and C. Kirret dis- \ncussed Theory of Antiferroelectric Crystals.\n\nDuring the annual meeting, at a symposium \nof the Division of Electron Physics held Feb- \nruary 1, J. A. Hornbeck spoke on Distinguish- \ning Atomic from Molecular Ions: A Review of \nPositive-ion Mobilities In the Noble Gases. On \nthe morning of February 3 at a session on \nSemi-conductors; Photoconductivity; Lattice \nDefects, with W. Shockley presiding, K. G. \nMcKay presented a paper on Efficiency of \nAlpha-Bombardment Conductivity in Ger- \nmanium; E. J. RypER spoke on Mobility of \nElectrons in High Electric Fields; and Dr. \nShockley gave a talk on Mobility of Electrons \nin High Electric Fields: Theory. In the after- \nnoon of February 3, J. BARDEEN spoke on \nSupraconductivity and Lattice Vibrations. At \nthe same time, in another session, MATILDA \nGorrTz gave a paper entitled Heat Treatment \nof Iron-Silicon Alloys in a Magnetic Field. \nMiss Goertz was introduced by R. M. Bozortu. \nAlso on February 3, P. W. ANDERSON spoke on \nTheory of Paramagnetic Reasonance Line \nBreadths in Diluted Crystals.\n\n\u201cIn recognition of his initiative and interest \nin furthering communication progress in Japan\u201d \nduring the two years that he served as Director \nof the Research & Development Division of \nthe Civil Communications Section of General \nMacArthur's staff, Frank A. Polkinghorn of \nMilitary Electronics has been elected an Hon- \norary Member of the Institute of Electrical \nCommunication Engineers of Japan. Formal \nconferment will be made in the Spring.\n\nP. E. Mills shown here with his bass viol, is one \nof the seven Laboratories people who are mem- \nbers of the Metropolitan Bell Symphony Or- \nchestra. This orchestra, a full symphony or- \nganization, composed of Bell System people of \nall types of occupations, will offer its third \nCarnegie Hall concert Friday, March 30, at \n8:30 p.m. Guest soloist will be the distinguished \nconcert pianist, Amparo Iturbi. Tickets for the \nconcert, priced from $1.00 to $2.50, can be \nobtained by filling out the application blanks \nthat will be distributed to all members of the \nLaboratories.\n\nAmong the projects carried out under Mr. \nPolkinghorn\u2019s guidance was the organization of \nthe Electrical Communications Laboratory, the \nresearch and development body of the Ministry \nof Telecommunications, which was initiated by \nKing Gould, and carried forward by R. D. \nParker. This laboratory is under the leadership \nof Dr. Goro Yoshida who visited the Labora- \ntories last July. Another project was the organi- \nzation of electrical engineering professors for \nthe improvement of engineering education. \nThis project was responsible for the sending of \nfive professors to America to study. All of these \nmen were visitors to the Laboratories in 1950. \nA third project was the supervision of courses \nin industrial management given for the benefit \nof the communications industry. One of the \ntwo leaders of this course was Charles W. \nProtzman of the Western Electric Company at \nPoint Breeze.\n\nThe October issue of the Journal of the Insti- \ntute of Electrical Communication Engineers of\n\nJapan contains an article (in Japanese) on the \nImprovement of Telecommunications Engineer- \ning Education in Japan by Mr. Polkinghorn. \nThis is a translation of a talk which he gave to \nprofessors and students at several universities \nin furtherance of a program for obtaining more \neffective engineers. The translation was made \nby Dr. I. Koga, neg of communication \nengineering at Tokyo University and Tokyo \nInstitute of Technology, who was a visitor to \nthe Laboratories last Spring and who is well \nknown in the United States for his researches \non quartz crystals.\n\nThe August, 1950, issue of the Journal of \nthe Institute of Electrical Engineers of Japan \nalso contained an article on engineering eco- \nnomics by Mr. Polkinghorn. The September, \n1950, issue of The Bridge of Eta Kappa Nu \npublished an article on \u201cThe Occupation in the \nLand of the Rising Sun\u201d by Mr. Polkinghorn \ndescribing the purpose and policies of the \nOccupation as well as life in Japan since the \nend of the war.\n\nSince October 1, Ann Williams has been the \nregistered nurse in attendance at Whippany. \nShe is also trained as an X-ray Technician, and \nalternates with Gertrude Thomas of Murray \nHill in providing that service there \u2014 an ar- \nrangement that assures X-ray coverage. Mrs. \nWilliams is a graduate of the Robert Parker\n\nin Sayre, Pa. She lives in Upper Montclair, \nN. J., with her husband and four and a half \nyear-old son. Mr. Williams is a supplier of \nmedical equipment.\n\nC. W. Halligan has been appointed head of \na new department which has been set up for \nthe duration of a special project; he will re- \nport to W. H. Martin. Mr. Halligan will con- \ntinue supervision over his previous group as \nstation systems development engineer. K. E. \nGould is transferred from Carrier Systems \nEngineering to the new department, reporting \nto Mr. Halligan.\n\nCommercial and Staff Service formerly re- \nporting to H. J. Wallis in a \u201cjunior\u201d position \nhas been assigned to W. C. Pitman, Jr., in ad- \ndition to the project and local order service \ngroup and the plant and shop service groups \nwhich he is retaining. Mail, Central Corre- \nspondence and Catalog Files, formerly report- \ning to Mr. Pitman, have been assigned to Gen- \neral Service reporting to R. C. Carrigan.\n\nShirley Lawton, formerly in charge of Ap- \nparatus Files, has been made supervisor of \nPatent service at Murray Hill. Her place has \nbeen taken by Marguerite Johnston, super- \nvisor of Apparatus Records; and Miss Johnston \nin turn has been replaced by Lillian Schou.\n\nJ. F. Neill has replaced the late Edward \nAlenius as supervisor of the Illustrations group.\n\nMary Regan, a clerk in Apparatus Records, \nhas become a junior draftsman in the Appara- \ntus Drafting Department.\n\nThe Sixth Annual Exhibition sponsored by \nthe Laboratories\u2019 Arts and Crafts club will open \non May 22 in the Game Room (Section 1H) at \nWest Street, to run for three successive days. \nDeadline for entries is May 18.\n\nEvery member of the Laboratories is invited \nto submit entries and to attend the exhibit. \nWork submitted must have been made by the \nentrant, and not have been entered in previous \nexhibitions at West Street or Murray Hill. \nClasses for entry are sculpture and ceramics, \noil paintings, watercolors, pastels and mono- \nchromes; and handicrafts. The last group may \ninclude metal, leather, wood, stone, plastics, \nneedlework and china painting. Each exhibitor \nmay submit an unlimited number of works. \nNot eligible for showing are commercial art, \ntechnical illustrations, photographs, collections \nsuch as stamps or coins, and art, sculpture or\n\nIna wagon drawn by \u201cBabe\u201d\u2014a possible descendant \nof Paul Bunyan\u2019s famous blue ox\u2014four Bell System \nmen start for the woods to study pole production in \nthe southern pine forests. Seated, F. F. Farnsworth \nand guide. Standing, F. R. George, Purchasing Divi- \nion, and L. L. Bulter, Supplies Service Division, of \nWestern Electric, and G. Q. Lumsden of the Labora- \ntories-Photo courtesy of Mr. Lumsden\n\nceramic ware which is a facsimile of an exist- \ning work. Full rules covering the exhibition, \nand a list of prizes to be awarded will be an- \nnounced soon by the exhibition committee. In- \nformation can meanwhile be obtained by call- \ning Alice Loe, Ext. 560, or Mrs. C. H. Hamil- \nton, Ext. 1179.\n\nThe annual election of the Murray Hill Popu- \nlar Orchestra held January 4 resulted in the \nfollowing officers being elected: Executive \nChairman, U. A. Matson; Secretary-Treasurer, \nF. L. Crutchfield; Orchestra Director, L. J. \nSpeck; and Librarian, W. H. Kossman. Because \nof the increase in the activity of the orchestra, \na new officer\u2014Talent Director\u2014was created \nwhose duties include the recruiting of musical \ntalent at Murray Hill and the rehearsing of \nsuch talent for performance with the organiza- \ntion. H. G. Geetlein was elected to this office.\n\nRESPONSIBILITY for Western Electric sales \nmade directly to the United States Government \nof all wire communication products and of \nradio communication products of a type identi- \ncal or similar to those sold to Bell Telephone \nCompanies has been transferred from Western \nElectric\u2019s Radio Division to its Telephone Divi- \nsion. H. N. Willets, presently Commercial Re- \nlations Manager, Telephone Division, has been \ndesignated Manager, Government Communi- \ncation Sales. The post of Commercial Relations \nManager will be taken by S. Hubbard,\n\nNEED of the Bell System and other pole users \nfor the urban type joint-use poles\u2014those that \nsupport both telephone and power wires or \ncable\u2014has raised a question as to whether \nsouthern pine forests can continue indefinitely \nto supply sufficient quantities of these poles. \nThe western species such as western larch, \nDouglas fir and western red cedar are being \nused to supplement the short supply of southern \npine poles in those sizes under great demand. \nHowever, it takes at least 100 years for some \nof these western poles to grow, whereas the \nsouthern pine poles of these same sizes are full \ngrown at 30 to 35 years. Because southern pines \ngrow rapidly in those areas under strict forestry \nmanagement, particularly where forest fires are \ncontrolled, the production of pole timbers is \nbecoming an increasingly important activity. \nResults in certain areas, particularly in the long- \nleaf-slash pine region i the deep South, indi- \ncate that much can be accomplished by inten- \nsive forest management, Recently F. F. \nFaRNSworRTH and G. Q. LUMSDEN visited one \nsuch area, the Satilla Forest, belonging to the \nGeorgia Forest Products Company at Wood- \nbine, Georgia, to examine the results of the \nmethods successfully used there. On the same \ntrip they looked over treating plants in neigh- \nboring states.\n\nAT A CONFERENCE to discuss the work of the \nPhysical Electronics Research group held in \nthe Arnold Auditorium at Murray Hill January \n17, H. W. ALLIson gave a talk on The Adsorp- \ntion of Strontium on Tungsten and G. E, Moore \nspoke on The Thermionic Emission of Barium- \nStrontium Oxide without Metallic Support. A \nsecond conference held January 24 included a \ntalk by H. D. Hacstrum on The Ejection of \nElectrons by Positive Ion Impact\u2014Singly and \nDoubly Charged Helium Ions on Molybdenum \nand another by J. A. Becker on The Use of the \nField Emission Electron Microscope to Study \nthe Adsorption of Tungsten on Tungsten and \nBarium on Tungsten.\n\nMrs. Entrabartolo, after attending business \nand secretarial schools, began her Bell System \ncareer in 1929 when, as Mary Studney, she was \nemployed as a stenographer at Western Elec- \ntric in Kearny. After a few months she ac- \ncepted a position as a clerk at West Street. In \nOctober, 1949, she transferred to Murray Hill \nwhere from May of last year until the time of \nher death she worked as a secretary. Mrs. En- \ntrabartolo devoted much of her spare time to \ncharitable work. As one of her benevolent proj- \nects, she \u201cadopted,\u201d through the Commission \nfor Children\u2019s Relief, a small Polish boy who \nhad been seriously wounded early in World \nWar II. She sent money and numerous pack- \nages of clothing to help the lad, his brother\n\nKaiku Uljas Kalske, an assembler and wire- \nman, joined the Laboratories in September, \n1944. He was graduated from Theodore Roose- \nvelt High School, New York City, in 1938, and \nworked as a carpenter for various companies, \nincluding Bethlehem Steel Company in Hobo- \nken before he came to the Laboratories. At \nMurray Hill, he was on assignment to various \nengineering laboratories, and recently worked \non the wiring of components of the L-3 carrier \nsystem.\n\nK. K. Darrow gave a talk on Electricity in \nMetals and Semi-Conductors before the Pough- \nkeepsie (New York) Sections of the A.LE.E. \nand I.R.E. on January 9. Dr. Darrow has been \nreelected Secretary of the American Physical \nSociety, beginning his eleventh year in that \noffice of the Society.\n\nLast May, A. G. JENSEN gave a lecture on color \ntelevision before the Science Engineering Club \nat Kearny. Since that time, the a of \nthis lecture has been attested by the number \nof times he has repeated it. Capacity audiences \ngreeted him twice at West Street and at Mur- \nray Hill and once at Whippany during June, \n1950. A number of Sections of the I.R.E. and \nA.I.E.E. have also heard the talk.\n\nTHe DEcEMBER 9, 1950 issue of the English \npublication Nature contains a review of W. P. \nMason\u2019s new book Piezoelectric Crystals and \nTheir Application to Ultrasonics. Recognition \nis given to Dr. Mason\u2019s extensive contributions \nto the researches on piezoelectric crystals; the \nreviewer states that Dr. Mason has compiled a \n\u201clucidly written book.\u201d\n\nW. L. Mraz presented a paper on Automatic \nGain Control and Automatic Frequency Con- \ntrol as Feedback Problems at the U. S. Naval \nAir Station, Willow Grove, Pa.\n\nG. N. Vacca has been made a member of the \nWire and Cable Technical Committee of the \nIndustry Operations Bureau, National Produc- \ntion Authority. Among the committee\u2019s objec- \ntives is to recommend ways of allocating stra- \ntegic rubber for wire and cable use. H. PETERS \nworks on a similar committee for the hard \nrubber industry. Substitutes for scarce metals \nwere discussed by E. E. Scuumacuer, D. H. \nWenny and J. H. Scarr at Hawthorne.\n\nDURING A RECENT VISIT to Hawthorne, E. L. \nFIsHER, with the cooperation of the E of M and \nQuality Control engineers, established observa- \ntional standards for use in inspecting the new \ncylindrical protector blocks used in the recently \ndeveloped station protectors and protected dis- \ntribution cable terminals. In today\u2019s heavily \nloaded plant, occasionally distribution cable ter- \nminals are over filled with drop wires. While \nin Chicago, Mr. Fisher discussed with repre- \nsentatives of the Illinois Bell Telephone Com- \npany the design features of a supplementary \ncover which might be used to provide more \nwiring space in these congested locations.\n\nAT THE JANUARY meeting of the Deal-Holmdel \ncolloquium C. H. E:rmenporr described the \nnew L-3 carrier system for coaxial cables and \nits transmission objectives. This system will en- \nable the transmission of 1800 telephone chan- \nnels or alternatively 600 telephone channels \nand one television channel. Mr. Elmendorf \nstressed the need of planning the system and \ndesigning its equipment for ready conversion \nof existing systems from L-1 to L-3 operation. \nAt a February meeting A. M. SKELLETT, for- \nmerly of the Bell Laboratories, and now with \nthe National Union Radio Corporation, spoke \non Recent Developments in Dark Trace \nCathode Ray Tubes.\n\nAS DEMANDS for communication apparatus in- \ncrease, particularly apparatus to be used in \nairborne equipment, space and weight limita- \ntions become more and more important. Efforts \nare constantly being made to \u201cminiaturize\u201d not \nonly capacitors and resistors, but is extending \nto transformers and other power coils. Ter- \nminals for hermetically sealed power coils re- \nquired visits during January to several plastics \ncompanies by L. W. Kirxwoop together with \nH. R. Bosworth, C. J. Rix, R. A, Walker and \nW. Stolberg of the Western Electric Company. \nMr. Kirkwood also visited the Westinghouse \nElectric Corporation at Sharon, Pa., to review \nproduction of 400-cycle power coils.\n\nComMoN cause of relay contact failure is a \nmicroscopic deposit of dust which mechanically \nprevents closure. Studies are underway to de- \ntermine the types of dust which do the most \ndamage and how best to keep them away. T. F. \nEcan, H. J. Keerer, H. W. HERMANCE and \nG. T. KouMan visited the Lehigh and Home- \nstead Central offices in Pittsburgh where The \nBell Telephone Company of Pennsylvania is \ncooperating in analyses of atmospheric dust in \nrelation to contact performance.\n\nHIGHER RESISTANCE subscriber loops in step-by- \nstep areas will become possible in the near \nfuture as a result of Laboratories developments. \nBy taking advantage of more precise control of \nthe per cent break period of a new dial, and of \nthe use of full selective ringing sets employing \n313A vacuum tubes, it has been possible to re- \nadjust the step-by-step relays at the central of- \nfices to secure satisfactory operation on loops \nof more than 18 per cent higher resistance than \nhas been possible before. A. S. Kinc and F. B. \nLaMBERTY of the step-by-step equipment and \ncircuit groups, respectively, were recently in \nHawthorne to discuss with the Western Elec- \ntric Company the best ways of getting this im- \nprovement into the field promptly.\n\nSeated on the spring board during a lull in the swim- \nming club activities are (left to right) Doris Ketchow, \nThelma Gradwell, Elizabeth Hull, Priscella Westphal,\n\nThe Murray Hill swimmers do not let the \nweather interfere with their activities, as their \nprogram continues the year round. Having a \nlarge competent staff of Red Cross instructors, \nand with the objective of teaching greater \nsafety in the water, the club provides training \nin all phases of Red Cross water safety. During \nthe fall and winter season the program em- \nphasizes beginner and intermediate swimming \nskills. In the spring, life saving training is added \nto the program. Early in the summer the in- \ndoor program moves from the Plainfield, N. J., \nY.W.C.A. pool to the outdoor lake at the Schiff \nScout Reservation near Mendham, N. J. An in- \ntensive training course in boat and canoe han- \ndling skills rounds out the over-all program. \nOccasionally, public demonstrations have been \ngiven of swimming, life saving skills and the \nhandling of canoes.\n\nPracticing lifesaving skills are, at the top, W. C. \nWestphal being towed by R. W. Hull; in the center, \nW. C. Buckland is being \u201crescued\u201d by J. B. De Coste; \nin the foreground, H. Peters is towing Mildred Read.\n\nCatHopes that don\u2019t wear out are a steady \ngoal of Laboratories tube research, H. E. KERN \nand R. T, Lyncu reported some new slants in \ntheir paper Emission and Life of Planar Diodes \nas Related to the Reducing Agent Content of \nCathode Nickel before the Symposium on Elec- \ntron Emission, City College, New York.\n\nIN oRDER to reduce labor costs involved in \nunderground conduit construction, the stand- \nard lengths of the more commonly used conduit \nsections have been increased by six inches, This \nwill result in more duct footage per manufac- \nturing operation and a comparable decrease in \nthe number of joints required. A further change \nis the elimination of the glaze from the conduit.\n\nThere is a need for qualified American \nRed Cross First Aid Instructors to meet \nan anticipated demand for Out-Of-Hour \ninstruction in First Aid in connection with \nCivil Defense.\n\nPreparations are being made to con- \nduct a combined First Aid Advanced and \nInstructor Reactivating Course at West \nStreet and Murray Hill to qualify former \ninstructors whose certificates have lapsed.\n\nCurrently qualified instructors inter- \nested in teaching First Aid, and former \ninstructors interested in re-qualifying and \nteaching, may obtain information regard- \ning registration from L. E. Coon, Exten- \nsion 1713, at West Street.\n\nThese changes have introduced some new prob- \nlems in manufacturing routines. J. H. Gray \nvisited a supplier\u2019s plant in Indiana recently to \nconsult with the manufacturer and to examine \nsome of the conduit produced.\n\nSUBSCRIBERS DROP WIRES have long been made \nwith a composite conductor embodying a steel \ncore for strength and an outer copper sheath \nto provide conductance. Western Electric engi- \nneers have now developed a method of making \nthis conductor in which the steel core wire is \ncontinuously plated with copper to provide the \ndesired mechanical and electrical characteris- \ntics. After \u201celectroforming\u201d the conductor is \ninsulated and jacketed in the usual manner. \nRecently, J. B. Drxon, W. K. Oser and I. V. \nWILLiAMs visited Point Breeze to observe the \noperation of the pilot plant which is making \nthis conductor.\n\nGROWING INTEREST in telephone switching has \ncalled a number of our switching engineers to \nthe lecture platform in recent months. As has \nbeen mentioned in an earlier issue of the \nRecor, A. E. is giving a course on \nswitching at M.I.T., and he has recently had \ntwo guest lecturers from the Laboratories. On \nJanuary 9 E. G. ANprews talked on Switching \nTechniques in Digital Computers, and on Janu- \nary 15 and 16 W. KeisTer talked on Automatic \nTelephone Switching Systems. Mr. Keister also \naddressed an A.I.E.E. meeting held in Cincin- \nnati on January 25 on Fundamentals of Tele- \nphone Switching.\n\nBELL SysTEM MAINTENANCE PRACTICES are \nprepared by the Laboratories to guide the \nOperating Companies in the maintenance of \ntheir circuits and equipment. To make sure \nthat the material and form of presentation is \nbest suited to the needs of the local mainte- \nnance force, discussions are being scheduled \nwith key men of the Operating Companies to \nsecure comments and criticisms, Recently F., S. \nEntTz, Switching Systems Development engi- \nneer, CHARLES BREEN, Switching Maintenance \nengineer, and L. A. LEATHERMAN of the Power \nDevelopment group were in Indianapolis, to- \ngether with a representative of A T & T, to get \nfrom the Plant and Engineering Department \nrepresentatives of the Indiana Bell Telephone \nCompany, their comments on the current type \nof Practices.\n\nW. H. BENDERNAGEL and V. MANSUETTO \nwere recently in West Haven, Conn., to discuss \nmodifications in the fluorescent lighting system \nwhich is now being applied to step-by-step as \nwell as to crossbar and panel offices.\n\nA WEATHER ANNOUNCING SYSTEM has been \noperating in Cleveland for the past year on \nmore or less of a trial basis. E. Von DER LINDEN \nand R. O. L. Curry recently visited the in- \nstallation to arrange for final modifications.\n\nFoR THE NEW 302A power plant, designed for \nthe larger central offices, a greatly improved \nmotor-operated switch for cutting in and out \nemergency cells is being manufactured by \nAlbert & J. M. Anderson Company in Boston. \nH. T. LANGABEER and H. M. Spicer, who sug- \ngested the design, were recently in Boston to \ngo over the final drawings.\n\nF, P. Wicur of the trial installation group and \nR. W. Burns of the switching maintenance \ngroup were recently in Philadelphia to arrange \nwith the Bell Telephone Company of Pennsyl- \nvania for a trial installation of a line insulation \ntest frame in the Media No. 5 crossbar office. \nWhen set in operation by a maintenance force,\n\nthis frame automatically runs through all the \nlines in the office and within a few minutes has \nobtained a measurement of the resistance to \nground of each line.\n\nDuRING THE WEEK beginning January 19, J. W. \nMcRae gave several talks before I.R.E. and \nA.I.E.E. Sections in Detroit, Toronto, Cedar \nRapids and Des Moines on the subject of \nMicrowave Super Highways for Television and \nTelephony.\n\nA FIFTEEN-WEEK Seminar and Public Lecture \nSeries on Prevention of Corrosion, sponsored by \nthe Graduate school of Stevens Institute of \nTechnology began on February 7. This series \nof lectures is in co-operation with the National \nAssociation of Corrosion Engineers. One of \nthese lectures, to be given May 16, 1951, will \nbe by R. M. Burns, on the subject of Paints \nand Organic Coatings.\n\nH. S. BLack was recently appointed to the \nBoard of Editors of the I.R.E. and to the Ad- \nministrative Committee of the Board of Editors. \nJ. G. Kreer, JR. was appointed chairman of \nthe Task Group on Transducer Definitions.\n\nJ. R. Townsenp addressed the York, Pennsyl- \nvania Section of the American Society for \nMetals on Plastics and Metals in Engineering, \nand the Pittsburgh Section of the American \nChemical Society on the Role of the Tech- \nnologist in the Utilization of Materials in the \nModern World.\n\nIN KEEPING with the Bell System\u2019 Program to \nconserve strategic metals, it is planned to sub- \nstitute steel in place of aluminum for the \ngreater part in future production of antennas \nfor the TD-2 Microwave Radio Relay System.\n\nR. R. ANDRES, in company with C. B. Cottrell \nof Western Electric, visited the Tennessee Air- \ncraft Company in Nashville to discuss design \nand fabrication of these antennas.\n\nA. H. Lince visited the Western Electric Com- \npany at Winston-Salem to discuss tests on wave \nguide pressure windows for the TD-2 Radio \nRelay System. Mica, heretofore used for these \nwindows, is becoming difficult to obtain and \nglass-resin laminates are being investigated as \na substitute for mica.\n\nWirTH soME of the materials used in permanent \nmagnets becoming difficult to obtain, Labora- \ntories people, in cooperation with Western \nElectric, are investigating the use of substi- \ntutes. P. P. Crorri visited Hawthorne recently \nto discuss these developments. He was also at \nthe new Indianapolis plant to assist in solving \nproduction problems arising in the use of a \nnew alloy for the ring armature telephone \nreceiver.\n\nLABORATORIES ENGINEERS are continually striv- \ning to hold down apparatus costs, but without \nimpairing telephone service. In addition, there \nnow is the problem of finding substitutes for \nstrategic materials which are becoming more \ndifficult to obtain. A. C. Ekvaut visited Haver- \nhill recently to discuss cost saving items and \nsubstitute materials on coils and transformers. \nIN THE M1 CARRIER SYSTEM, a new unit has \nbeen developed by which several rural sub- \nscribers can be served by one carrier terminal. \nVoice-frequency lines are run to the subscriber \nand a conventional installation made at the \nsubscriber\u2019s location, A. J. Wier and D. T. \nOscoop visited the Western Electric Plant at \nWinston-Salem in connection with the manu- \nfacture of these voice-frequency extension units. \nTHE JANUARY MEETING of the Movie Division \nof the Murray Hill Photographic Club featured \na talk on movie planning and making by Mr. \nSidney Moritz, a well-known amateur movie \nmaker and a winner of awards by the Amateur\n\nWhen the Coil Shop moved from West Street \nto Murray Hill this Fall, its potentiometer test- \ning laboratory was removed to Whippany, ac- \ncompanied by Marya Motowski, standing, who \nhad worked on it during the war. She is shown \nwith Lorraine Prothero checking tests on an \nelectronic component.\n\nCinema League and Metropolitan Motion Pic- \nture Club. The Murray Hill Photographic Club \nmeets every other Monday at 12:15 P.M. in the \nArnold Auditorium.\n\nTo ACQUAINT a group of large-scale users of \nLong Lines facilities with the background of \nBell System Research and Development, mem- \nbers of the Communications Managers Associ- \nation of New York spent January 10 at the \nLaboratories. This association is made up of \nrepresentatives of large industrial users of com- \nmunication facilities. They were accompanied \nby representatives of the Commercial Organi- \nzation of the A T & T Long Lines Department. \nAfter an introductory talk by R. K. Honaman, \nvisits were made to several locations within \nthe West Street and Murray Hill buildings for\n\nA Seminar on Prevention of Corrosion spon- \nsored by the Graduate School of Stevens Insti- \ntute of Technology in cooperation with the \nNational Association of Corrosion Engineers, \nbegan on February 7. This series of 15 lectures \nincludes a talk by R. M. Burns entitled Paints \nand Organic Coatings scheduled for May 16.\n\nON A RECENT TRIP to the Pacific Coast, Joun G. \nFeRGusON visite. the Hewlett-Packard Com- \npany and the Stanford Research Institute to \ndiscuss new developments on measuring ap- \nparatus. While on the West Coast, he gave \nshort talks before transmission groups at the \nSan Francisco and Los Angeles Headquarters \nof the Pacific Company.\n\nTHE SEVENTH annual national technical con- \nference of the Society of Plastic Engineers was \nheld at the Hotel Statler in New York January \n18-20. C, J. Froscu was chairman of the Com- \nmittee on Prize Papers.\n\nStupENTS from the Bell Telephone Company \nof Pennsylvania, who have been receiving \ntraining in telephone transmission, have visited \nthe Murray Hill Laboratories to hear talks in \nthis field. The third group of these students \nvisited Murray Hill January 26, where they \nwere given a series of talks on transmission \nsubjects.\n\nJ. B. Fisk has been chosen as a member of the \n1951 Nominating Committee of the American \nPhysical Society.\n\nR. G. Treutinc conferred with a group in \nCleveland, including Professor W. M. Bald- \nwin of Case Institute of Technology, to arrange \nan educational symposium on residual stresses \nin metals for the 1951 Fall Meeting of the Na- \ntional Metals Congress. Mr. Treuting has been \ninvited to give the introductory lecture. Re- \nsidual stresses are important in apparatus parts \nas affecting dimensional stability and sometimes \nlife of the parts.\n\nIN THE FACING ADVERTISEMENT S, R. KING o> \nOutside Plant Development demonstrates the \ntechnique of loading a 152-pair, 22-gauge \nAlpeth cable. Loading coils are spaced at inter- \nvals of 6000 feet.\n\nMANY more wires can be crowded into a \ncable sheath when the wires are fine. But \nnormally, wires don\u2019t transmit as well when \nthey are fine and closely packed.\n\nBell engineers long ago learned to make \nwires do better work by loading them with \ninductance coils at regular intervals. The \ncoils improve transmission and let messages \ntravel farther. But originally the coils them- \nselves were large, heavy and expensive. The \ncases to hold them were cumbersome and \ncostly too.\n\nSo year after year Bell scientists squeezed \nthe size out of coils. To make magnetic cores\n\nof high permeability they developed Permal- \nloy. Tough but extra-thin insulation per- \nmitted more turns to a core.\n\nNew winding machines were developed by \nthe Western Electric Company. Coil size \nshrunk to one-fiftieth. Some\u2014like the one \nshown above\u2014can be mounted right in cables \nthemselves.\n\nThe 15,000,000 coils in the Bell System \ntoday mean thinner wires, more wires in a \ncable\u2014more economical service for you. They \ndemonstrate once more how Bell Telephone \nLaboratories work continually to add to your \ntelephone\u2019s value.", "title": "Bell Laboratories Record  1951-03: Vol 29 Iss 3", "trim_reasons": [], "year": 1951}
{"archive_ref": "sim_record-at-t-bell-laboratories_1956-06_34_6", "canonical_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1956-06_34_6", "char_count": 139351, "collection": "archive-org-bell-labs", "doc_id": 960, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc960", "record_count": 158, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1956-06_34_6", "split": "test", "text": "Mechanized Memory and Logic \u2014 What Electronics Can Do, J. H. Felker 201 \nUse of Transistor in New Military Telephone System, H.C. Fleming 4 207 : \nThe Speakerphone, W. Clemency 208 \nUltrasonic Delay Lines, J. b. May, Jr. 212 \nImproved Key Handle for PBX Switchboards, H. J. Smith 217 : \nArcing at Telephone Relay Contacts, I\u2019. Kisliuk 218 \nRepeaters Military Telephone, Jr. 227 \n4 The 1956 A. & Stockholders Meeting\n\nat Murray Hill Laboratory 234 \nTHE COVER: Mrs, R. A. MacAllister of the Outside Plant EDITORIAL BOARD \nDevelopment Department conducts sea water bleaching tests \non jute protective covering used in submarine cable. F. J. Singer, Chairman \nA. Hornbeck \nF. A. Korn \nE, T. Mottram \nThe BELL LABORATORIES RECORD is published monthly by Bell Telephone R. J. Nossaman \nLaboratories, Incorporated, 463 West Street, New York 14, N. Y., M. J. KELLY, W. E. Reichle \nPresident; M. B. LONG, Seeretcry and Treasurer. Subscriptions: $2.00 per A. H. White \nyear; Foreign, $2.60 per year, Checks should be made payable to Bell Labora- \ntories Record and addressed to the Circulation Manager. Printed in U. 8. A. EDITORIAL STAFF \n\u00a9 Bell Telephone Laboratories, Incorporated, 1956. A D. Tebo, Editor\n\nG. E. Schindler, Jr., Assistant Editor, Murray Hill \nR. C, Sanford, Assistant Editor\n\nwho are familiar only with present-day telephone equipment. Despite its \nseeming complexity, however, electronic switching promises a wholesale sim- \nplification of telephony. Laboratories engineers believe that in the future \nit will be a faster and more economical way to provide telephone service.\n\nIn many magazines and newspapers you may see \nclaims about the \u201cmagic\u201d of electronics. As tele- \nphere people, we are not interested in magic, but \nwe are interested in the capabilities of electronics, \nbecause electronics will penetrate into areas of the \ntelephone business that either have never been \nmechanized before or have been mechanized by \nother means. Perhaps we can sharpen our apprecia- \ntion of the part electronics can play by first examin- \ning the characteristics that cause people to associate \nwords like \u201cmagic\u201d with it.\n\nLooking for fundamentals, we may inquire, \n\u201cWhat puts the magic into modern electronics?\u201d \nThe answer comes in two parts: (1) the ability to \ndevelop tremendous but stable amplification, and \n(2) the ability to perform complex operations in \nmillionths of a second. The first of these is familiar \nto telephone people, for it is the basis of our mod- \nern telephone transmission plant. Electronic ampli-\n\nfiers are the muscles of our transmission system. \nThe second is being applied extensively in new \nswitching systems now under development.\n\nThe automatic tracking radars developed at Bell \nLaboratories and manufactured by Western Elec- \ntric are a splendid example of these two abilities \nof electronics. The Nike guided missile system, \nFigure 1, naturally excites a writer to the use of \nwords like \u201cmagic\u201d and \u201cgiant brain\u201d when he \nattempts to describe its performance, Without the \nmagic of almost unlimited amplification, the radar \nwould never be able to hear the radio echo which \nbounces off an airplane many miles away and \nspreads out into space, because only an extremely \nsmall fraction of the energy returns to the radar \nantenna. This small echo is amplified billions and \nbillions of times and is strengthened to where it \ncan drive motors which cause the radar antenna to \nfollow the enemy plane as it goes across the sky.\n\nAt the proper instant a missile is launched by the \nNike system, and here the system shows its brain \npower. By exploiting the ability of electronics to \nperform complex operations in microseconds, an \nelectronic computer guides the missile toward the \ntarget. Even if the target should attempt to evade \nthe missile, the very high speed computer takes\n\naccount of the evasion, and the missile proceeds \ninexorably to the interception.\n\nSince the telephone plant is not a radar system \nand since telephone customers do not need to be \nconnected together in microseconds, how shall the \nsecond magic-giving characteristic of electronics \nbe exploited in the telephone plant? The answer is, \n\u201cThrough time sharing\u201d. Now, time sharing is just \nanother way of describing the common-control \noperations that we introduced into the telephone \nplant with the dial system many years ago, With \ncommon control, one expensive piece of equipment \nperforms in sequence the same function for a num- \nber of different customers. We are limited in time- \nsharing electromechanical devices because of their \nlimited speed. When we go to electronics, however, \nwe can design complex equipment which completes \na function in a few millionths of a second and is \nthen ready to do something else. To appreciate the \nfull power of such electronic speeds, consider a \nmachine that can complete an operation every \nmillionth of a second. Assume also that a man may \ntake a minute to perform the same operation, At\n\nthese speeds the machine could perform as many \noperations in a minute as a man could by working \nnight and day for an entire lifetime. Such speed \nresults in great economic advantages for electronics.\n\nWe can develop some insight into the fundamen- \ntals of time sharing by analyzing the case of the \nchessmaster, who sometimes plays and wins as \nmany as 30 games of chess simultaneously. Now he \ndoesn't really make the individual moves of these \ngames simultaneously. He moves quickly from \nboard to board and takes on each opponent in turn. \nHe has memorized the positions of the pieces and \nthe previous history of each game, and all he has \nto do is detect the new move made by an opponent. \nHe rapidly decides upon his next move, executes \nthe move, and then goes to the next board. Note \nwhat this chessmaster has been doing, First of all, \nthe whole secret of his success is his ability to \nremember a tremendous amount of information. In \nfact, he carries around with him a memory of all \nthe games. The other ingredient in his success is \nthe ability to make decisions very rapidly. As a \nresult, his 30 opponents are, in effect, engaged con- \ntinuously. We can do this kind of time sharing in \ntelephony with modern electronics.\n\nImagine now a central office which has a built-in \n\u201cchessmaster\u201d, This electronic wizard plays a game \nwith each telephone customer who wants service. \nHis goal is to meet the service needs of each cus- \ntomer, The \u201cboards\u201d the \u201ctelephone-master\u201d uses \nare the lines of 10,000 or so telephones that come \ninto the office. The \u201ctelephone-master\u201d continu- \nously scans these lines to determine if service is \nrequested, He remembers what progress he has \nmade in carrying out the wishes of each of the \ncustomers, and he makes decisions on what his next \nmove with each customer should be. He does all \nthese things so rapidly that every telephone user \nthinks he has the undivided attention of the central \noffice equipment.\n\nThe chessmaster used both memory and logic, \nand it is primarily because memory and logic have \nbeen mechanized in modern electronic computers \nthat people compare these machines to the human \nbrain, Since mechanized memory and logic will \nbe at the center of future telephone switching \nsystems, it is important to understand their funda- \nmentals and applications.\n\nWe may call to mind one example of mechanized \nmemory from the folklore of the wild West. Every \ntime a gun-fighter killed a man he cut a notch on \nthe stock of his gun, Any time he wanted to refresh \nhis memory he could go to the \u201cmechanized mem-\n\nory\u201d on the gun stock and run his finger down it. \nThis may seem like a rather trivial example, but it \nillustrates some of the fundamental principles of \nmechanized memory systems. A modern counter- \npart of the notched gun stock is the punched card \nused in modern accounting machines. This card \nhas holes punched in it, and the positions of these \nholes represent information. A punched card can, \nfor example, remember a man\u2019s social security num- \nber or it can remember his rate of pay.\n\nTo understand the ideas behind electronic mem- \nory it is well to keep in mind the idea of a notch, or \na hole, or some other physical modification as the \nbasis for mechanized memory. Other key ideas will \nbe explained with reference to Table | in which \nare listed some of the words and phrases used in de- \nscribing electronic memory systems.\n\nA very basic word is Brr\u2014a_ contraction of \nBINARY bicrr. A Bir is the elementary unit of infor- \nmation. It is the yes or no answer to a question, The \npower of such answers is indicated by the familiar \ngame of 20 questions. As you know, by asking no\n\nmore than 20 questions it is generally possible to \nfind out which object out of all the objects in the \nworld a person has in mind. The following example \nshows how yes-no answers can be used to repre- \nsent numbers. Suppose someone has a number \nbetween zero and 31 in mind and we ask this person \nin succession the following questions and get the \nanswers indicated.\n\n\u201cIs the number greater than 15?\u201d \u201cYes.\u201d \u201cIs the \nnumber greater than 23?\u201d \u201cYes.\u201d \u201cIs the number \ngreater than 27?\u201d \u201cNo.\u201d \u201cIs the number greater than \n25?\u201d \u201cNo.\u201d \u201cIs the number greater than 24?\u201d \u201cYes,\u201d \nNow from this pattern of answers \u2014 Yes, Yes, No, \nNo. Yes \u2014 we know that the decimal number is 25. \nWe can write the number twenty-five in binary no- \ntation by saying that \u201cyes\u201d corresponds to the digit \none and a \u201cno\u201d corresponds to the digit zero. Thus \nthe binary number twenty-five is written 11001. By \nasking five questions we have determined the magni- \ntude of a number between zero and thirty-one. By \nasking ten questions we could have identified any\n\nnumber between zero and a thousand. By asking \n20 questions we could have identified any number \nbetween zero and a million. Thus any number be- \ntween zero and a million could be written as a 20- \ndigit binary number.\n\nWhy bother with binary representation of num- \nbers? The reason is that a binary digit is something \nlike a notch on a gun or a hole in a paper card. It \nis very definite. There is either a hole in a card or \nthere isn't. Similarly a binary digit is either a one \nor it isn\u2019t. Binary digits can be convenie ntly repre- \nsented by a variety of components. For example, \na relay is either operated or not operated, We can \nsay that the operate condition represents a one \nand the unoperated condition represents a zero. \nSimilarly, a transistor can be allowed to represent \na one when it conducts or a zero when it is not con- \nducting. A capacitor can represent a one when it \ncontains stored electrical energy, whereas an un- \ncharged capacitor can be used to represent a zero. \nA magnet can be magnetized in such a way that the \nnorth pole is at the top or it can be magnetized \nso that the north pole is at the bottom. Thus the \nnumber twenty-five might be represented as a row \nof magnets that are magnetized North-North-South- \nSouth-North.\n\nIn addition to the problem of providing the indi- \nvidual memory cells, the designer of a mechanized \nmemory system must provide many other features \nindicated by some of the words in Table 1, These \nfunctions will be explained with reference to Figure \n2, which shows a pigeonhole memory system. As- \nsume each pigeonhole has a card in it, Figure 2 (a), \nand that information or remembered \non each card. Going to the memory and selecting a \nparticular card is the access operation. The neap \noperation is simply the operation of looking at the \ncard, Clearly, the access operation must precede \nthe reap operation. Similarly, to wrrre into the \nmemory involves an access operation, because we\n\nA \u201cpigeonhole \nin holes and access to address D3; \nholes and binary digits 11001 registered in Row 4,\n\nmust first find the card we are interested in. To \ngain access to information systematically, it is cus- \ntomary to provide addresses which give instructions \non how to get to each pigeonhole. appress ps, for \nexample, says go out to the p column and then pro-\n\nceed down to the third row to information cell ps. \nWith the aid of such a pigeonhole memory system, \nwe have developed the fundamental ideas of mech- \nanizvd memory \u2014 the idea of appress, the problem \nof access, the concept of Binary picrrs and BIT OF \nINFORMATION, the warre operation and, finally, the \nREAD operation.\n\nAn electronic system, however, could not be \nbased on a man running around writing on cards, \nso a more suitable record is needed to put in \nthe pigeonholes. Earlier, when developing the ide: \nof bits of information, it was pointed out that a \ncapacitor could be used to store information, To \ndevelop a memory system, we can obtain lots of \n\u2018apacitors and put one in each pigeonhole, Figure \n2(b). Suppose you want to write the binary num- \nber 11001 into the memory. First of all, at what \naddress will the information be written? Assume \nthat it is to be written in Row 4, Proceed down to \nRow 4. The first digit is a one, so charge the first \ncapacitor, Since the second digit is also a one, \ncharge the second capacrror. Since the third and \nfourth digits are zeros, leave the third and fourth \ncapacirors uncharged, The fifth digit is one, so \ncharge the fifth caracrrorn, Other numbers might \nbe stored in other rows. To read the number at \nAddress 4 at a later time we would short circuit the \ncapacrrors in Row 4; whenever we got a spark we \nwould know that a one had been stored.\n\nThis leads to another basic idea. After the infor- \nmation has been read, where has the information \ngone? It has been destroyed because short circuiting \n\u2018apacitors Was DESTRUCTIVE READING process. \nMemory designers must design REGENERATION into \nmemory systems in which reading is destructive. \nIf you want to read information from the capacitor- \nmemory and still leave the information ready to be \nread again at a later time, you must put energy \nback into the capacitor if it was storing a one. This\n\nAn electronic system with capacitors in the \npigeonholes is not complete because someone has \nto walk around charging and discharging the \ncapacitors, To see how the access problem can be \nsolved with electronics, consider the ordinary tele- \nvision set. As you know, the picture is drawn by a \nspot of light that chases back and forth across the \nface of the picture tube. This spot of light is a kind \nof electronic pencil that is time-shared to draw the \nTV picture. Imagine the face of the picture tube \ndivided into pigeonholes as in Figure 3. The spot \nof light on the picture tube can be deflected by \nvoltages to any desired address in a millionth of a \nsecond, This rapid deflection of an electron beam \nprovides almost instantaneous access to any one of \na group of pigeonholes.\n\nHow can information be stored on the screen of \na cathode ray tube? Physicists have developed a \nnew kind of tube called a BARRIER-GRID MEMORY \nruBe. This tube is like a TV picture tube except \nthat it has literally thousands of tiny capacitors \ndeposited on the screen. The electron beam in this \ntube can charge or discharge capacitors. Thus, in \nthe memory tube, Figure 4, we have a means of \ngaining access not to just 25 but to over 16,000\n\nFig. 4 W. W. Baldwin welding leads to contacts \non glass envelope of barrier-grid memory tube.\n\npigeonholes. The memory tube then has a memory \ncapacity of 16,000 bits. To get an equivalent mem- \nory capacity with relays would require 16,000 \nrelays. A memory tube is a splendid example of \nthe compactness that can be achieved through \nelectronics. It has, in addition, the tremendous ad- \nvantage that reading and writing into the tube \nrequires only a few millionths of a second.\n\nFig. 5 \u2014 Magnetic-core memory: direction of magnetism \nof selected cores is reversed by currents in wires.\n\nmagnetic memory. This is another important method \nof storing large amounts of information in a small \nspace. It would be cumbersome if conventional \nmagnets were used, but physicists have discovered \nways of making up a ceramic-like magnetic material \n(ferrite) into little doughnuts called \u201ccores\u201d. In a \nmagnetic-core memory, Figure 5, electrical con- \nductors run through the holes of numbers of these \ndoughnuts mounted in an array. Currents passing \nthrough certain of the wires will magnetize or de- \nmagnetize selected cores.\n\nThe chief advantages of magnetic-core memories \nare that the ferrite cores can be mass produced \nvery cheaply, and that we can easily gain access to \nthe stored information through transistorized cir- \ncuits. At present, there is a competition between \nmagnetic-core memories and memory tubes. Many \nof our telephone engineers believe that the me mory \ntube is a little faster and a little cheaper than the \ncore memory, but they recognize that as further \nimprovements are made in the art the situation may \nbe likely to change.\n\nHaving seen how electronic memory systems will \nstore thousands and even millions of bits of in- \nformation, it is appropriate to consider mechanized \nlogic. We have had logie circuits in the telephone \nplant for years \u2014 they are the circuits that deduce \nthe logical results of the inputs to the circuits. As \nan example of logic, consider the plight of a fellow \nnamed Joe who belongs to a riding group. Every \nthird day it\u2019s Joe\u2019s turn to drive to work. But things \ndon't often work out that way because somebody \nis always having to take someone else's turn for one \nreason or another. Suppose that Joe comes home \nfrom work some evening and his wife says: \u201cHoney, \ndid you ask the fellows if one of them could drive \nfor you tomorrow?\u201d\n\nJoe says, \u201cSure I asked them. When I asked Pete \nif he\u2019d take my turn, he was just leaving for Phila- \ndelphia but he said he'd be glad to drive tomorrow \nif he didn't have to stay over in Philadelphia and \nhis wife didn\u2019t need their car. Anyway, Oscar will \nprobably be able to drive, because his wife has \nbeen staying home lately and he will drive her \ncar, if she doesn't go to work. He also said that \nsince his own car is due back from the garage \ntomorrow, he can drive it even if his wife does \nuse hers provided the garage gets the car back to \nhim. But if this cold of mine gets any worse Um \ngoing to stay home even if those guys have to walk \nto work and you'd certainly have the car if 1 am \ngoing to be home.\u201d\n\nTo simplify all this, Joe might draw a logical \nblock diagram like Figure 6 for his wife. The block \ndiagram has an output, a signal present, whenever \nJoe will drive, It has signal inputs on the left. We \nhave a signal present on the first lead, for ex- \nample, if Peter's wife needs their car. The \u201cor\u201d \ncircuit is called an \u201cor\u201d circuit because it will have \nan output whenever there is a signal on the top on \nbottom lead, The \u201cand\u201d circuit will have an output \nif there is a signal on its top ANp its bottom lead. \nThe third type of circuit, called an iNeuerror cir- \ncuit, will have an output if there is a signal on the \ntop lead, unless there is a signal on the bottom lead. \nNote that if Joe\u2019s cold is worse, the inuiBrron \nfunctions and regardless of the signal on the top \nlead there will be no output because Joe just isn't \ngoing to drive if his cold is worse.\n\nJoe could mechanize the circuit with either relays \nor transistors. With the use of transistors, the \ncircuit could determine its output in a millionth \nof a second, In our telephone plant of the future, \nthere will be an extensive use of transistors for the \nkind of logic functions that have in the past been\n\nperformed with relays. Since the transistor can be \nmade to operate in a millionth of a second, we can \ntime-share these circuits and, like the chessmaster, \nreach very fast decisions.\n\nSo far we have explored the characteristics that \nput the magic in electronics. You have seen how \nwe can mechanize functions which lead people to \nrefer to electronic machines as electronic brains. \nBy analogy with the chessmaster you saw how we \nmight, through time sharing, build a new type of \nelectronic switching plant. It is appropriate now to \nbe more specific about the use of mechanized mem- \nory and logic in building this plant.\n\nIn new electronic switching systems which Bell \nTelephone Laboratories engineers are now plan- \nning, mechanized memory and logic circuits are \ntime-shared to control a high-speed switching net- \nwork, Extensive memory will be used in such \nswitching systems. Information is stored in both \ntemporary and permanent memories. The kind of \ninformation which might be stored in the tem- \nporary memory can be determined by going back \nto the chessmaster or telephone-master analogy. \nThe telephone-master would store in the memory \ntube certain information regarding the lines coming \ninto the office. Suppose the office is a 10,000 line \noffice. The machine would look at each line and \nobserve whether or not the receiver was off the \nhook and would then store in the memory a binary \ndigit which answers the question, \u201cDoes the line \ncorresponding to this memory cell want service or \ndoesn't it?\u201d The machine would also store for each \nline the answer to the question, \u201cIs the line getting \nservice?\u201d If the answer is \u201cyes\u201d the machine would \nnot worry about the line. If the answer is \u201cno\u201d the \nmachine would then prepare to give the line service. \nIf dial pulses are coming in over the line the \nmachine will store in its memory the number of \ndial pulses received, Note that relay registers might\n\nJ. H. Feixer attended Washington University in St, Louis, Missouri, and \nreceived the B.S. degree in Electrical Engineering in 1941. Mr. Felker served \nin the U, S. Army from 1942 to 1945, first as a radar maintenance officer and \nlater as an Army publications officer. He joined the Laboratories in 1945, \nwhere, in the Military Systems Engineering Department, he was in charge of \nthe development of the Tradic transistor digital computer. In 1955, Mr. Felker \nwas appointed Director of the Special Systems Engineering II Department, \nwhere he is responsible for long range planning in data processing and trans- \nmission. He is a member of the Institute of Radio Engineers, and is past Chair- \nman of that organization's Professional Group on Electronic Computers.\n\nhave been provided to store the dial pulses. Instead, \nthe memory is provided wholesale in a big economi- \ncal memory system where the dial pulse information \nis stored along with the other information. \nBecause of these new system concepts and new \napparatus such as the transistor and electronic \nmemories, we can expect to see electronics as the \nbasis of telephone switching as well as transmission. \nBecause all functions are achieved by sequences of \nlogic and memory operations, we expect our equip- \nment to become more flexible. In the future the \nsame equipment used to set up telephone calls may\n\nAND | INHIBITOR 2) \nL OUTPUT: \nOSCAR'S CAR IS NOT | WHEN SIGNAL \nBACK FROM GARAGE | AND IS PRESENT\n\n\u201cJOE'S COLD 1S WORSE \nFig. 6 \u2014 Block diagram illustrating logic behind \nanswer to question, \u201cWill Joe drive?\u201d\n\nalso be used to perform billing and accounting func- \ntions, to prepare service orders, and to do circuit \nlayout engineering. At some stage in the future, we \nmay have a telephone plant in which the boundaries \nbetween separate functions are not very distinct \nbecause the same machinery is carrying out all of \nour operations. Many engineers are confident that \nthis is the way to simplify and to make cheaper \nthe cost of doing the telephone job.\n\nA point-contact transistor plays a unique role in \nthe new military telephone system recently designed \nfor the Signal Corps.* With this system, twelve \nspeech channels are transmitted simultaneously over \nfour conductors twisted into what is termed \u201cspiral- \nfour\u201d cable. To amplify the signals along the cable \nrun, electronic repeaters are spaced at intervals of \nabout 5% miles; many of these repeaters do not re- \nquire attendance by operating personnel.\n\nAt these unattended repeaters, however, installa- \ntion and maintenance tests are sometimes required, \nwhich means that occasionally someone must visit \nthe site of the repeater and talk to attendants at \nother points along the system. A portable test set in \ncombination with a field telephone is used for this \npurpose, The test set includes a transistor which is \nused in an oscillator circuit that permits the line- \nman performing the tests to signal attended re- \npeaters or terminals along the line.\n\nTo establish a talking circuit to these points, the \nlineman connects the portable field telephone to the \ntest set. This telephone includes a hand generator \nwhich, when rotated, produces an alternating vol-\n\ntage of low frequency (about 20 cycles per second ). \nSuch a signal normally suffices for signaling over \nthe field wire used for other types of military tele- \nphone connections. However, the circuit arrange- \nments of the new system do not permit the trans- \nmission of these low frequencies, and the receiving- \nsignaling circuits will not respond to them.\n\nThe test set is therefore provided with circuits \n(Figure 1) that rectify the 20 cps signal and apply \nthe resulting de potential to a point-contact tran- \nsistor, arranged to oscillate at 1,600 cps. This tone is \ntransmitted by the cable and is of the correct fre- \nquency to cause a response by the receiving-signal- \ning circuits at the attended points. The reception of \nthe signal is indicated by a lamp and buzzer.\n\nThe special characteristics of transistors are used \nto great advantage in this circuit. In contrast with \na circuit using electron tubes, the oscillator requires \nno warm-up time. Rotating the hand generator \ncrank less than two turns is sufficient to produce \nlamp and buzzer indications at the attended points. \nAlso, since the small amount of power necessary to \ninitiate the signal is easily obtained from the hand \ngenerator, no battery power is required, Further, \nbecause of the size advantage of transistors, the \nsignaling circuit adds little to the volume and weight \nof the complete test set. Without the transistor, \nsignaling could have been accomplished only with \nmuch bulkier and more complex circuits.\n\nTELEPHONE TRANSISTOR \nSET \nRECTIFIER \n\u201cai | 20% ANO \n\u2014_ LIMITER \nCIRCUITS -- \n1600, \nHAND TO SPIRAL- \nGENERATOR FOUR CABLE\n\nFig. 1 \u2014 Porta- \nble telephone \nand test set sig- \nnaling arrange- \nment, Transistor \ncircuit\u2019 supplies \n1,600 cps tone to \ncable,\n\nAs the telephone has become a more and more valuable part of normal \nAmerican life, the need for many new services has increased. Part of the \nLaboratories\u2019 responsibility, acting on the advice of the A.T.&T. Co., is to \nanticipate these needs and design the instruments and systems to fulfill them. \nOne good example of this work is the Speakerphone designed to provide \n\u201chands-free\u201d telephone service under most conditions,\n\nEver since the invention of the telephone, the \ntrend of design has been to increase the ease with \nwhich a customer can carry on his telephone con- \nversation, Following this trend a new telephone \nset, the Speakerphone, that provides hands-free \ntelephony has recently been introduced for Bell \nSystem use, The customer, using the Speakerphone, \ntalks into a transmitter about eighteen inches from \nhis mouth, and listens to a loudspeaker about the \nsame distance away. The customer can now readily \nwrite notes or refer to charts, reference books, and \ncorrespondence during a telephone call. He con- \nverses with the distant caller as if the latter were \nseated across the desk from him; this gives the user \nconsiderable freedom of movement. The Speaker- \nphone also provides for group participation. During \na call, a number of people seated closely around \na desk can talk freely \u2014the transmitter picks up \nthe conversation and the replies are heard from the \nloudspeaker. This hands-free set is also of consid- \nerable value to the physically handicapped.\n\nThe location of the acoustic instruments at a \ngreater distance away from the user as required \nfor a hands-free telephone set makes the perform- \nance of the Speakerphone more dependent on \nroom acoustics and room noise than that of a \nhandset telephone, The handset is used at such \nclose distances, about one inch for the transmitter \nand directly on the ear for the receiver, that the \nordinary environmental acoustic conditions have \nlittle influence. As the distance between the user \nand the telephone instruments is increased, how-\n\never, room reverberation ansl room noise become \nimportant. Hence the quieter and less reverberant \nthe room, the better the Speakerphone performance.\n\nA typical desk arrangement for the components \nof the Speakerphone is shown in the headpiece of \nthis article. The housing, similar in appearance to \nthe 500-type telephone set, contains the transmitter, \non and orF buttons, and the loudspeaker volume \ncontrol, in addition to the handset and other usual \ncomponents of the 500-type set. The loudspeaker \nis located in the small plastic housing shown in the \nupper center of the headpiece. The circuit of the \nSpeakerphone permits independent use of either \nthe handset or the Speakerphone.\n\nThe Speakerphone is easy to use. To place a \ncall, it is only necessary to momentarily press the \non button which lights up when the set is con- \nnected to the line. Then, when dial tone is heard \nfrom the loudspeaker, the number is dialed in the \nnormal manner. When answering a call, the on \nbutton is momentarily pressed, again causing it to \nlight up, and the conversation can start. The \ntransmitter picks up the voice of the user, and the \nother person is heard on the loudspeaker. The \nvolume of the incoming speech is adjusted by \nturning the small knob of the volume control. At \nthe end of the call, the orr button is pressed, dis- \nconnecting the set from the line.\n\nIf privacy is desired during a call with the Speak- \nerphone, or if some transmission difficulty is encoun- \ntered, lifting the handset automatically transfers \nthe call to the handset, and turns off the Speaker-\n\nphone. If the handset is being used, and it is \ndesired to transfer to the Speakerphone, the on \nbutton is depressed while the handset is being \nreturned to the cradle. The Speakerphone is de- \nsigned for a talking distance of ten to thirty inches. \nThe loudspeaker is placed about three feet from \nthe transmitter and so located that the normal \nseated position of the user is midway between the \ntwo instruments.\n\nIn addition to the two desk instruments shown \nin the headpiece, an apparatus box containing \namplifiers, power supply, and associated circuitry \nis part of the Speakerphone. The interior of this \napparatus box is shown in Figure 1. This box is \nmounted following usual Bell System practices.\n\nTwo amplifiers, each having approximately 50-db \nmaximum gain, are required, one for the transmitter \nand one for the loudspeaker. These two amplifiers \nare assembled on a_ printed circuit card which \nplugs into a socket in the apparatus box as shown \nin Figure 1. When these amplifiers require servicing, \nthe printed circuit amplifier card can be quickly \nreplaced. The power supply of the Speakerphone \nis energized from a 115-volt, 60-cps power outlet \nand consumes 8 watts when the Speakerphone is \nin use and 0.6 watt when it is in standby condi- \ntion. The handset of the Speakerphone operates \nindependently of the power supply and will per- \nform in its normal manner in case of power line \nfailure. The other components in the apparatus box \nare a relay, a hybrid coil, and input and output \ntransformers. The instruments shown in the head- \npiece and the apparatus box shown in Figure 1, \nwhich together constitute the Speakerphone, are \ncoded as the 595-type telephone set.\n\nThe basic circuit of the Speakerphone is shown \nin Figure 2. The hybrid coil, which is a form of \nWheatstone bridge, couples the two-way telephone \nline to the output of the one-way transmitting \namplifier and to the input of the one-way receiving \namplifier, and introduces a loss between them (the \noutput of the transmiting amplifier and the input of \nthe receiving amplifier). As indicated in Figure 2, \nthe unavoidable acoustical coupling through the \nair path between the loudspeaker and the trans- \nmitter forms a closed loop with the rest of the \ncircuit, This closed loop, when there is power gain \nin it, acts as an oscillator causing the loudspeaker \nto produce a sustained tone called \u201csinging\u201d. Be- \ncause of this, there is an upper limit on the total \npermissible gain which can be used in the trans- \nmitting and receiving amplifiers.\n\nimpedance and the balancing network impedance \ndetermines the loss introduced by the hybrid coil \nbetween the output of the transmitting amplifier \nand the input of the receiving amplifier. If a perfect \nbalance could be obtained at all frequencies, this \nloss would be infinite, and singing could not occur. \nThe impedance of telephone lines varies consider- \nably, however. It may be resistive, capacitive, or \ninductive depending on the length of loop to the \ncentral office, the length and type of trunk between \ncentral offices, and in the case of short connections,\n\nalso somewhat on the length of loop to the distant \nparty. This is especially true for a line on a PBX \nwhich, on internal calls, may have a short loop, but \non calls to an external telephone may have a long \nloop to a central office. There is, fortunately, a \nrough correlation between loop impedance and \nloop direct current. The balancing network is de- \nsigned to be adjusted by this loop direct current, \nwhich produces a voltage across the terminals of \nthe network. A varistor in the network varies its \nresistance in response to this voltage and regulates \nthe impedance of the network. The network is \ndesigned to produce a better balance on long loop \nconnections where more loudspeaker gain is re- \nquired and singing becomes a limiting factor. Be- \ncause both the transmitting and receiving amplifiers \nare in the singing loop, it is important that the \ngains in these amplifiers be properly apportioned. \nThe gain of the transmitting amplifier is adjusted \nat the factory so that the Speakerphone transmitting \nlevel at a talking distance of eighteen inches is\n\nsomewhat below the transmitting level of the 500 \nset with its transmitter at about one inch from the \nuser's lips. The loudspeaker volume control, ad- \njustable by the user, regulates the gain in the \nloudspeaker amplifier circuit. With the fixed gain \nin the transmitting circuit, the permissible gain in \nthe loudspeaker amplifier before singing occurs\n\nis determined by the balance of the hybrid coil \ncircuit, and the acoustical coupling between the \ntransmitter and the loudspeaker. As shown in Fig- \nure 3, an increase in distance between the trans- \nmitter and the loudspeaker allows more gain to be \nused in the loudspeaker circuit. However, beyond a \ncertain separation, further increase in the distance \nbetween the instruments moves the loudspeaker \nfurther from the listener and hence causes a de- \ncrease in the signal at his ears for the same loud- \nspeaker circuit gain. The optimum practical ar- \nrangement is to have the user seated approximately \nmidway between the instruments which are sepa- \nrated by about three feet on the desk top.\n\nFigure 3 also indicates that the acoustical prop- \nerties of the room influence the acoustical coupling \nbetween the transmitter and the loudspeaker, In \nreverberant rooms having very little sound absorp- \ntion, the permissible gain in the loudspeaker cir- \ncuit before singing occurs is less than that in a room \nwith sound absorption resulting from acoustical \ntreatment on the ceiling, carpet on the floors, and \nperhaps window drapes. The acoustical properties \nof the room also influence the transmitting quality \nof the Speakerphone.\n\nBecause the transmitter of the Speakerphone is \neighteen inches from the lips instead of one inch \nas with the handset transmitter, the ratio of direct \nspeech to reverberant speech, caused by reflections \nfrom the walls and other solid surfaces in the room,\n\nis less with the Speakerphone than with the handset. \nIn a room having little sound absorption, these room \nechoes are a disturbing influence to the listener at \nthe other end of the line. In direct person-to-person \nconversation, the binaural effect resulting from our \nhaving two ears reduces the effect of such room \nechoes. In rooms having good sound absorption, the \nechoes are reduced and Speakerphone transmission \nis relatively free of this reverberant quality. Further- \nmore, if the loudspeaker received signal is high, \nit is picked up by the transmitter and returned to \nthe distant talker as a delayed sidetone of his \nspeech. This effect, caused by the transmitter-to- \nloudspeaker acoustical coupling, is small in rooms \nhaving good acoustical properties. To minimize \nthis delayed sidetone at the called customer's tele- \nphone, the loudspeaker volume should not be \ngreater than is required for comfortable listening.\n\nA thermistor is used in the loudspeaker ampli- \nfier to limit the volume of the sound from the loud- \nspeaker, and to limit the voltage applied to the tele- \nphone line if the Speakerphone should sing because \nthe volume control is advanced too high. The ther- \nmistor, designed especially for this application, is \nused as a variable resistance shunt across the loud- \nspeaker transformer as shown in Figure 2. As the \nalternating voltage developed by the amplifier in- \ncreases, the thermistor decreases in resistance and\n\nFig. 3\u2014 Permitted loudspeaker gain versus sep- \naration between loudspeaker and transmitter,\n\nacts to limit the voltage. This action reduces the \nsinging tone from a disturbing how] to a soft audible \ntone, and also limits the voltage applied to the line \nto that of the transmitted speech level. In addition, \nby acting as a volume limiter on received loud- \nspeaker signals, the thermistor reduces the magni- \ntude of the delayed sidetone at the distant telephone.\n\nThe transmitting and receiving frequency char- \nacteristics of the Speakerphone are illustrated in \nFigure 4. By experiment these response curves have \nbeen found to be good compromises for speech \nquality, for the reduction of the effects of room \nechoes, and for the control of the singing char-\n\nacteristics of the set. The decrease of receiving \nresponse at the higher frequencies produces a more \npleasant quality for speech received from a con- \nnected handset telephone over short lines without \nproducing a serious degradation of articulation on \nlong lines. This is due to the fact that the higher \nfrequencies are augmented by the reflection of \nsound waves at the listener\u2019s head so that the re- \nsponse in terms of ear canal sound pressure actually \nrises at frequencies between 1,500 and 3,000 cycles \nper second. These reflections do not occur with the \nregular telephone receiver held close to the ear. - \nAmbient room noise influences the performance of \nthe Speakerphone to a greater extent than it does \nthe performance of the handset. This results from \nthe greater talking distance of the Speakerphone \ntransmitter, and the fact that the loudspeaker is\n\n-\u201cONE MICROBAR SOUND \nPRESSURE IN FREE FIELD J \n-15 \nTRANSMITTER <a-25 \nPOSITION ~~--\u2014> / \nus \n7-35 \nFig. Speakerphone \na 1000-OHM -40 \nfrequency characteris set | Vinw | ARTIFICIAL \ntics; top, transmitting, 4 \u2014 \nand bottom, receiving. \nng Vine = OBV 4a 0 \nMeasurement details are MAXIMUM VOLUME as \nCONTROL POSITION 58 \ngiven at the left. \n22 \nSOUND PRESSURE _ om \nMEASURED AT \nTHIS POSITION \nSe, $e -15 \nBo / \nw \n18\u201d LOUDSPEAKER -20 \nSu \nSu \n-2s \ni | -30 \n0 200 400 600 600 1000 2000 3000\n\nheard through an open air path. With a handset, \nthe receiver cap, when pressed against the ear, \nattenuates the ambient noise considerably. With \nthe Speakerphone, the ambient room noise enters \nthe ear with the received speech from the loud- \nspeaker and may interfere with reception.\n\nGreater appreciation of quietness, and the use of \nmodern sound absorptive materials in office and \nhome construction have made rooms having good \nacoustical properties quite prevalent. A recent field \nsurvey, in a business area, shows that a high per- \ncentage of Speakerphone installations are in private \noffices having good acoustical properties. In these \noffices, the Speakerphone generally provides satis- \nfactory telephone service with the benefits of hands- \nfree telephony and permits a small group to par- \nticipate in the conversation.\n\nW. F. CLemency joined the Research Department of the Western Electric \nCompany in 1923. He was engaged in studies of transmitter carbon and in the \ndevelopment of improved methods of manufacturing this carbon. Since the \nincorporation of the Laboratories, he has been a member of the Apparatus \nDevelopment Department and has worked on the development of telephone \ntransmitters, receivers, and other electro-acoustical devices. He is now engaged \nin the application and development of loudspeaking telephone sets. Mr. \nClemency received a degree in Electrical Engineering from Brooklyn Poly- \ntechnic Institute in 1934,\n\nAs modern electronic equipment is called upon to perform increasingly intri- \ncate operations, electrical signals must frequently be delayed within a circuit. \nSince electromagnetic waves travel at a speed of more than 186,000 miles \nper second, most methods of providing such delay require considerable \nspace or equipment, Recently, however, Bell Laboratories has been develop- \ning ultrasonic delay lines which occupy a relatively small space, and can \nprovide time intervals ranging from a fraction of a microsecond to several\n\nTime delay of electrical signals is useful in many \nBell System communication circuits, Such delay is \nneeded, for example, to produce series of pulses \naccurately spaced in time for various coding sys- \ntems. Another general field of application is in \nstorage devices wherein an electrical signal is \nstored for use at a later time, as in electronic \ncomputers,\n\nIt has been found that ultrasonic delay lines pro- \nvide a convenient method of providing these delays \nin a relatively small space. In such a device, time \ndelay is achieved by converting an electrical signal \ninto an acoustical wave, which is propagated along \na suitable path and then reconverted to an electrical \nsignal, These delays can range from a fraction of a \nmicrosecond to several thousand microseconds, and \nfor these, bandwidths of from three to six mega- \ncycles are easily obtainable. Since frequencies in \nthe megacycle range are usually employed, the \nname \u201cultrasonic\u201d is applied to these devices,\n\nUltrasonic delay lines were developed during \nWorld War II for use in radar systems, and H. J. \nMcSkimin of the Laboratories\u2019 Mathematical Re- \nsearch Department has made valuable contributions \nto their theory and design, Because of their inherent \naccuracy and wide-band characteristics, these ultra-\n\nsonic delay lines provided a convenient means of \nproducing series of accurately spaced pulses for \nradar range calibrations. Long delay lines of this \ntype were also essential to the development of \nmoving target indicators for radar systems, and \nshort lines are now being used in various coding \nsystems for assuring communication privacy.\n\nOther methods of delaying an electrical signal \ninvolve transmitting an electromagnetic wave in a \ncoaxial cable or an electrical network and hence \nrequire much more space. A video delay of 10 \nmicroseconds, for example, would require 6,480 \nfeet of ordinary coaxial-cable, 16 feet of specially \nconstructed \u2018delay cable,\u201d or an electrical network \noccupying 56 cubic inches. The same delay with \nequivalent bandwidth can be provided by an ultra- \nsonic delay line only 1.48 inches long. Acoustic \nwaves travel some 100,000 times slower than electro- \nmagnetic waves, and therein lies the space advan- \ntage over electrical delay lines.\n\nA simple delay line consisting of two transducers \nand a delay medium is illustrated in Figure 1(a). \nThe transducers in this device convert electrical \nsignals into mechanical stresses, or vice versa, by \nthe piezoelectric effect. The mechanical stress ap- \nplied to the delay medium travels through a pre-\n\nA single-ended delay line is a simple rod with \na transducer on one end. The signal travels the \nlength of the rod, is reflected, and travels back to \nthe transducer where it produces a delayed pulse. \nSome of the energy is reflected at the transducer, \nhowever, and again travels twice the length of the \nrod before producing a second output pulse. Thus, \nin response to a single input pulse, a series of \nequally spaced pulses are produced which can be \nused for time calibration purposes. In a double- \nended delay line with two transducers, as in Figure \nl(a), the signal travels the length of the rod only \nonce and produces only one output pulse de layed \nin time with respect to the input pulse. Any signals \narising from reflections at the transducers are, in \nthis case, unwanted responses, and efforts are made \nto suppress them. To conserve space for longer \ndelays, the transmission path can be folded so that \nthe ultrasonic wave hits one or more reflecting \nfaces before arriving at the output transducer. \nDelay lines of this kind are called multiple reflec- \ntion types.\n\nThe transducers are most efficient over a fre- \nquency band near their resonant frequency, and \nbandwidths of about 30 percent of the center fre- \nquency can be realized. To provide sufficient band- \nwidth for delaying video signals, this center fre- \nquency must be 10 me or higher. A carrier fre- \nquency, equal to the transducer resonant frequency, \nis modulated by the video signal to be delayed, This \nmodulated carrier signal is applied to the input trans-\n\nFig. 1 \u2014 Typical simple delay line configurations: \n(a) rod type; (b) single reflection type; (c) triple \nreflection type.\n\nducer of the delay line, and the signal appearing at \nthe delay line output can be demodulated to re- \ncover the original video information.\n\nThe transducers can be either crystalline materials \nhaving intrinsic piezoelectric properties such as \nquartz crystals, or non-crystalline materials made \nto have these properties by a polarization process, \nas in the case of barium titanate ceramic. For delay \nline applications, a solid medium is preferred to a \nliquid because of its mechanical stability and sim- \nplicity. The solid used almost exclusively because \nof its low attenuation, is vitreous silica, an optical \ngrade of fused quartz. Vitreous silica has a tem- \nperature coefficient of delay amounting to about \n100 parts per million per degree centigrade, and \nhence for some applications the delay line must be \ntemperature controlled,\n\nwaves or transverse waves. In longitudinal waves \nthe particle motion is along the direction of propa- \nThis is the type of wave transmitted by\n\nFig. 2 The three units at \nthe left correspond to the type illustrated in Fig. \nl(a), the unit being held corresponds to the type \nshown in Fig. 1(b), and the unit at the right cor-\n\ngases and liquids which gives rise to the sensation of \nsound, Solid materials, however, can transmit not \nonly longitudinal waves but also transverse waves \n\u2014waves which the particle motion is perpen- \ndicular to the direction of propagation. Transverse \nwaves are similar to light waves in that they can \nbe polarized; that is, they can be arranged so that \nall particle displacements lie in the same plane. In \ngeneral, when either a longitudinal or transverse \nwave is reflected at a solid-air interface, the re- \nflected wave has two components, one longitudinal \nand the other transverse \nof reflection.\n\neach with a different angle \nFor the special case of transverse \nwaves polarized perpendicular to the plane of\n\nFREQUENCY IN MEGACYCLES PER SECOND \nFig. 3\u2014 Frequency response of a 25-microsecond\n\ndelay line using barium titanate transducers ter- \nminated in a 62 ohm load. The spurious response \nin this unit is 20 db below the main response.\n\nFig. 5 \u2014 Configuration of a 990-microsecond delay \nline containing 30 reflection paths,\n\ncidence, the incident wave is totally retlected as a \ntransverse wave with the angle of reflection equal \nto the angle of incidence. Hence, this type wave is \nused for delay lines that require many reflections.\n\nclusively up to the present time, can be so oriented \nas to produce either longitudinal or transverse \nwaves. Delay lines constructed with quartz crystal \ntransducers are characterized by a relatively low \ncapacitance and a high characteristic impedance, \nhundreds to thousands of ohms. Because of the low \ncoupling of quartz crystals, such a delay line must \nbe terminated in a low impedance in order to obtain \na 30 per cent bandwidth, This results in a delay \nline of high loss, some 30 to 50 db. The recently \ndeveloped barium titanate ceramic transducer, how- \never, has a much higher coupling, which enables \nthe construction of delay lines of 30 per cent band- \nwidth with much lower loss. The loss ranges from \n6 to 20 db when the line is terminated near its \ncharacteristic impedance, which is a low value of \n30 to 100 ohms. The transducer capacitance is high, \n400 to 800 micromicrofarads. For practical reasons, \nthe barium titanate ceramic transducers cannot be \npolarized readily for generation of transverse waves, \nand their use is therefore limited to designs based on \nthe use of longitudinal waves.\n\nFrom the preceding discussion, it might appear \nthat it would be desirable to use polarized trans- \nverse waves in all designs because of their simple \nreflecting properties. However, for low-loss appli- \nations it is of great advantage to use barium titanate \nceramic transducers because of their high coupling, \nand for this reason longitudinal wave types are used \nwherever possible. The performance of designs \nusing longitudinal waves is good, provided the angle \nof reflection and the number of reflections are kept \nsmall. Hence, the designs illustrated in Figures 1(b) \nand l(c), while feasible for transverse waves, are \nalso good designs for longitudinal waves, and the \nlow-loss barium titanate transducers can be used \nin this application.\n\nSome typical short delay lines are shown in Fig- \nure 2. At the left are three variations of the single \nrod type with transducers on both ends as dia- \ngrammed in Figure l(a), One of the transducers \ncan be seen as a light disk soldered to the end-face \non each of these units. The solder is allowed to \nflow over the entire end-face to act as an absorber \nfor end-to-end reflections. The other transducers \nare mounted on the opposite end-faces not visible \nin Figure 2. The light areas on the other faces of \nthe second, third, and fourth units from the left are\n\nfired silver electrodes for the electrical ground con- \nnections. They are insulated from each other by the \nuncoated area appearing as a dark grey band passing \naround the rod, as in the second unit. The other \nelectrical connection is made to the electrode on \nthe exposed face of the transducer.\n\nSingle-rod type designs are useful for delays in \nthe 1 to 5 microsecond range. For the 5 to 25 micro- \nsecond range the single reflection pattern of Figure \nl(b) is used. In these units, the transducers are \nmounted on faces making equal angles with the far \nend so that an input at one transducer is reflected \nat the far end and arrives at the output end on a \npath perpendicular to the output transducer, An \nexample of this type is shown in Figure 2. Here \nboth transducers are visible on adjacent faces. The \ngrey strips in the light silver coated areas are again \nuncoated areas to isolate the grounding electrodes, \nFor still longer delays, up to about 100 microsec- \nonds, the three reflection pattern of Figure 1(c) is \nused. Here both transducers are located on the \nsame face, and the beam is reflected from a tilted \nface at the opposite end back to the center of the \ntransducer face. From there the path is symmetrical, \ngoing to another tilted face and then to the output \ntransducer. A delay line of this type is shown at \nthe right in Figure 2. On this unit, solder has been \nadded to the silvered electrode areas to act as an \nabsorber for energy striking the major faces.\n\nTypical delay-line performance is shown by the \nbandwidth curve in Figure 3 for units with delays \nup to about 25 microseconds, This design has a 4.8- \nmec bandwidth centered at 14.8 me with a mid-band \nloss of 5.5 db. Spurious responses, due to end-to-end\n\nMrs. M. S. Libby engaged in bonding a \ntransducer to an ultrasonic delay line.\n\nreflections or reflections from the sidewalls, are kept \nat least 20 db below the main response. For units \nof this design using longitudinal waves, the path \nlength is 0.235 inches per microsecond of delay, \nand for the rod type the cross section can be as \nsmall as % inch in diameter.\n\nFor delays greater than 100 microseconds, pat- \nterns containing many reflections are necessary to \nsave space. Properly polarized transverse waves are \nused in these applications because of their afore- \nmentioned reflecting properties, and because of \ntheir lower velocity. The path length is 0.148 inches \nper microsecond of delay, The \u201cdouble-wedge\u201d re- \nHection pattern of a 500 microsecond delay line is \nshown in Figure 4. In this design, a beam is re- \nflected back and forth between the two faces that \nform a wedge until it is finally reflected back on \nitself and returns to the input end. There, it strikes \na turning face that sends it into another wedge at \nright angles to the first. Part of the ultrasonic beam \nactually spreads beyond the geometric path out- \nlined by the shaded areas in Figure 4, This part of \nthe beam follows undesired paths through the delay \nline, some of which terminate at the output trans- \nducer. These give rise to unwanted signals occurring \nat times other than the designed delay value. How- \never, with 20-mc transducers, the geometry of the \ndouble wedge, having a relatively open pattern near\n\nthe transducers, produces good unwanted response \nperformance of 40 db below the main response for \nthe 500 microsecond size.\n\nA more compact design used for a 990-microsec- \nond delay line is shown in Figure 5, where 30 re- \nflections are used to fold twelve feet of path into \na polygon 5% inches in diameter. Starting at trans-\n\nline using quartz crystal transducers terminated in \na 500 ohm load. The unwanted spurious response \nin this unit is 34 db below the main response.\n\nducer X, the beam tollows the reflection pattern in \nnumerical order and finally arrives at the output \ntransducer on path 31. This is one of the most com- \npact designs consistent with good unwanted re- \nsponse, about 34 db in this case. A typical band- \nwidth curve for this design using quartz crystal \ntransducers is shown in Figure 8. Such a delay line \nwould be packaged in a hermetically sealed con- \ntainer, which would result in a total weight of 5 \npounds, 10 ounces.\n\nBy suitably scaling up the dimensions of the \npolygon, longer delays can be attained\u2014up to \n3,300 microseconds. This limit is established by \nthe largest available disk of vitreous silica, which is \n17 inches in diameter. Recently, some experimental \n2,000-microsecond units have been built.\u00b0 Although \nthe loss and bandwidth are similar to those in the \n990-microsecond design, the unwanted response is \nimproved to 38 db, due in part to the larger re- \nflecting surfaces.\n\n* The photograph at the head of this article shows the \nauthor testing an experimental 2,000 microsecond unit.\n\nMay, Jn., received the B.A. degree in physics from Wesleyan Uni- \nNavy, where he was engaged\n\nin the deve \u2018lopment of radio-controlle d aircraft and sidtiome radar, followed by \ntwo years at the Naval Research Laboratory, Boston Field Station, in the devel- \nopment of ultrasonic delay lines, He received the M.A. degree in physics from \nTufts College in 1949 and in 1952 received his Ph.D. degree in physics from\n\nYale University, publishing a thesis on nuclear physics research with the Yale \ncyclotron, Mr, May became a member of Bell Telephone Laboratories in Octo- \nber r, 1952, where he has been concerned with fundamental studies of ultrasonic \ndelay lines, He is a member of Sigma Xi, Sigma Pi Sigma, the Institute of Radio\n\nDr. Claude E. Shannon, research mathematician \nin charge of Communication Theory, Discrete Sys- \ntems and Special Research at the Laboratories, was \nelected a member of the National Academy of Sci- \nences at its 93rd annual meeting held in W ashing- \nton, D. C., on April 24. Dr. Shannon, serving now \nas a visiting professor of electrical communications \nat Massachusetts Institute of Technology, has been\n\nThe National Academy of Sciences is a private, \nnon-profit organization that is restricted to 600 mem- \nbers. Other members from Bell Laboratories in- \nclude President M. J. Kelly, Executive Vice Presi- \ndent J. B. Fisk, and J. R. Pierce, Director of Re- \nsearch, Electrical Communications.\n\nSome years ago, the 555 PBX switchboard\u00ae was \ndeveloped to fill the particular requirements of a \nnumber of telephone customers. The design was \npointed toward flexibility of equipment, ease of \nmaintenance, adequate range for trunk and station \nfacilities, and low power consumption. One feature \nof the switchboard was the replacement of lever- \naction keys by push buttons for ringing and less con- \nventional \u201cpush or turn\u201d keys for the usual PBX \nfunctions. All cords, lamps, and ringing push but- \ntons are mounted on a sloping shelf, leaving the op- \nerator\u2019s desk free for writing or other work. The \n\u201cpush or turn\u201d answering keys are mounted on the \nvertical face of the board, below the sloping shelf.\n\nTo answer a call, the operator turns the key \nhandle to the right where a detent holds it in posi- \ntion until it is restored to the normal vertical posi- \ntion. For a \u201cthrough-dial\u201d or night connection, the \nkey is pushed in toward the face of the board where \nit is also held by a detent. Locks are provided so \nthat when the handle is turned, it cannot be pushed \nin, and when pushed in, it cannot be turned.\n\nExperience indicated that some modifications of \nthe key handle would add to the operator's comfort \nand to her efficiency. Suggestions were made to (1) \nremove the slight projection at the end of the handle \nsince this projection partially masks the white ref- \nerence line, making it difficult to see, and also offers \nsome discomfort to the fingers when operating the \nkey; (2) enlarge the \u201cpushing\u201d area and increase \nthe radius around the edge; and (3) provide a suit- \nable indicator for the \u201cin\u201d and \u201cout\u201d positions, since \nthe black handle against the black key-shelf makes \nthe position difficult to determine.\n\nThese comments were taken into consideration \nwhen the new handle was designed. Figure 1 shows \nthe original handle along with the new version.\n\nThe projection, indicated on the original handle by \nthe pencil, has been removed and all edges are \nrounded. The lever part of the handle is longer, \noffering a better grip, and blends into the pushing \narea so as to provide a large smooth surface, The \nwhite reference line is longer and wider for in- \ncreased visibility, and is no longer masked by the \nprojection. The pushing area is larger, tapering \ntoward the back to provide a grip for pulling when \nthe pushed-in key is restored to normal.\n\nAlternate black-and-white stripes at the back of \nthe handle serve as a \u201cflag\u201d to indicate the position \nof the key. When the lines are visible in front of the \nkey-shelf, the handle is free to turn. When it is \npushed in for \u201cthrough-dial\u201d service, the lines dis- \nappear, indicating that it cannot be turned. These \nhandles are being regularly supplied on new switch- \nboards and are available separately for replacement \nin existing switchboards.\n\nNot only is Bell Telephone Laboratories concerned with developing new sys- \ntems and devices for improving telephone service, but it is also constantly \nendeavoring to improve components in the existing equipment, One im- \nportant aspect of this work deals with fundamental research aimed at ex- \ntending the useful lifetimes of relay contacts. The many millions of such \ncontacts used in telephone central offices throughout the country make them \na significant item in the maintenance costs of the Bell System.\n\nTelephone switching systems depend upon the \nreliable operation of a large number of relay con- \ntacts. Many of these carry relatively large currents \n~about one-half ampere \u2014 and operate relatively \nfrequently. They are subject to an erosion of the \ncontact metal, which severely limits their useful \nlifetimes, This erosion is primarily due to electrical \narcing, and hence satisfactory contact lifetimes can- \nnot be achieved unless the energy dissipated in ares \non make and break is limited to a small fraction \n(about one ten-thousandth) of the energy stored \nin the inductance of the usual relay loads.\n\nFrom work on electrical breakdown in gases it \nwas believed about ten years ago that there can be \nno electrical breakdown, and therefore no arc, unless \nthe potential difference between electrodes is above \na certain \u201cminimum sparking potential\u201d which is \nroughly 300 volts for air. Very early work per- \nformed about 1901 gave evidence to the contrary, \nbut this had apparently not been incorporated into \nrecent literature. Thus, until a few years ago, one \nexpected that there might be arcs at the breaking \nof contacts in an inductive circuit, but that there \ncould be no discharge at all on closure with po- \ntential differences less than 300 volts. Neverthe- \nless, experimenters in the field of contact erosion \nwere aware that arcing occasionally takes place on\n\nthe closure of contacts when the potential difference \nbetween them is only 48 volts. This was explained \nby assuming that small projections on the surface \ntouch first and that the resulting current evaporates \nthe points of contact with explosive violence, initi- \nating an arc in the hot vapor.\n\nOscilloscopic studies at Bell Telephone Labora- \ntories revealed no evidence to indicate that con- \ntacts touch preceding arcs on closure, and one is \nled to conclude, in agreement with the very early \nwork, that arcs occur before there is any metallic \ncontact. Furthermore, consideration of the proc- \nesses expected to take place when contacts close \nat voltages considerably above the ionization po- \ntential of the atoms of contact metal (around \n10 volts), but below the minimum sparking poten- \ntial for air, indicates that a new type of breakdown \nprocess will take place. Even for a very low poten- \ntial difference, the electric field between approach- \ning contacts necessarily becomes extremely high. \nWhen the field at the negative electrode, or cathode, \nbecomes great enough, in the neighborhood of ten \nmillion volts per centimeter, both theory and experi- \nment show that appreciable current must flow from \nthe cathode. This \u201cfield emission\u201d current is not \nin itself a \u201cbreakdown,\u201d but is the first step in the \nbreakdown process.\n\nField emission current is extremely sensitive to \nsmall changes in the field, so that small regions of \nthe cathode surface, where the field is somewhat \nabove average, will deliver nearly all the current. \nBecause no metal surface can be perfectly smooth, \nprojections on the cathode provide such regions of \nincreased field. The flow of the field emission cur- \nrent, through the resistance offered by the small \npoints, heats the projections from which the current \nis drawn. A slightly larger region of the positive \nelectrode, or anode, is heated also, in this case by \nelectron bombardment.* When the heated region \u2014 \neither of the anode or of the cathode \u2014 reaches the \nboiling point of the contact metal, the density of \natoms and molecules in the gap is increased, so that \neach electron now experiences a greater number of \ncollisions in crossing the gap. If the electron energy \nis above the minimum ionization energy of the \natoms in the gap, ionization will take place, creating \nions and new electrons. These electrons are quickly \ndrawn out of the gap, leaving behind the relatively \nmassive and slow-moving positive ions. The pres- \nence of these ions increases the electric field at the \ncathode surface, which results in greater field emis- \nsion current. The increase in electron current in- \ncreases the amount of ionization, and the process \nrapidly builds up to a \u201cbreakdown\u201d; that is, the \ncurrent increases until it is limited only by external \ncircuit elements. From an analysis of this process \nit has been shown that a number of ions equal to\n\n* The electrons do not spread a great deal in crossing the \ncontact gap, because in the range of voltage under consid- \neration the distance at which appreciable field emission \nappears is less than 10\u00b0* cm. Each electron experiences only \none or two collisions with gas molecules in this distance.\n\n(1) (2) \nAnode Cathode \nAres Ares \n1. Anode Appearance Hole Nothing \n(Fig, 2, left) (or shallow hole) \n2. Cathode Appearance Roughened Scratches\n\n3. Occurrence: \nvoltage above 400 Never Always \nvoltage below 300 Sometimes Sometimes \n4. Weld Frequently Never \n5. Are voltage about 11 about 16 \nvolts volts\n\nonly a few percent of the number of field emission \nelectrons is more than sufficient to cause break- \ndown, and that the elapsed time from the first \nmeasurable currents to the completion of the break- \ndown is too short to be detected by oscilloscopic \nobservations.\n\nDetailed experimental confirmation of the break- \ndown process just outlined is difficult in air, because \none does not expect this process to be effective \nabove the minimum sparking potential (approxi- \nmately 300 volts), and at lower voltages the times,\n\nFig. 1 \u2014 (a) Oscillogram of a sustained are on \nbreak followed by an open circuit, Starting from \nthe left, the contacts are closed at first and the \ntrace is a horizontal line at zero voltage. The con- \ntacts then open into an are at 15 volts which lasts \nfor about 300 microseconds, When the are goes out, \nthe voltage jumps to a new value, then charges up \nmore slowly because of the current from the ex- \nternal source. (b) \u201cShowering\u201d on break followed \nby a glow discharge. After the portion of the trace \nat zero voltage, there is a region of rapid charge-ups \nand breakdowns lasting for about 30 microseconds. \nThis is followed by a steady glow discharge at 300 \nvolts lasting for about 70 microseconds, When the \nglow discharge goes out, the voltage begins to fall \nas the capacitance discharges through the power \nsource, which supplies only 50 volts.\n\nspacings, and pre-breakdown currents are extremely \nsmall, Furthermore, the effects of dirt and oxide \nfilms are difficult to estimate. Each step in the pro- \nposed process, however, is also operative in a \nvacuum, Experiments have been carried out in high \nvacuum, and the evidence gathered confirms the \ntheory of breakdown caused by field emission en- \nhanced by ionic space charge.\n\nBreakdowns of this type may take place either on \nclosure or break of telephone contacts. Depending \nupon the circuit and the condition of the metal sur- \nfaces the arcs which follow breakdown may last \nfor many microseconds and be easily observable on \nan oscilloscope, or they may be so brief as to be \nunresolvable, leading to many rapid cycles of charg- \ning up and breaking down of local capacitance. This \nlatter condition is called \u201cshowering.\u201d It is during \nthese periods of either continuous arcing or show- \nering that most of the damage to relay contacts \ntakes place. Typical oscilloscope traces of the volt- \nage across electrical contacts in these two forms of \ndischarge are reproduced in Figure 1.\n\nKnowing th\u00e9 conditions which lead to arcs on \nmake and break is a promising start in the predic- \ntion of the useful life of relay contacts. However, \nattempts to understand quantitatively the amount \nof metal transferred by arcing at contacts have fre- \nquently been frustrated in the past by inconsistent \nand contradictory experimental results. A consid- \nerable step toward understanding erosion was made \nby discovering, in experiments performed on palla-\n\nTypical cathode mark due to a single \n\u201ccathode\u201d arc. There is frequently no mark at all\n\ndium contacts, that there are two distinct types of \narcs. One type, called the \u201canode\u201d arc, leaves a \nmelted crater on the anode surface and a roughened \ncathode spot. The other, called a \u201ccathode\u201d arc, \nleaves many tiny melted spots on the cathode, usu- \nally along scratch lines when the cathode is a pol- \nished surface, and little or no mark on the anode. \nPhotomicrographs of the damage done to anode \nand cathode surfaces in single typical arcs of the \ntwo types are shown in Figures 2-4. One might \nexpect, from examination of these photographs, that \nan anode arc transfers metal from anode to cathode, \nand a cathode are from cathode to anode. These\n\n- (Left) Photomicrograph of a typical anode pit due to a single \u201canode\u201d arc. Magnification\n\n2000 diameters, (Right) typical cathode mark due toa single \u201canode\u201d arc. Magnification 2000 diameters.\n\nFig. 4\u2014 Electron micrograph of a portion of the \ncathode mark due to a single \u201ccathode\u201d arc. Magni- \nfication 47,000 diameters,\n\nexpectations have been confirmed both by deter- \nmining the transfer occurring in single arcs of each \ntype by radioactive tracer methods, and by weigh- \ning on a microbalance after many closures under \ncircumstances known to produce predominantly one \nor the other type of are.\n\nTo understand contact erosion and to predict con- \ntact lifetimes, it is thus necessary to understand the \nfactors affecting the occurrence of the two types of \narc. In general, for striking voltages above 400 \nvolts, only cathode arcs occur, and the probability \nof occurrence of anode ares increases as the voltage \nis lowered, Cleanliness of the cathode surface favors \nFig. 5 \u2014 Oscillograms of arcs on closure discharg- \ning 200 feet of 50 ohm cable charged to 200 volts. \nTrace length is 0.8 microseconds, (a) Are followed \nby a \u201cweld\u201d. (b) Are followed by an open circuit. \nThere is also a trace at zero voltage.\n\nIt is now believed that, on clean surfaces, a \ncathode are follows a breakdown initiated by the \nevaporation of a small projection on the cathode \nby resistive heating. An anode are occurs when the \ncathode projection is so large that the anode reaches \nthe boiling point before the cathode projection. In \nthe cathode arc, metal vapor necessary for the are\n\nPhotomicrograph of an anode pit due to \ndischarging a cable where the are has gone out and \nhas been re-ignited. This is to be compared with \nthe left part of Figure 2, where the are did not \nstrike a second time.\n\nis supplied by the cathode, but in the anode arc it \nis supplied by both electrodes with the larger frac- \ntion coming from the anode. Some of the distin- \nguishing characteristics of the two types of arcs are \nshown in Table 1.\n\nA further complication in the prediction of con- \ntact erosion is due to the occurrence of \u201cwelds.\u201d \nAnode arcs are frequently terminated before the \ncircuit is discharged, by short-circuiting the con- \ntacts by molten metal from the anode spot. Oscil- \nloscope traces of arcs which do and do not termi-\n\nnate in welds are shown in Figure 5. When welding \noccurs, the erosion depends upon the time of weld- \ning as well as upon the relative numbers of the two \ntypes of are,\n\nThe problem of time of welding should properly \nbe considered together with other phenomena that \noccur at the termination of the arc. In general, \nneither type of arc can persist when the current \nfalls to indefinitely small values. Observed \u201cmini- \nmum are currents\u201d are of the order of 0.3 to 1.0 \nampere for both types of are on clean metals. The \nreasons for this minimum are poorly understood for \ncathode ares, but for anode arcs it appears that the \nanode spot grows with time, and thus the power \nnecessary to maintain the surface at the boiling \npoint also increases. When the current is no longer \nsufficient to supply this power, the arc goes out. \nSince the size of the anode spot depends on the \ncourse of the current up to the time of extinction, \nthe minimum arc current is a function of the his- \ntory of the arc up to that point. When an arc is \nextinguished, the electric field at the anode, which \nwas very small during the arc, is suddenly increased \nand the pool of molten metal is pulled toward the \ncathode. Thus, welds occur most frequently at the\n\nend of the growth of one of the anode spots when \nthe are is extinguished. If the circuit is still capable \nof delivering considerable current, however, the \nweld may be blown up and the are reignited. The \nnature of an anode pit when a second arc strikes \nin discharging a cable is shown in Figure 6. If the \npool of metal in the anode spot is either too small \nor cools too quickly, the anode metal may not be \nsufficient to bridge the gap, and the anode arc may \nend in an open circuit, just as do all cathode ares. \nAlthough the results of many tests are under- \nstood as described, it is not possible to predict con- \ntact life with great reliability. In view of the ob- \nserved dependence of the erosion upon atmospheric \nconditions, upon the previous history of the con- \ntacts, and upon small changes in circuitry, it is dif- \nficult to believe that operating conditions can ever \nbe sufficiently well specified in a reasonably concise \nmanner to make such predictions possible. Never- \ntheless, certain conditions favorable to long lifetimes \nare understood. They involve, among other things, \navoidance, where possible, of the injurious organic \nvapors, prevention of the discharge of any consid- \nerable capacitance on closure, and keeping the \nenergy stored in the load inductance to a minimum.\n\nPaut Kisiiuk received a B.S. degree in chemistry from Queens College, \nNew York City, in 1943. After three years in the armed forces, during which \nhe was engaged in maintenance of repeater and carrier telephone equipment \nfor the Signal Corps overseas, he resumed his education at Columbia Univer- \nsity. He received an M.A. degree in physics in 1947 and a Ph.D. degree, also \nin physics, in 1952. Mr. Kisliuk joined the staff of Bell Telephone Laboratories \nin 1952, and is a member of the Physical Research Department, where he \nhas been engaged in research in contact physics. He is a member of the \nAmerican Physical Society and of Sigma Xi.\n\nA. G. Jensen, Laboratories Director of Television \nResearch, was recently elected a Fellow of the \nBritish Television Society in recognition of his out- \nstanding contributions to the development of tele- \nvision. The Television Society, founded in 1927, \nis \u201cthe first society in the world for the furtherance \nof research in television and allied problems.\u201d\n\nMr. Jensen has received a number of other honors \nfor his work in television research including the \nDavid Sarnoff Gold Medal from the Society of \nMotion Picture and Television Engineers, and the \nG. A. Hagemann Gold Medal for Industrial Re- \nsearch from the Royal Technical College in Copen- \nhagen, Denmark.\n\nThe Centralized Automatic Message Accounting system serves many local\n\ntelephone offices by recording information on calls outside the customer's \nlocal dialing area. In a CAMA installation, trunks from the local offices are\n\nperiodically tested by a circuit which helps insure service that is free of inter- \nruption or other difficulty. Operation is automatic and does not require \nassistance from maintenance personnel in the local offices.\n\nThe Centralized Automatic Message Accounting \nsystem has required the development of several \nnew circuits and the adaptation of many earlier \nunits.* This equipment is installed in a centralized \nlocation (a crossbar tandem office), so that it can \nserve a large number of surrounding local offices. \nWith the CAMA feature, customers can dial many \nextended-area and long-distance calls even though \nthe cost of installing separate Automatic Message \nAccounting equipment in each local office is pro- \nhibitive. The service is much like the previously \ndeveloped AMA operation, except that after the \ncustomer has dialed his number, a special CAMA \noperator asks for the originating telephone number. \nOtherwise, the recording of information relative \nto the telephone call is entirely automatic.\n\nBetween the crossbar tandem office and the sur- \nrounding local offices there are a large number of \npairs of wires which are used for the transmission \nof telephone conversations, and which are usually \nreferred to as \u201ctrunks.\u201d More exactly, however, a\n\n\u201ctrunk\u201d consists of the pair of wires plus the relay \ncircuits at each end. At the tandem office, such a \ntrunk relay circuit must perform many functions: \nit must work with other parts of the crossbar switch- \ning system to set up the connection to the called \ntelephone office, and it must later take charge of \nthe connection and maintain it until appropriate \nIn the CAMA \nsystem it must also perform several tasks in asso- \nciation with the AMA recording equipment.\n\nDescribed here is an automatic test circuit that \ntests all the operations of the trunk relay equip- \nment connected to a cable pair on a call coming \nin from a local office. The wide variety of tests \nperformed, and the rapidity of operation, make \nthis test circuit a very valuable part of the com- \nplete CAMA installation. As will be seen later, \nmuch of this value derives from the fact that, in the \ntandem office, maintenance personnel do not need \nto ask for assistance of the various local offices in \narranging for or conducting the tests.\n\nIn Figure 1, the automatic test circuit is repre- \nsented in the shaded portion of the block diagram, \nand the other units are those parts of the CAMA\n\nsystem and the crossbar tandem office with which \nthe test circuit is concerned, The trunk relay equip- \nment to be tested is represented in the upper left \ncorner of the diagram, What happens on a regular \nservice call is briefly this: the originating customer \nat the left is connected through the crossbar tan- \ndem office via the incoming trunk relay equipment, \nand during the progress of the call this trunk cir- \ncuit performs the tasks that have been mentioned. \nThe function of the test circuit is simply to simu- \nlate, in as exact a manner as possible, an actual \nservice call as dialed by a customer, and to observe \nwhether the trunk circuit is performing properly. \nConditions applied to the trunk circuit during the \nsimulated call are more severe than those usually \nencountered in service. The test circuit, however, \ndoes not automatically test transmission (that is, \nthe transmission characteristics of the voice path), \nnor does it request the services of the CAMA \nspecial operator.\n\nLocal offices send called-number pulses in sev- \neral different ways, termed Panel Call Indicator \n(PCI), Dial (DP), and Multifrequency (MF)\n\npulsing. At the present time, the test circuit simu- \nlates the panel call indicator and dial pulse types \nof calls, and the multifrequency type is to be added \nlater. The test circuit recognizes the type of trunk \ncircuit it is to test, and alters the test procedure \naccordingly. It contains generators for producing \nboth the panel call indicator and dial pulses.\n\nIn terms of the called-number pulses and other \ntypes of signals, the test circuit is thus a simulated \noriginating office. It is also a simulated terminating \noffice because the outgoing or destination trunk \nused is actually a \u201creturn test line,\u201d which has as \nits destination the test circuit. Signals that the ter- \nminating office sends back to the tandem office are \nby this means also simulated, and the effects on the \nincoming trunk circuit observed.\n\nCertain tests are made as the trunk circuit re- \nsponds to the normal sequence of events of a simu- \nlated telephone call. The test circuit tests the ef- \nfects of \u201canswer\u201d and \u201cdisconnect\u201d signals as they \nare normally received from the terminating office. \nThe effects of \u201cseizure\u201d and \u201cdisconnect\u201d signals \nfrom the originating office are also tested. In addi-\n\nORIGINATING | | \nOFFICE l CROSSBAR TANDEM OFFICE | TERMINATING OFFICE \n| \n| yest | \u2014 | + +\u2014 \nLINK | SENDER MARKER | EQUIPMENT \n| \n-- \n| H INDE XER \n\\ \nLINE \n| CONNECTOR ANO | \n| ; ; \nINCOMING | \nTRUNK RESERVE TRUNK | \n| OUTPULSING \nvast CIRCUITS | \n| SEQUENCE | \n| CONTROL \nTESTS | \nSEM! ~AU TOMATIC \n| Tests: | \n| THROUGH | \nAUTOMATIC TRANSMISSION \nTR K TEST \n\u2014 \n| INDICATOR \n| DISPLAY | \nFig. 1. \u2014 Block diagram showing the automatic trunk-test circuit and its association with CAMA crossbar\n\ntandem equipment; test circuit acts as a simulated originating and terminating office.\n\ncorder circuit at the time the entries are made. \nThe test call places a special non-billable entry on \nthe AMA paper tape.\n\nAll these operations are performed in testing a \nsingle incoming trunk circuit. If no trouble is found, \nanother trunk will be selected automatically, and \nthe test circuit will proceed in this manner through \nall the incoming trunk circuits in the office. Trunk \ncircuits that are busy with actual telephone calls \nmay be passed by and tested later when they are \nidle. If trouble is encountered, te sting will stop \nand an alarm will sound. An array of lighted lamps \nindicates the point at which the test was blocked, \nwhich is evidence of the nature of the trouble.\n\nPotomac Telephone Company at trunk-test circuit \nconnector frame in Washington, D. C., installation. \nROM \nTOMER \n\u2014 \ntion, extra tests are inserted between these normal Panera heme \nevents. For example, the way the trunk responds to ET AES \na busy signal is tested while the trunk is waiting ee eee \nfor the simulated call to be \u201canswered.\u201d il \nTANOEM TRUNK FA Ties \nSeveral additional tests performed by this circuit Ba ! \nSWITCHES NG, ar - \u2014 \nshould also be mentioned. Under certain circum- & \nstances, when a first attempt to establish a tele- i; -- rauNe \nphone connection is unsuccessful, a second trial Tare Bu57 /\n\nThe test circuit determines whether the trunk cir- \ncuit properly handles such requests for a second (b) OF TRUNK USING \"MAKE BUSY\" CIRCUIT\n\ntrial. It also checks that there are no trouble con- Fig. 3\u2014 Two types of trunk connections between\n\nditions on the outgoing leads of the trunk circuit, \nand connects other leads into the trunk circuit to \nobserve the performance of individual relays. Cer- \ntain relays must operate and release within speci- \nfied time intervals, and the test circuit checks such \ntiming characteristics. Also, the test circuit deter- \nmines whether the incoming trunk circuit is able \nto cause the proper recording of entries on the paper \ntape of the Automatic Message Accounting equip- \nment. This is done by a test connection to the re-\n\nlocal and tandem offices; both are transferred \nwithout assistance from the local telephone office.\n\nful in checking some of the office wiring which pro- \nvides information necessary for routing the call and \ndetermining the charges.\n\nSome of the interesting aspects of the new cir- \ncuit are the features associated with the transfer of \nWhen a trunk\n\nmust be made that customers will not be denied \nservice by having their calls run into a \u201cdead end.\u201d \nThis is commonly done by making the trunk appear \nbusy; that is, personnel at the originating local of- \nfice arrange the originating trunk circuit on that \nend of the cable pair to appear busy to all calls. \nIn this way, calls are diverted to other trunks dur- \ning the test procedure. Such manual \u201cmake-busy\u201d \narrangements at the originating end are not prac- \ntical when an automatic test circuit is located at \nthe tandem office.\n\nThis problem is solved in two different ways, as \nillustrated in Figure 3. The first method has to do \nwith originating offices (crossbar or panel) that \nuse the PCI type of out-pulsing. As seen in Fig- \nure 3(a), the incoming telephone call is in this case \ndiverted to a \u201cspare\u201d or reserve trunk circuit, the \ntransfer being accomplished automatically at the \ncrossbar tandem end of the connection. The re- \nserve trunk circuit is a duplicate of the incoming \ntrunk circuit; it is associated with test circuit identi- \nfication circuits which provide the same rate-class \n(charging) and class-of-service (routing) informa- \ntion as with a regular trunk circuit normally serving \na particular cable pair. Arrangements are also \nmade that if the testing of the regular trunk circuit \nends before the end of the customer\u2019s call, the trans- \nfer condition is maintained as long as the reserve \ntrunk is busy. Under this management the customer's \ncall will not be interrupted if the test being con- \nducted ends before he is finished.\n\nA second method of diverting customer calls is \napplied when the originating local office (step-by- \nstep switching equipment) uses the DP type of \noutpulsing. This is illustrated in Figure 3(b). \nHere, outgoing trunk circuits in the originating \nstep-by-step offices are arranged so that the test cir-\n\ncuit in the crossbar tandem office can make them \nappear busy. Calls are therefore diverted to other \ntrunk circuits in the originating office, and no re- \nserve trunk circuits in the crossbar tandem office \nare needed in this case. As with the previous method \nof transfer, the arrangement here permits automatic \ncontrol by the test circuit in the crossbar tandem \noffice, and thus eliminates the need for manual as- \nsistance in the originating office.\n\nAs mentioned earlier, the test circuit proceeds \nmethodically through the incoming trunk circuits \nin the crossbar tandem office. Now, if a faulty trunk \ncircuit is detected, a feature is provided to hold \nit out of service for subsequent maintenance action. \nOn the key and lamp panel of the test circuit, there \nare a number of trunk-hold-busy (THB) keys. \nWith these keys faulty trunks detected by the test \ncircuit can be held out of service pending mainte- \nnance action, meanwhile permitting the test cir- \ncuit to test other trunks. A THB key is associated \nwith the trunks of one AMA recorder. Operation \nof the key after one of the trunks has been seized \nby the test circuit will lock that trunk to the key. \nThe test circuit busy test insures that only idle \ntrunks can be thus taken out of service. If further \ntesting of the transferred trunk is required, the test \ncire*t can automatically scan all trunks associated \nwit . a particular THB key and locate the trans- \nferred trunk. The trunk may be released from its \nlocked-in condition at any time merely by restoring \nthe associated THB key.\n\nThe test circuit can automatically test 2,000 in- \ncoming trunk circuits, a capacity that is an im- \nportant factor in the economical maintenance of \ncontinuous service to customers. A significant part \nof the economy is derived from the elimination of \nthe need for manual make-busy arrangements.\n\nin 1948, and he joined the Laboratories in the same year as a student in the \nCommunications Devleopment Training Program. Rotational work assignments \nperformed as part of that program included duties in the Switching Apparatus \nDevelopment Department and in the Transmission Development Department. \nSince 1949 he has been in the Switching Development Department where he \nhas worked on switching circuit design, first for the Automatic Message \nAccounting Center and then for the Tandem Office CAMA project. Later he \nworked on the design of a portable traffic usage recording circuit and recently \nhas joined the group assigned to the toll office CAMA project. Mr. Dusenberry \nis an Associate Member of the American Institute of Electrical Engineers and \nis a member of Tau Beta Pi and Phi Beta Kappa.\n\nTactical military communication has recently been much improved by the \nLaboratories\u2019 development of a light-weight system for transmitting twelve \ntelephone conversations over a single four-conductor cable. Two types of \nrepeaters have been developed to restore signal strength and permit the \nterminals to be placed as much as 200 miles apart.\n\nAs a part of the development of a twelve-channel \ncable carrier telephone system for the military,\u00b0 \ntwo types of repeaters were designed. The first of \nthese is officially designated Telephone Repeater \nAN/TCC-11;, it operates without attendance by per- \nsonnel. The other is an attended repeater, officially \ndesignated Telephone Repeater AN/TCC-8. Both \nunits, like all the equipment used in the complete \nsystem, were designed to achieve a minimum in \nweight and a maximum in performance. The equip- \nment thus represents a great improvement over \nprevious tactical military carrier telephone com- \nmunication systems. The unattended repeater in par- \nticular contains some interesting and unique fea- \ntures, and a description of these will comprise most \nof this article. The attended repeater is more con- \nventional in design and will be described in some- \nwhat less detail.\n\nA photograph of the unattended repeater is shown \nin Figure 1. These repeaters are normally spaced at \n5&-mile intervals along the cable to amplify signals \nin the frequency range from 12 ke to 99 ke. The gain \nor amplification of the repeater is equalized to com- \npensate for the attenuation of a 5%-mile cable span \nat all frequencies in the 12 to 68 ke (message and \npilot) range. In addition, the repeater gain is auto- \nmatically regulated to compensate for changes in\n\ncable loss resulting from temperature variations. The \n12- to 99-ke band includes twelve telephone message \nchannels occupying the 12-ke to 60-ke range, a pilot \nor controlling frequency of 68 ke, and tones used for \ntest purposes at 83, 91, and 99 ke. When a field tele- \nphone set (EE-8) and a portable test set (TS-712) \nare connected to the repeater, personnel may ring \nthe attended points of the system and converse over \na voice-frequency order-wire circuit. Voice fre- \nquencies are not amplified in the unattended re- \npeater, however.\n\nA block diagram of the repeater is shown in Fig- \nure 2. Power from a constant-current 0.1 ampere de \nsupply is sent along the cable from an attended \npoint, and is delivered to the two amplifiers used for \nthe two directions of transmission. As indicated in \nthe diagram, the tube heater elements are wired in \nseries. Current from the power supply flows through \nthe heaters, and the resulting voltage drop is used \nas plate voltage for the tubes. One power supply can \npower as many as three repeaters. The power Loop. \nING switcH in the center of the diagram permits \nconnecting the de power path so that the current \neither flows through the tube heaters and along the \nnext length of cable to the succeeding repeater, or, \nif the repeater is the last of a series of three, \u201cloops \nback\u201d and returns to the low side of the power sup- \nply. This arrangement requires good electrical in-\n\nAN/TCC-11 \npeater on a pole, This repeater unit can also be \nplaced on the ground along the cable run.\n\nsulation between the circuit \u201cgrounds\u201d and the out- \nside case, since the potential of the circuit \u201cgrounds\u201d \nrelative to earth ground may be as high as 450 volts.\n\nOf course, prime requisites for tactical military \nequipment are that it be portable, rugged, and \nreliable. This repeater is designed especially for an \nexposed environment so that it may lie on the \nground or be strapped to a pole (see Figure 1) with- \nout shelter, in either arctic or tropic climates. The \nelectrical apparatus is shock-mounted inside a \nwaterproof cylindrical aluminum case 28%\u201d long and \n10\u201d in diameter, The complete assembly weighs only\n\n83 Ibs. Removable covers over ports at each end \nprovide access to cable connectors and test points. \nThe power looping switch and a three-position \nREPEATER SWITCH (discussed later) are also reached \nthrough covered ports on the end sections. The ease \nof installing and removing the repeater merely by \nconnecting the ends of the cable sections to the \nmating conductors on the case simplifies the main- \ntenance of the system.\n\nIt is beyond the scope of this article to describe \nthe component parts of the repeater in detail, but a \nfew facts and figures deserve mention. The gain of \nthe amplifiers at the pilot frequency (68 kc) is ap- \nproximately 22.4 db at a temperature of 45\u00b0F, the \nassumed average temperature. Equalization is built \ninto the amplifiers so that at average temperature, \nthe shape of the loss vs frequency characteristic of \n5% miles of cable is compensated within + 0.06 db at \nall frequencies in the message band. Regulation of \nthe gain characteristic of each amplifier is accom- \nplished by means of an ambient temperature-sensi- \ntive thermistor, which controls the loss characteristic \nof a network associated with the amplifier. If the \nthermistor temperature follows the cable tempera- \nture, compensation for the cable loss variations will \nbe obtained. This compensation is only approximate, \nof course, since the temperature of the cable and the \ntemperature of the thermistor will seldom be iden- \ntical. However, errors in the transmission character- \nistic will accumulate to only a moderate amount \nbetween attended points (2.4 db at 68 ke at the end \nof a 6-repeater span for a 20\u00b0F difference in cable \nand thermistor temperature at each repeater ). More-\n\nFILTERS f TERS \nLOW VOICE FREQUENCY ORDER w f LOW \nPASS PASS | \nal bo \nae |! , \nPASS DELAY BA PASS | \n| REPEATER \nSWITCH \nTUBE HEATER ELEMENTS THERM \n\u2018 \nver | WER \n' LOOP PING \nTER \nER LOOPIN Fig. 2\u2014 Block \nJ \ngram of AN/TCC-11 \nTHERMIS \n6 unattended repeater. \nFILTERS | EQUALIZER! FILTERS \nBA LAY HIGH \nLOW \nFREQUENCY ORDER WIRE PAT pace \n>\n\nover, at the attended points additional equalization, \nboth manual and automatic, is supplied to com- \npensate for the accumulated error.\n\nFigure 4(a) shows the predicted transmission \ncharacteristics of a 200-mile system under the condi- \ntion that the repeater and cable temperatures are \nequal at 45\u00b0F, and under the further conditions that \nthe repeater temperature is 20\u00b0 above or below the \ncable temperature. The closeness of the three curves \nis an indication of the regulator performance, and \nthe rather modest departure from uniform gain over \nthe frequency band shows the extent to which the \nsystem transmission is equalized. Figure 4(b) shows \nthe variation of the same 200-mile system character- \nistic when the ambient temperature varies + 30\u00b0F\n\nfrom an initial temperature of 125\u00b0. Equally good \nperformance is obtained when the initial tempera- \nture is \u201430\u00b0F.\n\nAn interesting and important feature of the \nAN/TCC-11 repeaters is the built-in fault-locating \napparatus, Each repeater contains a tunable band- \npass filter ( called a \u201clooping filter\u201d in the right hand \npart of Figure 2) bridged across the two directions \nof transmission. This filter provides a tuned coupling \npath between the two directions of transmission at \n83 ke, 91 ke, or 99 ke. A three-position REPEATER \nswitcH selects the proper pass band, which depends \nupon whether the repeater is the first, second, or \nthird from the attended point at which power is \nsupplied. If a repeater fails, the attendants at the \nnearest attended point can apply the $3-, 91-, and\n\n3 \n\u2018 CABLE 20\u00b0F \nCOLDER THAN \nAMPLIFIERS > \n=~ AMPLIFIERS CABLE 20\u00b0F \n_\u2014 AND CABLE WARMER THAN \nAT 45\u00b0F AMPLIFIERS \n4 \n-2 \nVU \nw-3 \nra) \nz 3 \n125\u00b0 F + 30\u00b0F =155\u00b0F \n> } \n< \n1 \n0 \n~ | \n125\u00b0F \u201430\u00b0F = 95\u00b0F \n~3 \n10 5S 20 25 30 35 40 45 50 55 60 65 70 \nFREQUENCY IN KILOCYCLES PER SECOND \nFig. 4 (a) Gain-frequency characteristics of 200-\n\nmile system, showing changes when cable is colder \nor warmer than amplifiers; (b) characteristics for\n\n99-ke test tones to the line and measure the signal \nreturned in the opposite direction of transmission. \nThe faulty repeater will not return the applied tone.\n\nCable breaks in general produce power alarm \nindications at attended points and can be located \nby resistance measurements of the line. If power \nfails, relays in the repeaters release and connect \n100,000-ohm resistors across the line. If, then, a \nbreak has occurred between the first and second \nrepeater, a resistance measurement will indicate a \nresistance of about 100,000 ohms. If the break is \nbetween the second and third repeater, the resist- \nance will be about 50,000 ohms, and so on.\n\nAt intervals up to 40 miles along the cable, \nthe attended AN/TCC-S5 repeater is inserted in the \ntransmission path, This repeater is shown in Figure \n3, Like the AN/TCC-11, it amplifies, equalizes, and \nregulates signals in the 12- to 68-ke range (a block- \ning filter stops incoming high-frequency test signals \nbefore amplification ). In addition, this repeater in- \ncludes voice-frequency order-wire amplification, \nequalization, and signaling circuits; a test set is also \nprovided which enables measurements to be made \nat various test points in the power and transmission \ncircuits. A 200-volt power supply delivers de power \nto the attended equipment, and two 600-volt, 0.1- \nampere constant-current supplies deliver power over \nthe cable to as many as three unattended repeaters \nin each direction of transmission.\n\nCABLE CABLE \n| \u201cBUILDING \"BUILDING \npass Out\u201d WW OuT\" \nFILTER gh at DEVIATION \nNETWORK LA) LSLOPE) (DELAY) (BULGE) | BASIC EQUALIZER NETWORK \ni \nSPIRAL- \nFOUR \nCABLE \nREGULATING \naan PASS BUILDING AND = = = BUILDING \nEQUALIZER ULGE} |DELAY| |sLopEe} | FLAT NETWORK\n\nFigure 5 is a simplified diagram of the components \nin the carrier frequency transmission path. Man- \nually adjustable equalizers permit accurate compen- \nsation for manufacturing or other variations in the \ncable and repeater characteristics, Equalizers with \nthe same loss characteristics as various lengths of \ncable are provided so that short lengths of cable, \nwhich may precede or follow the attended point in \na given direction of transmission, can be \u201cbuilt out\u201d \nto the nominal 5%-mile length. The regulating system \nat the attended point operates under the control of \nthe 68-ke pilot transmitted from the terminal, The \n68-ke pilot is picked off at the output of the at- \ntended station amplifier through a narrow-band \nfilter, then amplified and rectified. The rectified \nvoltage is used to control the loss of the regulating \nnetwork by varying a thermistor element. The reg- \nulator network, like that in the unattended repeater, \nhas a loss-frequency characteristic that compensates \nfor the cable attenuation variations. This shape also \nmatches the transmission deviations not compen- \nsated for by the unattended repeater. These devia- \ntions are \u201cmopped up\u201d at the attended points. Reg- \nulators operate over a + 5-db input variation to \nproduce less than a + 0.5-db output variation.\n\nBlock diagram of the carrier frequency transmission paths of the AN/TCC-8 telephone repeater.\n\nAgain, the military requirements of portability, \nreliability, and ruggedness were important con- \nsiderations from the standpoint of equipment de- \nsign. The attended stations will always be provided \nwith some sort of shelter, so waterproof containers \nare not essential. However, with the covers latched \nin place, these containers are rainproof and are suit- \nable without added protection for storage and trans- \nportation over long distances in all climates. As \nshown in the photograph (Figure 3), the complete \nrepeater consists of five boxes. The heaviest (the \none containing the repeater panel) weighs 118 Ibs., \nand the total weight of the complete repeater is 511 \nlbs. The boxes have specially notched edges so that \nthe individual units can be stacked as shown in the \nfigure. Inter-unit cables and cable connectors are \nused to interconnect the various circuits quickly \nand easily.\n\nThe Signal Corps has conducted trials of both \nrepeaters. They were subjected to severe vibration, \nshock, \u201cbounce,\u201d heat, and humidity tests, and elec- \ntrical tests were made on the units in an experi- \nmental cable system. These trials furnished data for \nfinal adjustments in design so that, as produced, the \nrepeaters give excellent performance in all respects.\n\nJoun M. Barstow, Jn. received his degree of B.S. in Electrical Engineering \nfrom the University of Kentucky in 1949. He joined the technical staff of Bell \nTelephone Laboratories in February of that year and enrolled in the Labora- \ntories Communications Development Training Program. In 1950 he joined the \nTransmissions Systems Development Department where he was engaged in the \ndevelopment of military carrier systems and in the development of the trans- \natlantic submarine cable system. Prior to his resignation from the Laboratories \nin January 1956, he was associated with military communication systems studies,\n\nThe Laboratories will institute a program of fel- \nlowship awards to allow a few outstanding students \nin the Communications Development Training Pro- \ngram to acquire their Ph.D. degrees while continu- \ning as regular Laboratories employees. Awards will \nnormally be made at the end of the students\u2019 first \nyear in C. D. T. A small number will be made \neach year, depending upon the size of the C. D. T. \nclass and the merit of the candidates. This year one \nor two additional awards will be made to candi- \ndates who completed their first year in C. D. T. \nprior to this year. First selections will be made \nlater this year.\n\nNominations for fellowships will be made by the \nDepartment of Education and Training, and awards \nwill be made by the C. D. T. Policy Committee, \nconsisting of E. I. Green, Chairman, H. W. Bode, \nR. R. Hough, A. D. Knowlton, F. D. Leamer, M. B. \nMcDavitt, and S. B. Ingram, Secretary.\n\nFellows will be selected from the upper scholastic \nlevel of C. D. T. with consideration being given \nto performance on rotational assignments, poten- \ntial as Laboratories employees, and promise of bene- \nfit from advanced academic training. The fellow- \nship program is expected to strengthen the \nLaboratories staff by the further academic devel- \nopment of outstanding Communications Develop- \nment Training students.\n\nChoice of institution will be made jointly by the \nfellow and the C. D. T. Policy Committee, with \nproximity to the Laboratories being one considera- \ntion. Fellows will continue as Laboratories em- \nployees, drawing full salary and receiving regular \nemployee benefits. Allowances will be made for \ntuition, books, and certain other expenses, During \nacademic vacations, Fellows will return to the \nLaboratories where they will work on projects in \nvarious technical departments.\n\nA wind tunnel, normally employed in aerody- \nnamic studies, was recently used by the Labora- \ntories to simulate the effect of hurricane conditions \non some telephone apparatus. Acting on a request \nby the A.T.&T. Co., members of the Outside Plant \nDevelopment and General Staff Departments at the \nPoint Breeze Laboratory conducted a series of tests \nto determine what effect hurricane winds would \nhave on a \u201cReady-Access\u201d cable terminal. This \nterminal is a device used to provide entrance from\n\nD. Gross, Director of the Univ. of Maryland wind tunnel, \ncenter, with P. P. Kaliss, left, and W. J. Fullerton of BTL \nin the test room of the Univ. of Maryland wind tunnel.\n\nTo perform these tests, arrangements were made \nwith the University of Maryland to use the wind \ntunnel on their campus at College Park, Maryland. \nCable terminals were set up in this wind tunnel \nand subjected to various wind velocities up to and \nincluding 200 miles per hour. The terminals sur \nvived the tests without damage.\n\n\u201cReady-Access\u201d cable terminal after it was sub- \njected to 200-mile-per-hour winds.\n\nCleo F. Craig, President of the A.T.&T, Company, \ngreeted share owners at the 1956 Annual Meeting.\n\nBell Companies will spend more than $2 billion \nfor new construction this year, Cleo F. Craig, presi- \ndent of the American Telephone and Telegraph \nCompany, told share owners at the company\u2019s 7st \nannual meeting in New York City, Such an amount \nfor growth and improvement in any one year is the \nlargest on record for the System. It is also the \nlargest ever planned for a single year\u2019s expendi- \nture by any American business.\n\n\u201cSo far this year new orders are considerably \nahead of 1955,\u201d Mr. Craig reported to the share \nowners. Barring unforeseen developments, the Bell \nSystem should be celebrating 50 million telephones \nin service when the share owners meet in 1957. \nThis compares with about 47 million today,\n\n\u201cLong distance business also continues to show a \ngood healthy growth,\u201d Mr. Craig said. This is only \npartly due to the increase in telephones, although \nthat is an important factor, We increase long dis- \ntance usage by going out and selling it, Mr. Craig \npointed out, and we also promote service by making \nit more attractive and convenient to use. Mr. \nCraig cited direct distance dialing as a powerful \nincentive for our customers to use long distance. \nAll this growth and improvement is the reason for \nthe slightly more than $2 billion in new construc- \ntion by the Bell System companies in 1956.\n\n\u201cFor most of their new equipment,\u201d Mr. Craig \nsaid, \u201cthe companies rely on Western Electric, the \nmanufacturing and supply organization of the Sys-\n\ntem.\u201d \u201cLast year,\u201d as the Annual Report pointed out, \n\u201cWestern Electric did a remarkable job of increas- \ning production. This year they are turning out even \nmore cable, dial equipment and various types of \ntelephone instruments.\u201d\n\nThe share owners met in a new location \u2014 the \nsixth floor at 50 Varick Street in downtown New \nYork. A.T.&T. purchased the building in 1953 to \nprovide more room for sections of the Treasury \nand the Comptroller's departments, Earlier this \nyear the Company completed a three-floor addition, \nand space on the sixth floor can be used from time \nto time for meetings of this kind.\n\nIn answer to questions from the floor concerning \nthe 1955 Annual Report, Mr. Craig replied that \nlast year\u2019s advertising expense was less than one \nper cent of operating expenses; that the recent anti- \ntrust suit consent decree would not disturb the \nrelationship among Bell Telephone Laboratories, \nWestern Electric and the Associated Bell Com- \npanies; and that profits from work done for the \nDepartment of Defense average about four per cent.\n\nShare owners re-elected each of the 19 incum- \nbent A.T.&T. directors, with votes in excess of \n41,000,000. They also re-elected the firm of Lybrand,\n\nFred R. Kappel, President of the Western Electric \nCompany, introduced a motion picture about \nWestern Electric at the: stockholders meeting.\n\nRoss Brothers & Montgomery, auditors. During the \ndiscussion following nominations for directors, Mr. \nCraig announced that last year the directors them- \nselves had decided to establish a retirement age \nof 72, to be fully effective by 1960. While ballots \nfor directors and auditors were being counted, \nFred R. Kappel, president of the Western Electric \nCompany, introduced a color film which dramatic- \nally told of the important role played by Western \nElectric in serving both the Bell System and na- \ntional defense.\n\nDuring the meeting, more than a score of the \nshare owners present used the microphones pro- \nvided to ask questions or make comments. In the \ngeneral discussion period at the end of the meeting, \nattention centered on earnings and dividends, and \nof whether the stock should be split.\n\nMr. Craig stated the present position of the Board. \nHe said in part: \u201cWe have made very careful studies \nto find out what happens to companies when they \nsplit their stock. Where there is an announcement \nof a stock split, the price of stock goes up. There \nis a flurry in the stock market and a rise in the \nprice of the stock.\n\n\u201cBut if there is no substantial dividend increase \nat the time of the stock split, or shortly thereafter, \nthat stock goes down in the market. In almost \nevery case it goes down below where it was before. \nThere is a spurt, and then a feeling of quite acute \ndisappointment. The stock winds up in a poorer \nplace, financially, after the split than before the split, \nif the trend of the market is taken into account.\n\n\u201cIf the telephone company were to split without a \ndividend increase,\u201d Mr. Craig continued, \u201cthe price \nof the stock would go up on the market tempo- \nrarily. That would benefit some people who bought \nthe stock on a short-term basis. It would not bene- \nfit the people who want to keep the stock as a per- \nmanent investment,\u201d Mr, Craig stated. \u201cA stock split \nwithout any increase in dividends to share owners \nwould be a very short-sighted and damaging step \nfor the company to take. Unless there is a divi- \ndend increase at the same time there is a stock split, \nthe speculators may be able to make a profit on the \nrise and fall in the market. The Board of Directors \nis not going to take action with respect to the stock \nmerely to push the price up and down on the \nmarket.\u201d\n\nIn discussing the question of a dividend increase, \nMr. Craig pointed out that American Telephone and \nTelegraph stock bought on the market today brings \na 4.9 per cent return on the money invested. The \ncompany, however, has been able to put only a lim-\n\nited amount per share into undistributed earnings. \nSix years ago there were 25 million shares out- \nstanding and today there are 55 million shares out- \nstanding. On the average, about five million shares \nhave been added each year. In 1949 there was \n$14.23 in undistributed earnings per share \na year and a half's dividend protection, There has \nto be a certain amount of undistributed \nto protect the whole financial situation. \nThe company has expanded by nearly 30 million \nshares. In that period undistributed earnings per \nshare have risen from $14.23 to $19.99. This means \nnot only that nearly $6 per share has been added \nto undistributed earnings, it means the business has \nearned enough to put into surplus for each new \nshare an amount equivalent to the surplus per share \nthat existed before those new shares were issued. \nMr. Craig pointed out that in 1955 \nstarted with $18.19 in undistributed \nshare.\n\nthe company \nearnings per \nRoughly, six million shares were added. So \nfor each of the six million shares, the company had \nto earn, in addition to the dividend payable on \nthose shares, $15.19 for surplus in order to protect \nall the previous shares outstanding. What was left \nafter that could be added to the surplus per share, \nand $1.80 per share was added last year.\n\nLast year the company, for the first time in a \nlong time, paid out only 69 per cent of earnings in \ndividends. The Board of Directors is trying to take \nwhatever is left over and put it back into the busi- \nness, and make that earn for the share owners.\n\nDr. Mervin J. Kelly declared to engineering edu- \ncators last month that our nation is engaged in a \ntechnological race for strength, and urged that we \n. move ahead as fast as we can.\n\nSpeaking at a conference of more than 300 mem- \nbers of the American Society for Engineering Edu- \ncation, Dr, Kelly described the nation\u2019s expanding \nneed for scientists and engineers and emphasized \nthe need for more and better training of both re- \n: search and development people. He also suggested \n\u2018 that educators consider establishing a fifth year for \nthe basic engineering course and then recommend \nit to all planning a career in research and develop- \nment so that these students would receive more \nengineering fundamentals. He also urged that many \nmore students continue their training to the doc- \ntorate level. The conference was the Spring Meet- for a Career in Industry.\u201d Dr. Kelly's keynote ad- \ning of the A.S.E.E. Middle Atlantic Section, held dress was entitled \u201cPreparation for a Career in De- \nMay 12 at Bell Telephone Laboratories, Murray velopment and Research.\u201d\n\nHill, N. J. In stressing the need for more and better engi- \n: Concerned with the critical shortage of engineer- neering training, Dr. Kelly pointed out that the four- \nAe ing graduates, the all-day conference had as its year engineering graduate generally lacks sufficient \ntheme the \u201cPreparation of Engineering Students training. He showed how Bell Laboratories\u2019 belief \nin more education is exemplified in the Communi- \ncations Development Training program, established \nin 1948 as a three-year course for newly employed \nengineers with bachelor\u2019s and master\u2019s degrees.\n\nIn suggesting a five-year engineering course, Dr. \nKelly explained that it could be arranged so that \nstudents could stop at the end of four years or con- \ntinue for the fifth year. This fifth year would be \nadvisable, he said, regardless of whether the stu- \ndent went on for a doctor's degree. Dr. Kelly also \nurged industry to place greater emphasis on basic \nresearch and recommended that research activities \nbe separated from development and design activities.\n\nDr. Kelly described the successful experiment of \nBell Laboratories which 20 years ago separated re- \nsearch from the development and design functions \nwithin the company. He explained that the Labora- \ntories worked to create an environment of freedom \ncomparable to that found on university campuses \u2014 \nMembers of the American Society for Engineering Educa- freedom in research, freedom to publish, and free- \ntion attended morning meeting in the Arnold Auditorium dom to intermingle with scientific people through- \nat the Murray Hill laboratory. out the world. \u201cThe results of the work done in this\n\nDr. M. J. Kelly addresses the spring meeting of \nMiddle-Atlantic Section of the American Society \nfor Engineering Education at Murray Hill.\n\narea of basic research have created a reservoir of \nknowledge which, for a long time, will contribute \nto improvements in communications,\u201d he said, Dr. \nKelly also insisted in his address that the research \nscientists must have \u201cthe freedom to wander into \nthe area of tomorrow.\u201d\n\nIn urging that the nation move ahead technolog- \nically as fast as it can, Dr. Kelly told of experiences \nrelated to him by a professor of aGerman Technische \nHochschule. The professor had been forced to \nwork in Russia, and his evidence indicated that top \nRussian engineering students are more advanced \nthan their counterparts in America.\n\nDr. Kelly was introduced by R. Karl Honaman, \nDirector of Publication. In welcoming members of \nthe society to Murray Hill, Mr. Honaman expressed \nthe belief that \u201cthe community of interest between \nindustry and education is stronger in the field of \nengineering than in any other area of professionals \nand their educators. Through common member- \nships in professional societies and as individuals in \ncontact with each other in both groups, this com- \nmunity of interest is strengthened.\u201d\n\nImportant problems in the field of engineering \neducation were discussed at three afternoon meet- \nings, as follows: \u201cPreparation of Engineering Grad- \nuates for Industry\u201d \u2014 Chairman; Sydney B. Ingram, \nDirector of Education and Training at Bell Labora- \ntories. Speakers: Morris D. Hooven, Public Service \nElectric & Gas Company and President of the \nA.I.E.E.; Henry N. Meixner, General Assistant Di-\n\nrector, Mechanical Development Laboratory, E. 1. \ndu Pont de Nemours & Company, Inc.; and Frank \nW. Miller, Vice President \u2014 Manufacturing, Yarn- \nall-Waring Company, Philadelphia.\n\n\u201cPreparation of Secondary School Graduates for \nEngineering\u201d \u2014 Chairman; Elmer C. Easton, Dean \nof Engineering, Rutgers University. Speakers; Lynn \nL.. Merrill, Dean of the Faculty, Stevens Institute of \nTechnology; Harold K. Work, Director of the Re- \nsearch Division, College of Engineering, New York \nUniversity; and Ablett H. Flury, Assistant Commis- \nsioner of Education for the State of New Jersey, \nwho delivered a paper by Frederick Raubinger, \nCommissioner of Education.\n\n\u201cCurrent Engineering Manpower Needs in Indus- \ntry and the Colleges\u201d \u2014 Chairman: Donald S. Bridg- \nman, Director of College Relations, American Tele- \nphone and Telegraph Company. Speakers: William \nT. Cavanaugh, Executive Secretary, Engineering \nManpower Commission of the Engineers Joint Coun- \ncil; and Robert W. van Houten, President, Newark \nCollege of Engineering.\n\nProjects, Western Electric Company, gave a re- \ncently declassified account of construction of the \nDistant Early Warning Line which guards our Are- \ntic continental approaches,\n\nArrangements for the conference were made by a \nLaboratories committee headed by Frank D. Lea- \nmer, Personnel Director.\n\nDr. M. J. Kelly was elected a director of The \nEconomic Club of New York at a meeting in New \nYork City April 16. His term of office is for three \nyears beginning June 1, 1956,\n\nThe Economic Club is an association whose ob- \njectives are: \u201cto contribute toward building a \nstronger business leadership in America; to provide \na platform for discussing those problems that af- \nfect the business community, and to develop an \ninformed opinion among all citizens as to the aims \nand achievements of our economic system.\u201d\n\nDr. Kelly also recently accepted a post on the \nNew York City Board of Education's Advisory \nCommittee on Science Manpower. In this capacity, \nhe is chairman of a subcommittee on the Role of \nIndustry and Colleges.\n\nThe first field trial of a rural telephone system \nmaking use of transistors and the Bell Solar Bat- \ntery, held in Americus, Georgia, has been con- \ncluded with satisfactory results. The Bell Solar \nBattery was installed on a part of this trial system \nin October, 1955, as an experimental substitute for \nordinary batteries, the conventional source of elec- \ntrical power for such a system where power lines \nare not readily available. Power engineers report \nthat the Bell Solar Battery, first device to change \nsunlight directly and efficiently into useful amounts \nof electricity, has lived up to expectations.\n\nBell System engineers have ascertained from the \nGeorgia tests that, from the standpoint of reli- \nability and effective operation, the Bell Solar Bat- \ntery mounted on a pole can be used to furnish elec- \ntricity for rural telephone equipment. The favorable \noperation report on the Solar Battery, however, does \nnot mean that regular commercial use of the device \nis as yet practical. Until the ultra-pure silicon used \nin the device becomes less expensive, it will be \nmore economical to use conventional power sources \nfor telephone systems. Hence, no further trials of the \nBell Solar Battery are planned immediately.\n\nOf more significance for telephone customers \nand electronics engineers is the future indicated \nfor the transistor, The Georgia tests have justified \nthe confidence placed in the transistor as a key to the \nplanning of future telephone systems.\n\nAbout 275 transistors were used in the Americus \ntrial. The short-haul, rural telephone system that\n\nwas tried out derives its economy largely from the \nuse of transistors instead of electron tubes which \nshould help make it possible to provide improved \ntelephone service in thinly populated rural areas.\n\nAnother device, used for the first time in the \nexperimental system, made transistor use practical. \nThis is a lightning protector which affords the low- \npower transistor special protection against elec- \ntrical surges. This new device \u2014 called a \u201csilicon \naluminum junction diode\u201d \u2014in conjunction with \nstandard protectors will protect transistors from \nordinary lightning damage. This lightning pro- \ntector evolved from the same basic research as the \ntransistor itself and belongs to the same general \nfamily of semiconductor devices.\n\nSan Diego First Large City Added to \nNationwide Customer Dialing Network\n\nThe recent cutover of the University Crossbar \nTandem Office in San Diego, California, marked \nanother major step along the road toward nation- \nwide direct distance dialing. This was the first oc- \ncasion in which an entire large multi-office city, \nserved by the Bell System, was arranged for direct \ndistance dialing. (In 1950, San Diego had a popula- \ntion of about 435,000, and in 1953, it had about \n130,000 telephone customers. )\n\nThe San Diego cutover also marked the first oc- \ncasion in which a city served by step-by-step cen- \ntral offices was included in the nationwide dialing\n\nThe May, 1956, issue of The Bell System \nTechnical Journal contains the following:\n\nChemical Interactions Among Defects in \nGermanium and Silicon by H. Reiss, C. S. \nFuller and F. J. Morin.\n\nSingle Crystals of Exceptional Perfection | \nand Uniformity by Zone Leveling by D.C. | \nBennett and B. Sawyer.\n\nA Laboratory Model Magnetic Drum Trans- \nlator for Toll Switching Offices by F. J. \n_ Buhrendorf, H. A. Henning and O. J. Murphy. \nTables of Phase of a Semi-Infinite Unit At- \n| tenuation Slope by D. E. Thomas.\n\nplan. This was made possible by new centralized \nautomatic message accounting and toll features in \nthe crossbar tandem switching system which will \nultimately provide service for 10,000,000 customers \nin about 100 step-by-step cities.\n\nThe University Tandem office serves 31 local \ncentral offices in the San Diego area, It is a pri- \nmary outlet in the nationwide dialing plan.\n\nThe Laboratories, at the request of the National \nAcademy of Sciences, presented a number of dem- \nonstrations at the annual meeting of the Academy \nheld in Washington, D. C., recently. Among these \nwere demonstrations that showed the applications \nof new ferroelectric memory devices.\n\nOne of these was a ferroelectric shift register\u2014 \na device in which stored binary digits can be simul- \ntaneously moved from one cell to another without \ninterfering with each other. This apparatus can \nperform a number of basic operations that make \nit useful for data processing applications, such as \ndigital computing and automatic switching. These \ninclude; data storage; pulse counting; the ability \nto transfer information between devices with widely \ndifferent pulse rates; and the ability to accept pulses \none-at-a-time, store them until the sequence is com- \npleted, and then transfer them all-at-once.\n\nTo permit definite planning to meet the Labora- \ntories\u2019 needs for more working space, the Board of \nDirectors has authorized the preparation of plans \nand specifications for major building facilities to \nbe constructed on the Laboratories\u2019 property at \nHolmdel, New Jersey.\n\nPresently authorized construction, including the \nproposed office building and the single-story ware- \nhouse-type building at Murray Hill and additional \nspace at Western Electric locations, are expected to \nprovide for space needs through 1958, The Labora- \ntories is now authorized to prepare detailed plans \nand specifications for a building unit at Hoimdel \nto be started within the next year to provide for \nLaboratories space needs for a reasonable period \nbeyond 1958, It is expected that this first unit \nwould provide space for about 1,500 people. Also \nauthorized is the preparation of an over-all building \nlayout for Holmdel so that if future Laboratories \nspace needs justify it Holmdel could be built up\n\nJ. Reid Anderson, of the Laboratories Switching Research \nDepartment, demonstrates a ferroelectric shift register at \nthe annual meeting of the National Academy of Sciences.\n\nunder an orderly and efficient plan to become a \nmajor location about the size of Murray Hill.\n\nPreliminary studies indicate that Holmdel is a \nhighly desirable location for Laboratories purposes. \nThis location is near enough New York, Murray \nHill and Whippany for necessary inter-location \ntravel, yet sufficiently separated to avoid overcon- \ncentration in any one area. It is believed undesir- \nable to expand sizably beyond present plans at \neither Murray Hill or Whippany. Similarly, there \nappears to be good reason for not enlarging the \nNew York location. Transportation, concentration, \navailability of suitable manpower, recruiting of pro- \nfessional people, accessibility of residential areas, \nand many other factors have been we ighed in the \ndecision to enlarge at Holmdel.\n\nHolmdel, where there are now about 130 Labora- \ntories employees, is about 25 miles from Murray \nHill and about 45 minutes in time by automobile \non the Garden State Parkway. It is only a little \nfarther from Whippany and about an hour from \nNew York. There is also rail and bus transportation \nfrom New York. The Laboratories now owns about \n450 acres at this site.\n\nThe authorization given by the Board of Di \nrectors includes funds for: surveying, engineer- \ning and architectural services necessary to plan the \nlong-term, over-all program; engineering and archi- \ntectural services to produce complete drawings \nand specifications for the first building unit; and \ndrilling of wells at Holmdel required to prove the \nadequacy of the water supply to meet the needs of \nthe program. It is expected that the planning can \nbe completed in time to start construction of the \nfirst step in the project late in 1957 or early 1955.\n\nPatents Issued to Members of Bell Telephone \nLaboratories During March\n\nBrooks, C, E., McGuigan, J. H., and Murphy, O. J. \u2014 Mag- \nnetic Drum Dial Pulse Recording and Storage Registers \n2,738,382.\n\nFox, A. G. \u2014 Wave-guide Impedance Elements \nGrisdale, R. O., and Sauer, H. A. \nConducting Coils \u2014 2,739,371. \nHolden, W. H. T. \u2014 Station Number Identifier \nHoutz, C. C., and McLean, D. A. \ntrolytic Capacitors \u2014 2,739,275, \nHunt, L. F., and Schafer, 5. P. Method and Apparatus \nfor Detecting and Correcting Amplitude Distortion\n\nNewby, N. D. \u2014 Magnetic Scanning Arrangement Providing \nCompensation for Battery Variation and Variation of \nOther Components \u2014 2,739,183.\n\nNewby, N. D. \u2014 Rotational Use of Register Circuits in Tele- \nphone Switching Systems \u2014 2,740,003.\n\nWallace, R. L., Jr. \u2014 Transistor Amplifiers and Circuit Ar- \nrangements Therefor \u2014 2,739,190.\n\nFollowing is a list of the authors, titles, and place of publication \nof recent papers published by members of the Laboratories.\n\nAhearn, A. J., and Law, J. T., Russell Effect in Silicon and \nGermanium, |, Chem, Phys., Letter to the Editor, 24,\n\nBashkow, T. R., DC Graphical Analysis of Junction Tran- \nsistor Flip-Flops, Comm, and Elec., 23, pp. 1-6, Mar., \n1956,\n\nBennett, W. R., Synthesis of Active Networks, Proc, Symp, \nModern Network Synthesis, MRI Symposia Series, 5, \npp. 45-61, 1956,\n\nBennett, W. R., Electrical Noise. Il \u2014 Noise Generating \nEquipment, Electronics, 29, pp. 134-137, Apr., 1956. \nBommel, H. E., Mason, W. P., and Wainer, A. W., Dislo- \ncations, Relaxations and Anelasticity of Crystal Quartz,\n\nBozorth, RK. M., Quelques Propri\u00e9t\u00e9s Magn\u00e9tiques, Elec- \ntrioques Et Optiques Des Films Obtenus Par Electrolyse \nEt Par Evaporation Thermique, Le |. De Physique Et \nLe Radium, 17, pp. 256-262, Mar., 1956.\n\nClogston, A. M., Suhl, H., Walker, L. R., and Anderson, \n?. W., Possible Source of Line Width in Ferromagnetic \nResonance, Phys. Rev,, Letter to the Editor, 101, pp. \n903-905, Jan. 15, 1956.\n\nDe Leeuw, K., Moore, E. F., Shannon, C, E., and Shapiro, \nN., Computability by Probabilistic Machines, Automata \nStudies, Princeton Univ. Press, pp. 183-212, Apr. 1956,\n\nHaynes, J. R., and Westphal, W. C., Radiation Resulting \nfrom Recombination of Holes and Electrons in Silicon, \nPhys. Rev., 101, pp. 1676-1678, Mar. 15, 1956.\n\nLaw, J. T., and Francois, E. E., Adsorption of Gases on \nSilicon Surface, J. Chem. Phys., 60, pp. 353-358, Mar., \n1956,\n\nLloyd, S. P., and McMillan, B., Linear Least Squares Fil- \ntering and Prediction of Sampled Signals, Proc. Symp., \nP.1.B., 5, pp. 221-247, Apr. 1955.\n\nMoore, E. F., Gedanken-Experiments on Sequential Ma- \nchines, Automata Studies, Princeton Univ. \n129-153, Apr., 1956.\n\nRemeika, J. P., and Merz, W. ]., Guanidine Vanadium \nSulfate Hexahydrate: A New Ferroelectric Material, Phys. \nRev., Letter to the Editor, 102, p. 295, Apr. 1, 1956,\n\nSullivan, M. V., and Eigler, J. H., Five Metal Hydrides \nas Alloying Agents on Silicon, |. Electrochem, Soc., 103, \npp. 218-220, Apr., 1956.\n\nTurner, D. R., The Anode Behavior of Germanium in Aque- \nous Solutions, |. Electrochem. Soc., 103, pp. 252-256, \nApr., 1956.\n\nUhlir, A., Jr., High-Frequency Shot Noise in PN Junctions, \nProc, LR.E., Correspondence, 44, pp. 557-558, Apr., 1956.\n\nVan Haste, W., Statistical Techniques for a Transmission \nSystem, Comm. and Elec., 23, pp. 50-54, Mar., 1956,\n\nVan Haste, W., Component Reliability in a Transmission \nSystem, Elec. Engg., 75, p. 413, May, 1956.\n\nVan Roosbroeck, W., Theory of the Photomagnetoelectric \nEffect in Semiconductors, Phys. Rev., 101, pp. 1713-1724, \nMar. 15, 1956.\n\nWahl, A. J., and Kleimack, |. J., Factors Affecting Relia- \nbility of Alloy Junction Transistors, Proc. LR.E., 44, pp. \n194-502, Apr., 1956,\n\nWeisbaum, C., and Boyet, H., A Double-Slab Ferrite Field \nDisplacement Isolator at 11 KMC, Proc, LBRLE., 44, pp. \n554-555, Apr., 1956.\n\nDuring April, a number of Laboratories people gave talks before professional and \neducational groups. Following is a list of speakers, titles, and places of presentation.\n\nBiondi, F. J., Status of Semiconductor Technology. \nGuldner, W. G., see Thurmond, C. D.\n\nHaynes, J. R., Radiation Resulting from the Recombination \nof Holes and Electrons in Silicon.\n\nLaw, J. T., The High Temperature Oxidation of Germanium, \nSmith, K. D., see Veloric, H. S.\n\nThurmond, C. D., The Distribution of Copper Between Ger- \nmanium and Ternary Melts Saturated with Germanium, \nThurmond, C. D, Guldner, W. G., and Beach, A. L., Mydro- \ngen and Oxygen in Single Crystal Germanium as Deter\n\nAnderson, J. R., Ferroelectric Devices and Some Circuit Ap- \nplications, R.C.A. Research Laboratories, Princeton, N. J.\n\nAnderson, O. L., The Structure of Inorganic Glass in View \nof Recent Measurements on Volume Flow, Brooklyn Poly- \ntechnic Institute, Chemistry Symposium, New York City.\n\nAnderson, P. W., and Suhl, H., Nonlinearities in Ferromag- \nnetic Resonance at High Power, Armour Research Founda- \ntion, Relaxation Phenomena in Ferromagnetic Materials \nSymposium, Chicago, Il.\n\nBaker, W. O., Materials for Future Electronics, New Eng- \nland Radio-Electronics Meeting, Boston, Mass.\n\nBangert, J. T., The Transistor as a Network Element, \nA.LE.E. \u2014 LR.E., Lehigh University, Bethlehem, Pa.\n\nBiggs, B. S., Effect of Air Pollution on Elastomeric Products \nin the Telephone Plant, Detroit Rubber and Plastics Group, \nDetroit, Mich.\n\nBudlong, A. H., Mechanized Intelligence, Montclair Society \nof Engineers, Upper Montclair, N. J.\n\nCioffi, P. P., Rectilinearity of Electron Beam Focusing Fields \nfrom Transverse Component Determinations, A.LE.E., \nGreat Lakes District Meeting, Fort Wayne, Ind.\n\nClogston, A. M., Disorder Scattering of Spin Waves, Armour \nResearch Foundation, Relaxation Phenomena in Ferromag- \nnetic Materials Symposium, Chicago, IIL.\n\nDillon, J. F., Jr., Motion of Single Domain Walls in Man- \nganese Ferrite, American Physical Society, Washington.\n\nFay, C. E., Ferrite-Tuned Resonant Cavities, Microwave \nProperties and Applications of Ferrites Symposium, Har- \nvard University, Cambridge, Mass.\n\nFerrell, E. B., The Control Chart \u2014 Modifications and Ex \ntensions, American Society for Quality Control, Montreal \nSection, Canada.\n\nFinch, T. R., Science and Tomorrow's Communications, lowa \nState Science Fair, Cedar Falls, lowa.\n\nFoster, F. G., Preparation Methods for Materials Microscopy, \nMetropolitan Microchemical Soc iety, American Museum \nof Natural History, New York City.\n\nFrost, G. R., Logic of Switching Circuits, Queens College, \nPhysics Club, Flushing, N. Y.; and Columbia University, \nNew York City,\n\nFthenakis, E., A Voltage Regulator Using High Speed of \nResponse Magnetic Amplifiers with Transistor Driver, \nA.LE.E LR.E.\u2014LS.A., Special Technical Conference \non Magnetic Amplifiers, Syracuse, N. Y.\n\nGalt, J. K., Ferromagnetic Domain Wall Motion, Physics \nColloquium, New York University, New York City.\n\nGeballe, T. H., Thermomagnetic Effects in Germanium, Uni- \nversity of Pennsylvania, Philadelphia.\n\nGilleo, M. A., Magnetism in the Perowskite Structure System, \nSolid State Physics Seminar, Physics Department, Uni \nversity of Pennsylvania, Philadelphia.\n\nGoldstein, H. L., and Lowell, RK. J., Magnetic Amplifier Con- \ntrolled Regulated Rectifiers, AA.E.E. \u2014 LBB. \u2014 LS.A,, \nSpecial Technical Conference on Magnetic Amplifiers, \nSyracuse, N. Y.\n\nGordon, J. P., The Maser, LR.E. Long Island Section, Garden \nCity, L. L; and Union of Radio Scientific Internationale \nConvention, Washington, D. C\n\nHagelbarger, D. W., SEER\u2014A Sequence Extrapolating \nRobot, Seminar on Digital Computers, Joint Student \nBranch, A.LE.E, \u2014 LBLE., Polytechnic Institute of Brook \nlyn, New York City; and A.LE.E. \u2014 LBLE., Joint Student\n\nHamming, R. W., and Rhodes, Mrs. Ida, General Use of \nDigital Computers, General Services Administration Audi- \ntorium, Washington, D. C.\n\nHardy, F., Lubricants and How to Use Them, A.LE.E. New \nYork Section, New Jersey Division, Newark,\n\nHaynes, J. K., Radiation Resulting from the Recombination \nof Holes and Electrons in Silicon, Massachusetts Institute \nof Technology, Physics Department Colloquium, Cam- \nbridge, Mass.\n\nB Sample; Picking the Best of K Binomial Processes; and \nComparison of Two Univariate Distributions, Army Chem- \nical Center, Edgewood, Md.\n\nKock, W. E., Undergraduate Honors \u2014 Precursors to Suc- \ncess, 50th Anniversary Celebration of Cooperative Engi- \nneering, Honorary Fraternities, University of Cincinnati, \nOhio; and Polarized Sound Waves, Physics Colloquium, \nBrown University, Providence, R. 1.\n\nKudlich, R. A., TRADIC \u2014 A Transistor Digital Computer, \nLR.E., Akron Chapter, Ohio.\n\nLandgren, C. R., Transistors \u2014 Their Development and Ap- \nplications, John Ericsson Society, Engineer's Club, New \nYork City,\n\nLewis, W. D., and Vaughan, H. E., Electronic Switching and \n\u20ac Data Transmission, American Management Association, \nIntegrated Data Processing Seminar, New York City.\n\nphone Company, New York City. \nLogie, J. R., Jr, NIKE 1 \u2014 A Guided Missile System for AA \nDefense, \u20141.R.E., Student Activities Day, East- \nerm District Papers Competition, Rutgers University, New \nBrunswick, N.J.\n\nLong, T. R., The Elastic Constants of Magnesium and Mag- \nnesium Alloys, National Bureau of Standards, American \nPhysical Society Meeting, Washington, D.C.\n\nMattson, R. H., Properties of Junction Transistors, IRE, \nBoston Section, Mass.\n\nMcDonald, H. S., Signal Theory Applied to Speech Proces- \nsing, Johns Hopkins University, Electrical Engineering \nSeminar, Baltimore, Md.\n\nMcKay, K. G., The Interaction Between Research and De- \nvelopment at Bell Telephone Laboratories, A.AE.E. \nPittsburgh Section, Pa,\n\nMealy, G. H., Deterministic and Probabilistic Prediction \nTheir Relation to the Theory of Automata, Columbia Uni- \nversity, Discrete Sequence Transducers Seminar, New \nYork City,\n\nMonro, S., Design of Experiments, Lehigh University, Grad- \nuate Seminar, Bethlehem, Pa.\n\nMorgan, S. P., Control of Spurious Modes in Multimode \nWaveguides by Use of Foam Dielectric Inserts, Union of \nRadio Scientific Internationale Meeting, Washington, D.C.\n\nPaul, C. E., NIKE I \u2014 A Guided Missile for AA Defense, Sea \nExplorer Scouts, Ship 20, Livingston, N. J.\n\nRead, W. T., Jr., Dislocation Theory, Maryland Institute of \nMetals, Baltimore; and Solid State, Conference on New \nDevelopments in Engineering, University of Pennsylvania, \nPhiladelphia,\n\nReiss, H., Chemical Interactions Among Defects in German- \nium and Silicon, General Electric Company, Knolls Labora- \ntory, Schenectady, N.Y.\n\nRiesz, R. P., Mr. Meticulous, A.L.E.E. New York Section, \nCommunications Divisions, New York City.\n\nRowe, H. E., and Manley, J. M., General Energy Relations in \nNon-Linear Inductors and Capacitors, Non-Linear Circuit \nAnalysis Symposium, Brooklyn Polytechnic Institute, New \nYork City,\n\nSchroeder, M. R., Information Theory and Speech, Siemens \nund Halske Zentrallaboratorium, Munich, Germany.\n\nSeidel, H., Anomalous Propagation in Ferrite-Loaded Wave- \nguide, Ferrite Symposium, Harvard University, Cambridge, \nMass.\n\nSharp, W. O., Submarine Cable Tube, Royal Arch Masons, \nCorinthian Chapter, Westfield, N. J.\n\nShive, J. N., The Bell Solar Battery, Council of Agricultural \nand Chemurgic Research, Chicago, Il; and Transistors \nand Solar Batteries, New York Telephone Company, New \nYork City.\n\nSparks, M., Chemistry in Modern Communications, Waynes- \nboro, Sweet Briar, Va; Greensboro, Wake Forest, Kinston, \nCharlotte, N. C.; Ashville, Columbia, 8. C.; and Tallahas- \nsee, Lakeland, Gainesville, Fla.\n\nSuhl, H., Non-Linear Behavior of Ferrites Under High Peak \nPowers, Microwave Properties and Applications of Ferrites \nSymposium, Harvard University, Cambridge, Mass.; and \nSpin-Wave Effects in Ferromagnetic Resonance, Physics \nColloquium, Columbia University, New York City.\n\nTukey, J. W., Mathematics, Statistics and Computers, Amer- \nican Statistical Society, Pennsylvania State Chapter, Penn- \nsylvania State University; and Society for Industrial and \nApplied Mathematics, Central Pennsylvania Section.\n\nVanUitert, L. G., The Dielectric Properties of and Conduc- \ntivity in Ferrites, Microwave Properties and Applications \nof Ferrites Symposium, Harvard University, Cambridge, \nMass.\n\nWaltz, M. C., The Time Behavior of Transistors in Useful \nCircuits, A.1.E.E., Winston-Salem, N. C.\n\nWarner, R. M., Jr., Survey of Transistors for High Frequency \nUse, LR.E. Pittsburgh Section, Mellon Institute, Pa.\n\nWilkinson, R. L, Some Queueing Theory for Engineers, \nAmerican Society for Quality Control, Metropolitan Sec- \ntion, N. Y. C.\n\nWinslow, F. H., Electrons in Organic Insulators, American \nChemical Society, Philadelphia Section, University of \nPennsylvania.\n\nYoung, J. A., and Morgan, S. P., Helix Waveguide, Union of \nRadio Scientific Internationale Meeting, Washington, D. C.\n\nNew radio relay systems for telephone and \ntelevision now in the making will employ an \ningenious device invented by Bell scientists. \nThe device, known as an \u201cisolator,\u201d senses \nwhich way microwaves are traveling in a wave- \nguide, and stops those going the wrong way.\n\nIn the new systems a klystron wave gen- \nerator sends signals through a waveguide to \nthe antenna. The klystron must be shielded \nfrom waves reflected back along the wave- \nguide by the antenna. The isolator stops re- \nflections, yet allows the transmitted signals to \ngo through clear and strong.\n\nThis isolator is a slab of ferrite which is \nmounted inside the waveguide, and is kept \nmagnetized by a permanent magnet strapped \nto the outside. The magnetized ferrite pushes \naside outgoing waves, while unwanted re- \nflected waves are drawn into the ferrite and \ndissipated. This \u201cfield displacement\u201d action \nresults from the interplay between micro- \nwaves and a ferrite\u2019s spinning electrons.\n\nThis is another example of how Bell Tele- \nphone Laboratories research works to im- \nprove American telephony and telecommuni- \ncations throughout the world.\n\nDr. S. Weishaum assembles an isolator which he \ndeveloped for use in a new microwave system. Dr. \nWeisbaum is a Ph.D. in microwave spectroscopy \nfrom New York University. He is one of many \nyoung men at Bell Telephone Laboratories ap- \nplying the insight of the physicist to develop \nnew systems of communication.\n\nthe isolator is a \nferrite slab. \nGeometric pattern \nis a carbon layer \nwhich dissipates \nreflected signals.\n\n4 At a radio relay \nstation an isolator \nassures one-way \ntransmission \nfrom the output \nof the amplifier \nto the antenna.", "title": "Bell Laboratories Record  1956-06: Vol 34 Iss 6", "trim_reasons": [], "year": 1956}
{"archive_ref": "sim_record-at-t-bell-laboratories_1957-12_35_12", "canonical_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1957-12_35_12", "char_count": 128894, "collection": "archive-org-bell-labs", "doc_id": 979, "document_type": "journal_issue", "id": "bella-qwen-pretrain-doc979", "record_count": 153, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/sim_record-at-t-bell-laboratories_1957-12_35_12", "split": "test", "text": "Propagation of Millimeter Waves Through the Atmosphere, \nA. B. Crawford and D. C. Hogg\n\nTHE COVER: Converging beams of red, blue and green light \nproduce image for Bell Laboratories\u2019 study of subjective as- \npects in color television. (See opposite page.)\n\nThe BELL LABORATORIES RECORD is published monthly by Bell Telephone \nLaboratories, Incorporated, 463 West Street, New York 14, N. Y., M. J. KELLY, \nPresident; M. B. LONG, Secretary and Treasurer. Subscriptions: $2.00 per\n\nEDITORIAL BOARD \nF, J. Singer, Chairman \nJ. A. Hornbeck \nF. A. Korn \nE. T, Mottram \nR. J. Nossaman \nW.E. Reichle \nA. H. White\n\nyear; Foreign, $2.60 per year. Checks should be made payable to Bell Labora- \ntories Record and addressed to the Circulati Manager. Printed in U. S. A.\n\nEDITORIAL STAFF \nG. E. Schindler, Jr., Editor \nW. W. Mines, Assistant Editor, Murray Hill \nA. G. Tressler, Assistant Editor, Murray Hill \nA. B. Hilton, Assistant Editor \nR. L. Shepherd, Production Editor \nT. N. Pope, Circulation Manager\n\nColor television bears out well the adage, \u201cbeauty is in the eye of the be- \nholder,\u201d since only the human eye can measure \u201csharpness.\u201d This purely \nsubjective quality of video depends on the many complex parameters of a \ntelevision system. Because Bell System transmission facilities are frequently \na part of such systems, a color television simulator has been devised at the \nLaboratories to help answer some of the basic questions about sharpness.\n\nFrom time immemorial man has made pictures \nin color. Before modern times all pictures were \nmade by hand, and any faults in form or color could \nbe attributed to the artist\u2019s lack of skill. As the \nscientific understanding of picture-making in- \ncreased, this better understanding led to the intro- \nduction of new techniques. Complete freedom \nfrom hand reproduction became possible about a \ncentury ago, with the invention of photography. \nThis revolutionary advance brought with it, of \ncourse, new kinds of picture faults that required \nincreased scientific understanding. Photography \nhas progressed rapidly \u2014 nowadays we make pic- \ntures that move, we make them in full color, and \nmost of us take them quite for granted.\n\nThe youngest of the arts of picture-making is \ncolor television. This new art has released us from \nsome of the limitations imposed by the dyes and \nprocesses used in color photography, but not with- \nout cost. The very complexity of television appa- \nratus results in a host of new picture faults to be \nunderstood and overcome.\n\nTo increase our understanding of some of these \nfaults, the Visual and Acoustics Research Depart- \nment at Bell Laboratories has for some time used \na \u201cdefocusing\u201d projector. The chief function of this \nprojector is to make color pictures that are meas- \nurably blurred. This unusual research tool can \nalso produce a full color image that looks like the \npicture one might see on a color television receiver. \nThe blurring that results from out-of-focus projec- \ntion can then be used to simulate the blurring that \nresults when a color television picture is trans- \nmitted over a circuit of limited bandwidth.\n\nIn the future, the Bell System may be called \nupon to transmit television signals which will pro-\n\nduce color pictures sharper than those presently \nbroadcast for home entertainment. By being able \nto simulate such color television pictures right now, \nit is possible to study some of the factors which \ninfluence sharpness.\n\nColor television pictures are produced by a three- \ncolor additive process. The picture tube produces \na separate and complete image in each of the pri- \nmary colors \u2014 red, green and blue \u2014 and these three \npictures are superimposed on the screen of the tube. \nThe controlled addition of three colors at every \npoint on the screen can produce whatever color is \nrequired at each point to build up a full color \nimage on the viewing screen.\n\nThe defocusing projector, shown in Figure 1, \nalso makes color pictures by an additive process,\n\nFig. 1\u2014 Front view of the defocusing projector \nshowing the four individual projector lenses.\n\nFig. 2\u2014 The vertical reference lines on the screen show that these two pictures are the same size.\n\nas a proper simulator should. The projector, or \nmore accurately, the projection machine, is actually \ncomposed of four projectors, with independent \nfocusing adjustments, directed at a common pro- \njection screen. The machine projects separate pic- \ntures, one in red light, one in green light and one \nin blue light onto the screen, where the additive \nprocess takes place. The viewer sees only the sum \nof these three pictures, or a full color image. The \nfourth projector is available for showing a com- \nparison picture if needed.\n\nA basic requirement, when separate images are \nprojected to make a color picture, is that the sepa- \nrate images remain in good register on the screen. \nThe machine would not be useful for sharpness \nstudies if the effects of focus adjustment were to \nbe contaminated by misregistration. In other words, \nimage size and centering must both remain fixed \nwhile the focus changes.\n\nThe optical principle used to maintain constant \nmagnification with variable focus is a simple one. \nTaking an ordinary home slide projector as an ex- \nample, we know that the picture gets larger as \nthe projector is moved farther from the screen. We \nalso know that the picture goes in and out of focus \nas the lens is moved back and forth on the pro- \njector. Likewise, movement of the lens causes the \npicture to change size. If one could move the whole \nprojector back and at the same time move the lens \nforward on the projector by the proper distance, \nhe could keep the picture from changing size. The \nincrease in size due to moving the projector back\n\nwould cancel out the decrease in size due to moving \nthe lens forward, yet the picture would go out of \nfocus. Figure 2 illustrates projections of the same \nslide and shows that the two images are exactly the \nsame size, even though the picture on the right is \nobviously out of focus.\n\nTo make possible the precise motions and opti- \ncal relationships necessary to get a constant-size \npicture of variable focus, a well-known mechanism\n\nFIXED MEMBER \nFig. 3 \u2014 Pantograph maintains spacing of projector \nand lens for in- and out-of-focus projection.\n\n\u2014the pantograph\u2014is used. The pantograph, \nwhich is simply a geometric parallelogram with \npivots at the corners, provides exactly the motions \nrequired. Figure 3 shows how the pantograph was \nadapted to this use. The pantographs as they ap- \npear in the simulator are shown in Figure 5. A bel- \nlows, seen between the slide carrier and the lens, \naccommodates the relative motion of these parts \nas the projector moves in its focusing and defocus- \ning action. The bellows stretches about % of an \ninch when the whole projector moves back two \ninches. The two projectors on the other side are \nmounted in the same way.\n\nThe four projectors are mounted so that the pro- \njection lenses and lantern slides are all squared \nonto, or held parallel to, the screen. The centers \nof the lenses are closer together than the centers \nof the slides. The resulting projection is known\n\n\u2018ig. 4 \u2014 The observer, E, Averbach, tells the operator, G. \nNielsen, when he thinks the sharpness of this picture \nnatches that of a comparison picture.\n\nas off-axis projection, because the center of the \nimage is a few degrees off the optical axis of each \nlens. This kind of projection insures perfect regis- \ntration of the images on the screen at all times and \neliminates \u201ckeystone\u201d distortion \u2014 an image irregu- \nlarity that would result if the projectors were not \nsquared onto the screen. The keystone effect, to \ndescribe it simply, is the tapered shape a rectangu- \nlar picture assumes when projected at an angle. The \nslightly tapered keystone shape, which is ordi- \nnarily not objectionable, would make image regis- \ntration impossible. The illuminating systems are \n\u201ctoed-in\u201d so that they all point toward the center \nof the screen to improve the uniformity of coloring \nover the field.\n\nFig. 5\u2014 Side view shows top pantograph as it \nactually looks in the projection machine.\n\nThis machine performs well because all of its \nparts have been made and aligned with precision. \nThe lengths of corresponding sides of the panto- \ngraphs differ by less than one thousandth of an \ninch. The lantern slides and the lenses are squared \nonto the screen to better than five minutes of arc. \nThe in-focus magnifications of the four projectors \nare equalized to better than one part in 2,000. \nAs a result, the separate images on the screen, \nwhether in sharp focus or not, are in register to \nwithin the limits of unaided vision.\n\nA most important part of this projection machine \nis its control system. Each of the four projectors \nis actuated by small hydraulic controls which move \nthe projectors slowly and smoothly and hold them \nfirmly in position. Small indicating units monitor \nthe individual projector positions and transmit \nthese positions to indicator dials mounted on the \ncontrol panel.\n\nThe indicator dials, in effect, relate the focus \ncondition of individual projectors to the sharpness \nof the color picture on the screen. The sharpness \nof a three-color additive picture \u2014 a purely subjec- \ntive quality \u2014 can only be judged by an observer. \nIn a typical experiment, the observer examines a \npicture in the \u201ctest condition\u201d (the three pro- \njectors out of focus by different amounts) and then \nexamines a picture in the \u201cgauging condition\u201d (the \nthree projectors out of focus by the same amount). \nThe observer then adjusts the gauging condition \nto be equally as sharp as the test condition. Fig-\n\nure 4 shows the projector as it is used with an \nobserver. The indicator dial readings called for \nby the observer in making this sharpness match are \nthen recorded by the operators and can be trans- \nlated into bandwidth information by a straight- \nforward calibration procedure.\n\nFig. 6 \u2014 Equivalent bandwidths of a 10-mc, three- \ncolor picture show the line of maximum sharpness.\n\nA three-dimensional plot of the sharpness-match \ndata from ten observers, statistically analyzed and \nput into terms of bandwidth, is presented in Fig- \nure 6. The percentages of the total band of ten \nmegacycles allotted to blue and green are shown \non the base plane. Red would, of course, occupy \nthe remainder of the band. The equivalent band- \nwidth of the gauging picture (all three projectors \nfocused identically ) is plotted vertically. All of the\n\npoints on the vertical, or equivalent-bandwidth, \naxis form the top surface of the plot. Wherever \nthe surface lies above ten megacycles, the color \npicture looks sharper than would the picture trans- \nmitted by a ten-megacycle band equally appor- \ntioned. The maximum sharpness \u2014 equivalent to a \n15.6-megacycle picture \u2014 occurs when 63 per cent \nof the total band is allotted to green. In other \nwords, a ten-megacycle test picture (each pro- \njector out of focus by a different amount) with \n6.3 megacycles devoted to green looks just as sharp \nas a 15.6-megacycle picture with 5.2 megacycles \ndevoted to each color. It is also interesting to note \nthat the remaining 37 per cent of the total band \ncan be allotted to red and blue in any proportion.\n\nMany experiments of this type have been per- \nformed by engineers at the Laboratories and else- \nwhere, some of them earlier than this one, and \nthey all point toward the same conclusion: tele- \nvision in three colors can be made to look just as \nsharp as television in monochrome without paying \nthe price of three times the bandwidth. Such in- \nformation is the foundation of the latest system of \ncolor television transmission adopted by the Fed- \neral Communications Commission.\n\nIn practical terms, television researchers want \nto know whether the color picture sent by one \ntelevision system will look sharper than that sent \nby another. This color television simulator is help- \ning to fit together some of the pieces of this larger \npuzzle. In addition to increasing our understanding \nof some of the complex problems involved in the \ntransmission of color television signals, this pro- \njection machine has also made some valuable con- \ntributions in the field of visual acuity in color.\n\nM. W. Ba.pwin, Jr., a native of Portland, Maine, received the E.E. degree \nfrom Cornell in 1925 and the M.A. degree from Columbia in 1928. He joined \nBell Laboratories in 1925, and has been associated with the Research Department \nsince that time. His primary concern has been television research, with particular \nemphasis on subjective aspects. During the war he was involved in the develop- \nment of an automatic tracking anti-aircraft radar. He is a member of Eta Kappa \nNu and of Tau Beta Pi. He is a Fellow of the I.R.E., and is currently Chairman \nof its Standards Committee.\n\nPrivate Branch Exchange (PBX) customers differ widely in their telephone \nrequirements \u2014 some have only a few extensions and others have thousands \nof extensions in hotels, department stores, and large industrial establish- \nments. Most businesses, however, require less than sixty extensions, and de-\n\nmand has been increasing for a modern dial PBX system to serve such \ncustomers. The new 756A PBX is a small \u201cpackaged\u201d design that matches \nthe decor of modern offices. In addition to the conventional PBX services, \nthe 756A provides a \u201ccamp-on\u201d feature to permit automatic connection\n\nTo most people, a PBX is a telephone switch- \nboard, frequently located in or near the reception \nroom of a business establishment and used for \nservicing calls over the company\u2019s telephone exten- \nsions. In a dial PBX, however, the switchboard is \nonly one part of the total equipment; much of the \nwork of interconnecting telephones is done auto- \nmatically by electromechanical switches. The \nswitchboard in a dial installation might thus be \nconsidered as a sort of \u201ccontrol panel,\u201d with the \nrelays, switches, and other equipment out of sight.\n\nIn small manual PBX systems, much or even all \nof the \u201cbehind-the-scenes\u201d equipment can be \nmounted within the switchboard itself. On the \nother hand, even a small dial system requires \nswitching equipment and power equipment \nmounted in additional cabinets. As a consequence, \nthe installation of present-day systems occasionally \noffers difficulties because a particular building may \nhave narrow doors, too small an elevator. or in- \nsufficient floor bracing. These problems have been \nsolved by the new 756A crossbar dial PBX devel- \noped at the Laboratories.\n\nMany telephone customers requiring dial PBX \nservice have between 20 and 60 extension lines and\n\n10 or fewer central-office trunks, and the most \neconomical arrangement for serving this field is \nto provide only one or two \u201cpackaged\u201d systems. \nThe 756A is a small packaged design supplied in \ncapacities of either 40 or 60 lines and equipped for \n6 central-office trunks. Four additional central-office \ntrunks can be added as desired. The basic switching \ncomponents are crossbar switches and wire-spring \nrelays. The 756A is also the first equipment to use \nthe recently-developed \u201c2-in-1\u201d wire-spring relay \nthat provides two relays in the space normally re- \nquired for one.\n\nPhysically, the entire PBX is housed in two modu- \nlar cabinets of steel and aluminum (Figure 1). \nThe depth and height are similar to those of \nstandard office filing cabinets, and all maintenance \nis done from the front. Each module contains \nthree small relay racks arranged so that they can \nbe pulled forward out of the module like vertical \nfile drawers. Each rack mounts on ball-bearing \ntelescoping slides, Interlocking latches prevent more \nthan one rack being pulled out at a time and, \nsince each weighs aboui the same, the combined \nweight of the other two racks is sufficient to keep \nthe cabinet steady. For this reason there is no\n\nThe PBX cabinets are designed for mounting in \nregular office space, along with file cabinets or other \noffice furniture. Since an extended rack projects no \nfurther than a standard file drawer, the aisles do \nnot need to be widened. The cabinets are of modern \nstyling, finished in a shade of beige-gray that has \nbecome popular in office furniture. Sound-absorbing \nmaterial reduces the noise of the switching equip- \nment. The PBX is connected to the office wiring \nby plugs and pre-arranged jacks.\n\nA total of three equipment modules are provided, \nfrom which the twin-module 40-line or 60-line 756A \nmay be constructed. One module is common to both \n40 and 60 lines, and the second contains additional \nequipment for either the 40-line or the 60-line sys- \ntem. A new modern attendant\u2019s console, shown in the \nillustration at the head of this article, has sufficient \ncapacity to handle all ordinary calls requiring the \nassistance of an attendant. This console features \npush-button keys and simplified operation. Since \nit is assumed that the attendant will have other \nduties as a secretary or receptionist, the console is \nsmall enough for table- or desk-top mounting.\n\nSmaller systems may use a six-button telephone \nset instead of the console. With some installations \na switchboard may be more desirable than either \nthe console or the six-button set. In such cases the\n\nThe power supply is completely self-contained, \nand is mounted in the first or basic module. Com- \nmercial 115-volt, 60-cycle ac is stepped down in \nvoltage, rectified, and filtered to produce 48 volts \nde for use in the PBX. The filter contains a large \ncapacitor that stores sufficient electrical energy to \nmaintain the voltage during momentary interrup- \ntions of the commercial supply. Where interruptions \nare apt to be of appreciable length, an additional \nmodular cabinet can be supplied. This cabinet \ncontains storage batteries as a reserve power source, \nand a rectifier keeps the batteries charged. Ringing \ncurrent, signaling tones, and flashing interruptions \nare generated by equipment in the power-supply \nunit regularly furnished.\n\nWhile most present-day dial PBX\u2019s use step-by- \nstep switches, the 756A achieves small size and \nfaster operation by using crossbar switches and \ncommon-control principles. It operates essentially \nas a small crossbar office; register circuits connect \nto calling lines and trunks to supply dial tone and \nreceive called numbers. The called numbers are \npassed to a marker which determines the busy and \nidle state of called points, performs an idle-hunting \nfunction for grouped lines and trunks, and controls \nthe establishment of talking connections. The cross- \nbar switches establish all connections.\n\nAlthough only one marker is supplied, it actually \ncontains two independent marker channels, used \nalternately on successive calls. Should trouble be \nencountered, the marker \u201ctimes out\u201d and makes a \nsecond trial using the other channel. The two regis- \nters are also used alternately unless one is busy. \nBusy tone is normally supplied by a busy-tone \ntrunk but, should this trunk itself be busy, the \nmarker operates a relay in the appropriate register, \nand the register supplies busy tone. Troubles are \nindicated by lamp displays.\n\nAll conventional PBX services are available with \nthe 756A. Calls to extensions are placed by dialing \ntwo digits; with two to seven as the number in \nthe initial digit. Calls requiring the attendant are \nplaced by dialing zero. Dialing an initial 9 causes \nconnection to an idle central-office trunk and the \ncalling station line receives a second dial tone \nfrom the central office. Where certain extensions \nare restricted from direct outside dialing, users \nmust dial 0 for the attendant and ask her to get \nthe number. If they dial 9, they will be intercepted \nand transferred to the attendant. Also, certain ex- \ntensions, while not restricted from dialing local \noutside calls, will also be intercepted if used to dial\n\nInward calls from a central office first cause \nthe attendant to be signaled. She answers a trunk \nby its key appearance on the console and asks for \nthe desired called party. She then operates a key \nto summon a register which furnishes dial tone, \nand then either keys or dials the desired extension \nnumber into the PBX equipment. The trunk is \nconnected to the desired station which receives \nringing current, and subsequent talking battery \n_from the central office. The attendant may split the \n\u2018connection when desired until after conversation \nwith the called party begins. After the attendant \nreleases her keys, she may be recalled by the called \nparty operating his switch-hook momentarily. Dis- \nconnects take place automatically without involving \nthe attendant.\n\nA \u201ccamp-on\u201d feature has been included in the \n756A for inward calls from a central office. If the \ncalled station is busy, the attendant is so notified and \ncan advise the calling party that he may wait if \nhe wishes. After the attendant releases from the \ncall, the trunk will \u201ccamp-on\u201d until the line is \nfree. As soon as the first call disconnects, the trunk \nis automatically cut through to the called line and \napplies ringing current. No further assistance is \nrequired from the attendant. This feature reduces \nwaiting time for inward calls to busy lines and \nfrees the attendant for other duties.\n\nThe central office provides talking current for \nboth inward and outward calls, and also provides \ncurrent to ring the PBX attendant on inward calls. \nFor intra-PBX calls \u2014 those between two extensions \n\u2014ringing and talking current are supplied by the\n\nFig. 2\u2014 The 756A PBX circuitry being inspected \nat the Cleveland, Ohio, trial installation.\n\ntrunk, or junctor, is brought in on this type of call \nto supply the necessary currents and signals. Out- \nward trunks to other PBX\u2019s are reached by dialing \n8, and calls over these trunks are handled in the \nsame manner as outgoing calls.\n\nThis new PBX, modern in design and appearance, \nshould fill the telephone needs of today\u2019s smaller \nbusinesses. The ease of installation and mainte- \nnance, the adaptability of its size and modular \nconstruction to the requirements of modern offices, \nand the new features offered should make the \n756A system attractive both to customers and to the\n\nPBX. An additional circuit, called an \u201cintercom\u201d Operating Telephone Companies.\n\nO. H. Wiuirorp, whose home town is Greenwood, Miss., joined the Labora- \ntories in 1920 and initially engaged in laboratory and field testing of step-by- \nstep, panel, manual and No. 1 crossbar systems. During World War II, he was \nassociated with various military projects, and at the close of the war he \nbecame concerned with the development of the No. 5 crossbar system, particu- \nlarly in the design of the maintenance facilities. He was an active member of the \ngroup representing the Laboratories during the installation and cutover of the \nfirst No. 5 system in Media, Pa. Transferring to the Systems Engineering Depart- \nment, he prepared engineering information on the No. 5 system for the use of \nOperating Company engineers in planning central offices. He was active in the \nEnglewood DDD trial and thereafter in engineering studies related to the ex- \npansion of the customer dialing network. He is now in charge of a Special Systems \nEngineering group studying requirements for new PBX features and systems.\n\nA new concept in memory devices has recently \nbeen announced at Bell Laboratories. As a result of \nexploratory work by A. H. Bobeck of the Device \nDevelopment Department, it may now be possible \nto design memory systems that are simpler to fab- \nricate and more economical to manufacture than \nexisting systems. The new concept, which has been \ngiven the name \u201cTwistor\u201d, may find extensive appli- \ncation in computers and electronic switching sys- \ntems where rapid-access, high-capacity memories \nare necessary.\n\nSet-up for testing magnetic wire: I, supplies cir- \ncular field and I, in solenoid supplies longitudinal \nfield, resulting in helical magnetization of the wire.\n\nA. H. Bobeck tests memory characteristics of a \n\u201cTwistor\u201d device, which incor porates magnetic wires.\n\ncould be constructed in the familiar grid-like or \ncoordinate pattern. This is the way present mag- \nnetic-core memories are arranged \u2014 horizontal and \nvertical wires are interwoven, with a donut-shaped \nmagnetic core encircling each intersection. A pulse \nof energy along one wire is chosen to be insufficient \nfor magnetizing a core, but coincident pulses along \nboth a horizontal and a vertical wire will magnetize \nthe core in a circular direction. Pulses of opposite \npolarity magnetize the core in the reverse direction \naround the donut. In this way, one \u201cbit\u201d of informa- \ntion is stored at one coordinate point in the array.\n\nA \u201cTwistor\u201d array would be similar in appearance \nto the magnetic-core memory, but would have no \ncores. It would be constructed merely by interweav- \ning copper wires with wires made of a magnetic \nmaterial, much as window screen is woven.\n\nThe \u201cTwistor\u201d gets its name from a characteristic \nof magnetic wire. Normally, such wire can be mag- \nnetized most easily in the longitudinal direction (in \na straight line along its length). But if a torsional \nforce is applied to the wire, the magnetization will \nprefer to lie along a helical, rather than longitudinal, \npath. Torsion may not have to be applied to the \nmagnetic wire in a final device. The preference for \nhelical magnetization may be \u201cfrozen\u201d into the wire \nduring processing.\n\nIn a \u201cTwistor\u201d memory, the required helical field \ncan be obtained by applying coincident pulses of\n\ncurrent through the copper and magnetic wires. \nSince current causes a circular magnetic field \naround a conductor, the circular component is read- \nily supplied with a pulse of current through the \nmagnetic wire. The field resulting from current in \nthe copper conductor is also circular, but at an in- \ntersection where the copper wire touches the mag- \nnetic wire at right angles, this field has a component \nlongitudinal to the magnetic wire. Thus, to \u201cwrite\u201d \ninformation into 10 x 10 array, for example, current \ncould be applied to wire number 7 of 10 vertical \nmagnetic wires, and to wire number 3 of 10 hori- \nzontal copper wires. This would produce helical \nmagnetization in the magnetic wire in the immedi- \nate vicinity of the (7,3) coordinate point. In this \nway, a \u201cbit\u201d of information can be stored in the \n\u201cdirection\u201d of the helical magnetization.\n\nAlthough two pulses of current are needed to \n\u201cwrite\u201d information into a \u201cTwistor\u201d memory, only \na single pulse is required for \u201creadout\u201d. The current \nis applied to one of the copper wires so that it will \noverdrive the longitudinal field in the reverse direc- \ntion at any intersection where a bit is stored. A \n\u201creadout\u201d pulse of voltage will then be induced in \nthe associated magnetic wire. Because the lines of \nmagnetic flux along the helical path \u201cwrap\u201d the \nmagnetic wire many times, a favorable increase in\n\noutput signal is obtained. Thus, the magnetic wire \nis used both as the storage medium and as the \nsensing element.\n\nInvestigations are now under way to determine \noptimum size and composition for the magnetic \nwires. It appears that a conductor plated with mag- \nnetic material will be most useful. Diameters as \nsmall as one thousandth of an inch appear to be \nusable. At least 10 bits per inch may be stored along \nsuch a magnetic wire without adverse interaction.\n\nIn conventional magnetic-core memory devices, \nconductors must be threaded through the cores to \nmake up a suitable matrix. When a ferrite sheet is \nused, either a threading or a plating operation is \nnecessary for suitably locating the conductors. How- \never, with the \u201cTwistor\u201d, the ferrite material is com- \npletely eliminated and no threading or plating is \nnecessary. Speed of operation and output are com- \nparable to those of ferrite systems.\n\nPresent indications are that the drive circuits for \na \u201cTwistor\u201d array can be readily transistorized. \nThus, a memory system using the \u201cTwistor\u201d concept \nwill retain all of the advantages of ferrite-core or \nsheet systems, and will be much simpler and more \neconomical to fabricate.\n\nAn experimental version of the \u201cTwistor\u201d memory: information-storage section in center consists only of \nhorizontal magnetic wires and vertical copper wires. Magnetic wire serves for both storage and sensing.\n\nTo handle air traffic in the vicinity of air bases, the Air Force is installing \na number of Radar Approach Control (RAPCON) Centers. These Centers, \nwhich maintain high safety standards of aircraft control, require highly \nspecialized communications service, both internally and with other control \npoints. A modified version of the Bell System\u2019s 102A key equipment is \nbeing installed to provide safer, more efficient air-traffic service of this type.\n\nSince World War II the United States Air Force \nhas greatly extended the use of radar for con- \ntrolling military air traffic. Personnel operating a \nradar-control system are given a graphic display of \nthe location, direction of travel and speed of all air- \ncraft within the range of the radar set. To use the \nadvantages of radar methods, Radar Approach Con- \ntrol (RAPCON) Centers have been installed at \nmany Air Force Bases. These centers use radar \nfor departure and landing control of all traffic \nwithin the traffic pattern of the air base, and ex- \ntend this control from the visual limit of an airport \ncontrol tower to about 50 miles from the base.\n\nThe chief purpose of air-traffic control is to pro- \nvide safe separation of aircraft in the control area, \nespecially those under instrument flight rules. Mod- \nern aircraft compound the safety problem by virtue \nof the complexity of operation. The time element \nhas become increasingly important because many \njet planes now fly over 600 miles per hour. At this \nspeed, the aircraft is traveling one mile every six\n\nseconds, or ten miles per minute. Another important \nconsideration is that the fuel consumption of jet \naircraft is very high, particularly below 20,000 feet. \nPositive and definite landing preparations are \ntherefore necessary, since even a lapse of a few \nseconds in traffic-control procedures could mean \nthe difference between safety and disaster.\n\nThe success with which RAPCON Centers con- \ntrol these high speed aircraft depends to a large \ndegree on a rapid and efficient voice-communica- \ntion system. RAPCON Centers are furnished with \nthe No. 102A key equipment system. With this sys- \ntem, several attendants, or controllers, can establish \u00a9 \ndirect communication with any one of many internal \nand external lines. This key equipment was orig- \ninally designed for the Civil Aeronautics Adminis- \ntration to control civilian air traffic without the use \nof radar. Certain features have been added for the \nswitching and other arrangements necessary to \nmeet the operational procedures required by the \nRAPCON Center method of air-traffic control.\n\nNo. 102A key equipment was originally arranged \nfor one attendant at an operating position who had \naccess to telephone lines only. Two controllers \u2014 \na duty controller and an assistant \u2014 are normally \nrequired at each RAPCON Center position, and \nthey have a relatively large number of direct air- \nground radio contacts. Thus the two attendants \nmust be able to share the telephone facilities and \nto integrate closely the telephone and radio equip- \nment. They can independently use either the same \nor separate telephone and key circuits, and either \nthe same or separate radio equipment at the con- \nsole. Each attendant has a transfer key so that the \nsame telephone headset may be used on either tele- \nphone lines or Air Force-owned radio equipment; \nrapid transfer is possible from one to the other. The \nconvenience and savings in time gained by this ar- \nrangement are of great value in the over-all opera- \ntions. The transfer keys, line key units, and at- \ntendant\u2019s telephone-set jacks are mounted in the \nAir Force-owned consoles as shown in the illus- \ntration on page 490.\n\nRapid and positive intercommunication between \nthe operating personnel in a RAPCON Center is \nimperative. For example, when the controller in \ncharge of departures (departure controller) finds \nthat an aircraft has been inadvertently taxied onto \nan active runway, he must immediately contact the \ncontroller in charge of incoming traffic (final con- \ntroller) to warn him of the existing hazardous con- \ndition. The final controller can then direct any air- \ncraft on final approach to a temporary holding point \nto avoid a collision. An \u201coverride\u201d feature permits \nthis fast and positive intercommunication between \noperating personnel. It is arranged so that any at- \ntendant may have immediate communication with \nan attendant at any other position in the RAPCON \nCenter, even though the called attendant may be \ntalking on a radio channel or a telephone line.\n\nFigure 2 is a schematic diagram of the inter- \ncommunicating circuits between two positions. \nEither the duty or assistant controller, who share \nthe equipment at position No. 1, can override the \nassistant controller at position No. 2 by operating \nthe associated line key. This key connects the con- \ntroller\u2019s telephone circuit at position No. 1 \u2014 through \nan induction coil and an operated relay \u2014 directly \nto the telephone circuit of the assistant controller \nat position No. 2. Operation of the key also lights \nthe line lamp at the overridden position to inform \nthe controller that he has been overridden by posi- \ntion No. 1.\n\nWhen the assistant controller\u2019s transfer key at \nposition 2 is in the telephone position, a two-way\n\ntalking circuit is immediately established between \nthe controllers, whether or not the controller at \nposition 2 is connected to a telephone line. When \nthe key is operated for radio transmission, a relay \nopens the transmitting circuit from the controller \nat position 2 so that a one-way talking circuit is \nestablished from the controller at position 1 to the \nheadset receiver of the controller at position 2. \nThis circuit is not connected to the input of the \nradio transmitter because the message may be \nmisunderstood by the pilot and confuse him. \nIf the controller at position 2 wishes to talk to \nposition 1 over this circuit, he operates the key to \nthe telephone-lines position. The calling controller \nmay interrupt any conversation in progress. (Both \nsides of the called controller's conversation, either \nradio or wire, are heard by the overriding controller \nthrough the sidetone circuit.) Similarly, either the \nduty or assistant controller at position 2 can over- \nride the assistant controller at position 1.\n\nThe override feature is normally arranged so \nthat override calls can be made to only one (as- \nsistant controller) of the two attendants at a posi- \ntion. During light-load periods, however, the posi- \ntion may be operated by only one attendant (duty \ncontroller). For this case, the assistant controller \nhas a third position of his transfer key, as shown \nin Figure 2. This position is called \u201csingle operator.\u201d \nWhen the duty controller is alone at the position, \nthe assistant controller's transfer key is operated\n\nFig. 1\u2014 G. A. Giddings examines arrangement of \nmodified 102A key unit for RAPCON operation.\n\nto this single-operator position. This connects the \nassistant controller's receiver circuit to the duty \ncontroller's receiver circuit and extends the over- \ntide features to the duty controller.\n\nIt is very important to record the conversations \nof the attendants, especially when they talk to \npilots. A recorder jack panel is provided so that \nany one of the Air Force-owned continyously-run- \nning tape recorders may be connected to record\n\nall the radio and telephone conversations concern- \ning aircraft control.\n\nAll radio transmitters use a push-to-talk arrange- \nment and, to permit hands-free operation, this is \nusually done with a footswitch at the position. How- \never, this footswitch may be bypassed by operating \na footswitch-shorting key in the front of the console. \nWhen this is done, the radio transmitter is actually \nkeyed with the locking or nonlocking push-to-talk\n\nPOSITION NO. 1 POSITION NO. 2 \nASSISTANT \nCONTROLLER \n! \n-- \u2014 TRANSFER KEY \nSINGLE \n\u2018RADIO TELEPHONE OPERATOR! \n\u2014> \nINDUCTION | OVERRIDE \nCOIL OUCTION = RADIO ; H \nDUTY OR TA RELAY RELAY \nASSISTANT = \nCONTROLLER \n| RI \n= LINE ' \nINDUCTION \nTO af DUTY \n| TELEPHONE <= \nCONTROLLE \n| -- \nTRANSFER KEY \nRADIO TELEPHONE \ni < \n! \n! \n!\n\nFig. 2\u2014The override feature provides rapid communication between controllers in emergencies.\n\nLINE 3 \n! ) \n\u2014> TO OTHER LOUDSPEAKER \nSL RELAYS THROUGH \n_LINE 1 INTERMEDIATE \nT SL RELAYS \n/ \nLINE 2 / 5 \nLOUD- | <\u2014 y \nSPEAKER T \nTO OTHER \nSL RELAYS \nSe | OTHER POS \nASSOC WITH \n| SAME TRANS \nA INE \n| \nLINE LAMP LINE LAMP \nSWITCHHOOK |\n\nFig. 3 \u2014 During eS of heavy traffic or in an emergency, the \u201chot line\u201d arrangement makes possible \ndirect communication between the RAPCON Center and the airport control tower.\n\nswitch in the cord of the telephone headset. The \n53-type headsets used have a 12-foot retractile \ncord, which permits the attendant to move from \nhis position and still talk on radio channels. \nOne-way \u201chot\u201d lines (lines continuously open for \ntransmission ) are installed in the system to provide \nfast, direct and positive communication between a \nRAPCON Center and the associated airport control \ntower. This arrangement is needed in periods of \nvery heavy traffic, and especially so in case of an \nemergency. Normally one hot line runs from \nRAPCON to the tower and either two or three from \nthe tower to RAPCON. Figure 3 shows the special \ntermination arrangement for these hot lines. The \ntower controller can talk immediately to a position \nin the center by lifting his telephone handset and \noperating a line-selection key to the line associated \nwith the called position. The switchhook contacts in \nthe handset transfer the \u201creceive\u201d line from the \nloudspeaker to the handset receiver and connect the \ntalking battery to the handset transmitter. The con- \ntroller at position 1 in the RAPCON center can \ntalk immediately to the tower by operating a line \nkey which activates a relay associated with that \nposition. The relay performs four functions. (1) It \nlights the line lamps associated with the transmit \nline at that and all other positions to indicate that \nthe line is in use. (2) It transfers the receive line \nat that position from the loudspeaker to both the \nhead telephone set of the controller's receiver and \na bus circuit in the receive line to other relays asso- \nciated with the same receive line. (3) It connects \nthe controller's head telephone set transmitter to \nthe transmit line. And (4) it connects two resistors \nbetween the transmit line and the controller's re-\n\nceiver circuit to provide sidetone to the controller. \nIn a similar manner, the controllers at other posi- \ntions can talk to the control tower by operating the \nline key associated with the transmit line.\n\nBoth controllers at a position in the RAPCON \nCenter have individual loudspeakers in the console \nfor monitoring radio channels when their headsets \nare connected to the telephone lines. When a con- \ntroller\u2019s headset is connected to radio channels, he \nmay operate a switch in the console to cut off the \nloudspeaker and to permit him to listen with his \nheadset receiver only. However, the key equip- \nment is arranged so that if the loudspeaker-cutoft \nswitch is left on when the attendant returns to tele- \nphone-line operation, the loud-speaker is automati- \ncally connected to the output of the radio receiver. \nThis ensures that the radio channels will always be \nmonitored.\n\nThe Air Force plans to use RAPCON Centers at \noverseas locations as well as in the Continental \nUnited States. At all U. S. installations, Operating \nTelephone Companies install and maintain the RAP- \nCON telephone equipment. Key equipment for the \noverseas Centers, however, is pre-assembled and \npackaged in kit form by the Western Electric Com- \npany and is sold directly to the Air Force. It is in- \nstalled and maintained, therefore, by the Air Force.\n\nThese additions to No. 102A key equipment, for \nuse at both United States and overseas RAPCON \ninstallations, speed the control and promote the \nsafety of handling the high-speed military aircraft \nof today. They comprise another step in a series of \nmeasures being taken to keep communications in \npace with the continually growing traffic control \nproblem of the Air Force.\n\nG. A. Gipp1ncs, a native of Pontiac, Mich., received a B.S. degree in Electrical \nEngineering from Michigan State University in 1941. He immediately joined \nthe Michigan Bell Telephone Company, and worked in their Plant Department \nfor twelve years, except for three years during World War II when he was on loan \nto the Laboratories for work on war contracts. In 1953, he transferred to the \nLaboratories and was concerned with the development of station systems for \nmilitary applications. Mr. Giddings recently transferred to the O. & E. Depart-\n\nFor the large information-carrying capac- \nities required of future communications \nsystems, it is necessary to go to shorter \nwavelengths and greater bandwidths. The- \nory and experiment have shown, however, \nthat millimeter waves are sometimes se- \nverely attenuated by rain and by certain\n\n\u2018 | gases in the air. To get precise information \non this point, a very accurate method has \n; Vi. \u00a5 been devised for measuring such losses\n\nMost long-distance telephone and television traf- \nfic in the Bell System is handled by coaxial cable \nand microwave radio-relay systems. Coaxial cable, \nbecause of its shielding and underground installa- \ntion, is relatively immune to atmospheric disturb- \nances, but radio services must always be designed \nwith careful consideration of the effect of the at- \nmosphere on the propagation of radio waves.\n\nMicrowave radio-relay systems \u2014 TD-2, TH (now \nunder development) and TJ \u2014 use the atmosphere \nas a low-loss medium through which the radio energy \nis propagated. This may give the impression that \nthe atmosphere is transparent to all radio frequen-\n\ncies; actually, however, as the frequency increases, \n\" attenuation is severe in certain bands.\n\nTransmission losses encountered by radio waves \nin the atmosphere can be divided into two general \ncategories: those due to precipitation \u2014 rain, snow,\n\nprincipally oxygen and water vapor. At 4,000 mega- \ncycles, the frequency of the TD-2 radio-relay system, \nsuch losses are so small that they can be neglected \nfor most practical purposes. But when the operating \nfrequency is increased to about 10,000 mc, the \nfrequency of the TJ system,* attenuation by rain\n\nbecomes an important consideration. This type of \nloss increases rapidly as one proceeds to still higher \nfrequencies, and at 60,000 mc it has a value of 20 \ndecibels per mile for a moderately heavy rain. For- \ntunately, fog causes much less attenuation, princi- \npally because there is a lesser amount of precipitate \nper unit volume of air. Likewise, because of its \ncrystalline structure, snow attenuates radio waves \nmuch less than rain.\n\nAt the higher frequencies, the second category \nof atmospheric loss is also important. Even if it is \nnot raining or snowing, the oxygen and water vapor \nthat constitute part of our atmosphere attenuate \nradio waves appreciably in the region of millimeter \nwaves (frequencies between 30,000 and 300,000 \nmc). The oxygen molecules in the air become res- \nonant to the frequencies of mm waves and absorb \nmuch of the propagated energy. As shown in Figure \n1, absorption occurs chiefly in the band from 50,000 \nto 70,000 mc, with an attenuation of more than 20 \ndb per mile at 60,000 mc. At this frequency, there- \nfore, oxygen attenuation of mm waves is about the \nsame as that from a moderately heavy rain. Also, \nduring rain, the two attenuations are additive, so \nthat transmission in the region of 60,000 mc would \nbe very difficult.\n\nthat frequencies higher than those employed in \npresent microwave systems are impracticable, but \nlosses may limit the use of mm waves in the at- \nmosphere to very short ranges, Also, transmission \nlosses can be avoided through the use of waveguide \nsystems, and the Laboratories is now concentrating \non this type of system in which the \u201catmosphere\u201d \nor propagation medium within the waveguide can \nbe controlled.*\n\nAny research into possible mm-wave systems, \nhowever, depends upon an accurate knowledge of \nthe transmission losses. Described here is a precise \nmethod devised for making an accurate determina- \ntion of atmospheric loss. It has proved to be reliable \nfor studies of attenuation over a broad range of \nfrequencies in the mm-wave region. The method \nis demonstrated by experiments that determine ab- \nsorption by the oxygen in the air. In these experi- \nments, mm waves are propagated through the at- \nmosphere toward two reflectors located at different \ndistances from the transmitter, so that the difference \nin the energies of the two reflected waves is a \nmeasure of the absorption.\n\nIn principle, attenuation by the atmosphere can \nbe determined simply by radiating a mm wave \nfrom a transmitting antenna to a receiving antenna. \nIf the gains of the antennas are known accurately, \nand if the antenna beams are narrow enough to pre- \nclude reflections from the ground and from objects \nalong the path of propagation, one can calculate \nthe power that would be received provided the at- \nmosphere had no loss (often called the free-space \npower ). Thus, the calculated power minus the ob-\n\n24 \nj \na \no YY Y Y \nY \nWH \n= 12% \nY ABSORPTION BY-~ \nEo OXYGEN IN THE AIR \n(SEA-LEVEL \nz PRESSURE) \nuw 7 P J \n|\n\nsorption and calculated curve, showing close agree- \nment between theory and test results.\n\nFig. 2\u2014R. A. Desmond inspecting polyethylene \nlens on conical horn antenna. Antenna served for \nboth transmitting and receiving the test signals.\n\nserved power provides a measure of the atmospheric \nattenuation between the transmitter and receiver.\n\nUnfortunately, the accuracy of this one-way trans- \nmission method is seriously affected by slight in- \nstabilities at the transmitting and receiving termin- \nals. Moreover, if one wishes to measure a number \nof frequencies, the method is time-consuming, since \noscillators at both the transmitter and receiver must \nbe adjusted for each measurement.\n\nA more accurate and more convenient experi- \nmental procedure is indicated in Figure 3. This is \nthe two-way transmission method presently used \nat the Holmdel Laboratory. The mm waves at vari- \nous frequencies are generated by a mm-wave oscil- \nlator and are directed by an antenna toward two \ncorner reflectors R; and Ro. These reflectors are of \nthe tri-hedral type investigated at the Laboratories. \nThey have the configuration of an inside corner of \na cube, and they direct the incident beam back \nupon itself. Reflectors of this type were used be- \ncause they are less sensitive than flat reflectors to \nminor misalignments in position.\n\nThe transmitting antenna also receives the re- \nflected signals. First, the two reflectors are placed \nside by side facing the antenna, and their relative \nreflecting properties are determined by comparing \nthe reflected power when the reflectors are covered \nalternately with a highly absorbent, non-reflecting \nmaterial. This calibration is carried out at all fre- \nquencies of interest.\n\nThe reflectors are then separated as shown in \nFigure 3. Knowing the relative reflecting properties\n\nand the distances d, and d. from the antenna to \nthe two reflectors, one can readily calculate the \nratio of the powers that would be reflected from \nR; and Re if there were no loss in the atmosphere. \nThe difference between the measured and calculated \nratios then represents the attenuation of that portion \nof the atmosphere between Rr; and Rp. Since the \nmeasurement is in terms of ratios rather than ab- \nsolute values, it is independent of the gain of the \nantenna and of the absolute operating levels of the \nmeasuring set.\n\nThe method of measuring the reflected signals \nis illustrated in Figure 4. The transmitted mm-wave \nsignal is frequency modulated in a \u201csawtooth\u201d man- \nner, and the modulating frequencies have a small \ntotal excursion, Fr. The transmitted signal \u2014 solid- \nline sawtooth curve in Figure 4(a) \u2014is radiated \nfrom the antenna to the near corner reflector, Rj. \nUpon being received, the reflected signal is delayed \nwith respect to the transmitted signal by a time, ty, \nequal to twice the distance to the reflector divided \nby the velocity of light. This delayed signal is indi-\n\nFig. 3\u2014The two-way transmission method: re- \nflectors are first placed side-by-side and then spaced \nat distances d,; and d, from the antenna.\n\nFig. 4\u2014 Transmitted and reflected signals fer (a) \nnear reflector R, and (b) far reflector Re.\n\nFig. 5 \u2014 D. C, Hogg, left, and A. B. Crawford ad- \njusting waveguide assembly used in experiments.\n\ncated by the dashed-line sawtooth curve. Ducing \na portion of each cycle, 1,-t;, the transmitted and \nreceived signals differ in frequency by a constant \nvalue, f. Power at this difference or \u201cbeat\u201d fre- \nquency is amplified in a narrow-band amplifier \ncentered on f. Because during time t, there is a \nmuch larger value of beat frequency, the amplifier \nwill have no output for this period. The result, \ntherefore, is a series of pulses at frequency f, each \nof length 1,-t,, and with a repetition rate 1/1.\n\nFor the far corner reflector, Ro, the period of the \nsawtooth modulation is increased in proportion to \nthe increase in distance, as indicated in Figure 4(b). \nConsequently, the difference frequency, f, is identi- \ncal to that obtained with the near reflector. Also, \nthe frequency excursion, F, is held constant, which \nmeans that the average power output of the trans- \nmitter remains unchanged. Thus, the same amplifier \ncan be used for both the near and far reflectors, \nand a single meter can be used to compare the \ndifference between the received powers. Actually, \nthe signals are adjusted to give identical meter \nreadings for the two cases, and the power difference \nis read from a precision attenuator.\n\nA block diagram of the electronic apparatus is \nshown in Figure 6. In brief, the sawtooth generator \nin the upper left of this illustration modulates the \nmm-wave oscillator at the lower left, and power is \nfed to the antenna at the lower right. Part of this \npower, however, is delivered through a 6-db coupler \nto a balanced converter. The reflected signal is \nreceived by the antenna and is also delivered to \nthe converter, via the 3-db coupler. This balanced \nconverter, which uses two wafer-type millimeter \nrectifier units, mixes the transmitted and delayed \nsignals to produce the difference or intermediate\n\nThe nF source used for most of these measure- \nments was a low-voltage reflex klystron developed \nby E. D. Reed of the Laboratories.* It has an av- \nerage power output of about 12 milliwatts over \nthe 50,000 to 60,000 mc range. The mr amplifier is \ncentered on a frequency, f, of 750 ke and has a \nbandwidth of 300 ke. This bandwidth is narrow \nenough to give a good signal-to-noise advantage \nand is wide enough to take care of any non-linearity \nin the sawtooth modulation. Backward-wave oscilla- \ntorst are also used as sources of mm-wave power \nfor this type of measurement.\n\nThe measurements shown by the points in Fig- \nure 1 were made with this equipment. They were \ntaken during winter days when the air was dry \nenough to eliminate effects due to water vapor. The \nsolid curve in the illustration represents a theoreti- \ncal calculation based on absorption by oxygen. Be-\n\ncause of the good agreement between the measured \nand calculated values, attenuation by dry air in \nthe measured band of frequencies is believed to \nbe due mainly to absorption caused by resonance \nof the oxygen molecules. In this band, water vapor \nattenuates radio waves much less than oxygen. Even \nin conditions of relatively high humidity, water \nvapor increases the attenuation only about one-\n\nFig. 6\u2014 Transmitted and reflected signals are \nmixed in a balanced converter, amplified, and de- \nlivered to an oscilloscope and meter or recorder.\n\nin Electrical Engineering at Ohio State University in 1928 and joined the Radio \nResearch Department of the Bell Telephone Laboratories the same year. For a \nnumber of years he was concerned with ultra-short wave measuring techniques \nand propagation studies. During the war years he worked on various microwave \ncomponents for radar systems. In recent years he has headed a subdepartment \nin Radio Research at the Holmdel Laboratories working on microwave antennas, \nline-of-sight and beyond-the-horizon radio propagation and point-contact recti- \nfiers. He is a Fellow of the Institute of Radio Engineers and a member of Sigma Xi.\n\nD. C. Hoce, a native of Saskatchewan, Canada, spent five years with the Cana- \ndian Army during World War II in radar and telecommunications work in Britain \nand Europe. After the war, he received the B.Sc. degree from the University of \nWestern Ontario, London, Canada, in 1949, and the M.Sc. and Ph.D. from \nMcGill University, Montreal, in 1953. At McGill University, his research was \ncentered upon microwave diffraction. Mr. Hogg joined Bell Laboratories in 1953. \nHe has been concerned with research on over-the-horizon microwave radio \npropagation and with atmospheric attenuation of millimeter waves.\n\nAn automatic instrument has been developed at \nBell Laboratories for rapidly measuring and record- \ning very small imperfections in pitch unifonnity of \nrepetitive structures. The instrument, termed a \n\u201cmicrodeviometer,\u201d is presently being used for eval- \nuating the helices of traveling-wave tubes, but it can \nbe used equally well for measuring the pitch uni- \nformity of other periodic structures such as preci- \nsion screws and the grids of electron tubes.\n\nIn a traveling-wave tube, the wave to be ampli- \nfied is launched on one end of a helix, which is \nwound in such a manner that the wave is effectively\n\nslowed down along the axial direction of the helix. : \nWhen the axial velocity of the wave equals the \u2014\n\nvelocity of an electron beam focused through the \ncenter of the helix, amplification results. Thus, the \nhelix must be very precisely dimensioned if the tube \nis to meet design requirements.\n\nIn the microdeviometer, the output is a pen re- \ncording which tells at a glance the deviation of each \nsuccessive relevant part of the periodic structure\n\nH. T. Closson examines record of precision measurements \nof pitch uniformity obtained with the new microdeviometer.\n\nfrom the ideal, the ideal structure being one in \nwhich the length of each repeated section is exactly \nthe same as every other. Longitudinal displace- \nments are indicated to an accuracy of one micron \n(1/1000 mm). A traveling-wave tube helix can be \ncompletely evaluated in less than 5 minutes, com- \npared to more than two man-days necessary by \npreviously available techniques.\n\nThe apparatus incorporates a combination of opti- \ncal, electronic and mechanical technology. Essential \nfeatures include: (1) an optical grating which pro- \nvides a very accurate distance scale, (2) an optical \nsystem for ascertaining the position of the part to \nbe measured with respect to the chosen scale, and \n(3) an electronic method of analyzing this informa- \ntion and recording it on a strip chart.\n\nThe part to be measured is rigidly mounted on a \nmoving platform on which is also mounted the pre- \ncision optical grating. This grating moves past a \nsimilar fixed grating, so that a beam of light is in- - \nterrupted when the rulings on the two gratings are \nsuperimposed. The light beam falls on a photocell, \nwhere the interruptions produce electrical pulses \nrepresenting accurate distance increments of one \nmicron. Another combination of a fixed beam of \nlight and a photocell produces an output pulse \nwhen an element of the periodic structure passes \na fixed reference point. In other words, pulses from \nthe helix are obtained for comparison with the very \naccurate pulses from the precision gratings.\n\nThe pulses are fed to an electronic computer \nwhich evaluates the deviation from the ideal. This \ninformation is then recorded on a chart in such a \nmanner as to provide a direct reading of the actual \ndeviation of each element.\n\nIn the accompanying photograph, the mechani- \ncal-optical section of the microdeviometer is in the \ncenter, and the specially designed electronic com- \nputer is at the left. The pen-recording strip chart \nis obtained from the unit seen at the lower right. \nThe inset view shows a typical record, where the \npositions of the horizontal bars represent helix-turn \ndeviations from the ideal.\n\nSince outside plant makes up about 37 per cent of the total Bell System \nplant investment, Laboratories\u2019 development effort directed toward the \nimprovement of outside plant facilities and techniques pays large \ndividends. The \u201cpunched-sleeve\u201d method of cable splicing is a good \nexample of this type of development. The new method results in the \nsplicer\u2019s making better joints more rapidly and efficiently than by the\n\nEach year the Operating Companies of the Bell \nSystem add about 50,000 sheath miles of telephone \ncable to their outside plant. These cables vary \nin size from the smallest aerial distribution types \ncontaining 6 pairs of wire to the big feeder cables \ncontaining as many as 2,100 pairs, running in \nunderground conduit systems. The cables are in- \nstalled in lengths that range from less than 200 \nfeet at one extreme to 2,000 feet and more at the \nother. A work force of more than 14,000 cable \nsplicers must make more than 300 million conduc- \ntor joints each year in splicing together several hun- \ndred thousand sections of cable.\n\nAlthough many construction and maintenance \noperations of the outside plant forces have been \nmodernized, and although some have been highly \nmechanized, the techniques for joining cable con- \nductors have remained virtually unchanged since \nthe beginning of the telephone business. The time- \nhonored method of joining two or more paper or \npulp-insulated cable conductors is simple and re- \nquires a minimum of equipment. The cable splicer, \nor his helper, slips a cotton sleeve over one of the \nconductors before the joint is made. He then brings \nthe conductors together, cuts them to proper length, \nremoves a few inches of insulation, twists the bare \nconductors together, cuts them to final length,\n\nand slips the cotton sleeve over the joint for insula- \ntion to replace that previously removed.\n\nThe mechanical and electrical stability of joints \nproduced by this method is generally satisfactory \nfor today\u2019s exchange distribution plant which is \nsubjected to ac ringing voltages and sizeable trans- \nmitter currents. Because high-resistance joints do \noccasionally develop, however, the twisted conduc-\n\nFig. 1\u2014 Pneumatic presser and sleeve used in \npunched-sleeve cable splicing. The sleeve assembly \nconsists of a tube made of the same metal as the \nconductors to be joined (aluminum or copper) with\n\nThe Bell System is constantly seeking ways to \ncut construction costs, and methods which consist \nprincipally of hand operations are special targets \nfor improvement. This, together with the need for \nhigher quality joints to meet the requirements of \nmodern and future circuits, which operate at low \nde voltages and small currents, has stimulated a \nsearch for new methods of joining conductors.\n\nIn the search for better splicing methods, one \nmust remember that the splicer seldom does his \nwork under ideal conditions. He must work in \nhot or cold, dry or wet weather, or in clean or dirty \nsurroundings, and so must his equipment. He must \ncarry all the equipment to the job and it must be \nrugged enough to stand frequent handling. If he \nneeds electric power, it must usually come from a \nportable generator. Thus many joining methods \nwhich would be satisfactory in the laboratory would \nnot be practical under field conditions.\n\nThe groundwork for the new approach to the \nconductor-joining problem was laid by analyses of \ntime and motion studies of the old method and of \nlabor and material costs. Guiding objectives were \ndeveloped as follows:\n\n4. It should be possible to join the various com- \nbinations of conductor gauges with a minimum \nnumber of sizes of parts.\n\n. Power should be used to reduce worker fatigue. \n. The method must be adaptable to the varied \nconditions of field splicing.\n\nFig. 2 \u2014 Cross-section of aluminum punched-sleeve \nsplice showing how insulation on conductors is dis- \nplaced and metal-to-metal contact is made between \nconductors and metal sleeve.\n\nFig. 3\u2014L. W. Faulkner checking resistance of \njoints made by punched-sleeve method.\n\nLong and favorable experience in the Bell System \nwith compreggion types of joints on bare conductors \nsuggested a Bethoa in which a simple metal sleeve \ncould be slipped over the ends of the conductors \nand then deformed in such a way as to penetrate \nthe insulation and produce a joint. Early experi- \nments with such sleeves and with toothed dies for \ndeforming them, showed that the general approach \nwas promising.\n\nStudy of the proposed method showed that sev- \neral requirements would have to be met. The tangs \nproduced by the die teeth must penetrate the insu- \nlation to reach every conductor, but must not sever \nany conductors. The contact areas between the \ntangs and the conductors must be adequate in size \nto furnish low resistance paths, and must be held \ntogether by stresses sufficient to prevent corrosion. \nThe stressed parts involved in the contacts must \nhave elastic reserve to provide permanent joints. \nThe insulation must not be held under compression \nin such a way that changes in temperature and \nhumidity could result in release of the stresses in \nthe deformed parts.\n\nTo obtain optimum performance, a program of \nintensive comparison testing was begun for many \ndifferent designs of toothed dies and variations of \nsleeves. Many types of tests for determining the \nquality of the joints produced, particularly from the \nstandpoint of permanence, were studied. The most \nuseful test has been one in which the joints are \nsubjected to cycles of temperature change. Changes \nof temperature can be expected to cause displace- \nments in the joints, unless the stresses involved are \nsufficient to prevent them. If the joint can withstand \nrepeated wide temperature fluctuations, the stresses\n\nIn this test, the joints are cycled between an oven \nmaintained at 150\u00b0F and a bath of liquid nitrogen \nat \u2014324\u00b0F. The higher temperature is normally \nreached by joints in an aerial cable splice under \nexposure to the sun. The lower temperature, while \nit bears no relation to any service condition,: was \nchosen partly as a matter of testing convenience, \nbut also because it provides a wide temperature \nrange with considerable thermal shock when the \nwarm joints are plunged into the liquid. The test \nis carried out automatically by a special machine, \nand joints are normally subjected to 100 of the \ntemperature cycles. Measurements of joint resist- \nance are made before and after the cycling. Cor- \nrelation has been obtained between the results of \nthese tests and actual aging under field conditions.\n\nThe method finally developed involves slipping \na sleeve over the ends of the two or three conduc- \ntors to be joined, and \u201cpunching\u201d the Peeve between \ntoothed dies in the jaws of a pneumatic tool, pow- \nered from the same tank of compressed nitrogen \nthat the splicer uses to pressure test cables and \nsplices. The sleeve consists of a tube made of the \nsame metal as the conductors to be joined (alumi- \nnum or copper), with a jacket of plastic that is \nclosed at one end. The punching operation deforms \nthe sleeve wall and the conductors in such a way \nthat a combination of shearing and compressive \nforces displaces the insulation, making metal-to- \nmetal contacts. This operation produces punctures \nin the plastic covering of the sleeve. These punc- \ntures are not detrimental, however, since they do \nnot lower the dielectric strength between the paper- \ninsulated conductors. Because the jacket material \nsprings out at the punctures, there is no possibility \nthat the metal sleeves on two joints will touch.\n\nFor optimum performance it has been found nec- \nessary to use different types of dies for punching\n\naluminum and copper sleeves. This is due to the \ndifferent mechanical properties of the two metals \nrelative to those of the insulation. Only two sizes \nof aluminum sleeves and three sizes of copper \nsleeves are required to handle the range of con- \nductor combinations encountered in the field, from \n19 gauge (0.036\u201d) to 26 gauge (0.016\u201d).\n\nThe introduction of a power tool into the cable \nsplicing art is an important step in mechanizing \nthis basic technique of cable plant construction. \nThe punched-sleeve method produces conductor \njoints of improved electrical quality in comparison \nwith the twisted joints. The evidence of extensive \nfield trials also gives promise of substantial reduc-\n\nFig. 4\u2014 Pneumatic presser being used to make \npunched-sleeve joint in paper-insulated conductors.\n\ntions in cable-splicing costs. For the future, the \nadvent of cable with plastic-insulated conductors \nintroduces new and more exacting requirements \nand makes further work necessary on the challeng- \ning problem of conductor joining.\n\nW. C. Kietrevper had been engaged in outside plant tool and wire develop- \nment at the Laboratories for five years when he received the B.S. degree in M.E. \nfrom Cooper Union in 1934. In 1939, still a member of Outside Plant, he trans- \nferred from projects in wire and wire attachments to cable joining and mainten- \nance, in which he has worked on problems such as the splicing of coaxial cables, \nsplicing methods for cables having new sheath structures, and improved methods \nfor joining cable conductors. This work was interrupted during World War II \nwhen he spent three years in the development of military equipment. He has \nrecently been in charge of the group working on cable joining methods.\n\nThe telephone industry must show the commis- \nsions and the public that the only way to telephone \nprogress is through telephone prosperity, A. T. & T. \nPresident Frederick R. Kappel said in a talk before \nthe United States Independent Telephone Associ- \nation in Chicago on October 14.\n\n\u201cTelephone service has been a low-earning enter- \nprise now for a long time. It ought not to be,\u201d Mr. \nKappel told the annual convention. He said a large \nsegment of the public and \u201ctoo many\u201d commissions \nseem to think that low earnings mean low rates and \ngood earnings mean high rates; whereas the fact is \nthat good earnings, in the long run, mean quality \nservice at lower cost to the user.\n\n\u201cIs it asking too much of regulatory commissions,\u201d \nhe continued, \u201cto exercise in fullest measure their \npractical judgment, their imagination and their \npolitical courage?\n\n\u201cI am sure it is not. And I am especially sure of \nthis when I reflect that the problems which confront \nthe regulator when a business is successful... are \nas nothing compared with the problems which con- \nfront him when a business doesn\u2019t have the money \nto do what ought to be done.\u201d\n\nMr. Kappel said he \u201ccannot believe\u201d that many \nstates expect to attract other industry and raise \nemployment and prosperity \u201cas they regulate tele- \nphone expansion and employment down.\n\n\u201cIt is our job to demonstrate that every state \nneeds our financial good health, not only for what \nthis means in direct telephone employment and \nwages, but equally or even more for what we can \ndo to help make the state attractive to others.\n\nTurning to the field of labor relations, Mr. Kappel \nsaid that what telephone employees really know \nand believe about their company depends on how \ndiligent the company is in giving employees the \nfacts and demonstrating its sincerity.\n\n\u201cI saw a union editorial the other day that had the \nBell System waxing rich and practically choking on \nfat profits,\u201d he said. \u201cThis kind of thing is so absurd \nyou may wonder why I even mention it. The reason \nis that the problems which unions and manage- \nments are concerned with are so important that \nthere just isn\u2019t room for careless or misleading talk.\n\n\u201cYou and I know that the telephone companies \nare bound to provide wages and working conditions \nthat compare favorably with other industry and\n\noffer attractive opportunity for willing and indus- \ntrious men and women.\n\n\u201cFrivolous and unconsidered statements can only \ncause harm\u2014the more so if by silence we seem to \ngive them our consent. What we need is to get such \na steady stream of truth flowing that there\u2019s no \nroom in the river for anything else.\u201d\n\nMr. Kappel referred also to competition in provid- \ning communication services. \u201cIt seems there are \nquite a few non-communication companies today \nthat would like to provide a considerable part of \ntheir communications themselves, without depend- \ning on the common carriers,\u201d he said.\n\n\u201cOf course, this goes against the basic principles \nand experience which have shown that the public \ninterest depends on common carrier service. Indis- \ncriminate licensing of non-common carriers to build \ntheir own systems would not only sacrifice the most \nefficient use of the radio spectrum; it could very \nseriously interfere with the ability of the telephone \ncompanies to serve the public at reasonable prices.\n\n\u201cWe are sure,\u201d he continued, \u201cthat common car- \nrier-service results in lower costs to the public. We \nare confident it also makes possible much more effi- \ncient use of the radio spectrum. We know a strong \ncommon-carrier network is a much greater asset in \ntime of emergency or war than a fragmentized \nhodge-podge of private systems could ever be.\u201d\n\nBuilding future telephone management, Mr. Kap- \npel said, \u201cis the responsibility of every boss in the \nbusiness. ... The great essential is to have the kind \nof working atmosphere that gives people air and \nroom and freedom and incentive to grow. We'll do \nthe best job, I\u2019m sure, when\u2014and only when\u2014every \nboss acts on the understanding that an indispensable \npart of his assignment is to . . . encourage the \ngrowth of his subordinates.\u201d\n\nMr. Kappel congratulated the USITA on its 60 \nyears of progress and its vital part in the total \naccomplishment of the industry. \u201cThe way things \nare nowadays, none of us have much time to be \nlooking back. The future isn\u2019t waiting for us. It\u2019s \nrushing right at us,\u201d he said.\n\n\u201cWe in the telephone business must meet these \nchanges together, for ours is one single industry and \nour service is indivisible. Far more than in other \ntypes of business, the progress we make depends on \nhow effectively we combine our efforts.\u201d\n\nWith the expansion of direct distance dialing \nand automatic alternate routing, transmission \nmaintenance requirements for long-distance \ntrunks are more exacting. To assist the main- \ntenance forces in gathering transmission data \nmore efficiently, new circuits, including a com- \nputer unit, have been incorporated into the \nintertoll trunk-test equipment. This equipment \nspeeds the analysis of transmission perform- \nance and thus expedites maintenance of trunks.\n\nLong-distance telephone channels, called intertoll \ntrunks, must be maintained at high transmission \nefficiency if service is not to be interrupted or im- \npaired. This is and always has been a general re- \nquirement for long-distance facilities, but it is even \nmore important now that completely automatic and \nhighly versatile switching systems are enabling \nmore and more people to dial their own long- \ndistance calls without operator assistance.\n\nA long-distance communication path may require \nseveral separate transmission links connected in \ntandem (seven links is the usual maximum), and \nwith fuily automatic alternate routing, the path \nitself will be only one out of several possible alterna- \ntives. When operators were involved in all long- \ndistance calls, they could frequently detect any \ndegradation of transmission quality and report it as \nan aid to the maintenance forces. Now that tele- \nphoning is becoming more completely automatic, \nhowever, the discovery and measurement of trans-\n\nmission deviations must also become more auto- \nmatic. To insure that necessary adjustments are \npromptly made, the maintenance force must more \nclosely observe intertoll trunk performance. This is \ntrue both of the operational aspects of trunks \u2014 the \nmany switching and regulating actions that take \nplace in setting up the trunk connection \u2014 and of \ntransmission performance \u2014 whether voice and sig- \nnals are transmitted over the trunk without exces- \nsive deviation from design values.\n\nTo aid maintenance, several types of testing and \nmeasuring circuits have been developed. As shown \nin simplified form in Figure 2(a), tests can be per- \nformed from a toll test board at an originating \n(\u201cnear-end\u201d) office, by making connection to an \nautomatic transmission-measuring and noise-check- \ning circuit at the terminating or \u201cfar-end\u201d office. \nHowever, because personnel at the near-end office \nset up the tests and record the data, this arrange- \nment is not fully automatic. Consequently, addi-\n\n<\u2014 MINUS PLUS \nMAGNITUDE OF DEPARTURE FROM DESIGN VALUE \nFig. 1\u2014 Normal distribution curve for random \nvariation of transmission values about design value.\n\nThe automatic transmission test and control cir- \ncuit and the recording teletypewriter printer circuit \nin Figure 2(b) have also been added to the near- \nend test equipment. These circuits, in conjunction \nwith the AOIT, perform fully automatic two-way \ntransmission-loss measurements and noise checks, \nand also provide a printed record of the operational \nor transmission-test results.\n\nWith this test gear, however, a large amount of \ndata-plotting is still required, and more recently a \ncomputer-type circuit has been added to the test \nfacilities to simplify the presentation of data in \nusable form. But before describing this circuit, it is \nwell to review the sort of data-gathering operation \nrequired in the transmission-testing of trunks.\n\nTransmission performance of intertoll trunks is \nmeasured in terms of how much a given trunk \ndeviates from a specified or design loss. For ex- \nample, over a certain trunk the transmitted energy \nmay decrease in power by 6 db when it is desired \nthat the actual loss should be only 5 db. Transmis- \nsion performance is then described by saying that \nthis trunk has a transmission deviation of +1 db. \nHowever, trunk use becomes, to a considerable \nextent, a matter of chance selection, and for this \nreason, performance is considered by statistical anal- \nysis of a large number of trunks or by groups of \ntrunks, rather than on an individual trunk basis.\n\nExperience has shown that when a sufficiently \nlarge number of transmission measurements are \nmade, the variations, taken as a whole, will follow a \nnormal distribution curve as shown in Figure 1. \nMost of the trunks in a group will be fairly close to\n\n* TERMINATING OFFICE MAY BE NO. 4 OR NO. 5 CROSSBAR; \nCROSSBAR TANDEM, OR STEP-BY-STEP INTERTOLL\n\nFig. 2 -\u2014 Semi-automatic and full-automatic circuits for \nperforming transmission and operational tests of long-dis-\n\ntheir specified loss values, while some trunks may \ndeviate 2 or 3 db or more. In the illustration, this \nfact is reflected by the peak seen in the curve \nat zero deviation with gradually decreasing num- \nbers of measurements on either side. However, be- \ncause the specified losses are not necessarily the \nsame for each trunk in a large group, the distribu- \ntion of deviations usually will not be symmetrical \nabout zero, but will center on an algebraic average \nof all deviations. The displacement from the normal \nis known as \u201cbias\u201d and indicates the general trend \nof deviations in the group.\n\nStatistical analysis of all deviations, both above \nand below the specified values, provides the \n\u201cstandard deviation\u201d (o). As indicated in Figure 1, \nthe standard deviation is the cross-hatched area \nunder the curve. With normal deviation, about 65 \nto 70 per cent of all deviations will occur in this \nregion, and in transmission parlance it is called the \n\u201cdistribution grade.\u201d If the distribution grade is low, \nthe group of trunks is good, with most trunks near \ntheir assigned values. If it is high, some of the \ntrunks need readjustment.\n\nAlthough the trunks are thus considered as a \ngroup, the data for the curves must be obtained \nfrom separate measurements on each trunk. When \ntransmission grade is calculated manually, the pro- \ncedure of recording a large number of measure- \nments is somewhat simplified by using a special \ndata sheet. Part of one of these sheets can be seen \nin Figure 4, The procedure consists first of subtract- \ning the specified loss from the actual measured loss \nof the trunk. For each of the resulting deviation \nvalues, a \u201cstroke\u201d or mark is then placed on the data \nsheet. For example, suppose the first deviation value \nmeasured fell between \u20141.25 and \u20141.75 db. On the \nhorizontal line representing this band, a stroke \nwould be placed just to the left of the vertical line \nfor tally number one. All successive values in this \nsame \u20141.25 to \u20141.75 db range would later be \nstroked in the later tally positions along the hori- \nzontal line, and all other deviation values would be \nstroked along the other horizontal lines in the same \nmanner. This is continued until all the deviations \nhave been calculated and recorded, so that the total \nnumber of strokes as observed from the tally posi- \ntion of the last stroke on each line is the total num- \nber of measurements within each of the one-half db \ndeviation bands. It will be seen that an envelope \nenclosing these strokes is similar to the distribution \ncurves of Figure 1, except that here the graph is \noriented horizontally. The data sheets also give \nvisual clues to areas where maintenance effort\n\nFig. 3\u2014 Registers which score number of deviations \nin various decibel-bands, seen above telety pewriter \nwhich records results of tests.\n\nThere are a number of other operations per- \nformed with the data sheet before the information \nis brought to its final form, but from this description \nit is obvious that when large numbers of deviations \nare involved, manual subtraction and totalization of \nthe data can become prohibitively time-consuming. \nThe new computer included in the near-end testing \nequipment greatly reduces the amount of manual \nwork required. This circuit automatically calculates \nall the individual deviations and totalizes them in \none-half db ranges according to sign. That is, the \noperations of subtracting the specified from the \nactual values, and of adding by strokes, are per- \nformed automatically. The maintenance people \nmerely read the total number of strokes for each \none-half db band from message registers. Only the \nfinal stroke in each range is placed on the data \nsheet, after which the sheet is handled in the cus- \ntomary manner.\n\nIn addition, the associated recording teletype- \nwriter provides a record of each test. A portion of \nsuch a record is shown in Figure 5; it gives main- \ntenance personnel a number identifying the trunk \nthat was tested, the specified loss, and the deviation \nfrom this specified loss in each direction of trans- \nmission (in and out of the office). For example, the\n\nspecified loss for trunk number 1,378 is 7.5 db. The \ntest results indicate transmission deviations of \u20145.2 \ndb in the far-to-near direction and +-0.7 db in the \nnear-to-far direction. For simplicity, the decimal \npoint between the tenths and units digits for speci- \nfied loss and deviation entries is not recorded by the\n\nFig. 4\u2014 Portion of special data sheet used in \nrecording of transmission deviations.\n\nprinter. In addition, for an entry of less than 10 db. \nthe tens digit is omitted, again for simplicity.\n\nThe record also gives a symbol to indicate when \nmeasurements exceed transmission or noise limits, \nor when circuit failures occur to prevent satisfactory \ncompletion of the test. For instance, on trunk num- \nbered 1,377, the nois\u00e9 exceeded limits in the far-to- \nnear direction, as indicated by the symbol n. On \ntrunk numbered 1,378 the deviation exceeded wide \ndeviation limits in the far-to-near direction, as indi- \ncated by the vu in the appropriate column. The \nrecord for trunk numbered 1,267 indicates, by the \nsymbol y, that far-end equipment failure prevented \ncompletion of the test. In Figure 5, the character B \non the test record indicates that the trunk was busy \nat the time a test was attempted, and a indicates \nthat the test call did not successfully complete to \nthe terminating office.\n\nBasically, the computer duplicates the operations \ndescribed above for the manual determination of \ntransmission deviations and their proper sign (+ or \n\u2014). This information is passed to the teletypewriter \nfor recording. A \u201cclass\u201d relay associated with each \ntrunk supplies the specified loss value, and this is \nsubtracted from the measured loss value. The meas- \nured loss values are stored in relay memory circuits \nthat have registered the test results. The values are \nin decimal form \u2014 between zero and 19.9 db \u2014 and\n\nbefore being fed into the computer, they are trans- \nlated into their biquinary equivalents. The biquin- \nary system is a method of encoding any digit from 0 \nto 9 in such a way that it can be registered by oper- \nating two out of seven relays. Five of these relays \n(the \u201cquinary\u201d part) represent either 0, 1, 2, 3, 4, or \n5, 6, 7, 8, 9, depending on which of the remaining \ntwo relays (the \u201cbi\u201d part) is operated. Since two and \nonly two relays must be operated for each digit, the \ncomputer can be made self-checking to the extent \nthat if more or fewer than two relays operate, or if \ntwo relays operate in an invalid combination, an \nerror is automatically detected. After a subtraction \nis completed in the computer, the information is \npassed to the teletypewriter in a form suitable for \nits operation.\n\nIn the design of the computer, it was decided that \nit would be convenient to have a circuit that would \nonly add and that would perform the required sub- \ntraction by the \u201cnines complement\u201d method. It may \nbe recalled from arithmetic studies that a \u201cnines\u201d \nmethod is sometimes used to check a subtraction \nby adding the complements. The computer uses a \nversion of this device to perform subtraction by \naddition of the complements.\n\nFor example, a trunk might have a specified loss \nof 7.2 db; this quantity would appear in the com- \nputer as 99.9 minus 07.2 or 92.7. Suppose now that \nthe actual loss of this trunk is 9.1 db. The computer \nadds 09.1 to the 92.7 to get a sum 101.8. Or, suppose \nthe actual loss of the trunk is 5.2 db, in which case \nthe arithmetic is 92.7 plus 05.2 equals 97.9 db. It \nwill be noticed that the sum is greater or less than \n100.0, depending on whether the actual loss is\n\nFig. 5 \u2014 Teletypewriter record of computer-ana- \nlyzed results of transmission tests. Data are used \nfor rapid maintenance of trunks.\n\ngreater or less than the specified loss. The mathe- \nmatical expression is (99.9 \u2014 S) -+ M = D, where S \nis the specified loss, M is the actual measured loss, \nand D is the \u201cdeviation.\u201d\n\nThis \u201cdeviation,\u201d however, does not represent the \ntrue sum, which must be the actual deviation from \nthe specified loss. With a number over 100 like the \n101.8, the first digit \u201c1\u201d is dropped and 0.1 is added, \nto result in the true deviation of plus 01.9 db. With \na number below 100 like the 97.9, the nines com- \nplement of the sum (99.9 minus 97.9) yields the \ntrue result of minus 2.0 db. In this manner, the com- \nputer is able to give an indication of the correct \nsign of the result, and all values are recorded by the \nteletypewriter with their appropriate + or \u2014 signs. \nSeveral entries of this type can be seen in the tele- \ntypewriter record shown as Figure 5.\n\nThirty-three message registers at the near end \ntotalize the individual transmission deviations for \neach 0.5 db increment from +8 db to \u20148 db accord- \ning to sign. This is done so that the values will agree \nwith the increments on the special data sheet used \nfor transmission analysis. An additional message \nregister totalizes the number of individual one-way \nmeasurements, and its reading should be equal to \nthe total of the readings on the thirty-three devia-\n\ntion registers. Each register, equipped with a man- \nual reset key, is reset to zero prior to the start of a \ntest cycle and is read immediately on completion of \nthe test cycle.\n\nDuring normal operation of the test circuit, two- \nway measurements and a noise check are made on \neach trunk. The computer first calculates the devia- \ntion and scores the proper register for the far-to- \nnear measurement. When the printer completes this \nentry, the computer is restored to normal so as to \nservice the near-to-far measurement in a similar \nmanner. Should the printer not be functioning, the \ncontrol circuit will automatically recycle the com- \nputer for the near-to-far calculations. At the end of \na test cycle, the deviation register readings can be \ntranscribed as the final tally on the data sheet.\n\nPrior to standardization, successful field trials of \nthe near-end equipment were conducted in the \nWashington, D. C. No. 4A toll switching office, in \nconjunction with far-end equipment trials in several \nother cities throughout the United States. This \nequipment is now in production by the Western \nElectric Company, and it will greatly aid main- \ntenance forces in the analysis of transmission per- \nformance of long-distance trunks and will expedite \ncorrective action by maintenance forces.\n\nR. C. Nance, a resident of Wood-Ridge, N. J., joined the Laboratories in 1936, \nand became a draftsman before taking military leave from 1941 to 1945 to serve in \nthe Signal Corps in the Pacific Theater. Shortly after his return to the Laboratories \nhe did design work on switchboard trunk circuits and toll crossbar trunk circuits. \nIn 1953 he became a Member of Technical Staff, concerned with the design of \nautomatic trunk test frames and associated recording printer circuits for No. 4 and \ncrossbar tandem offices. More recently his work has involved design of sender test \nframes for CAMA toll crossbar installations, Mr. Nance attended evening classes \nat Newark College of Engineering to receive the Associate E.E. degree in 1940.\n\nAt Bell Laboratories, continuing studies are di- \nrected at one of the oldest and most fundamental \nproblems of the communications industry \u2014 the \naction of a pair of contacts in an electromechanical \nrelay or switch as these contacts make or break an \nelectrical circuit. Thousands of contact operations \nmay be involved in a single telephone call.\n\nOf special interest has been the phenomenon of \narcing \u2014 the discharge of electrical energy when \ntwo contacts are close together. Frequently, arcs \ncause welding of the contacts. According to recent \nresearch studies at the Laboratories, it has been \nfound that welding at relay contacts takes place \nthe instant the arc is extinguished. This work, by \nJ. L. Smith and W. S. Boyle, sheds new light on the \nmechanism of the welding process.\n\nThe process was studied by short-circuiting vari- \nous short lengths of charged transmission line \nthrough an arc formed by closure of a pair of clean \nrelay contacts. It was found that the contacts tend \nto weld at the time when the line has just been com- \npletely discharged. Since the arc is known to pro- \nduce a small pool of molten metal on the contact \nsurface, this suggests that molten metal is drawn \nacross the contact gap by the redistributed field \nafter the arc is extinguished, resulting in a weld. \nFor longer transmission lines or more complex\n\nJ. L. Smith (left) and W. S. Boyle studying weld- \ning at relay contacts. The coil and capacitor com- \nbinations in the center of the picture are simulated \nlengths of transmission wire used in the studies.\n\ncircuitry, welding is most probable at the instant \nthe arc is terminated. In such cases, however, termi- \nnation is due to a fundamental instablity in the arc \nitself. This arises because the diameter of the molten \npool of metal increases as the cube root of the \nenergy which has been dissipated in the arc, while \nthe heat losses from the pool increase linearly with \nits diameter. In the constant-current arc, therefore, \nthere occurs a critical time after which it can no \nlonger be maintained because of the heat losses.\n\nThis time of extinction can be computed from the \nphysical constants of the contact metal and the \ncurrent and voltage of the arc. The extinction times \nthus computed for several contact materials at vari- \nous voltages and currents agree with the observed \ntimes to weld with a maximum deviation of about \n30 per cent.\n\nPrior to these studies, it had been generally be- \nlieved that since the volume of molten metal in- \ncreases with time, welding occurred only after \ndissipation of a critical amount of energy and hence \nfor a critical volume of melt. These studies have \nnow shown that welding occurs both for much \ngreater and much smaller energies. Welding time \nbears no direct relationship to power dissipated in \nthe arc, but it is very well correlated with the time \nof extinction of the arc.\n\nIn addition to providing fast, efficient and economical communications, two \nfundamental requirements in the Bell System are safety for telephone users \nand employees and uninterrupted service. This applies of course, to the \nvast TD-2 microwave radio-relay networks throughout the nation. TD-2 \ncarries a large percentage of the country\u2019s telephone and television trans- \nmission traffic, and antenna sites must be protected from damage by light- \nning. Such careful attention has been given to this problem that there has \nnot been a single injury or service interruption from lightning strokes.\n\nTo obtain the height required for line-of-sight \ntransmission, TD-2 radio-relay stations are located \non hilltops or on tall supporting structures \u2014 and \nfor this reason they are prime targets for lightning. \nWith about 600 such installations in a microwave \nnetwork of some 21,000 route miles, the likelihood \nof a thunderstorm occurring over some part of \nthis vast network is very great. The probable light- \nning-stroke incidence for these 600 structures is \nestimated at over 500 per year, which, without \nadequate protective measures, would constitute a \nconsiderable hazard to personnel, equipment and \nservice continuity.\n\nA structure hit by a stroke of lightning becomes \npart of the path over which the current passes \nbetween cloud and earth. The magnitude of the \nstroke current is principally a function of meteor- \nological conditions and is not significantly affected\n\nby the impedance of the structure. The crest values \nof current in lightning strokes vary over a wide \nrange, as shown by the distribution curve in Figure \n1. It may be noted from this illustration that the \naverage value is about 16,000 amperes, but that \nstrokes may occasionally reach magnitudes of cur- \nrent exceeding 200,000 crest amperes. On the \naverage, stroke currents rise to peak value in 1 to \n2 microseconds and decay to half value in 40 to 50 \nmicroseconds. Because of the rapid rate of rise of \nthe surge front and because of the high currents \ninvolved, inductive potentials of many thousands \nof volts can develop, even in structures with neglig- \nible resistance.\n\nSuch inductive potentials could cause arcing to \nadjacent conducting objects and thus introduce fire \nand safety hazards. Where appreciable resistance is \npresent in the surge path \u2014 as might occur at joints\n\nFig. 1 \u2014 Distribution of lightning-stroke currents; \naverage is about 16,000 amperes, but some strokes \nmay exceed 200,000 amperes.\n\nand connection points \u2014 another fire hazard could \nexist from arcing produced by stroke currents of \nlarge magnitude. It is therefore necessary to pro- \nvide, between the stroke point and earth, paths \nhaving sufficiently low inductance to limit the build- \nup of inductive potentials, and also having adequate \ncurrent-carrying capacity to prevent arcing effects.\n\nFortunately, the antenna towers commonly used \nin the TD-2 system are metallic structures having \nwaveguides and wiring conduits for tower lights \nin good electrical contact with the tower at fre- \nquent intervals. These parallel paths provide a good \nlow-impedance path for the lightning currents.\n\nConsiderable metal is also present in the equip- \nment buildings in the form of structural members, \nreinforcing bars, additional wiring conduits, and \nequipment racks, These may be interconnected to \nsome extent in the normal course of construction, \nbut more adequate interconnection must be assured \nby specific bonding connections. Connections are \nestablished not only between the various conducting \ncomponents, but also between them and the area- \ngrounding structure. The basic approach has been \nto simulate as far as economically possible a type \nof protective structure known as a \u201cFaraday Cage.\u201d \nThis is accomplished by means of many conducting \npaths and frequent bonding connections.\n\nIn the absence of electrical arcing, a surprisingly \nsmall wire (No. 10 AWG copper) will carry most \nlightning surges without fusing. It is customary,\n\nmechanical strength, No. 2 copper is employed. To \nreduce potential drops due to the impedance of \nsuch conductors, it is preferable to provide multiple \npaths to ground for stroke currents, especially with \nrespect to those parts of an installation housing \nequipment and personnel. These factors have been \nwell recognized in the design of protection for \nmicrowave installations.\n\nThe antenna horns employed in the TD-2 sys- \ntem are of heavy metal construction capable of \nwithstanding the fusing effects of a lightning stroke. \nWhen the horn assembly is supported on a steel \ntower, which is the most common arrangement, \nthere is no problem of the electrical conductivity \nto earth. Some supporting structures are of con- \ncrete construction, however, in which case particu- \nlar care is taken to provide adequate electrical \nconductivity through the reinforcing bars and other \nsteel members.\n\nThe circuits required for aircraft-warning lights \nare enclosed in metal conduits supported on the \nstructural steel of the tower. The topmost fixture \nis provided with an \u201cair terminal\u201d in the form of \na 5-foot vertical stainless steel rod to shield the \nfixture against direct strokes.\n\nFig. 2 \u2014 Representation of buried-ground network \nto protect TD-2 sites from lightning damage.\n\nIn a typical installation, the transmitting and \nreceiving equipment is housed in a building near \nthe base of the tower. The waveguides from the \nhorns are electrically bonded at intervals to the \nsteel tower, and between the tower and the build- \ning such waveguides are supported on a steel rack. \nThe rack provides additional conductivity in parallel \nwith the waveguides between the tower and the \nbuilding ground.\n\nThe grounding arrangement for a microwave in- \nstallation is solely for protection purposes, since it \nis not required for microwave transmission. Usually, \na buried metallic piping system, such as a water \nsystem, is not available, so it is necessary to con- \nstruct a \u201cmade\u201d ground. Because of adverse ground- \ning conditions at many microwave sites, however, it \nis uneconomical to attempt the construction of a \nvery low-resistance ground. Rather than specify \na resistance value, Bell System practices require \na buried network dimensioned in relation to the \nsize of the installation. A general idea of such \nburied-ground networks is given in Figure 2.\n\nThis grounding network \u2014 supplemented by bond- \ning connections to the tower footing ground, fuel \ntank and other buried metallic objects \u2014 effectively \neliminates potential differences around the tower \nand equipment building. For high stroke currents, \nthe area adjacent to the station will attain a high \npotential with respect to remote earth, but will not \nendanger personnel or equipment within the sta- \ntion area. Power and local communication lines \nfeeding the station will, however, be subjected to \nthis rise in station potential and will thus require \nspecial protection measures in the form of arresters \non the power conductors and protectors on all com- \nmunication circuits. The grounding terminals of \nthese arresters and protectors are all connected to \nthe station ground ring. Additional protection may \nalso be provided at intervals on these facilities to\n\nFig. 3 \u2014 J. B. Hays, left, and author discuss results \nof Laboratories study of electrical protection.\n\nequalize potentials and to provide an adequate \npath to remote ground for a sizeable proportion of \nthe stroke current.\n\nA final but very important link in this chain of \nprotective measures is the extensive grounding and \nbonding of conduits and equipment racks in the \nstation to eliminate potential differences in the area \noccupied by personnel.\n\nD. W. Bop.e, a native of Huguenot, N. Y., entered the Development and Re- \nsearch Department of the A.T.&T, Co. in December, 1929, where he studied in- \nductive coordination and joint-use problems. He became a member of the Lab- \noratories in 1934. Mr. Bodle\u2019s experience has been chiefly in the field of protec- \ntion of equipment against foreign potentials and the protection of personnel from \nelectric shock, and he has engaged in several field investigations of the character- \nistics of natural lightning. During World War II he was concerned with the design \nof cathode-ray indicators for airborne radar. Mr. Bodle received a B.S. degree \nin Electrical Engineering from New York University. He is a licensed professional \nengineer and a member of the A.I.E.E. and of the L.R.E.\n\nIn the Automatic Message Accounting system, the two functions of printing \nand comparing paper tapes formerly required separate machines. Now, \nthese two functions are performed by a single AMA unit, which in addition \ncan automatically scan tapes for special information. The newly developed \nAMA printer-comparer-scanner thus increases the efficiency of message ac- \ncounting procedures and also simplifies the tasks of operating personnel.\n\nThe new development resulted from the need \nfor a simpler means of transcribing AMA perfora- \ntor tape information into printed form. During the \nearly stages of the design it became apparent that \nat very little additional cost the machine could \nalso be made to perform functions of the AMA \ntape comparer as well as provide a new require- \nment of tape scanning. The AMA printer-comparer- \nscanner combines these functions in a single ma- \nchine. The machine, shown in Figure 1, consists of \none bay of relays, one printer cabinet and two read- \ner cabinets.\n\nThe work performed by the printer-comparer- \nscanner can best be understood in terms of what\n\nthe AMA system has been designed to do. Included \namong the many AMA processing steps are provi- \nsions for printing data,* for comparing paper tapes,t \nand now, scanning paper tapes. The printing func- \ntion is necessary in some instances so that informa- \ntion which appears as punched holes in a paper \ntape can be recorded in a form readable by ac- \ncounting personnel. The comparing function is used \nin test procedures whereby an AMA paper tape is \ncompared with a known or standard tape to check \nthat the equipment is functioning properly. The \nnew function of scanning provides a method where- \nby a tape containing call information may be in- \nspected to observe irregularities, answer questions \nraised by customers, aid maintenance, or compile \ndata for the few cases where bills are made out on \nirregular dates.\n\nIn the initial stages of AMA development, toll- \ncall rating (calculation of charges for long-distance \ncalls) was handled manually. AMA call records,\n\nthe call was made, were processed through various \nAMA machines, and the information necessary to \ncalculate the charges on each call was assembled \nas a single entry on a tape. The AMA printer then \ntranslated the numerical information on the tape \n\u2018into printed letters and numbers and prepared in- \ndividual toll slips which were used by accounting \npersonnel to prepare customers\u2019 statements. The \ntremendous growth of long-distance traffic han- \ndled by AMA (approximately 32 million toll mes- \nsages per month) has made manual methods im- \npractical. A present step towards the automation \nof toll rating is the use of punched card methods. \nThis change in method of handling toll calls re- \nmoves the necessity for the complex AMA printer. \nThe only call records which must be printed un- \nder the present method of AMA operation are the \n\u201cstraddle\u201d records. This term is significant because \nwhen calls are in process at the time the central \noffice tape is cut, the records will straddle the cut- \nting point. The assembler-computer will process \nthe records on each side of the cutting point at \ndifferent times and therefore will be unable to per- \nform the necessary computations. The assembler- \ncomputer perforates the straddle call records on a \nseparate tape. Also included on this tape are rec- \nords of calls which exceed the capacity of the as- \nsembler-computer for calculating conversation time \nor message-unit charges. The tape is printed and \nIBM cards are punched manually from the printed \nrecord. The calls are then merged with the rest of \nthe records for automatic processing. An example \nof a straddle record, depicted in Figure 2, shows no \ntranslation of either the calling or called customer\u2019s \noffice code. The information is printed in the same \nnumerical form used by the switching equipment. \nAnother use of the printer is in obtaining mes- \nsage-unit summary information for customers who \nchange or disconnect their lines during the billing \nmonth. In the normal billing procedure, summaries \nof the message-unit usage of each customer are \nmade several times a month and the record is kept \nin tape form. A summary tape is prepared for each \ncentral office code. For each customer who has made \nany message-unit calls during the period of time \ncovered by the tape, an entry is made giving the \ntotal message-unit charges which he has accumu- \nlated. The entries are arranged in ascending order \nof customer line number. These summaries are ac- \ncumulated through the billing month and at the \nbilling date an IBM card is punched from the tape \nfor each customer, showing his total usage for the \nmonth. This card, together with the cards for toll\n\nmessages made by the customer during the month, \nis used in making the final bill.\n\nWhen a customer is to be billed prior to the \nregular billing date, however, it is necessary to de- \ntermine his usage before the cards are punched. \nAs a normal procedure, a printed record is made of. \nthe intermediate summary tapes which are pre- \npared during the month. The records are available \nwhen they are needed for billing a particular cus- \ntomer without preparing cards for all the customers \nin the office.\n\nAs a further application of the printer function, \nit may be necessary, for maintenance reasons, to \nprint a verbatim record of an AMA tape. The print- \ner-comparer-scanner can do this in either of two \nways. Figure 3 shows a portion of central office \ntape printed in the two forms. The form to the left \nshows each line printed as it appears on the AMA \ntape. The form to the right shows each entry \nprinted across a single line of the page. When the \n\u201cin line\u201d form is used, the zeros, which are always\n\nFig. 1\u2014The printer-comparer-scanner unit includes a \nprinter cabinet, left, a bay of relays, center, and two reader \ncabinets. Several functions are combined in the machine.\n\nthe first digits of a supplementary line, are omitted \nfrom the printed copy. This form is essentially the \nsame as that used to print straddle records.\n\nThe second function of the printer-comparer- \nscanner \u2014 tape comparing \u2014 is used in testing AMA \nmachines. Since those machines pass intelligence \nfrom one tv another in the form of punched paper \ntapes, the machines are best tested by means of \nknown test input tapes which will produce known \noutput tapes. The test output tape can then be \nchecked against a master tape. The printer-com- \nparer-scanner checks the two tapes, line by line, for \nidentity of the perforation. This function was pre- \nviously performed by the AMA tape comparer.\n\nThe third function of the new AMA unit \u2014 scan- \nning \u2014 fills a need long felt in the field for a means \nof obtaining special information from AMA tape \nrapidly and accurately. The alternatives to an au- \ntomatic tape scanning device are: using a hand \nreader, or printing the entire tape and then scanning \nthe result by eye. Both of these methods are time \nconsuming and susceptible to human error. When \nthe printer-comparer-scanner is used, instructions \nrequired to locate the desired information are set \ninto the machine on switches and keys located on\n\nFig. 2 \u2014 A typical straddle record with key to iden- \ntify each numerical group in the various entries.\n\nFig. 4\u2014 The author introduces into the control \npanel information which will initiate the scanning \nfeature of the printer-comparer-scanner.\n\nthe control panel (see Figure 4). When the tape \nis inserted, the machine starts scanning the tape a \nline at a time until the desired information is found. \nThe tape moves along at the rate of about eight \nfeet per minute. When the desired information is \nencountered, the machine blocks and sounds an \nalarm to attract the attention of the attendant. At \nthis point, the attendant operates a key to obtain a \nprinted record of the information.\n\nOne possible use of the scanning function pro- \nvides an alternative to the printing of the intermedi- \nate summary tapes which was discussed earlier. \nWhen message-unit summary information is re- \nquired for billing a particular customer, the four \nswitches, shown in the lower left row of Figure 4, \nare set to correspond to his line number. The tape \ncontaining his record is inserted and the machine \nstarts scanning at its usual rate of eight feet per\n\nFig. 5\u2014A_ printed \nrecord of the scanned \ntape. Customers\u2019 line \nnumbers appear in \nleft column and mes- \nsage unit charges ap- \npear in right column,\n\nit encounters either the desired line number or a \nhigher line number. This feature makes it unnec- \nessary to scan the entire tape if there is no entry \ncorresponding to the desired line number.\n\nAs an example, suppose that information on line \nnumber 3953 is desired. All numbers from 0000 to \n3952 will be passed over and the machine will stop \non 3953 or, if 3953 is missing, on 3954 or the next \nnumber higher than 3953 which it encounters. Lamp \nindications inform the attendant whether the ma- \nchine has stopped because the desired number has \nbeen found or because a higher number has been \nencountered. Under key control, the customer's \nline number and his message units are printed. The \nmachine can then be reset to scan for some higher \nnumber. A tape containing message-unit summary \ninformation for 10,000 customers can be scanned \nin less than half an hour, whereas four hours would\n\nbe required to print the complete information. The \nprinted record is shown in Figure 5. The informa- \ntion is arranged in two columns with the customers\u2019 \nline numbers appearing in numerical order in the \nleft hand column and the associated message unit \ncharges in the right hand column.\n\nMany other scanning functions can be performed, \nand in every case the procedure is the same. The \nrequired data are set into the machine and the tape \nis scanned until the information is found. The ma- \nchine can be set to step a message register, instead \nof stopping when it encounters the desired infor- \nmation, and to continue scanning for subsequent \nappearances of the same information. A possible \nuse of this feature might be to determine the \namount of traffic handled by a given trunk in a \ncentral office. New applications of the printer- \ncomparer-scanner are continuously being studied.\n\nafter receiving her A.B. degree from Vassar College. She has been mainly con- \ncerned with AMA development, primarily circuit. design and laboratory testing \nof the accounting center machines used in this system. She participated in ac- \ncuracy tests in 1948 at Philadelphia where the first No. 5 crossbar installation \nwith AMA was made, and a year later in similar tests when No. 1 crossbar and \nAMA were used together for the first time. She also participated in pre-cutover \ntests of nationwide customer dialing at Newark and Englewood in 1951. After \na short period devoted to design of crossbar tandem circuits for use with AMA, \nshe returned to accounting center circuit design work to develop the Printer- \nComparer-Scanner, From 1955 until she left the Laboratories recently, she was \nengaged in designing circuits for the Common User Group, part of the air-ground \ncommunication system for SAGE.\n\nThe widely acclaimed Bell System Science Series \nprogram, \u201cOur Mr. Sun\u201d, will be telecast in color \nover the NBC television network on Sunday, De- \ncember 15, at 5:30 P.M., EST (4:30 P.M., CST; \n3:30 P.M., MST; and 5:30 P.M., PST). Starring Dr. \nFrank Baxter as \u201cDr. Research\u201d and Eddie Albert \nas \u201cFiction Writer\u201d, the program dramatizes many \nfacts about the sun and shows scientists at work on \nsolar studies throughout the world.\n\nFirst telecast in November, 1956, \u201cOur Mr. Sun\u201d \nwill be the second in a series of four Bell System\n\nScience programs over NBC-TV for 1957-58. \u201cThe \nStrange Case of the Cosmic Rays\u201d was seen on \nOctober 25, and \u201cThe Unchained Goddess,\u201d a new \nprogram about weather, will be seen on Wednesday, \nFebruary 12. \u201cHemo the Magnificent\u201d, about the \nblood and its circulation, will be seen for the sec- \nond time on Sunday, March 16. All four of these \nprograms were produced and directed by Academy \nAward Winner Frank Capra. In addition to their \nuse on television, these films have been shown to \nabout six million students in classrooms during 1957.\n\nW. O. Baker, Vice-President in charge of research \nat Bell Laboratories, received the honorary degree \nof Doctor of Science from Washington College on \nOctober 20. The degree was awarded in Chester-\n\nDr. Daniel Z. Gibson, right, President of Washing- \nton College, presenting honorary Doctor of Science\n\ntown, Maryland, at a convocation marking the 175th \nanniversary of the College, of which George Wash- \nington was the first benefactor.\n\nCollege President Daniel Z. Gibson made the \npresentation and cited Dr. Baker as \u201cone of the most \ndistinguished young scientists of the United States.\u201d \nThe citation continued by recognizing Dr. Baker \nas \u201can authority on the chemistry of high polymers \nand in the development of synthetic rubber. . . . He \nhas served as visiting lecturer at Princeton, North- \nwestern, Western Reserve and Brooklyn Polytechnic \nInstitute, is a consultant for the Office of Naval Re- \nsearch and the Army Quartermaster Corps, and is a \nmember of the Visiting Committee in Chemistry \nat Princeton University, and the Science Advisory \nCommittee of the Executive Office of the Presi- \ndent. We are proud to welcome home a native \nson and to recognize his distinguished achieve- \nments.\u201d\n\nDr. Baker was graduated from Washington Col- \nlege in 1935 and continued study at Princeton, \nwhere he received the Ph.D.\n\nJ. R. Townsend, formerly of the Laboratories and \nnow special assistant to the Assistant Secretary of \nDefense for Research and Engineering, received \none of the highest awards of the American Stand- \nards Association on November 14. He was awarded \nthe gold Standards Medal at the Association\u2019s An- \nnual Award Dinner during the Eighth National \nConference on Standards held in San Francisco. \nThis Medal is awarded for leadership in the actual \ndevelopment and application of standards.\n\nWith the Bell System for more than 38 years, Mr. \nTownsend is a nationally known materials expert \nwho has been called on frequently to render service \nto the highest levels of government. On leave from \nBell Laboratories since 1952, he served as Director \nof Materials Application Engineering for the Sandia \nCorporation until his recent appointment to the \ngovernment post. At the Laboratories, Mr. Town- \nsend was active in the initiation and promotion of \nthe materials testing laboratories as well as the \nx-ray, optical, welding and metallurgical labora- \ntories. Among his many official positions, he has \nserved as President of the American Society for \nTesting Materials and as a member of the A.S.A. \nBoard of Directors.\n\nThe September, 1957, issue of THe BELL SysTEM \nTECHNICAL JOURNAL contains the following articles:\n\nOceanographic Information for Engineering Sub- \nmarine Cable Systems by C. H. Elmendorf and \nB. C. Heezen.\n\nResistance of Organic Materials and Cable Struc- \ntures to Marine Biological Attack by L. R. Snoke.\n\nDynamics and Kinematics of the Laying and Re- \ncovery of Submarine Cable by E. E. Zajac.\n\nTheory of Curved Circular Waveguide Contain- \ning an Inhomogeneous Dielectric by S. P. Morgan.\n\nCircular Electric Wave Transmission Through \nSerpentine Bends by H. G. Unger.\n\nCioffi, P. P.\u2014 Traveling Wave Tube Apparatus Including \nMagnetic Structures \u2014 2,807,743.\n\nEdwards, C. F.\u2014 Impedance Matching Devices for Wave- \nGuide Hybrid Junctions \u2014 2,806,210.\n\nArcher, R. J., Optical Measurement of Film Growth on Sili- \ncon and Germanium Surfaces in Room Air, J. Electro- \nchem., Soc., 104, pp. 619-622, Oct., 1957.\n\nBaker, A. N., and Webber, D. S., Hydrogen Vibration Spec- \ntra of Rochelle Salt, J. Chem. Phys., 27, pp. 689-692, \nSept., 1957.\n\nBurns, F. P., and Fleischer, A. A., Piezoresistive Effect in \nIndium Antimonide, Phys. Rev., 107, pp. 1281-1282, Sept. \n1, 1957.\n\nChynoweth, A. G., The Pyroelectric Behavior of Coleman- \nite, Acta Crystallographica, 10, pp. 511-514, Aug., 1957.\n\nDunn, H. K., and Young, R. W., On the Interpretation of \nCertain Sound Spectra of Musical Instruments, J. Acous. \nSoc. Amer., 29, pp. 1070-1073, Oct., 1957.\n\nEllis, W. C., Williams, H. J., and Sherwood, R. C., Evidence \nfor Subgrains in MnBi Crystals from Bitter Patterns, \nJ. Appl. Phys., Letter to the Editor, 28, pp. 1215-1216, \nOct., 1957.\n\nFeher, G., Fuller, C. S. and Gere, E. A., Spin and Magnetic \nMoment of P\u00ae by the Electron Nuclear Double-Resonance \nTechnique, Phys. Rev., Letter to the Editor, 107, pp. \n1462-1464, Sept. 1, 1957.\n\nFeinstein, J., and Kino, G. $., The Large-Signal Behavior of \nCrossed-Field Traveling-Wave Devices, Proc. I.R.E., 45, \npp. 1364-1373, Oct., 1957.\n\nFollowing is a list of the authors, titles and places of publication \nof recent papers published by members of the Laboratories:\n\nFleischer, A. A., see Burns, F. P. \nFoster, F. G., see Williams, H. J. \nFuller, C. S., see Feher, G. \nGere, E. A., see Feher, G.\n\nGeller, S., and Gilleo, M. A., The Crystal Structure and \nFerrimagnetism of Yttrium-Iron Garnet, Y,Fe.(FeO,)3, \n]. Phys. and Chem. Solids, 3, (Nos. 1-2), pp. 30-36, 1957.\n\nGordon, J. P., and White, L. D., Experimental Determina- \ntion of the Noise Figure of an Ammonia Maser, Phys. \nRev., Letter to the Editor, 107, pp. 1728-1729, Sept. 15, \n1957.\n\nHobstetter, J. N., and Breidt, P., Jr., Detection of Both Va- \ncancies and Interstitials in Deformed Germanium, J. Appl. \nPhys., Letter to the Editor, 28, pp. 1214-1215, Oct., 1957.\n\nKelley, E. M., see Williams, H. J. \nKino, G. S., see Feinstein, J. \nKisliuk, P., The Sticking Probabilities of Gases Chemisorbed\n\nLax, M., Frequency Dependence of the AC Resistance of \nThin Semiconducting Films, Phys. Rev., 107, pp. 650-655, \nAug. 1, 1957.\n\nMatthias, B. T., and Corenzwit, E., Superconducting Alka- \nline Earth Compounds, Phys. Rev., 107, pp. 1558-1559, \nSept. 15, 1957.\n\nMatthias, B. T., and Remeika, J. P., Ferroelectricity of Dical- \ncium Strontium Propionate, Phys. Rev., Letter to the Edi- \ntor, 107, p. 1727, Sept. 15, 1957.\n\nMertz, P., Information Theory Impact on Modern Communi- \ncations, Comm. and Electronics, 32, pp. 431-437, Sept., \n1957.\n\nPfann, W. G., Techniques of Zone Melting and Crystal \nGrowing, Solid State Physics, Vol. IV, Academic Press, \nNew York, 1957.\n\nRomanow, W. J., Crystallographic Angles for Manganese \nBismuthide, |. Metals, 209, p. 1284, Oct., 1957.\n\nTreuting, R. G., and Arnold, S. M., Orientation Habits of \nMetal Whiskers, Acta Metallurgica, 5, p. 598, Oct., 1957.\n\nWernick, J. H., Effects of Crystal Orientation, Temperature \nand Molten Zone Thickness in Temperature Gradient \nZone Melting, J. Metals, 9, pp. 1169-1173, Oct., 1957.\n\nWernick, J. H., and Benson, K. E., New Semiconducting \nTernary Compounds, J. Phys. Chem. Solids, Letter to the \nEditor, 3, (Nos. 1-2), pp. 157-159, 1957.\n\nWilliams, H. J., Sherwood, R. C., Foster, F. G., and Kelley, \nE. M., Magnetic Writing on Thin Films of MnBi, J. Appl. \nPhys.. 28, pp. 1181-1184, Oct., 1957.\n\nWooley, M. C., Passive Components for Submarine Cable \nTelephone Repeaters, 1.R.E., Trans. on Reliability and \nQuality Control, pp. 14-24, Nov., 1957.\n\nFollowing is a list of talks given before professional and \neducational groups by Laboratories people during October.\n\nAmron, I., Corby, W. J., Craft, W. H., Koontz, D. E., and \nPondy, P. R., A Method of Wide Applicability for Clean- \ning and Etching Electronic Materials.\n\nBurcham, N. P., and Miller, L. E., Stabilization of Germani- \num Alloy Junction Transistor Characteristics Through Sur- \nface Treatment.\n\nClosson, H. T., Danielson, W. E., and Nielsen, R. ]., Auto- \nmatic Measurement of Micro-Deviations in Periodic Struc- \ntures,\n\nEarly, J. M., and Sevick, J., Characteristics and Structure of a \nDiffused-Base Germanium High Frequency Amplifier\n\nFeder, D. O., Craft, W. H. and Koontz, D. E., Simple Tech- \nniques for Storing Ultraclean Electron Tube Components.\n\nDavid, E. E., Jr., Guttman, N., and van Bergeijk, W. A., \nSome Factors Governing the Lateralization of High Fre- \nquency Complex Waveforms, (Presented by W. A. van \nBergeijk).\n\nMason, W. P., and Bommel, H., Frequency and Temperature \nDependance of Internal Friction in Pure Copper.\n\nMcSkimin, H. J., Measurement of Dynamic Shear Im- \npedance of Liquids at High Ultrasonic Frequencies.\n\nvan Bergeijk, W. A., see David, E. E., Jr. \nWeinreich, G., Acoustoelectric Effect in Germanium.\n\nForster, J. H., and Veloric, H. S., The Effect of Variations \nin Surface Potential on Junction Characteristics.\n\nHoward, B. T., Phosphorus Diffusion in Silicon. \nSingleton, J. B., and Silverman, S. J., Techniques for Pre-\n\nSmith, K. D., Semiconductor Materials and Processes. \nStruthers, J. D., see Bemski, G.\n\nAbraham, R. P., see Kirkpatrick, R. J. \nBlecher, F. H., Transistor Feedback Amplifiers.\n\nHewitt, W. H., and von Aulock, W. H., A Reciprocal Ferrite \nPhase Shifter for X-Band.\n\nMcDavitt, M. B., 6,000 Mc/Sec Radio Relay System for \nBroad-Band, Long Haul Service in the Bell System.\n\nScheinman, A. H., A Numerical-Graphical Method for Syn- \nthesizing Switching Circuits.\n\nBaker, A. N., Recent Developments in Semiconductors, New \nYork Telephone Company, White Plains, N. Y.\n\nBashkow, T. R., and Desoer, C. A., Digital Computers and \nNetwork Theory, 1.R.E.-PG Conference on Circuit Theory, \nSyracuse, N. Y.\n\nBenson, R, J., Wear Studies of Fine-Pitch Gear Materials, \nSemi-annual Meeting, American Gear Manufacturers As- \nsociation, Chicago, Ill.\n\nBlecher, F. H., Properties of Junction Transistors, Chicago \nSection of I.R.E., Chicago, Ill.\n\nBleicher, E., Life Testing of Electronic Components, The \nAmerican Society for Quality Control, Bridgeport, Conn.\n\nBradley, W. W., The Sea Bottom as an Environment for \nMetals, National Association of Corrosion Engineers, San \nDiego, Calif.\n\nChapin, D. M., Solar Energy and the Bell Solar Battery, \nAmerican Society of Mechanical Engineers, New London, \nConn.\n\nCompton, K. G., Sources of Underground Corrosion Poten- \ntials, S. E. Regional Meeting, National Association of Cor- \nrosion Engineers, Birmingham, Ala.\n\nDickieson, A. C., Trends in Transmission Systems Develop- \nment at Bell Telephone Laboratories, Tokyo Section of \nLR.E., Tokyo, Japan.\n\nFerrell, E. B., The Control Chart \u2014 Modifications and Ex- \ntensions, Metropolitan Section of American Society for \nQuality Control, Newark, N. J.\n\nFisher, C. E., Quality Assurance, Waco Section, American \nSociety for Quality Control, Waco, Texas; San Antonio \nSection, San Antonio, Texas; Dallas-Ft. Worth Section, \nFt. Worth, Texas, and South Texas Section, Houston, \nTexas.\n\nGalt, J. K., Cyclotron Absorption in Bismuth and Graphite, \nRutgers University, New Brunswick, N. J.\n\nGause, G. R., The Quality Survey -- An Essential Part of \nan Overall Quality Program, Montreal Section of Ameri- \ncan Society for Quality Control, Montreal, Canada.\n\nGeils, J. W., Research and Development in Industry Today, \nNewark College of Engineering, Newark, N. J.\n\nGraham, R. E., Review of Paris Symposium on Physical \nProblems of Color Television, Society of Motion Picture \nand Television Engineers Convention, Philadelphia, Pa.\n\nHarmon, L. D., Computer Simulation of Pattern Recogni- \ntion, Symposium on Pattern Recognition, University of \nMichigan, Ann Arbor, Mich.\n\nHarvey, F. K., The Physics of Hearing and Music, New Jer- \nsey Division of New York Section, A.LE.E., Allenhurst, \nN. J.\n\nHawkins, W. L., The Behavior of Antioxidants in Autoxida- \ntion, Chemical Symposium, Stevens Institute, Hoboken, \nN. J.\n\nHebel, L. C., Jr., Nuclear Spin Relaxation in Superconduc- \ntors, 1.B.M. Watson Laboratory, New York City and Phy- \nsics Colloquium, Princeton University, Princeton, N. J.\n\nHelm, H. A., The Analysis of Digital Control Systems, Con- \nference on Computers in Control, A.IE.E., Atlantic City, \nN. J.\n\nHerbst, R. T., Computer Programing Techniques, 2nd An- \nnual Symposium, Piedmont Subsection, L.R.E., Greens- \nboro, N. C.\n\nHornbeck, J. A., Device Research at Bell Telephone Labora- \ntories, Joint Session, Tokyo Section I.R.E.-Institute Elec- \ntrical Communications Engineers of Japan, Tokyo, Japan.\n\nIngram, S. B., Graduate Training for the Young Engineer at \nBell Telephone Laboratories, 4th General Assembly, Joint \nMeeting of the Engineers Council for Professional De- \nvelopment and the Engineers Joint Council, New York \nCity.\n\nIngram, S. B., The Utilization and Training of Engineers in \nIndustry, Management Division, Cleveland Engineering \nSociety, Cleveland, Ohio.\n\nKatz, D., Magnetic Regulated Power Supplies, Northern \nNew Jersey Section, I.R.E., Montclair, N. J.\n\nLaudise, R. A., Hydrothermal Growth of Oxide Crystals, \nConference on Growth and Evaluation of Single Crystals \nof Ferrimagnetic Ceramics, Air Force Cambridge Research \nCenter, Bedford, Mass.\n\nLawson, C. C., Buried Distribution of Telephone Circuits, \nAnnual Convention of the Wire Association, Chicago, Ill.\n\nLegg, V. E., Basic Magnetic Components, Northern New \nJersey Section, I.R.E., Montclair, N. J.\n\nLundberg, J. L., Pyrolyses and Photocyses in the Flash Illu- \nmination of Polymers, Chemistry Department Seminar, \nNew York University, New York City.\n\nMacNair, W. A., Industrial Research and Development, U.S. \nArmy Signal School, Ft. Monmouth, N, J.\n\nMcClure, B. T., Total Ionization by Low Energy Electrons \nin Neon, Gaseous Electronics Conference, Cambridge, \nMass.\n\nMcMahon, W., Birdsall, H. A., Johnson, G. R., and Camilli, \nC. T., Degradation Studies of Mylar, Conference on Elec- \ntrical Insulation, Pocono Manor, Pa.\n\nMayo, J. S., Analysis of an On-Off Digital Control System, \nSymposium on Feedback Control, A.LE.E., Atlantic City, \nN. J.\n\nMoore, E. F., A General Introductory Survey of Automata, \nInformal Seminar on Automatic Computers and_ their \nCapabilities, University of Pennsylvania, Philadelphia, Pa.\n\nMoore, G. E., Physics in Industrial Laboratories, Physics \nClub, Fordham University, New York City.\n\nMorrison, J., Gas Collection and Analysis System Employed \nin Vacuum Tube Problems, 1957 4th National Vacuum \nSymposium, Boston, Mass.\n\nNielsen, J. W., The Use of Molten Lead Oxide as a Solvent \nfor Magnetic Crystals, Symposium on Crystal Growth, Air \nForce Cambridge Research Center, Bedford, Mass.\n\nPierce, J. R., Fancies and Fallacies of Space Travel, Joint \nLR.E., A.LE.E. and American Rocket Society, New York \nSection, New York City; Griffiiss Air Force Base Club, \nRome Air Development Center, Rome, N. Y.\n\nPierce, J. R., The Challenging Field of Engineering Writing \nand Speech, 1st National Symposium, I.R.E.-PG on Engi- \nneering Writing and Speech, New York City (Presented \nby H. S. Black).\n\nPfann, W. G., New Methods of Purifying Solid Materials, \nStuyvesant High School, New York City.\n\nPollak, H. O., The Problem of Minimal Connecting Net- \nworks, Computation Laboratory, National Bureau of \nStandards, Washington, D, C.\n\nSchawlow, A. L., The Intermediate State of Superconduc- \ntors, Physics Colloquium, Johns Hopkins University, Bal- \ntimore, Md.\n\nSchroeder, M. R., Artificial Stereophony Using Single Input, \n1957 Convention of the Audio Engineering Society, New \nYork City.\n\nScovil, H. E. D., The Solid State Maser, PGMTT, PGED \nand Northern New Jersey Section, I.R.E., Murray Hill, \nNew Jersey.\n\nSharpe, L. H., Electrolytic Regeneration of Cupric Chloride \nEtching Solutions, Joint Bel! Telephone Laboratories- \nWestern Electric Company, Printed Wiring Symposium, \nGreensboro, N. C.\n\nSobel, M., On a Nonparametric Definition of the Represen- \ntativeness of a Sample with Tables for Applications, Co- \nlumbia University, New York City.\n\nStephens, Miss S. J., Preparation and Properties of Clean \nHigh-Area Metal and Alloy Evaporated Films for Use in \nSurface Studies, 4th National Vacuum Symposium, Boston, \nMass.\n\nWarthman, K. L., Bell Telephone Laboratories Optical \nTracking System for NIKE Radar, Society of Motion \nPicture and Television Engineers, Philadelphia, Pa.\n\nWeinreich, G., Acoustoelectric Effect in Germanium, Physics \nColloquium, Yale University, New Haven, Conn.; Uni- \nversity of Minnesota, Minneapolis, Minn.\n\nWenk, H. A., Trips toe DEW Line and White Alice, Mountain \nLakes School, Mountain Lakes, N. J.\n\nWilkinson, R. L, Some Engineering Applications of Queue- \ning Theory, Administrations Applications Conference, \nAmerican Society for Quality Control, Columbia Uni- \nversity, New York City.\n\nWintringham, W. T., Tailoring a Facsimile System to Its \nApplication, Savings Bank Research Group, New York \nCity,\n\nWright, S. B., Life Along the DEW Line, Circa 1957, St. \nCloud Presbyterian Couples Club, West Orange, N. J. \n(Presented by R. C. Newhouse. )\n\nYounker, E. L., Storage Devices: Characteristics and Tech- \nniques, 2nd Annual Symposium, Piedmont Subsection, \nLR.E., Greensboro, N. C.\n\nOne method of thermo-compression bond- \ning. A heated wedge presses a wire against a \nheated semiconductor with enough force to \ndeform the wire. Adhesion occurs in seconds.\n\nThermo-compression bonding pro- \nvides a new way to attach a wire \nto a semiconductor. It calls for \nheat and pressure \u2014 nothing else. \nThe wire and the semiconductor \nare moderately heated, then pressed \ntogether under moderate pressure. \nThe resulting bond is very strong\u2014 \nstronger actually than the wire. No \nchemical flux or molten metal is \nrequired.\n\nEliminating molten metal pro- \nvides an enormous advantage in \nfixing electrical connections to \ntransistors. That\u2019s because molten \nmetal tends to spatter and spread \nuncontrollably over the surface of \na semiconductor. And it may alloy\n\nWire bonded to germanium by thermo-compression technique (en- \nlarged). Wires only 1/10 the breadth of a human hair have been \nsuccessfully anchored to germaniun) wafers only three hairs thick. \nThe bond may be an ohmic contact or rectifying contact by adding \u2014 \nsuitable impurities to the wire and the semiconductor.\n\nwith the semiconductor to alter iis \nall-important crystalline structure \nand chemical purity. Thermo-com- \npression bonding easily and quickly \nmakes a strong permanent electri- \ncal connection without damaging \nthe semiconductor. Furthermore, \nthe lead may be attached to micro- \nscopic areas and precisely posi- \ntioned, a most valuable aid in the \nconstruction of high-frequency \ntransistors.\n\nThermo-compression bonding \nwill speed the production of tran- \nsistors . . . the transistors needed to \nfill all the new jobs Bell Labora- \ntories finds for them in the quest \nfor still better telephony.\n\nAt Bell Labs Howard Christensen and \nOrson Anderson discuss their discov- \nery of new bonding principle with \nPeter Andreatch, Jr., who collaborated \nin the studies.", "title": "Bell Laboratories Record  1957-12: Vol 35 Iss 12", "trim_reasons": [], "year": 1957}
{"archive_ref": "CPS-JK17", "canonical_url": "https://archive.org/details/CPS-JK17", "char_count": 9333, "collection": "archive-org-bell-labs", "doc_id": 1116, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1116", "record_count": 10, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/CPS-JK17", "split": "test", "text": "WHEN CHANGES ARE MADE IN THIS DRAWING \nONLY THOSE SHEETS AFFECTED WILL BE \nREISSUED.\n\nTHIS SHEET INDEX WILL BE REISSUED ANC \nBROUGHT UP 10 DATE EACH TIME ANT SHEET \nOF THE DRAWING IS REISSUED. OR A NEW \nSHEET IS ADDED.\n\nTHE ISSUE NUMBER ASSIGNED TO A CHANGED \nOR NEW SHEET WILL BE THE SAME ISSUE \nNUMBER AS THAT OF THE FIRST SHEET.\n\n5. THE LAST ISSUE NUMBER Of THE FIRS' SHEET \nINOEX IS RECOGNIZED AS THE LATEST ISSUE \nNUMBER OF THE DRAWING AS A WHOLE.\n\nRESISTANCE VALUES ARE IN OHMS \nCAPACITANCE VALUES ARE IN MICROFARADS \nVALUES PRECEDED BY THE SYMBOL *(PLUS) \nOR -(MINUS) ARE IN VOLTS\n\nNOT FOR USE OR DISCLOSURE OUTSIDE THE BELL \nSYSTEM EXCEPT UNDER WRITTEN AGREEMENT.\n\nTHE READ CIRCUIT CAN OPERATE IN EITHER OF TWO MODES, \nNORMAL DATA TRANSFER OR MAINTENANCE DATA TRANSFER. IN \nTHE NORMAL MODE DATA IS TRANSFERRED FROM THE CARTRIDGE \nTAPE TRANSPORT (CTT). IN THE MAINTENANCE MODE, THE REV \nANO WRITE CIRCUITS CAN BE EXERCISED WITHOUT OPERATING \nTHE CTT. DATA WILL ENTER DIRECTLY FROM THE OUTPUT OF \nTHE WRITE CIRCUIT LOCATED ON JK18 (SEE FIG. 1). THE \nDATA AND ITS ASSOCIATED CLOCK ENTERS THE READ CIRCUIT \nTHROUGH THE DATA SELECTOR ANO THE CLOCK SELECTOR \nCIRCUITS, RESPECTIVELY. THE DATA CLOCKING CONTROL \nCIRCUIT WHICH RECEIVES CLOCK PULSES FROM BOTH THE CLOCK \nSELECTOR CIRCUIT AND A FREE RUNNING CLOCK (OPERATING AT \nTHE NORMAL DATA FREQUENCY) HAS TWO BASIC CLOCK OUTPUTS. \nTHE OUTPUT LABELED SHIFT PULSES BEGINS CLOCKING THE 24- \nBIT DATA REGISTER ANO THE DATA FRAME CIRCUITRY (THE DATA \nFRAME CIRCUITRY INCLUDES LOGIC FOR STRIPPING THE PRE- \nAMBLE AND POSTAHBLE) AS DATA INITIALLY ENTERS THE READ \nCIRCUIT. AS THE PREAMBLE (FIFTEEN ZEROS FOLLOWED BY A \nSINGLE ONE) IS SHIFTED THROUGH THE 24-BIT REGISTER, THE \nOUTPUT OF THE DATA FRAME CIRCUIT INHIBITS CLOCK PULSES \nTO EOTH THE C?C CHECK CIRCUIT AND THE BUFFER CIRCUITS \n(THE BUFFERS ARE CLOCKED BY THE LEAD LABELED CLOCK OUT). \nBY MONITORING THE OUTPUT OF THE 24-BIT REGISTER, THE \nDATA FRAME CIklUIT DETECTS THE FIRST ONE (LAST BIT OF \nTHE PREAMBLE) AND THEN ENABLES THE CRC CHECK CIRCUIT \nAND THE CLOCK OUT LEAD. THIS ALLOWS THE FOLLOWING DATA \nTO BE CLOCKED INTO THE CTC CHECK CIRCUIT AND TO THE \nBUFFER CIRCUITS. THIS IN EFFECT, STR I PS THE PREAMBLE \nFROM THE DATA BY SIMPLY IGNORING IT AS IT FALLS OUT THE \nEND OF THE 24-B'T REGISTER.\n\nAS IN A NORMAL READ OPERATION. WHEN THE DATA FRAME \nCOUNTER REACHES A COUNT OF 16, ALL OPERATIONS WILL \nBE TERMINATED, LEAVING THE BUF FILL LEAD ENABLED.\n\nTHE READ DATA CLOCK LEAD (RDCLKO) WHICH COMES FROM THE \nCTT. FEEDS THE 1>is HONOPULSER RDCLK1. THIS MONOPULSER \nPROVIDES A DELAY T. ENSUR r THAT THE DATA LEAD (RDDATO) \nFROM THE CTT IS STABLE BEFORE CLOCKING INFORMATION INTO \nTHE OVERROW REGISTER 0FLREG. THE ZERO OUTPUT OF RDCLK1 \nSERVES AS A TEST POINT WHICH IS BROUGHT OUT TO A CONNEC- \nTOR TERMINAL. THE ONE OUTPUT OF RXLK1 FEEDS THE NAND \nGATE TSTO WHICH IS ENABLED BY BOTH MAINTO ANO ENRD1. \nMAINTO IS AN OUTPUT OF THF MAINTENANCE RIP-ROP ON JK16. \nIT IS AT A HIGH LEVEL WHEN T HE CTTC IS NOT IN A MAIN- \nTENANCE MOOE. ENRD1 IS AN OUTPUT OF THE READ RIP-ROP \nON JK16. IT IS AT A HIGH-LEVEL WHENEVER THE CTTC IS \nHANDLING DATA (REAO, WRITE AND CRC SHIFT OPERATIONS). \nTSTO FEEDS RDST1 WolCH CLOCKS THE DATA THROUGH THE THREE \n8-BIT REGISTERS SSRREG, PSTREG1A, ANO PSTREG1B. THE \nDATA !S FED INTO BRREG BY RDATA1 WHICH PASSES DATA FROM \nEITHER THE CTT OR FROM THE WRITE CIRCUIT (JK18) DEPEND- \nING UPON THE ENABLING OF EITHER TDATAO OR HOATAO BY MAINTO OR \nMAINT1, RESPECTIVELY. HAINT1 IS THE SECOND OUTPUT OF THE \nMAINTENANCE FLIP-FLOP ON JK16.\n\nIN THE MAINTENANCE MCOE THE READ AND WRITE CIRCUITS CAN \nBE EXERCISED WITHOUT OPERATING THE CTT. MAINTO WILL BE \nAT A LOW LEVEL AND HAINT1 WILL BE AT A HIGH LEVEL. THIS \nENABLES A DATA PATH FROM WDATA1 AND A CLOCK PATH FROM \nWB11 THROUGH MSTO. THIS STATE DISABLES THE OATA AND \nCLOCK PATHS FRC? THE CTT. WDATA1 ANO W011 ARE DATA AND \nASSOCIATED CLOCK LEADS FROM THE OUTPUT OF THE WRITE \nCIRCUIT.\n\nINPUT DATA WHETHER CLOCKED FROM THE WRITE CIRCUIT OR FROM \nTHE CTT IS CLOCKED THROUGH THE THREE REGISTERS BRREG, \nPSTREG1A, AND PSTREG1B. THESE ARE 41T 8-BIT -1HIFT \nREGISTERS WHIG\" MAKE UP A 24-BIT SERIAL SHIFT REGISTER. \nTHE OUTHJT OF THIS REGISTER RDDATAB IS ANDeo WITH ROST1 \nBY DAFRENO. DAFRENO WILL PROVIDE A TOGGLE PULSE WHICH \nCLEARS DAFR ON THE FIRST REAO-CLOCK PULSE /'FTER RDDATAB \nGOFS TO THE ONE STATE. THE FIRST WORD OF A DATA 8; OCK \nIS THE PREAMBLE (15 \"ZEROS\" FOLLOWE3 BY A SINGLE \"ONE\"). \nTHEREFORE, THE CLEARING OF DAFR IMPLIES THAT THE FIRST DATA \nBIT FOLLOWING THE PREAMBLE IS AT THE OUTPUT OF THE 24-BiT \nREGISTER. WHEN THE CTTC IS IN THE MAINTENANCE MODE, \nTHE FIRST WORD FROM THE WRITE CIRCUIT IS ALSO THE PRE- \nAMBLE. ALL DATA-CLOCK PULSES WHICH FEED EITHER THE \nCRC CIRCUIT OR THE BUFFER CIRCUITS WJST PROPAGATE \nTHROUGH THE NAND GATE, RDSTG. RDSTO HAD BEEN INHIBIT- \nED BY THE SET STATE ON DAFR UNTIL AFTER THE LAST BIT OF \nTHE PREAMBLE. SliiCE NO CLOCK PULSES ARE PERMITTED \nTHROUGH RDSTO, UNTIL AFTER THE PREAMBLE HAS BEEN SHIFTED \nFROM THE 24-SIT REGISTER, THE PREAMBLE IS STRIPPED FROM \nTHF DATA STREAM. THE CLEARED STATE OF DAFR ENABLES \nR^STO. THE OUTPUT OF RDSTO PROVIDES DATA CLOCK PULSES \nTO THE 16-STATE WORD FRAMING COUNTER f WFRCT), THE READ \nSHIFT PULSE MONOPULSER (RSTPUL), THE CLOCK INPUTS TO THE \nCRC REGISTER, AND SHCRC1.\n\nIN THE NORMAL DATA TRANSMISSION MODE, DATA WHICH APPEARS \nAT THE Q8 OUTPUT OF PSTREG1B PROPAGATES THROUGH RDDATO, \nTD0UTO, AND DflUTI TO THE D CELL OF CRCRGA. IT IS LOADED \nONTO THIS RIP-ROP BY SHCRCO. THE OUTPUT OF THIS CELL \n(RDBUT1) IS THE DATA OUTPUT SIGNAL FROM THE CTTC'S RE/1 \nCIRCUITRY. SHCRCO ALSO CLOCKS THE Q8 OUTPUT OF PSTRE-IB \nTHROUGH THE CRC ERROR DETECTION CIRCUIT.\n\nAS DATA IS BEING SHIFTED THROUGH THE CRC CIRCUITS AND THE \nOUTPUT RIP-ROP, RDSTO IS ALSO PROVIDING CLOCK PULSES \nTO THE 1iis MONOPULSER, RSTPUL AND THE 16-STATE WORD FRAME \nCOUNTER, WFRCT. THE OUTPUT OF RSTPUL (RSP1) SERVES AS \nTHE REAO-OATA CLOCK TO THE BUFFER CIRCUIT. DATA IS CLOCKED \nINTO THE BUFFERS ON THE TRAILING EDGE OF RSP1 . THIS ALLOWS \nDATA FROM PSTREG1B TO BE SETTLED IN THE D CELL OF CRCRGi \nBEFORE THE BUFFER IS CLOCKED.\n\nWHEN THE READ CIRCUIT IS IN THE IDLE STATE (NO DATA IS \nSEING TRANSFERRED THROUGH IT), D\u00abD1 THE ONE OUTPUT \nOF THE REAL FLIP-ROP ON JK16 IS AT A LOW LEVEL. THIS LOW \nLEVEL MAINTAINS A SET STATE ON THE D-TYPE RIP-FLOP \nRADJEN AND BY DRIVING THE OUTPUT OF RDCL1 HIGH, IT MAIN- \nTAINS A LOW LEVEL AT THE OUTPUT OF RCLKO. THE LOW LEVEL \nAT THE OUTPUT OF RCLKO RETAINS A CLEARED STATE ON THE \nOUTPUT OF THE D-TYPE RIP-ROP ADJSYN AND A SET STATE \nAT THE OUTPUT OF DAFR. THE LOW LEVEL AT THE ZERO \nOUTPUT OF DAFR HOLDS WFRCT IN ITS CLEARED OR INITIAL \nSTATE. THIS ENSURES THAT THE COUNTER WILL BEGIN COUNTING \nAT THE ZERO STATE ON THE FIRST CLOCK PULSE AFTER THE \nPREAMBLE HAS FALLEN FROM PSTREG1B. AS A READ OPERATION \nBEGINS THE CLEAREO STATE OF ADJSYN ENABLES CLKINH AND \nDISABLES ENCLKO. THE ENABLED CLKINH ALLOWS READ-CLOCK \nPULSES FROM RDST1 TO PROPAGATE THROUGH ADJCLK1 TO THE \nINPUT OF ENCLKO WHICH IS DISABLED, AND TO ADJCLKD. THE\n\nIN THE MAINTENANCE STATE THE READ CIRCUIT OPERATES IN A \nVERY SIMILAR MANNER. MAINTO WILL BE LOW AND MAINT1 WILL \nBE HIGH. DATA THAT COMES DIRECTLY FROM THE WRITE CIRCUIT \n(JK18) WILL APPEAR ON WDATA1. THE CLOCK WILL APPFfR ON \nU011 (ALSO FROM JK18). AT THE END OF THE INPUT DATA \nRDADJ1 WILL BE DRIVEN HIGH BY THE WRITE CIRCUIT. THIS \nALLOWS THE REAO ADJUST CLOCK CLK011 TO PROPAGATE THROUGH \nRAWAOJO.\n\nTHE CRC CIRCUIT MONITORS ALL DATA READ INTO THE READ CIR- \nCUIT WHEN THE CTTC IS IN THE NORMAL OPERATING MODE, THE \nCRC REGISTER IS CLEARED BY A PULSE ON THE RDCLO (OUTPUT\n\nTRANSFER OPERATION. WHEN THE CTTC IS IN THE MAINTENANCE \nMODE THE CRC REGISTER IS CLEARED BY ISSUING A CLEAR CRC \nORDER WHICH APPEARS AS A PULSE ON THE CLRCRCCO INPUT \nLEAD. CLRCRCOO IS AN OUTPUT FROH THE CTTCs COMMAND DE- \nCODER LOCATED ON JK16. THE CRC REGISTER IS HADE UP OF: \nCELLS A.B.AND C OF CRCRGA, ALL OF CRCRGB.AND LRLt 1,2,5, \n4 ANO 5 OF CRCRGC. CRCRGA IS OPERATED IN A PARALLEL HOOE \nWHERE THE OTHER TWO REGISTERS ARE OPERATED IN THE SERIAL \nMODE. THE 16 BITS OF THE CRC REGISTER ARE ORGANIZED IN \nFOLLOWING MANNER; BITS 0,1, AND 15 ARE CRLS A,B , AND \nC OF CRCRGA, RESPECTIVELY. BITS 2,5, 4,5,6,7,8, MO 9 \nARE CELLS 1,2,5,4,5,6,7, AND 8 OF CRCRGB, RESPECTIVELY.\n\nBITS 10,11,12,15, AND 14 ARE CRLS 1,2,5,4, AND 5 OF CRCRGC, \nRESPECTIVELY. CRC1D1 PERFORMS AN EXCLUSIVE 0R FUNCTION ON' \nRDDATAB (INPUT DATA TO THE CRC CIRCUIT) AND BIT 15 OF THE \nCHECK REGISTER. THE OUTPUT OF CRC1D1 IS ENABLED THROUGH \nGATES CRCDENO AND CRCD1 BY SHEN. THE OUTPUT OF CRCD1 \nFEEDS BIT OF THE CHECK REGISTER, ANO IS EXCLUSIVELY 0Red \nWITH BIT 1 BY CRC501 TO FEED BIT i, AND BIT 14 BY CR15D1 \nWHICH FEEDS BIT 15 OF (HE CHECK REGISTER. THIS CIRCUIT \nIS A CYCLIC-REDUNDANCY CHECKER FOR THE GENERATOR POLYNGMINAL \n, + x 2 + x 15 \u2666 X 14 . ER1A0, ER1B0, ER1C0, ER1D0, ER1E0, \nER1F0, ER1G0, ER1H0, AND ESDET1\"0R\"THE 16-OUTPUT BITS OF \nTHE REGISTER TOGETHER AND INDICATE IF ANY OF THE 16 CELLS \nCONTAIN A \"ONE.\" (THiS INDICATES AN ERROS CONDITION \nAFTER AN ENTIRE BLOCK OF OATA HAS PASSED THROUGH THE \nCIRCUIT.) THE OUTPUT OF ER0ET1 IS ENABLED THROUGH \nCRCERO BY EREKIN1. ERENIN1 IS HIGH WHENEVER THE CTTC IS \nNOT IN A READ OR REAO-AFTER-WRITE MOOE (THE READ CIRCUIT \nIS NOT TRANSFERRING DATA).", "title": "JK17 Circuit Pack, cartridge tape transport, board B controller circuit", "trim_reasons": [], "year": 1974}
{"archive_ref": "CPS-JL16", "canonical_url": "https://archive.org/details/CPS-JL16", "char_count": 2615, "collection": "archive-org-bell-labs", "doc_id": 1156, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1156", "record_count": 4, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/CPS-JL16", "split": "test", "text": "UNLESS OTHERWISE SPECIFIED: \nRESISTANCE VALUES ARE IN OHMS \nCAPACITANCE VALUES ARE IN MICKOFARADS \nVALUES PRECEDED BY THE SYMBOL \u2666 (PLUS) \nOR - (MINUS) ARE IN VOLTS.\n\n2. FOR REISSUES, A CHANGED \nOR NEW SHEET IS ASSIGNED \nTHE SAME ISSUE NUMBER AS \nSHEET 1.\n\n. . THE ISSUE NUMBER OF \nSHEET 1 IF RECOGNIZED AS \nTHE ISSUE NUMBER OF THE \nWHOLE DRAWING.\n\nCIRCUIT DESCRIPTION/TIMING \nJL16 WRITE TiHING EXAMPLE \nADDRESS (CEA0I1 THROUGH CEA2I1)\n\nTHE JL16 MEMORY PLANE CONTAINS RANDOM-ACCESS STORAGE ELEMENTS \nMADE UP OF 16K BY 1 BIT (K \u25a0 1024) INSULATED GATE FIELD \nEFFECT TRANSISTOR (IGFET) fAJAL IN-LINE PACKAGES (OIPS) \nARRANGEO IN A 12BK BY 2-BI f ARRAY. THE MEMORY IS VOLATILE, \nIN THAT THE STORED INFORMATION WILL BE LOST IF POWER IS \nINTERRUPTED. AND DYNAMIC, IN THAT THE S T ,rfED INFORMATION \nMUST BE REGENERATED (REF\u00abSHEO) AT SPECIFIED INTERVALS. IN \nADDITION TO STORACE ELEMENTS, THE MEMORY PLANE CONTAINS \nADDRESS AND DATA BUFFERS, CONTROL LOGIC, DEVICE SELECT \nDECODERS. ALL INPUTS AND OUTPUTS ARE +5 V TTL COMPATIBLE.\n\nADORESS BITS CEA0I1 THROUGH CEA2I1 ARE GATED BY ONE OF THE \nTIMING SIGNALS RASIO AND ARE USED TO SELECT 2-0F-16 MEMORY \nDIPS. THE 16K RAM REQUIRES 14 ADDRESS BITS WHICH ARE \nMULTIPLEXED TO THE MEMORY PLANE ON INPUTS GRCAOO THROUGH \nGRCA60. THESE SIGNALS ARE BUFFERED AND CONNECTED TO ALL \nMEMORY D*PS. A SECOND TIMIN3 SIGNAL, CASIO, IS NECESSARY TO \nACHIEVE PROPER OPERATION. THIS SIGNAL IS BUFFERED INTO TWO \nBRANCHES WHICH EACH DRIVE EIGHT DEVICES IN THE MEMORY ARRAY.\n\nTHE MEMORY PLANE IS PUT INTO THE \"WRITE MODE\" BY FORCING A \nLOW LEVEL ON RWIO. DATA IS WRITTEN FROM INPUTS WRTDOIO AND \nWRT01I0 DURING THE TIME WHEN BOTH RASIO AND CASIO ARE LOW.\n\nDATA IS READ FROM THE MEMORY DEVICES WHEN RWIO IS HIGH AND \nBOTH RASIO AND CASIO ARE LOW. DATA IS THEN GATED TO OUTPUTS \nPTWV1 AND RD010. THE DATA OUT TRANSITIONS ARE SHOWN IN THE \nREAD TIMING EXAMPLE, CASIO STAYS HIGH DURING REFRESH CYCLES.\n\nEACH STORAGE CEIL ON THE MEMORY RANE MUST BE REFRESHED AT \nLEAST ONCE EVERY 4.4 ins, OR ELSE THE STORED CHARGE WILL LEAK \nOFF CAUSING THE DATA TO BE LOST. THIS REFRESH OPERATION CAN \nBE ACCOMPLISHED BY ENABLING THE RASIO INPUT IN THE SAME \nMANNER THAT IT IS PERFORMED DURING A READ CYCLE WHILE THE \nCASIO INPUT IS INHIBITED. THIS MUST BE PERFORMED FOR EACH \nOF THE 128 ROWS OF EACH OF THE MEMORY DIPS AT LEAST ONCE \nEVERY 4.4 \u25a0\u00bb. THE REFRESH OPERATION FOR THIS MEMORY PLANE \nCAN BE PERFORMED IN 128 CYCLES BY FORCING REFIO TO THE LOW \nSTATE AND EXECUTING A CYCLE THAT ENABLES RASIO BUT INHIBITS \nCASIO AT EACH OF THE 128 STATES OF ADDREFS INPUTS RCAOIO \nTHROUGH RCA6I0 AT LEAST ONCE EVERY 4.4 at.", "title": "JL16 Circuit Pack, 128K x 2 bit memory plane circuit", "trim_reasons": [], "year": 1978}
{"archive_ref": "WesternElectricPowerTools1915", "canonical_url": "https://archive.org/details/WesternElectricPowerTools1915", "char_count": 4970, "collection": "archive-org-bell-labs", "doc_id": 1233, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1233", "record_count": 12, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/WesternElectricPowerTools1915", "split": "test", "text": "These drills are thoroughly reliable, strongly built, light in weight and convenient in shape. They \noperate on either alternating or direct current and are reversible. Attachment is made to the ordinary \ndrop cord or incandescent lamp socket.\n\nModel Bs Has the breakover feature for changing drills or taps instantly by hand without the use of \nchuck wrench or key.\n\nModel C: Has two speeds, the gears running in grease. It has the offset spindle, allowing close quarter\n\nThis tool is not only adapted for grinding centers on lathes, but also for grinding dies, reamers and \ncutters and for universal grinding as well.\n\nTk*\u00ae t\u00b0\u00b0l very useful for buffing and polishing brass, steel and other metals hnd when equipped with \nthcsmall emery wheel is w$ll adapted for light grinding also.\n\n\u2666Delivery F. O. B. Factory, Leipsic, O. For warehouse deliveries write nearest house.\n\nModel A Drill. This tool was designed for drilling in either metal or wood. It will run on either \nalternating or direct current. The offset spindle allows drilling in close corners. The main driving \nspindle is run in a tobin bronze bushing, two inches in length. The lower handle can be readily removed \nwhen necessary.\n\nThe body is made of aluminum thoroughly strengthened and supported where the strain is most severe. \n^ gements are made for positive and effective oiling for the motor shafts, driving spindle and all parts \nare subjected to wear. The gears are accurately machined and run in grease. This drill is very light \nW C \u00b0 nVen ^ en ^ handle. It is especially useful for drilling in sheet metal of all kinds. Also for drilling \ng plate and oiler holes and for light drilling in wood.\n\n. 0 Drill. Specially designed for drilling in cast iron, brass and other soft metals. Also for light\n\ncente ^ ^ enera ^ construction is similar to Model A drill except in Model 0 drill the chuck spindle is in the \nspindle ^ e mo ^\u00b0 r body and is directly connected to the motor shaft without gear reduction. The chuck \ncurrp rUDS m bronze bushing of ample bearing surface; will operate on either alternating or direct\n\nThese hammers will do the work of pneumatic tools of like capacity at about 15% of the power cost and \nwithout the expense and inconvenience of compressor, air piping nose, etc. Over hand work, the economy \nis from 80% to 90%, and it is by no means uncommon for a tool to save its cost in a week. Every tool is \ncontrolled by a switch mounted in the handle and equipped with flexible cord and plug. They may be \nattached to any lamp socket.\n\nThese hammers are built in two types: For direct current and for universal service. The Universal \ntools operate on either direct current or alternating current of any commercial frequency. These tools are \nparticularly adapted for putting up fixtures on a concrete ceiling, dressing Frencn burr stones in a paint \nmill, cleaning scale off of condenser tube, for channeling plaster in an office building, for bush hammering \na concrete wall, for drilling solid concrete, drilling for pipe racks in a brick wall, chipping gray iron castings, \ncutting openings for I beams, etc.\n\nSupporting stands are strongly recommended for all overhead work. The use of these stands takes the \njar and strain off the operator and keeps the tool pressed tight against the work.\n\nThe proper selection of steels to be used in different kinds of material is extremely important.\n\nFor stand to be used with the above electric hammer, literature and prices fur nish ed on application.\n\nNote: The capacities given are conservative and approximate. The tools cannot be forced, the \nstrength of the blow is constant and the capacity depends upon economical drilling speeds only. There is \nno danger of burning out the machines.\n\nEach tool is complete with 10 feet of cord and plug and runs from a lighting socket. \nDelivery F. O. B., Chicago, Ill. For warehouse deliveries write nearest house.\n\nare \u201csynchronous\u2019* for alternating current motors and \u201cfull load\" for direct current\n\nrl_Jf 0t0r8 hsted above are the so-called domestic buffing and grinding outfits and consist of totally en- \nc \u00abtor mounted on a high base and furnished with separable straight shaft extension attachments \ntogether with a 3 l A inch rag buffing wheel and a 3 inch emery grinding wheel, which are \nmeiudea at prices indicated; and may be used as separate sale prices as well as allowances.\n\nice angle phase motors are provided with solid armatures, i.e., without clutches. The above motors \n^ not furnished for two or three-phase circuits.\n\np_jTjmnr j. ra ? buffing wheel, B\u20143 inch emery grinding wheel, C\u2014Straight spindle shaft attachment,\n\neluded at in ftnrr\u00abF\\ 1 ,' 8 nna inch shafts are furnished complete with 6y% men rag oumnj,\n\nNot*; tu. ttachnient may be used for light drilling purposes, drills not furnished. \nttDeliverv F n o ^ or % H inch shafts are not applicable to oil ring bearing rnoto \nj' \u2022 o, o. Factory, Lynn, Mass. For warehouse deliveries write nearest house.", "title": "Western Electric : Electric Drills : Electric Hammers : Buffing and Grinding Outfits : 1916", "trim_reasons": [], "year": 1916}
{"archive_ref": "bstj18-2-255", "canonical_url": "https://archive.org/details/bstj18-2-255", "char_count": 14500, "collection": "archive-org-bell-labs", "doc_id": 1241, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1241", "record_count": 10, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bstj18-2-255", "split": "test", "text": "ORIGINALLY, the ceramic parts used in the telephone and \nassociated equipment were not manufactured by the Western \nElectric Company because the technical requirements and volume of \nconsumption of such parts did not warrant the development or estab- \nlishment of processes or the facilities for manufacture. The later \ndevelopment of such manufacturing processes for some of the ceramic \nparts has been necessitated largely by inability to secure an adequate \nsupply of parts meeting the close limits required for satisfactory \nfunctioning of the apparatus, although there have usually been other \ninfluencing factors. Such developments have, in most instances, \nbeen advantageous from an economic standpoint. The experimental \nwork has been confined to that required fdi*' the above ends and \nonly a very limited amount of research work has been done. Some \nof the major projects for which it was necessary to develop new methods \nof processing to obtain the desired quality at a satisfactory cost are \noutlined.\n\nSwitchboard Lamp Cap Lenses \nThe first major project undertaken was the development for manu- \nfacture of switchboard lamp cap lenses of the types shown in Fig. 1. \nOne factor necessitating this undertaking was the difficulty experienced\n\ntional compositions were then made of talc and clay with and without \nsodium silicate and feldspar. It was found that most satisfactory \nresults were obtained with a ball clay, kaolin, and talc body in the \nproportions of forty per cent, ten per cent and fifty per cent, and this \nbody was therefore adopted.\n\nOriginally, attempts were made to cut the groove in the fired core \nwith a diamond tool but this resulted in excessive chipping of the\n\ngroove. A chaser with alternate teeth was then tried out, with the \nthought that the gradual cutting action would prevent chipping. \nThis also proved unsuccessful. The use of a circular saw, emery \nwheel, and a phosphor bronze disc charged with diamond dust were \nalso considered. These methods were not completely satisfactory \nalthough better results were obtained. Rolling the thread in the core \nwhile in the leather hard state after extrusion was then tried with \ngood results and a suitable machine was developed for performing \nthis operation.-\n\nWith this machine, an extruded blank of slightly oversize diameter \nwas placed on a revolving mandrel. An arm was provided to hold a \nshaving tool ahead of a disc which formed the thread. This arm was \nattached to a segment of a nut and the movement of the arm when \nthe nut segment was engaged with a thread integral with the mandrel, \nshaved the core to exact diameter and carried the disc longitudinally \nacross the core forming the spiral groove. An auxiliary arm carrying \ntwo knives was then engaged which cut the core to exact length and\n\nfactory^ results could be obtained with 10-24-mesh material and a \n12.75 per cent \u00b1 .75 per cent moisture content. This moisture content \nwas sufficient to give a compact part and body mix still could be fed \nsatisfactorily to the dies. Methods of^processing the body mix were \nthen worked out to hold it within these limits.\n\nThe use of material within these close size limitations involved \nconsiderable effort to establish economical methods of production. \nCommon practice consisted of slaking the clay in water, adding the \nother body ingredients, mixing thoroughly with water, filter pressing, \ncomplete drying, addition of water to obtain the desired moisture \ncontent, aging and screening. The effect of aging was investigated \nand found negligible with the moisture content to be used and it was \ntherefore decided to dry the material after filter pressing to the \nrequired approximate twelve per cent moisture content before the \ndisintegrating and sizing operations. Methods of handling were \nevolved to obtain a maximum percentage of material between 10 and \n24-mesh and to regranulate the fines without again mixing them with \nexcess water.\n\nVarious methods of economically removing fins after forming were \ninvestigated and initially the fired parts were tumbled with small \nporcelain balls. This method removed fins and produced smooth \nsurfaces. Another advantage of the method was the automatic \nelimination of any weak or flawed parts by breakage during the \ntumbling. Later, further developments in methods of firing described \nhereafter made it more economical to remove the fins in the raw state \nby vacuum brushing the parts in multiple after they were arranged on \ntrays at the pressing machines.\n\nInitially the parts were fired using the practice then commonly \nfollowed in the industry. With this method, the parts were placed \nin saggers and fired in an intermittent kiln. This method involved \ncostly handling, heat losses due to heating and cooling the furnace at \neach firing, and considerable expense from sagger replacements. A \nsmall continuous kiln was therefore installed in which the parts were \ncarried in layers on top of cars through successive preheating, firing, \nand cooling zones which were continuously maintained at definite \ntemperatures, the heat from the cooling fired ware being used to heat \nthe incoming ware.\n\nSummarizing, the method of manufacture finally developed for \nporcelain blocks consisted in mixing feldspar, clay and flint with water \nto get an intimate mixture, filter pressing, drying to proper moisture \ncontent, sizing, automatically molding the parts, removing fins in \nmultiple, and firing in a continuous kiln. This method resulted in a\n\npossess these properties and excellent results were obtained from their \nuse. 3 These substances were made from kelp. Their most interesting \nproperty as a suspending medium was the ability of the alginates when \nadded to water even in small percentages to make solutions of high \nviscosity. For example, water solutions of ten per cent ammonium \nalginate would stand stiff. Some of the advantages in our use of \nalginates for suspending number plate enamel were: (1) uniformity of \ncomposition resulting from the alginates being a manufactured product \nrather than a natural mineral ; (2) the fact that dried sprayed coats of \nalginate suspended enamel were less subject to damage from handling; \n(3) a low decomposition temperature which resulted in the material \nbeing driven off before fusion of enamel, thus avoiding bubbles in the \nenamel ; and (4) increased resistance of the finished enamel surfaces to \nchemical attack and their ability to withstand greater mechanical shock \nand distortion without damage, since any refractory materials present \nwhen the enamel was fired would not be completely fused or incor- \nporated into the glass, leaving points more readily attacked chemically \nas well as lines of mechanical weakness.\n\nUsing alginate suspended enamels, suitable manufacturing processes \nwere developed for the application and firing of enamel and the appli- \ncation of characters to the fired plates. A machine was devised for \nthe application of the sifted coating, and rotary continuous furnaces \nwere installed for the firing operations.\n\nOriginally the decalcomania method was used for character applica- \ntion. In this process, the enameled parts were first coated with a thin \ncoat of sizing and, after partial drying, they were placed in a locating \nfixture mounted on a small arbor press and pressure was applied to a \nproperly located transfer by means of a soft rubber pad. The paper \nbacking of the transfer was then removed by soaking in water and, to \ninsure contact, the characters were repressed with a silk covered pad. \nThe sizing was then baked off before firing to remove organic materials \nand eliminate shadows around the characters. This method was \ncostly and even well trained, careful operators did not produce satis- \nfactory plates.\n\nTo eliminate these defects, an offset printing and dusting method \nwas developed in which an electrotype printing plate was covered with \nprinter's ink and an impression was transferred to the number plate \nby means of a rubber transfer pad. Powdered vitrifiable colors were \nthen dusted over the entire surface of the plate and the unprinted \nareas of the part brushed clean with a camel's hair brush. In printing \ntwo color plates, the black letters were printed and dusted first, after\n\nceramic printing ink covered tests on various printing vehicles to deter- \nmine what vehicle or mixture of vehicles would be most suitable. \nDifficulty was encountered in incorporating a sufficient amount of \ninert, finely pulverized, intensely colored glasses into a vehicle and \nstill retaining the properties essential for offset printing by transferring \nan impression from an electrotype plate to a vitreous enamel. This \nproblem was finally solved by the use of a relatively large percentage \nof uncalcined ceramic material in combination with light and heavy\n\nink varnishes. 4 Motor driven presses using this ink were also devel- \noped to facilitate the printing operations. 5\n\nThe manufacture of these parts was undertaken primarily to \neliminate an undesirable supply situation but Western Electric manu- \nfacture resulted in improvements in quality of enameling, quality of \nprinting, and in mechanical strength. This latter characteristic was \nimportant since it reduced assembly losses from cracked plates.\n\nWhile vitreous enameled copper base number plates have been \nreplaced by other types, the developments outlined were the basis of \nsubsequent enameling developments.\n\nVitreous Enameled Iron Base Number Plates \nThe low level of illumination at some pay stations led to the design \nby the Bell Telephone Laboratories of a large iron base number plate, \nshown in Fig. 6, to be mounted flush with the finger wheel of the dial. \nSince the demand for these plates was relatively small, they were \noriginally made by the usual process followed in the industry in enamel- \ning similar articles. This process consisted of applying and firing one \nground coat for adherence and then applying and firing two sprayed \ncover coats to obtain the whiteness and opacity desired; all being \nfelspar enamels. The whiteness was not as good as that obtained on \nthe copper base plates with lead enamels and in addition considerable \ndifficulty was experienced in the field due to the fading of the characters \nas a result of chemical action on plates exposed to corrosive gases such \nas sulphurous fumes in certain locations. The process was also costly. \nSince maximum whiteness and opacity was obtainable in the lead- \narsenic type of enamels previously described when applied by dusting \non dry, it was desirable that the coating be applied in this manner. \nIn order to avoid several enamel applications and firings, it was also \ndesirable that other portions of the plate be protected by some corrosion \nresistant coating other than vitreous enamel which would necessarily \nhave to retain such corrosion resisting properties after exposure to a \ntemperature of 1500\u00b0 F. for six minutes and also be capable of being \nenameled with satisfactory results. Numerous coatings were tried and \nit was found that a Western Electric black oxide finish on iron would \nsatisfactorily meet all requirements. 6 Using this finish, it was possible \nto fuse the enamel directly on the upper surface of plates, to retain \ncorrosion resistant qualities on all other exposed surfaces, and to reduce \nthe number of process operations. A number plate of greatly improved \nappearance and durability also resulted. In addition, the curved\n\nfired parts to the desired thickness of 0.030 inch. This operation was \ncostly and losses from breakage were high. The narrow dimensional \nlimits of \u00b1.002 inch on thickness were also hard to maintain because \nof the difficulty of keeping the lapping surfaces parallel. In view of \nthis, it was decided to machine the parts from natural talc rod or lava.\n\nThe mineral talc or lava, being soft, was easy to machine and the \nfiring shrinkage was only one per cent as compared to about ten per \ncent with dry pressed porcelain. While less difficulty with warpage \nand dimensional variations was experienced, the machined surfaces, \nwhile reasonably smooth and accurate, were not equal in quality to \nsurfaces obtainable with molded parts. The chief difficulty with the \nprocess was in obtaining a satisfactory raw material free from flaws \nand fissures. The first work was done with domestic lava which was \nsomewhat granular in structure but large rejections resulted from pitted \nsurfaces and chipped edges. A survey of domestic lavas showed that \nonly a small percentage was sufficiently dense. Chinese white lava \nwas found to be homogeneous and fine grained but of uneven shrinkage. \nBest results were obtained with Italian green lava and this material \nwas used in commercial production. Due to breakage because of \nfissures, the number of good insulators per foot of rod was very low and \nthe manufacturing cost was therefore excessive.\n\nIn view of this, various domestic manufacturers of glass, porcelain, \nlava and other types of ceramic parts were canvassed but no source \nof supply that could meet the required quality limits could be located. \nIt was therefore decided to make a thorough investigation of new \nmolding compositions for the job. As a first step in this study, it \nwas necessary to do away with drying shrinkage which required \na binder which would give sufficient strength in the raw state to \nwithstand the various finning and handling operations prior to firing. \nIt was also desirable that such a binder should not affect the fired \nstructure of the parts. Various organic substances such as pitches, \nphenolic resins, asphalts, paraffins, and waxes were tried in both \nhot and cold molded bodies. It was found that a large percentage \nof these binders could be incorporated into a body without defor- \nmation during firing. 9 As a mixture of paraffin and carnauba wax \nwas found satisfactory for cold molding and in addition possessed \nsufficient hardness to furnish the necessary molded strength, this \ncombination of materials was chosen for the binder. 10\n\n9 W. J. Scott Patent 1,847,102, \"Ceramic Material,\" March 1, 1932. W. J. Scott \nPatent 1,977,698, \"Ceramic Material and Method of Making the Same,\" October 23,\n\n10 L. I. Shaw and W. J. Scott Patent 1,847,197, \"Ceramic Material and Method of \nMaking the Same,\" March 1, 1932.\n\n12 A. G. Johnson and L. I. Shaw, \"Ceramics in the Telephone,\" Industrial and \nEngineering Chemistry, Vol. 27, pp. 1326-1332, November, 1935.", "title": "BSTJ 18: 2. April 1939: Some Ceramic Manufacturing Developments of the Western Electric Company. (Johnson, A.G.; Shaw, L.I.)", "trim_reasons": [], "year": 1939}
{"archive_ref": "Western_Electric_Tube_Manual_1963", "canonical_url": "https://archive.org/details/Western_Electric_Tube_Manual_1963", "char_count": 63122, "collection": "archive-org-bell-labs", "doc_id": 1287, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1287", "record_count": 153, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/Western_Electric_Tube_Manual_1963", "split": "test", "text": "Dimensions \u2014The dimensions and outline diagrams are given in Figures 1 and 2, The overall di\u00ac \nmensions are:\n\nMounting \u2014The 301A emplo \nElectric 143B or similar socket \nthe base terminals are shown in\n\nThe tube should be mounted in a ^rtical position with the base end down. There should be \na free circulation of air^Sl^ngWhe tube. No object should touch the glass bulb.\n\niin thrust type base suitable for use in a Western \nions and the arrangement of electrode connections to\n\nThe thisppube is designed to operate on a voltage basis from an alternating-current\n\nsupply. The vd^taWshould be maintained to within 5% of its rated value (5.0 volts). Operation\n\nof the filament at a voltage above the upper limit will definitely reduce the life of the tube while a \ndecrease in voltage below the lower limit may cause immediate failure.\n\nSufficient time must always be allowed for the filament temperature to reach its normal operat\u00ac \ning value before the anode potential is applied. If filament circuits with good regulation are \nused, this time is 30 seconds. If the tube is operated at ambient temperatures below 20 \u00b0 C., a \nlonger period of time is required for the purpose of bringing the mercury vapor pressure to a satis\u00ac \nfactory operating value. The minimum filament warming time as a function of ambient tempera\u00ac \nture is shown in figure 3 .\n\nFor proper distribution of the mercury a period of 10 to 15 minutes filament warming time \nshould be allowed when the tube is used for the first time or if it has been reinserted in the apparatus \nafter having been removed.\n\nMaximum operating ambient temperature range. . . \nRecommended operating ambient temperature range\n\nThe anode-cathode potential drop is substantially independent of the plate current. The \nexact value varies from tube to tube and during the life of a given tube. Within the specified \nambient temperature range and plate current range, it may vary from 5 to 25 volts.\n\nThe anode-cathode drop as a function of temperature is shown on fig. 4 for a typical 301A tube \nafter reaching temperature equilibrium and when passing the rated plate current.\n\nThe maximum permissible peak plate current (L 0 ampere) is a limitation on the instantaneous \nvalue that the tube can carry safely in the direction in which it is designed to conduct and should \nnot be exceeded. The maximum direct load current is not fixed but will depend upon the wave \nform required by the load and filter circuit.\n\nThe maximum permissible peak potential between electrodes (1800 volts) is a limitation on \nthe instantaneous value that the tube can stand safely. If it is exceeded, an arc-back may result \nwhich will injure the tube. The maximum direct potential available is not fixed but will depend \nupon the type of circuit used.\n\n301x4 vacuum tubes may be operated in parallel if some provision is made to insure a proper \ndivision of the load current. Current dividing reactors or ballasting resistors in series with each \nanode, may be used for this purpose. The size of the reactors or resistors depends upon the circuit \ndesign.\n\nIn most cases the termination of the useful life of the 301A tube is due to the loss of filament \nactivity. This causes the tube to fail by arcing between the electrodes. Failures of this kind \nshould be safeguarded by proper fuse protection to prevent injury to other tubes in the circuit and \nto the auxiliary equipment.\n\nTypical Rectifier Circuits \u2014The 301A vacuum tube may be used in any standard high vacuum \nrectifier circuit subject to its current, voltage and temperature limitations. Typical circuits are \nshown below. The approximate direct output current and voltage for each type of rectifier circuit \nwhere tubes are operated at maximum permissible plate current and inverse voltage are given in \nTable 1. The values listed are average values of the pulsating current and voltage for an \nunfiltered circuit.\n\nCORRESPONDING AMEilENT TEMPERATURE \nAFTER REACHING TEMPERATURE EQUILIBRIUM \nIN DEGREES CENTIGRADE\n\nA. development of Bell Telephone Laboratories, Incorporated, \nthe research laboratories of the American Telephone and Tele\u00ac \ngraph Company and the Western Electric Company\n\nThis tube was designed primarily for use as an audio-frequency amplifier or modulator; or as \na radio-frequency oscillator or amplifier.\n\nDimensions \u2014Dimensions and outline diagrams are shown in Figures 1 and 2 and are:\n\nMounting\u2014 Large four-pin bayonet base for use in a W. E. 113A or similar socket, for either \nvertical or horizontal mounting. If mounted horizontally the plane of the filament, which is \nindicated in Figure 2, should be vertical.\n\nCharacteristics \u2014Performance data given below are based upon a typical set of conditions. \nVariations can be expected with different circuits and tubes.\n\nFigures 3 and 4 give the static characteristics of a typical tube plotted against grid and plate \nvoltages.\n\nAverage Characteristics at 1600 volts direct plate potential and minus 160 volts grid bias, \nI b - .167 ampere.\n\nThe above are maximum ratings which do not apply simultaneously but depend on the \ntype of service as specified below.\n\nMechanical\u2014 Figures 1 and 2 show the overall dimensions and basing arrangement for the \ntube.\n\nThe tubes should not be subjected to mechanical shock or excessive vibration. Mechanical \nvibration may cause breakage of the thoriated tungsten filaments.\n\nA free circulation of air must be provided to insure adequate cooling of the glass during operation.\n\nElectrical\u2014Overload protection should always be provided for the plate circuit. A suitable \nfuse or circuit breaker should remove the plate voltage if the plate current exceeds 425 milliamperes. \nAlthough the tube is sufficiently rugged to withstand momentary overloads, a prolonged overload \ncaused by inefficient adjustment of the circuit, may damage the tube. When adjusting a new \ncircuit, reduced plate voltage or a series resistance of 1000 to 5000 ohms in the plate circuit should \nbe used until it is operating properly.\n\nThe filament should always be operated at the rated voltage, measured at the tube terminals. \nA 5 % decrease in filament voltage reduces the thermionic emission approximately 25%. Either \ndirect or alternating current may be used for heating the filament. If direct current is used, the \nplate and grid circuit returns should be connected to the negative filament terminal. If alternating \ncurrent is used, the circuit returns should be connected to the center tap of the filament heating \ntransformer winding or to the center tap of a resistor placed between the filament terminals. \nA resistance of 30 to 40 ohms of ten watt rating is suitable.\n\nIn cases where severe and prolonged overload has temporarily impaired the electronic emission \nof the filament, the activity may be restored by operating the filament, with the plate and grid \nvoltages off, 30% above normal voltage for 10 minutes followed by a longer period at normal voltage.\n\nGrid bias may be obtained from the drop across a resistance in the plate current return \nor from a battery or rectifier supply.\n\nPlate dissipation allowable for this type of service is generally lower than is safe for other \nuses since the energy is dissipated in the plate in smaller areas due to relatively high voltage \ndrop in the tube.\n\nThe plate dissipation is equal to the plate voltage multiplied by the normal plate current. \nPerformance data are based upon the use of a resistance load. Undistorted output is calculated \non the basis of 5% second harmonic distortion.\n\nClass B\u2014Grid bias practically at cut-off and grid driving voltage higher than the bias.\n\nTwo tubes may be used in a balanced circuit. An adequate driving stage and an input \ntransformer with good regulation must be used so that the grid current drawn during positive \ngrid swings does not produce appreciable distortion. The output transformer must transform \nthe load impedance to the proper value for the tubes used. The power output obtainable will \nbe determined by the quality of the transformer used and the amount of distortion which can \nbe tolerated. The grid bias must be held constant and therefore cannot be obtained by grid \nleak or series resistor methods. A battery or other source having good regulation is necessary.\n\nThe power required of a modulator for complete modulation of a Class C amplifier is \none-half the direct power input to the plates of the Class C amplifier.\n\nThe Class B radio-frequency amplifier is used to amplify a modulated radio-frequency \ncarrier wave without appreciable distortion. It operates similarly to the Class B audio \namplifier except that a single tube may be used, the tuned output circuit serving to preserve \nthe wave shape. The push-pull circuit, however, eliminates the even order harmonics and \nthus increases the efficiency slightly.\n\nThis type of operation is suitable for telegraphy, or the production of a continuous flow \nof radio-frequency power for purposes other than communication.\n\nThis type of operation is for use when the modulating voltage is superimposed on the \nplate supply voltage and to obtain good quality the output power should vary as the square of \nthe plate voltage. For complete or 100% modulation, the plate voltage varies from zero to twice \nthe applied direct value during a cycle of the audio frequency. With no modulation applied, \nthe plate voltage is, of course, the direct value and the carrier power output is one-fourth of \nthe peak power output under 100% modulation. In this case, since the plate voltage varies \nwith modulation, the direct value must be rated lower than for other types of operation.\n\nThe frequency limits specified under maximum ratings are based on the tube being used as \nan oscillator. The tube may be used at full rating up to 1.5 megacycles. When operating at higher \nfrequencies, the dielectric losses, charging currents and lead-in heating are increased greatly. \nThe plate voltage and hence plate dissipation must be reduced to values specified for the upper \nfrequency limit and for frequencies between these two limits the plate voltage should be propor\u00ac \ntionately reduced.\n\n1-J-3G-35C the research laboratories of the American Telephone ami Tele* V. T. DATA SHEET 308B\n\nBELL SYSTEM PRACTICES \nTransmission Engineering and Data \nElectron Tube Data\n\nThe 310A is an indirectly heated cathode type pentode having a separate \nsuppressor grid connection. It is intended for use in audio, carrier and radio- \nfrequency voltage amplifiers, oscillators or modulators.\n\n^Operation with the heater at 9.0 volts is permissible \nonly when such voltage is regulated to \u00b1 1.0# or better.\n\nThe 310B Is an indirectly heated cathode type pentode haying a separate \nsuppressor grid connection. It is intended for use in audio, carrier and radio\u00ac \nfrequency voltage amplifiers, oscillators or modulators. This tube, except for \nhaving special design features to minimize microphonic noise and hum and having \nan appreciably lower maximum heater-cathode voltage rating, is identical to the \n310A.\n\nBELL SYSTEM PRACTICES \nTRANSMISSION ENGINEERING AND DATA \nELECTRON TUBE DATA\n\nThe 311A is a suppressor grid power pentode having an indirectly heated cathode. \nIt is designed for use as an audio, carrier or radio-frequency amplifier.\n\nA development of Bell Telephone Laboratories* the research laboratories of the \nAmerican Telephone and Telegraph Company and the Western Electric Company\n\nThe 3113 is a suppressor grid power pentode having an indirectly heated \ncathode. It is designed for use as an audio, carrier or radio-frequency amplifier.\n\nDimensions and pin connections shown in outline drawing on Page 4 \nMaximum Ratings, Absolute System (Note 2)\n\nScreen Grid Voltage. . \nPlate Dissipation. . . \nScreen Grid Dissipation \nCathode Current.... \nHeater-Cathode Voltage\n\nTYPICAL OPERATING CONDITIONS AND CHARACTERISTICS \nSingle Title Amplifier - Class A\n\nNote 2: In the \"Absolute System\" the maximum ratings specified are limiting \nvalues above which the serviceability of the device may be impaired \nfrom the viewpoint of life and satisfactory performance. Maximum \nratings, as such, do not constitute a set of operating conditions \nand all values may not, therefore, be attained simultaneously.\n\nBELL SYSTEM PRACTICES \nTransmission Engineering and Data \nVacuum Tube Data\n\nClassification \u2014Double gap, cold cathode, gas-filled tube for use as a relay, rectifier or voltage \nregulator in special circuits.\n\nThe elements of the 313A tube consist of two similar control electrodes and one anode. The \nconduction path between the control electrodes is known as the control gap. The conduction path \nbetween either control electrode and the anode is known as the main gap.\n\nThe glass bulb has been given an opaque coating so that the discharge is not visible while the \ntube is operating. In the photograph at the right the coating has been removed to show the tube \nelements.\n\nDimensions \u2014The dimensions and outline diagrams are given in Figs. 1 and 2. The overall \ndimensions are:\n\nMounting \u2014The 313A vacuum tube employs a standard four-pin thrust base suitable for use in \na Western Electric 143B or similar socket. The arrangement of electrode connections to the base \nterminals is shown in Fig. 2.\n\nThe \u201cmaximum peak control-electrode current\u201d is the maximum value of current which may \nbe drawn from either control electrode when it is acting as a cathode.\n\nThe \u201cmaximum average control-electrode current\u201d is the maximum value of current (averaged \nover 1 second' 1 which may be drawn from either control electrode when it is acting as a cathode.\n\nThe \u201cmaximum peak reverse current in the main gap\u201d is the maximum value of current which \nmay be drawn from the anode in the reverse direction, that is when it is acting as a cathode. The \nreverse current rating is intended for use in designing rectifier circuits and is the maximum inverse \ncurrent which it is permissible to draw- from the tube in such circuits.\n\nThe \u201ccontrol gap breakdown voltage\u201d is the potential required to initiate ionization, thereby \nstarting conduction in the control gap. Once ionization has occurred the potential across the gap \nwill be reduced to the \u201ccontrol gap sustaining voltage\u201d and will be approximately independent \nof the current. It is this property of the tube which enables it to be used as a voltage regulator.\n\nThe \u201cmain gap breakdown voltage\u201d is the potential required to start conduction in the main \ngap w'hen no ionization is occurring in the control gap. After breakdown, conduction will take \nplace at the \u201cmain gap sustaining voltage\u201d and will be practically independent of current.\n\nThe \u201cmain gap sustaining voltage\u201d is substantially independent of current when the current \npasses through the tube in the forward direction. When the current passes through the main gap \nin the reverse direction the sustaining voltage increases rapidly with increasing current. It is this \nasymmetry in the properties of the main gap of the 313A tube wdiich enable it to be used as a \nrectifier. The current voltage characteristics of the main gap of a typical 313A tube in both forw r ard \nand reverse directions as shown in Fig. 3. This curve was obtained with a cathode ray oscillograph.\n\nWhen the anode potential is maintained at a value intermediate between the \u201cmain gap break\u00ac \ndown and sustaining voltages\u201d the passage of a small amount of current in the control gap will \nproduce ionization sufficient to initiate conduction in the main gap. It is this property of the tube \nwhich enables it to be used as a relay. The amount of current m the control gap required to initiate \nconduction in the main gap is known as the transfer current. This quantity varies considerably \nfrom tube to tube and during the life of a given tube but will in general be less than o microamperes \nand usually only a few tenths of a microampere.\n\nThe deionization time is the time during which the voltage must be removed from the tube in \norder that the discharge shall not be reestablished when the voltage is restored. This time increases \nwith increasing applied voltage and wdth increasing current through the tube before the deioniza\u00ac \ntion period. This rate of increase of deionization time is such that the tube will not deionize wdth \na 60 cycle sine wave main gap voltage if the load is inductive or if the peak voltage is near the \nmain gap breakdown voltage or the current near the maximum rated value.\n\nThe \u201ctransfer time\u201d is the time during which the control gap must be energized in order that \nthe discharge may transfer to the main gap. It depends upon the amount of current flowing in \nthe control gap and on the main gap voltage. For a control gap current of 10 microamperes the \n\u201ctransfer time\u201d is approximately 200 microseconds.\n\nCircuit A shows a circuit using the control gap of the 313A as a voltage regulator.\n\nCircuit B shows a circuit using the 313A as a relay. The anode voltage should be intermediate \nbetween the main gap breakdown and sustaining voltages and the control anode may be biased \nat any desired potential less than the control gap sustaining voltage. The resistance R, in the \ncontrol anode circuit should be of the order of 100,000 ohms. This circuit possesses a \u201clock-in\u201d \nfeature, since the anode potential must be removed momentarily to restore the tube to a non\u00ac \nconducting condition. When supplied from alternating current this circuit does not possess a \n\u201clock-in M feature unless the frequency of the supply voltage is so high that the tube is not allowed \na sufficient interval to deionize.\n\nCircuit C shows a circuit using the 313A as a rectifier. The rectifying properties of the main \ngap are used but the control gap should be connected into the circuit as indicated through a high \nresistance. This will cause conduction in the forward direction to begin at a voltage much below \nthe main gap breakdown voltage. It is important to note that as a rectifier the 313A tube possesses \na unique property not common to other rectifiers in that its impedance is infinite for voltages \nbelow the breakdown voltage. In many applications that is of importance since the tube may be \nused to pass current at the higher potentials without placing a bridge across the line for signals of \nlower voltage.\n\n1-J-36-73C the research laboratories of the American Telephone and Tele- V T. DATA SHEET 313A\n\nThe 313C is a three-electrode, inert-gas-filled, cold cathode tube for use in \nrelay, voltage regulator, or rectifier circuits. This tube is especially suitable \nfor use in control circuits such as in triggering, counting, or switching apparatus.\n\nThis tube contains a small amount of krypton-85 gas which is a by-product \nradioactive material. The amount of krypton-85 is less than five microcuries, which \nis too small an amount to require any special care in use.\n\nAtomic Energy Commission regulations require that the individual tube carton \nfor tubes containing by-product radioactive material be appropriately marked. The \nmarking includes the statement that tube disposal should be in approved manner.\n\nApproved instructions for disposal of tubes containing krypton-85 are as \nfollows;\n\nTubes to be disposed of should be broken or crushed in a well \nventilated place releasing any resulting vapors to the outside \natmosphere. The residual broken or crushed tubes should be dis\u00ac \nposed of in a normal public trash disposal system. Tubes should be \ndisposed of at a rate of not more than 100 each week from any one \nlocation. Avoid breathing vapors from broken tubes.\n\nNote l: In the ''Absolute System 1 ' the maximum ratings specified are limiting values above which the \nserviceability of the device may be impaired from the viewpoint of life and satisfactory \nperformance. Maximum ratings, as such, do not constitute a set of operating conditions and \nall values may not, therefore, be attained simultaneously.\n\nNote 2: Sufficient resistance must be used in series with the tube to assure that the electrode \ncurrents do not exceed the maximum rated values.\n\nNote 3: Limits apply immediately after tube has conducted current. If tube has been idle, these \nvalues initially may be as much as 3 volts higher or lower.\n\nNote 4: with 15 volts overvoltage (15 volts above starter breakdown voltage) with tube in total \ndarkness.\n\nNote 5: Negative anode voltage applied through 3000 ohms. Starter connected to anode through \n100,000 ohms.\n\nA development of Bell Telephone Laboratories, the research laboratories of the American Telephone\n\nThe 313CA is a three-electrode, inert-gas-filled, cold cathode tube for use in \nrelay, voltage regulator, or rectifier circuits. This tube is especially suitable \nfor use in control circuits such as in triggering, counting, or switching apparatus.\n\nStarter Voltage Drop at 20 milliampere \nAnode Voltage Drop at 20 milliamperes\n\nThis tube contains a small amount of krypton-85 gas which is a by-product \nradioactive material. The amount of brypton-85 is less than five microcuries, which \nis too small an amount to require any special care in use.\n\nAtomic Energy Commission regulations require that the individual tube carton \nfor tubes containing by-product radioactive material be appropriately marked. The \nmarking includes the statement that tube disposal should be in approved manner.\n\nApproved instructions for disposal of tubes containing krypton-85 are as \nfollows;\n\nTubes to be disposed of should be broken or crushed in a well \nventilated place releasing any resulting vapors to the outside \natmosphere. The residual broken or crushed tubes should be dis\u00ac \nposed cf in a normal public trash disposal system. Tubes should be \ndisposed of at a rate of not more than 100 each week from any one \nlocation. Avoid breathing vapors from broken tubes.\n\nNote 1: In the \"Absolute System\" the maximum ratings specified are limiting values above which the \nserviceability of the device may be impaired from the viewpoint of life and satisfactory \nperformance. Maximum ratings, as such, do not constitute a set of operating conditions and \nall values may not, therefore, be attained simultaneously.\n\nNote 2: Sufficient resistance must be used in series with the tube to assure that, the electrode \ncurrents do not exceed the maximum rated values.\n\nNote 3: Limits apply immediately after tube has conducted current. If tube has been idle, these \nvalues initially may be as much as 3 volts higher or lower.\n\nNote 4: With 15 volts overvoltage (15 volts above starter breakdown voltage) with tube in total \ndarkness.\n\nNote 5: Negative anode voltage applied through 8000 ohms. Starter connected to anode through 100,000 \nohms.\n\nThe 313CB is a three-electrode, inert-gas-filled, cold cathode tube for use in \nrelay, voltage regulator, or rectifier circuits. This tube is especially suitable \nfor use in control circuits such as in triggering, counting, or switching apparatus.\n\nNote 1: In the \"Absolute System\" the maximum ratings specified are limiting values above which the \nserviceability of the device mav be impaired from the viewpoint of life ml satisfactory \nperformance. Maximum ratings, as such, do not constitute a set of operating conditions and \nall values may not, therefore, be attained simultaneously.\n\nNote 2: Sufficient resistance must be used in series with the tube to assure that the electrode \ncurrents do not exceed the maximum rated values.\n\nNote 3: Limits apply immediately after tube has conducted current. If tube has been idle, these \nvalues initially may be as much as 3 volts higher or lower.\n\nNote 4: With 15 volts overvoltage (15 volts above starter breakdown voltage) with tube in total \ndarkness.\n\nNegative anode voltage applied through 8000 ohms. Starter connected to anode through 100,000 \nohms.\n\nThe 313CC is a three-electrode, inert-gas-filled, cold cathode tube for use in \nrelay, voltage regulator, or rectifier circuits. This tube is especially suitable \nfor use in control circuits such as in triggering, counting, or switching apparatus.\n\nNote 1: In the \u2019\u2019Absolute System\u201d the maximum ratings specified are limiting values above which the \nserviceability of the device may be impaired from the viewpoint of life and satisfactory \nperformance. Maximum ratings, as such, do not constitute a set of operating conditions and \nall values may not, therefore, be attained simultaneously.\n\nNote 2: Sufficient resistance must be used in series with the tube to assure that the electrode \ncurrents do not exceed the maximum rated values.\n\nNote 3: Limits apply immediately after tube has conducted current. If tube has been idle, these \nvalues initially may be as much as 3 volts higher or lower.\n\nNote 4: With 15 volts overvoltage (15 volts above starter breakdown voltage) with tube in total \ndarkness.\n\nNote 5: Negative anode voltage applied through 8000 ohms, starter connected to anode through 100,000 \nohms.\n\nA development of Bell Telephone Laboratories, the research ianoratories of the American Telephone \nand Telegraph Company and the Western Electric Company.\n\nThe 315A is a half-wave, mercury-vapor rectifier tube for use in high-voltage rectifier \ncircuits.\n\nCondensed Mercury Temperature 20 to 55 C \nCondensed Mercury Temperature 20 to 65 C\n\nMounting . . . This tube should be mounted in a vertical position only, with the\n\nbase end down. Sufficient clearance should be maintained around \nthe tube to insure free air circulation.\n\nel ectrode^y^cury-vapor and gas-filled thyratrcn with a \neristic. This tube is designed for use in regulated\n\n2bb\u201cl\u00b00 volts;TEg=40C;grid overvoltage-5 volts \nEbb*100 volts;THg*80C;grid overvcltage=25 volts\n\nCritical Grid Current at 220 Anode Volts .... \nChange in Critical Grid Voltage at\n\nEquilibrium Condensed Mercury Temperature \nRise above Ambient, Approximate\n\n1. For starting conditions only. Equilibrium operation is limited to +20 c \nminimum condensed mercury temperature.\n\n2. Deionization time decreases with an increase in negative grid voltage or \nwith a decrease in (a) condensed mercury temperature (THg), (b) grid resis\u00ac \ntance or (c) anode current immediately preceding the end oi conduction.\n\n3. Ionization time decreases with an increase in (a) anode voltage, (b) con\u00ac \ndensed mercury temperature (TEg) or (c) grid overvoltage. Grid overvoltage \nis defined as the magnitude by which the applied voltage exceeds, in a \npositive direction, the critical grid voltage value. Critical grid voltage \nis the instantaneous value of grid voltage at the time when anode current \nstarts to flow.\n\nTYPICAL CONTROL CHARCTER1STICS. \nSHADED AREA SHOWS RANGE OF CHAR\u00ac \nACTERISTICS, CONDENSED MERCURY \nTEMPERATURE -55\u00b0C. TO +80\u00b0C.\n\nA development of Bell Telephone Laboratories, the research laboratories of the \nAmerican Telephone and Telegraph Company and the Western Electric Company.\n\nBELL SYSTEM PRACTICES \nTransmission Engineering and Data \nVacuum Tube Data\n\nThe 327A vacuum tube is designed to supply direct current from an alternating-current supply \nin power systems for battery charging and for other purposes.\n\nDimensions \u2014The dimensions and outline diagram are given in Figure 1. The overall dimen\u00ac\n\nMounting-\u2014 This tube employs a 3 connection, skirted medium screw base. Overall dimensions \nare shown in Figure 1.\n\nThe tube may be mounted either in a vertical or a horizontal position. There should be a free \ncirculation of air around the tube. No object should touch the glass bulb.\n\nThe filament of this tube is designed to operate on a voltage basis from an alternating-current \nsupply. The voltage should be maintained to within 10% of its rated value (2.0 volts). Operation \nof the filament outside of these limits may cause the tube to become inoperative. Filament and \nplate voltage may be applied to the tube simultaneously.\n\nThe anode-cathode potential drop is substantially independent of the plate current. The \nexact value may vary from tube to tube and during the life of a given tube over the range from 5 to \n10 volts.\n\nThe ignition voltage is the voltage required to start conduction within the tube. In a-c. circuits \nthis is a function of frequency, increasing as the frequency increases. The values are given for \n60 cycles since this frequency will be most generally encountered in circuit design.\n\nThe maximum permissible peak plate current (6 amperes) is a limitation on the instantaneous \nvalue that the tube can carry safely in the direction in which it is designed to conduct and should \nnot be exceeded. The maximum average load current (2 amperes) is the maximum direct output \ncurrent which may be obtained from a half-wave circuit using one tube. A full-wave circuit using \ntwo tubes will supply a maximum of 4 amperes.\n\nThe maximum permissible peak inverse potential (275 volts) is a limitation on the instan\u00ac \ntaneous value that the tube can stand safely. If it is exceeded, an arc-back may result which may \ninjure the tube. All circuits should be adequately fused to prevent injury to the equipment in \nevent of arc-back due to line surges. The maximum output voltage obtainable in either the half\u00ac \nwave or full-wave circuit is approximately 75 volts.\n\n327A vacuum tubes may be operated in parallel if some provision is made to insure a proper \ndivision of the load current. Current dividing reactors or ballasting resistors in series with each \nanode, may be used for this purpose. The size of the reactors or resistors depends upon the circuit \ndesign.\n\nthe research laboratories of the American Telephone and Tele- V. T. DATA SHEET 327A\n\nClassification\u2014Voltage-amplifier, suppressor-grid pentode with indirectly heated \ncathode\n\nThe electrical characteristics of the 328A tube are identical with those of the 310A tube except \nfor the heater voltage and current.\n\nThis tube is intended primarily for use in audio, carrier and radio-frequency voltage amplifiers, \noscillators or modulators. The connection for the suppressor grid has been brought out to an external \nterminal, thus making the tube more flexible in its applications.\n\nDimensions\u2014 Dimensions, outline diagrams of the tube and base, and the arrangement of the \nelectrode connections to the base terminals are shown in Figures 1 and 2.\n\nBase\u2014 Small, six-pin thrust type with pins silver-plated. A small, metal cap control-grid terminal \nis located at the top of the bulb.\n\nSocket \u2014Standard, six-contact type, preferably provided with silver-plated contacts such as the \nWestern Electric 144B socket.\n\nThe heater element of this tube is designed to operate on a voltage basis and should be operated at \nas near the rated voltage as is practicable.\n\nCathode Connection \u2014Preferably direct to the heater. If voltage must be applied between \nthe cathode and heater, it should not exceed 150 volts.\n\nCharacteristics\u2014 Plate current and screen-grid current characteristics of a typical 328A tube are \nshown in Figures 3 and 4, respectively, as functions of control-grid voltage for several values of \nscreen-grid and plate voltage and zero suppressor-grid voltage. The screen-grid voltage for these \ncharacteristics is equal to the plate voltage. Corresponding amplification factor, plate resistance, \nand transconductance characteristics are given in Figures 5, 6 and 7. Plate current and screen- \ngrid current characteristics as functions of plate voltage are given in Figures 8 and 9, respectively, \nfor several values of control-grid voltage, a screen-grid voltage of 135 volts, and zero suppressor- \ngrid voltage. Corresponding amplification factor, plate resistance, and transconductance charac\u00ac \nteristics are shown in Figures 10, 11 and 12. Plate current, screen-grid current, plate resistance, \nand transconductance characteristics are shown in Figures 13, 14, 15 and 16 as functions of plate \nvoltage for several values of suppressor-grid voltage, a screen-grid voltage of 135 volts, and a control- \ngrid voltage of \u20143 volts. These last characteristics are of particular interest in modulator applica\u00ac \ntions where separate inputs are applied to the control and suppressor grids.\n\nMaximum cathode current (screen-grid current plus plate current) 10 milliamperes \nMaximum screen-gnd current. 2,5 milliamperes\n\nOperating Conditions and Output \u2014Nominal performance data are given in the table below \nfor a number of typical operating conditions. Less severe operating conditions should be selected \nin preference to maximum operating conditions wherever possible. The life of the tube at maximum \nconditions may be shorter than at less severe conditions.\n\nThe performance data include the fundamental voltage or power output for the indicated \nvalues of load resistance and input voltage, and the maximum second and third harmonic levels \nfor input voltages no greater than the indicated values. The voltage output is given in peak volts, \nthe power output in milliwatts, and the harmonic levels in decibels below the fundamental.\n\nCurves showing the fundamental power and voltage output and the second and third harmonic \nlevels as functions of input voltage for a number of values of load resistance and a typical operating \ncondition are given in Figures 17, 18, 19 and 20.\n\nTHIRD HARMONIC IN DB SECOND HARMONIC IN DB FUNDAMENTAL OUTPUT FUNDAMENTAL POWER OUTPUT\n\nThe 328A is an indirectly heated cathode type pentode having a separate \nsuppressor grid connection. It is intended for in use in audio, carrier and radio\u00ac \nfrequency voltage amplifiers, oscillators or modulators. This tube, except for \nhaving a different heater voltage and current rating, is identical to the 310A \ntube.\n\nScreen Grid Voltage . . \nPlate Dissipation . . . \nScreen Grid Dissipation \nCathode Current . . . \nHeater-Cathode Voltage\n\nClassification\u2014Low-power, suppressor-grid pentodes with indirectly heated cathodes\n\nThese tubes are intended primarily for use as audio, carrier or radio-frequency power amplifiers \nwhere power outputs of approximately two watts are required and where the plate voltage is not in \nexcess of 180 volts. The suppressor grid is permanently connected to the cathode within the tube.\n\nDimensions and Connections\u2014 Dimensions, outline diagrams of the tubes and bases, and the \narrangement of the electrode connections to the base terminals are shown in Figures 1 and 2.\n\nBase and Mounting \u2014These vacuum tubes employ small five-pin thrust type bases with silver \nplated pins. They are adapted for use in standard five-contact type sockets, preferably those pro\u00ac \nvided with silver-plated contacts such as the Western Electric 141A socket. A small metal cap \ncontrol-grid terminal is located at the top of the bulb.\n\nThe heaters should be operated on a voltage basis and at as near the rated voltage as \npracticable.\n\nCharacteristics\u2014Figures 8 and 4 respectively, show plate current and screen-grid current as \nfunctions of control-grid voltage for several values of screen and plate voltage. In all curves the \nplate voltage is equal to the screen voltage. Amplification factor, plate resistance and transcon- \nductance curves for the conditions corresponding to those of Figures 3 and 4 are given respectively \nin Figures 5, 6 and 7.\n\nPlate current and screen-grid current are shown as functions of plate voltage in Figures 8 and \n9 respectively, for a screen-grid voltage of 135 volts and for several values of control-grid voltage. \nCorresponding curves for amplification factor, plate resistance and transconductance are given in \nFigures 10, 11 and 12 respectively.\n\nCurves showing the fundamental power output and the second and third harmonic levels as \nfunctions of input voltage for a number of values of load resistance for typical operating conditions \nare given in Figures 13, 14 and 15 respectively.\n\nMaximum cathode current (plate current plus screen-grid current) 60 milliamperes \nMaximum direct screen-grid current . 12 milliamperes\n\nOperating Conditions and Output \u2014 Nominal performance data are given in the table \non page 3 for a number of typical operating conditions. Less severe operating conditions should be \nselected in preference to the maximum conditions wherever possible. The life of the tube at maxi\u00ac \nmum conditions will be shorter than at the less severe conditions.\n\nThe performance data include the fundamental power output for the indicated values of load \nresistance and input voltage, and the maximum second and third harmonic levels for input voltages \nnot exceeding the indicated values. Under certain conditions the maximum second harmonic level \noccurs at a lower input voltage than that given in the table. The power output is given in watts, \nand the harmonic levels in decibels below the fundamental.\n\nThe 333A is a three-electrode, inert-gas-filled, cold cathode tube for use in relay, \nvoltage regulator, or rectifier circuits. This tube is especially suitable for use in control \ncircuits such as in triggering, counting, or switching apparatus.\n\nPeak Anode Voltage \nAverage Cathode Current \nAverage Life, approximate \nTransfer Current\n\nRequired Transfer Current at 130 Anode Volts (D.C.) \nDeionization Time, approximate\n\n\u00a5 Limits apply immediately after tube has conducted current. If tube has been idle, \nthese values initially may be as much as 3 volts higher or lower.\n\nNegative anode voltage applied through S,000 ohms. Starter connected to anode \nthrough 100,000 ohms.\n\nThe 336A is a suppressor grid, power penthode with an indirectly heated cathode. \nIt is designed for use as an audio-frequency power amplifier in Class A x and ABi\n\nBELL SYSTEM PRACTICES \nTransmission Engineering and Dat \nElectron Tube Data\n\nThe 337A is a variable-mu pentode of the unipotential cathode type* The sup\u00ac \npressor grid is connected to a separate base pin to provide flexibility in usage. \nThis tube is designed for use as an audio, carrier or radio-frequency amplifier, \noscillator or modulator.\n\nIt is primarily a rectifier of J^w internal impedance whose conduction cycle is determined by \nthe relative instantaneous grid aitd abode potentials. It is intended for use in special circuits as a \nrelay or trigger-action device. A f&w^pf its other possible uses are: as a controlled-frequency \noscillator giving a squar^wa^|ornlSA a voltmeter or volume level-indicator, as a source of \nsweep-voltage for a linear tm^e^^ oj^as a variable-voltage rectifier.\n\nMounting\u2014This vacuum tube employs a standard five-pin thrust type base suitable for use in \na Western Electric 141A or similar socket. The arrangement of electrode connections to the base \nterminals is shown in Figure 2.\n\nIt may be mounted in either a vertical or horizontal position, although the vertical position \nis preferable.\n\nThe heater element of this tube is designed to operate on a voltage basis from a direct or \nalternating current supply. The voltage should be maintained to within 5% of its rated value \n(10 volts). Operation of the heater element above the upper limit will definitely reduce the life \nof the tube, while a decrease below the lower limit may cause immediate failure.\n\nSufficient time should always be allowed for the cathode temperature to reach its normal \noperating value before anode current is drawn. Failure to allow sufficient time may result in \nimmediate failure.\n\nThe characteristics of the 338A tube are such that, for any given anode potential, there is a \ncritical grid potential. If the grid is held more negative than this value and the tube is non-con\u00ac \nducting, the anode current will remain zero. If it is made less negative, the current will assume a \nvalue determined by the applied potential and the resistance in the anode circuit. To extinguish \nthe discharge and return the current to zero, the positive anode potential must be removed. \nWhen current is flowing a visible discharge occurs in the tube. Under this condition, the tube \nvoltage drop is practically independent of the value of both the anode current and the grid \npotential. A protective resistance should always be included in the circuit to limit the anode \ncurrent to the rated values. A typical curve relating the critical grid potential to the anode \npotential is shown in Figure 3. This charactersitic may vary from tube to tube and during the \nlife of a given tube.\n\nSufficient resistance must always be included in the grid circuit to limit the negative grid po\u00ac \ntential to 10 volts when anode current is flowing. Failure to observe this precaution will result in \nshort tube life.\n\nThe tube may be used in a variety of circuits adapted to the application of thyratrons. Two \ngeneral types are common. One use of the tube is to produce a saw-toothed, current wave. The \ncircuit for this application is shown in Figure 4. The resistance R should, ordinarily, be at least \n100,000 ohms, and the product RC (C in farads) approximately equal to the desired fundamental \nperiod.\n\nThe second general use for the tube is as a relay device. In this application the anode may be \nsupplied from either alternating or direct current. When supplied from direct current, the circuit, \nFigure 5, possesses a \u201clock-in\u201d feature, since the anode potential must be removed momentarily \nin order to restore the tube to the non-conducting condition. When supplied from alternating \ncurrent, the circuit possesses no \u201clock-in\u201d feature, but the average anode current may be controlled \nby the relative phase of grid and anode potentials. The schematic circuit for this application \nis shown in Figure 6. Figure? is a simplified circuit employing a photoelectric cell in place of the \nresistance, R, used in the phase shifting device in Figure 6. The photoelectric cell, however, is \nequivalent to a variable resistance in the sense that the current passed will depend upon the \namount of light falling upon it.\n\nA development of Bell Telephone Laboratories, Incorporated, \nthe research laboratories of the American Telephone and Tele\u00ac \ngraph Company and the Western Electric Company\n\nThe 345A is a full-wave rectifier with indirectly heated cathodes. It is designed to \nsupply direct current from an alternating current source or to rectify radio-frequency \ncurrents for feedback purposes in broadcast transmitters.\n\nA development of Bell Telephone Laboratories, the research laboratories of the \nAmerican Telephone and Telegraph Company and the Western Electric Comjwun\n\nThe 346 b is a three-electrode, inert-gas-filled, cold cathode tube for use \nin relay, voltage regulator, or rectifier circuits. This tube is especially \nsuitable for use in control circuits such as in triggering, counting, or \nswitching apparatus.\n\nDimensions and pin connections shown in outline drawing on page 4 \nNote 1: Sufficient resistance must be used in series with the tube to \nassure that the electrode currents do not exceed their maximum \nrated values.\n\nNote 2: Limits apply immediately after tube has conducted current. If the \ntube has been idle, these values initially nay be as much as 3 \nvolts higher or lower.\n\nNote 3: With 15 volts starter overvoltage (15 volts above Starter Breakdown \nVoltage) with tube in total darkness.\n\nNote 4: Negative anode voltage applied through 8COO ohms. Starter connected \nto anode through 100000 o hm s.\n\nWestern Electric cold cathode tubes contain a minute amount of radium bromide \nwhich is a radioactive material. The amount in most types is too small to require \nany special care in use, handling or disposal.\n\nA few types contain a larger quantity of radium bromide in which the radium \napproximates that found on a luminous watch dial. These types bear a red three- \nbladed propeller-shaped symbol on the tube envelope. Instructions for handling \nsuch tubes are given below and also in Bell System Practices for Central Office \nmaintenance.\n\nInstallations ordinarily require no precautions against radiation. However, \nquantities of the tubes should not he so installed, or so stored outside the \nshipping carton ; that they will he within a few inches of personnel or in proxi\u00ac \nmity uo photographic film for extended periods of time. For example, however, a \n4-0-hour week exposure at about one (l) foot from a bank of 5CO tubes (covering \nan area of 20 inches x 45 inches) is well within the accepted tolerance limits \nfor personnel. Reasonable care should be exercised in handling and disposal of \nbroken tubes. In general, attention should be given to the following:\n\n(c) Use wet rag to pick up broken parts. Wrap broken parts in rag and tie \nsecurely sc as to form a package. Thoroughly wash hands after disposal.\n\nOne or two tubes at a time may be disposed of with normal waste material. \nAccumulation of tubes in one concentrated area of the place of final \ndisposition should be avoided.\n\nA development of Bell Telephone laboratories, the research laboratories of the \nAmerican Telephone and Telegraph Company and the Western Electric Company.\n\nThe 346C is a three-electrode, inert-gas-filled, cold cathode tube for use in \nrelay, voltage regulator, or rectifier circuits. This tube is especially suitable \nfor use in control circuits such as in triggering, counting, or switching apparatus.\n\nThis tube contains a small amount of krypton-85 gas which is a by-product \nradioactive material. The amount of krypton-85 is less than five roicrocuries, which \nis too small an amount to require any special care in use.\n\nAtomic Energy Commission regulations require that the individual tube carton \nfor tubes containing by-product radioactive material be appropriately marked. The \nmarking includes the statement that tube disposal should be in approved manner.\n\nApproved instructions for disposal of tubes containing krypton-85 are as \nfollows;\n\nTubes to be disposed of should be broken or crushed in a well \nventilated place releasing any resulting vapors to the outside \natmosphere. The residual broken or Crushed tubes should be dis\u00ac \nposed of in a normal public trash disposal system. Tubes should be \ndisposed of at a rate of not more than 100 each week from any one \nlocation. Avoid breathing vapors from broken tubes.\n\nNote 1: In the \"Absolute System\" the maximum ratings specified are limiting values above which the \nserviceability of the device may be impaired from the viewpoint of life and satisfactory \nperformance. Maximum ratings, as such, do not constitute a set of operating conditions and \nall values may not, therefore, be attained simultaneously.\n\nNote 2: Sufficient resistance must be used in series with the tube to assure that the electrode \ncurrents do not exceed their maximum rated values.\n\nNote 3: Limits apply immediately after tube has conducted current. If the tube has been idle, these \nvalues initially may be as much as 3 volts higher or lower.\n\nNote 4; With 15 volts starter overvoltage (15 .volts above Starter Breakdown Voltage) with tube in \ntotal darkness.\n\nNote 5: Negative anode voltage applied through ROOO ohms. Starter connected to anode through 100,000 \nohms.\n\nBELL SYSTEM PRACTICES \nTRANSMISSION ENGINEERING AND DATA \nELECTRON TUBE DATA\n\nThe 347A is a triode designed for use as an audio-frequency amplifier where exception\u00ac \nally low tube noise is required. Special design features minimize both the microphonic \nnoise and the hum produced by a.c. operation of the heater.\n\nPlate Voltage \nPlate Dissipation \nPlate Current \nHeater-Cathode Voltage\n\nTYPICAL OPERATING CONDITIONS AND CHARACTERISTICS-CLASS Ai AMPLIFIER\n\nUnder typical operating conditions, and \nwith the cathode of the tube connected to \nthe mid-point of the heater circuit, the \nequivalent hum voltage in the grid circuit \nwill be less than 12 microvolts at the supply \nfrequenc\\ r and less than 6.0 microvolts at\n\nIf the insulation leakage and capacitance be\u00ac \ntween the external grid and heater connec\u00ac \ntions are kept reasonably low, a resistance of \n2 megohms may be used in the grid circuit \nwithout materially affecting the hum level.\n\nTRANSCONDUCTANCE IN MICROMHOS PLATE RESISTANCE IN OHMS AMPLIFICATION FACTOR\n\nThe 348 a is an indirectly heated cathode type pentode having a separate \nsuppressor grid connection. It is intended for use in audio, carrier and radio- \nfrequency voltage amplifiers, oscillators or modulators. It has special design \nfeatures to minimize microphonic noise and hum. This tube, except for having a \ndifferent base, top cap, heater voltage and current rating, is identical to the \n310B,\n\nScreen Grid Voltage . . \nPlate Dissipation . \u00ab \nScreen Grid Dissipation \nCathode Current .... \nHeater-Cathode Voltage\n\nThe 349A is a suppressor grid, power pentode with an indirectly heated cathode. \nIt is designed for use as an audio-frequency power amplifier in Class A t and AB t \nservice.\n\nPlate Voltage \nScreen Grid Voltage \nPlate Dissipation \nScreen Grid Dissipation \nCathode Current \nHeater-Cathode Voltage\n\nThe 350B is a beam power tetrode of the neater-cathode type. It is designed for use as an \naudio-frequency amplifier or as a radio-frequency oscillator.\n\nThe 351A is an octal based full-wave rectifier with indirectly heated cathodes. It is \ndesigned to supply direct current from an alternating current source or to rectify radio\u00ac \nfrequency currents for feedback purposes in broadcast transmitters.\n\nD-C Output Volts, Approximate, at Input to Filter \nTotal Effective Plate-Supply Impedance per Plate \nFilter Input Condenser .\n\nA development of Bell Telephone Laboratories, the research laboratories of the \nAmerican Telephone and Telegraph Company and the Western Electric Company\n\nThe 352A tube comprises three distinct vacuum tube units which are independent of each \nother except that sections of a single cathode structure supply electron emission for all three. Two \nof these units are diodes. The other is a triode.\n\nApplications\u2014Diode detector, diode rectifier for automatic volume control voltage, and triode \naudio-frequency amplifier. If desired the two diodes may be used for full-wave rectification or they \nmay be connected in parallel to provide a lower impedance half-wave rectifier. The former con\u00ac \nnection requires about twice as high an input voltage as the latter to give equal detector output.\n\nDimensions and Connections\u2014Outline diagrams of the tube and base giving the dimensions \nand the arrangement of the electrode connections to the base terminals are shown in Figures 1 and 2.\n\nBase and Mounting\u2014This tube employs a small six-pin thrust type base suitable for use in a \nWestern Electric 144B or similar socket. The base pins are silver plated. The triode grid terminal \nis a small metal cap located at the top of the bulb.\n\nThe heater of this tube is designed to operate on a voltage basis and should be operated at \nas near the rated voltage as practicable.\n\nCathode Connection \u2014Where alternating heater voltage is used the cathode should pref\u00ac \nerably be connected directly to the mid-point of the heater transformer winding or to the mid\u00ac \npoint of a low resistance connected across the heater terminals. For direct current operation the \ncathode may be connected to either end of the heater. If voltage is applied between the heater \nand cathode, it should be kept low and must not exceed 50 volts.\n\nTriode Characteristics\u2014 Typical curves showing triode plate current as a function of grid \nvoltage for several values of plate voltage are shown in Figure 3. Corresponding amplification \nfactor, plate resistance and transconductance characteristics are given in Figures 4, 5 and 6 re- \nspectivelv. Figure 7 show's plate current as a function of plate voltage for several values of grid \nvoltage.\n\nPermissible operating plate and grid voltages are included within the area, ABCD, in Figure 3. \n.Amplification factor, plate resistance, transconductance and performance data for a number of \ntvpical operating conditions are given in the table. Recommended conditions or others of no \ngreater severity should be selected in preference to maximum conditions wherever possible. The \nFife of the tube at maximum operating conditions will be shorter than at less severe conditions.\n\nIn the last four columns of the table are given the fundamental power output, P m , in milliwatts, \nthe fundamental voltage output, E pm , in peak volts and the second and third harmonic levels, F 2m \nand F im , in db below the fundamental, for the indicated values of load resistance. The peak \nvalue of the sinusoidal input voltage, E gm , is numerically equal to the grid bias in each case. \nWhere the level of the third harmonic is lower than 45 db below the fundamental, its value may \nbe widely different from tube to tube. The values given represent a typical tube.\n\nFor a smaller input voltage, E g , the fundamental power and voltage outputs and the harmonic \nlevels are given approximately by the following relations:\n\nTriode Microphonic Noise\u2014 With a plate voltage of 135 volts, a grid bias of -6 volts, and a \nload resistance of 100,000 ohms, the mean microphonic noise output level of the triode section of \nthe tube, measured in a laboratory reference test set, is 45 db below 1 volt. The range of levels of \nindividual tubes extends from 20 to 60 db below 1 volt. Since microphonic noise depends on the \ntype and intensity of the mechanical disturbance which produces it, the values given here are useful \nchiefly for comparison with the levels of other tubes which have been tested in the same way.\n\nDiode Characteristics\u2014 The current-voltage characteristic of a single diode is shown in Figure \n8. Rectification characteristics for a single diode are shown in Figure 9 for a number of values of \nimpressed alternating input voltage. Each of these characteristic curves gives the relation between \nthe direct voltage impressed on the diode plate and the average diode current as indicated by a \ndirect-current microammeter for a constant impressed alternating input voltage of the value speci\u00ac \nfied. Where the diode is used as a detector with the usual condenser-resistance circuit, the direct \ncomponent of the voltage developed across the resistance by any alternating-voltage input is given \nby the intercept of the load line with the rectification characteristic corresponding to the input \nvoltage. Load lines for zero fixed bias are shown in Figure 9 for load resistance values of 0,25, \n0,5 and 1.0 megohm.\n\nThe potential of each diode plate with respect to the cathode on the positive swing of the input \nvoltage should be limited to a maximum value of +10 volts.\n\nA development of Bell Telephone Laboratories, Incorporated \n1-J-39-61C the research laboratories of the American Telephone and Tele-\n\nBELL SYSTEM PRACTICES \nTransmission Engineering and Dat^ \nElectron Tube Data\n\nThe 353A is a three-electrode, inert-gas-filled, cold cathode tube for use in \nrelay, voltage regulator, or rectifier circuits. This tube is especially suitable \nfor use in control circuits such as in triggering, counting, or switching apparatus.\n\nNote 3: Limits apply immediately after the tube has conducted current. If tube has been idle, these- \nvalues initially may be as much as 3 volts higher or lower.\n\nNote 4: With 15 volts starter overvoltage (15 volts above starter breakdown voltage) with tube in \ntotal darkness.\n\nNote 5: Negative anode voltage applied through 8000 ohms. Starter connected to anode through 100,000 \nohms.\n\nA development of Bell Telephone Laboratories, the research laboratories of the American Telephone \nand Telegraph Company and the Western Electric Company.\n\n1. Deionization time decreases with an increase in negative grid voltage \nor with a decrease in (a) condensed mercury temperature (THg), (b) grid \nresistance or (c) anode current immediately preceding the end of conduction.\n\n2. Ionization time decreases with an increase in (a) anode voltage, (b) condensed \nmercury temperature (THg) or (c) grid overvoltage. Grid overvoltage is defined \nas the magnitude by which the applied voltage exceeds, in a positive direction, \nthe critical grid voltage value. Critical grid voltage is the instantaneous \nvalue of grid voltage at the time when anode current starts to flow.\n\nSHADED AREA SHOWS RANGE OF CHARACTERISTICS\u00bb \nCONDENSED-MERCURY TEMPERATURE +30 # T0+70 # C\n\nA development of Bell Telephone Laboratories, the research laboratories of the \nAmerican Telephone and Telegraph Company and the Western Electric Company*\n\nCritical Grid Current at 220 Anode Volts . \nChange in Critical Grid Voltage at\n\n1. For starting conditions only. Equilibrium operation is limited to +2GC \nminimum condensed mercury temperature.\n\n2. Deionization time decreases with an increase in negative grid voltage \nor with a decrease in (a) condensed mercury temperature (THg), (b) grid \nresistance or (c) anode current immediately preceding the end of conduction.\n\n3. Ionization time decreases with an increase ih (a) anode voltage, (b) condensed ( \nmercury temperature (THg) or (c) grid overvoltage. Grid overvoltage is \ndefined as the magnitude by which the applied voltage exceeds, in a positive \ndirection, the critical grid voltage value. Critical grid voltage is the \ninstantaneous value of grid voltage at the time when anode current starts to \nflow.\n\nTYPICAL CONTROL CHARACTERISTICS \nSHADED AREA SHOWS RANGE OF CHARACTERISTICS \nCONDENSED MERCURY TEMPERATURE - 55* TO + 80*C\n\nThe 357B is a three-electrode tube designed \nfor use as a radio-frequency amplifier or \noscillator, audio-frequency amplifier or \nmodulator. The anode is capable of dissi\u00ac \npating 400 watts. The tube is cooled by radi\u00ac \nation at frequencies below 40 megacycles.\n\nForced-air cooling of the envelope is neces\u00ac \nsary at higher frequencies. The tube is ca\u00ac \npable of operating up to 100 megacycles at \nmaximum ratings and up to 150 megacycles \nat reduced ratings. The cathode is a thori- \nated tungsten filament.\n\nRequired Air flow on Envelope \nWhen Operated Above 40 Megacycles . \nMaximum Incoming Air Temperature\n\n1. Represents maximum usable cathode current for tube \nas plate current plus grid current for any condition of \noperation.\n\n2. Radiation cooling is adequate \u2019when the tube is operated \nbelow 40 megacycles and with a free circulation of air \naround the tube. If operated in a confined space or at a \nfrequency above 40 megacycles, forced-air cooling is \nnecessary. Satisfactory air cooling will be obtained from \na blower delivering approximately 40 cubic feet of air \nper minute from a 2-inch diameter nozzle. The nozzle \noutlet should be placed approximately 3 inches from the\n\ntube and directed toward the central point of the enve\u00ac \nlope, midway between the plate and grid terminals.\n\nThe plate terminal connector shall be of a design that \nwill readily conduct heat from the plate terminal.\n\n3. This test is equivalent to a JAN-1A Pendulum Bump \nTester 15' test. The data given represent the maximum \ncapabilities of the tube without electrical potentials ap\u00ac \nplied and should not be construed to mean that the tube \nis capable of withstanding an infinite number of shocks \nof this magnitude.\n\nMAXIMUM RATINGS AND TYPICAL OPERATING CONDITIONS \nAUDIO-FREQUENCY POWER AMPLIFIER AND MODULATOR-CLASS B \nMAXIMUM RATINGS, ABSOLUTE VALUES\n\nD-C Grid Current, approximate \nDriving Power, approximate 6 . \nPower Output, approximate .\n\n5. As high level modulator for 1000 watt transmitter. Total harmonics approxi\u00ac \nmately 1 . 5 % at full output.\n\nPLATE MODULATED RADIO-FREQUENCY POWER AMPLIFIER\u2014CLASS C TELEPHONY \nCarrier conditions per tube for use with maximum modulation factor of 1.0\n\nD-C Grid Current, approximate \nDriving Power, approximate . \nPower Output, approximate\n\nKey-down conditions per tube without amplitude modulation 8 \nMAXIMUM RATINGS, ABSOLUTE VALUES\n\nD-C Grid Current, approximate \nDriving Power, approximate . \nPower Output, approximate .\n\nMaximum ratings apply up to 100 megacy cles. \nThe tube may be operated at higher frequen\u00ac \ncies provided the maximum values of plate \nvoltage and plate input are reduced according\n\nto the tabulation below. Other maximum rat\u00ac \nings are not affected. Forced-air cooling of the \nenvelope with an air flow of approximately \n40 cfm is required at these frequencies.\n\n8. Modulation essentially negative may be used if the positive peak of the \nenvelope does not exceed 115 per cent of its unmodulated value.\n\nBase pin positions shatl be \nheld to tolerances such that \npins will fit a flat - plate \ngauge having a thickness \nof .250\" with 2 holes of \n.391\" \u00b1 .0005\u201c dia. and \n1 hole of .469'i 0005\" dia \nAll holes shall be located \non a 1,938\" \u00b1 .0005\" dia \ncircle at specified centers.\n\nThe 358A is a two-electrode, inert-gas-filled, cold cathode tube designed to \nprovide a visual signal for telephone work. When the tube is conducting, a glow will \nappear near the surface of the negative electrode. If a d.c. voltage supply is used \nthe negative polarity should be applied to the upper electrode. The upper electrode \nterminal may be identified by the adjacent circular dot of contrasting color on the \nbase. When the tube is operating with an alternating voltage a glow will appear on \nboth electrodes.\n\nThis tube possesses a unique property not common to filamentary type lamps in \nthat its impedance is essentially infinite for voltages below breakdown. In some \napplications this is an advantageous feature since the tube may be used to pass \ncurrent at*the higher potentials without placing a conducting path across the line \nfor signals of lower voltage.\n\nUnlike filamentary type lamps the light output of this tube is proportional to \nthe current through the tube instead of varying as a power of this current. This \ntube is well adapted to furnishing a visual signal from a varying voltage source.\n\nNote 1: In the \"Absolute System\" the maximum ratings specified are limiting values above which the \nserviceability of the device may be impaired from the viewpoint of life and satisfactory \nperformance. Maximum ratings, as such, do not constitute a set of operating conditions and \nall values may not, therefore, be attained simultaneously.\n\nNote 2: When the tube is operating from a direct current supply, the upper electrode shall be used \nas the cathode.\n\nThe 359A is a three-electrode, inert-gas-filled, cold cathode tube for use \nprimarily as a relay in communication circuits. It is also suitable for use in \ncontrol circuits such as in triggering, counting or switching apparatus and as a \nvisual indicator. This tube, by reason of small size and provision for wiring \ndirectly into the circuits, may be used to advantage in equipment having limited \nspace for components.\n\nPeak Anode Voltage, Maximum \nAverage Cathode Current \u2022 \u2022 \nAverage Life, Approximate *\n\nThe 372A is a three-electrode, inert-gas-filled, cold cathode tube for use in relay, \nvoltage regulator, or rectifier circuits. This tube is especially suitable for use in control \ncircuits such as in triggering, counting, or switching apparatus.\n\n* Limits apply immediately after tube has conducted current. If tube has been idle, \nthese values initially may be as much as 3 volts higher or lower.\n\nNegative anode voltage applied through 8,000 ohms. Starter connected to anode \nthrough 100,000 ohms.\n\nThe 373A is a filamentary type suppressor grid pentode. It is designed for use as an audio, \ncarrier or radio-frequency voltage amplifier, oscillator or modulator.\n\n1. Pin #6 is connected internally to pin #5 and is approximately 3/32 inch shorter \nthan the other pins to minimize noise when changing tubes while in service.", "title": " Western Electric Tube Manual 1963", "trim_reasons": [], "year": 1960}
{"archive_ref": "bellsystem_BSRS_353.317", "canonical_url": "https://archive.org/details/bellsystem_BSRS_353.317", "char_count": 5040, "collection": "archive-org-bell-labs", "doc_id": 1469, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1469", "record_count": 1, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bellsystem_BSRS_353.317", "split": "test", "text": "This index lists the information that forms a part of, or supplements this specification, and indicates the \nauthorized issues thereof.\n\nPAGE 1 .ISSUE\n\nHMIIIM\n\nISSUES AUTHORIZED ON ABOVE DATE\n\nTITLE\n\nWiring Diagram\n\nCordin\n\na- No longer required\n\nb\u2014The D Series Bell System Practice of the same number shall be used until the BSRS is available, \nc\u2014Information in accordance with that currently authorized.\n\nBell Telephone Laboratories, Incorporated - Dept 5222\n\nPrinted in O.S.A\n\nBSRS-353. 317 \nPage 2, Issue 3\n\n1. GENERAL\n\n1. 01 This specification covers the connections and cording for 592-type\n\ntelephone sets.\n\n2. CONNECTIONS AND CORDING\n\n2. 01 Connections for the 592-type telephone sets shall be in accordance \nwith drawing LA-540 264. \u2022\n\n2. 02 Cording for the 592-type telephone sets shall be in accordance with \ndrawing LA-540265.\n\nCO-+Q333Z\n\np/ngep\n\nnotes :\n\n/ - CONTACT SEQUENCE :\n\nREMOVING HANDSET\n\n/. cA Closes before de \nz. e 3 opens \nRESTORING HANDSET \n/. fg CLOSES \n2 de. OPENS BEFORE zb'.\n\n2\" FOR MANUAL SERVICE, REPLACE DIAL WITH APPARATUS\n\nBLANK, AND GREEN LEAD FROM TERMINAL !Z CF TERMINAL \nSTRIP ASSEMBLY SHALL BE TRANSFERRED FROM (F) \nTERMINAL TO (RR) TERMINAL AT NETWORK .\n\nCONNECT TO TERMINAL \n'F~ ON NETWORK. (.SEE - \nNOTE 2)\n\nPW\n\nli:ll\n\nMTS CORD {SEE TABLE \nPOP CONNECTIONS)\n\nTERMINAL\n\nSTRIP\n\nTERMINAL STRIP\n\ncn microphone e\n\nCONTROL UNIT ASSEMBLY\n\nS 6 7 : 8 A '10 II 12 j!3\n\n5) (c) c ) Lo) ro^ K3) to) ig) I) !\n\nVIEW FROM TOP OP SET \nLESS MICROPHONE (j CONTROL UNIT ASSEMBLY\n\n* TERMINAL CN NETWORK\n\n! telephone:\n\nKEY T MTG~\\\n\nCOLOR OF MOUNTING CORO CONDUCTORS TO TERMINALS ON MICROPHONE<1 CONTROL UNIT ASSEMBLY\n\nI 1\n\nSET\n\nFEATURES' CORD\n\nBk 3K\n\n'BAY BA-BL'BB-W\n\nP-W \\ G-Y RrBK ! / ! G '3R-G\\R-G R-3L~R-Y | BR-R\n\n12\n\n\u25a0\n\nBL\n\nW i\n\n1 bk 1\n\nVAN UAL ! VIAL\n\ni\n\n_\n\nTjv2A-3 S92B-3\n\nON-OFF D/8F-3\n\n3>\n\nS I 6 7\n\nS': 9 | 10 // iz\\ IS \u25a0 IS 17 \\ 17 \\ IS\n\n19\n\n20\n\n2-! \\\n\ni RP !\n\nr- ' r\n\n.... L _\n\n~ r\n\n! ! I\n\ni 1 J\n\nMICROPHONE\n\nVIEW PROM UNDER SIDE OP \nM/CROPHONE tj CONTROL UNIT ASSEMBLY\n\nNOTES:\n\nA. (L FOR METHOD OF CONNECTING AND SOLDERING LEAD\n\nWIRES TO CONTACT SPRINGS AND EYELET TYPE TERMINALS \nREFER TO LA850M2-33, FIGS. 84 & 85A RESP.)\n\n8. (L FOR METHOD OF WRAPPING AND SOLDERING LEADS TO \nCONTACT SPRING TERMINALS REFER TO LA850112-37 \nFIG. ioi.;\n\nC. ALL OTHER SOLDERING SHALL 8E PER KS524 METHOD C.\n\nNOTES CONTINUED\n\nD. ALL WIRES ARE LEADS OF INDIVIDUAL APPARATUS, \nUNLESS DESIGNATED WITH ITEM NUMBERS.\n\nE. ITEMS 5, 6, 24, 25, 26, 27,428 \u00ab=*\u00bb ARE \nSPECIFIED IN STOCK LIST OF LP14A241\n\nL DESIGNATES GUIDANCE INFORMATION\n\nDIMENSIONS UP TO AND INCLUDING 72 INCHES EXPRESSED IN INCHES. * \nNON-LIMITED DIMENSIONS. OTHER THAN SIZE OF RAW MATERIAL. SHALL' \n\u25a0E HELD WITHIN FRACTIONAL DECIMAL\n\nDssui \n4 CHI\n\nUNDER SIDE, POT TO \nR6REE WITH LP-HAi5+, \nISS.E. TERM STRIP TO \nAGREE WITH if-J4A242, \nI iSS+BK&W LEAPS \nFROM EQUALIZER \nWERE SHOWN CONNECTED \nWITH CORD TIPS TO \nSCREW TERMS. 20 & 2/\n\nRESP. Of term strip\n\nON TOP SIDE, \nRemoved . resistor\n\nDESIGNATED item 42,\n\nafTWfCA/ SOLOe\u00ab\n\nTEK MIS . Z 4 *\n\n-T\u00a3/z*7. \u00abr/?/p. por.\n\nEeads \u00bb*: Ey weed\n\nSHO Wa/ PP.OF7\n\nTERM*. Z RESP. ON\n\nTOP S/DG OF TEPFJ.\n\nST/ZIP.\n\nI CHANGED: SAI VIEW FROM\n\n\\T0P ETC. TERM. STRIP TO\n\nI AGREE WITH IP-I4A242.\n\nISS. 4.\n\n\u2022 ' ' \u25a0 T.'\n\n1\n\n1\n\n1\n\nW.M\n\nlEZMQB\n\nmm\n\nAS-\n\nA-\n\ns&/9/R/ee~rn\n\nE /\n\nCEO.\n\n22L.\n\nCH4NOC4 ' IHAIOMT *t\u00a3W \nleaf/ f#ov re*M /a ro \n\u2022OM~ KEV \u00a3 FKO ht T tiKM \n/\u2666 TO Off KfT -\n\nm\n\nCO - 164**1\n\n*4*\n\nISSUE 4\n\nCHAN CEO . IN V/EWEROm\n\n\u2022 *0\n\n592 TYRE \nTELEPHONE \nSETS\n\nWIPING DIAGRAMS\n\nWESTERN ELECTRIC CO. INC\n\nENGINEER OF MANUFACTURE\n\nBell Telephone \nLaboratories. Inc\n\nLA- 54 - 02 . 64 -\n\nmm\n\n* LEADS ARE EROM MICROPHONE \nCONTROL UNIT ASSEMBLY.\n\nS S-R\n\n\\\n\nS-R .\n\nw .\n\nBK \u25a0\n\nVIAL TERMU)\n\nSWITCH\n\nHANDSET CORD\n\n*\u25a0 BR-BK \nBR-BK\n\n* BR-BL\n\n* BR-W\n\nRINGER\n\nMTS 'CORD \n(SEE NOTE A )\n\nNETWORK\n\nW\n\nW\n\nBK\n\nR\n\nN\n\nW\n\nBL\n\nR\n\nG*\n\nBR-R\n\nS-Y\n\nTERMINAL strip \nON MICROPHONE \nC CONTROL UNIT\n\nG\n\nY\n\nR- BK \nS-BR \nG-Y \n\u25a0 Z\n\nR-W\n\nBK\n\nNOTE :\n\nA THE MOUNTING CORD SHALL BE DRESSED TO \nAPPROACH AS NEARLY AS POSSIBLE THE \nMETHOD OF DRESSING AS SHOWN.\n\n60 T\n\nIDENTIFICATION NO.\n\nDESCRIPTION\n\nI REF DESIG\n\nG*%lS0mde:ir\\G AS \\ c flJ\n\n61\n\ntiSSUE /\n\n59\n\nw/ K\n\nSTOCK LIST\n\nSUPPLEMENTARY DRAWINGS\n\nSIGNIFICANCE OF SYMBOLS AND ITEM NUMBER SUFFIXES\n\n\u2014PARTIAL ASSEMBLIES OR OTHER \nDRAWINGS WITH STOCK LISTS \n\u2014FARTS SHIPPED LOOSE\n\nITEMS NOT SPECIFIED AS \n\u2022 OR \"D\" PARTS IN ITEM COLUMN \nARE INDIVIDUAL (\"A\"l PARTS.\n\nWHERE ITEMS HAVE REFERENCE DESIGNATIONS. THEY MAT BE \nUSED IN LIEU OF ITEM NUMBERS ON THE DELINEATION.'\n\n\u2713 IN \u201cRCO\u201c COLUMN OF STOCK LIST\u2014QUANTITY AS REQUIRED\n\n\u2713 IN \u201cREQ\" COLUMN OF SUPPLEMENTARY DWG LIST\u2014DRAWING REQUIRKO \n\u2022 DENOTES MANUFACTURE DISCONTINUED\n\n\u25a0 DENOTES MANUFACTURE LIMITED \nAS\u2014OENOTES APPARATUS BOOK ORAWINOS\n\nA-\n\nCO- U-OS33Z\n\nissue z\n\ntifTSS\n\n59 C TYPE \nTEL SETS \nCOP DINS \nDIAGRAM\n\nWESTERN ELECTRIC CO. INC\n\nENGINEER OF MANUFACTURE\n\nBell Telephone\n\nLABORATORIES. INC\n\nDIMENSIONS UP TO AND INCLUDING 7E INCHES EXPRESSED IN INCHES. \nNON-LIMITED DIMENSIONS. OTHER THAN SIZE OF RAW MATERIAL. SHALL \nBE HELD WITHIN FRACTIONAL DECIMAL\n\n3S\n\nLA- 540 P 65 -", "title": "Telephone Sets - 592 Type - Connections and Cording", "trim_reasons": ["leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise"], "year": 1956}
{"archive_ref": "bellsystem_BSRS_105.001", "canonical_url": "https://archive.org/details/bellsystem_BSRS_105.001", "char_count": 20945, "collection": "archive-org-bell-labs", "doc_id": 1497, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1497", "record_count": 105, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bellsystem_BSRS_105.001", "split": "test", "text": "This index lists the information that forms a part of. or supplements this specification, and indicates the \nauthorized issues thereof.\n\nBell System Repair Specifications which have been covered in \nBSRS Lists320 through328 prepared by the Indiana Publication Center. \nIt is the intent that BSRS Lists issued subsequent to List 328 are to \nbe considered as supplementary to this checking list until it is \nreissued. It is the plan to reissue this checking list semiannually.\n\n1.02 The \"Issue\" column of this checking list indicates the authorized \nissue of BSRS Page 1 Indexes. Since BSRS's do not have an issue \nnumber for the entire document due to the method of revision whereby \nindividual pages and drawings are reissued, the issue of the BSRS \nPage 1 Index is used for reference and record purposes.\n\n1.03 Changes and additions are indicated by asterisks (*) in the left- \nhand margin.\n\nSCCPE ANC PLAN FOR ISSUANCE CF BELL SYSTEM REPAIR SPECIFICATIONS \nALPHABETICAL INDEX - ALL DIVISIONS \nCHECKIN!-, LIST\n\nSTATICN AND OUTSIDE PLANT APPARATUS AND EQUIPMENT - GENERAL REQUIREMENTS \nCIRCUIT PACKS AND PRINTED WIRING BOARD ASSEMBLIES - GENERAL REQUIREMENTS \nTELEPHONE SETS - GENERAL REQUIREMENTS\n\nCOIN COLLECTORS AND CCIN TELEPHONE SETS - GENERAL REQUIREMENTS \nAPPEARANCE STANDARDS - STATICN APPARATUS - GENERAL REQUIREMENTS \nSTATICN APPARATUS - APPLICATION CF CORROSION INHIBITOR \nSTATICN PLLSS -274 AND 233 TYPE - CORDING \nTELEPHONE SETS - 592 TYPE - CONNECTIONS ANC CCRCING \nMULTISLDT COIN COLLECTCRS - CONNECTIONS\n\nHAND TELEPHONE SETS ANC TELEPHONE SETS - MODIFIED AND RAFIC RECOVERY \nTELEPHONE SETS - 59? TYPE - RECOVERY REQUIREMENTS\n\nMUL TI BUT TON K r Y TELEPHONE SETS - 500, i5C0, 25C0, ANC 35C0 SERIES - RECOVERY REQUIREMENTS \nTELEPHONE SETS - 532, 533, 535, AND 536 TY PES \nT EL C PHONE SETS, 5 54,555 , r 56,557 5 5B,ANC 559 TYPES\n\nTELEPHONE SETS 6C0,1fcOC,260C AND 3600 SERIES (CALL CIKECTCRSl- RECOVERY REQUIREMENTS \nTELc-PHCNE SETS - PRINCESS (R) TY FES - RECOVERY REQUIREMENTS\n\nTELEPHONE CONSOLES, SELECTOR CONSOLES, 12-14 TYP.E APPARATUS UNITS ANC 600 SERIES TEL'PHUNE SETS ATTENDA \nNT CONSOLES - RECOVERY REQUIREMENTS\n\n5 BlA TELEPHONE S r T EASE G F-56659 G E-56660 TELEPHONE SET- RECOVERY REQUIREMENTS \nTELEPHONE SETS-A NT I CUE-MCDIFICATICN ANC RECOVERY REQUIREMENTS \nTELEPHONE C1NSCLES-14 AND 15 TYPFS-RECOVERY REQUIREMENTS \nTf'L = PHONE SETS-2671 TYFE- RECOVERY REQUIREMENTS\n\nDIALS - 21 TYPE - RECOVERY RECUIREMENTS \nCIALS 22-, 25-, 35ILCI-, AND 66-TYPE \nDIALS-26.36.AND 67 TYPFS RFCCVERY REQUIREMENTS \nEILTFRS - RADIO FREQUENCY SUPPRESSION TYPE \nDIALS to, 7, 3, 9, 10, AND 11 TYPES \nDIALS-10C8,1025,1035,1036 AND 1066 TYPE \nDIALS - 51 TYPE \nDIALS - 3 J AND 82 TYPE\n\nAND C16Cli3 (SPECIAL NUMBER 228A) CCNNECTING BLOCKS - 18 TYPE \nl RELAYS - 114, 124, 126, 174, AND 198 TYPES - RELAY COVERS IA ANC 2A\n\nRELAYS - 2 9 0 TYPE \nR rlAYS - ?99, 2 5 C\u00bb AND 293 TYPES \nRFLAYS - 156 TYPE \nRELAYS - GENERAL RECUIRFMFNTS \nN1. 49 JACKS AND ASSOCIATFC JACK MOUNTING \nNOS. 92 AND 292 JACKS AND ASSGCIATEC JACK MOUNTINGS\n\nSTEP-BY-STEP SYSTEMS WIRFD ASSEMBLIES WITH 197 ANC 198 TYPE SWITCHES \nSTEP-BY-STEP SYSTEMS - SWITCH BANK MULTIPLE ASSEMBLES \nPANEL SELECTOR COMMUTATORS \nBRUSHES - 11, 12, 13 ANC 15 TYPES \nTO 11 TYPE CLUTCHES \nPANEL RACKS\n\nC). AND JU CARRIER TELEPHONE - GFCUF RECEIVING AND R E PE A T t R-AMPL I F I ER UNITS \n11 CARRIER TcLFPHONE-GPPUP CSCILLATCR UNIT\n\nCl AND C\\ll CARRIER TELEPHONE - REPEATER CSCILLATCR ANC MISCELLANEOUS - OSCILLATOR UNITS \nLX CARRIER AMPLIFIERS\n\nCN CARRIER ( J 9 F 7 06) - RECFIVING AMPLIFIER AND CEMCCULATOR ANC TRANSMITTING AMPLIFIER ANC MODULATOR UNIT \nS\n\nON /K CARRIER (J987C9) - RECEIVING AND TRANSMITTING AMPLIFIER UMTS \nCN/K CA.RRI = R (J987C9) - ADJUSTABLE DEVIATION FCUALIZER UMTS \nA 6 CHANNEL RANK EQUIP W FNT-J68853A MCCEM UNIT\n\nSECONDARY GROUP AMPLIFIER AM) DISTRIBUTICN MOGUL F EC-50105-31 - CONSISTING OF 230A AMPLIFIER AND CISTRI\n\n'\u2022.-TYPE COUNTERS - 4A, 4B, 4C \n2 c 4A SWITCH \nH) CIRCUIT PACK \nH 2 CIRCUIT PACK \n25 3 A SWITCH \n2 5 SWITCH\n\nJ 6 a 64 7 VOICE-FREQUENCY AMPLIFIERS \nEl, r2\u2022 AND E 3 TELEPHONE REPEATERS \nTELEPHONE REPEATER ( J9 9 2 5 ? )\n\nM\u2019\u00bb350 3 TYPE N 2 CARRIER (J99272) - N 2 WM 1 WIDE-eAND MODEMS - TRANSMITTING AND RECEIVING UNITS\n\n411.301 1 Hi DATA CARRIER TERMINAL UNIT (J7C.52AB AND J7C152BBI (PLUG INBOARDS ONLY! ANC POWER SUPPLY\n\n41 1.302 1. 81 DATA CARRIER TERMINAL - CARRIER SUPPLY UNIT - (J7C152AA AND J7C1528A1 - (PLUG IN eCARCS CNLY1\n\n411.443 , c JfcPR c 6AP-i,LIST 1,3,4 6 J6RB58AP-2 C-2C RECEIVING SUPERGROUP REGULATED AMPLIFIER -L-MULTIPLEX\n\n411.445 1 F 452=4 THROUGH F 55257 SUPERGROUP DEMODULATOR NETWORKS ANC F 55256 THROUGH F 55?Hi SUPFRGROUP MCOLLATC\n\nJ6885EAC SUPERGROUP MODULATOR CR DEMODULATOR, CARRIER FILTER \u00a3 BANCPASS FILTER MODLLPS\n\nJ68818 - TONE IDENTIFICATION AND BALANCING UNIT - L2 CARRIER SYSTEM (PEARL) - INCLUDING 62 TYPE CSCILLA\n\nED-5C657-30 MAJOR CARRIER ALARM - MASTERGROUP MULTIPLEX \nMASTERGROUP MULTIPLEX - ED-506C5-31 128 TO 512 KHZ MULTIPLIER ASSEMBLY \nMASTERGROUP MULTIPLEX - EC-50605-32 1024 TO 4C56 KHZ MULTIPLIER ASSEMBLY \nL4 C Afifi I \u00a38-1 8 A MEMORY LMT \nL4 CARRIER - ISA MEMORY LMT\n\nL4 CARRIER - J68877AF-), LIST 1 THRU LIST 6 - TEST OSCILLATOR ANC LOGIC UNIT \nL4 CARP. IER-J66877AG COMMAND RECEIVER\n\nMASTERGROUP MULTIPLEX - FC-506C5-33 8152 TC 16384KHZ MULTIPLIER ANC ALARM ASSEMBLY \nEC-5 C615-30 TRANSFER CONTROL CIRCUT - MASTERGROUP MULTIPLEX \nEC-50622-31 OSCILLATOR ALARM CIRCUIT - MASTERGROUP MULTIPLEX\n\nL~4 CARRIER-J66877AS-1.LIST l.LIST 2A G LIST 3 TEST CSCILATOR AND LOGIC UNIT \nMASTERGROUP MULTIPLEX - EC-50621-30 PILOT GENERATOR ALARM AMPLIFIER \nED-5C62 2-30 PILOT ALARM SY STEM-MASTERGROUP MULTIPLEX \n4A ECHO SUPPRESSOR - CIRCUIT BCARCS - (J68514!\n\nL4 CAPRIcR-J6E877AB-2 LIST 7 REGULATING REPEATER \nCATV EUUIPMENT-JERROLO ELECTRONICS CORP.-\u2022STARL INF\u2022\n\nDATA SYSTEMS CENTRAL OFFICE - WLR4 AND WLR5 WICEeANC LOOP REPEATERS J70171AC, J70171RA AMPLIFIERS\n\nDATA SYSTEMS CENTRAL OFFICE - WLR4 AND WLR5 WIDEBAND LOOP REPEATERS J70171AC GROUP PRFAMPLIFIEKS AND J7\n\nDATA SYSTEMS CENTRAL OFFICE - WLR4 6 WLR5 W ICEPAND LCOP REPEATERS J7C171AG, LI GROUP CONTROL TESTS G J7 \nJi 7 iAT, LI SUPER GRUUP CCNTRCL TESTS\n\nDATA SYSTEMS CENTRAL OFFICE - WLR4 6 WLR5 WIDEBAND LCCP REPEATERS J7C171AR GROUP PILOT OSCILLATORS G J7 \nC171AL SUPER GRCLP PILOT CSCILLATCRS\n\nDATA SYSTEMS CENTRAL OFFICE - WLR4 G WLR5 WICEeAND LCOP REPEATERS J7C171AJ-L1, L2 LOCAL POWER SELECTCRS\n\nJ7C171AK-LL. L2 RFMCTE PCWER SELECTORS J7C171AW- LCCAL FCWEP SELECTCRS ANC J7jl 7 ]AY- o EMCTC POWER \u00abELE\n\nCATA SYSTEMS CENTRAL CFFICE - WLP4 6 WLR 5 WICEEANC LCCP REPEATERS J7C171AL ALARM \nFWA.FWH, G FUA SIGNALING UNTTS-(J59335WA,J59335WB,G J9R335UA1\n\nRYA ?6CC H7 TCNE SUPPLY AND TRANSFER CIRCUIT (J55335YA-1) - FYG CARRIER GROUP ALARM CONTROL CIRCUIT (J-\n\nL4 CARR IER-J68E77AC-LIST 2 REGULATING REPEATER \nL4 CARR IFR-J6EE77AN-LIST 2 REGULATING REPEATEP\n\nL4 CARRIER - J6PF77AE, LIST 1, LIST 2 ANC LIST 2A MEMORY UNITS \nL4 CARRIER - J6F 877AR, LIST 1 G LIST 2 MEMORY UMTS\n\n14 CARRIER J6S877AB-2 LIST 0 REGULATING REPEATER \nL4 CARRIER J60677AC-1 LIST 3 REGULATING REPEATER \nL4 CAPRI\u00a3R-J6E\u20ac77AM-1\u00bbLI$T2-PEGULATING REPEATER \nL4 CARRIER J66E77AN-1 LIST 3 REGILATING REPEATER \nL4 CARRIER - 28A GATE \nL4 CARRIER - 6 0 A CONTROL LMT\n\nLAMP SOCKET MOUNTINGS - 136 TYPE, NO. 137B, 236 TYPE, 275 TYPE AND NCS. 278B, 281A, 283A, 290A AND 251A \nLAMP SOCKETS - NOS. 12, 12B, 13, 41A, 43A, 47B, 45A, 50A, 51B, AND 53A \nELFCTRIC CLOCKS - NOS. 1A, le ANC 1C \nCALCULAGRAPHS - MODEL 6 \nCALCULAGRAPHS - MODEL 3C \nCALCLLAGRAPHS - KS-7765 - MCCEL 33\n\nALTOMATIC MESSAGE ACCOUNTING - REAOERS KS-13S35, LISTS L, 2, 3, 4, 5, 6 , AND 7 \nAMA TAPE REELS, BINS ANC RECEPTACLE\n\nCONVERSION OF AMA PERFORATOR CABINET EQUIPPED - WITH OUTPUT TAPE BINS (ED-40028-011 TO CABINET EQUIPPED \nWITH OUTPUT TAPE REELS (EC-40025-01 ANO EC-26301-11)\n\nAC/OC DUPLEX MOTORS - KS-5C2C, KS-5021, ANC KS-5407 REPAIRED AC/OCDUPUEX MOTORS - KS-5440 \nFREOLENCY GENERAT0RS-1C7 TYPE \nJ072C2A, LIST 1 RECTIFIERS \nJ86738A, LIST 1 POWER SLPPLY UNITS\n\nENCLOSED LEAD-ACID TYPE STORAGE eATTERIES- KS5361, KS15577, KS15754 AND KS15886 \n8C7D TONE POWER PLANT - FLUG-IN UNITS - IJ86825C, 0, E, AND C-)\n\n404A-G TONE GENERATORS - 1CA ANC 8 TONE MONITORS, ANC 174A INTERRUPTER \nJ86815C. C, E, F, G, AND H FLUG-IN UMTS \n24-TYPF POWER UNIT \n23A POWER UNIT\n\nJ67266F 2C-HZ INVERTER - FLUG-IN UMTS - (J87266F, G, ANC HI \n158 TYPE REGULATOR \n\u2022 2 TYFE POWER LMT \n39A POWER UNIT\n\nJ 8 729 7 A,3, ANC C - OC TC DC CONVERTER \nJF72FPA-CC TO DC CONVERTER REGULATOR \nRECT IF IER-J872 C5A.8 \nPOWFR PLANTS-(J8t485)\n\nPOWER SUP\u2019LY ANC RINGING AND TONE FOR PCCA FriX - J <E f: 3 5 A ANC B \n1 C1 A ANC ICIB-FRfcOLENCY GENERATOR\n\nI12A frequency gfneratcf \nP 8 X SWITCHED ARCS - 5C5 ANF 5C6 TYPES \nCORDLESS POX SWITCHBOARDS - 507 TYPE \nPRX SYSTEMS - NO 5 5 8A PBX\n\nPBX SWITCHBOARDS - COR r TYFE (INCLUCING PLATE FORMS I-GENERAL REQUI REPENTS \nPBX SWITCHBOARDS - NO. 6G5A ANC 6C7 TYPFS \nR 8 > 5YSTFYS - 6 C 8 A PBX \nRwITCH FRAMES - 74C TYPF FPX \nP IX SYSTEMS NC. 7 * 5 A P?X\n\nS * ITCHING ,r CUlPRHNT - NC. 7C1A ANC 713A F8XS \nPBX SYSTEMS NC, 7 5 8 A PCX \nS*ITCHING EQUIPMENT, 7C1B ANC 711H PRXS \nPBX SYSTEM NJ.757A PPX\n\n?A AUTOMATIC CALL DISTRIBUTING SYSTEM - (EQUIPMENT CABINETS, CABLING AND CONSOLES \n7 11 A PRX - Hv)TEL-MCTE L SERVICE\n\nTFIEPK'Nf ANSWERING SYSTEM NC. 1A - (ECUIFMENT CABINETS, CABLING, ANC CONSOLES!\n\nATTENDANT CABINETS ISED WITH NC. 740A, 7406, AND 740C PBXS \nSCLOERING COPPERS - KS-P74C ANC KS-14440\n\nTEST OSCILLATCR-HEWLETT-PACKARD CCMPANY-MCCEL 65CA \nRELAY TIMING TFST SET JS4713A\n\nCONNECTOR, SELECTOR, ANC TRUNK TEST SET J34722B \nPORTABLE AC RELAY AND SIGNALING TEST SET-J68602AJ Ll \nPORTABLE TRUNK TEST SET-J54729A\n\nMAGNFTIC LATCHING RELAY TIMING TEST SET - J54735A \nICOA.e C COMMUNICATION SET \nRESISTORS - 18 AND 19 TYPES \nKEY SPACES\n\nOPERATOR CHAIRS IMETAL TYFE) - AND PURSE HOLDERS \nSUPERVISOR CHAIRS (METAL TYPE) - ANC SUPPLY CABINETS \nDIODES AND VARISTORS\n\nCOMMON SYSTEMS-LINE CONCENTRATOR N0.1A-REMCTE UMTS \nTOUCH-TONE CALLING-RECEIVER CIRCUIT-J99266(TYFE A2)\n\nJ9512CB MF RECEIVING UNIT - H295-119, H295-12C ANC H295-121 \nPBX TOUCH-TONE CALLING-RECEI VtNG CIRCUIT-J58844A ANC J58844B \nPB X SYSTEMS - SCLIC STATE CCNVERTER J58847AC \n2A ANC 28 - LINE CCNCENTRATCR - CIRCUIT PACKS \nELECTRONIC DRAWER ASSEMBLIES - E0-71291-10, G1 THROUGH G13 \nAE1, AE2, AE3 ANC AE4 CIRCUIT PACKS - (3A CCMMUNICATICN SYSTEM)\n\n* 421.001 11 AGP TYPE CIRCUIT PACKS G NC.l ESS LINES G TRUNKS-CENERAL REQUIREMENTS G INDEX\n\n421.C 66 1 NC.l ESS HIGH CURRENT PLLSER 6 PATH CHECK CIRCUIT - ELECTRONIC FERREfcC PULSER\n\n4 2 1 .00* 4 \\T). 101 ELECTRONIC SWITCHING SYSTEMS - HETERCCFNCUS FULSt-TESTEC CIRCUIT PACKAGES\n\nh 2 , .OOA 7 NO. 10! ELECTRONIC SWITCHING SYSTEM - OSCILLATOR ANC TONE CENERATQR CIRCUIT PACKAGES\n\n4 2 3 \u2022 C C G 2 N). 1C! ELECTRON 1C SWITCHING SYSTEM - MCNCFULSEP AND TIMFR C IF. CU ITPACK AGE S\n\niil.J'J l RL-CTRONIC SWITCHING SYSTFMS - NC. 101 - LCGIC Tf.STCr S-TYPE CIRCUIT PACKS\n\nh 2 7 \u2022 Ill 1 ELECTRONIC SWITCHING SYSTEM - NC. 101 - S-TYPE CIRCUIT PACKS - OSCILLATORS AND TON\" GE N>\" R A TCP S\n\n4?j.on 1 ELECTRONIC SWITCHING SYSTFMS - NC. 101 - S-TYPE CIRCUIT PACKS - AMPLIFIERS AND SWITCHES\n\n423.104 1 SCOA PBX-CIRCLIT PACKS AC1C.AC13 C AC51.REC-ISTER CIRCUIT FOR DIAL PULSING 6 TOUCH-TONE CALLING\n\n423.105 2 SCOA PBX - CIRCUIT PACKS AC23 AND AC59 - CNE-WAY OUTGOING CENTRAL OFFICE TRUNK CIRCUIT\n\n423.106 ? SCOA PBX CIRCUIT PACKS AC20 AC22. AC52 AC58, AC63 AC66 CNE WAY INCOMING OR TWO WAY CENTRAL OFFICE\n\n423.111 1 SOOA PBX - CIRCUIT PACK ADI - FUSE, ALARM, EMERGENCY TRANSFER AND TEST CIRCUIT\n\n423.112 1 SCOA PBX - CIRCUIT PACKS AC31, AC32 AND AC33 - ATTENDANT CONTROLLED CONFERENCE CIRCUIT\n\n4 3 2.012 1 ELECTRON 1C SWITCHING SYSTEMS - NC. 101 - S - TYPE CIRCUIT PACKS -NONCPULSERS AND TIMERS\n\nKS-16378 TELEPHCNE ANSWERING SET ANC ACCESSORIES \nANNOUNCEMENT SETS KS-16765 ,LIST 1 AND LIST 2 \nKS-16523 TELEPHCNE ANSWFRING SETS AND ACCESSORIES \nKS-19245 TELEPHCNE ANSWERING SETS ANC ACCESSORIES \n1A TELEPHCNE REPORTING SET \nCONTROL LMT 62A\n\nK$-6846\u00bbKS-81C8\u00bbKS-81C9 ANC KS-8110 eUZZERS, 4 AND 7 TYPES EXCEPT 7G AND KS-1378? ANC 7 TYPE BELLS \nEU7/FRS -0 TYPE\n\nTRANSFORMER RELAY SFTS KS-7338, ANC KS-8233 AND RELAY SETS KS-16626 AND KS-16763 \nCODE SFN0IN3 STATIONS AND TRANSFORMER UNIT - KS-14468 \nSIGNAL 5 - KS-16 3C1\n\nK S-19120 SCHCCL-IC-HCME SETS \nLOUDSPEAKER SETS -ICO TYPE \nAMPLIFIER - KS-16754 LISTS 1,2,3 AND 4 \nLOUDSPEAKER SEIS-lC6A , B ,C , 0 , E , F \nLOUDSPEAKER SET-137 TYPE\n\nTRANSFORMERS - KS-15675, KS-16184, KS-16368, KS-16886, 2C12 ANC 2075 TYPES \nTRAN\u00abFGRMERS-3S3 TYPE \nLCUOSPFAKERS-759A ANC 759P TYPES \nLOUDSPEAKERS - 76CA TYPE\n\nCONNECTING BLOCKS - 66 TYPE - 72ei-50 TYPE & 1A1 MATRIX BLOCK \nAMPLIFIER - KS-15220 TYPE \nAMPLIFIER - KS' 15 2211 LIST 1\n\nSTAPLERS - HELLER CCNVEPTEC MODEL T ANC MODEL T2 \nB AUTOMATIC DRILL \nSTAPLER-HELLER MODEL TM\n\nFLASHLIGHTS, HEAD LAMPS, SPOTLIGHTS, ELECTRIC HAND LANTERNS, AND DUAL BATTERY LIGHTS\n\nTELETYPEWRITER COMPONENTS-GENEPAL REQUIREMENTS \nTELETYPEWRITER MOTORS, MOTOR UNITS, AND GOVERNORS \nTELETYPEWRITER RECTIFIERS\n\n714 TYPE REPERFORATOR TRANSMITTER UNITS 114 TYPE REPERFORATOR TRANSMITTER UNITS 114 TYPE REPERFGRA\n\nNCN-TYPING S TYPING PERFORATOR \u00a3 REPERFORAT OR,RECEIVING SELECTOR fiTAPE TYPING UNITS - 28 TYPE\n\nTYPING UNIT, SUB-BASE 6 DISTRIBUTOR 6 ANSWER-PACK MECHANISMS - ASSOCIATED WITH 33 TYPE TELETYPEWRITER-S \nETS\n\nNCN-TYPING AND TYPING REPERFCRATCR, RECEIVING SELECTOR AND TAPE TYPING UNITS - 35 TYPE\n\nSELECTCR MAGNET DRIVERS - TF182630 AND TP1S3440 \nLOUDSPEAKER AMPLIFIER - TP18275C \nCALL CCNTRCL LMTS - 33 \u00a3 35 TYPES\n\nCATASPEEC - 1A, 2A AND 5C SENDER CABINETS - IB ANC 5B RECEIVER CABINETS AND 5A-APPAHATUS BOX ASSEMBLY ( \nTP196C22 )\n\nHIGH SPEED WlRF SPRING PEACERS CXI ,CX2 ,CX3 ,CX4,0X800,CX8C1 \u00a3 DX802 \nEDGE PUNCHED CARC REACEP UMTS 35 TYPE \nTYPING UNIT - 37 TYPE\n\n37 TYPF TELETYPEWRITER - KEYBOARC ANC BASE NOT E9 EARLY UNIT YK80C WAS DROPPED AND NEW UNITS ADDED \nNCN TYPING ANC TYPING REPERFCRATCR UNITS - 37 TYPE \nELFCTRICAU SERVICE UMTS - 37 TYPE\n\n456.7C7 2 TPl529C3 AND TP15478C MODIFICATION KITS TC FRCVIOE HORIZONTAL TABULATION AND TRANSMITTER OISTRIBL\n\n456.7CB 3 TP154760 THROUGH TP15477C MODIFICATION KITS TC PPCVICE VERTICAL TABULATION ANO TRANSMITTER-DISTRIBUTOR\n\n456.7C9 1 TP17e3l6, TP178344, TP178345 ANC TP178346 MCCIFICATICN KITS TC PROVICE ANSWER BACK ANC MCTCR CONTRCL\n\n456.710 1 1P 152324 ANC TP154147 MCDIFICATICN KITS TC FPCVICE LCCAL PAPER FEEC CUT MCTOR START CN 26 TELETYPE\n\n456.713 1 TP161815 MODIFICATION KIT LSED FOR MOUNTING CN AUXILIARY TYPING REPERFORATOR\n\n456.716 1 MODIFICATION KITS LSED TO CONVERT 28 TYPING UMTS FROM FPIcflON FEED TO SPROCKET FEED ANC FROM SPROCKET\n\n456.718 1 TP173443 MODIFICATION KIT - TO PROVIDE REMOTE CONTROL - NON-INTERFERRING LETTERS FEED-OUT - 28 TYPING R\n\n456.719 1 TP16241 ANO TP163359 VARIABLE SPEEC GEAR SHIFT MCCIFICATION KIT 28 TELETYPEWRITER BASE (KSR ANC RC)\n\n456.721 1 TP-164564 MODIFICATION KIT - TC CONVERT FROM CHADLESS - TC FLLLY PERFORATED TAPE - 28 REPERFORATOR\n\n456.722 1 TP16249Q MULTIPLE WIRE CLTPUT - MCCIFICATICN KIT - TRANSMITTER DISTRIBUTOR UNIT - 28 TYPE\n\n456.725 l TP1736ei FIVE-LEVEL PARALLEL - CCCE READING CONTACTS - MODIFICATION KIT - MODEL 28 TYPING UNIT\n\nTP174216 MOO IFICATICN KIT - TO ECUIP MARK 11 KEYBOARD - WITH ANSWER-EACK - MODEL 28 KEY80ARD \nTP175724 MODIFICATION KIT - TC PROVIDE TWO-CCLCR PRINTING CN MOOEl 28 TYPING UNIT\n\nTP174235 AND TP178834 MCOIFICATICN KIT - TC CONVERT CHADLESS TO FULLY PERFORATED TAPE - MCOEL 28 TYPING \nREPCRATOR\n\nTP 333672 MOUNTING FRAMF ANC \u2022 STLNTRCNIC\u2022 (SA120AG1 CATA TERMINAL- ACCESSORY ASSEMBLY \nSTUN TRONIC S A113 CATA TERMINAL ACCESSCRY PARITY ERKCP CETECTCR\n\nSTUNTRONIC SA120 DATA TERMINAL ACCESSOR IES,PAR ITY ERROR OETECTORS ANC PARITY ERROR INSERTORS \nJ70112 A CHANNEL TERMINALS-43A1 (CARRIED TELEGRAPH ECU IPMENT)\n\nNETWORK-453 6 454 TYPES-(LSED WITH 43A1 CARRIER TELEGRAPH TERMINAL EGUIPMENT1 \n144 TYPE COUPLING UNITS AND 56A1 LCCP REPEATERS \n43B1 VOICE FRECLENCY - CARRIER CATA SYSTEM\n\nASSEMBLED TELETYPEWRITER STATIONS M28 ECUIPPEC WITH/WITHCUT SUB SET OF DATA SET FOR PRIVATE LINE SERVI \nCE\n\n15-ANC 15-TYPE ASSFMBLEC STATIONS - 60-WCRC-PER-MINUTE TWX \nASSEMBLED TTY STATIONS - 33-TYPE\n\nTELEGRAPH SYSTEMS - 83B3 SELECTIVE CALLING SYSTEM - ASSEMBLEC 28TYPE OUTLYING STATIONS\n\nTELEGRAPH SYSTEMS - P3B3 SELECTIVE CALLING SYSTEM - ASSEMBLEC 28ASR CONTRCL STATIONS\n\n3A1, ANC FBI CATA SELECTIVE CALLING RC ST AT I CN (35EE, 35BF1 \nCATASPEEC-COMPLETE STATION TYPE 4\n\nASSEMBLED TTY STATICNS M35 TTY SETS ECUIPPEC WITH/WITHCUT SUBSETS CR CATA SETS FOR PRIVATE LINE \nSERVICE\n\nCATASPEEC(R)MAGNETIC TAPE TERMINALS STANC ALCNE CR USED WITH A LOW SPEED ADJUNCT\n\nCATA LINE CONCENTRATOR SYSTEM - CATA-PHCNE - PRIVATE LINF - El A ANC CIRECT NEUTRAL INTERFACE TERMINAL S \nT ATICNS - M 3 3 \u00a3 M35 SElF-CCNTA INEC LCGIC SETS \nCCMMLNICATIONS DISPLAY - TERMINAL \nCLUSTER CCNTRCLLER - TERMINAL\n\nTELEPHONE WIRING, BCCTHS AND SIMILAR MOUNTING ARRANGEMENTS \nSHELF - 20 TYPE \nMOUNTING - KS-167C5\n\nKS-19425 ANO KS-19580 - TELEPHONE BOOTHS ANC ACCESSORIES \nCOIN RECEPTACLES AND CCIN RECEPTACLE COVERS\n\n457.220 2 COIN COLLECTORS -MULTISLOT TYPES - CONVERSIONS TO PULL BUCKET RETURN CHUTE\n\n457.223 9 MULTISLOT COIN COLLECTORS - HANOSET TYPES - CCNVERSICNS FOR USE INNETWORK CIRCUIT\n\n457. 220 4 CCIN CCLLECTORS,MULT ISLCT TYPES, CONVERSION TO SLOW RELEASE CCIN RELAYS\n\n472.012 3 55A RADIO RECEIVER AND ACCESSORIES (8 TYPE BATTERY CASE, 32 TYPE RECTIFIER AND 215 TYPE SELECTORS)\n\n473.014 2 PORTABLE RAOIC - LCW PCWER TRANSMITTER-RECEIVER - HALL I CRAFT ER LITTLEFONE HT-22 TYPE\n\n473.015 1 PORTABLE RADIO-LCW-POWER TRANSM ITTER-RECEIVER-INCUSTRIAL RADIO COMPANY(PAK-FONE)\n\n1 M J MOBILE RADIC TELEPt-CNE SYSTEM - RACIO TRANSM ITTER-RECEIVER - KS-1S6J9 (MOTOROLA)\n\n2 OSCILLATORS - HEWLETT-PACKARD COMPANY-MODELS 200C, 23JCD, ANC 200H \n4 2iA TRANSMISSION MEASURING SET - (J94021A)\n\n4 VIDEO MONITORS KS-5799,LI S7S 1, 2, AND 3 (7-INCH) - KS-15654, LISTS 1 ANC 2 (12-INCH) \n4 2M REPEATER SWITCHING SET (J940J2M)\n\nMOTOFCl A P-81CC 1EST SET \nPULSING TEST SET - J34717A \nMERCURY RELAY TEST SET - J54725A \nCCNTRCL SET - J64717\n\nFREQUENCY AND MODULATION METERS - GENERAL ELECTRIC 4ST-13A ANO 4ST13A-WE \n1 AH TUEE TEST SET - (J640C1AH)\n\n3 7B TRANSMISSION MEASURING S ET-(J64037BI \nCOLD CATHODE TUBE TEST SETS - J24754A ANC J94731A \nAUXILIARY TEST PANEL ANC TELEPHONE SET - J68649C \nSELECTOR TEST SET - JS47C6A\n\nVACUUM TLBE VOLTMETER - MEASUREMENTS CORPORATION - MODEL 62 \nVACUUM TUBE VOLTMETERS - HEWLETT-PACKARD COMPANY - MODELS 4iOA,ANC41CB \nREPEATFR ACTIVITY TEST SET - J68825A \nTEST SET FOR TIMING TESTS - J24753A\n\n53A MOBILE OSCILLATOR, J64C53A, LISTS 1, 2 ANC 3 (CCNSISTING OF THERMOCOUPLE PANEL) J86221A RECTIFIER,\n\nVOLTMETERS - WESTON TYFE 531 \nPER CENT BREAK METER - KS-7361 \nTHFRMISTER CCNTRCL SETS - J6E839A ANC J68839e \nANALYZER - WESTON MODEL 779\n\n1U AMPLIFIER - RECTIFIER - IJ640Q1U) - (MODIFIED FOR USE IN 406 MOBILE TRANSMISSION MEASLRING CIRCUIT) \n32A TRANSMISSION MEASURING SET (X66C65A ANC J64032A)\n\nSIGNAL GENERATORS - FERRIS INSTRUMENT COMPANY - MCCELS 220 ANC 22CS \nFM MODULATION MONITORS - DOOLITTLE RADIO FC-9 ANC FC-9B ANC JAMES KNIGHTS FD-12\n\n1 OSCILLOSCOPE - HEWLETT-PACKARD - MODEL 130A \n1 5L SERVICING CENTER TEST SET - (JSe7C5U) \nl ECHO SUPPRESSOR TEST SET - J68605M\n\n1 TRANSMISSION MEASURING SYSTEM - J64047A - (J64047BI SENDING UNIT AND J64047C RECEIVING UNIT\n\n3 MODELS 4 ANC 4A TRANSMISSION TEST SETS (NORTHEAST ELECTRONIC CORPORATION!\n\n1C 5C2-SERIES CATA TEST SET (J79902A ANC 81 - X-67945MB DISTORTION MEASURING AND ERRCR CETECTCR SET\n\n3 36A VICEO VISUAL GAIN AND DELAY CISTCRTICN MEASURING SET- (J64C36AA TRAN EMITTER!-!J64C36AB\n\n3 36R VIDEO VISUAL GAIN AND DELAY DISTORTION MEASURING SET - (J64036eA TRANSMITTER ) - (J64C368B RECEIVER)\n\nJ-693920-1-ATTENUATCR,JACK,AND MARKER OSCILLATOR PANEL \nJ-633S2E-1-SWEEP CCNTRCL PANEL \nJ-68392F-1-IF DETECTOR UNIT \nSIGNAL GEN6RATCR-KS-2C146 \nDISTORTION CETECTOR-KS-20147\n\nSLA,8, 6 E AUTOMATIC TRANSMISSION MEASURING SYSTEM - J94051A,8,C,E - CIRCUIT PACKS \n26A MICE3AN0 GAIN AND DELAY SET (J94026A!\n\n23 SIGNALING TEST SET-!J64730B> 6 PULSE REPEAING ADAPTER(J64730CI \nKS-19763 TV hAVEFCRM CSCILLCSCCPE \nKS-1963 3-TV PICTURE MCMTCR \nKS-205C1 RETURN LOSS MEASURING SET\n\nCATA SETS 601A, 601A1, 6C1A2 AND CATA SETS 6018, 60 iei, ANO 601B2 \nCATA SET 202 7YPE-TRANSMITTER-RECEIVER-!J1C202A ANC J1C2C2B)\n\nCATA SET 201 TYPE-TRANSMITTER-RECEIVER \nJ70148 - PLUG-1N ELECTRONIC UMTS \nCATA SETS - 1C1 TYPE - 1J1C1C1I \nCATA SET 4C2A TYPE TRANSMITTERS\n\nDATA SET 3318 - TRANSMITTER - RECEIVER - (J1C301BI \nCATA SET 1038 - (J1D103B) ANC J87239A, 8 RECTIFIER \nCATA AUXILIARY SETS 801C1, EC1C2-At TCM AT I C CALLING UMTS \nCATA SETS 6 32A (ALL LISTI AND 602Al-TRANSMITTER-RECE IVER\n\nCATA AimiARY S ET S-804 A1,80 4A 2,804 A3,804 A4,804 A 5,8 04 A6,804 A 7 C 804AE \nDATA SETS 105A TYPE, CIRCUIT PACKS ANC RECTIFIER (JE7240B, LIST 1>\n\nCATA SET 1C3F TYPE ANC JE7239A ANC e RECTIFIERS \nCATA SET 202C/202D TRANSMITTER RECEIVER \nCATA SET-TYPE 4020\n\nDIGITAL CATA TRANSMITTER JIGCGCBG ANC DIGITAL CATA RECEIVER JIGOOOeG", "title": "Bell System Repair Specifications - Checking List", "trim_reasons": [], "year": 1974}
{"archive_ref": "bellsystem_BSP_002-800-901_PN", "canonical_url": "https://archive.org/details/bellsystem_BSP_002-800-901_PN", "char_count": 2836, "collection": "archive-org-bell-labs", "doc_id": 1519, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1519", "record_count": 14, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bellsystem_BSP_002-800-901_PN", "split": "test", "text": "1.01 This section is reissued to define the circuit \npack systems now included in the circuit pack\n\npool in the Washington-Idaho Area (1.03a) and to \nmake a minor revision in the wording of paragraph \n4.05. Changes are indicated by arrows.\n\n1.02 The SPCS circuit pack pools replace defective \ncircuit packs for the No. 101 Electronic\n\nSwitching System (ESS), No. 1 ESS, No. 2 ESS, \nTraffic Service Position System (TSPS), Electronic \nTranslator System (ETS), and solid state Traffic \nmeasuring devices, eg, Alston and PMC.\n\n1.03 Defective packs are replaced with spares from \nthe SPCS circuit pack pools.\n\n1.04 Twenty-four hour service is provided. The \nmethod of shipment to the field is determined \nby the urgency of the requirement.\n\n2.01 Replacement packs are ordered from the pool \nby calling the above numbers at any time.\n\n2.03 The central office, after receiving the \nreplacement pack, does the following:\n\n(a) Completes the P-7991, provides the \nteletypewriter diagnostic printout if \navailable. Places the completed form and \nprintout in the shipping container with \nthe defective circuit pack.\n\n(b) Attaches the address label to the carton \nand ships the items to the pool, reusing \nthe carton and packing materials.\n\nBCb q This material is prepared for Pacific Northwest Bell Tel. Co. purposes and is in no sense a publication. Neither the material nor any \n\u2022J\" portion thereof is to be reproduced in any form by others without the written permission of the Pacific Northwest Bell Tel. Co.\n\n2.04 The pool receives and forwards the defective \npack to Western Electric (WE) with Form\n\n2.05 Repaired packs and no-trouble-found packs \nreturned from Western Electric may be\n\n(b) Packs are returned to the original office \nand tested in the same equipment \nlocation in which they originally failed. If \nthe packs do not fail, they are left in place \nand the replacement packs are returned \nto shelf stock in the pool.\n\n3.01 Obtain replacement packs as soon as possible \nto prevent accumulation of defective units.\n\n3.02 Make sure the completed Form P-7991 and \nteletypewriter diagnostic are returned with the\n\n3.03 Make sure the pack is actually defective before \nrequesting a replacement.\n\n3.04 Make sure defective packs are returned to the \npool as soon as replacements arrive.\n\n4.01 Ship packs (same pack issue, if appropriate) to \nthe field as soon as they are requested.\n\n4.02 Verify accuracy of paper work accompanying \nshipments to and from the field.\n\n4.05 Have defective units repaired, or dispose of \nthem as obsolete or unrepairable items (eg,\n\n4.07 Maintain an adequate spare circuit pack stock \nto provide the field with any pack that may be\n\n4.08 Be responsible for preparation of all paper \nforms involving pooled items:\n\n4.09 Supply Plant field forces with enough packing \nand shipping materials to insure adequate", "title": "Stored Program Control Systems (SPCS) / Circuit Pack Pool", "trim_reasons": [], "year": 1974}
{"archive_ref": "bellsystem_SD-1C904-01", "canonical_url": "https://archive.org/details/bellsystem_SD-1C904-01", "char_count": 37939, "collection": "archive-org-bell-labs", "doc_id": 1535, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1535", "record_count": 127, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bellsystem_SD-1C904-01", "split": "test", "text": "1. WHEN CHANGES ARE MADE IN THIS DRAWING. \nONLY THOSE SHUTS AFFECTED WILL BE \nREISSUED.\n\n2. THIS SHEET INDEX WILL BE REISSUED AND \nBROUGHT UP TO DATE EACH TIME ANY SHEET \nOF THE DRAWING IS REISSUED. OR A NEW \nSHEET IS ADDED.\n\n3. THE ISSUE NUMBER ASSIGNED TO A \nCHANGED OR NEW SHEET WILL BE THE \nSAME ISSUE NUMBER AS THAT OF THE\n\nCIRCUIT PACK SCHEMATIC \nEQUIPMENT DRAWING \nKS-21447 MINIRECORDER CKT \nDC-DC CBNVERTER (J87421A)\n\n* - SCHEMATICS OF ALL FA,FB,FC AND JK-COOED \nCIRCUIT PACKS USED IN THIS CIRCUIT ARE \nSHOW ON DRMINGS NUMBERED WITH A CPS \nPREFIX FOLLOWED BY THE CODE OF THE PACK, \ne.\u00ab, CPS-JK10\n\nNOTICE- NOT FOR USE OR DISCLOSURE OUTSIDE THE BELL \nSYSTEM EXCEPT UNDER WRITTEN AGREEMENT.\n\nREQUEST TO THE SP! TO TRANSMIT AN ERROR \nSTART CODE IN THE CURRENT STATUS REPLY \nTO THE CC. LOW ACTIVE\n\nREOUEST TO BT TO GENERATE PARITY OVER \nTHE STATUS REPLY CURRENTLY RESIDING ON \nTHE COMMON PARALLEL BUS\n\nCOMMON PARALLEL BUS PARITY BIT (ODD) \nOVER THE 8 HIGH-ORDER DATA BITS. LOW = \nLOGICAL ONE\n\nCOMMON PARALLEL BUS PARITY BIT (ODD) \nOVER THE S LOW-ORDER OATA BITS. LOW = \nLOGICAL ONE\n\nADDRESS DEVICE TO RECEIVE THE DATA ON \nTHE COMMON PARALLEL BUS, ACTIVE LOW\n\nADDRESSED DEVICE TO SEND OATA TO THE SPI \nON THE COMMON PARALLEL BUS. ACTIVE LOW\n\nADDRESSED DEVICE REOUESTED TO GATE A \nSTATUS REPLY ON THC COMMON PARALLEL BUS. \nACTIVE LOW\n\nA SYNCHRONIZING SIGNAL TO THE SPI WHICH \nINDICATES ThAT A DEVICE HAS SENSED A \nCOMMAND ON THE PARALLEL BUS. ACTIVE LOW\n\nA REGUEST TO THE SPI TO WAIT UNTIL THE \nDEVICE HAS HAD TIME TO ACT UPON A \nCOMMAND BEFORE REMOVING THAT COMMAND \nFROM THE COMMON PARALLEL BUS. ACTIVE LOW\n\nINDICATES T\"AT A DEVlCt HAS SENSED A \nACKNOWLEDGE INTERRUPT COMMAW. ACTIVE COMMAND ON THE P < RALLEL BUS. ACTIVE LOW \nLOW\n\nWAITO A REQUEST TO THE SPI TO WAIT UNTIL THE \nSERIAL BUFFER BUS CARRY PULSE INDICATING DEVICE KAS HAD TIME TO ACT UPON A \nTHAT THE ON-LINE (ACTIVE) DATA BUFFER CCHHAKu BEFCf . REMOVING THAT COMMAND \nHAS BEEN EITHER FILLED OR CIFT.cD. FROM THE COMMON PARALLEL BUS. ACTIVE LOW \nACTIVE LOW\n\nSERIAL BUFFER BUS 16 BIT CARRY PULSE \nINDICATING THAT A 16-BIT WORD HaS BEEN \nTRANSFERRED BETWEEN THE ON LINE DATA \nBUFFER AND THE DEVICE CONNtCTED TO IT, \nACTIVE LOW\n\nREOUEST TO THE SPI TO TRANSMIT AN ERROR \nSTART CODE IN THE CURRENT STATUS REPLY \nTO THE CC. ACTIVE LOW\n\nCOMMON PARALLEL BUS PARITY BIT (ODD) \nOVER THE 8 HIGH-ORDER DATA BITS. LOW = \nLOGICAL ONE\n\nCOMMON PARALLEL BUS PARITY BIT (ODD) \nOVER THE 8 LOW-ORDER DATA BITS. LOW = \nLOGICAL ONE\n\nADDRESS DEVICE TO RECEIVE THE DATA ON \nTHE COMMON PARALLEL BUS AND INTERPRET \nTHEM AS A COMMAND. LOW ACTIVE\n\nADDRESSED DEVICE TO RECEIVE THE DATA ON \nTHE COMMON PARALLEL BUS. ACTIVE LOW\n\nADORESSED DEVICE TO SEND DATA TO THE SPI \nON COMMON PARALLEL BUS. ACTIVE LOW\n\nADDRESSED DEVICE REOUESTED TO GATE A \nSTATUS REPLY ON THE COMMON PARALLEL BUS, \nACTIVE LOW\n\nSERIAL BUFFER BUS CARRY PULSE INDICATING \nTHAT THE ON-LINE (ACTIVE) DATA BUFFER \nHAS BEEN EITHER FILLED OR EMPTIED, \nACTIVE LOW\n\nSERIAL BUFFER 8US 16 BIT CARRY PULSE \nINDICATING THAT A 16-BIT WORD HAS BEEN \nTRANSFERRED BETWEEN THE ON-LINE DATA \nBUFFER AND THE DEVICE CONNECTED TO IT, \nACTIVE LOW\n\nSERIAILY SHIFT THE INTERMEDIATE TRANSFER \nREGISTER ON THE TRAILING EDGE OF A \nLOW-GOING PULSE\n\nENABLE DATA TRANSFER - GATE THE \nINTERMEDIATE TRANSFER REGISTER ONTO THE \nCOMMON PARALLEL BUS. ACTIVE LOW\n\nREQUEST TO THE SPI TO TRANSMIT AN ERROR \nSTART CODE IN THE CURRENT STATUS REPLY \nTO THE CC, LOW ACTIVE\n\nCOMMON PARALLEL BUS PARITY BIT (ODD) \nOVER THE * LOW-ORDER DA A BITS. LOW = \nLOGICAL ONE\n\n16 BIT LARRY Pl'LSE INDICATING THAT A \n16-BIT WORD HAS BEEN TRANSFERRED BETWEEN \nTHE OFF-LINE BUFFER AND THE INTERMEDIATE \nTRANSFER REGISTER. ACTIVE HIGH\n\nSERIAL DATA INPUT TO THE OFF-LINE BUFFER \nFROM THE INTERhEDlATE TRANS-ER REGISTER, \nLOW = LOGICAL 1\n\nLOAD THE INTERMEDIATE TRANSFER REGISTER \nfRDM THE COMMON PARALLEL Btfr ON A HIGH \nTO LOW TRANSITION\n\nLOAD OPERATION PARITY CHECK DELAY. \nACTIVE LO\u00ab. GUARANTEES THAT .. PARALLEL \nPARITY ERROR IS REPORTED IN THE \nFOLLOWING STATES REPLY\n\nENABLES THE LOADING OF THE PARALLEL \nPARITY BITS INTO THE INTERMEDIATE \nTRANSFER REGISTER FROM THE COMMON \nPARALLEL BUS\n\nCLOCK THE OFF-LINE SUFFER AND ITS \nASSOCIATED COUNTER ON A HIGH TO LOW \nTRANSITION\n\nCLOCK THE ON-LINE BUFFER AND ITS \nASSOCIATED COUNTER ON A LOW TO HIGH \nTRANSITION\n\nPARALLEL PARITY ERROR IN THE HIGH ORDER \nBITS DUPING !TR REGISTRATION. ACTIVE LOW\n\nPARALLEL PARITY ERROR IN THE LOW ORDER \nBITS DURING 1TR REGISTRATION. ACTIVE LOW\n\nADDRESS DEVICE TO RECEIVE THE OATA ON \nTHE CGMMON PARALLEL BUS AND INTERPRET \nTHEM AS A COMMAND, LOW ACTIVE\n\nADDRESSED DEVICE TO RECEIVE THE DATA ON \nTHE COMMON PARALLEL BUS. .CT1VE LOW\n\nSERIAL DATA INPUT TO INTERMEDIATE \nTRANSFER REGISTER FROM THE OFF-LINE \nBUFFER. LOU \u00ab LOGICAL 1\n\nADDRESSED DEVICE TO SEND DATA TO THE SPI \nON THE COMMOX PARALLEL BUS. ACTIVE LOW\n\nADDRESSED DEVICE REQUESTED TO GATE A \nSTATUS REPLY ON THE COMMON PARALLEL BUS. \nACTIVE LOW\n\nA SYNCHRONIZING SIGNAL TO THE SPI WHICH \nINDICATES THAT A DEVICE HAS SENSED A \nCOMMAND ON THE PARALLEL BUS. ACTIVE LOW\n\nTOGGLE THE FIRST ELEMENT OF THE OFF-LINE \nBUFFER SEQUENCER CHAIN ON A HIGH TO LOW \nTRANSITION\n\nINHIBIT PARALLEL PARITY CHECKING WHILE \nUNLOADING THE INTERMEDIATE TRANSFER \nREGISTER ONTO THE COMMON PARALLEL BUS. \nACTIVE LOW\n\nA REOUEST TO THE SPI TO WAIT UNTIL THE \nDEVICE HAS HAD TIME TO ACT UPON A \nCOMMAND BEFORE REMOVING THAT COMMAND \nFROM THE COMMON PARALLEL BUS, ACTIVE LOW\n\nTIMING DIAGRAMS F0R C0MM0I, PARALLEL BUS \nCISHMUNICA T I0N AND 0FF - LINE SEQUENCING\n\nSTART OF OFF-LINE SEQUENCE \n(DASHED LINES INDICATE SEQUENCER \nACTION DURING LOAD OPERATION WITH PARALLEL PARITY ERROR)\n\nON-LINE SHIFTING \n(DASHED LINES INDICATE ACTION LEAOING TO OR RESULTING FROM OVERFLOW)\n\nTHE BUFFER UNIT (BUF) IS A TEMPORARY SERIAL MEMORY DEVICE \nFOR BUFFERING DATA TRANSFERS BEIVJEEN THE 3ACC AND SERIAL \nDEVICES ON THE TUC COMMON PARALLEL BUS. MX IS ESSENTIALLY \nBUILT AROUND TWO 1024-BIT SHIFT REGISTERS (B'JFO,BUF1 ) , \nWHICH ARE MAINTAINED IN OPPOSITE BUT SWITCHABLE ON-LINE/ \nOFF-LI N\u00a3~STATES. AN ON-LINE BUFFER IS THST SHIFT REGISTER \nWHICH IS CONNECTED TO THE SERIAL 3\"FFER BUS PRESENTLY SERV- \nINfi THE CARTRIDGF TAPE TRANSPORT CONTROLLEP 'CTTC). AN OFF- \nLrtJE BUFFER IS THAT SHIFT REGISTER WHICH IS CONNECTED IN- \nDIRECTLY (THROUGH A SERIAL/PARALLEL CON\u00abERTER) TO THE COM- \nMON PARALLEL BUS AND MAY BE SERVICED BY THE 5AC,:. THE DUAL\n\nTHE 0FF - LINE BUFFER AT ITS CONVENIENCE WHILE A CTTC/ \nON-LINE BUFFER SERIAL CONVERSATION PtfOGPESSES AT THE CTTC \nDATA RATE THE ON-LI NE /OFF-LINE BUFFEK STATUS SWITCHES AT \nTHE COMPLETION OF EACH 1024-BIT DATA TRANSFER ON THE SERIAL \nBUS TRANSPARENTLY PROVIDING THE CTTC WITH A FRESH BUFFER \nWHILE FLAGGING AN OFF-LINE BUFFER SERVICE REQUEST. BUF IN- \nCLUDES FEATURES FOR TRANSFERRING VARIABLE LENGTh DATA BLOCKS \nIN 16-BIT WORD INCREMENTS.\n\nTHE B'JF F'lNCTIONA! PARTITION DIRECTLY CORRESPONDS TO THE \nPHYSICAL PARTITION OF THE CIRCUIT ONTO FOUR BOARDS, AS \nSHOWN IN COMPOSITE DIAGRAM 1. BUF-A (J* 1 \"} I NTERFACES Wl l\"H \nTHE COMMON PARALLEL BUS AND PERFORMS B'JF ADCPfSS AND COM- \nMAND DECODING. B'IF-C (JM2) PROVIDES THE SERIAL /PARALLEL \nCONVERSION CAPABILITY NECESSARr TO TRANSFER DATA BETWEEN THE \nCOMMON PARALLEL BUS AND THE 1024-RIT SHIFT REGISTERS LO- \nCATED ON BUF-C (JK13). BUF-C ALSO INCLUDES PARITY CHECKING \nAND GENERATION CIRCUITS. BUF-B (JK1 1 ) PERFORMS THE SE- \nQUENCING INVOLVED IN STEERING DATA THROUGH THE BUFFER.\n\nBUF RESlOES AS A PERIPHERAL DEVICE ON 'HE COW.1N KARALLEl \nBUS TOR WHICH THE COMMUNICATIONS PROTOCOL HAS Bt\"-'N DESCRlUfcO \nIN FS 1 AND FS 2. BUF IS ASSIGNED DEVICE CODE 15. THF AP- \nPEARANCE OF THIS DEVICE CODE ON THE SIX LOW-ORDER BUS IK- \nFORMATION LEADS, ACCOMPANYING AN ACTIVE X BUS CONTROL LEAD. \nAODRESSES LIF AND ENABlFS T!\u20ac DECODING CI COMMAND INFORK- \nTION ON TKF TEN HIGH-ORDER BUS INFORMATION LEADS. ONCE \nSELECTED Bl'F LATCHES INTO AN ADDRESSED S'\"ATE, TURN INC 1 1- \nSELF ON fo FATHER PARRLEL b'JS COMMUNICATION. BUT IS DE- \nSELECTED UPON THE RECrPTION OF AN RC SIGNAL ACCOMPANIED BY \nOTHtR THAN A tj DEVICE CODE.\n\nCODED COMMAND INFORMATION IS BROUGHT INTO THE COMMAND DE- \nCODER ON THE THREE HIGitST ORDER BUS INFORMATION LEADS. \nTHE DECODER RESPONDS BY DRIVING CONTROL CIRCUITS WHICH Pl?- \n- FORM OR INITIATE THE ACTION SPECIFIED BY THE CO^E. THE \nFIGHT *-BIT CODES FOLLOW:\n\nCLEARS THE BR (BUFFER READY) FLAG, INDICATING \nTHt 3ACC HAS COMPLETED SERVICING THE OFF-LINE \nBUFFER. (THE BR IS RAISED TO REQUEST 5ACC SER- \nVICING OF THE OFF-LINE BUFFER.)\n\nINITIATES AN ON-LINE BUFFER FILL OPERATION. A \nFILL OPERATION CAUSES THE DATA QUEUE IN THE ON- \nLINE BUFFER TO BE STEPPED TO THAT BUFFER'S OUT- \nPUT IF NOT ALREADY THERE.\n\nTHE STATE REGISTER B1I KStGmtlCSS FOLLOW. THE FIRST \nTHREE A^E \"JSFI TC S\u00a3l\u00a3\" CUFFcREM OFF-LINE BUFFER OPERA-\n\nSTIFF . A ST\u00bbJFF OPERATION CAUSES TtC DATA QUEUE \nSN THt 5F-L'\u00ab B.FFER TO BE STEPPED TO THAT \nBUTTER'S OUTPUT IF !\u00a3,T ALREADY THERE.\n\nTHE IOTEWCMITE TWUBTES REGISTER (ITR) PERFORMS \nSERIAL '='-RALLEL CONFUSION IN AL t DATA TRANSFERS BETWEEN \nTHt OFF-LINE KFFE8 MB TIC JACC. IN A LOAD OPERATION, \nTHE TT R ACCEPTS VA'S. A?\u00abAR!\u00bbC ON THE 16-PARALLEL BUS \nINFORMATION LEADS. BE BUS PARITY BITS ARC LOADED INTO \nAN AUXILIA*' 2-B' I AGISTER. THE PARALLEL LOADING OF\n\nTKX*LEi\" RTM (NOTE TVT BUF WJST BE ADDRESSED). PARITY \nrPE\u00a3S WJiGlfiG Ot. THE ITR CCTRTS patch THE PARALLEL DATA \nPARITY Wt\u00bb Tt\u00a3 S\u00a33ISTETS) PARITY STS TO CHECK FCR BUS \nPARITY BCORS. BHCMKS THE OCCURENCE OF A PARALLEL \nPARITY ERSCw THE ITR ENTERS A SERIAL SHIFT MODE AND THE \n16 O-'TA 3!T3'fc'>; 3UXX2 ll.TO T\u20ac OFF-LINE BUFFER UNDER \nCONTROL\" OF THE S\u00a32.3\u00a3ER ON SLF-e. A COUNTER \nASSOCIAT'D WITH T>\u20ac C-F-Li'.E BUFFER SIGNALS THE SEOUENCER \nWHEN \"!'. = ITS HME F3* T5AKSFFRRE0. A DETECTED PARALLEL \nPARITY ERROR WILL ISSHSSiT SERIAL SHIFTING OF THE ITR.\n\nTH r err- i;.e buffer hjtc t>\u00a3 its (khich is in a serial \nshift \u2022*:-:?;. the M&m trees se&erate odd parity over\n\nTh\" utC-~*'iD LOW ^TtS OF WE IATA WORD. \"HE DATA _ ANO \nPARITY *-\u00a3 GATED KTt I\u00bb\u20ac RHtALLEi BL5 BY BUS CONTROL \nSIGNAL 5Kv\n\nsEOuENCiv;, cal-sfs rE=-:ES tc E\u00a3 shifted into the off- \nline Et'TEF TO ?*? OCT A BUFFER CONTAINING LESS THAN \n1C24 DATA SITS.\n\nTHE ON-LINE BL'FFEs; fW?i Vi OAT* PORT OF A SERIAL BUF- \nFT! BL- WICM PRESEVTLY iERVES OBLT THE CTTC. THE SERI- \nAL BUFFER BUS l\u00abXUDES SEVERAL CONTROL AND STATUS LEADS \nVWICH A 'CW THF CTTC TC tOiVC. THE MOVEMENT OF DATA INTO \nAMD OUT CF THE 0V-U3L\" a=-FES W1THCLT INTERFERING \nBUF PF'-LINE CPE?JtT10 , i <L OFt EVE*. PECLHRING BUF TO BE AD- \nDRESSED IN PARTIS-'LAF. THE CTTC-DRIVBJ BUFFER SHIFT \nPULSE CCKTRO\" LFJC (5SW1 DIRECTLY CLOCKS THE ON-LINE \nBUFFER. \u00bb DATA BIT APPEARING C*i THE SERIAL DATA INPUT \nLEAD 1\u00ab CLOCKED 8' \"\" TME tWFFER BY BSHPC WHILE A DATA \nHIT AT Tr\u00a3 BLIFFT T\u00bb.:T IS PRESENTED ON THE SERIAL DATA \nOUTPUT \u201eU2. A CCL^TS ASSSCiRTEC WITH THE ON-LINE BUF- \nFER PRC-VIDFS A HASIH? PLLSE (EA16CR0) ON THE SERIAL BUS \nWHICH WO! GATES V* CCPRETIW. OF EACH 16-BIT ON-LINE \nDATA TRANSFER. A t-^4_F.iT RARKS? R'LSE (BACRO) IS ALSO \nAVAILABLE TO DEVICES C*. THE SERIAL BUFFER BUS.\n\nA FILL OPERATION &X S\u00a3 IMTIAJE3 BY THE CTTC (VIA \nSERIAL SuTEJr tfci iEJ\u00a3 tlllO) R) PAD CUT A BUFFER TO \nWHICH ir\u20ac CTTC MAS WMffTDHICT) LESS THAN 1024 BITS. \nFIL^ CAN AL& 5E ISIT1ATETJ SY A 5ACC COWAND FOR HAIN- \nTEKAN\" PLBPCSES. !>\u25a0: EITTCS CASE, THE OM.INE BUFFER IS \nADVANC\u2014 ?/ THE ft- S\u20ac3jE>iCER L9.TIL BACRO IS DETECTED. \nTHE C*i-LINE bLIFFES \u00abY 55 RECIRCULATEC OF gNES PADOED \nOUR INC A FILL 0?E\u00ab-Ti3i, 0EFB3IIIG ON THE STATE OF THE \nSERIAL CiTA 1MR-T i&C\n\nDATA AND CLOCK INFORMATION IS STEERED TO OR FROM THE \nPRCPER BUFFER BY MULTiaEXER SWITCHING DIRECTLY CONTROLLED \nBY THE STATE OF THE ACT F/F. BUFFER INITIALIZATION \nFORCES ACT INTO STATE SHOWN IN FIGURE 1 (SHT B3GB) WITH BUF0 OFF \nLINE AND BUF1 ON-LINE. COUNTER CNTO (ASSOCIATED WITH \nBUFO) IS CONFIGURED TO THE OFF-LINE CLOCK ALONG WITH \nBUFO WHILE CNT1 AND BUF1 ARE CONFIGURED TO THE ON-LINE \nCLOCK SOURCE. TOGGLING ACT RESULTS IN A REVERSAL OF \nTHE ON-LINE/OFF-LINE STATUS OF BUFO.BUFI, AND THEIR \nASSOCIATED COUNTERS. SUBSEQUENT CHANGES IN THE STATE OF \nTHE ACT F/F YIELD SIMILAR BUrFER REVERSALS. THE BR aAG \nIS SET BY THE ON-LUE BUFFER COUNTER BACRO SIGNAL AND \nINDICATES THAT THE ON-LINE BUFFER HAS BEEN CLOCKED 1024 \nTIMES AND NOW REQUIRES SERVICING BY THE 5ACC. RAISING \nTHE BR FLAG NORMALL\" ASSERTS THE PARALLEL BUS INTERRUPT \nLEAD (INTPO) UNLESS STATE REGISTER BIT INTOFF IS SET. \nRAISING THE BR FUG ALSO TOGGLES THE ACT F/F, SWITCHING \n. THE BUFFERS. THE 3ACC MAY THEN SERVICE THE NEW OFF-LINE \nBUFFER AND RETIRE THE BR FLAG UPON COMPLETION. IF THE \n3ACC DOES NOT MANAGE TO RETIRE THE BR FLAG BEFORE CURRENT \nON-LINE BUFFER OPERATIONS ARE COMPLETE (INDICATED BY \nBACRO), A BUFFER SWITCH WILL MOT OCCUR, THE BBFL F/F \n(BUFFER OVERFLOW) IS SET, AND ALL SUBSEQUENT CLOCK SIG- \nNALS TO THE ON-LINE BJFFER ARE CUT OFF. THE STRANDED \nON-LINE BUFFER MAY BE BROUGHT OFF-LINE BY A 3ACC COMMAND \nTO SWITCH BUFFERS, TOGGLING ACT THROUGH THE COMMAND DE- \nCODER. DATA IN THIS BUFFER MAY THEN BE RETRIEVED BY SUC- \nCESSIVE UNLOAD OPERATIONS.\n\nBUF STATUS, ACCOMPANIED BY A 13 DEVICE CODE, IS REPORTED \nON THE PARALLEL BUS IN RESPONSE TO A SST CONTROL SIGNAL. \nTHE STATUS BIT DESIGNATIONS FOLLOW:\n\nTHIS EXAMPLE. A SERIES OF LOAD OPERATIONS TRIGGERED BY \nRD cgw-:a;;ds BEGINS, EACH RD MOVING A 1 6-BIT DATA WORD \nTFROUGH THE ITR AND INTO BUFO. WHEN THE 64-WORD CAPACITY \nOF BUFO IS REACHED (ASSUMING A DATA BLOCK GREATER THAN \n64 WORDS FOR SIMPLICITY), BUFO IS SWITCHED ON-LINE AND \nTHE CTTC IS ALERTED THAT IT IS TO RECEIVE A DATA BLOCK- \nLOAD OPERATIONS CONTINUE ON BUF1. WHEN &UF1 HAS BEEN \nLOADED TC ITS 64-WORD CAPACITY, OFF-LINE BUFFER OPERA- \nTlCiS ARE SUSPENDED AND BUF STATUS REQUESTS MUST BE \nPERIODICALLY HADE TO CHECK THE STATE OF THE BR FLAG \nSINCE THE INTERRUPT FACILITY IS INHIBITED.\n\nTHE CTTC BEGINS TO DEPLETE THE ON-LINF BUFFER. A STATUS \nREPLY INDICATING THAT THE BR RAG HAS BEEN RAISED IMPLIES \nA BUFFER SWITCH HAS OCCURRED AND OFF-LINE LOADING OPERA- \nTIONS CAN CONTINUE. AS BUF COMPLETES THE LOADING OF AN \nOFF-LINE BUFFER, THE BR RAG IS CLEARED. IT IS IMPORTANT \nTHAT THE OFF-LINE LOADING OF THE SECOND BUFFER BE CON- \nRETE3 BEFORE THE FIRST BR FLAG IS RAISED; OTHERWISE. AN \nOVERROU SITUATION OCCURS (A PREMATURE BUFFER SWITCH) \nBUT MR DOESN'T GET SET.\n\nFOR DATA BLOCKS WHICH ARE NOT MULTIRES 01 64 WORDS, THE \nFINAL BUFFER IS PARTIALLY LOADED, THE COUNTER ASSOCIATED \nWITH THAT OFF-LINE BUFFER IS CLEARED TO MARK THE Nll'^ER \nOF DATA WORDS IN THE BUFFER, A STUFF OPERATION IS i \nIT1ATED. AiiD THE BR FLAG IS CLEARED AT THE COMPLETIu: OF \nTHE STUFF.\n\nAN EXA.MRE BUF COMMAND SEQUENCE FOR MOVING A BLOCK OF \nDATA FROM THE CTTC TO THE 3ACC IS FLOWCHARTER IN \nFIGURE 5. BUF SHOULD BE INITIALIZED AT THE START. AN \nACKNOWLEDGED BUF INTERRUPT INDICATES THAT THE BR HAS \nBEEN RAISED AND BUF1 HAS JUST BEEN SWITCHED OFF-LINE \nFOR SERVICING. THE UNLOAD BIT IS SET IN THE STATE \nREGISTER WHICH ADVANCES THE FIRST DATA WORD INTO THE \nITR AN SO COMMAND WILL GATE THE ITR ONTO THE PARALlEL \nBUS\" AliO ADVANCE THE NEXT WORD INTO THE ITR. 64 SDCOM- \nHANOS ARE REQUIRED TO MOVE A FULL BUFFER OF DATA TO THE \n5AC- AFTER WHICH THE OFF-LINE COUNTER IS CLEARED. THE \nBR RAG IS RETIRED AND THE 3ACC WAITS FOR THE NEXT IN- \nTFPCTiPT NOTE THAT THE UNLOAD STATE MUST BE SET EACH \nTIME\" A BUFFER IS TO BE DUMPED TO THE 3ACC IN ORDER TO \nGET THE FIRST WORD OF EACH BUFFER INTO THE ITR.\n\nrur r-n- \"\"ST REQUEST A FILL OPERATION ON THE LAST ON- \ni'l'NE'ajFFER IF THt BLOCK LENGTH IS NOT A MULTIPLE OF 64 \nWORDS.\n\nCOMPOSITE DIAGRAM 8 PR6VIDES A FUNCTIONAL REFERENCE, \nORAUIH& FOR TIMING DIAGRA-SS THAT FOLLOW, AND AS SUCH \nIS NOT MEANT TO OFFER AN ACCURATE GATE LEVEL DESCRIP- \nTION. THE BUF CPS DRAWINGS SHOULD BE REFERENCED FOR \nGATE LEVEL RESOLUTION. A TIMING DIAGRAM MNEMONI- ENDING \nIN OR 1 DIRECTLY CORRESPONDS TO A \"WS'^AL CIRCUIT \nLEAD WHOSE LOGIC POLARITY WILL BE CORRECTLY INDICATED ON \nTHE DIAGRAM-\n\nBITS 5-0 CONTAIN THE 13 DEVICE CODE. BUF BUS TIMING \nLOGIC USES PARALLEL BUS LEAD GPO TO REQUEST PARITY \nGENERATION OVER THE STATUS WORD BY THE BUS TERMINATOR.\n\nBUF SHOULD BE INITIALIZED INTO A KNOWN STATE VIA A TDC \nSYSTEM INITIALIZATION SIGNAL ON PARALLEL BUS LEAD INITO \nOR BY A BUF RESET COMMANO. THE LOAD BIT MUST THEN BE \nSET IN THE STATE REGISTER TO ENABLE THE REGISTRATION OF \nPARALLEL PARITY BITS. THE INTeFF BIT IS ALSO SET IK\n\n, glF RECEIVES A COMMAND FRO-; THE 3ACC VIA THE RC SEQUENCE \n&OM IN FIGURE 4 (SHT B5GC). THE 3UF DEVICE CODE AND \nCOMMAND INFORMATION APPEARS ON TH-: BUS BEFORE THE LEAD- \nING EOCE OF RCO, ALLOWING ATDRESS WO COMMAND DECODER \nSET UP TIME. THE LEADING CDGE OF RCO WILL LATCH THE \nUOR F/F AUO ENAffE Or;' CF THE EIG'-T CurfMAND DECODER \nOUTPUTS, BUF RETURiiS SYNCO AND ErO ON THE BUS AT THIS \nTIME TO VERIFY THAT A CONTROL SIGNAL HAS BEEN RECEIVED \nAND TO CHECK FOR PROPER OPERATION OF THE ERO LEAD, RE- \nSPECTIVELY. THE TRAIL IMG EDGE OF RCO ALLOWS SYIJCO TO \nDROP OFF. EF.0 WILL ALSO DROP OFF AT THIS TIME UNLESS \nyuc PTTR F/F (PARITY ERROR) HAS BE r N SET IN A PREVIOUS \nRUF OPERATION. IN THIS CASE, ERO WILL REMAIN ACTIVE \nUNTIL BUF IS RESET OR DESELECTED.\n\nTHE BUS INFORMATION WILL REMAIN STEADY FOR A PERIOD \nBEYOND THE TRAILING EDGE OF RCO. THIS EDGE IS INDIRECT- \nLY USED TO CLOCK LEADS INF09-12 INTO THE STATE REGISTER \nIF SUCH AN OPERATION HAS BEEN SPECIFIED BY THE COMMAND \nOFCODER. IF THE UNLOAD OR STUFF BIT IS TO BE SET IN THE \nSTATE REGISTER, THE COMMAND DECODER LOAD STATE REGISTER \nOUTPUT PRODUCES WISE TG. SEO WHOSE TRAILING EDGE IS USED \nTC TOGGLE THE FIRST ELEHB,T \\U AN OFF-LINE SEQUENCE TIM- \nING CHAIN. A BUSY STATE IS ENTERED FOR THE DURATION CF \nTHE OFF-LINE SEQUENCE INITIATED.\n\nBUF RECEIVES DATA FROM THE 3ACC VIA RD SEQUENCES, BUF \nMIST BE ADDRESSED WITH THE STATE REGISTER LOAD BIT SET \nPRIOR TO A VALID DATA TRANSFER FRO!-' 3ACC TO BUF. IF BUF \nIS NOT ADDRESSED, CONTROL SIGNALS RO, SO, OR SST ARE IG- \nNORED. IF BUF IS ADDRESSED BUT NOT IN THE LOAD STAU, \nA 1-BIT BUFFERING REGISTER AT THE IfR SERIAL OUTPUT \n(LDO) IS HELD CLEAR PREVENTING DATA SHIFTED OUT OF THE \nITR FROM REACHING THE OFF-LINE BUFFER. IN ADDITION TO \nTHE LOCK UP OF LDO, THE AJXILIARY 2-BIT PARITY REGISTER \nISALSOHLLD CLEAR, RESULTING IN A PARITY ERROR AS SOON \nAS AN EVEN PARITY BYTE IS REGISTERED IN THE ITR.\n\nTHE LEADING EDGE OF RED CAUSES SYNCO TO BE RETURNED AND \nTG.SEQ TO BE ASSERTED. THE TRAILING EDGE OF RDO REG- \nISTERS DATA ON THE PARALLEL BUS INTO THE ITR AMD ITS \nASSOCIATED PARITY REGISTER. SYNCO AND TG. SEQ DROP OFF \nAND THE BUSY STA^ IS ENTERED FOR THE DURATION OF THE \nLOAD SEQUENCE.\n\nTHE 3ACC RECEIVES DATA FROM BUF VIA SO SEQUENCES, BUF \nMUST BE ADDRESSED WITH THE STATE REGISTER UNLOAD BIT SET \nPRIOR TO A VnLID DATA TRANSFER. SETTING THE UNLOAO BIT \nPRIMES THE ITR FOR THE FIRST SO.\n\nTHE LEAUlNG EDGE OF THE SDC CAUSES BUF TO ASSERT SYNCC \nAND TG. SEQ. THE ITR IS GATED ONTO THF PARALLFL BUS FOR \nTHE DURATION OF SOO. THE TRAILING EDGE OT SDO REMOVES \nTHF ITR FROM THE BUS AND NEGATES SYNCO. THE BUSY STATE \nIS ENTERED FOR the DURATION OF THE UNLOAD SEQUENCE.\n\nTHE 5ACC REQUESTS BUF STATUS VIA THE SST SEQUENCE ON \nFIGURE 5. AN AWRESSED BUF RESPONDS TO THE ASSERTION \nOF SSTO BY ACTIVATING SYNCC, WAITO, AND GPO AND BY GATIKG \nSTATUS INFORMATION ONTO THE BUS. LDPCKD DELAYS FURTHER \nACTION IN A SST WHICH FOLLOWS A RO UNTIL THE RESULTS OF \nTHE LOAD OPERATION PARITY CHECK CAN BE REPORTED. LDPCKD \nIS ONLY ACTIVE IN LOAD OPERATIONS. WHEN RELEASED FRCf. \nTHE DELAY, A 2-STAGE TIMING CHAIN IN THE BUF SST CIRCUIT \nLOOKS FOR THE NEXT NEGATIVE TRANSITION ON PARALLEL BUS \nCLOCK LEAD CLK AND RELEASES GPO ON THE FOLLOWING POS- \nITIVE TRANSITION INSURING THAT GPO IS OF SUFFICIENT \nWIDTH TO OPERATE THE PARITY GENERATION CIRCUITS IN THE \nBUS TERMINATOR (BT). BT RESPONDS WITH GPRO, INSTRUCTING \nBUF TC REMOVE ITS STATUS AND ITS WAITO FROM THE BUS. \nBUF LATER REMOVES SYNCO AFTER SSTO HAS BEEN NEGATED.\n\nPROGRESS TO CONTINUE. AN EXCEPTION TO THE ABOVE IS AN \n-- | jr ,p COMMAND IN WHICH A CQH1AH0 Dl CODER OUTPUT IS USED \nTC INHIBIT WAITO. RC MOT CAM BE USED TC ADDRESS BUF \nWITHOUT HANGING UP THE BUS IF BUT IS BUSY. THE ABSENCE \nC-F THE BUSY CONDITION IS REFLECTED BY THE STATE OF THE \nTOY STATUS BIT.\n\nBUF REPORTS AN .NTERNUPT TO THE- 3ACC WHEN THE BR 'LAG IS \nSET AND Tr'f STATE REGISTER BIT INTC-F IS NOT SET. AS- \nSERTION OF BUS CONTROL LEAD ACKIC GATES BU C 'S INTERRUPT \nIDENTIFICATION RESPONSE ONTO INF010. THIS LEAD IS RE- \nLEASED UPON RECEPTION OF GPRO.\n\nCOMPOSITE DIAGRAM B SHOWS BUF SEQUENCING TO BE DIVIDED \nBETWEEN TWO DISTINCT SEQUENCERS. THE OFF-LINE SEQUENCER \nEaSEQ HANDLE! LOAD, UNLOAD, AND STUFF OPERATIONS. 0FLSE3 \nIS BASICALLY A FOUR ELEMENT TIMING CHAIN (BFLSEQA-D) \n05110 BY A FREE-RUNNING BUF CLOCK (BCLK). BCLK PROVIDES \nA SQUARE UAVE PULSE TRAIN WITh A NOMINAL 600 MS PERIOD.\n\nBPSEG ACTION BEGINS WITH A Y6.SFQ PULSE. TG. SEO IS \nDERIVED FROM SO OR RO PULSES OR RC PULSES WHICH SET THE \nSTAT r AGISTER STUFF OR UNLOAC BITS. THF iHAILING EDGE \nOF TG,'.n TOGGLES BFLSEOA WHICH THEt; FNABLE3 THl CONSECU- \nTIVE ACTUATION CF TIMING CHAIN ELEMENTS BFLSEQ6-D ON \nSUCCESSIVE NFGAT'VF TRANSITIONS OF BCLK, FIGURE 6 (SHT B5GC). \nA BUSY CONDITIO* EXISTS WHENEVER 0R.SEQA OR C ARE ACTIVATED. \nTHF ACTIVATION OF BFLSEQB FORCES THF ITS INTO TIC SERIAL \nSHIFTING MODE. BFLSEQO ENABLES AN OFF-LINE CLOCK SIGNAL \n(CLKITRO 0FLCIK1) DERIVED FRCM BCLK. A NEGATIVE TRAN- \nSITION OF THE OFF-LINE CLOCK SIGNAL CLOCKS THE STATE OF \nTHE 'TR SERIAL OUTPUT INTO LDD. THE FCLIOWING POSITIVE \nTRANSITION HCVES THE OFF-LINE BUFFER OUTPUT INTO THE ITR, \nCLOCKS THE STATE OF LCD INTO THE BUFFER iNPUT, AND IN- \nCREMENTS THE BUFFER COUNTER.\n\nTHF SEQUENCE OF FIGURE 6 IS PERFORMED AT THE START OF ALL \nOFF-LINE OPERATIONS WITH THE EXCEPTION OF THE LOAD OPERA- \nTION WHICH REQUIRES THE ADDITIONAL PARITY CHECKING FUNC- \nTION LDPCKD (DFLAYS A SST OPERATION) IS ACTIVE FRC.-I \nT iic POSITIVE 0FLSEQA TRANSITION TO POSITIVE BFLSEQC TRAN- \nSITION AT WHICH TIME PARALLEL PARITY IS CHECKED. A PAR- \nITY EUROS AT THIS TIME SETS FTER F/F CAUSING THE TIMING \nCHAIN ELEMENTS TO SEQUENTIALLY DROP OFF AND INHIBITING \nANY OFF-LINE CLOCK PULSES. BFLSEO IS LOCKED OFF UNTIL \nTHE PTER F/F IS CLEARED VIA BUT RESET OR WIT. THF. AB- \nSENCE OF A PARITY ERROR AT THE BFLSEQC TRANSITION ALLOWS \neaSEQ TO CONTINUE THE LOAD OPERATION.\n\n3ELSEQD ENABLES THESHIFTING PHASE OF AN OTF-LINE OPERA- \nTION. FIGURE 7 (SHT B3GD) SHOWS THE SHIFTING ASSOCIATED \nWITH A LOAD OR UNLOAD OPERATION. SIXTEEN DATA BITS ARE \nSERIALLY TRANSFERRED BETWEEN THE ITR ANO THE OFF-LINE \nBUFFER BY THE OFF-LINE CLOCK. THE OFF-LINE BUFFER COUN- \nTER GENERATES A 16-BIT CARRY PULSE, H6BC, WHOSE TRAILING \nEDGE TERMINATES SHIFTING AND INCREMENTS STUFF COUNTER \nSCNT' SCNT TALLIES THE NUMBER CF IT6BC PULSES (16-BIT \nTRANFERS) DETECTED SINCE THE LOAD STATE WAS LAST SET. \nBaSEQ DROPS OFF AS SHOWN AT THE COMPLETION OF SHIFTING. \nSCNT RECYCLES TC ZERO ON THE RECEPTION OF THE 64TH H6BC \nPULSE.\n\nSHINING IN A ST FF OPERATION (FIGURE 8 - SHT 63CE) PR0CEEDS CON- \nTINUOUSLY OVER 1 TIT WORD BOUNDARIES UNTIL DATA IN THE \nOFF-LINE BUFFER I'.hS BEEN PIGHT-ACJUSTED TO THE BUFFER \nOUTPUT, AS INDICATED BY 5CI!T. LDD IS HELD CLEAR DURING \nSTUFF SHIFTING, PADDING THE OFF-LINE BUFFER WITH ALL \nZEROES.\n\nFOLLOWING IS A DESCRIPTION OF THE LEADS WHICH COMPRISE \nTHE SERIAL SUFFER BUS. THESE LEADS ARE ALL TERMINATED \nAT THE BT.\n\nRSHPO IS THE PRIMARY CLOCK SOURCE FOR THE ON-LINE \nBUFFER. CLOCK PULSES SUPPLIED BY THE CTTC ON THIS \nLEAD SHUT DATA INTO AND OUT OF THE ON-LINE BUFFER.\n\nDATA SHIFTED OUT OF THE ON-LINE BUFFER APPEARS ON \nSOCTO FOLLOWING A NEGATIVE TRANSITION ON BSHPC, \nFIGURE 9 (SHT B5GF). DATA IS NOT VALID PRIOR TO THE \nLEADING EDGE OF THE FIRST CLOCK PULSE.\n\nDATA IS SHIFTED INTO THE ON-LINE BUFFER FROM THIS \nLFAO ON THE TRAILING EDGE OF BSHPO. INPUT DATA \nMUST BE STEADY FOR A PERIOD BRACKETING THIS\n\nBA16CR0 IS A MARKER PULSE GENERATED BY THE ON-LINE \nBUFFER COUNTER WHICH COINCIDES WITH EVERY 16TH \nON-LINE CLOCK PULSE.\n\nBACRC IS A HARKER PULSE GENERATED BY THE ON-LINE \nBUFFER COUNTER INDICATING THE ON-LINE BUFFER HAS \nBEE!. CLOCKED 1024 TIMES.\n\nBR FLAG CAN BE RETIRED. AN OVERFLOW CONDITION \nLOCKS OUT ON-LINE CLOCK SIGNALS.\n\nTHE CTTC ACTIVATES FILLO TO INITIATE A FILL OPERA- \nTION, RIGHT ADJUSTING DATA IN THE ON-LINE SUFFER.\n\nBCLK IS A FREE-RUNNING CLOCK WHICH DRIVES 9UF \nSEQUENCERS AND WHICH IS BROUGHT OUT ON THE SERIAL \nBUFFER BUS FOR OPTIONAL USE BY DEVICES ON WIS \nBUS, BCLK IS A SQUARE WAVE PULSE TRAIN WITH A \nNOMINAL 600 NS PERIOD.\n\nCTTC SHIFTING OF THE ON-LINE BUFFER IS SHOW; IN FIGURE \n9 (SHT B3GF). THE TRAILING EDGE OF BACRO FIRES IW \nOt'-LINE BUFFER COMPLETE MOHOPULSER 0NLCMP, HHICH RAISES \nTHE SR F..AG AND SWITCHES THE BUFFERS WITHOUT INTERFERING \nCTTC/BUF DATA TRANSFER. NOTE THAT IF THE BR PJ.Z IS \nSTILL RAISED AT THE TRAILING EDGE OF 0NLCMP, AN OVER- \nFLY conditio;: exists, the buffers do not switch and on- \nline BUFFER CLOCKING IS INHIBITED.\n\nTHE ON-LINE SEQUENCER BNLSEQ HANDLES CTTC OR 5ACC RE- \nQUESTED FILL OPERATIONS. BNLSEQ IS A THREE ELEMENT \nTIMING CHAIN (BNLSEQA-C) DRIVEN BY BCLK. ACTION SEGINS \nWITH CTTC ASSERTION OF FILLO OR THE ACTIVATION OF THE \nCOMMAND DECODER FILL OUTPUT, EITHER OF WHICH SET BNLSEQA, \nFIGURE 10 (SHT B3GG). SUBSEQUENT NEGATIVE TRANSITIONS \nOF BCLK SET ELEMENTS BNLSEQB AND C CONSECUTIVELY 0NLSFQC \nENABLES ON-LINE BUFFER SHIFTING BY A SERIES OF CLOCK \nPULSES DERIVED FROM BCLK INFORMATION AT THE BuF^ER OUT- \nPUT WILL BE RECIRCULATED INTO THE BUFFER INa'TOURINC' \nFIU OPERATION (NORMALLY PADDING THE BUFFER WI\"H ZEROES/ \nUN'LESS THE CTTC H0LDS SDWO L0W. IN WHICH CASE \nTHE BUFFER IS PADDED WITH ALL 0NES. THE TRAILING \nEDGE BF BACRO INDICATES THE RIGHT ADJUSTMENT BE \nTHE BN - ' INE BUFFER IS CBMPLETE EY FIRING flNLCHF. \nWHICH IN TURN RAISES THE BR FLAG AND SWITCHES \nBUFFERS. BNLSEQ SUBSEQUENTLY DR0PS 0FF, TEkkINATI >- \nFURTHER SHIFTING.\n\nWTA10 f WTA00,WREtiAB0,TTSR0 f TTSf0,TTSEL0,TTINUC, \nTTREHCO^TT^aO.TTFFQ^RTAIO.RTAOO (12 LEADS)\n\nTHAT THE ON-LiNE (ACTIVE) DATA BUFFER \nHAS BEEN EITHER FILLED OR EMflED. \nACTIVE LOW\n\nCTTC'S CLEAR THE CYLIC REDUNDANCY CHECK \nCIRCUIT COMMAND BEING DECODED. LOW \nACTIVE\n\nERROR INDICATION OUTPUT OF THE CYCLIC \nREDUNDANCY CHECK CIRCUIT. ACTIVE LOW\n\nINHIBITS THE DATA DETECT STATUS SIGNAL \nFROM THE CTT UNTIL THE TAPE HAS \nACCELERATED TO FULL SPEED DURING \nBACKSPACE. READ-A-BLOCK. ASD \nREAD-AFTER-WR'TE OPERATIONS. ACTIVE LOW\n\nENABLE THE RIAC CIRCUITS OF THE CTTC TO \nACCEPT READ DATA FROM THE CTT. ACTIVE \nHIGH\n\nREQUEST TO THE SPI TO TRANSMIT AN ERROR \nSTART CODE IN THE CURRENT STATUS REPLY \nTO THE CC. LOW ACTIVE\n\nCTTC'S TAPE MOTION REOUEST PEG1STER IS \nSET TO FAST OR SLOW FORWARD. ACTIVE LOW\n\nREQUEST TO BT TO GENERATE PARITY OVER \nTHE STATUS REPLY CURRENTLY RESIDING ON \nTHE COMMON PARALLEL BUS\n\nTDC INITIALIZE COMMAND EXECUTE \n(LOGICALLY EQUIVALENT TO INITO). ACTIVE \nLOW\n\nSTATUS PULSE FROM CTT TO INDICATE THAT \nTHE CTT HAS SENSED THE LOAD POINT OR \nEARLY WARntNG HCi MARK ON THE TAPE. \nACTIVE LOW\n\nCTTC IS IN THE MAINTENANCE MODE OF \nOPERATION. (COMPLEMENT OF MAINTO) ACTIVE \nHIGH\n\nREOUEST TO THE CTT TO ENABLE THE \nOPERATORS REWIND AND UNLOAD MANUAL PUSH \nBUTTONS\n\nGATE THE CTTC'S PRIMARY STVTUS REPLY TO \nTK. PARALLEL INFORMATION BUS, HIGH \nACTIVE\n\nRECEIVE THE DATA ON THE COMMON PARALLEL \nBUS AND INTERPRET IT AS A COMMAND, LOW \nACTIVE\n\nREAD DATA ADJUST FOR PROPER 16-BIT WORD \nFRAMING (COMPLEMENT OF RDADJ1). ACTIVE \nLOW\n\nREAD DATA ADJUJT FOR PROPER 16-BIT WORD \nFRAMING (COMPLEMENT OF RDADJO), ACTIVE \nHIGH\n\nCLEAR THE CRC CHECK CIRCUIT FOR A READ \nOR READ-AFTER-WRITE OPERATION, ACTIVE \nLOW\n\nCTTC'S TAPE MOTION REOUEST REGISTER 'S \nSET TO FAST OR SLOW REVERSE. ACTIVE LOW\n\nREAD SHIFT PULSE USED TO GENERATE DATA \nSHIFT PULSES TO THE BUFFER. ACTIVE HIGH\n\nCTT IS PERFORMING A TAPE REWIND SEQUENCE \n(LOGICALLY EQUIVALENT TO RWDINGBO). LOW \nACTIVE\n\nCTT IS PERFORMING A TAPE REWIND SEQUENCE \n(LOGICALLY EQUIVALENT TO RWDINGAO). LOW \nACTIVE\n\nCTTC'S SHIFT CYCLIC REDUNDANCY CHECK \nCIRCUIT COMMAND BEING DECODED. LOW \nACTIVE\n\nA SYNCHRONIZING SIGNAL TO THE SPI WHICH \nINDICATES THAT A DEVICE HAS SENSED A \nCOMMAND ON THE PARALLEL BUS. ACTIVE LOW\n\nTAPE IS MOVING STATUS BIT =RCM CTT \n(LOGICALLY EQUIVALENT TO TIM30). ACTIVE \nLOW\n\nTAPE IS MOVING STATUS BIT FROM CTT \n(LOGICALLY EQUIVALENT TO TIMAO). ACTIVE \nLOW\n\nPHYSICAL END-OF-TAPE STATUS BIT FROM CTT \n(LOGICALLY EQUIVALENT TC TTEOTBO). \nACTIVE LOW\n\nCTTC'S 3-0UT-0F-6 DEVICE SELECTIONS CODE \nIS PRESENT ON THE PARALLEL INFORMATION \nBUS LINES INFOOO THROUGH INF350 DURING \nSCO. ACTIVE LOW\n\nSPl'S. RECEIVE COMMAND SIGNAL. BUFFERED \nFOR USE BY THE CTTC (LOGICALLY THE \nCOMPLEMENT OF RCO). ACTIVE HIGH\n\nA REOUEST TO THE SPI TO WAIT UNTIL TW \nDEVICE HAS HAD TIME TO ACT UPON A \nCOMMAND BEFORE REMOVING TH*' COMMAND \nFROM THE PARALLEL BUS. ACTIVE LOW\n\nWRITE DATA BEING PRESENTED TO THE CTTC'S \nPHASE ENCODING CIRCUITRY, HIGH = \nLOGICAL ONE\n\nOUTPUT CF THE CTTC'S WRITE CIRCUITRY \nENABLE FLIP-FLOP (LOGICALLY EQUIVALENT \nTO WRENABO). ACTIVE LOW\n\nWRITE CIRCUITRY ENABLE REQUEST LINE TO \nTHE CTT (LOGICALLY EQUIVALENT TO \nWENAAO). ACTIVE LOW\n\nWRITE DATA DELAY PULSE USED TO DELAY THE \nSTARTING OF THE WRITE DATA OPERATION \nUNTIL THE TAPE IS MOVING AT FULL SPEED \nAND THE PROPER LENGTH 1BG HAS BEEN \u2014 \nWRITTEN, ACTIVE LOW\n\nCTTC'S WR1TE-A-BL0CK COMMAND IS BEING \nDECODED AND A NONWRITE FROTECTED TRACK \nIS SELECTED. ACTIVE LOW\n\n1 DATA P0STAMBLE IS STRIPPED BY BEING LEFT IN THE DELAY \nREGISTER AT THE TERS1INATI8N 0F A READ ADJUST BPERATI0N.\n\n1. LEADS INDICATED SHALi BE PAIRED AND TWISTED WITH A \nSIMILARLY NAMED GROUND LEAD ittMHMG FP*4 THE \nCARTRIDGE TAPE TRANSPBRT T0 AN INTERMEDIATE \nCTF STAGING AREA.\n\nSTATUS PULSE FROM CTT TO INDICATE THAT \nTHE CTT HAS SENSED THE LOAD POINT OR\n\nREQUEST TO THE CTT TO ENABLE THE \nOPERATORS REWIND AND UNLOAD MANUAL PUSH\n\nSWITCH SI IS AN ALTERNATE ACTI3N SWITCH \nSUCH THAT IT IS DEPRESSEO SINCE T3 CL0SE \nALL C0NTACTS ANO DEPRESSED AGAIN T0 \n0PEN ALL C0NTACTS.\n\nNPA \u00bb5V CONVERTER POWER ALARM OUTPUT AND \nPOWER ALARM RESET INPUT. ACTIVE LOW\n\n20u'2400 bps, switched networt or private line \n20ea 1900 bps. 4 wire private line \n20sb 4800 bps, switched network \n209a 9600 bps. 4 wire private line\n\nTHAT THE ON-LINE (ACTIVE) DATA BUFFER \nHAS BEEN EITHER FILLED OR EMPTltD. \nACTIVE LOW\n\nrequest to the spi to transmit an error \nstart code in 'he current status reply \nto the cc. low active\n\nreouest to bt to generate parity over \nthe status reply currently residing on \nthe common parallel bus\n\nADDRESS DEVICE TO RECEIVE THE DATA ON \nT\u00abE COMMON PARALLEL BUS AND INTERPRET \nTHEM AS A COMMAND. LOW, ACTIVE\n\nAWRESSED DEVICE TO RECEIVE THE DATA ON \nTHE COMMON PARALLEL BUS. ACTIVE LOW\n\nADDRESSED DEVICE REQUESTED TO GATE A \nSTATUS REPLY ON THE COMMON PARALLEL BUS. \nACTIVE LOW\n\nA S'iNCHRONIZlNG SIGNAL TO THF. SPI WHICH \nINDICATES THAT A DEVICE HAS SENSED A \nCOMMAND ON THE PARALLEL BUS. ACTIVE LOW\n\nA REQUEST TO THE SPI TO WAIT UNTIL THE \nDEVICE HAS HAD TIME TO ACT UPON A \nCOMMAND BEFORE REMOVING THAT COMMAND \nFROM THE COMMON PARALLEL BUS. ACTIVE LOW\n\n1.1 THE SDSC CIRCUIT PACKS ARE A USER SPECIFIED OPTION TO THE \nTDC CIRCUIT ANO MAY BE PURCHASED SEPARATELY. TDC'S BUILT \nAFTER HID 1977 WILL HAVE BACKPLANE WIRING FACTORY INSTALLED. \nWIRING LISTS ARE AVAILABLE FOR USERS WISHING TO RETROFIT \nEXISTING TDCS. THE SDSC IS REQUIRED FOR 2B ESS OFFICES \nUSING THE EF2 GENERIC.\n\nBY A USER SUPPLIED DATA SET WHICH HAY BE LOCATED UP TO 50 \"IRE \nFEET AWAY FROM THE TDC. THE TDC BACKPLANE CONTAINS A \nKS-19088,L14 CONNECTOR TO HATE WITH ALL E1A BINARY \nSYNCHRONOUS HALF DUPLEX DATA SETS WITH A USER SUPPLIED CABLF.\n\n1.5 THE SDSC HAS THE CAPABILITY TO OPERATE AT 2K.2.4K.4.8K AND \n9.6K BITS PER SECOND. HO ALTERATION OF THE SOSC IS REQUIRED \nTO OPERATE AT THE VARIOUS DATA RATES. IT IS NECCESSARY THAT \nSPEED AND FUNCTION COMPATIBLE DATA SETS BE USED AT OPPOSITE \nENDS OF THE DATA LINK.\n\n1.4 PRIOR TO ATTACHING A DATA SET TO THE SDSC THE FOLLOWING \nOPTIONS INTERNAL TO THE DATA SET SHOULD BE CHECKED PER THE \nAPPROPRIATE BSP. ALL OTHER OPTIONS AS SUPPLIED BY THE \nFACTORY.\n\n1.4.1. GROUNDING - FOR ESS USE FRAME GROUND MUST BE SEPARATED \nFROM SIGNAL GROUND.\n\n1.5 THE SDSC WILL ONLY OPERATE III HALF-DUPLEX MODE. E.G. IT \nMAY NOT BE USED TO SIMULTANEOUSLY TRANSMIT ANO RECEIVE.\n\n1.6 THE SDSC IS TRANSPARENT TO THE PROTOCOL USED. IT IS ALSO \nTRANSPARENT TO THE DATA TRAWSHITTED WITH THE EXCEPTION OF\n\nTHE FIRST CHARACTERS IHICH MET BE EIA STANDARD SYN CHARACTERS. \nTHESE ARE USED FOR MESSAGE DETECTION AND DISCARDED. ALL \nOTHER DATA MANIPULATIONS MUST BE DONE BY THE PROGRAM.\n\n2.1 \u00a9LINE INACTIVE, TRANSHIT CLOCK IS ALWAYS PRESENT FROM THF \nDATA SETS TO THE CUSTOMER EQUIPMENT (SDSC) AT BOTH ENDS OF \nTHE LINE.\n\n2.2 THE CALLING END HAS DIALED AMD THE CALLED PHONE IS RINGING \n\u00a9THE LINE LIGHT ON THE CALLED PHONE IS FLASHING.\n\n2.5 THE LINE KEY IS DEPRESSED\u00ae (PLACING THE PHONE IN THE VOICE \nMODE), THE HANDSET IS PICKED IIP, AND VOICE CONTACT IS \nESTABLISHED.\n\n2.4 THE FIRST PROGRAM ACTIOS TWHI IS TO SET DATA TERMINAL READY \n\u00a9 THE \"TR\" LAMP ON A 201-C DATA SET WILL BE LIT WHEN DTR \nIS ACTIVE.\n\n2.5 THE DATA (RaEASE) KEY OB THE TEL. SET IS THEN DEPRESSED\u00a9 \nTHE LINE LIGHT SHOULD THEM ILLUMINATE AND THE HANDSET HAY BE \nHUNG UP WITHOUT DROPPING THE LINE.\n\n2.6 APPROXIMATELY 55ms LATER THE DATA SET WILL ACTIVATE ITS \nREADY BIT\u00a9 INDICATING DATA MODE AND ILLUMINATE THE \"MR\" \nLAMP (201-C).\n\n2.7 THE OH STATE OF THE \"MR\" LAW SHOULD NOT BE INTERPRETED TO \nMEAN THAT A DATA COMMtHI CAT IONS PATH EXISTS. THE CALLING \nSTATION MUST ALSO BE III THE DATA MODE.\n\nTHE PROTOCOL) BUT FOR THE SAKE OF THIS EXAMPLE WE WILL ASSUME \nTHE CALLING END DOES SO.\n\n2.9 THE CALLED (RECEIVING} END MET SET ITS' RECEIVE BIT\u00a9 \nENABLING THE RECEIVE MESSAGE REGISTER AND THE CALLING \n(TRANSMITTING) END TMEB SETS ITS TRANSMIT BIT WHICH ASSERTS \nREQUEST TO SEND TO THE DATA SET \u00a9 .\n\n2.14 AS THE FIRST NON-SYN CHARACTER IS DETECTED THE SDSC WILL \nASSERT \"NEW MESSAGE\" \u00a9 OHF070) .\n\nMAY DISABLE REQUEST TO SEND EITHER MANUALLY OR AUTOMATICALLY \nVIA THE ABE BIT. THIS IS IMMEDIATELY REFLECTED BY THE DATA \nSET REMOVING CLEAR TO SEND.\n\n2.17 THE RECEIVING OATA SET SIMILARLY CONTINUES TO RUN FOR SEVERAL \nCYCLES \u00a9 FOR THE SAKE REASON AND THEN DISABLES CARRIER \nDETECT TO THE CALLED SDSC.\n\n2.18 THE SDSC SEES DCD GO AWAY AND REMOVES THE RECEIVE CLOCK \n(IF STILL PRESENT) CAUSING NEW MESSAGE AND SYNC TO GO \nINACTIVE. THIS TRIGGERS AN END OF MESSAGE PULSE AND FORCES \nA FILL OF THE TDC BUFFER CIRCUIT IF THE BUFFER READY FLAG \nIS NOT SET.\n\n2.20 COMMUNICATIONS TYPICALLY CONTINUE LIKE THIS WITH REPETITIVE \nLINE REVERSALS UNTIL ALL DATA IS TRANSMITTED \u00a9 .\n\n2.21 DATA TRANSFER MAY BE INTERRUPTED AT ANY TIME TO RE-ESTABLISH \nVOICE CONTACT (SEE F\"G. 2) OR TERHINATE THE CALL. TRANSFER \nTO VOICE MODE OR CALL TERMINATION MUST OCCUR AT BOTH ENDS\n\n2.22 THE CALL IS TERMINATE^ BY EITHER \u00a9 TRANSFERING TO VOICE \nMODE AND HANGING UP Or , IF THE DATf SET IS OPTIONED FOR DTR \nCONTROL, BY REMOVING DATA TERMINAL READY Q .\n\n2.23 THE LINE IS ACTUALLY TERMINATED 15ms LATER \u00a9 WHEN DATA \nSET READY GOES INACTIVE.\n\n2.11 APPROXIMATELY 5ms LATER \u00a9 THE CARRIER SENT BY THE \nCALLING END IS RECEIVED AND SYNCHRON I ZED BY THE CALLED \nDATA SET WHICH BEGINS TO SEND CLOCK INFORMATION ANO \nCARRIER DETECT TO THE CALLED SDSC.\n\n2.12 ANY TIME AFTER CTS GOES ACTIVE THE CALLING SDSC IS FREE \nTO SEND DATA TO ITS* DATA SET \u00a9 USING THE TRANSHIT \nCLOCK TO SHIFT DATA FROM THE TDC BUFFER CIRCUIT.\n\n2.13 AFTER DETECTING TWO SW CHARACTERS (026 OR 226) \u00a9 \nCALLED SOSC INDICATES SYNCHRONIZATION (INF 060).\n\nTHESE SIGNALS OCCUR IDENTICALLY AT THE \nOPPOSITE END OF THE LINE DURING THE \nREVERSE DATA TRANSFER. ALL SHOULD BE \nINACTIVE DURII.G NORMAL CALL TERMINATION.\n\n1. the point at which dtr is set is mot critical as long \nas it is active prior to putting the tel set into data\n\n3. RINGO IS ONLY ACTIVE AT THE TIME THE CALLED TEL SET IS \nACTUALLY RINGING (THE SAME TIME AS THE TEL SET LINE \nLAMP IS LIT).\n\n5. UNLESS OTHERWISE STATED, SIGNALS SHOWN ARE THE CALLED \nEND OF THE LINE (THE END WHICH RECEIVES THE FIRST \nMESSAGE).\n\n6. RECB1 MUST BE SET PRIOR TO RDATA1 GOING ACTIVE AND \nSHOULD BE SET PRIOR TO DCDO GOING ACTIVE.\n\n8. ALTHOUGH CTS GOES INACTIVE SIMULTANEOUSLY WITH RTS \nTHERE IS AN EFFECTIVE INTERVAL OF 1-2ms WHEN RTS Mtfi \neS REASSERTED WITHOUT INVOKING THE NORMAL 150-\u00bbs DELAY.\n\n9. LINE DISCONNECT IS ACHIEVED IN DATA MODE BY REHOVING \nDTR, OR IN VOICE MODE BY HANGING UP THE HANDSET.\n\n10. DELAY INDICATED AT THESE POINTS IS VARIABLE ID A \nFUNCTION OF PROGRAM TIMING.\n\n3 1 A CALL HAY BE ORIGINATED EITHER MANUALLY AS SHOWN IN \nFIGURE 1, OR AUTOMATICALLY WITH AT! AUTOMATIC CALLING \nUNIT (ACU). THE ACU MUST BE INTERFACED SEPARATELY \nFROM THF DATA SET.\n\n1. the actual point at which dtr is set is not critical \nas long as it is set prior to transferring into data\n\n201. UNLESS OTHERWISE SPECIFIED. ALL HIRING SHALL BE 30 GAUGE. \nAUTOMATICALLY INSTALLED, 0-4 TTPE HIRING.\n\n207 THE DATA SET MAY BE LOCATED lil ANY SPACE AVAILABLE AS LONG AS \nTHE CONNECTING CIRCUIT CABLE DOES fiOT EXCEED 50 FEET IN L\u00a3t\u00abTH.\n\n302. IN SDSC APPLlCAIIONS OF THE TDC REFER TO \nAPPROPRIATE SD FOR INFORMATION ON USER SUPPLIED \nDATA SET. SEE NOTE 1 ON FS 7 FOR SOSC COMPATIBLE \nRECOMMENDED DATA SETS.\n\n10 THIS TERHINA1 IS A STAJSARD COAXIAL TERMINATION FIELP(CTF) GROUND PIN. IT'S GROUND IS COMMON WITH ALL GROUND \nTERMINALS VIA ThE WitTliAYER PRINTFE HIRING BGARC(MLPWB), E04C017-30, GROUND PLANE.\n\n11. THE FOLLOWING SHOWS Tt\u00a3 SYJSCLIC EQUIVALENT OF THE TABULAR REPRESENTATION\".\n\n13. THE FOLLOWING SHOWS THE SYMBOLIC EQUIVALEHT CF THE TABULAR REPRESENTATION;\n\n14. THE FOLLOWING SHOWS THE SYMBOLIC EQUIVALENT CF THE TABULAR REPRESENTATION;\n\n15. THE FOLLOHING SHOWS THE SYMBOLIC EQUIVALENT OF THE TABULAR REPRESENTATION:\n\n16. THE FOLLOWING SHOWS THE SYMBOLIC EQUIVALENT OF THE TABULAR REPRESENTATION:\n\nOPT NOTE DESTINATION DESIG METHOD SYM TERMINAL DESIG TERMINATION TERMINAL\n\nHTAOOO CA3 \nRTAOOG CA3 \nkTA'OG CA3 \nRTA10G CA3 \nURENABOG CA3 \nHROATAG CA3 \nRDDATAOG CA3 \nROCLKOG CA3 \nDATDETOG CA3 \nOATDETO CA3 \nHTAOO \nRTAOO \nHTA10 \nRTA10 \nHFCNABO \nURDATA \nRDDATAO \nRDCLKO", "title": "Common Systems, Tape Data Controller circuit", "trim_reasons": [], "year": 1975}
{"archive_ref": "bellsystem_CD-1A210-01", "canonical_url": "https://archive.org/details/bellsystem_CD-1A210-01", "char_count": 7740, "collection": "archive-org-bell-labs", "doc_id": 1545, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1545", "record_count": 1, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bellsystem_CD-1A210-01", "split": "test", "text": "APPENDIX 3D\n\nDWG ISSUE 14D\n\nDISTN CODE 1T99\n\nFEB 2\n\nS 1977\n\n^\n\nCHANGES\n\nD. Description of Changes . \u25a0\n\nD.l Changed the rating of option W from A&M Only to AT&TCo\n\nStandard for No. 3 ESS only. Option W remains A&M Only . \nfor all other systems.\n\nD.2 Added Circuit Note 111 which details the limited use of\n\noption w. i \u25a0 '\n\nD.3 Changed Circuit Note 104 and FS 1 to reflect the changes \nin D . 1 . .'\n\nD.4 Corrected a drawing error in Circuit Note 102.\n\nBELL TELEPHONE LABORATORIES, INCORPORATED\n\nDEPT 5317-GJS-GH\n\n\\^F\n\n^i!P\n\nPrinted in U.S.A.\n\nPage 1 \n1 Page\n\nfi\n\nJi.\n\n<^\n\nCIRCUIT DESCRIPTION\n\nCD-1A210-01\n\nISSUE 7D\n\nAPPENDIX 2D\n\nDWG ISSUE 13D\n\nvy\n\nELECTRONIC SWITCHING SYSTEMS\n\nNO. 1, lA, 2, OR NO. 3 \nARRANGED WITH 2 -WIRE FEATURES\n\nREMOTE MASTER SCANNER \nAPPLIQUE CIRCUIT-\n\nFEB 1 8 1975\n\n1^1\n\nCHANGES\n\nD. Description of Changes\n\nD.l Revised FS I to show the necessary connecting Information \nfor the application of this circuit into ESS No. 3-\n\nD.2 Corrected the lead Index and Circuit Notes 102 and 104 on \nFS 1 and CAD 1 to reflect the changes in D.l.\n\nBELL TELEPHONE LABORATORIES, INCORPORATED'\n\nDEPT 5317-GJS-GH\n\n'kJ\n\nPrinted in U. S. A,\n\nPage 1 \n1 Page\n\nCIRCUIT DESCRIPTION\n\nCD- 1A2 10-01\n\nISSUE 7D\n\nAPPENDIX -IB\n\nDWG ISSUE 12B\n\nELECTRONIC SWITCHING SYSTEMS\n\nFEB 2 4 1975\n\nNO. 1 OR NO. 2 \nARRANGED WITH 2-WIRE FEATURES\n\nP EMOTE MASTER SCANNER \nAPPLIQUE CIRCUIT\n\n^\n\n^^f'\n\n. CHANGES\n\nB . Changes in Apparatus \nB. 1 Added\n\nCR10 through CR15\n\nApp Fig. 1 \nCR20 through CR25\n\nApp Fig. 1\n\n(6) Diodes 4 46K - \n(6) Diodes 446K -\n\nD . D escription of Changes\n\nD. 1 Revised FS 1 to show the addition of \nbattery isolating diodes, option R. The \ndiodes will eliminate circulating currents \nbetween paralleled battery plants caused by \npotential di f f erences .\n\nD.2 The option index. Circuit Notes 102 \nand 104, and App Fig. 1 have been corrected \nto reflect the addition of the battery \nisolating diodes.\n\nD.3 Added Circuit Note 110 and Equipment\n\nNote 204-\n\nF. Changes in Description of Oger a tign\n\nF.I In Section II, under 4-. BATTERY_AND\n\nGROUND FOR OUTSIDE ESS OFFICE, add:\n\n(d) Option R when the circuit is part of \na multiple monitoring a single signal \nsource in conjunction with other \ncollocated No. 1 ESS offices.\n\nF.2 In Section II, unde^ 6-_ PURPOSE_OF\n\nCOMPONENTS, add:\n\n6.03 Diodes CRI and CR2 (option R) are \nprovided to eliminate circulating currents \nbetween multipled No. 1 ESS offices \nresulting from battery potential \ndifferences.\n\nw\n\n\\i\n\n!\\\n\n1 \ni\n\n( I i\n\nBELL TELEPHONE LABORATORIES, INCORPORATED\n\nDEPT 5317-GJS-GH\n\nPrinted in U.S.A.\n\nPage 1 , 1 Page\n\nf \nL\n\n-=\u00bbi\u00abivrTr- --^1\n\nCIRCUIT DESCRIPTION\n\nCD-1A210-01\n\nISSaE 7D\n\nDH3 ISSUE 11D\n\nELECTRONIC SWITCHING SYSTEMS\n\nFEB 2 ^ 1975\n\nNO. 1 OR NO. 2 \nARRANGED WITH 2 -WIRE FEATURES\n\nREMOTE MASTER SCANNER \nAPPLIQUE CIRCUIT\n\nSECTION I\n\nGENERAL DESCRIPTION\n\n^:z7\n\nSw^\n\n#\n\n\"<!\u25a0\n\n1. PURPOSE OF CIRCUIT\n\n1.01 The purpose of this circuit is to \nprovide means to connect miscellaneous \ncircuits outside the central office to the \nmaster scanner or universal trunk scanner \ncircuit.\n\n2. GENERAL METHOD OF OPERATION\n\n2.01 Connecting circuits requiring the \nuse of this circuit must connect on a loop \nbasis and be capable of opening or closing \nthe loop.\n\n2.02 Battery and ground can be supplied \nby either the No. 1 ESS or by the \ntransmission facility, but not both.\n\n2.03 When battery and ground is \nsupplied by No. 1 ESS, it can be supplied \neither by the master scanner circuit or by \nthis circuit, but not both.\n\n2.04 The remote master scanner applique \ncircuit can be used for connecting circuits \nwithin the office.\n\nSECTION I I - DETAILED DESCRIPTION\n\n1 . LOOP SIGNAL S\n\n1.01 Connecting circuits requiring the \nuse of this circuit must supply an open or \nclosed loop signal. An open loop provides \nno current; a closed loop produces current \nin the associated ferrod sensor control \nwinding.\n\n2. FERRO D SENSOR\n\n2.01 One ferrod sensor is associated \nwith one applique circuit. The, applique \ncircuits are designated through 7. There \nare eight circuits per unit.\n\n3i MASTER SCANNER OR. UNIVERSAL\n\nTiyNK_SCANNER ~ '\n\n3.01 The scanner circuit is under \nprogram control and reports the loop \ncondition to the system.\n\nii BATTERY, AND .GROUND, FACILITIES FOR\n\ngUTSIDE_ESS_2FFICE\n\n4.01 Transmission facilities outside \nthe central office should be compatible \nwith one of trie follo#fing options:\n\n(a) Option Y when battery and ground is \nprovided at the scanner circuit\n\n(b) Option W when battery and ground is \nnot provided at the scanner circuit \nor fro.Ti the distant office\n\n(c) Option Z when battery and ground is \nprovided by the distant office.\n\n5i. BATTERY. AND\u201eGROUNp\u201eFACILITIES_FOR\n\nINSIDE THE ESS OFFICE\"\n\n5.0 1 Connecting circuits within the \noffice should be compatible with one of the \nfollowing options:\n\n(a) Option W when battery and ground is \nnot provided by the scanner circuit\n\n(b) Option Y when battery and ground is \nprovided Dy the scanner circuit.\n\n6i. PURPOSE. OF_COM PONE NTS\n\n6.01 Resistors R1 and R2 limit the \ncurrent and provide lightning protection \nand a balanced loop for each circuit.\n\n6.0 2 Contact protection network CP \nreduces the peak voltage across the ferrod \nsensor for each circuit.\n\nPrinted in U.S.A.\n\nPage 1\n\nCD-1A21C-01\n\nISSUE 7D\n\nSECTION III - REFERE NCE DATA \nIs WORK ING LIMI TS\n\n1.01 Maximum external circuit \nresistance is 6.8 kilohms. Minimum\n\npsulation resistance is 30 kilohms.\n\n1.02 Voltage limits: \nVoltage Minimum Maximum\n\n-48 42.75 52.5\n\n2. FUNCTIONAL DESIG NATIONS\n\n2.01 Associated Scanner Ferrod Sensors\n\nIsiqnat ion \nthrough 7\n\nFUNCTION\n\nMe ani ng\n\nFerrod sensors associated \nwith this circuit are \nnumerically designated \nfor program reference.\n\n3.01 The basic function of this circuit \nis to provide the facilities for connecting \nmiscellaneous circuits outside the central \noffice to the master scanner or universal \ntrunk scanner circuit.\n\nt. CONNECTI NG_CIRrnTTS\n\n4.01 When this circuit is listed on a \nkeysheet, the information thereon is to be \nfollowed.\n\n(a) Master Scanner Circuit - \nSD-1A1 18-01.\n\n(b) Miscellaneous circuits for all \nFrames\n\n- SD-1A129-01.\n\n(c) Master Scanner circuit - \nSD-1A209-01.\n\n(d) Universal Trunk Scanner - \nSD-2H161-01.\n\n(e) Route Transfer and Make Busy \ncircuits - SD-56048-01.\n\n(f) Auxiliary Signal Circuit - \nSD-63842-01.\n\n(g) Toll Switchboard No. 1, 3, or 3C \nEmergency Ringback Circuit - \nSD-64244-01.\n\n(h) Alarm Circuit - SD-81223-01.\n\n(i) Power Systems - SD-8 1626-01-\n\n(j) N2 Carrier Group Alarm Circuit - \nSD-97118-01.\n\n(k) N2 Carrier Group Alarm Circuit - \nSD-97166-01.\n\nii MANUFACTURING TESTING REQUIREMgNTS\n\nIntermediate R equire ments\n\n5.01 None. \nEnd Req uirem ents\n\n5.02 This circuit should be tested to \nverify that it is wired in accordance with \nthe schematic and wiring drawings, and that \nthe circuit is capable of performing all \nfunctions stated in this circuit \ndiscription,\n\n6^ ALA RM SYSTEMS\n\n6.01 This circuit has a single fuse to \nthe -1*8 volt supply in the bay fuse panel. \nThere is one fuse for every eight circuits \nin a unit.\n\n6.02 A blown fuse causes the FA relay \nto operate an alarm. In No. 1 ESS the FA \nrelay is part of the miscellaneous circuit. \nIn No. 2 ESS the FA relay is part of the \ntrunk peripheral decoder circuit.\n\n2i rAKING_E2yiPMENT_0UT_0F_SERVICE\n\n7.01 Instructions for taking this \ncircuit out of service can be found in \nBSP-231-130-301.\n\nSECTION_IV_::_REAS0NS_FOR_REISSUE \nPi. De script ion of_Changgs\n\nD.I Changed the CADs to provide means \nby which this circuit and non-ESS frames, \nin the same building, can be connected \ndirectly.\n\nr>\n\n^\n\n^\n\nD.2 Corrected CAD 1 \nlead designations.\n\nto show the A, B\n\n^kL TELEPHONE LABORATORIES, INCORPORATED \nDEPT 5723-GJS-GH\n\nr>\n\nPage 2, 2 Pages\n\n\u25a0-WUI^^^BII.a^ VIM^,,\n\n\u2022JiVBf-i\n\n*\u2022 >. r\n\nr-", "title": "Electronic Switching Systems, Common, Remote Master Scanner, Applique Circuit", "trim_reasons": ["leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise"], "year": 1977}
{"archive_ref": "bellsystem_CD-3H205-01", "canonical_url": "https://archive.org/details/bellsystem_CD-3H205-01", "char_count": 5142, "collection": "archive-org-bell-labs", "doc_id": 1604, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1604", "record_count": 1, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bellsystem_CD-3H205-01", "split": "test", "text": "\"THE 'INFORMATION \"CONTAINED KERCtbJ iSHDULD \n>fc)T -BEOISCLOSED TO UNAUTHOfcim*^RSDNS. \n>fT <iS '-MEANT SOLELY FOR USE BY AUTW\u00a9ft+2\u00a3D \n\u2022W^&I'F-JC-' NORTHWEST BELL EMPLOYEES.\n\nCD-3H205-01\n\nISSUE 1\n\nDWG ISSUE 1\n\nDISTN CODE 7T11\n\nELECTRONIC SWITCHING SYSTEMS \nNO. 3\n\nj Ul 8 1976\n\nDIAL-TONE-FIRST COIN LINE \nCIRCUIT\n\nm.\n\ni ;\n\nSECTION I\n\nGENERAL DESCRIPTION\n\n1.\n\nPURPOSE OF CIRCUIT\n\n1.01 The \nspecify the \nJ3H00EE uni \nwiring and \ncircuit pac \npotential f \nAlso provide \n-48 volt dc \ndc talk bat \n\u266648 volt \ndial-tone- fi \nthe miscella\n\npurpose of this circuit is to \nwiring information for the \nt, which provides the means of \nphysically mounting eight \nks to provide +48 volts dc \nor coin lines in No. 3 ESS. \nd are the fuses needed for\n\nsignal battery, the + 48 volts \ntery, and the filters for the\n\ndc talk battery. All \nrst units will be mounted on \nneous frames.\n\n2^ GENERAL DESCRIPTION OF OPERA TION\n\n2.01 The operation of this circuit \ninvolves placing +4 8 volts dc potential \ntoward the coin line on the ring lead and \nkeepinq the tip lead at ground. Also, the \ncircuit repeats the supervision of tha coin \nline back to the junctor circuit, where it \nis maintained.\n\nSrcTigN_II_-_pF^AILED_DESCRIP2:iON \n1_. DIAL TONE FIRST (DTF) CIRCUIT\n\n1.01 This circuit is used to connect a \ncoin telephone to a +48 volt dc source to \ndisable the TOUCH-TONE* pad and to release \na single nickel dropped into the coin \ntelephone. There are two DTF circuits on \none circuit pack. Distribution fuses for \nboth -48 and +48 Vdc and the filters of the \n\u266648 Vdc are provided on this unit.\n\nPOWER DISTRIBUTION\n\n1.02 The +48 volt talk battery, via an \nA and B bus, comes to the DTF unit from the \nmiscellaneous power frame, and is filtered \nat the unit by C1, R1 for the A bus and C2,\n\nR2 for the B bus. After the filter, power \nis distributed to the circuit packs by \nfuses DFAO through DFA3 for the A bus and \nfuses DFBO through DFB3 for the B bus.\n\n1.03 The -48 volt signal battery, via \nan A and B bus, comes to the DFT unit from \nthe miscellaneous frame circuit. \nDistribution of the -48 volt battery to the \nDTF circuit packs is done through fuses \nDCA0 through DCA3 for the A bus and DCBO \nthrough DCB3 for the E bus.\n\nDTF CIRCUIT \u2022\n\n1.04 \ncircuit, \ntelephone \noperation \nDTF cin \nindeoende \ncircuit \ntelephone \njunctor \nobtained \nthe coin\n\nThe DTF circuit,\n\nis connected dir\n\nthat is set for\n\nThe circuit pa\n\nuits that are oper\n\nntly of each ot\n\nin the bypass s\n\nis connected d\n\ncircuit from wh\n\nand supervision i\n\ntelephone.\n\nbeing a line \nectly to a coin \ndial- tone- fir st \nck contains two \nated completely \nher. With the \ntate, the coin \nirectly to the \nich battery is \ns maintained for\n\n* Registered U.S. Patent Office.\n\n1.05 With the circuit in the reverse \nbattery state, +48 volt talk battery is \nplaced on the ring lead to the coin \ntelephone and ground is maintained on the \ntip lead. Capacitors C1 and C2 are placed \ninto the circuit to pass ac signals and to \nprovide dc isolation between +48 volt \nbattery and the -U8 volt battery coming \nfrom the junctor circuit. Loop current to \nthe coin telephone causes relay K2 to \noperate, thus closinq a dc current path \nthrough L2 back to the junctor circuit. L2 \nprovides a low impedance for dc continuity \nwhile maintaining a high impedance to voice \nsiqnals. Supervision is maintained at the \njunc+or circuit with dc current.\n\n1.06 One relay, designated A and \noperated by the distributor circuit under \nprogram control, is used to provide the \nnecessary states for the various functions \nof the circuit. F.Ach state is defined by a \nparticular relay operating or releasing.\n\nPrinted in U.S.A.\n\nPage 1\n\n?D-3H205-01\n\nISSUE 1\n\nSECTIOW III - REFEREM Cff D ATA\n\n1. WO RKING LIM TfS\n\n1.01 The maximum external loop\n\nresistance is 1000 ohms . The minimum\n\ninsulation resistance is 10,000 ohms.\n\n_FUNCTIONAL_DESI3NATIONS\n\n2.01 Pel ay s\n\nD esigna tion Meaning\n\nA This relay is alpha-\n\nbetically designated \nfor prograrr reference.\n\n3. FUNCTIONS\n\n3.0 1 Provides a direct connection \nitween the coin telephone and the junctor \ncircuit.\n\n3.0 2 Provides the *HS volts to the ring \nlead and repeats suDervision to the junctor \ncircuit from the coin telephone.\n\nfij CQNNECTING_CIRCUITS\n\nU.01 When these circuits are listed on \nthe keysheets, the connecting information \nthereon is to be followed.\n\n(a) Distribute \nSD-3H150-01 .\n\nPoint\n\n(b) Network Frame Circuit \nAppearance) - SD-3H901-01.\n\nCircuit - \n(Network\n\nf~\\\n\n5.= MANyFACTJJRING_TJSTING_PE2yiREMFNTS\n\nl2^rmediate_geauirements\n\n5.01 None. \nEnd_Be_guirements\n\n5.02 This circuit should be tested to \nverify that it is wired in accordance with \nthe schematic and wiring drawing, that the \nrequirements of the circuit requirements \ntable are met, and that the circuit is \ncapable of performing all functions stated \nin this circuit description.\n\nL: TAKING .EQUIPM ENT OUT OF SERVICE\n\n6.01 Information for taking this \ncircuit out of service is found in IM-3H000 \nand OM-3H000. Also, the associated fuses \nmust be removed before removing the PTF \ncircuit from this unit.\n\nBELL TELEPHONE LABORATORIES, INCORPORATED\n\nDEPT 5341-WLH-LFG\n\n^\\\n\n^\n\nPage 2, 2 Pages", "title": "Electronic Switching Systems, No. 3, Dial-Tone-First Coin Line circuit", "trim_reasons": ["leading_ocr_noise", "leading_ocr_noise", "leading_ocr_noise"], "year": 1960}
{"archive_ref": "bellsystem_CD-3H402-01", "canonical_url": "https://archive.org/details/bellsystem_CD-3H402-01", "char_count": 16358, "collection": "archive-org-bell-labs", "doc_id": 1608, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1608", "record_count": 20, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bellsystem_CD-3H402-01", "split": "test", "text": "points. These signals are the directory \nnumber of a customer to which a distant \noffice is calling. In addition, this \ncircuit checks the continuity of loop \ntrunks and transmits a battery- reversal \nstart-pulsing signal (wink) to the MF \ntransmitter of the distance office. This \ncircuit also provides a high degree of \nprotection against false operation by \nvoice-frequency components contained in \nspeech or noise picked up by an operator \ntransmitter and/or coupled into the \ntransmission facility as crosstalk.\n\n2.01 When a call is from a distant \noffice, the local office must be sent the \ncalled number so that it can complete the \ncall. An MF receiver is the circuit that \nreceives the called number in the form of \ntone signals from the distant office. \nAfter the calling customer has dialed the \ncalled number and the distant office has \ndetermined that the call is to another \noffice, an idle MF transmitter will be \nseized and connected to the appropriate \ntrunk circuit. The trunk circuit will then \nsend a seizure (start) signal to the local \noffice. The local office will connect an \nidle MF receiver to the trunk circuit via a \nnetwork path through a bypassed junctor. \nTo verify the connection, the processor \nwill scan the continuity scan point of the \nMF receiver. Whan the MF receiver is ready \nto receive digits, the MF receiver sends a \nwink (reverse battery) signal to the \ndistant MF transmitter. The MF transmitter \nwill detect the wink signal and the distant \noffice will cause the MF transmitter to \nstart sending the digits in the form of \ntone signals conforming to the 2-out-of-6 \ncode. The MF receiver will detect and \ndecode the signals into the frequency \ncomponents.\n\n2.02 Several timers are used to verify \nthat the signals detected are true digits. \nWhen the signal input exceeds a minimum \nvalue, a timer is started. Once the MF\n\nreceiver decodes the signal into at least \ntwo freauency components, a second timer is \nstarted. When both timers time out, the \nsignal-present scan point is operated. The \nprocessor will then scan the frequency scan \npoints to determine the digit sent and \ncheck that only two of the six frequencies \nare present. If the signal input does not \nexceed the minimum valve or goes away \nbefore the timer times out, the timer will \nbe reset and no signal-present indication \nis given. If only one frequency is \npresent, the second timer will not be \nstarted and no signal-present indication \nwill be given when the first timer times \nout. To prevent double registration of a \ndigit, a third timer is started when the \nsignal input is removed. The third timer \nwill hold the signal-present scan point \noperated for a period of time to prevent \nnoise from causing a second signal-present \nindication to be given on the same digit. \nAfter all digits have been received, the \nprocessor idles the MF receiver and \nreleases it.\n\n1.01 MF signaling uses one group of six \nfrequencies in the speech band (700, 900, \n1100, 1300, 1500, and 1700 Hz). A valid MF \nsignal consists of exactly two of these six \nfrequencies. This signaling code is known \nas the 2-out-of-6 code. There are only 1 5 \nvalid combinations, which can be used as \ndigits or other signals.\n\n1.02 When a call is from a distant \noffice, the distant office will locate and \nconnect an idle MF transmitter to the \nappropriate trunk circuit. This connection \nwill then cause the trunk circuit to send a \nseizure (start) signal to the local office.\n\n1.03 The local \nconnect an idle MF-re \ntrunk circuit via a \nbypassed junctor, \nsend a start-pulsing \nMF transmitter when i \ntrunk. When the MF t \nwink signal, the \noperate the MF-tra \nrelays and begin send\n\noffice will locate and \nceiver circuit to the \nnetwork path through a \nThe MF receiver will \nsignal (wink) to the \nt is connected to the \nransmitter detects the \ndistant office will \nnsmitter tone^select \ning the digits.\n\n1.0i\u00bb The MF receiver will receive the \nsignals from the distant office and decode \nthe individual frequencies. The receiver \nwill time the input signal and the \nfrequency combinations to verify that a \ndigit has been received (not noise or \nspeech). When the input signal persists \nfor a minimum time, with at least two \nfrequency components, the signal-present \nscan point is operated. The operation of \nthe signal-present scan point will cause \nthe processor to scan the frequency scan \npoints. The processor then determines that \ntwo out of the six frequencies (2-out-of-6\n\ncode) are present and that the frequency \ncombination is a valid combination. If \neither test fails, the processor aborts the \ncall. To prevent double registration of a \ndigit (noise interference) , another timer \nis started when the input signal is \nremoved. This timer will hold the \nsignal-present scan point operated for a \nperiod of time to prevent causing the \nprocessor to res can the frequency scan \npoints for the same digit. The processor \nwill monitor the signal-present scan point \nfor each successive operate and release (to \ninitiate a frequency scan) until all digits \nare received.\n\n2.01 The relationship between digits \nand other signals and the MF-signaling \nfrequencies in the 2-out-of-6 signaling \ncode is shown in Table A.\n\nMF receiver has one state \nrelay (A), which is controlled by the \ndistributor circuit. The processor \ndetermines the state required and, via the \nperipheral decoder circuit, operates the \nrelay.\n\n3.02 Relay A, driven by -48 V, has a \n1000-ohm resistor, with one of the break \ncontacts in parallel with it, in series \nwith the coil to initially provide rapid \noperate. This arrangement also limits the \nmaximum current through the distribute \npoint when the relay is operated. ro limit \nthe transient voltage spike when the relay \nis released and to dissipate the relay-coil \nenergy when released, a diode is connected \nfrom the distribute point to -48- V with the \nanode of the diode to -48 V.\n\nlamo. The remainder of the circui< \ncomponents will be discussed in the order \nin which the signal passes through them.\n\n4.02 The input transmission components \nare L, T, C5, C6, C7, and C8. Inductor L \nconnects the continuity scan point (SCO) \nwhich suDplies battery and ground, to the \nnetwork via the tip-ring reversing contacts \nof relay A. Transformer T has a very lowe- \ncapacitance between the primary and \nsecondary windings, which provides a high \ndegree of protection from longitudinal \nlightning surges to the circuit. \nCapacitors C5, C6, C7, and C8 provide \nprotection against metallic surges. C5 and \nC6 also provide dc blocking in the \ntransformer primary. The input \ntransmission circuitry, when terminated \ninto the variolosser circuit (P/o circuit \npack A260) , presents a very high return \nloss to the network.\n\n3.04 When the MF receiver is in the \nidle or wink state, the A relay is released \nand -4 8 V and ground are connected to the \ntip and ring respectively (reverse \nbattery) . Operating the A relay connects \n-48 V and ground to the ring and \nrespectively.\n\n3.05 This is a tvpical state sequence \nof this circuit. See BSP 233-151-105 for \nmore details.\n\n4.01 An MF-receiver circuit is \ncomprised of 11 circuit packs, input \ntransmission circuitry, and a \nvolt age- divider power supply (R1 \nthrough R5) . Because all operating \npotentials are proportional to the battery \nvoltage, all potential changes are as well \nThus, overall battery variations have \nlittle effect on receiver performance. \nFilter capactors (C1 through C6 on circuit \npack A1024) smooth the output voltage of \nthe voltage divider. The power-off (PWR \nOFF) key removes battery from the circuit \nto permit circuit pack removal and testing. \nA resistor (R3 on circuit pack A1024) \nlimits the current in the PWR OFF indicator\n\n4.03 The variolosser circuit (P/o \ncircuit oack A26 0) provides a controlled \nloss to the input signal and passes the \nsignal to the AGC amplifier (P/o circuit \npack A261). A dc control current from the \nAGC will change the shunt impedance (and \nthus the gain) of the variolosser circuit. \nThis feedback control assures an almost \nconstant steady-sta*-e ac signal at the \nvariolosser output.\n\n. 4.04 Any feedback \nfinite time to respond \nchanges in the input \ncalled the attack time \nThe attack time of \napproximately 5 ms. If \nis received, the vario \nexcessively-high input \namplifier until the \nreduce the variolosser\n\n4.05 After the receiver sends the wink \nsignal (reverse battery) , relay A is \noperated to place the circuit in the \nreceive-digits state. Operating relay A \nconnects ground and battery to tip and ring \nrespectively, and also creates high-level \nvoice- frequency transients, which appear as \nan input to the receiver. These transients \ncould cause the variolosser circuit to be \nplaced in a high- loss condition until the \nAGC system responds. This high-loss \ncondition could persist long enough to \nprevent detection of the low-level keypulse \n(KP) signal from the distant office. To \nprevent this problem an inhibit amplifier \n(P/O circuit pack A26 2) stops the feedback \ncontrol from operating for approximately \n20 ms after relay A operates. The 20-ms \ndelay is sufficient to allow for the \nbattery-reversal transients to die out and \nno longer affect the operation of the \nreceiver.\n\nit. 06 The AGC amplifier amplifies the \nsignal from the variolosser and supplies \nthe amplified signal to the signal driver \n(P/O circuit pack A261) , the guard driver \n(P/O circuit pack A260) , and the control \nrectifier of the variolosser, which \ngenerates the direct current to control the \nvariolosser gain.\n\n4.07 The guard driver provides a \nf reguency-dependant dc bias to the \nMF-channel detectors (A152s) to prevent \ndigit simulation (see 5.01 through 5 .01) . \nThe guard filters (P/O circuit packs A26 2 \nand A263) short signaling frequencies out \nof the guard-driver input. All other voice \nfreguencies are amplified and rectified to \nprovide the dc guard bias.\n\n4.08 The signal driver provides power \ngain to the signal so that the channel \nfilters (circuit packs A2 64 and A26 5) can \nrespond. These six filters are a parallel \ncombination of series- re sonant circuits. \nThis configuration offers low impedance at \nsignaling frequencies and requires that the \nsignal driver provide high (ac) current.\n\n4.09 The channel filters separate the \nsignal into the frequency components, \ncorresponding to the MF-signaling code, and \napply the individual frequencies to the \nMF-channel detectors. The channel \ndetectors (dc) will convert, the frequencies \napplied to a direct current only if the \nguard bias does not exist. A direct \ncurrent would operate the frequency scan \npoints (SC2 through SC7) . When two channel \ndetectors operate, a valid signal is \npresent.\n\n4.10 The signal-present timer (circuit \npack A266) is two separate timers ANDed \ntogether. The first timer is activated by \nthe ac signal from the signal driver. If \nthe input persists continuously for 20 ms, \nthe timer will time out and activate one \nside of the AND gate. The second timer \nwill be activated when at least two \nchannel-detector outputs are operated. If \nthe channel- detector outDUts are operated \ncontinuously for 10 ms, the timer will time \nout and activate the other side of the AND \ngate. If the input to either timer is \nremoved before the timer times out, the \ntimer will be reset. When both inputs to \nthe AND gate are activated, the AND gate \nwill operate the scanner driver (P/O \ncircuit pack A1024), which will operate the \nsignal-present scan point (SC1). The \noperating of the signal-present scan point \n(SC1) will cause the processor to scan the \nfreguency scan points (SC2 through SC7) and \ninterpret the digit received. To prevent \ndouble registration of the same digit, \nbridge timing is used. When the input to \nthe timers is removed after time-out, the \ntimers will time for 20 ms before releasing \nthe AND gate. The 20-ms delay will prevent \nnoise from causing the signal-present scan\n\n5.01 The MF receiver uses guard action \nto reduce digit simulation or the response \nof the receiver to signals that are not \ngenuine MF pulses. The guard driver \napplies a dc bias to the channel detectors \nthat varies in magnitude according to the \nfrequency and amplitude of the received \nsignal. With guard action, genuine MF \npulses excite the guard driver only weakly \nand strongly excite the proper channel \ndetectors. Thus, the receiver is able to \nrespond. In contrast, most speech signals \ncontain many frequency components that \nexcite the guard driver more strongly than \nthe channel detectors. Guard bias is \ngenerated to inhibit the response of the \nchannel detectors and the receiver tends \nnot to respond to signaling frequencies in \nthe presence of other signals.\n\n5.02 The AGC amplifier applies a signal \nto the guard driver. The MF guard filter \neffectively shorts the signaling \nfrequencies out of the guard driver. Any \nvoice-frequency signal not shorted out will \nbe amplified and rectified to produce a \ndc-bias voltage. Since the MF-signaling \nfrequencies are shorted out, the dc bias \nwill only appear as a result of the \npresence of nonsignaling frequencies. This \nbias raises the threshold of the channel \ndetectors and inhibits the recognition of \nsignaling frequencies. Consequently, if a \nsiqnal is to be recognized, frequencies in \nthe 2-out-of-6 code must be present and \nother signals must be either absent or at a \nmuch lower level.\n\n5.04 Another \nfor ac-signal lev \nor greater than \nsignals. Lower 1 \ntimer at all. Th \nac signal be pre \ndigit can be \naction and two \nprevent digi \nvoice-frequency e\n\nlonger timing is provided \nels approximately equal to \nthe minimum expected MF \nevels cannot activate the \nis timer requires that an \nsent for 20 ms before a \nrecognized. Thus, guard \nseparate timing operations \nt simulations by \nnergy.\n\n3.01 Provides for reception of \nmultifrequency signals from operator \nkeysets or automatic sending equipment and \nconverts the received signals into outputs \nsuitable for detection by the scanner.\n\n3.02 Provides proper response to MF \npulses of 30-ms duration with 30-ms \nintervals between pulses.\n\n3. .03 Provides response to MF pulses \nwith received level within the limits of \n-22 dBm and -2 dBm per tone. The receiver \nis able to respond properly in the presence \nof 6. C-dB twist (difference) between any \ntwo MF tones.\n\n^.0 4 Provides 6c output signals of at \nleast 11 ms for each of the signaling \nfrequencies, regardless of input-signal \nduration, when a valid signal is received.\n\n3.05 Provides a signal-present output \nsignal, which starts about 2 ms after a \nvalid input signal is received and remains \noperated until about 20-ms after the input \nsignal is terminated.\n\n3.06 Provides tolerance for variation \nin the received signaling frequencies of \n\u26661.5 percent +10 Hz about the nominal \nMF-signalinq frequencies.\n\n(c) A signal-validity check reguiring \nthe operation of two (or more) \ndetectors for a timed interval \nbefore the signal-present output is \ndelivered.\n\n(d) A fast-cycling timer, which forces \nan input signal above a threshold \nlevel to persist uninterrupted for a \nrequired time interval before the \nsignal-present output is delivered.\n\n(e) Signal-present output holding to \nprevent recycling when noise causes \nshort breakups in the input signals.\n\n3.08 Provides means to prevent a \nreverse-battery start-pu] sing transient \nfrom affecting the AGC feedback loop; the \nfeedback loop is inhibit ed for a timed \ninterval after operation oi the A relay.\n\n4.01 when this circuit is listed on a \nkeysheet, the connecting information \nthereon is to be followed.\n\n5.01 Before circuit packs are inserted \nin the unit, it should be verified that the \nunit is wired in accordance with the \nschematic and wiring drawings to prevent \ndamage to the circuit packr.\n\n5.02 This circuit should be tested to \nverify that it is wired in accordance with \nthe schematic and wiring drawings, that \nrequirements of the circuit requirements \ntable are met, and that the circuit is \ncapable of performing all functions stated \nin this circuit description.\n\n6.01 This circuit is fused individually \nwith one fuse to the -\u00abJ8 V talk supply. if \nthe fuse blows, it will cause an FA relay, \nin the frame that it is mounted on, to \noperate as alarm.\n\n7.01 Information on taking this circuit \nout of service can be found in BSP IM-3H0 \nand OM-3H000.", "title": "Electronic Switching Systems, No. 3, Multifrequency Receiver Circuit", "trim_reasons": [], "year": 1960}
{"archive_ref": "bellsystem_CD-3H901-01", "canonical_url": "https://archive.org/details/bellsystem_CD-3H901-01", "char_count": 3826, "collection": "archive-org-bell-labs", "doc_id": 1615, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1615", "record_count": 11, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bellsystem_CD-3H901-01", "split": "test", "text": "order to break the ground loop between battery boost \ncircuits. This is reflected in Equipment Note 202.\n\nD.l Diodes were added to cable assembly CA02 to break up the input \nnode multiple and to confine the effects of a cross between the \ntip/ring path and the pulse path.\n\n2.01 The detailed operation of all \ncircuits listed in Table A may be found in \nthe circuit description for each one. Any \ncircuit that is shown only in the Network \nFrame Circuit St will be explained in \ndetail in this document.\n\n1.01 The system control periodically \nalternates the on- and off-line states of \nthe two halves cf the ringing and tone \n(P6T) plant. Pa this occurs, the RTS relay \non each network traire 1b operated or \nreleased. This causes the audible and\n\noverflow tones from the on-line side to be \nconnected to one set of splitting \nresistors, which are connected to the \njunctors. The relay operation causes \neither the RST transfer alarm (0) or R6T \ntransfer alarm (1) scan point to saturate \nin the master scanner. If both saturate, \nthis indicates that a relay on one of the \nnetwork frames did not transfer the tones.\n\n2.01 A CONV OFF key on the control \npanel is used to turn the battery boost \nconverter off and on. A lamp in the key \nlights when the OFF key is depressed. \nThere are two keys: one for converter \nand one for converter 1. Two diodes are \nmounted in the base of the frame. These \nallow -48 volt power to be supplied to the \nload if the battery boost converters fail \nor are removed.\n\n3.01 The TELA and TELB jacks and the SP \njack are mounted on the front of the \ncontrol panel. These jacks are multiplied \ntogether on all network frames. Any other \nTELA and TELB jacks are also connected in \nparallel and finally connected to the \nperipheral test circuit on the test frame. \nAny other SP jacks are also connected in \nparallel. These jacks are used for a \ngeneral-purpose beltline. A pair of jacks \non the back of the control panel provides \n-48 volt power for testing purposes.\n\n4.01 There are three types of battery \npower provided to network fraire circuits. \nSignal battery is the -4 8 volt power, \nprovided to user circuits, which is \nunfiltered after entering the frame. Talk \nbattery is filtered on the network frame \nand then distributed to user circuits. \nBoosted talk battery la talk hatt\u00ab>ry that \nis boosted above the battery plant voltaqe \nand is used to extend the dc-signaling\n\nrange to 1600 chirs without special per-line \ntreatment. Table B contains a list of \ncircuits and how the power supplied to them \nis fused.\n\n4.02 The LT fuse is a 0.75-A fuse which \nsupplies power tc the -48 volt test jack. \nThe MC fuse is a 0.5-A fuse which supplies \npower to the CONV OFF lamps, the FA lamp, \nand the PTS relay.\n\n4.03 When any fuse supplying power to \nthe 15A remreed grids blows, the FAA relay \noperates. This lights the FA lamp in the \nfuse panel and sends a contact closure to \nthe line-attending-element fuse alarm scan \npoint in the iraster scanner iratrix circuit.\n\n4.04 When any other fuse blows, the FAB \nrelay operates and also lights the FA lamp. \nA contact closure is sent to the trunks, \njunctors, and service circuits common fuse \nalarm scan point.\n\n4.05 Both the \nno ise- suppr ession \ncoils. Pesistors \nbus and -48E bus \nfuse on each bu \nboth relays, but \nbattery boost fus \nThus, additional \nbuses separate an \nseparate froir the\n\n3.01, This circuit shows the \ninterconnection of all circuits on the \nnetwork frame and the connections to any \nother circuits on other frames.\n\n4.01 When these circuits are listed on \nthe keysheets, the connecting information \nthereon is to be followed.\n\n5.02 The manufacturing testing \nrequirements are specified in the x-79037 \nspecification.\n\n6.01 Information for taking any circuit \non this frame out-of-service is found in \nIM-3H000 and OM-3H000.", "title": "Electronic Switching Systems, No. 3, Network Frame Circuit", "trim_reasons": [], "year": 1960}
{"archive_ref": "bellsystem_BSP_007-220-304", "canonical_url": "https://archive.org/details/bellsystem_BSP_007-220-304", "char_count": 18147, "collection": "archive-org-bell-labs", "doc_id": 1680, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1680", "record_count": 19, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bellsystem_BSP_007-220-304", "split": "test", "text": "1.01 The purpose of the Preliminary Design Phase \nis to develop the overall architecture of the\n\nsystem and to produce specifications reflecting the \nlogical view of the system.\n\n1.02 Whenever this section is reissued, the rea- \nson(s) for reissue will be included in this para- \ngraph.\n\n1.03 This section is a guideline. It provides ex- \npanded information in support of the concepts\n\nof Total System Development specified in Section \n007-220-300*, Total System Development - Mile- \nstones.\n\n1 .04 During Preliminary Design, the system model \nselected at the completion of the Feasibility\n\nPhase must be reevaluated to determine if that ap- \nproach is still valid and can be satisfied by the system \nrequirements developed during Definition Phase. \nAlso, the reasonable architectural options must be \nexamined in terms of their technical, operational, \nand economic feasibility. Once a generalized archi- \ntectural approach has been established, the design of \nthe system must be developed. Also during this de- \n* Check Divisional Index 007 for availability.\n\nsign activity, system control features and perfor- \nmance and reliability requirements must be \nintegrated into the total system design.\n\n1 .05 During this phase, the strategies and require- \nments for system test and conversion should\n\nbe established. Typically, the selected conversion \nstrategy will influence selection of a specific test \nstrategy. For example, a flash cut of the entire sys- \ntem would require that each portion of the system be \nfully tested before conversion. On the other hand, \nsequential conversion of subsystems or system ver- \nsions would dictate a phase and cumulative test \nstrategy. Because of the high interrelation between \nthe system test and conversion functions, planning \nfor the two should be a coordinated effort.\n\n1.06 Throughout the Preliminary Design process, \ndecisions will be made concerning the re- \nsources required to perform specific system functions \nor sets of functions. As the functional or logical de- \nsign is firmed up, the total resource requirements for \nthe system must be determined for both conversion \nand on-going operation. Requirements should be de- \nveloped for personnel, equipment, facilities, trans- \nportation, hardware, software, and the \ncommunications network. As these requirements are \nprepared, they should be reviewed with each appro- \npriate planning and/or support organization to as- \nsure resource availability and conformance to \norganizational resource plans.\n\n1.07 By the end of the Preliminary Design, there \nwill be sufficient information to develop a rea- \nsonably precise view of the costs and benefits of the \nnew system. Development and conversion costs, oper- \national expenses, anticipated maintenance costs, and \nthe magnitude of system benefits should be \nrecalculated. Significant differences (10 to 15 per- \ncent) from Feasibility Phase estimates should be ex- \namined and the reasons for the increase(s) or \ndecrease(s) identified.\n\n1.08 The composition of the project team for the \nPreliminary Design Phase is critical to pro-\n\nducing an effective system design. Depending upon \nthe type(s) of system architecture that is to be em- \nployed, a variety of expertise may be required \u2014 per- \nsonnel subsystem and human performance, computer \nsubsystem, data base, data communications, soft- \nware, hardware, etc. For large projects or for projects \nwith heavy technical involvement in any of these \nareas, it is usually advisable to have technical experts \nactually assigned as project personnel. Where techni- \ncal demands are not so great, it may be possible to \nobtain part-time assistance from technical support \ngroups.\n\n1 .09 It is important that all project personnel work \nas a team throughout the Preliminary Design\n\nPhase. The selection of a system architecture and the \ndevelopment of an optimum system design will re- \nquire a great number of technical decisions and \ntrade-offs utilizing inputs from each of the specialty \nareas represented on the team, eg, PSS, CSS, data \nbase, and data communications. This interdisciplin- \nary approach to design is beneficial throughout the \nsystem process. Additionally, every effort should be \nmade to colocate all project personnel during Prelim- \ninary Design in order to foster the decision-making \nprocess and the effective flow of information.\n\n1.10 There should be a high degree of interaction \nbetween the project team and the technical\n\nsupport and planning groups during Preliminary \nDesign. The various support groups can provide in- \nformation and ideas concerning design alternatives. \nIn addition, in most companies, the support organiza- \ntion is responsible for the actual selection of equip- \nment, software, hardware, etc, based on the total \nrequirements across projects/applications. Also, the \ndesign of the communications network and data \nbases for the system will typically be a support re- \nsponsibility based on requirements from the applica- \ntion. Because of these shared responsibilities, the \nproject manager should assure a close day-to-day \nworking relationship with all affected planning and \nsupport organizations.\n\n1.11 It is important that a design review be held at \nthe completion of the Preliminary Design\n\nPhase. The review should focus on the operational, \ntechnical, and economical characteristics of the sys- \ntem and should typically involve the following orga- \nnizations as required:\n\n1.12 The purpose of this design review is to assure \nthat all facets of the design have been prop- \nerly addressed and satisfy the system requirements \nand to gain the acceptance, concurrence, and/or ap- \nproval of the various organizations that are or will he \ninvolved in the development, installation, or opera- \ntion of the new system. Thus, this review point is a \nmajor milestone in the system development process.\n\n2.01 The first step in addressing the design of the \nsystem structure is the review and validation\n\nof the proposed system model. The system require- \nments are then reviewed to determine if they ade- \nquately represent the system model. If changes to the \nmodel or requirements are required, they should be \ndocumented.\n\n2.02 Following these reviews, design activities can \nproceed. The emphasis at this point is twofold:\n\n(a) To finalize the logical data views into the logi- \ncal record, segment, and data base specifica- \ntions\n\n(b) To structure the system down to a level of \nmodule and program descriptions for machine\n\nprocessing; task and work module (position) de- \nscriptions for human processing.\n\n2.03 The logical data structure, usually developed \nin the Definition Phase, may be presented in\n\nproducts such as linkage diagrams, entity depen- \ndency diagrams, and usage views. Based on this data\n\ninformation and system functions information from \na high-level flow diagram, the logical record specifi- \ncations and segment and data base specifications can \nbe developed.\n\n2.04 The detailed functional architecture of the \nsystem can be developed in parallel with the\n\nwork on the data structure and specifications. Al- \nthough the architecture is heavily related to the data \nstructure, there is sufficient information on the \nviews of the logical data from the Definition Phase \nto start identifying the functional architecture for \nthe system. During this activity, several architecture \noptions may be identified. In developing these op- \ntions, the following influences in addition to the data \nstructure must be considered:\n\nEach identified architecture option must be evalu- \nated in terms of its feasibility [ie, technical, opera- \ntional, and economical feasibility (both \ndevelopmental and operational)].\n\n2.05 Identifying the functional architecture op- \ntions basically involves the following three\n\n2.06 When allocating the system functions, some \nanalysis of these functions will be required in\n\norder to clearly identify to which processor, person or \nmachine the function or subfunction should be allo- \ncated. Those functions allocated to people should be \nreviewed to determine that the results will be mean- \ningful work for them. Also, allocated functions \nshould be continually reviewed throughout the de- \nsign process to assure the allocation is still appropri- \nate. For example, it may become evident in later\n\ndesign that the volume of work, accuracy required, or \nprocessing criticality is such that people cannot per- \nform effectively or efficiently, and the function \nshould be reallocated to the machine. Conversely, a \nfunction assigned to the machine may be performed \ninfrequently, require judgmental decisions, and \ncould be performed more effectively by people. In this \ncase, the function should be reallocated to the human \nprocessor.\n\n2.07 Once an architectural option has been selected \n(with the functions appropriately allocated),\n\neach function can be further analyzed. The low level \nprocesses identified through this analysis can then be \nsynthesized into modules and programs for machine \nprocessing or tasks and work modules (positions) for \nhuman processing. During this analysis process, the \ndata structures and function considerations must be \nconsidered in order to develop an effective design.\n\n2.08 System control and reliability requirements \nmust be examined to determine the processes\n\n(machine and human) which need to be interwoven \nwith the transaction oriented functions. The various \nareas which require system controls and the means \nfor providing such controls will determine the control \nprocesses to be designed and integrated into the over- \nall system design. The processes to satisfy reliability \nrequirements also must be designed and integrated \ninto the overall design. The design of these two sets \nof processes will be influenced by the architectural \noption selected and must also consider the data and \nfunction structures.\n\n3.01 At this point in Preliminary Design, enough is \nknown about the system requirements; system\n\nfunctions; personnel requirements; hardware, soft- \nware, and equipment requirements; and the system \nschedule to develop the initial system test and con- \nversion strategies. Both of these strategies are de- \npendent on the nature of the system that has been \ndesigned and both can be influenced by the dictates \nof the other strategies (ie, testing can be influenced \nby the conversion strategy or vice versa).\n\n3.02 For testing, an overall system test plan for \ndynamic testing will be developed and re- \nviewed. This plan will include general test objectives, \ntesting environment (verification, validation, and \ncertification) for each level of testing (unit, integra- \ntion, and system), strategies and techniques to be\n\nused, testing schedule, resource estimates for testing, \nand test data base requirements. This overall test \nplan should be reviewed and evaluated. Once the plan \nis accepted, specific test plans will be developed in the \nDetail Design Phase for the PSS and CSS units, inte- \ngration, system, conversion, and acceptance testing.\n\n3.03 Static testing of phase products will be \nplanned and conducted using techniques such\n\nas desk top reviews, peer reviews, andwalk-throughs. \nAlthough each phase of TSD will include these types \nof product reviews, they are especially important \nduring Preliminary Design because of the develop- \nment of the overall test plan. During these product \nreviews, an objective is to ensure the design specifica- \ntions contain specific and measurable criteria for \nunambiguous testing because these specifications \nwill be the basis for developing the overall test plan.\n\n3.04 Another factor to consider in developing the \noverall test plan is the selected conversion\n\nstrategy. Different strategies may require adjust- \nments to various aspects of the test plan such as the \nschedule, resources, test cases, and test data base.\n\n3.05 As previously noted, the selection of an appro- \npriate conversion strategy is done at this point\n\nas a part of the initial view of conversion require- \nments. The specifics for the conversion plan will be \ndeveloped in Detail Design based on these conversion \nrequirements.\n\n3.06 The data conversion considerations developed \nin the Definition Phase should be reviewed\n\nand updated if required. This information plus \nknowledge of the system architecture and data struc- \ntures will be used to select a conversion strategy (eg, \na phased conversion or flash cutover), and to identify \nif conversion can be accomplished using the system \nfunctions or if special conversion functions must be \ndesigned or a conversion subsystem is needed.\n\n3.07 Other factors to consider in developing the \nconversion strategy and requirements are the\n\nimpact on the business operations during conversion, \nsuch as reorganization in user work units, special \ntraining requirements for conversion procedures, \nand resource requirements (eg, unique or special \nhardware or software and personnel requirements \nfor conversion).\n\n4.01 With the basic design of the system available, \na more definite set of physical resource re-\n\nquirements can now be determined. These resource \nrequirements should address the following as appro- \npriate for the system:\n\nThese resource requirements must include both con- \nversion and operational needs. In many cases, the \nsame resources will be used for both conversion and \ndaily operations. However, when special or addi- \ntional resources are required for conversion, they \nmust be clearly highlighted, and the length of time \nthey will be required should be identified.\n\n4.02 The cost analysis of these resource require- \nments are discussed in Part 5, Refined Eco- \nnomic Analysis. However, when special or additional \nconversion resources are required, their potential \ncosts should be reviewed in conjunction with the \nlength of time they will be required based on the con- \nversion strategy selected. If potential costs are high \nfor these resources, it would be appropriate to exam- \nine the entire proposed conversion process to deter- \nmine if adjustments can be made to reduce these \ncosts. Although conversion costs are not on-going, \nthey can be a major expenditure when bringing in a \nnew system.\n\n4.03 Each of the seven potential resource require- \nments must be defined based on the design of\n\nthe system at this point in the development cycle and \nthe system requirements with any updates incorpo- \nrated during this phase. As the various resource re- \nquirements are defined, the current set of \nassumptions and constraints must be reviewed to \ndetermine if there are any conflicts. If conflicts are \nfound, they must be resolved by clearing or rewriting \nassumptions, negotiating changes to constraints, or \nmaking adjustments to the design. This may require \nobtaining agreement to changes in the system re- \nquirements.\n\n4.04 Various modeling, simulation, and quantitive \nanalysis techniques are available to assist in\n\ndetermining each type of resource requirement. The \nselection of techniques to use is influenced by a num- \nber of factors:\n\n(e) Skill and knowledge of project members in \nusing a specific technique or availability of\n\n4.05 As the various resource requirements are pre- \npared, they should be reviewed with each ap- \npropriate planning and/or support organization to \nassure resource availability and conformance to or- \nganizational resource plans. If the required resources \nare not available or are not in comformance with re- \nsource plans, then adjustments must be negotiated \neither to the plans to provide for the resources or to \nthe system under development.\n\n5.01 At this point in the development cycle, the \neconomic analyses prepared in the Feasibility\n\nPhase are reviewed and refined but only for the sys- \ntem alternative being developed. The same economic \nanalyses and method of calculation used during feasi- \nbility should be used again at this point to assure a \ncommon basis for comparing any variances in the \nresults. If additional analyses are deemed useful, \nthey should be prepared. However, if additional anal- \nyses are prepared, those results which are \nrecalculations of the Feasibility analyses should be \nclearly indicated.\n\n5.02 The cost and benefit calculations may still be \npresented in ranges; however, the difference\n\nbetween the low and high ends of a range calculated \nduring Preliminary Design should be less than calcu- \nlated during Feasibility. With more specifics known \nabout the system during this phase, the refined eco- \nnomic analyses results should be more precise.\n\n5.03 The refined economic analyses results should \nbe compared with the same calculations pre- \npared during Feasibility. Any results which differs\n\nsignificantly (10 to 15 percent) from the Feasibility \nestimates should be examined, and the reasons iden- \ntified for the increase(s) or decrease(s).\n\n6.01 During this phase, various types of product \nreviews should have been conducted. However,\n\na final review of the overall system preliminary de- \nsign should be conducted at this point to assure logi- \ncal continuity and completeness of design based on a \ncomparison with the system requirements produced \nin the Definition Phase and any updates to it. Testing \nand conversion information, resource requirements, \nand refined ecomonic analyses results should also be \nincluded in this end-of-phase review. This review will \ntypically involve such groups as users, support ap- \nproval, operation, or any other group that will be \nimpacted by the system or that can contribute tech- \nnical or subject matter expertise to the review.\n\n6.03 The status and recommendations plus other \nsupporting information required by the ap- \npropriate project approval entity are submitted to \nthat group to obtain authorization to proceed. The \nproject approval entity should evaluate the project in \nthe following ways:\n\n(a) Compare the refined economic analysis with \nthe economic results from the Feasibility\n\n(b) Determine if the business goals and objectives \naddressed by the system are still viable and\n\n7.01 The following sections will provide additional \ninformation relevant to the Preliminary De- \nsign Phase:", "title": "Total System Development / Preliminary Design Phase Guidelines", "trim_reasons": [], "year": 1983}
{"archive_ref": "bellsystem_BSP_007-220-305", "canonical_url": "https://archive.org/details/bellsystem_BSP_007-220-305", "char_count": 27563, "collection": "archive-org-bell-labs", "doc_id": 1681, "document_type": "technical_report", "id": "bella-qwen-pretrain-doc1681", "record_count": 37, "release_policy_version": "hf_public_v1", "rights_status": "public_domain", "selected_extraction_backend": null, "selected_extraction_score": null, "source_family": "archive_org", "source_url": "https://archive.org/details/bellsystem_BSP_007-220-305", "split": "test", "text": "1.01 The primary objectives to be satisfied during \nthe Detail Design Phase are:\n\n(b) Produce the detailed specifications that are \nrequired to construct the system.\n\n1.02 Whenever this section is reissued, the rea- \nson(s) for reissue will be included in this para- \ngraph.\n\n1.03 This section is a guideline. It provides ex- \npanded information in support of the concepts\n\nof Total System Development specified in Section \n007-220-300*, Total System Development - Mile- \nstones.\n\n1.04 Many of the activities within this phase are \ninteractive. Because major activities may be\n\nassigned to several groups of people, each designer or \ndesign group must be sure to identify all critical de- \nsign interfaces and monitor the progress of individu- \nals responsible for related system modules. The \ndesign of a module may also be affected by decisions \nmade in performing activities later in the phase. For \nthese reasons, the designer may find the design or at \nleast the specification of a module may not be \nfinalized until near the end of the phase.\n\n1 .05 Although they are not highlighted as a sepa- \nrate activity because they are embedded in the\n\nsystem design, such things as error control, system \ncontrols and examination, reliability, and recovery \nmust be considered throughout the Detail Design \nPhase. \u25a0 \u25a0 :-.,. ;-,;;_ .\u2022\n\n1 .06 It is recognized that some detail design activi- \nties may actually be performed by support\n\norganizations; for example, physical data base de- \nsign, data communications specification, facility \nplanning, etc. Each project must determine its spe- \ncific responsibility for each of these design functions \nand provide a means for overall coordination of the \ndesign effort. , ,. > ,.,,,; . : ..^ruy,:r:\n\n1 .07 Control of change during this phase is critical. \nThe design change requests that are received\n\nduring this phase and the next will usually impact \neither the system design, the system development \nschedule, or both. In order to properly control such \nchanges, procedures to document, evaluate, put in \norder of priority, and determine the disposition of \neach design change request must be strictly followed.\n\n2.01 Procedures must be developed for all of the \nwork modules and functions of the system.\n\n2.02 Tasks within each work module or function \nmust be analyzed to determine the complete,\n\ndetailed processing requirements. This may involve \nbreaking each task down into further levels of detail \nuntil all required activities for accomplishing the \ntasks have been identified.\n\n2.03 The designer must also analyze each task and \nstep to determine the conditions under which\n\nmalfunctions or errors are likely to occur. The objec- \ntive of this analysis is to identify each possible error \ncondition; determine how frequently it is likely to \noccur; and decide how to prevent, minimize, or cor- \nrect such occurrences. If there is no way to prevent \ncertain types of contingencies, it may be necessary to \ndesign corrective procedures to handle these prob- \nlems.\n\n2.04 The resulting procedure should optimize both \nprocessing flow and human performance. The\n\nprocessing sequence should be logical and efficient. \nAll decision points should be clearly identified and all \nsubsequent processing steps specified. Control steps \nshould be provided as necessary to avoid unnecessary \nor incorrect processing. Whenever possible, feedback \nmechanisms should be provided to inform the opera- \ntor of how well the function is being performed.\n\n2.05 Once each manual procedure has been devel- \noped, the designer should determine the need\n\nfor exhibits, decision or statement tables, perfor- \nmance aids, forms, reference materials, or other tools\n\n2.06 When forms are to be used in the work mod- \nule, each form should be designed and a speci- \nfication prepared, stating: ' \u2022 ' \u25a0 \u25a0\n\nWhen forms are to be stored in manual files, the con- \ntent, organization, and physical attributes of the file \nmust be determined.\n\n2.07 Some work modules require unique equip- \nment, furniture, or facilities. Such special\n\nwork station requirements must be identified and \nreflected in the development of the manual proce- \ndures. This is particularly important for Visual Dis- \nplay Terminal (VDT) or equipment positions as these \nmechanisms may have special capabilities or restric- \ntions on usage. , . ^\n\n2.08 Once the procedures, forms, files, and facili- \nties have been designed, the means by which\n\neach procedure will be documented must be decided. \nAll procedures should be documented, both to prop- \nerly communicate required operating methods and to \nprovide for system integrity, auditability, and con- \ntrol. However, the means selected for this documen- \ntation may vary depending upon system \nrequirements. Procedures may be captured in:\n\nThe designer must select the best means of documen- \ntation in light of user requirements, the nature of the\n\nprocedure itself, on-the-job usage of the product, and \ndocument maintenance considerations.\n\n2.09 Section 007-230-210, System Deliverable Doc- \numentation, provides content and packaging \nspecifications for documentation delivered with a \ncentrally developed system.\n\n3.01 At this point in the Detail Design Phase, the \ndetailed logic procedures for each program or\n\nmodule must be developed. Potential techniques to be \nutilized include detailed functional analysis, stepwise \nrefinement, use of a program definition language, \nflowcharts, and structure charts.\n\n3.02 Although largely independent of the program- \nming language that will be used for implemen- \ntation, the design should take the capabilities and \nlimitations of the language used into consideration. \nThe limitations in particular may require \nrestructuring the functions of a program or module \nso they coincide with a physical structure that can be \nimplemented in the language.\n\n3.03 At this point, the designer must specify any \nmessages or codes that may be produced by\n\n3.04 It is also critical that the designer determine \nhow each module is to be implemented. The\n\ndesigner must specify any interfaces explicitly and \ndesign the internal data structures that will be used \nby the module.\n\n4.01 Detailed specifications must be developed at \nthis time for all the data and physical struc- \ntures used by the mechanized portion of the system. \nThese specifications will serve as documentation of \nthe system and should provide a vehicle to ensure \nsystem integrity, auditability, control, and mainte- \nnance.\n\n4.02 Kach data group and element should be as- \nsigned a unique identifier or alias that will be\n\nused in programs that refers to data. The designer \nmust analyze each use of a group or element to deter- \nmine the best formats (including length and data \ntype) for storage and display, taking into account\n\nspace requirements, the need for conversion, efficien- \ncy, versatility, and readability. Specific codes, includ- \ning record type designators, should be assigned to all \napplicable elements. ,\n\n4.03 The physical layout of each record or segment \nmust be designed now so the precise record\n\n4.04 The organization and use of each file that will \nnot reside under a data base management sys- \ntem must also be specified.\n\n4.05 The design of the physical data bases must be \ncompleted. The designer must select the spe- \ncific data base physical structures and internal ac- \ncess methods that will be used to store and retrieve \nthe data. The designer must also provide for data se- \ncurity and recoverability of the data base as required \nby the Preliminary Design specifications.\n\n5.01 In performing this activity, designers must \nconsider how their data requirements fit into\n\nthe overall company network. Much money and re- \nsources can be wasted if the characteristics of the \ncorporate environment are overlooked. Therefore, \nthe proposed network design should be reviewed with \nPlanning, Technical Support, Computer Operations, \nand Marketing to assure proper compatibility and \ncompleteness. \u2022 \u2022' -\n\n5.02 If an existing corporate communications net- \nwork specification does not meet the needs for\n\nthis system, a new data network specification must \nbe prepared. A data network specification must in- \nclude all the elements of the data system. In addition \nto the communication lines, the designer also must \nconsider who is communicating, what is being trans- \nmitted and where, the type of transmission, telepro- \ncessing software packages and access methods, \nnetwork speed, control, security, backup and recov- \nery.\n\n5.03 The capabilities and services to be provided by \nthe application system will definitely influ- \nence the decisions made by the network designer. For \nexample, line selection, terminals, and response \ntimes will vary depending on whether the application \nis batch, inquiry/response, message switching, data \nentry, or data collection. A further consideration is \nwhether or not the network will use on-line, off-line, \nor interactive processing. . ^\n\n5.04 In planning a data communications system \nthe following seven basic factors should be \nconsidered:\n\n(a) Function: The designer should carefully re- \nview and/or further define the overall objec- \ntives of the teleprocessing system in terms of the \nfunctions it must perform.\n\n(b) Distribution: The designer must identify all \nlocations to be served by the network. The lo- \ncations may be within the same building or in sep- \narate buildings, at short or long distances. Once \nthe locations and the required functions are \nknown, lines and terminals can he considered. The \nline selected could be dial up or private, switched \nhalf duplex or full duplex. Also to be considered is \nthe pattern of flow, in terms of who receives and \nwho transmits.\n\n(c) Volume: Next, the amount of information to \nbe communicated should be determined. Ana- \nlyze both the number and length of messages, as \nwell as peak volumes and potential growth. When \nestimating the probable volumes, retransmission \nrequirements resulting from line or operational \nerrors must also be considered.\n\n(d) Urgency: Closely associated with volumes is \nthe need to know when the information is\n\navailable for transmission, when it is required, \nand what response times are allowable. Speed of \ntransmission will be dependent upon the types of \nmedia, terminals, and lines used. Backup, recov- \nery, and equipment or hardware availability (up^ \ntime) will also require consideration in the net- \nwork design.\n\n(e) Language: The language required by the \nuser (the data and programs) is another im- \nportant consideration. In this respect, particular \nsoftware or hardware and conversion or interface \nrequirements may have to be met. In addition, \nhuman performance characteristics should be con- \nsidered in the design of the human/machine inter- \nface, input coding, output interpretation, etc.\n\n(f) Accuracy: The designer must consider the \nperformance aspect of the teleprocessing net- \nwork. The requirements set for transmission accu- \nracy are critical to the selection of the proper \nphysical components of the communication sys- \ntem. In addition, the teleprocessing software facil- \nities for editing, checking, balancing, and data \ncorrection must be compared to the actual func- \ntional requirements of the application system.\n\n(g) Cost: After weighing all of the factors above, \nthe designer must evaluate the cost of the net- \nwork that has been designed. Trade-offs may be \nnecessary, based on what is required and provided \nversus how much it costs and how much is avail- \nable, with particular attention paid to the true \nvalue of the total system.\n\n5.05 The communication network specification \nshould define the configuration and compo- \nnents of the network, the data to be shared via the \nnetwork with other systems, network utilization and \nperformance requirements, and any special proce- \ndures for network operation.\n\n6.0 1 Once work module and/or work station design \nhas been completed, all equipment require- \nments should be fully detailed. Specific types and \nquantities of equipment items will have been deter- \nmined and vendor negotiations initiated.\n\n6.02 Any requests for equipment modification or \nredesign to be done by the vendor must be\n\nspecified so these requirements may be stated in the \ncontract agreement. The same is true for any special \nwork that will be done withfn the eompany or by a \nlocal contractor.\n\n6.03 An equipment specification should be devel- \noped for each equipment item that will be used\n\nwith the system. The specification should include a \ngeneral description of the item, any required modifi- \ncations, delivery and test requirements, maintenance \ncharacteristics, installation locations, quantity, and \ncost. This specification will serve as a useful checklist \nduring system test and conversion. Therefore, all ac- \ntivities that must be performed to obtain the equip- \nment and assure that it is operational should be \nincluded in the specification.\n\n6.04 The term \"facilities\" refers to the total physi- \ncal space that will be occupied by the person- \nnel and machine components of the system. The \nfacilities requirements should include the amount of \nspace, the physical location, and the environmental \ncharacteristics of the facility allotted to each compo- \nnent of the system.\n\n6.05 In most cases, planning or engineering will be \ninvolved in the detailed planning of machine\n\nfacilities. However, the facilities requirements are \ncritical to assuring that the actual physical facilities \nwill properly support the operational system. For\n\nmachine facilities, the following factors should be \nconsidered in preparing the requirement:\n\n6.06 If the facility is to be occupied by personnel, \nthe following factors should be considered:\n\n6.07 The final descisions concerning these types of \nwork facilities will probably be made by the \nuser organization. In fact, the facilities requirements \nthemselves may be developed by the user representa- \ntives on the project team. However, if the design deci- \nsions and facilities preparation is to be the \nresponsibility of the user department, the system \ndesigners must identify all the facilities characteris- \ntics that will impact overall system performance and \nassure that the final design adequately supports sys- \ntem operation.\n\n7.01 Once the basic design of the system has been \ncompleted, the hardware and software specifi- \ncations can be finalized. This may require revision to \nthe requirements developed in Preliminary Design or \nit may necessitate further detailing of specific fea- \ntures or items.\n\n7.02 The hardware requirements developed in Pre- \nliminary Design have provided Information\n\nSystems Organization (ISO) Planning with an early \nestimate of the resources required for operation of \nthe new system. When program, data base, and file \ndesigns are complete, the original estimates should \nbe reviewed using the more detailed specifications \ndeveloped during this phase. Estimates of individual \nprogram sizes can be used to validate earlier esti- \nmates of module size, device requirements can be \nmodified as necessary, and the specific content of \nrecords within files can be used to recalculate data \ntransfer requirements and validate estimated run \ntimes.\n\n7.03 The planning organization will use detail de- \nsign estimates to adjust the composite projec- \ntion of processing requirements. If any modeling or \nsimulation of the application was done, those results \nshould also be forwarded to the planning group. Such \nresults may aid the planning group in viewing the \ntotal operations environment. This is a critical phase \nfor planning, because if the resources required by \nthis system or others are less than previously antici- \npated, a slippage in the equipment installation sched- \nule could be indicated. On the other hand, an increase \nin requirements may dictate an accelerated installa- \ntion plan, a different machine assignment, or a\" \nchange in the projected system conversion date.\n\n7.04 When the system is to be installed in more \nthan one location, hardware sizing guidelines\n\nshould be developed. The guidelines should define the \nvarious parameters, formulas, and/or processing \noptions that can be used to estimate resource require- \nments for a given processing site.\n\n7.05 Basic software requirements will probably \nremain constant after Preliminary Design. \nHowever, changes in the types of operating control \nsystem, release, or level may affect project plans for \ntest, conversion, and/or operation. New software or \nfeatures announced since the last software evalua- \ntion may be of significant benefit to the new system. \nTherefore, a review of software needs and the prepa- \nration of a final specification is recommended.\n\n8.01 The total personnel requirement to convert, \noperate, and maintain the system must be de- \ntermined. Personnel estimates should include these \ncategories:\n\nFor each category, the numbers of people, skill levels, \ndepartmental affiliation, time frame needed, etc, \nshould be specified. If the system is to be installed in \nmultiple locations, personnel staffing guidelines, \nparameters, or formulas may have to be developed.\n\n8.02 If work modules have been grouped into jobs, \neither with certainty or as a guideline, each\n\nthe work group, personnel and physical require- \nments, administrative requirements, etc.\n\n9.01 Several types of training may be required for \nsystem conversion and/or operation for:\n\n9.02 The designer should consider the specific \ntraining needs of these groups of people as\n\nwell as any others whose jobs may be affected by the \nnew system. The extent of training can be determined \nby comparing the existing skill&and knowledge of the \npeople to the skills and knowledge they will need in \norder to operate and manage the new system. The \nbasis for training then becomes those new capabili- \nties that personnel will have to develop. For each per- \nsonnel category, and for most jobs within a category, \na specific training package will be identified.\n\n9.03 Obviously, not all training packages will have \nto be developed from scratch. In some cases\n\nthere may be existing course materials that will \nserve perfectly well with some modification and/or \nadditions. The designer should review current train- \ning packages and determine where development time \nmight be save through minor revision to existing \ncourses.\n\n9.04 The designer should also determine the num- \nber of people who will require each type of\n\ntraining. The amount of time spent on the develop- \nment of a specific training package will depend to a \nlarge degree on the number of people who will use it.\n\n9.05 For each training package, whether for formal \nor informal instruction, the designer should\n\ndetermine what the people should know or be able to \ndo by the time they complete their training. Those \nthings they will need to know or do become the basis \nfor establishing course objectives.\n\n9.06 Course objectives should be stated so they can \nbe measured in terms of the trainee's actual\n\nperformance. Objectives that cannot be measured are \nof doubtful value for inclusion in the course specifica- \ntion. The setting of course objectives is important in \nassuring that all required topics are included in the \ncourse specification.\n\n9.07 The course objectives will help the designer to \ndecide on the best training method to use.\n\nExamples of training methods are seminars, lectures, \nprogrammed instruction, computer aided instruc- \ntion, w^orkshops, etc. The selection of the proper in- \nstruction method depends greatly on the content of \nthe course, the number of students predicted, the \nskill of the developer, the time available for develop- \nment or instruction, and the materials and equip- \nment available.\n\n9.08 With a method selected, the designer can de- \ntermine how the course should be presented.\n\nThe course could be taught in one continuous session \nor broken up into segments or modules that address \nspecific aspects of the course content. The duration \nof the training sessions are determined by the type \nof training being given, the method of instruction, \nand the number and location of the people being \ntrained. Judgment and common sense are perhaps \nthe best guides in making this determination.\n\n9.09 The materials required for the training pack- \nage must be identified. Examples of training\n\nmaterials are instructor guides, student workbooks, \nhandouts, audio and video tapes, slides, exercises and \ncase problems, quizzes, student and course evaluation \nfbrrtis, etc. In addition to these materials, the deliver- \nable user documentation may be used extensively in \nthe training environment. Such documents as user \nmanuals, administrative guides, work module in- \nstructions, performance aids, and forms can all be \neffectively used in the classroom. Any document that \nwill be used on the job should be introduced and re- \nviewed in the training session.\n\nlO.OT In preparation for the testing activities in \nthe Implementation and Conversion Phases, \ndetailed test plans for the following three levels of \ntesting must be prepared:\n\n(a) Unit testing examines each discrete compo- \nnent (module, program, work module or proce- \ndure) of the system.\n\n(b) Integration testing examines how each com- \nponent of the system interacts with each other\n\nas they are assembled in a stepwise manner. This \ntesting concentrates on chained programs or work \nmodules and human/machine interfaces.\n\n(c) System testing examines the operation of all \ncomponents of the system as a whole, accord- \ning to its system requirements and performance \ncriteria.\n\n10.02 As test plans for each level of testing detail \nare developed, the following testing strate- \ngies, based primarily on environment, should be con- \nsidered:\n\n(a) Verification examines the logical correctness \nof each component (module, program, work\n\nmodule, developmental component, or procedure), \neither individually or together, using controlled \ndata in a test environment.\n\n(b) Validation examines the logical correctness of \nthe system using controlled data in the operat- \ning environment or one that approximates the op- \nerating environment as closely as possible.\n\n(c) Certification examines the performance, qual- \nity, and reliability of the system to ensure that\n\nthe system meets its objectives and its perfor- \nmance requirements. This type of testing is con- \nducted in the actual operating environment using \nreal data. Certification emphasizes tlie perfor- \nmance aspect (eg, volume, response time, fallback, \nand recovery, etc) of the system rather than the \nlogical correctness of the system.\n\n10.03 Acceptance testing is conducted as an exten- \nsion of system testing and completes certifi- \ncation testing. It is performed by or on behalf of those \nwho will use and operate the system to ensure the \nuser's needs are met. It must be performed as a condi- \ntion of acceptance by organizations installing a Cen- \ntrally Developed System (CDS). The Project Manager \nwill coordinate acceptance testing requirements with \nthe recipients of the CDS.\n\n10.04 For each specific test, a test design is pre- \npared outlining objectives, techniques, me- \ndia, expected results, resources, and schedules.\n\n10.05 The design and creation of a test data base is \nalso a major testing activity in this phase.\n\n10.06 A test status report system should also be \nestablished to provide the Project Manager \nwith an effective means to observe and control the \ntest activity.\n\n1 1 .01 Although primary system performance spec- \nifications are established early in develop- \nment, the specific means by which these objectives \nwill be met are reflected in the detailed specifications \nof each of the system components. Because of the vol- \nume of such specifications, it is desirable to provide \na summary of the overall performance attributes of \nthe system as seen at the end of design.\n\n11.02 The following information should be pro- \nvided in the performance specification:\n\n(b) Processing performance criteria (accuracy, \nthroughput, resource utilization, etc)\n\nmeasures have been taken to assure that perfor- \nmance level, and the means by which system perfor- \nmance can be measured in the operational \nenvironment.\n\n12.01 The conversion plan should include all \npreconversion activities, a description of \neach step in the conversion process, and a conversion \nschedule. The schedule should indicate the start and \ncompletion date for all activities, personnel assign- \nments by name, both project and user personnel, and \ndependencies on nonproject events. It is of critical\n\nimportance that the project personnel and the users \nagree on the conversion plan, including the commit- \nment to provide personnel as identified on the con- \nversion schedule. \u25a0\n\n12.02 Since, in some cases, conversion may follow \nthe completion of the Detail Design Phase by\n\nonly a few weeks, the preparation of the converison \nplan has been included as a Detail Design activity. It \nis recognized, however, that for larger projects the \nImplementation Phase may last up to a year or lon- \nger. Under those circumstances, some of the detailed \ndecisions, actual personnel assignments, and the de- \ntermination of schedule dates will be postponed until \nsometime during the Implementation Phase when \nthey can be established with a greater degree of cer- \ntainty.\n\n12.03 The conversion plan and schedule encompass \na number of activities that affect both the\n\ndeveloper and the user. While those activities are \nvery dependent upon the project, the characteristics \nof the system, and the conversion approach, there are \na number of items which should be considered by any \nproject. They are:\n\n(a) Personnel staffing (schedules, method of han- \ndling displacements or acquisitions, etc)\n\nFor each conversion activity, the schedule and person \nor organization responsible should be identified.\n\n12.04 For central developers, the conversion plan \nmust include all those activities required for\n\nsystem installation and development of any interface \nfunctions. This information will be contained in the \nInstallation Planning Guide that is delivered to the\n\n12.05 Once the detailed plan for conversion is com- \npleted, conversion costs and resources should \nbe reviewed. Any significant deviations from the \noriginal estimates should be reviewed with affected \norganizations.\n\n13.01 For most projects, a Detail Design review is \ndesirable or required. However, most of the\n\ndocumentation produced during this phase is fairly \ndetailed and technical. For these reasons, \nmodularized design reviews of the system may be \nrequired.\n\n13.02 Within the project, it may be desirable to \nconduct walk-tbrougbs to determine the\n\nadequacy of individual design specifications. One or \nmore development groups may be involved in this \nprocess.\n\n13.03 Technical design reviews may also be con- \nducted to evaluate certain subsystems within\n\nthe total system. Technical support, planning, opera- \ntions, or the user may participate in these subsystem \nreviews.\n\n13.04 A nontechnical system design review with \nspecial emphasis on operational procedures,\n\nschedules, controls, reports, forms, etc, may be held \nwith user representatives.\n\n13.05 Whatever design review procedures are se- \nlected, the amount of information, technical\n\ncontent, and level of detail should be appropriate for \nthe reviewer's area of interest and technical exper- \ntise.\n\n14.01 The following sections will provide addi- \ntional information relevant to the Detail \nDesign Phase:", "title": "Total System Development / Detail Design Phase Guidelines", "trim_reasons": [], "year": 1983}