hash stringlengths 32 32 | doc_id stringlengths 7 13 | section stringlengths 3 121 | content stringlengths 0 2.2M |
|---|---|---|---|
abd3ec15d3b7f9b4dfa3b9b29f0899ad | 101 445 | 3.1 Definitions | For the purposes of the present document, the following terms and definitions apply: Adaptive Frequency Agile (AFA): radio system that changes its operating frequency to suit the environment in which it finds itself NOTE: AFA systems minimize the risk of interference between radio systems sharing common frequency bands. backscatter: system where a carrier signal is propagated or induced and in which the direction of the incident waves from the Interrogator/Reader are resolved by a Transponder along a reference direction, and are reflected back in a reverse direction to the Interrogator/Reader in varying amplitude as the impedance of the transmitter is modulated using the same frequency or a harmonic of that frequency Battery Assisted Tag: tag without an active transmitter, therefore still reliant on a technique such as backscatter, but that uses an on- board battery either to maintain on board power or boost the return signal Carrier Signal: radio (or inductive) signal generated by the Interrogator/Reader and received by (or induced in) the Tag to initiate and conduct the Transaction, and, in the case of a Passive Tag, to provide the energy for the Tag to function Channel: small frequency sub-band within the operating frequency band into which a Radio Signal fits NOTE: Commonly, a frequency band is divided into contiguous channels. assigned frequency band: frequency band within which the device is authorized to operate conducted measurements: measurements which are made using a direct 50 Ωconnection to the equipment under test dedicated antenna: removable antenna supplied and type tested with the radio equipment, designed as an indispensable part of the equipment fixed station: equipment intended for use in a fixed location full tests: all tests specified in the present document integral antenna: permanent fixed antenna, which may be built-in and is designed as an indispensable part of the equipment mobile station: equipment normally fixed in a vehicle portable station: equipment intended to be carried, attached or implanted radiated measurements: measurements which involve the absolute measurement of a radiated field transponder: a device that responds to an interrogation signal telemetry: use of radio communication for indicating or recording data at a distance wide band: equipment to be used in a non-channelized continuous frequency band covering more than 25 kHz, or to be used in a channelized frequency band with a channel spacing greater than 25 kHz ETSI ETSI TR 101 445 V1.1.1 (2002-04) 7 |
abd3ec15d3b7f9b4dfa3b9b29f0899ad | 101 445 | 3.2 Symbols | For the purposes of the present document, the following symbols apply: E Electrical field strength Eo Reference electrical field strength f frequency P Power R distance Ro Reference distance t time λ Wavelength ∆r Range resolution |
abd3ec15d3b7f9b4dfa3b9b29f0899ad | 101 445 | 3.3 Abbreviations | For the purposes of the present document, the following abbreviations apply: AFA Adaptive Frequency Agile ERC European Radiocommunication Committee FHSS Frequency Hopping Spread Spectrum GTAG Global Tag. EAN●UCC compliant R&TTE Radio and Telecommunications Terminal Equipment RF Radio Frequency Rx Receiver SRD Short Range Device UHF Ultra High Frequency RFID Radio Frequency IDentificator EAN Electronic Article Numbering Association UCC Uniform Code Council |
abd3ec15d3b7f9b4dfa3b9b29f0899ad | 101 445 | 4 Executive summary | RFID technology is used increasingly across a wide range of applications. There are many systems available operating at a number of different frequencies. The choice of frequency is a function of the specific application. For satisfactory operation in logistics and item management reading ranges of up to 2 m using passive tags are necessary. This is not feasible using either inductive or microwave RFID technology. However it is achievable at UHF. This System Reference Document proposes UHF as the frequency best suited to the RF identification needs of the materials handling industry. This is because UHF offers low path loss and high data transmission speed combined with relatively low field strength. The tags can be produced in high volume at low cost and operate at high efficiency. The proposed frequency range and power levels will enable RFID tags to be deployed within the supply chain on a global basis in accordance with a single ISO standard. UHF RFID tags can respond to frequencies anywhere within the band 860 MHz to 930 MHz. Therefore they can work with readers operating within the different UHF frequency bands assigned to SRDs by administrations in the three regions. In other words the same RFID tag can operate both in region 2 with a 902 MHz to 928 MHz FHSS reader and in region 1 where AFA readers operate in the 862 MHz to 870 MHz band. The GTAGprogram aims to provide a global standard for the radio identification of goods moving in the supply chain. This necessitates the assignment of a UHF sub-band of 2 MHz to 3 MHz where frequency agile RFID systems can be used. These readers will use "listen before talk" techniques. This means that prior to the selection of a certain channel the reader will first check that it is not occupied by another device. If a channel is busy then the reader will search until it finds a channel that is unoccupied. Frequency agile systems have been selected after carefully conducted tests. These compared different RFID modulation technologies such as wideband and FHSS systems with regard to compatibility with present users in the same or adjacent sub-bands. The results indicate that wideband frequency agile systems present least interference and highest spectrum efficiency. ETSI ETSI TR 101 445 V1.1.1 (2002-04) 8 The System Reference Document therefore defines the frequency and spectrum needs as well as power requirements. It further proposes that more complete compatibility studies should be carried out with other SRD equipment and services operating within the range 865 MHz to 868 MHz. |
abd3ec15d3b7f9b4dfa3b9b29f0899ad | 101 445 | 4.1 Status of the System Reference Document | This System Reference Document has been adopted by ETSI ERM. |
abd3ec15d3b7f9b4dfa3b9b29f0899ad | 101 445 | 4.2 Technical issues | System description See clause B.1 Applications See clause A.1 New technology (if any) The UHF reader and tags use state-of-the-art technology and available chip designs. Short market information: Increasingly major manufacturers, distributors and retailers are seeking ways to reduce their costs of distribution. Many of these companies are beginning to see the benefits that can be achieved by utilising RFID. With the present state of the technology this is particularly relevant at the logistical asset level. These same companies have also seen the necessity for RFID to operate in accordance with an ISO/IEC 18000-6 (see bibliography) presently in work in ISO/IEC JTC1/SC31 [3]. This would permit goods to be distributed freely throughout the distribution chain on an international basis. The Electronic Article Numbering Association (EAN) and the Uniform Code Council (UCC) is closely working with ISO on the realization of the standard and as a consequence the two organizations have jointly mounted the GTAG project in support of ISO/IEC 18000-6 (see bibliography). During their initial assessment of the user requirement, EAN/UCC researched the global needs of the materials handling industry. As a result of their study they specifically rejected conventional bar codes and battery-assisted tags. The shipment of palletized loads is a physically demanding process during which damage is frequently inflicted on external packaging. To effectively withstand abuse, the means of identification must be positioned inside the load where it can be properly protected. Since line of sight is no longer possible bar codes cease to be an option. In the case of battery assisted tags, end users had two fundamental objections. Firstly there is a practical problem in administering large populations of battery powered tags. Tagged items will frequently be dispersed around the world for extended periods outside the direct control of the owner. In such situations it will be impossible to manage the replacement of tags with dead batteries. Secondly end users were concerned about the additional associated cost - particularly where applications involved millions of tags. As a consequence of this study EAN/UCC determined that passive tags operating at UHF most closely met the needs of their members. Market size, forecasts and timing See clause A.2, Market size and Volume Spectrum requirement and justifications In order to meet best the needs of the industry and the EAN and UCC members, the GTAG Project team investigated the use of RFID at different frequencies across the spectrum. The result of their investigation showed that operation at UHF offered a system with features that most closely met the market requirement. The reasons for reaching this conclusion are summarized below. ETSI ETSI TR 101 445 V1.1.1 (2002-04) 9 It is proposed that the band 865 MHz to 868 MHz be assigned to wide-band non-specific SRDs. The band would be divided into 15 channels of 200 kHz. This will allow SRD manufacturers to develop equipment that will operate at high data rates without interfering with existing users. The equipment will be restricted to either fixed or frequency agile operation. For some RFID applications a maximum permitted power level of 500 mW will be sufficient. However situations will arise where to meet the needs of end users a higher power level is necessary. One way that this might be achieved while minimising the risk of interference to other SRDs is to permit different power levels in the different channels. An example of such scheme is shown in figure 1. 500 mW 2 W Cordless Audio Existing SRDs 863 865 866 867 868 870 MHz 100 mW 865.6 867.6 CT2 10 mW Proposed band for RFID Figure 1 Current regulations Technical and regulatory parameters are defined in the CEPT/ERC/REC 70-03 [1] in annex 1 (Generic applications) The existing UHF sub-bands do not permit sufficient bandwidth and power to operate RFID tags. The new band plan extends the frequency range from 868 MHz to 870 MHz to 865 MHz to 870 MHz. This would allow the operation of a number of wideband channels for non-specific SRDs as well as RFID devices. |
abd3ec15d3b7f9b4dfa3b9b29f0899ad | 101 445 | 5 Main conclusions | The UHF system as described meets the performance requirements of the market and provides global interoperability of RFID systems. Efficient use of the spectrum is assured by frequency agile technology avoiding or minimizing interference both to other co-channel systems and to users of adjacent bands. Preliminary compatibility studies were conducted with positive results using RFID systems at 500 mW together with both cordless audio and radio modems. To implement the use of RFID systems at UHF new ETSI documentation will be necessary as well as amendment of the CEPT/ERC/REC 70-03 [1]. Requested ERC actions - Compatibility evaluations and studies for services as defined under C1 and C2. - Update of CEPT/ERC/REC 70-03 [1] annex 1. ETSI ETSI TR 101 445 V1.1.1 (2002-04) 10 Annex A: Detailed market information A.1 Range of applications Item Management Pallet Management Logistical Assets Containers Returnable Trade Items Totes, Crates, Boxes Truck Trailers Paper Rolls A.2 Market size and value 0 1000 2000 3000 4000 5000 2002 2004 2010 2020 Millions of Euros Supply Chain Management Other (Parcel services, Emerging, Rental, Reusable media,..) Source: EAN International Figure A.1 A.3 Traffic evaluation The traffic evaluation can be estimated from the number of systems, the average emitted power based on antenna patterns and the data rate as reflected by the bandwidth. The power level will range from 100 mW to 2 W depending on the tag technology and the needs of each individual site. Battery assisted tags require lower power while batteryless tags generally need up to 2 W. The ratio of reader to tags typically will vary from 1 : 500 to 1 : 5 000. Most readers will operate only on receipt of an external trigger. ETSI ETSI TR 101 445 V1.1.1 (2002-04) 11 Annex B: Technical information B.1 Detailed technical description The system operates in half duplex using adaptive frequency agile techniques with "listen before talk". For much of the time the reader will be in its "rest" state during which the reader will be powered up but will not transmit its carrier. On the approach of a tag the reader can receive an external trigger command, which will switch it to its "operational" state. This will initiate the idle mode during which the reader will transmit a succession of pulses. Each pulse comprises a wake-up command followed by a period of emission of continuous carrier. A pulse will have a duration of approximately 5 ms. As soon as a tag reaches the interrogation field, providing its "application identifier" matches the one emitted by the reader, it will transmit its response. The interrogator will then enter into a dialogue with the tag during which it may read data from the tag's memory or write to it additional information. In the event that multiple tags are present in the interrogation field, they will reply in accordance with an anti-collision algorithm. The data transmitted by the interrogator is AM at 40 % depth of modulation at a data rate of 40 kBaud. The tag replies using AM modulated backscatter at 66 kBaud. Details of the spectrum mask are shown in the figure below. Spurious emissions will be less than 250 nW in accordance with the limit in EN 300 220-1 [2]. 33 dBm -36 dBm 200 kHz Figure B.1 B.2 Technical justifications for spectrum B.2.1 Power 2 W power is preferred in order for ISO/IEC 18000-6 (see bibliography) and GTAGto achieve the necessary reading range of 2 m using batteryless tags. The power level and possible mitigation techniques such as duty cycle have to be determined after completion of compatibility studies. ETSI ETSI TR 101 445 V1.1.1 (2002-04) 12 B.2.2 Frequency (See also spectrum requirements clause 4.2) UHF has been chosen by GTAGas the best compromise frequency, because it offers the greatest read range for a given transmit level. (e.g. vs. 2,45 GHz) and suffers less from the problems of range reduction with water and reflections than at 2,45 GHz. Because of the proximity of Tags to products that may be aqueous or metallic, these are important considerations. In addition an adequate allocation of bandwidth is necessary to permit the use of sufficient channels. This requirement is specified in the proposed Bandplan under clause 4.2. B.3 Information on current version of relevant ETSI standard At the time of drafting of the present System Reference Document existing SRDs in the UHF band are covered by EN 300 220-1 [2]. Appropriate documentation will be necessary to handle the particular requirements of RFID systems (e.g. operation with 2 W). ETSI ETSI TR 101 445 V1.1.1 (2002-04) 13 Annex C: Expected compatibility issues C.1 Coexistence studies Studies will be necessary between RFID and other systems using the same band or adjacent bands: - Tactical Radio Relay Link (TRRL) where or when applicable; - CT2 systems and with other SRDs such as tele-alarm, telemetry, DSSS, FHSS and cordless audio devices in adjacent bands and within the defined sub-bands; - Television (channel 69) and aeronautical in some countries (S5.323). Practical initial tests (See document ETSI/ERM(01)SA47R1 for details of GTAG compatibility assessment,) show no detrimental interference above distances of 30 cm. Theoretical data (see note): Class 1, Indoor = 5 m, Open Air = 26 m (Blocking) (see note): Class 2, Indoor = 30 m, Open Air = 300 m NOTE: Data provided by Felix Elsen, Philips (11 May 2001). The practical tests were performed with product designed for the US market, but using different parameters (e.g. Operating frequency). C.2 Current ITU allocations In Region 1: FIXED, MOBILE except aeronautical mobile. C.3 Sharing issues The compatibility studies should take into account SRD services and applications mentioned in the Strategic Plan for the use of SRD applications within the band 862 MHz to 870 MHz as adopted by the WG FM including spread spectrum systems across the band 863 MHz to 870 MHz. ETSI ETSI TR 101 445 V1.1.1 (2002-04) 14 Annex D: Item check list 1) Approval by TG28. 2) Review and Decision of System Reference Document ERM-RM_10(01)118R4 by ETSI TC ERM RM. 3) If accepted, Liaison Statement to SRD MG from ETSI TC ERM. 4) If accepted by SRD-MG, liaison statement to SE 24 to provide detailed compatibility studies. 5) Completion of compatibility studies. 6) Generate appropriate ETSI documentation. 7) Amend CEPT/ERC/REC 70-03 [1] annex 1. ETSI ETSI TR 101 445 V1.1.1 (2002-04) 15 Annex E: Bibliography Directive 1999/5/EC of the European Parliament and of the Council of 9 March 1999 on radio equipment and telecommunications terminal equipment and the mutual recognition of their conformity (R&TTE Directive). ETSI ETR 028: "Radio Equipment and Systems (RES); Uncertainties in the measurement of mobile radio equipment characteristics". ISO/IEC 18000-6: "Information Technology AIDC Techniques-RFID for Item Management Air Interface; Part 6: q Parameters for Air Interface Communications at UHF frequencies". ETSI ETSI TR 101 445 V1.1.1 (2002-04) 16 History Document history V1.1.1 April 2002 Publication |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 1 Scope | The present document applies to IP networks that provide voice telephony in accordance with any of the TIPHON Scenarios. The objective with the present document is to collect all results of various VoIP speech transmission quality tests and related information. This collection should be used for information and to review and discuss the values of the TIPHON QoS classes which are described in WG5 documents TR 101 329-1 [3] and TR 101 329-7 [6]. The separate measurements should give a very good opportunity to understand the goal of the measurement itself and the exact measurement Set Up conditions to understand under which framework the measurements were done. The present document covers measurement results provided to TIPHON during the years 1999 to 2001, which have contributed to the measurement methodologies in 101 329-5 [5] as well as providing design parameters in 101 329-7 [6]. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 2 References | For the purposes of this Technical Report (TR) the following references apply: [1] ETSI ETR 275 (1996): "Transmission and Multiplexing (TM); Considerations on transmission delay and transmission delay values for components on connections supporting speech communication over evolving digital networks". [2] ETSI TR 101 329 (V2.1.1): "Telecommunications and Internet Protocol Harmonization Over Networks (TIPHON); General aspects of Quality of Service (QoS)". [3] ETSI TR 101 329-1: "Telecommunications and Internet Protocol Harmonization Over Networks (TIPHON) Release 3; End-to-end Quality of Service in TIPHON systems; Part 1: General aspects of Quality of Service (QoS)". [4] ETSI TR 101 329-2: "Telecommunications and Internet Protocol Harmonization Over Networks (TIPHON) Release 3; End-to-end Quality of Service in TIPHON systems; Part 2: Definition of speech Quality of Service (QoS) classes". [5] ETSI TR 101 329-5: "Telecommunications and Internet Protocol Harmonization Over Networks (TIPHON) Release 3; End-to-end Quality of Service in TIPHON systems; Part 5: Quality of Service (QoS) measurement methodologies". [6] ETSI TR 101 329-7: "Telecommunications and Internet Protocol Harmonization Over Networks (TIPHON) Release 3; End-to-end Quality of Service in TIPHON systems; Part 7: Design guide for elements of a TIPHON connection from an end-to-end speech transmission performance point of view". [7] ETSI ES 201 168 (V1.1.1): "Corporate Telecommunication Networks (CN); Transmission characteristics of digital Private Branch eXchanges (PBXs)". [8] ETSI GTS 06.10 (V3.2.0): "European digital cellular telecommunications system (Phase 1); GSM Full Rate Speech Transcoding (GSM 06.10)". [9] ITU–T Recommendation G.107: "The E-Model, a computational model for use in transmission planning". [10] ITU–T Recommendation G.108: "Application of the E-model: A planning guide". [11] ITU–T Recommendation G.109 (1999): "Definition of categories of speech transmission quality". [12] ITU–T Recommendation G.113 (2001): "Transmission impairments due to speech processing". [13] ITU–T Recommendation G.131 (1996): "Control of talker echo". [14] ITU–T Recommendation G.165 (1993): "Echo cancellers". ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 8 [15] ITU–T Recommendation G.168 (2000): "Digital network echo cancellers". [16] ITU–T Recommendation G.711 (1988): "Pulse code modulation (PCM) of voice frequencies". [17] ITU–T Recommendation G.721 (1988): "32 kbit/s adaptive differential pulse code modulation (ADPCM)". [18] ITU–T Recommendation G.723.1 (1996): "Dual rate speech coder for multimedia communications transmitting at 5.3 and 6.3 kbit/s". [19] ITU–T Recommendation G.726 (1990): "40, 32, 24, 16 kbit/s adaptive differential pulse code modulation (ADPCM)". [20] ITU–T Recommendation G.727 (1990): "5-, 4-, 3- and 2-bit/ sample embedded adaptive differential pulse code modulation (ADPCM)". [21] ITU–T Recommendation G.728 (1992): "Coding of speech at 16 kbit/s using low-delay code excited linear prediction". [22] ITU–T Recommendation G.729 (1996): "Coding of speech at 8 kbit/s using conjugate-structure algebraic-code-excited linear-prediction (CS-ACELP)". [23] ITU–T Recommendation G.729A (Annex A - 1996): "Reduced complexity 8 kbit/s CS-ACELP speech codec". [24] ITU–T Recommendation G.729B (Annex B - 1996): "A silence compression scheme for G.729 optimized for terminals conforming to Recommendation V.70". [25] ITU–T Recommendation H.323 (2000): "Packet-based multimedia communications systems". [26] ITU-T Recommendation P.57: "Artificial ears". [27] ITU-T Recommendation P.58: "Head and torso simulator for telephonometry". [28] ITU–T Recommendation P.64 (1999): "Determination of sensitivity/frequency characteristics of local telephone systems". [29] ITU-T Recommendation P.501: "Test signals for use in telephonometry". [30] ITU-T Recommendation P.502: "Objective test methods for speech communication systems, using complex test signals". [31] ITU-T Recommendation P.581: "Use of head and torso simulator (HATS) for hands-free terminal testing". [32] ITU–T Recommendation P.800 (1996): "Methods for subjective determination of transmission quality". [33] ITU–T Recommendation P.861 (1998): "Objective quality measurement of telephone-band (300-3 400 Hz) speech codecs". [34] ITU–T Recommendation Q.13/12: "Rapporteur of Question 13 inside ITU-T Study Group 12; ETSI TIPHON 17TD135". [35] SG 16, Santiago, Chile, 17-28 May 1999; D.249 (WP 3/16): "A High Quality Low-Complexity Algorithm For Frame Erasure Concealment (FEC) With G.711" (Source: AT&T). [36] T1A1.7/99-012r3; Jul-28-1999: "Draft Proposed American National Standard - A Packet Loss Concealment Technique for Use with ITU-T Recommendation G.711" (Source: AT&T). [37] ETSI ETS 300 581-2: "Digital cellular telecommunications system (Phase 2) (GSM); Half rate speech; Part 2: Half rate speech transcoding (GSM 06.20 version 4.3.1)". [38] ETSI EN 300 726: "Digital cellular telecommunications system (Phase 2+) (GSM); Enhanced Full Rate (EFR) speech transcoding (GSM 06.60 version 8.0.1 Release 1999)". ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 9 [39] ITU-T Recommendation P.862 (2001): "Perceptual evaluation of speech quality (PESQ), an objective method for end-to-end speech quality assessment of narrowband telephone networks and speech codecs". [40] ITU-T Recommendation G.723: "Extensions of Recommendation G.721 adaptive differential pulse code modulation to 24 and 40 kbit/s for digital circuit multiplication equipment application". [41] ITU-T Recommendation G.723.1-A: "Speech coders : Silence compression scheme". [42] ITU-T Recommendation P.50: "Artificial voices". [43] ITU-T Recommendation G.114: "One-way transmission time". |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 3 Abbreviations | For the purposes of the present document, the following abbreviations apply: ACR Absolute Category Rating ASL Active Speech Level CAS Communication Analysis System (HEAD acoustics test system) CSS Composite Source Signal EC Echo Canceller EP Error Pattern GSM Global System for Mobile communications GSM EFR GSM Enhanced Full Rate Speech Coder GSM FR GSM Full Rate Speech Coder HATS Head And Torso Simulator IP Internet Protocol IRS Intermediate Reference System ISDN Integrated Services Digital Network JLR Junction Loudness Rating LAN Local Area Network MNRU Modulated Noise Reference Unit MOS Mean Opinion Score NLP Non-Linear Processor OLR Overall Loudness Rating OVL Over-Load Point PESQ Perceptual Evaluation of Speech Quality (see ITU-T Recommendation P.862) PLC Packet Loss Concealment PSTN Public Switched Telephone Network PVS PC Voice Switch QoS Quality of Service RLR Receive Loudness Rating SCN Switched Communications Network SLR Send Loudness Rating TMOS TOSQA Mean Opinion Score (output of TOSQA) TOSQA Telecommunication Objective Speech Quality Assessment VAD Voice Activity Detection ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 10 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 4 List of Measurement Results | Table 1: List of measurement results Nr. Document Source Document Introduction Date 1 Simulation Results of VoIP scenarios Deutsche Telekom Berkom t.scheerbarth@berkom.de; i.kliche@berkom.de ETSI TIPHON 11TD064 11/01/1999 2 APPENDIX I (to ITU-T Recommendation G.113 [12] Mark E. Perkins mperkins@att.com ETSI TIPHON 11TD084 11/01/1999 3 Speech Quality Test results of IP equipment in a LAN environment Robert Bosch GmbH Joachim.Pomy@Tenovis.com ETSI TIPHON 14TD081 16/07/1999 4 QoS Measurements of IP-Configurations HEAD acoustics, Robert Bosch GmbH, T-Nova (Deutsche Telekom) h.w.gierlich@head-acoustics.de ETSI TIPHON 15TD089 05/10/1999 5 Subjective Results on impairment effects of IP packet loss Nortel Networks paulcov@nortelnetworks.com ETSI TIPHON 17TD167 14/03/2000 6 Subjective and Objective Speech Quality Evaluation on Speech Data recorded at the SuperOp 99 event in Hawaii Rapporteur of ITU-T Recommendation Q.13/12 [34] ETSI TIPHON 17TD135 15/03/2000 7 Anonymous Test report of ETSI Speech Quality Test Event 2000 Deutsche Telekom, T-Nova; HEAD acoustics ESTI TIPHON 22TD38 26/03/2001 8 Problems with the behaviour of jitter buffers and their influence on the end-to-end speech quality Pieter Veenstra p.k.veenstra@kpn.com ESTI TIPHON 22TD47 26/03/2001 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5 General Measurement Results | |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1 Subjective Testing | |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.1 Simulation Results of VoIP Scenarios | Source: Deutsche Telekom Berkom; Simulation Results of VoIP Scenarios; ETSI TIPHON 11TD064. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.1.1 Introduction | ETSI TIPHON WG 5 has defined a methodology for testing VoIP End-to-End speech quality. This methodology was used as a basis model for the T Berkom simulation processing. Figure 2 shows the methodology used for simulation. Reference Codec G.711, G.726, GSM-FR, G.729 TIPHON Network Acoustic Part Electrical Part Terminal Reference Acoustic Device Test Point Subjective Assessment recorded speech database ITU-T speech signals Electrical Part Acoustic Part Terminal Simulation Figure 2: Simulation methodology for testing TIPHON speech quality ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 11 A set of speech signals designed according to ITU-T Recommendation P.800 [32] was used as input of the simulation path. The simulation path includes the terminal side (electrical part) and the network itself. The influence of the terminal side was focussed to the speech conversion and IP packet size issue. The influence of the network side was simulated by different packet loss rates. After the simulation the speech samples were recorded and stored in a database. The subjective assessment was carried out according to the ITU-T Recommendation P.800 [32] method. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.1.2 Measurement Set Up | |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.1.2.1 Basics | All source speech samples consisted of German sentences spoken by four talkers (2 male, 2 female). The input level taken for all scenarios was ASL = -26 dB Ovl. In the pre- and post-processing phase the speech samples were filtered with the modified IRS transmit and receive filter. For every test condition the speech file was encoded and then assembled in IP packets. These IP packets were assembled with different lengths, according to the concerned speech frame number per packet. For simulation of network influences in the case of packet loss, a common channel model was designed, realized by channel files which describe the network condition with the same time resolution as the source speech sample rate. So the network has a certain condition (good or bad) for every speech sample (every 125 µs), two adjacent network states were considered as statistically independent because the network speed was assumed to be much higher than the sample rate (8 000 samples per second). So for each packet loss rate one channel file was created using a random generator. The length of this channel file was exactly the same as the length of the speech file. In a further step the speech file, assembled in IP packets, was matched to the channel file. According to the length of the IP packet (10 ms, 20 ms,...) the channel file was checked every time when a packet was ready to send. That means if the packet size was 10 ms the channel file was checked also every 10 ms if the condition is good or bad. In a bad case the IP packet was lost, otherwise it was further processed. This information (IP packet lost or not) was stored in a description file which was the input of the re-assembler and speech decoder. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.1.2.2 Test Cases | The test cases consisted a group of single codec scenarios (references), phone to phone scenarios in fixed network environments and a group of tandeming conditions. The tandeming conditions based on real scenarios where a mobile customer is connected to an ordinary telephone via IP. For this cases the GSM Full Rate Codec (GSM-FR) and the GSM Enhanced Full Rate Codec (EFR) were used. In such scenarios mainly the influence of the IP network was taken into account. Only one condition was chosen to simulate a voice transmission from a mobile phone to an ordinary telephone via an impaired radio channel and via an IP network with packet loss. Table 2: G.711 [16] + Codec + G.7xx (Phone to Phone Scenario in fixed network environments) Codec Packet Loss Speech Frame Size Nr. of Frames/Packet Substitution G.711 [16] 5 %, 10 %, 15 %, 20 % 0,125 ms 80, 320, 480, 800 Silence G.729B [24] 5 % 10 ms 1, 4, 6, 10 G.729 [22] internal G.723.1 [18] (5.3) 0 %, 5 %, 10 %, 15 %, 20 % 30 ms 1, 2, 3 G.723.1 [18] internal G.728 [21] 0 %, 5 %, 10 %, 15 %, 20 % 0,625 ms 16, 64, 96, 160 proprietary ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 12 ISDN / PSTN IP G.711- Encoder G.72x Decoder G.711 Decoder packet loss TIPHON network class Encoder G.72x G.711 Enc + Dec simulation scenario: logical scenario: Decoder G.72x packet loss generator source destination ISDN / PSTN G.711 Encoder G.711- Decoder G.72x Encoder G.711 Enc + Dec Pre- Processing Post- Processing Figure 3: Processing scenario for single codec conditions Table 3: Tandem Configuration with GSM x FR (Phone to Phone Scenario in including mobile networks) Tandem with GSM-FR Packet Loss Speech Frame Size Nr. of Frames/ Packet Substitution GSM FR + G.723.1 [18] (5.3) 0 %, 5 %, 10 %, 15 % G.723.1 [18]: 30 ms 2 G.723.1 [18] internal GSM FR + G.729B [24] 0 %, 5 %, 10 %, 15 % G.729 [22]: 10 ms 6 G.729 [22] internal GSM EFR + G.723.1 [18] (5.3) 0 %, 5 %, 10 % G.723.1 [18]: 30 ms 2 G.723.1 [18] internal GSM EFR + G.729B [24] 0 %, 5 %, 10 % G.729 [22]: 10 ms 6 G.729 [22] internal GSM EFR EP2 + G.723.1 [18] (5.3) 0 %, 5 %, 10 % G.723.1 [18]: 30 ms 2 G.723.1 [18] internal ISDN / PSTN IP G.711- Encoder G.72x Decoder G.711 Decoder packet loss TIPHON network class Encoder G.72x G.711 Enc + Dec simulation scenario: logical scenario: Decoder G.72x packet loss generator source destination ISDN / PSTN G.711 Encoder G.711- Decoder G.72x Encoder G.711 Enc + Dec Pre- Processing Post- Processing 12345 GSM xFR Enc + Dec GSM xFR Enc + Dec Figure 4: Processing scenario for tandem codec conditions ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 13 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.1.3 Results | The following figures illustrate the subjective assessment results. On the y axis the MOS score from 1 (bad) to 5 (excellent) is shown. The x-axis shows the various end-to-end scenarios. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.1.3.1 G.711 + Codec + G.711 | 0,00 0,50 1,00 1,50 2,00 2,50 3,00 3,50 4,00 4,50 5,00 5 10 20 5 10 20 5 10 20 5 10 20 0 0 5 10 20 5 10 20 5 10 20 0 0 5 10 5 10 5 10 5 10 ref ref ref ref 5 10 15 20 25 30 99 Packet Loss % IP Packet (ms) Codec 10 40 60 100 G.711 30 60 90 G.711-G.723(5.3)-G.711 G.711-G.728-G.711 10 40 60 100 MNRU G.711 G.729 1x ADPCM32 4x ADPCM32 G.723(5.3) G.728 Figure 5: Subjective assessment result The assessment of single codecs under the influence of packet loss leads to assumptions as follows: • the packet loss rate of 5 % seems to be almost the quality threshold of MOS 3,0; • in all test cases the evaluation of voice signals with packet loss of >= 10 % the MOS scores are widely below the quality threshold of MOS 3,0. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.1.3.2 Tandem Conditions with GSM x FR | 0,00 0,50 1,00 1,50 2,00 2,50 3,00 3,50 4,00 4,50 5,00 0 5 10 0 5 10 0 0 0 5 10 0 5 10 0 5 10 ref ref ref 5 10 15 20 25 30 99 Packet Loss % Codec GSM FR- G.723(5.3) GSM FR- G.729 GSM EFR- G.723(5.3) GSM EFR- G.729 GSM EFREP2- G.729 GSM EFR GSM FR G.711 1x ADPCM 32 4x ADPCM 32 MNRU IP Packet: 60ms Figure 6: Tandem conditions with GSM x FR ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 14 The assessment of tandem codecs under the influence of packet loss for the G.7xx codecs, leads to assumptions as follows: • the quality threshold for tandem connection of GSM FR and G.723.1 [18] can be seen with less than or max. 5 % packet loss, under the precondition of an error free radio channel; • the quality threshold for tandem connection of GSM FR and G.729 [22] can be seen in the range of << 5 % packet loss, under the precondition of an error free radio channel; • the quality threshold for tandem connection of GSM EFR and G.723.1 [18] can be seen in the range of 5 % packet loss, under the precondition of an error free radio channel. No acceptable speech transmission quality for tandem connection of GSM EFR and G.723.1 [18] for 0 % packet loss can be provided if the radio channel induces errors. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.2 Subjective Results on impairment effects of IP packet loss | Source: Nortel Networks; ETSI TIPHON 17TD167, Subjective Results on impairment effects of IP packet loss. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.2.1 Introduction | With the growing interest in voice transport over Internet (IP) networks (voice over IP or VoIP), it is important to understand the effects of various impairments on voice quality. One of the important impairments is packet loss, which can be produced when voice packets are lost or delivered too late to be useful. A subjective experiment was conducted to investigate the effects of packet loss on voice quality. This experiment included a variety of codecs that are used in VoIP applications and packet loss rates ranging from 0 % to 5 %. Loss rates greater than 5 % were not included in this study because based on previous experience such large impairments in voice quality tend to skew subjective data. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.2.2 Measurement Set Up | |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.2.2.1 Subjects | Sixty-one listeners aged 16 to 68 years (mean age 37 years) participated in the experiment. All listeners were telephony users with self-reported normal hearing drawn from Nortel Network's Subjective Assessment Lab subject pool. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.2.2.2 Speech Processing | The source speech consisted of high-quality anechoic chamber recordings of North American English sentences spoken by six talkers (3 male, 3 female). Each speech sample consisted of four sentences uttered by the same talker. All the speech samples were transmit filtered before encoding and receive-filtered prior to being heard by listeners. The input level of the speech signals to the codecs was -20 dBm0. For non-G.711 [16] codec conditions, the codecs received G.711 [16] encoded/decoded speech as the input speech. The codecs tested in this experiment were G.711 [16], G.729 [22], G.729A [23], G.723.1 [18], and GSM EFR (06.60) [38]. For G.711 [16] and G.729 [22], speech frame sizes of 10 ms and 20 ms were tested, while G.723.1 [18] and GSM EFR were tested with their standard frame sizes (30 ms and 20 ms respectively). For all the codecs with a Voice Activity Detection (VAD) feature (G.729 [22], G.729A [23]. GSM EFR, G.723.1 [18] ) the VAD was set to "OFF". In addition, two Packet Loss Concealment (PLC) schemes for G.711 [16] were tested: one scheme described by AT&T in a submission to SG 16 [35] and proposed as an ANSI standard and a second proprietary scheme developed by Nortel Networks. For the packet loss conditions, frames were removed from the speech samples randomly with a frequency determined by the test condition (e.g. 1 % of the frames). A voice activity detector was used to ensure that losses always occurred during an active speech period. It is important to note that this technique for simulating packet loss may be different from other studies, so cross-experiment comparisons should be done with caution. A different random mask file was applied to the speech samples for each group of 3 listeners in order to randomize where the losses were occurring during the 4-sentence sample. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 15 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.2.2.3 Procedure | Samples were played back over one channel (one side) of high-fidelity headphones to simulate handset listening. The samples were played back at 79 dB SPL, measured at the ear reference plane. Listeners heard one sample during each trial, and entered their ratings by pressing a button on a response box. The order of presentation of the samples was randomized and the ratings were stored in a file for later statistical analysis. Eight additional samples, which were not counted in the data, were presented at the beginning of the session as a warm-up. All subjects heard and rated all the conditions presented by six talkers, resulting in a total of 366 observations per condition. Listeners rated the processed speech samples in an Absolute Category Rating (ACR) test. Samples were rated according to the telephony 5-point scale (excellent, good, fair, poor, bad). Mean-opinion-scores (MOS) were computed from the ratings assigned to each test case. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.2.2.4 MOS Data and MNRU Conditions | The MOS and MNRU equivalence data for each of the conditions are summarized in clause 5.1.2.3. The results of various conditions of interest are also summarized and discussed below. Figure 7 shows the MOS values as a function of the Q parameter used in the MNRU conditions. This function is typical of what is seen in many subjective tests of voice quality, and a curve fit using the standard MNRU equation and the parameters A = 1,6; B = 1,6; C = 9,5; Qm = 17,3 achieved an R2 value of 0,999. 1 2 3 4 5 0 10 20 30 40 Q MOS Figure 7: MOS as a function of Q in the MNRU conditions |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.2.3 Results | |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.2.3.1 Reference Codec Conditions | Figure 8 summarizes the MOS ratings for the codec conditions without packet loss impairments. These results are also typical of VoIP quality experiments, with G.711 [16] and GSM EFR receiving the highest ratings and G.723.1 [18] receiving the lowest rating. The only unexpected result was that the G.729A [23] score is slightly higher than the G.729 [22] (3,78 vs. 3,65), which is inconsistent with previous findings that show that G.729A [23] is rated as slightly below or equal to G.729 [22]. The difference in the current experiment is small and only occurred for the 0 % packet loss conditions. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 16 1 2 3 4 5 G.711 G.729 G.729 A G.723.1 GSM EFR Reference Codec MOS Figure 8: MOS values for each codec without packet loss impairments |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.2.3.2 Packet Loss Conditions | Figure 9 shows the effects of packet loss on MOS ratings for various codecs when the packet size was set to 10 ms. It can be seen that the quality of G.711 [16] (without PLC) deteriorates rapidly with increasing loss, while G.729 [22] and G.729A [23] are more robust to packet loss. This is due to PLC (frame erasure concealment) schemes built into the G.729 [22] codecs. The two PLC schemes for G.711 [16] tested in this experiment (algorithms produced by AT&T and Nortel Networks) had large and similar beneficial effects in preserving voice quality with packet loss. The results for 20 ms (and larger) packets are shown in figure 10. Again, G.711 [16] quality drops dramatically with increasing packet loss and this can be preserved using either of the PLC schemes. The new codecs in this figure are GSM EFR, which is somewhat resistant to packet loss, and G.723.1 [18], which deteriorates rapidly with increasing loss similar to the G.729 [22] codecs. 10 msec 1 2 3 4 5 0 1 2 3 4 5 Packet Loss Percentage MOS G.729 G.729 A G.711 G.711 Nortel PLC G.711 AT&T PLC Figure 9: Effects of packet loss on voice quality with 10 ms packets ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 17 20+ msec 1 2 3 4 5 0 1 2 3 4 5 Packet Loss Percentage MOS G.711 G.711 Nortel PLC G.711 AT&T PLC G.729 G.729 A G.723.1 (30 msec) GSM EFR NOTE: All packet sizes were 20 ms except G.723.1 [18], which was 30 ms. Figure 10: Effects of packet loss on voice quality with 20 ms or larger packets |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.1.2.3.3 Tables of Results | The parameters for the MNRU curve fit were: A = 1,6; B = 1,6; C = 9,5; Qm = 17,3. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 18 Table 4: MOS and MNRU equivalence scores Packet PLC Packet MNRU (ms) Codec bps Algorithm Loss MOS Q 10 G.729 [22] (8 kbit/s) included 1 % 3,50 23,37 10 G.729 [22] (8 kbit/s) included 2 % 3,38 22,34 10 G.729 [22] (8 kbit/s) included 3 % 3,05 20,08 10 G.729 [22] (8 kbit/s) included 5 % 2,67 17,70 10 G.729A [23] (8 kbit/s) included 1 % 3,37 22,27 10 G.729A [23] (8 kbit/s) included 2 % 3,10 20,40 10 G.729A [23] (8 kbit/s) included 3 % 2,85 18,78 10 G.729A [23] (8 kbit/s) included 5 % 2,57 17,13 10 G.711 [16] u-law (64 kbit/s) none 1 % 3,41 22,57 10 G.711 [16] u-law (64 kbit/s) none 2 % 2,82 18,60 10 G.711 [16] u-law (64 kbit/s) none 3 % 2,41 16,18 10 G.711 [16] u-law (64 kbit/s) none 5 % 2,00 13,55 10 G.711 [16] u-law (64 kbit/s) Nortel PLC 1 % 4,02 30,83 10 G.711 [16] u-law (64 kbit/s) Nortel PLC 2 % 3,97 29,45 10 G.711 [16] u-law (64 kbit/s) Nortel PLC 3 % 3,93 28,61 10 G.711 [16] u-law (64 kbit/s) Nortel PLC 5 % 3,49 23,23 10 G.711 [16] u-law (64 kbit/s) AT&T PLC 1 % 4,00 30,16 10 G.711 [16] u-law (64 kbit/s) AT&T PLC 2 % 3,99 30,03 10 G.711 [16] u-law (64 kbit/s) AT&T PLC 3 % 3,86 27,43 10 G.711 [16] u-law (64 kbit/s) AT&T PLC 5 % 3,63 24,60 20 G.729 [22] (8 kbit/s) included 1 % 3,38 22,34 20 G.729 [22] (8 kbit/s) included 2 % 3,18 20,89 20 G.729 [22] (8 kbit/s) included 3 % 2,77 18,28 20 G.729 [22] (8 kbit/s) included 5 % 2,40 16,13 20 G.729A [23] (8 kbit/s) included 1 % 3,32 21,90 20 G.729A [23] (8 kbit/s) included 2 % 3,00 19,74 20 G.729A [23] (8 kbit/s) included 3 % 2,87 18,89 20 G.729A [23] (8 kbit/s) included 5 % 2,39 16,07 20 G.711 [16] u-law (64 kbit/s) None 1 % 3,33 21,98 20 G.711 [16] u-law (64 kbit/s) None 2 % 2,83 18,66 20 G.711 [16] u-law (64 kbit/s) None 3 % 2,50 16,69 20 G.711 [16] u-law (64 kbit/s) None 5 % 1,92 12,99 20 G.711 [16] u-law (64 kbit/s) Nortel PLC 1 % 4,06 31,96 20 G.711 [16] u-law (64 kbit/s) Nortel PLC 2 % 3,82 26,82 20 G.711 [16] u-law (64 kbit/s) Nortel PLC 3 % 3,59 24,14 20 G.711 [16] u-law (64 kbit/s) Nortel PLC 5 % 3,33 21,96 20 G.711 [16] u-law (64 kbit/s) AT&T PLC 1 % 4,05 31,76 20 G.711 [16] u-law (64 kbit/s) AT&T PLC 2 % 3,86 27,34 20 G.711 [16] u-law (64 kbit/s) AT&T PLC 3 % 3,62 24,44 20 G.711 [16] u-law (64 kbit/s) AT&T PLC 5 % 3,47 23,07 20 GSM EFR (12,2 kbit/s) included 1 % 3,80 26,48 20 GSM EFR (12,2 kbit/s) included 2 % 3,57 23,94 20 GSM EFR (12,2 kbit/s) included 3 % 3,36 22,23 20 GSM EFR (12,2 kbit/s) included 5 % 3,01 19,80 30 G.723.1 [18] (6,3 kbit/s) included 1 % 3,42 22,64 30 G.723.1 [18] (6,3 kbit/s) included 2 % 3,05 20,08 30 G.723.1 [18] (6,3 kbit/s) included 3 % 2,67 17,74 30 G.723.1 [18] (6,3 kbit/s) included 5 % 2,30 15,47 MNRU n/a 6 dB n/a 1,26 MNRU n/a 12 dB n/a 1,81 MNRU n/a 18 dB n/a 2,66 MNRU n/a 24 dB n/a 3,59 MNRU n/a 30 dB n/a 3,90 MNRU n/a 36 dB n/a 4,11 G.729 [22] included none 3,65 24,74 G.729A [23] included none 3,78 26,26 G.711 [16] u-law none none 4,15 36,75 GSM EFR included none 4,13 35,43 G.723.1 [18] included none 3,57 23,94 ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 19 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2 Objective Testing | |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2.1 Objective Results of SuperOp event 1999 in Hawaii | Source: Rapporteur of ITU-T Recommendation Q.13/12 [34]; ETSI TIPHON 17TD135. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2.1.1 Introduction | This clause presents results from the ITU-T Recommendation Q.13/12 [34] evaluation for a new Recommendation for objective speech quality assessment algorithm, carried out on speech material that was recorded in a joint QoS measurement initiative between ETSI EP TIPHON and ITU-T Recommendation Q.13/12 [34] at the SuperOp Interoperability event on Hawaii in July 1999. Due to former agreements and restrictions for publishing individual data before finalization of the new ITU-T Recommendation, subjective scores as well as anonymized objective scores are shown. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2.1.2 Measurement Set Up | For speech input into the simulator as well as for recording of degraded speech, the CAS (Communication Analysis System) from HEAD-acoustics was used. It was kindly provided by Robert BOSCH GmbH. The CAS provides playback and recording of stereo signals. To allow delay measurement, the input speech signal (reference) was directly recorded by the left input channel. The transmitted signal (degraded speech signal) was recorded on the right channel. Due to technical problems and the fact that the participants of the SuperOp had scheduled very tough timeslots for all their tests, we have been able to only test one physical scenario with one set of equipment. To collect the speech files the TIPHON scenario 2 was Set Up. Telephone Sound Generator Network Simulator S W I T C H POTS Gateway A Recorder H.323 Endpoint CAS-System Figure 11: Recording of speech samples with TIPHON scenario 2 After recording of the speech samples, the 32 s files were separated into samples of approximately 8 s in order to allow per-talker-evaluations. In addition to the processed conditions, 7 MNRU reference conditions were used in order to ensure a proper instruction and training of the test subjects. Furthermore, MNRU references are used for further comparison and quality management. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 20 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2.1.3 Quality Assessment | Within the ITU-T Recommendation Q.13/12 [34] model evaluation, subjective tests on the Hawaii database were performed at the subjective quality test laboratories at T-Nova Deutsche Telekom Innovationsgesellschaft mbH in Berlin. The tests were designed according to ITU-T Recommendation P.800 [32]. Each of the 208 speech samples was assessed by 24 subjects. The listener's group consists of 12 female and 12 male persons of normal hearing with no specific knowledge in speech processing and speech quality assessment. Normal hearing capability was proven by pure- tone audiometry within 125 Hz to 8 000 Hz. The objective assessment procedure was done with 5 candidates for the new ITU-T Recommendation. The result of every objective method was compared against the subjective results and against the existing ITU-T Recommendation P.861 [33]. The correlation between the subjective and objective assessment methods for all candidates can be seen in table 5. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2.1.4 Results | Within the competition of objective speech quality measurement algorithms of ITU-T Recommendation Q.13/12 [34], a large speech database with more than 2 100 speech samples was used. The database that was produced during the Hawaii QoS measurement initiative contributed one of 22 experiments to this evaluation. The objective estimates were compared against the subjective measures from the Hawaii test, and correlation coefficients were derived for each model. Table 1 shows the correlation coefficients of the objective speech quality assessment models. In fact, only one VoIP test bed was used at the Hawaii QoS measurement initiative. Because of the good homogeneity of the recording scenario, the correlation coefficients of the best models exceed 95 % - a very good measurement result for unknown data. Table 5: Correlation coefficient for the ITU-T Recommendation Q.13/12 [34] objective speech quality assessment algorithm Rank order Correlation coefficient 1 0,9879 2 0,9859 3 0,9489 4 0,9275 5 0,8377 P.861 [33] 0,8706 Table 5 shows clearly that the P.861 [33] algorithm is less accurate then other instrumental assessment methods. P.861 [33] is withdrawn by ITU-T in 2001. It has been replaced by P.862 [39] in 2001. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2.2 Objective Results of ETSI Speech Quality Test Event 2000 | Source: Anonymous Test Report of the 1st ETSI Speech Quality Test Event; T-Nova and HEAD acoustics, 22TD38. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2.2.1 Introduction | This clause describes the test methodologies and the objectice results of the measurements which were carried out during the 1st ETSI Speech Quality Test Event. The tests were conducted by T-Nova Deutsche Telekom Innovations- gesellschaft mbH Berkom (T-Nova) in collaboration with HEAD acoustics GmbH (HEAD acoustics). The aim of the test event was to determine the speech quality of various voices over IP equipment under certain IP network conditions. During the test event, speech material as well as measurement data were collected by transferring voice samples and artificial signals across the Voice over IP Set Up. This material was analysed and the results are reported in this chapter. The one-way speech transmission quality was evaluated by processing real speech samples and analysing it using the TOSQA algorithm. TOSQA leads to MOS-comparable results. To validate the TOSQA results an auditory reference test was carried out. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 21 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2.2.2 Measurement Set Up | Measurements for one way speech transmission quality were mainly conducted by "electrical-electrical" measurements. For the "electrical-electrical" measurements two kinds of input signals were used: a) speech samples designed according to ITU-T Recommendation P.800 [32]; and b) test signals according to P.501 [29]. The input signals were transmitted and recorded simultaneously, i.e. the sending and receiving process were started at the same time. Therefore exact delay measurements were possible. For all kind of measurements a packet loss generator (NIST Net) and a packet loss monitor were used. Both entities were controlled by the PVS measurement equipment. The measurement set up is shown in figure 12. Gateway IP T1 / E1 / BRI NISTNet Packet MONITOR T1 / E1 / BRI Packet Loss DELAY Output HEAD Acoustics Testsystem CAS T-Nova Testsystem PVS Input Gateway IP IP ISDN DSS1 ISDN DSS1 ALCATEL PABX 4400 Figure 12: Electrical – Electrical Measurement Set Up For all kinds of "electrical-electrical" measurements the following IP network conditions were used. Table 6: Network Conditions for Electrical-Electrical Measurements Condition Packet Loss (Equal) Additional Delay (see note) Delay Variation 1 0 0 No 2 1 % 0 No 3 2 % 0 No 4 3 % 0 No 5 5 % 0 No 6 1 % 50 ms 20 ms NOTE: Additional IP network delay was introduced by NIST Net. The additional delay in condition 6 was intended to have a fix delay value in order to ensure proper delay variation (jitter) handling. In such jitter condition the test network can cause situations where packets are reordered, if the packet size is very small. This effect can be avoided with high probability by using a certain packet length, which should be at least three times higher than the delay variation itself. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 22 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2.2.3 Quality Assessment | The speech quality assessment of the speech samples was performed using the method for instrumental speech quality estimation of Deutsche Telekom, TOSQA. For verifying the instrumental results auditory reference assessments were performed using speech material in German. The auditory test was performed in the Berlin speech quality labs at T-Nova Deutsche Telekom Innovationsgesellschaft mbH and was carried out according to ITU-T Recommendation P.800 [32]. Each circuit condition was tested with four different speech samples (short, concatenated sentences, read by 2 male and 2 female speakers each). Speech samples were played back to the subjects via a conventionally shaped standard telephone handsets in a low-noise test cabinet (room noise floor < 30 dB(A)), at a listening level of about 79 dB SPL. Test subjects were required to judge the overall listening quality on a 5-point ACR quality scale, as recommended in ITU-T Recommendation P.800 [32], with German scale labels. Before starting the first test session, 8 example speech samples were played back to the test subjects, which were expected to range in quality from good to bad. In addition to the tested VoIP scenarios also 11 reference conditions were integrated in the test. These are single speech codecs (G.723.1 [18], G.729 [22]) as well as different MNRU conditions. To avoid effects caused by presentation order, groups of test subjects listen to the samples in different randomized orders. All in all 25 test subjects took part on the auditory test. Four votes per each condition and test subject yield the Mean Opinion Sore (MOS) for the condition under test. So the MOS per condition is an average of 100 single votes. In the similar manner as subjects assess the speech quality in the listening test, TOSQA rates the four speech samples for each condition. Here exactly the same speech material from the auditory test was used. TOSQA compares this transmitted and possibly distorted speech material with the clean input speech material as reference. The four TOSQA results (TMOS) of the four speech samples were averaged and are the basis for comparison with the MOS values gained by the auditory test. The graphic in figure 13 plot shows mean opinion scores gained by auditory experiments on the X axis. The Y axis shows the speech quality value TMOS calculated by TOSQA. The TMOS values were transformed in a common way by a third order monotonous mapping. In an ideal prediction of quality, the instrumental results would be equal to the MOS-values. In this case all symbols would be on the 45° line in the diagram. All in all TOSQA shows a good prediction of speech quality, the correlation coefficient after third order mapping is 0,916. This result is based on the four German samples which are used in the auditory reference test. In addition to the four German samples, also four English samples were transmitted via each network condition, followed by another 8 German samples; all of them were assessed by TOSQA. The resulting mean TMOS averaged over all 16 samples was also compared with MOS values gained by the auditory test. In this case the correlation coefficient between MOS and "mean TMOS" increases slightly to 0,921. This value should be also compared with the requirement which was defined in the ITU-T Q13/12 for such kind of conditions (cf. ITU-T document COM12-117E, March 2000). During the competition for the new standard for objective speech quality measurement a minimum correlation coefficient of 0,80 was required for real VoIP recordings if the test corpus was not used for training the objective model. This is exactly the same situation we had in our VoIP test here. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 23 1.0 2.0 3.0 4.0 5.0 1.0 2.0 3.0 4.0 5.0 auditory MOS TMOS Results for auditory reference test Figure 13: MOS versus TMOS for VoIP conditions and references Figure 14 shows a more detailed scatter plot with the G.711 [16] conditions as unfilled circles, all other VoIP conditions as filled circles and the reference conditions as grey diamonds. 1.0 2.0 3.0 4.0 5.0 1.0 2.0 3.0 4.0 5.0 auditory MOS TMOS G.711 conditions other codec cond. references Results for auditory reference test Figure 14: MOS versus TMOS for VoIP conditions and references, detailed plot TOSQA calculates not only a speech quality value which is focused on quality during speech activity, but also an additional value called "connection quality: CQ". This algorithm is still under test and therefore not referenced in ITU-T. Here besides other values, a special weighting of intervals without speech activity is taken into account for the connection quality result. If the current version of this algorithm of TOSQA "connection quality" will be used for assessment, the correlation increases further to 0,935. During the competition for a new standard for objective speech quality measurement within ITU-T Q13/12, also a VoIP database was assessed. This database was recorded during the ETSI Tiphon SuperOp, Hawaii in September 1999. Here a correlation of 0,949 between TOSQA and auditory MOS could be reached. In case of TOSQA "connection quality" this result is increasing to 0,965. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 24 These higher correlations are caused by the smaller range of included conditions in this test. This so called "SuperOp-test" contains only G.711 [16] and G.723.1 [18] conditions under different rates of packet loss and delay jitter. There only two coding algorithm were tested and only one manufacturer was involved in the test. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2.2.4 TOSQA Results | Voice samples were processed across the VoIP scenario for the estimation of the speech quality. Because of the random behaviour of the IP network emulation (NIST Net) 16 voice samples (8 seconds voice each) have been processed to average the random influence for each condition. Nevertheless, there may be still some cases where this random behaviour influences the results. For analysis of the conditions under test (C01...C06), the TMOS values of all 16 speech samples, German and English, were averaged. Tables 7, 8 and 9 show the speech quality estimation from the codec types which were mostly used in the participants VoIP equipment. The tables give also the minimum and maximum TMOS values for each condition. In some cases there is a large difference between TMOS min and TMOS max. This is caused by the random influence of packet loss, the point of time when the packet loss occurs, the influence of VAD and the PLC algorithm itself. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2.2.4.1 G.711 codec | Table 7 shows the average test results from C1 to C6 for the G.711 [16] codec. Table 7: Average Test Results for G.711 [16], VAD on, PLC on/off, PL = 20 ms to 30 ms Condition TMOS average TMOS min TMOS max C1: Drop = 0 %, Delay = 0 ms, Jitter = 0 ms 4,2 4,2 4,2 C2: Drop = 1 %, Delay = 0 ms, Jitter = 0 ms 3,7 3,3 4,0 C3: Drop = 2 %, Delay = 0 ms, Jitter = 0 ms 3,4 3,1 3,8 C4: Drop = 3 %, Delay = 0 ms, Jitter = 0 ms 3,4 2,8 3,7 C5: Drop = 5 %, Delay = 0 ms, Jitter = 0 ms 3,1 2,7 3,5 C6: Drop = 1 %, Delay = 50 ms, Jitter = 20 ms 3 2,7 3,3 As expected the TMOS values are decreasing with higher packet loss values. In general the results show that for the G.711 [16] codec the reached TMOS values, especially for C1, are fully comparable with the simulated reference values. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2.2.4.2 G.723 codec | Table 8 shows the test results from C1 to C6 for the G.723 codec. Table 8: Average Test Results for G.723, VAD on, PLC on/off, PL = 30 ms Condition TMOS average TMOS min TMOS max C1: Drop = 0 %, Delay = 0 ms, Jitter = 0 ms 3,3 3,2 3,5 C2: Drop = 1 %, Delay = 0 ms, Jitter = 0 ms 3,2 3,1 3,4 C3: Drop = 2 %, Delay = 0 ms, Jitter = 0 ms 3,1 3,0 3,2 C4: Drop = 3 %, Delay = 0 ms, Jitter = 0 ms 3,0 2,9 3,1 C5: Drop = 5 %, Delay = 0 ms, Jitter = 0 ms 2,8 2,7 2,9 C6: Drop = 1 %, Delay = 50 ms, Jitter = 20 ms 2,7 2,0 3,1 Normally one may expect better values, especially for condition C1. The result of condition 1 is mainly caused by some bad voice samples where the codec seemed to be desynchronized. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 25 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2.2.4.3 G.729 codec | Table 9 shows the average test results from C1 to C6 for all vendors for the G.729 [22] codec. Table 9: Average Test Results for G.729 [22], VAD on, PLC on/off, PL = 10 ms to 30 ms Condition TMOS average TMOS min TMOS max C1: Drop = 0 %, Delay = 0 ms, Jitter = 0 ms 3,3 3,1 3,5 C2: Drop = 1 %, Delay = 0 ms, Jitter = 0 ms 3,1 2,9 3,3 C3: Drop = 2 %, Delay = 0 ms, Jitter = 0 ms 3,0 2,9 3,2 C4: Drop = 3 %, Delay = 0 ms, Jitter = 0 ms 2,9 2,9 3,0 C5: Drop = 5 %, Delay = 0 ms, Jitter = 0 ms 2,7 2,6 2,9 C6: Drop = 1 %, Delay = 50 ms, Jitter = 20 ms 2,3 1,8 2,7 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.2.2.4.4 Summary and conclusion | Figure 15 again give a summarized overview of all TMOS results. It displays also the references used in this test to compare the results of the real processed voice samples against codec simulations. C1 C3 C5 C2 C4 C6 G.711 G.729 MNRU 18 30 C1 C3 C5 C2 C4 C6 C1 C3 C5 C2 C4 C6 G.723.1@6.3 24 12 06 direkt References 1 2 3 4 5 G.711 PL=20...30ms VAD on, PLC on/off G.723.1@6.3 PL=30ms VAD on, PLCon/off G.729 PL=10...30ms VAD on, PLC on/off G.711 reference G.723.1@6.3 ref. G.729 ref. TMOS Figure 15: Summary of all conditions (incl. reference conditions) |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3 Delay | |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.1 Delay between two analogue PBX subscribers | Source: Robert Bosch GmbH: ETSI TIPHON 14TD081. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.1.1 Introduction | The end-to-end speech transmission performance of two different IP gateways, which are commercially available on the marketplace, has been evaluated. In addition, two versions of one PC client software have been compared with respect to their impact on speech transmission quality. The tests have been conducted in an environment which is typical for business applications, i.e. the IP network consisted of a LAN (without traffic load) and the SCN consisted of PBXs (Private Branch Exchanges). The coding scheme was in all cases selected as G.723.1 [18] (6,3 kbit/s); the Voice Activity Detection (VAD) as well as the integrated echo cancelling devices were activated in all cases. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 26 In the following the IP Gateway from manufacturer A is denoted "Gateway type A" while the IP Gateway from manufacturer B is denoted "Gateway type B"; the two versions of the PC client software are denoted "SW version x" and "SW version y", respectively. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.1.2 Measurement Set Up | Measurements have been conducted with HEAD Acoustics' Communication Analysis System (CAS); E-Model calculations have been performed with Alcatels' software. The IP gateways are connected to the PBXs via 2 Mbit/s links with QSIG signalling; the measuring interfaces at the subscriber side of the PBX were analogue 4-wire interfaces in accordance with ES 201 168 [7]. Measurements at the PC client have been performed at the electric handset interface. The telephone sets used during the tests have been standard analogue sets which comply with the default values of the E-Model (as far as applicable). In order to keep the testing effort within reasonable limits, measurement of delay, jitter and functionality of the echo cancellers has not been performed between acoustical interfaces, because measurement of such parameters between electrical interfaces gives approximately the same results. Figure 16 shows the principal test Set Up. Analogue subscriber Analogue subscriber PBX PBX Gateway Gateway LAN Figure 16: Set Up between two analogue PBX subscribers |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.1.3 Results | The results of Delay measurements are summarized in table 10. Table 10: Measurement results of delay measurements of two different gateways Gateway A Gateway B The mean one-way delay for both subscribers is variable between 125 ms and 250 ms, with an arithmetical mean value at 160 ms. The mean one-way delay is constant with a value between 90 ms and 110 ms (see figure 17). ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 27 NOTE: Red: original send signal; Green: receive signal (transmitted via the test Set Up). Figure 17: Delay behaviour of the test Set Up for two analogue PBX subscribers Due to the selected measurement mode the original send signal (red) is displayed 150 ms later and reduced by 6 dB in level than in reality. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.1.4 Conclusion | For transmission planning purposes the mean value of the one-way delay of both transmission directions of a connection is calculated and expressed as "Mean one-way delay". In this respect Gateway type B with 100 ms is much better than Gateway type A with approx. 160 ms (which was variable, additionally). |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.2 Delay between two PABX systems with analogue subscribers | Source: HEAD Acoustics: QoS Measurements of IP-Configurations 15TD089. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.2.1 Measurement Set Up | On both networks no additional traffic was generated such that there was no packet loss. The System Configurations (PABX [Bosch], IP-Gateway [3rd party], PC-Terminals with commercial VOIP software) were provided by BOSCH Telecom. The acoustical access to the terminals was made using HATS (artificial head according to ITU-T Recommendation P.58 [27]) equipped with the artificial ear type 3.4 according to ITU-T Recommendation P.57 [26] and a handset positioning device according to new Recommendation P.64 [28]. This Set Up guarantees a most realistic Set Up for the terminals since all transmission properties are reproduced in a very realistic way. The measurements were conducted in "noisy" environments, this means the background noise in the laboratory rooms was present all the time. The electrical access to the configurations was made using the 600 Ω access pint of the PABX for configuration 1, for configuration 2 the soundcard was used for access. The level adjustment in this configuration was made in advance to the test in order to achieve realistic loudness ratings for sending and receiving. Although it should be noted that there was always AGC in sending and receiving present which could influence the settings during the measurements. Such settings were always checked during the measurements in order to not be misled by different settings of the devices. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 28 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.2.2 Result | • One way transmission delay: 70 ms. • Coding: ITU-T Recommendation G.729 [22]. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.3 Delay between two PC SW clients | Source: HEAD Acoustics: QoS Measurements of IP-Configurations, 15TD089. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.3.1 Measurement Set Up | On both networks no additional traffic was generated such there was no packet loss. The System Configurations (PABX [Bosch], IP-Gateway [3rd party], PC-Terminals with commercial VOIP software) were provided by BOSCH Telecom. The acoustical access to the terminals was made using HATS (artificial head according to ITU-T Recommendation P.58 [27]) equipped with the artificial ear type 3.4 according to ITU-T Recommendation P.57 [26] and a handset positioning device according to new Recommendation P.64 [28]. This Set Up guarantees a most realistic Set Up for the terminals since all transmission properties are reproduced in a very realistic way. The measurements were conducted in "noisy" environments, this means the background noise in the laboratory rooms was present all the time. The electrical access to the configurations was made using the 600 Ω access pint of the PABX for configuration 1, for configuration 2 the soundcard was used for access. The level adjustment in this configuration was made in advance to the test in order to achieve realistic loudness ratings for sending and receiving. Although it should be noted that there was always AGC in sending and receiving present which could influence the settings during the measurements. Such the settings were always checked during the measurements in order to not be misled by different settings of the devices. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.3.2 Results | • One Way Transmission Time: 400 ms to 530 ms. • Coding: ITU-T Recommendation G.723.1 [18]. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.4 Delay between an analogue PBX subscriber and a PC client | Source: Robert Bosch GmbH; Speech Quality Test results of IP equipment in a LAN environment; ETSI TIPHON 14TD081. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.4.1 Measurement Set Up | Analogue subscriber PC client with SW version PBX Gateway LAN Figure 18: Set Up between an analogue PBX subscriber and a PC client ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 29 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.4.2 Results | Figure 19 shows the delay behaviour of the test Set Up for the transmission direction from PC client to analogue subscriber. NOTE: Red: original send signal; Green: receive signal (transmitted via the test Set Up). Figure 19: Delay behaviour of the test Set Up for the transmission direction from PC client to analogue subscriber Due to the selected measurement mode the original send signal (red) is displayed 500 ms later and reduced by 6 dB in level than in reality. The mean one-way delay is constant but depending on the direction of transmission: • from analogue subscriber to PC client = 465 ms; • from PC client to analogue subscriber = 200 ms. Figure 19 shows that no significant delay jitter could be observed. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.3.4.3 Conclusion | The values depend on the direction of transmission and are in the range from 200 ms to 700 ms, where the highest values occur when the PC client is at receive side of a delay measurement. The PC client functions as a combination of gateway and telephony terminal and its contribution to the total delay - in principle - should not deviate significantly from the delay contribution of a gateway plus its interconnected telephone set. The very high delay contribution of the PC client found in this test may be caused by another approach of its delay variation buffer in receive direction. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.4 Echo | |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.4.1 Echo between two analogue PBX subscribers | Source: Robert Bosch GmbH, Speech Quality Test Results of IP equipment in a LAN environment, ETSI TIPHON 14TD081. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 30 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.4.1.1 Measurement Set Up | Analogue subscriber Analogue subscriber PBX PBX Gateway Gateway E C E C LAN Delay and Loss simulation Whole Echo Path of the EC in the right Gateway Figure 20: Measurement Set up for Echo measurements between two analogue PBX subscribers Two different cases of loss were simulated. • In case of simulated loss in the whole echo path of 7 dB: The delay and loss simulation device (see figure 20) has been adjusted such, that the loss of the whole echo path is 7 dB, because ITU-T Recommendation G.165 [14] specifies functionality of an EC for echo path loss > 6 dB. • In case of simulated loss in the whole echo path of 4 dB: The delay and loss simulation device (see figure 20) has been adjusted such, that the loss of the whole echo path is 7 dB, because ITU-T Recommendation G.168 [15] specifies functionality of an EC for echo path loss > 0 dB (which is different from G.165 [14], see above). |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.4.1.2 Results | Table 11 shows the results of simulated loss in the whole echo path. Table 11: Results of Echo loss simulation Gateway Simulated loss in the whole echo path 7 dB 4 dB A For simulated tail delay values up to 28 ms the residual echo level is below -65 dBm0 (test signal white noise, -10 dBm0). For increased values of the tail delay the residual echo level is more than -25 dBm0. Note, that the tail delay for the EC is identical with the delay along the whole echo path. The adaptation of the EC is very slow; sometimes the value of the remaining level of the echo signal goes below the threshold of the Non-Linear Processor (NLP) which instantly provides very good values of echo loss, because in that case the NLP clips off the remaining amplitude of the echo signal. B For simulated tail delay values up to 11 ms the residual echo level is below -85 dBm0 (test signal: white noise, -10 dBm0). For increased values of the tail delay the residual echo level is more than -25 dBm0. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.4.1.3 Conclusion | The echo cancellers in both gateways do not comply with the applicable ITU-T Recommendation G.168 [15]. For practical application the maximum tail delay of the echo canceller (as found in the tests) must be reduced by 6 to 8 ms; the remaining value must be divided by two (tail delay is a round-trip value) in order to get the correct value of the maximum mean one-way delay which an EC can handle. For Gateway type A this results to (28 ms - 6 ms)/2 = 11 ms, whereas for Gateway type B (11 ms - 6 ms)/2 = 2,5 ms are available, only. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 31 Small and medium sized private networks (e.g. corporate networks) which in a PSTN environment are operated without the deployment of separate echo cancellers, typically add 10 ms to 20 ms to the mean one-way delay of a connection; it should be recognized that the use of both types of gateways will require additional echo cancellers in such networks. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.5 Impairment Factors | |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.5.1 Transmission Impairments according to ITU-T G.113 | Source: ITU-T Recommendation G.113 [12]. ITU-T has provided tables of transmission impairment factors for a couple of speech codecs. These impairment factors are marked as Ie values in every table. Table 12 of Ie values refers to non-error conditions. For propagation errors and frame-erasures or packet loss, no definite values are available which would be valid for more than one codec or codec family. In order to help the transmission planner, examples of Ie values under conditions of packet loss are given in tables 13 and 15, and for propagation error patterns EP1 and EP2 in table 15. These values are provisional only as they were determined in single or a few experiments. Table 12: Provisional Planning Values for the Equipment Impairment Factor Ie Codec Type Reference Operating Rate kbit/s Ie Value PCM G.711 [16] 64 0 ADPCM G.726 [19], G.727 [20] 40 2 G.721 [17], G.726 [19], G.727 [20] 32 7 G.726 [19], G.727 [20] 24 25 G.726 [19], G.727 [20] 16 50 LD-CELP G.728 [21] 16 7 12,8 20 CS-ACELP G.729 [22] 8 10 G.729A [23] + VAD 8 11 VSELP IS-54 8 20 ACELP IS-641 7,4 6 QCELP IS-96a 8 19 RCELP IS-127 8 6 VSELP Japanese PDC 6,7 24 RPE-LTP GSM 06.10 [8], Full-rate 13 20 VSELP ETS 300 581-2 [37], Half-rate 5,6 23 ACELP EN 300 726 [38], Enhanced Full Rate 12,2 5 ACELP G.723.1 [18] 5,3 19 MP-MLQ G.723.1 [18] 6,3 15 ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 32 Table 13: Provisional planning values for the equipment impairment factor Ie under conditions of random packet loss, codecs G.729A [23] + VAD, G.723.1-A [41] + VAD and GSM EFR % Packet Loss G.729A [23] + VAD G.723.1-A [41] + VAD 6,3 kbit/s GSM EFR 0 11 15 5 0.5 13 17 — 1 15 19 16 1.5 17 22 — 2 19 24 21 3 23 27 26 4 26 32 — 5 — — 33 8 36 41 — 16 49 55 — NOTE: Number of frames per packet: • G.729A [23] + VAD: 2;• G.723.1-A [41] + VAD: 1;• GSM EFR: 1. Table 14: Provisional planning values for the equipment impairment factor Ie under conditions of packet loss, codecs G.711 [16] without and with Packet Loss Concealment (PLC) G.711 [16] w/ PLC Packet Loss % G.711 [16] w/o PLC Random Packet Loss Bursty Packet Loss 0 0 0 0 1 25 5 5 2 35 7 7 3 45 10 10 5 55 15 30 7 — 20 35 10 — 25 40 15 — 35 45 20 — 45 50 NOTE: Speech packet length: 10 ms. Table 15: Provisional planning values for the equipment impairment factor Ie under propagation error conditions, GSM codecs Codec type Error pattern Ie Range GSM-HR EP1 25 to 32 EP2 31 to 42 GSM-FR EP1 32 to 39 EP2 40 to 45 GSM-EFR EP1 15 to 22 EP2 26 to 35 NOTE 1: The range given results from the difficulties in deriving exact impairment factor values for these conditions. NOTE 2: EP1 is equivalent to 10 dB C/I, EP2 is equivalent to 7 dB C/I. C/I is the carrier to interference ratio. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 33 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.6 R-Values | |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.6.1 Analogue PBX- and SW Client scenarios | Source: Robert Bosch GmbH, Speech Quality Test Results of IP equipment in a LAN environment, ETSI TIPHON 14TD081. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.6.1.1 Introduction | The results given in clause 5.5.1 are taken as an input to the E-Model in order to predict end-to-end speech transmission performance in terms of the E-Model Rating R as perceived by the average user and to predict user satisfaction. The following three scenarios have been chosen to demonstrate the impact on end-to-end speech transmission performance, which are representative for the application of the gateways under test in private networks (e.g. corporate networks). In order to include the massive impact of the low syllable recognition rate into the E-Model calculation, a preliminary Ie value for this effect had to be estimated. The syllable cut-off due to a VAD can be compared with the effect of hands free telephony. For hands free telephony it is well-known that the impact perceived by the user lies in a range Ie = 10 to 20. Hence, for this evaluation an Ie value of Ie = 10 was chosen to consider the impact of the reduced syllable recognition rate. End-to-end speech transmission quality has been calculated according to ITU-T Recommendations G.107 [9] using the guidance given in ITU-T Recommendation G.108 [10] and compared with the categories of speech transmission quality defined ITU-T Recommendation G.109 [11]. An Advantage Factor A which is sometimes discussed for Internet-Telephony does not apply for business applications. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.6.1.2 Results | R values were calculated for different scenarios. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.6.1.2.1 Two Gateways type A with analogue telephone sets | The delay in the PBX network on the left side exceeds the capability of the EC (which is integrated in the gateway). Figure 21: Speech transmission quality as perceived by the user at the left terminal where the EC is out of his operational range ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 34 Table 16: Speech transmission quality as perceived by the user at the left terminal where the EC is out of his operational range R-Value Speech Transmission Quality Category User satisfaction 67 Low Many users dissatisfied Table 17: Speech transmission quality as perceived by the user at the right terminal where the EC is within his operational range R-Value Speech Transmission Quality Category User satisfaction 0 Values < 50 are not recommended All users dissatisfied Because the left EC is out of his operational range, the speech quality perceived by the user at the right terminal is decreased. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.6.1.2.2 Gateway type A with analogue subscriber behind PBX | In this scenario the analogue subscriber behind the PBX is on the left side and the PC client is on the right side. Both EC (which are integrated in the gateway and in the PC client) are within their operational range. Figure 22: Speech transmission quality as perceived by the user at the left terminal (analogue telephone set) Table 18: Speech transmission quality as perceived by the user at the left terminal (analogue telephone set) R-Value Speech Transmission Quality Category User satisfaction 38 Values < 50 are not recommended All users dissatisfied ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 35 Table 19: Speech transmission quality as perceived by the user at the right terminal (PC client) R-Value Speech Transmission Quality Category User satisfaction 53 Poor Nearly all users dissatisfied |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.6.1.2.3 Gateway type B with analogue telephone sets on both sides | The analogue telephone sets are used as a termination of the PBXs. The delay in the PBX network on the left side exceeds the capability of the EC (which is integrated in the gateway). Figure 23: Speech transmission quality as perceived by the user at the left terminal where the EC is out of his operational range Table 20: Speech transmission quality as perceived by the user at the left terminal where the EC is out of his operational range R-Value Speech Transmission Quality Category User satisfaction 68 Low Many users dissatisfied Table 21: Speech transmission quality as perceived by the user at the right terminal where the EC is within his operational range R-Value Speech Transmission Quality Category User satisfaction 8 Values < 50 are not recommended All users dissatisfied Because the left EC is out of his operational range, the speech quality perceived by the user at the right terminal is decreased. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 36 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.6.1.3 Conclusion | The results presented herein do not significantly differ from those we gained recently with other gateways and with other client software. Most manufacturers claim that they delivered state-of-the-art gateways and client software as an input for our evaluations. The speech transmission end-to-end performance which was found in this evaluation is not acceptable for long-term business applications. Hence, one may recognize a substantial gap between the TIPHON classes defined for speech quality and available solutions in practice. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7 Advanced Measurement Techniques | |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.1 QoS Measurements of IP-Configurations | Source: HEAD Acoustics; QoS Measurements of IP-Configurations; ETSI TIPHON 15TD089. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.1.1 Configuration and Measurement Set Up | For the test 2 IP configurations were available which can be found in figures 24 and 25. On both networks no additional traffic was generated such there was no packet loss. The System Configurations (PABX [Bosch], IP-Gateway [3rd party], PC-Terminals with commercial VOIP software) were provided by BOSCH Telecom. The acoustical access to the terminals was made using HATS (artificial head according to ITU-T Recommendation P.58 [27]) equipped with the artificial ear type 3.4 according to ITU-T Recommendation P.57 [26] and a handset positioning device according to new Recommendation P.64 [28]. This Set Up guarantees a most realistic Set Up for the terminals since all transmission properties are reproduced in a very realistic way. The measurements were conducted in "noisy" environments, this means the background noise in the laboratory rooms was present all the time. The electrical access to the configurations was made using the 600 Ω access pint of the PABX for configuration 1, for configuration 2 the soundcard was used for access. The level adjustment in this configuration was made in advance to the test in order to achieve realistic loudness ratings for sending and receiving. Although it should be noted that there was always AGC in sending and receiving present which could influence the settings during the measurements. Such the settings were always checked during the measurements in order to not be misled by different settings of the devices. 600 analog in Ω IP- Gateway IP- Gateway PABX 2 Mb Measurement System CAS & Signal-Conditioning MFE II 2 Mb PABX 600 analog out Ω Figure 24: Configuration 1 - One way transmission delay = 70 ms, Coding ITU-T Recommendation G.729 [22], "Standard" - handset telephone ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 37 Sound- Card Measurement System CAS & Signal-Conditioning MFE II Personal Computer, & IP-Client Personal Computer, & IP-Client IP IP Sound- Card Figure 25: Configuration 2 - One way transmission time = 400 ms to 530 ms, variable, Coding ITU-T Recommendation G.723 [40], PC with soundcard and headset Figure 26: Test set up with head set and hand set |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.1.2 Results | |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.1.2.1 Parameters in Single Talk Conditions | The following parameters are described as "Traditional" Parameters. Figures 27 and 28 show the measured frequency responses in sending and receiving for both, configuration 1 and configuration 2. In sending the frequency responses do not indicate any problem except that the headset does not show a high pass filtering needed to exclude low frequency room noise. This can be seen for the handset at about 300 Hz. The same statement can be made for the receiving direction. In addition the handset provides a smooth frequency response when adapted with a pressure force of 13 N to the artificial head. The headset provides a more high frequency emphasized frequency response. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 38 Figure 27: Sending frequency responses for configuration 1 (handset) and 2 (headset) Figure 28: Receiving frequency responses for configuration 1 (handset) and 2 (headset) In the following clauses more detail results in single talk conditions especially for switching, AGC characteristics and echo loss will be given. When evaluating the switching characteristics more in detail the following statements can be made. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.1.2.2 Level dependant input-output characteristics | In sending and receiving direction a VAD is detectable which for both configurations indicate no or only minor distortions caused by syllable clipping. The measurement was made using an increasing voiced sound of artificial voice (ITU-T Recommendation P.50 [42]), monitoring the output signal and comparing it to the input (measurement) signal. In receiving direction the configuration 1 does not show any non linear behaviour whereas configuration 2 introduces a quite strong companding for speech as it can be seen in figure 30. In addition –depending on the input signal level- strong level variations can be seen (beginning and end of the test sequence in figure 29). ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 39 NOTE: upper: time signal; lower: level vs. time. Figure 29: Level depending input - output characteristics in receiving, configuration2 NOTE: upper: time signal; lower: level vs. time. Figure 30: Level dependant input-output characteristics in sending, configuration 1 NOTE: upper: time signal; lower: level vs. time. Figure 31: Level dependant input-output characteristics in sending, configuration 2 ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 40 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.1.2.3 Echo loss and convergence | The echo loss was measured always from the electrical input to the electrical output. Certainly in single talk the characteristics of the echo canceller(s) are quite important for the subjective impression of a connection. Figure 32 and 33 show measurements using the composite source signal CSS as defined in ITU-T Recommendation P.501 [29] and ITU-T Recommendation G.168 [15] at configuration 1. Figures 32 and 33 show the convergence as a function of time, displayed in the spectral domain. Figure 32 shows the convergence. As typical for echo cancellers the echo signal is decreased, first in the low frequency domain. After about 2 s the echo signal is sufficiently low and the non linear processor (NLP) is activated. It however should be noted that the echo loss in this condition is only 32 dB, which is sufficient for 70 ms delay but not for higher delays which may occur in the connection. Figure 33 shows accidental switching of the NLP after the echo canceller was fully converged. This will result in "bursty" echoes during single talk and certainly degrades the speech quality. The echo loss in this condition is only 25 dB which is not sufficient according to ITU-T Recommendation G.131 [13]. The spectral components of the echo components are the same as they can be found in figure 32 for times where the echo canceller was not fully converged. Additional tests were carried out by introducing an additional delay between PABX and terminal and such "scratching" the limits of the processing window of the echo canceller. This leads (as expected) to poor echo loss of 12 dB (no echo loss enhancement provided by the echo canceller) and accidental switching of the NLP as seen in figure 29. NOTE: red: measurement signal; green: echo signal; upper: time signal; lower: echo loss vs. time. Figure 32: Convergence vs. time; configuration 1 NOTE: red: measurement signal; green: echo signal; upper: time signal; lower: echo loss vs. time. Figure 33: Echo loss vs. time, EC converged after 11 s; configuration 2 ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 41 NOTE: red: measurement signal; green: echo signal; upper: time signal; lower: echo loss vs. time. Figure 34: Echo loss vs. time when exceeding the processing window of the echo canceller, configuration 1 NOTE: red: measurement signal; green: echo signal; upper: time signal; lower: echo loss vs. time. Figure 35: Echo loss vs. time, configuration 2 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.1.2.4 Performance in sending direction in the presence of background noise | A very importance performance parameter is the system behaviour in the presence of background noise. Some measurement examples are shown in figures 36 and 37. For configuration 1 figure 36 shows, that the background noise (indicated in red) is transmitted only when a certain level is reached. After reaching this threshold the background signal is transmitted without interruptions. In contrast to that the background noise transmission in sending for configuration 2 is very poor. As it can be seen from figure 37, the background noise is constantly interrupted for various levels. NOTE: red: measurement signal; green: output signal at the line interface; upper: time signal; lower: level vs. time. Figure 36: Background noise transmission in sending level versus time, configuration 1 NOTE: red: measurement signal; green: output signal at the line interface; upper: time signal; lower: level vs. time. Figure 37: Background noise transmission in sending, level versus time, configuration 2 ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 42 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.1.2.5 Background noise performance in receiving direction | Configuration 1 behaves in receiving direction similar to the sending direction. Such no speech degradation should be expected. Configuration 2 however shows quite a strong AGC characteristic as it can be seen in figure 38. NOTE: green: measurement signal; red: output signal at the headphone); upper: time signal; lower: level vs. time. Figure 38: Background noise transmission in receiving, level versus time, configuration 2 From this evaluation it can be expected that even low level background noise signals are transmitted to the listeners ear with nearly the same level than speech! Furthermore accidental amplification of the background noise signal occurs as it can be seen e.g. in the time interval from 5 to 6 s in figure 38. This will result in a very bad background noise performance, especially if this system is connected with a terminal providing the intermittent transmission behaviour as shown in figure 37. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.1.2.6 Evaluation of Double Talk Conditions | The double talk evaluations were made using test sequences as described in new ITU-T Recommendation P.502 [30] and TR 101 329 [2]. The double talk test signal is shown in figures 39 and 40 and is inserted simultaneously in sending and receiving direction. Figure 39: Overview of double talk test signal ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 43 s(t) - test signal sending direction sdt(t) - double talk test signal in receiving direction Figure 40: Cut out of the complete measurement sequence with detailed view on the overlap of sending and receiving direction signal, principle arrangement By using this test signal the speech-like CSS Sequences are fed into sending and receiving simultaneously with varying level. The level is increasing (or decreasing) by 0,5 dB from CSS package to CSS package. Thus the level variation during double talk is 20 dB in both directions. Some measurement results for configuration 1 are given below. NOTE: red: measurement signal; green: output signal at the line interface Figure 41: Transmission during double talk in sending for high sending level, configuration 1 NOTE: red: measurement signal; green: output signal at the line interface Figure 42: Transmission during double talk in sending for low sending level, configuration 1 For high signal levels in sending direction and low levels in receiving no signal clipping or echo problems are obvious. For low signal levels in sending the sending signal is subject to clipping, the switching time is about 120 ms. A similar behaviour can be found in receiving direction. Due to the low signal levels this is not critical, speech will be transmitted in double talk with no or only minor audible degradation. The echo evaluations during double talk led to similar results than the ones found in single talk conditions and such are not discussed here in detail. Due to the very poor echo loss of configuration 2 already in single talk conditions the more detailed investigation of the double talk characteristics is discussed not more in detail. The performance of the system in double talk is as poor as it is in single talk condition. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 44 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.1.2.7 Relationship to subjective tests | The configurations investigated here were not subject to extensive subjective tests. Although some experts tests had been made which led to similar results than they were achieved in the objective measurement. All the effects which were found by the instrumental methods were the ones which were found subjectively as well. All the delay, echo and background noise problems as well as the switching found in configuration 2 were reported. In a similar way the configuration 1 was judged good which could be confirmed by the measurements conducted. The switching and echo problems for configuration 1 in the case when the processing window of the echo canceller was exceeded can be confirmed by the subjects in the same way. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.1.3 Conclusion | Two IP configurations were investigated by the test methods defined in TR 101 329 [2] and the new ITU-T Recommendations P.502 [30] and P.581 [31]. The test results show that the methods seem to work well for the diagnostic evaluation of end-to-end scenarios using IP-connections. A detailed investigation of various impairments especially in configurations including terminals is possible by using the new methodologies and signals. The background noise performance, the switching characteristics in various conditions, the double talk performance and especially the echo characteristics can be evaluated in great detail and such a very good estimation of the various impairments introduced is possible on an objective basis. Further work will be conducted in order to evaluate other end-to-end scenarios and to include the condition of packet loss and other impairments. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.2 QoS Measurements of ETSI Speech Quality Test Event 2000 | |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.2.1 Introduction | This clause describes the test methodologies and the results of the measurements which were carried out during the 1st ETSI Speech Quality Test Event. The tests were conducted by T-Nova Deutsche Telekom Innovationsgesellschaft mbH Berkom (T-Nova) in collaboration with HEAD acoustics GmbH (HEAD acoustics). The aim of the test event was to determine the speech quality of various voices over IP equipment under certain IP network conditions. During the test event, speech material as well as measurement data were collected by transferring voice samples and artificial signals across the Voice over IP Set Up. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.2.2 Measurement Set Up | Measurements at the acoustical interface were conducted with two measurement scenarios: • Acoustical transmitter interface – Acoustical receiver interface ("Acoustical-Acoustical"); • Electrical transmitter interface – Acoustical receiver interface ("Electrical-Acoustical"). Measurements were executed using mainly with artificial test signals (according to ITU-T Recommendation P.501 [29]). The artificial test signals were used for measuring of transmission parameters according the ITU-T Recommendation P.500 series. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 45 Gateway IP T1 / E1 / BRI NISTNet Packet MONITOR T1 / E1 / BRI Packet Loss DELAY Output HEAD Acoustics Testsystem CAS Input Gateway IP IP ISDN DSS1 ISDN DSS1 ALCATEL PABX 4400 Figure 43: "Acoustical – Acoustical" Measurement Set Up with Reference ISDN Terminals NISTNet Packet MONITOR Packet Loss DELAY Output HEAD Acoustics Testsystem CAS Input IP IP IP-Phone PC IP-Terminal Figure 44: "Acoustical – Acoustical" Measurement setup with IP Terminals (handsets or headsets) The reference terminals provided were standard digital handset terminals according to ETSI TBR 8. For all kind of measurements a packet loss generator and a packet loss monitor were included in the Set Up. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 46 Gateway IP T1 / E1 / BRI NISTNet Packet MONITOR T1 / E1 / BRI Packet Loss DELAY Output HEAD Acoustics Testsystem CAS Input Gateway IP IP ISDN DSS1 ISDN DSS1 ALCATEL PABX 4400 Figure 45: Measurement Set Up "Acoustical – Electrical" for Gateway to Gateway Configuration For the tests the handsets of the terminals are applied to the HATS using the positioning as described in ITU-T Recommendation P.64 [28] with defined pressure force. The test sequences, natural speech as well as the artificial sequences were automatically introduced and recorded by the test system CAS and stored on hard disc. Instead of the PABX telephone an IP telephone was used if provided in combination with a gateway which interfaces to the PABX. This is shown in figure 46. IP NISTNet Packet MONITOR T1 / E1 / BRI Packet Loss DELAY Output HEAD Acoustics Testsystem CAS Input Gateway IP ISDN DSS1 ALCATEL PABX 4400 IP IP Terminal (Phonre or PC) Figure 46: Measurement Set Up "Acoustical – Electrical" for IP-Terminal to Gateway Configuration ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 47 The IP network conditions for all kinds of acoustical measurements ("electrical – acoustical" and "acoustical – acoustical") were: Table 22: Network Conditions for All Kinds of Acoustical Measurements Condition Packet Loss (Equal) Additional Delay1) Delay Variation 1 0 No No 2 0 100 ms No 3 0 100 ms 20 ms 4 1 % 100 ms No 5 1 % 100 ms 20 ms 6 3 % 100 ms No NOTE: Additional IP network delay was introduced using NIST Net. Also for all kinds of acoustical measurements it was recommended to use a packet length of at least 60 ms for audio frames (cf. electrical-electrical measurements). This packet length had no influence on auditive or instrumental assessment methods. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.2.3 Results | |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.2.3.1 Packet loss | Figures 47 and 48 demonstrate two analysis examples obtained from recordings under network conditions with simulated packet loss. The figures show an enlarged time sequence from the recorded signal using the periodical repetition of a voiced sound as test signal. Figure 47: Occurrence of packet loss Figure 48: Packet loss concealment A resulting signal gap due to one lost IP packet can be analysed in figure 47. Packet loss concealment (PLC) is obviously not implemented in the equipment which was under test. A typical PLC implementation can be seen in figure 48. The lost IP packet obviously is substituted by a previous frame, but signal irregularities at the beginning and the end of this substituted frame can be analysed. These resulting peaks are audible and annoying. This result reflects one kind of PLC implementation which can be used to cover the influence of lost packets in the network. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 48 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.2.3.2 Transmission characteristics for background noise | The test signal consists of the spectral shaped noise signal (Hoth spectrum) with increasing test signal level vs. time. The measurement result derived from the test via one ISDN line and the PBX is shown in figure 49 for comparison. The red signal in the upper windows represents the test signal, the green signal the measured signal (transmitted signal). Note, that the overlapped red and green colour results in yellow. The analysis curves in the lower window show the calculated level vs. time (calculated in time domain using a 35 ms time constant). The transmission is linear, no level variations can be observed, the two curves (red for the original test signal level, green for the transmitted signal) are in parallel. Typical results derived from tests during the event are shown in figure 50. Figure 49: Transmission characteristic for background noise (PBX reference connection) NOTE: red: original test signal. Figure 50: Typical analysis results derived from tests with 3 manufacturers The results displayed in yellow, magenta and cyan represent different kinds of implementations. An activation level threshold (yellow), the influence of automatic gain control (AGC, cyan) and an implementation with adaptive comfort noise injection can be analysed clearly. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.2.3.3 Transmission performance under double talk performance | The transmission characteristics under double talk conditions were determined using the test signal consisting of two decorrelated composite source signals which are periodically repeated and applied with different signal levels to both sides of the connections. The recordings were carried out in one direction and consequently the signal which is fed at the other end of the connection should be transmitted and no additional signal components should occur. The following 4 figures demonstrate typical measurement results which were obtained during the event. In each figure the transmitted and recorded signal is shown in green and the original test signal is displayed in red. In order not to confuse the reader the double talk signal which is fed in the opposite transmission path is not given in these figures. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 49 Figure 51: Result for the PBX connection Figure 52: Typical result demonstrating echo components and short term clipping Figure 53: Typical result demonstrating echo components Figure 54: Typical result demonstrating echo components and clipping Figure 51 shows the example for the analysis obtained for the PBX reference connection and points out that all signal bursts are completely transmitted. Clipping does not occur. During the pauses between 2 signal bursts the channel noise floor is recorded. No additional signal components which could be identified as disturbances (like echoes or others) occur. The complete transmission of all signal bursts can also be seen in figure 52 except a short frontend clipping. Additional high level echo components (green) can be determined during the pauses between the signal bursts. Basically the same result can be analysed in figure 53 for another tested connection. One example for the occurrence of signal clipping (parts of the green signals are missing) is demonstrated in figure 54. An additional high level noise floor is recorded between the signal bursts in this example. These three measurement results demonstrate some disturbances which are introduced by the implemented echo cancellers and the non-linear processors. Figure 55 compares the power density spectra analysed during the signal pauses under double talk conditions. The red curve represents the curve for the original test signal which was applied in the opposite transmission path. The green curve was derived from the measurement of the PBX connection. The other curves in cyan, yellow and magenta represent the results for 3 different test connections. Obviously significant differences compared to the PBX connection occurred for the connections under test during the event. ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 50 NOTE: red: original test signal; green: PBX connection; others: typical results measured during the event). Figure 55: Comparison of power density spectra analysed in the pauses between two signals bursts during the double talk period Further analysis demonstrates that these signal components are caused by echoes which only occur during double talk periods. It should be noted that the test connection was completely 4-wire and therefore physically did not produce any echo. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.7.2.3.4 Detailed analysis of echo during double talk | A specific test signal according to ITU-T Recommendation P.501 [29] consisting of a two channel signal with a comb filter structure was used for a more detailed analysis of the echo disturbances during double talk. This signal is suited to determine and measure the echo more precisely. The test result which was obtained for the PBX connection is analysed in the example in figure 56. The time signals are shown in the upper window. The analysis window (lower window) shows the power density spectrum calculated by Fourier transform for the measured signal (green) and the original test signal in receiving direction (red). In principle signal components in the measured (green) signal which correlate to the excitation frequencies of the original signal (red, note that this was the original one which may cause echoes) can be analysed as echoes. The analysis in figure 56 shows that the PBX connection was echo-free under double talk conditions. A more detailed analysis example for the enlarged frequency range between 400 Hz and 1,2 kHz is given in figure 57. Again typical measurement results obtained for 3 different manufacturers are analysed together with the PBX connection (green). The figure shows significant echo components for some of the tested connections. These components can be detected in the frequency range which was excited by the original test signal (red). ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 51 NOTE: green: measured signal containing double talk signal and echo; red: original test signal. Figure 56: Determination of echo during double talk test result for the PBX reference connection upper window: time sequences, lower window: power density spectra NOTE: green: PBX reference connection; red: original test signal; others: typical results measured during the event. Figure 57: Enlarged frequency range with a comparison of typical test results ETSI ETSI TR 101 329-6 V2.1.1 (2002-02) 52 |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.8 Jitter Buffer | Source: KPN; Problems with the behaviour of jitter buffers and their influence on the end-to-end speech quality; ESTI TIPHON 22TD47. |
7f05a20ff9261fe60f521819e832606e | 101 329-6 | 5.8.1 Simulation | All results presented in this clause are only based on simulations, all the way from source to receiver. No measurements or real IP packets are involved, but all experiments were done within a discrete event driven simulator environment. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.