hash stringlengths 32 32 | doc_id stringlengths 7 13 | section stringlengths 3 121 | content stringlengths 0 2.2M |
|---|---|---|---|
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.1.1 Technical parameters and implications on spectrum | The radar systems described in the present document use a continuous transmission with frequency modulation. The systems are compliant with the technical parameters of ETSI EN 301 091-2 [i.6]. The following tables list typical parameters for a range of scanning fixed infrastructure radars for safety and security applications. Table 1: H1 Technical parameters Equipment Name Example H1 Frequency Band 76 - 77 GHz Occupied Bandwidth ~940 MHz Modulation FMCW Antenna type Scanning FoV Up to 360° Az, 1,8° El Instrumented Range (see Explanation 1) 300 m Peak Power 47 dBm e.i.r.p. Mean Power 24 dBm e.i.r.p. Antenna Scan Rate 4 Hz Azimuth Beam Width 1,8° Elevation Beam Width 1,8° Duty Cycle (see Explanation 2) 0,5 % Typical Mounting Height above Ground level 2 - 4 m Silent Time (see Explanation 2) 995 milliseconds per second ETSI ETSI TR 104 052 V1.1.1 (2025-06) 16 Table 2: H2 Technical parameters Equipment Name Example H2 Frequency Band 76 - 77 GHz Occupied Bandwidth ~940 MHz Modulation FMCW Antenna type Scanning FoV Up to 360° Az, 3,6° El Instrumented Range (see Explanation 1) 800 m Peak Power 42 dBm e.i.r.p. Mean Power 16 dBm e.i.r.p. Antenna Scan Rate 2 Hz Azimuth Beam Width 1,8° Elevation Beam Width 3,6° Duty Cycle (see Explanation 2) 0,5 % Typical Mounting Height above Ground level 2 - 6 m Silent Time (see Explanation 2) 995 milliseconds per second Table 3: H3 Technical parameters Equipment Name Example H3 Frequency Band 76 - 77 GHz Occupied Bandwidth ~700 - 960 MHz Modulation FMCW Antenna type Scanning FoV Up to 360° Az, 2,8° El Instrumented Range (see Explanation 1) 2 000 m Peak Power 51 dBm e.i.r.p. Mean Power 26 dBm e.i.r.p. Antenna Scan Rate 1 Hz Azimuth Beam Width 0,9° Elevation Beam Width 2,8° Duty Cycle (see Explanation 2) 0,25 % Typical Mounting Height above Ground level 4 - 8 m Silent Time (see Explanation 2) 997,5 milliseconds per second Table 4: Quayside Technical parameters Equipment Name Example Quayside Frequency Band 76 - 77 GHz Occupied Bandwidth ~700 - 960 MHz Modulation FMCW Antenna type Fixed FoV 120° - 160° Az, 30° El Instrumented Range (see Explanation 1) 300 m Peak Power 55 dBm e.i.r.p. max, typically 40 dBm e.i.r.p. Mean Power 50 dBm e.i.r.p. max typically 35 - 40 dBm e.i.r.p. Azimuth Beam Width As FoV Elevation Beam Width As FoV Duty Cycle (depends on use-case) (see Explanation 2) 20 - 50 % Typical Mounting Height above water level 1 - 5 m RCS of target 20 dBsqm for smaller ships up to 40 dBsqm for larger ones ETSI ETSI TR 104 052 V1.1.1 (2025-06) 17 Table 5: Quayside Technical parameters with reduced duty cycle Equipment Name Example Quayside Frequency Band 76 - 77 GHz Occupied Bandwidth ~700 - 960 MHz Modulation FMCW Antenna type Fixed FoV 120° - 160° Az, 30° El Instrumented Range (see Explanation 1) 300 m Peak Power 55 dBm e.i.r.p. max, typically 40 dBm e.i.r.p. Mean Power Typically 35 - 40 dBm e.i.r.p. Azimuth Beam Width As FoV Elevation Beam Width As FoV Duty Cycle (see Explanation 2) 10 % Typical Mounting Height above water level 1 - 5 m RCS of target 20 dBsqm for smaller ships up to 40 dBsqm for larger ones Explanation 1: Instrumented range is the range for which the equipment will deliver output. Small features may not be discoverable at maximum range display. Explanation 2: Duty cycle and silent time in the tables above are as experienced at a given point in the FoV. They include effects of the antenna pattern and rotation as well as signal duty cycle. Radiated powers are expressed as e.i.r.p. The actual power generated in the transmitter is of the order of 10 mW - a property of the semiconductors that are used. The e.i.r.p. is a combination of the transmitter power and the antenna gain in the relevant direction. In the tables above, peak and mean power are radiated powers in a given direction; i.e. they are the illumination as experienced at a point within the FoV. The values are as determined in clauses 4.3.2 and 4.3.3 of ETSI EN 301 091-2 [i.6]. Table 6 shows how the illumination is affected by scanning and fixed antennas. The effects of sector blanking and equipment activity factor have been excluded. Table 6: Effects of scanning and fixed antennas Antenna type Scanning Fixed Signal duty cycle Ton / (Ton+Toff) Ton / (Ton+Toff) Scanning duty cycle (azimuth) Beamwidth (H)/360° 100 % Illumination cycle Combination of signal duty cycle and scanning duty cycle Identical to signal duty cycle Peak power (e.i.r.p.) Peak antenna gain x peak Tx power (maximum in time and maximum in direction) Peak antenna gain x peak Tx power (maximum in time and maximum in direction) Mean power (e.i.r.p.) Time average of radiated power in direction of maximum (average over time and maximum in direction) Time average of radiated power in direction of maximum (average over time and maximum in direction) Mitigation strategy to protect vehicular radar Scanning antenna with a narrow beam illuminates individual points in space only rarely. Restrict installation to non-roadside locations without signal-related limitations, OR Reduced duty cycle to ensure minimum silent time. ETSI ETSI TR 104 052 V1.1.1 (2025-06) 18 |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.1.2 Status of technical parameters | |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.1.2.1 Current ITU and European Common Allocations | |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.1.2.2 Sharing and compatibility studies (if any) already available | In 2017 ECC report 262 [i.2] was published following a co-existence study conducted with SE24. The study related to surveillance radar equipment operating in the 76 - 77 GHz range for fixed transport infrastructure. The fixed radars considered in this study have a mounting location of approximately 5 m above the road surface and 2 - 3m laterally from the first running lane. The executive summary states that the incident power that may be received by an vehicular radar from this fixed radar installation is of the same order of magnitude as can be received from a second vehicular radar. The report concluded that the scanning nature of the FIR contributed to the co-existence with vehicular radar and as such ERC/REC 70-03 [i.3] Annex 5, Note 1 states: "Fixed transportation infrastructure radars have to be of a scanning nature in order to limit the illumination time and ensure a minimum silent time to achieve coexistence with vehicular radar systems". This recommendation is considered within the present document and is discussed in clause 5.2.5. The study considered but did not conclude on other methods that could help mitigate interference including sector blanking (ability to switch off the transmitter in azimuths outside of the area of interest), switching off the transmitters when not sampling, ability to add jitter to the ramp start frequency. It should be noted that the FSSA radar systems discussed in the present document have these mitigation methods available to minimize any potential interference. Other studies of interest are found in ECC Report 222 [i.12] (manned rotorcraft) and ECC Report 315 [i.13] (HDGB-SAR). |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.1.2.3 Sharing and compatibility issues still to be considered | ECC Report 262 [i.2] examined the use of FIR mounted at the roadside and used for TTT applications. One obvious consideration for FIR used for FSSA is whether there would be an increase in illumination directed at the road. The following points are noted: 1) TTT installations can be 1 m from the roadside. FSSA would not be this close; installations are expected to be a minimum of 10 m away. 2) FSSA for an area is concerned with detecting movement inside a perimeter not outside it. Installations are at the edge of the perimeter directed inwards. Sector blanking is used to avoid illumination outside the target area. A restriction preventing the direction of illumination towards roads would be acceptable. 3) Signal format. FSSA use narrower beamwidth and slower scan rates (e.g. 1 or 2 Hz) than TTT. The silence time between beams is greater which would assist compatibility with vehicular radars. 4) Deployment density. For airfield security, the expected density would be 2 sites around a city with 10 units per site. In conclusion, the proponents of the present document believe use of FSSA will not result in a noticeable increase in illumination of vehicles on the road. |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.1.3 Transmitter parameters | |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.1.3.1 Transmitter Output Power / Radiated Power | As noted in clause 7.1.1 above, the RF power generated in the transmitter is of the order of 10 mW. Further details of how this relates to radiated power are given below. |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.1.3.2 Scanning antennas | Three cases of equipment with scanning antennas are presented in clause 7.1.1. When active, these typically operate at close to 100 % duty cycle. ETSI ETSI TR 104 052 V1.1.1 (2025-06) 19 Peak radiated power is typically in the range of +42 dBm to +51 dBm on the antenna boresight. Mean radiated power is in the range of +16 dBm to +26 dBm which can be calculated from to the antenna duty factor. The three cases considered herein differ in their antenna beamwidth characteristics: • Example H1 has an antenna beamwidth of 1,8° × 1,8° (Az x El). • Example H2 has an antenna beamwidth of 1,8° × 3,6° (Az x El). • Example H3 has an antenna beamwidth of 0,9° × 2,8° (Az x El). The actual beam shapes are similar. The pattern for example H2 is representative of all three and is shown in Figure 1 below. Figure 1: Example 2 antenna pattern (source: Navtech Radar) |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.1.3.3 Fixed antennas | The fourth example considered has a fixed antenna of gain approx. 20 dBi. Together with current transmitter technology, a peak e.i.r.p. in the range of 37 - 40 dBm can be achieved. For such a radar, the mean e.i.r.p. is based on the signal duty cycle which is in the range of up to 50 %. The fifth example is almost identical to the fourth with regard to generated power and antenna gain but has a reduced duty cycle of maximum 10 %. With this adjustment, such a radar meets the timing requirements suggested by ECC Report 262 [i.2] for fixed installations to ensure co-existence with vehicular radars. I.e. it is equivalent in this respect to the emissions from a scanning antenna. |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.1.3.4 Operating Frequency | The current operating frequency is in the band 76 - 77 GHz. |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.1.3.5 Bandwidth | The overall bandwidth is defined by the FM sweep pattern. This is typically in the range of 700 - 940 MHz. ETSI ETSI TR 104 052 V1.1.1 (2025-06) 20 |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.1.3.6 Unwanted emissions | Unwanted emissions are within the limits specified by ETSI EN 301 091-2 [i.6] which is aligned with ERC/REC 74-01 [i.15]. |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.1.3.7 Duty Cycle/Mechanical Scanning | A narrow beam scanning antenna only illuminates a given target area intermittently. The radar boresight scans a horizontal plane parallel to the ground. The antenna duty cycle depends on the antenna beam width in azimuth systems and is between 0,25 - 0,5 %. The scan rate is model dependant but ranges between 1 - 4 Hz. A single fixed antenna illuminates the whole FoV whenever the transmitter is active. The duty cycle experienced in the target area depends on the signal transmitted. |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.1.4 Receiver parameters | The infrastructure radar includes either monostatic (single antenna) for transmit and receive (H1) or is a bi-static, dual antenna configuration for the H2 & H3 examples. The radar receiver includes an active mixer that converts the Radio Frequency signal into an Intermediate Frequency range which covers 50 kHz to 5 MHz. The receiver Noise Figure is typically 10 dB at 1 MHz. There is no receive only mode. |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 7.2 Information on relevant standard(s) | The following ETSI standards apply to short range radar equipment using the 76 - 77 GHz band: • ETSI EN 301 091-1 [i.5] "Short Range Devices; Transport and Traffic Telematics (TTT); Radar equipment operating in the 76 GHz to 77 GHz range; Harmonised Standard covering the essential requirements of article 3.2 of Directive 2014/53/EU; Part 1: Ground based vehicular radar". V2.1.1 was published by ETSI in 2017. • ETSI EN 301 091-2 [i.6] "Short Range Devices; Transport and Traffic Telematics (TTT); Radar equipment operating in the 76 GHz to 77 GHz range; Harmonised Standard covering the essential requirements of article 3.2 of Directive 2014/53/EU; Part 2: Fixed infrastructure radar equipment". V2.1.1 was published by ETSI in 2017. This is the standard applicable to the equipment described in the present document. ETSI has current Work Items to revise both the above standards. • ETSI EN 303 360 [i.8] "Short Range Devices; Transport and Traffic Telematics (TTT); Radar equipment operating in the 76 GHz to 77 GHz range; Harmonised Standard covering the essential requirements of article 3.2 of Directive 2014/53/EU; Obstacle Detection Radars for Use on Manned Rotorcraft". V1.1.1 was published by ETSI in 2017. • ETSI EN 303 661 [i.9] "Short Range Devices (SRD); Ground Based Synthetic Aperture Radar (GBSAR) in the frequency range 17,1 GHz to 17,3 GHz and High Definition Ground Based Synthetic Aperture Radar (HD-GBSAR) in the frequency range 76 GHz to 77 GHz; Harmonised Standard for access to radio spectrum". V1.1.1 was published by ETSI in 2024. ETSI ETSI TR 104 052 V1.1.1 (2025-06) 21 |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 8 Radio spectrum request and justification | No change to the spectrum allocations, in terms of frequency bands, power limits, etc., is requested. The purpose of the present document is to seek clarity and a harmonised position on the applications and use cases of radar equipment in the 76 - 77 GHz band. In particular, the request is that applications for security and safety are expressly permitted. At present, most radar equipment in the band is deployed and operated under the heading of Transport and Traffic Telematics (TTT). While many applications are clearly TTT, some applications are wholly or partly for purposes of security and safety. Manufacturers and operators find that some countries permit them anyway and some countries say they are outside the definition of TTT. The situation is further confused by grey area applications where it is difficult, for either manufacturers or administrations, to determine whether they are TTT or not. The proponents of the present document believe that the solution is to expressly permit additional applications such as security and safety in the 76 - 77 GHz band. Further details of the request are given in clause 9. It is accepted that an ETSI Systems Reference Document normally presents technology and use cases and that a spectrum allocation is the role of ECC. In this case, however, the 76 - 77 GHz band is already allocated for radar and the equipment under discussion is the same as that already deployed in the band. The proponents of the present document believe that the additional applications will not cause any compatibility issues. There are various mitigation techniques available including sector blanking, duty cycle, pulse jittering, etc. One compatibility concern is with vehicular radar. This particular concern is solved by geographic separation; the new applications will be away from roads and will not cause any significant increase in illumination of automobiles. The question at heart here is not so much finding new spectrum for a technology, but to create a harmonised market across CEPT for existing equipment. Such harmonisation is of obvious benefit to manufacturers, users and operators, but will also be of benefit in increasing the security and safety of citizens within CEPT countries. |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 9 Regulations | |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 9.1 Current regulations | Fixed infrastructure radars are in Annex 5 (TTT) of ERC/REC 70-03 [i.3] with a corresponding usage restriction in the EU Decision [i.4]. Frequency Band Power / Magnetic Field Spectrum access and mitigation requirements Modulation / occupied bandwidth ECC/ERC Deliverable Notes e1 76 - 77 GHz 55 dBm peak e.i.r.p. (see note) Not specified ECC Report 262 [i.2] 50 dBm average power or 23,5 dBm average power for pulse radar only. For ground based vehicle and infrastructure systems only. The frequency band is also included in Annex 4 of ERC/REC 70-03 [i.3]. NOTE: Fixed transportation infrastructure radars have to be of a scanning nature in order to limit the illumination time and ensure a minimum silent time to achieve coexistence with automotive radar systems. The 76 - 77 GHz band is included in another entry in ERC/REC 70-03 [i.3] Annex 5 for use on rotorcraft and also in Annex 4 for railway use. ETSI ETSI TR 104 052 V1.1.1 (2025-06) 22 |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 9.2 Proposed regulation and justification | |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 9.2.1 Additional applications | The applications described in clause 5.2 (Fixed Security and Safety Applications) raise an interesting question when FSSA is installed in locations such as airports and harbours. Is the application already harmonised as Transport and Traffic Telematics? If a radar illuminates a vehicle or a small boat, does the answer depend on whether the occupants are passengers or terrorists - i.e. whether the issue is transport or security? If they are terrorists, does it depend on whether they intend to disrupt transport or to create another type of mayhem? One of the intentions of the present document is to remove the need for such sophistry. It is requested that the usage restriction on infrastructure radars, which is currently TTT only, be relaxed to include security and safety radiodetermination. Specifically for ERC/REC 70-03 [i.3], where Annex 4 is for fixed railway applications, Annex 5 is for TTT and Annex 6 for Radiodetermination, the requested change might be achieved by a Note in Annex 5, but better would be a new entry in Annex 6. The current scope of Annex 6 is: "This annex covers frequency bands and regulatory as well as informative parameters recommended for SRD radiodetermination applications including Equipment for Detecting Movement and Alert. Radiodetermination is defined as the determination of the position, velocity and/or other characteristics of an object, or the obtaining of information relating to these parameters, by means of the propagation properties of radio waves. Radiodetermination equipment typically conducts measurements to obtain such characteristics." This scope perfectly describes the operation of FSSA. The proposal therefore is to create a new entry in ERC/REC 70-03 [i.3] Annex 6 for 76 - 77 GHz radar equipment, with the same technical parameters as in Annex 5. Frequency Band Power / Magnetic Field Spectrum access and mitigation requirements Modulation / occupied bandwidth ECC/ERC Deliverable Notes x 76 - 77 GHz 55 dBm peak e.i.r.p. (see note) Not specified 50 dBm average power or 23,5 dBm average power for pulse radar only. For fixed radar installations. The frequency band is also included in Annexes 4 and 5. NOTE: To ensure coexistence with Annexes 4 and 5 applications in this band, for fixed radars one of the following mitigations is required: - Not sited at the roadside. - Having a scanning antenna nature or a Low-Duty Cycle (LDC) to ensure similar silent times. A possible definition of "roadside" is discussed below in clause 9.2.3. |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 9.2.2 Scanning antenna restriction | The reasons for mandating scanning antennas are discussed above in clause 5.2.5. It is noted that this provision arose purely for the purposes of mitigation towards vehicular radar. The proposal therefore is that the requirement for antennas of a scanning nature is applied only to roadside installations. One option is to alter Note 1 in ERC/REC 70-03 [i.3] Annex 5. ETSI ETSI TR 104 052 V1.1.1 (2025-06) 23 Frequency Band Power / Magneti c Field Spectrum access and mitigation requirements Modulation / occupied bandwidth ECC/ERC Deliverable Notes e1 76 - 77 GHz 55 dBm peak e.i.r.p. (see note) Not specified ECC Report 262 [i.2] 50 dBm average power or 23,5 dBm average power for pulse radar only. For ground based vehicle and infrastructure systems only. The frequency band is also included in Annexes 4 and 6. NOTE: Roadside fixed infrastructure radars have to be of a scanning nature or transmit with a low duty cycle in order to limit the illumination time and ensure a minimum silent time to achieve coexistence with vehicular radar systems. This requirement does not apply to non-roadside installations. Alternatively, if the proposal in clause 9.2.1 for an entry in ERC/REC 70-03 [i.3] Annex 6 is adopted, this change may not be necessary as the radars described above could operate under that provision. Note it is also proposed to remove the word "transportation". |
c61edf069c94e8cd635d6dbdeb17b902 | 104 052 | 9.2.3 Roadside meaning | Inevitably, the above proposals lead to the question of what counts as roadside and non-roadside. The following ideas are offered for consideration. Road A paved way accessible to the public on which motorised traffic routinely exceeds 100 vehicles per hour and which is not subject to a speed limit of 20 km/h or lower. It is noted that the level of traffic may not be known to the installer/operator, but it is still felt that it is a useful distinction between an active road and one that is hardly used. Roadside Sited within 10 m of a road and with all or part of the 3 dB beamwidth of the antenna intersecting the road within 500 m. Alternatively, a field strength limit could be applied at the road edge, similar to that in ETSI EN 301 091-3 [i.7] for road/rail crossings. ETSI ETSI TR 104 052 V1.1.1 (2025-06) 24 Annex A: Fixed Radar Installations at 76 - 77 GHz A.1 Existing Installation Examples A.1.1 Bristol Airport, United Kingdom Figure A.1: Bristol airport (source: Navtech Radar) Bristol is the UK's 7th largest airport and uses a security radar to monitor a Critical Point Boundary to detect any movement from an open area used as an overflow carpark into the main airport grounds. A.1.2 Ostrava Airport, Czechia Figure A.2: Ostrava airport (source: Navtech Radar) ETSI ETSI TR 104 052 V1.1.1 (2025-06) 25 Ostrava Airport in Czechia has four radars deployed for wide area security sensing to detect vehicles, humans and animals and provide alarms when pre-determined rules are breached. Ostrava airport is Czechia's 3rd largest airport with over 300 000 passengers per year. A.1.3 Bologna Airport, Italy Figure A.3: Bologna airport (source: Navtech Radar) Figure A.4: Bologna airport (source: Navtech Radar) Bologna Airport in Italy has a number of radars deployed for wide area security sensing to detect any intrusion and track the intruder once inside the airport grounds. Bologna airport is Italy's 7th largest airport with over 8,5 million passengers per year. ETSI ETSI TR 104 052 V1.1.1 (2025-06) 26 A.1.4 Jersey Airport Figure A.5: Jersey airport (source: Navtech Radar) The busy Critical Part (CP) at Jersey Airport sits within an unrestricted area of the airport grounds. Unable to protect it with a physical barrier, the airport is using radar with its virtual alarm zones to secure the area. A.1.5 Other Notable Airport Installation Examples • Shannon Airport, Ireland Shannon airport is the third busiest airport in Ireland and has a number of HDR200 & HDR300 radars deployed for a PID system. • Istanbul Grand Airport, Türkiye Istanbul Grand Airport has approximately 10 of HDR200 & 300 radars for a PID system. • San Francisco Airport, USA San Francisco Airport has a number of HDR300 radars for a PID system. • East Midlands Airport, United Kingdom East Midlands Airport near Nottingham has radar systems installed for aircraft Surface Movement monitoring. A.1.6 Minas Gerais A.1.6.1 Stockpile Monitoring A Ground Based Vehicle installation in a mining environment in Brazil. The radar is integrated into the customer stacker reclaimer vehicle to profile the stockpile close to the bucket wheel to improve efficiency. ETSI ETSI TR 104 052 V1.1.1 (2025-06) 27 Figure A.6: Stockpile monitoring (source: Navtech Radar) A.2 Maritime & Shoreside Examples A.2.1 Khalifa Port, Abu Dhabi Khalifa Port is a state-of-the-art, world-leading, deepwater cargo handling facility, and as such has built a solid reputation. To maintain this status, it was recognized that tighter security measures surrounding Khalifa Port, both on land and sea needed to be implemented. Due to its complex geographical location, Khalifa Port required a security solution that would provide perimeter detection both over water and on land. This solution needed to accurately and quickly detect approaching vessels, boats, vehicles, and intruders, and to operate effectively 24/7 and in all weather and light conditions, providing 360° perimeter protection. It quickly became evident that Khalifa Port were not only looking for a suitable technology that could effectively operate across both land and sea, but they also required a solution that they could have complete confidence in. ETSI ETSI TR 104 052 V1.1.1 (2025-06) 28 Figure A.7: Khalifa port (source: Navtech Radar) Figure A.8: Khalifa port (source: Navtech Radar) ETSI ETSI TR 104 052 V1.1.1 (2025-06) 29 A.2.2 Collision Avoidance Radars used to detect ship infrastructure in order to avoid collisions between loading boom and ship. Figure A.9: Crane protection (source: Navtech Radar) A.2.3 Small Target Detection for Inland Marine The radar detects much smaller objects than traditional X-band radar, such as buoys and kayaks, supporting navigation in congested waterways. Figure A.10: Inland marine (source: Navtech Radar) ETSI ETSI TR 104 052 V1.1.1 (2025-06) 30 A.2.4 Quayside collision prevention Vessels manoeuvring too fast and/or on the wrong heading repeatedly cause damage to port infrastructure, piers, sluices, floodgates, fairway limitation/buoy and other ships. Figure A.11: Vessel and infrastructure (source: Robert Bosch GmbH) To prevent this, a radar on the infrastructure side could be foreseen as part of an assistance / protection system. Such system integrates highly robust maritime radar sensors that are installed at various points on the maritime infrastructure and is in direct contact with the crew on the ships via various end devices (signal lamps, displays, etc.). The information on local conditions helps the crew to navigate the ship safely even during adverse weather conditions affecting the ship captain's sight (fog, heavy rain, snow, nighttime, etc.) or affecting the ship's manoeuvrability (heavy winds / gusts). Figure A.12: Vessel and infrastructure (source: Robert Bosch GmbH) Figure A.13: Vessel and infrastructure (source: Robert Bosch GmbH) Such a system could be seen as part of TTT infrastructure but the antenna would be fixed not scanning. ETSI ETSI TR 104 052 V1.1.1 (2025-06) 31 A.2.5 Dynamic positioning High definition radar is especially useful in poor weather or GNSS denied areas/times. Figure A.14: Marine positioning (source: Navtech Radar) The radar provides situational awareness, for small object detection and high-resolution imaging of harbour infrastructure for vessels navigating inland waterways, near-shore environments and ports. E.g. autonomous navigation, berthing, tracking and collision avoidance. Another application is targetless dynamic positioning: to accurately hold position of a vessel when operating around infrastructure in ports and offshore e.g. windfarms, oil rigs, docking of autonomous vessels. "Targetless" means the system is self-contained and does not rely on co-operative devices on the infrastructure. Many ETSI members believe such an application falls under TTT. Some members, however, have encountered arguments that is it predominantly safety and thus not necessarily TTT. ETSI ETSI TR 104 052 V1.1.1 (2025-06) 32 Annex B: Change history Date Version Information about changes 3/2/24 V1.1.1_0.0.1 First draft 12/2/24 V1.1.1_0.0.2 For rapporteur meeting 15/2/24 11/3/24 V1.1.1_0.0.3 For TGUWB#67 5/5/24 V1.1.1_0.0.4 For rapporteur meeting 6/5/24 3/6/24 V1.1.1_0.0.5 For TGUWB#68 14/6/24 V1.1.1_0.0.6 Following TGUWB#68 25/8/24 V1.1.1_0.0.7 For rapporteur meeting 27/8/24 31/10/24 V1.1.1_0.0.8 Following TGUWB/TGSRR meeting on GBV 28/11/24 10/11/24 V1.1.1_0.0.9 Following rapporteur meeting 4/11/24 11/11/24 V1.1.1_0.0.10 18/11/24 V1.1.1_0.0.11 4/12/24 V1.1.1_0.0.12 Output of TGUWB#70 17/1/25 V1.1.1_0.0.13 Output of TGUWB#70bisD2 29/1/25 V1.1.1_0.0.14 Output of TGUWB#70bisD3 17/2/25 V1.1.1_1.0.0 TG UWB#71 Clean and accepted version 26/03/25 V1.1.1_1.0.1 Editorial corrections after internal quality check ETSI ETSI TR 104 052 V1.1.1 (2025-06) 33 History Document history V1.1.1 June 2025 Publication |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 1 Scope | The present document analyses the mechanisms that use cryptography in the specifications under ETSI TC SET responsibility. It describes the potential changes for a responsible industry transition to Quantum-Safe technology. |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 2 References | |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 2.1 Normative references | Normative references are not applicable in the present document. |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 2.2 Informative references | References are either specific (identified by date of publication and/or edition number or version number) or non-specific. For specific references, only the cited version applies. For non-specific references, the latest version of the referenced document (including any amendments) applies. • In the case of a reference to a TC SET document, a non-specific reference implicitly refers to the latest version of that document in the same Release as the present document. NOTE: While any hyperlinks included in this clause were valid at the time of publication ETSI cannot guarantee their long term validity. The following referenced documents may be useful in implementing an ETSI deliverable or add to the reader's understanding, but are not required for conformance to the present document. [i.1] ETSI TS 102 224: "Smart Cards; Security mechanisms for UICC based Applications - Functional requirements". [i.2] ETSI TS 102 225: "Smart Cards; Secured packet structure for UICC based applications". [i.3] ETSI TS 102 226: "Smart Cards; Remote APDU structure for UICC based applications". [i.4] Peter W. Shor: "Polynomial-Time Algorithms for Prime Factorization and Discrete Logarithms on a Quantum Computer". [i.5] Lov K. Grover: "A fast quantum mechanical algorithm for database search". [i.6] ANSSI: "ANSSI views on the Post-Quantum Cryptography transition". [i.7] BSI: "Quantum-safe cryptography – fundamentals, current developments and recommendations". [i.8] NIST: "Post-Quantum Cryptography, Frequently Asked Questions". [i.9] NSA statement: "Suite B Cryptography". [i.10] GlobalPlatform: "GlobalPlatform Technology, Confidential Card Content Management Card Specification v2.3 - Amendment A", Version 1.2. [i.11] GlobalPlatform: "Remote Application Management over HTTP, Card Specification v2.3 - Amendment B", Version 1.2. [i.12] GlobalPlatform: "GlobalPlatform Card Technology, Secure Channel Protocol '03', Card Specification v2.3 - Amendment D", Version 1.2. [i.13] GlobalPlatform: "GlobalPlatform Secure Channel Protocol '04' – Amendment K", Version 1.0.2. [i.14] GlobalPlatform: "GlobalPlatform Card Specification v2.3.1". [i.15] ENISA: "Post-Quantum Cryptography: Current state and quantum mitigation". ETSI ETSI TR 104 005 V1.2.1 (2026-01) 7 |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 3 Definition of terms, symbols and abbreviations | |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 3.1 Terms | Void. |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 3.2 Symbols | Void. |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 3.3 Abbreviations | For the purposes of the present document, the following abbreviations apply: AES Advanced Encryption Standard APDU Application Protocol Data Unit DAP Data Authentication Pattern ECC Elliptic Curve Cryptography KIc Key and algorithm Identifier for ciphering KID Key and algorithm IDentifier for RC/CC/DS PQC Post-Quantum Cryptography TLS Transport Layer Security |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 4 General presentation | Cryptography has become part of our daily life, securing most of our electronic activities ranging from web browsing to mobile communications or payments. Although cryptography is a key component of digital security, it has never experienced the ever-faster cycle of attacks and patches that characterizes cybersecurity in general. Quite the contrary, cryptography has evolved quietly, without significant hitches, as epitomized by the omnipresence in current systems of 45-year-old protocols such as Diffie-Hellman key exchange or RSA signatures. Arguably, this stability is due to a good understanding of the mathematical foundations of cryptographic systems, which enables to precisely assess their concrete security level but also to identify in advance potential new threats. Several decades ago, this approach led to identify vulnerabilities of those systems to quantum algorithms. The current public key cryptographic algorithms are proven to be compromised by the Shor's and Grover's algorithms (see note 1). The impacts on symmetric key cryptographic algorithms are still being analysed by security agencies and consensus for recommending an increase in key size has not been reached yet (see note 2). NOTE 1: See Peter W. Shor: "Polynomial-Time Algorithms for Prime Factorization and Discrete Logarithms on a Quantum Computer" [i.4] and Lov K. Grover: "A fast quantum mechanical algorithm for database search" [i.5]. NOTE 2: Some security agencies (e.g. ANSSI) recommend to double the key size whereas some others (e.g. NIST, BSI) suggest that no increase is needed, see [i.6], [i.7] and [i.8]. For a long time, this quantum threat has remained elusive because of the lack of large-scale quantum computers required to run those quantum algorithms. This situation is expected to change because of the investment of many companies supported by important governmental fundings. This has led to significant advances in the design of quantum computers which would make the threat tangible from 2030 onwards (see BSI: "Quantum-safe cryptography – fundamentals, current developments and recommendations" [i.7]). In 2015, NSA published a statement recommending to start planning the transition to quantum resistant cryptography, that is, cryptography immune to quantum algorithms (see NSA statement: "Suite B Cryptography" [i.9]). Since then, most of the security agencies worldwide have issued similar statements and recommendations to move to so-called "post-quantum algorithms". This led NIST to launch in 2017 a competition to select post-quantum standards for public key encryption and digital signatures. This competition ended in 2022 and the first standards have been published in 2024. In parallel, similar initiatives were launched by China and South Korea. ETSI ETSI TR 104 005 V1.2.1 (2026-01) 8 Post-quantum cryptography is the solution to the quantum threat. However, many security agencies and experts are reluctant to rely exclusively on those new algorithms because they have not been as scrutinized as classical ones. They therefore promote a phased transition where post-quantum algorithms will be used together with classical ones in a first stage so as not to weaken current security level. This "hybrid" approach is supported by several agencies, e.g. BSI, ANSSI, ENISA, among others (see [i.6], [i.7] and [i.15]). Although there is uncertainty surrounding the realization of large-scale quantum computers and the roadmaps mentioned above suggesting that it should not happen before 2030, migration process needs to start immediately. Indeed: • Both ANSSI and BSI recommendations state that retroactive attacks cannot be ruled out. An example of retroactive attacks is the "store now decrypt later" attack, where data is gathered now for later decryption, when quantum computers are available. NOTE 3: See BSI: "Quantum-safe cryptography – fundamentals, current developments and recommendations" [i.7] and ANSSI2: "ANSSI views on the Post-Quantum Cryptography transition" [i.6]. • Public key-based user authentication algorithms are not subject to retroactive attack. This means that classical algorithms could be still used, waiting for significant advances in the area of quantum computing before migrating to post-quantum ones. However, this assumes that the devices support migration features, which again calls for considering transition to post-quantum cryptography as soon as possible, at least for devices whose lifespan extends beyond 2030. The quantum threat is likely to lead to a complete overhaul of cryptographic systems. Even if no post-quantum standards are currently available, transition to post-quantum cryptography can already be initiated by inventorying cryptographic components in standards along with the assets they protect. |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 5 Analysis of ETSI TC SET specifications | |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 5.1 ETSI TS 102 224 | ETSI TS 102 224 [i.1] describes the functional requirements of security mechanisms in conjunction with the Card Application Toolkit for the interface between a Network Entity and a UICC. Regarding the cryptographic mechanisms, ETSI TS 102 224 [i.1], clause 6.2.2,contains only two high level requirements which are still valid in the context of post-quantum cryptography: "When the security of a cryptographic algorithm from the technical specification is considered compromised, it may be deprecated. When a new cryptographic algorithm becomes state of the art, its addition to the implementation specification shall be considered." At the time of the publication, no particular changes are foreseen for transitioning ETSI TS 102 224 [i.1] to post-quantum cryptography. However, clause 6.2.3 of ETSI TS 102 224 [i.1] related to the recommended combinations of cryptographic mechanisms needs to be evaluated. |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 5.2 ETSI TS 102 225 | ETSI TS 102 225 [i.2] specifies the structure of Secured Packets for different transport and security mechanisms. The following impacts are seen, together with remediation proposals for transitioning ETSI TS 102 225 [i.2] to post-quantum cryptography. ETSI ETSI TR 104 005 V1.2.1 (2026-01) 9 Table 1 Impacts Requirements to become Quantum-Safe Coding of the KIc (clause 5.1.2) Based on symmetric encryption: • AES with length of 128, 194 or 256 bits. AES key size security level is still under discussion by various cyber security agencies. Coding of the KID (clause 5.1.3) Based on symmetric encryption: • AES with length of 128, 194 or 256 bits. AES key size security level is still under discussion by various cyber security agencies. |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 5.3 ETSI TS 102 226 | |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 5.3.1 Introduction | ETSI TS 102 226 [i.3] defines the remote management of the UICC based on the secured packet structures specified in ETSI TS 102 225 [i.2], i.e.: • SMS and CAT_TP based packet structures, also known as SCP80; • HTTP-based using TLS cipher suites, also known as SCP81 and defined by GlobalPlatform in Amendment B to the GlobalPlatform Card Specification [i.11]. ETSI TS 102 226 [i.3] specifies the APDU format for remote management, as well as: • A set of commands coded according to this APDU structure and used in the remote file management on the UICC; • A set of commands coded according to this APDU structure and used in the remote application management on the UICC, based on the GlobalPlatform Card Specifications. |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 5.3.2 Analysis of the current content of ETSI TS 102 226 | The following impacts are seen, together with remediation proposals for transitioning ETSI TS 102 226 [i.3] to post-quantum cryptography. Table 2 Impacts Requirements to become Quantum-Safe Use of SCP81 According to GlobalPlatform PQC roadmap (see note), an update of Amendment B to the GlobalPlatform Card Specification [i.11] is not addressed yet but would rely on official cipher suites published by IETF in the future (RFC). Remote Application Management (ETSI TS 102 226 [i.3], clause 8) Cryptographic computations, e.g. DAP, are based on AES with length of 128, 194 or 256 bits. According to GlobalPlatform PQC roadmap (see note), an update of the GlobalPlatform Card Specification [i.14] introducing PQC is planned for beginning of 2026. AES key size security level is still under discussion by various cyber security agencies. Confidential loading (ETSI TS 102 226 [i.3], clause 10.1) Cryptographic computations are based on AES with length of 128, 194 or 256 bits. AES key size security level is still under discussion by various cyber security agencies. Additional application provider security (ETSI TS 102 226 [i.3], clause 10.2) Based on SCP03 defined in GlobalPlatform Amendment D [i.12]. According to GlobalPlatform PQC roadmap (see note), SCP03 is based on a symmetric algorithm (AES), is widely used and is considered quantum-safe. Then, SCP03 will not be deprecated for now and no update of Amendment D to the GlobalPlatform Card Specification [i.12] is expected. ETSI ETSI TR 104 005 V1.2.1 (2026-01) 10 Impacts Requirements to become Quantum-Safe Confidential setup of Security Domains (ETSI TS 102 226 [i.3], clause 10.3) Refers scenarios defined in GlobalPlatform Amendment A [i.10]: • Scenario #2.B (Push Model), based on RSA; • Scenario #1 (Pull Model) using the public key scheme, based on RSA; • Scenario #3 using ECKA-EG. According to GlobalPlatform PQC roadmap (see note), an update of Amendment A to the GlobalPlatform Card Specification [i.10] is planned in the middle of 2026. This update should introduce new PQC scenarios, should not deprecate ECC scenarios and might deprecate RSA scenarios or require a minimum key size (e.g. 3K). NOTE: Based on the liaison statement exchange between ETSI TC SET and GlobalPlatform about PQC in spring 2025 (in document SET(25)000017). |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 5.3.3 Other areas of improvement | |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 5.3.3.1 Secure Channel Protocol '04' (SCP04) | Secure Channel Protocol '04' (SCP04), defined by GlobalPlatform in Amendment K to the GlobalPlatform Card Specification [i.13] is designed to be crypto agile, i.e. algorithms may be replaced with less effort by other algorithms when vulnerabilities are found, or more secure algorithms become available. The current version of Amendment K to the GlobalPlatform Card Specification [i.13] includes algorithms SM3/SM4 in addition to AES. With respect to clause 10.2 [i.3], Additional application provider security, the details of the encapsulation of SCP04 in SCP80, SCP81 and SCP82 would have to be defined in updates of GP UICC configuration and ETSI TS 102 226 [i.3]. |
c07fbe808386ef471b0fb7f16ff9234c | 104 005 | 6 Conclusion and way forward | The present document provides analysis regarding the mechanisms that use cryptography in the specifications under ETSI TC SET responsibility. Potential changes for a responsible transition to Quantum-Safe technology are described. However, the impact on performance which may be caused by the introduction of Quantum-Safe mechanisms is not considered. Such effects may require the adaptation of the current mechanisms, i.e. a one-to-one replacement may not be feasible in all cases. For the mechanisms that use symmetric key cryptographic algorithms, the impacts are still being analysed by security agencies. ETSI TC SET needs to wait for their recommendations to make the appropriate changes to its own documents. For the mechanisms that use asymmetric cryptographic algorithms, the specifications under ETSI TC SET responsibility rely on GlobalPlatform specifications. GlobalPlatform has planned updates of their specifications in 2026 (according to SET(25)000017: LS response to ETSI LS SET(24)000157r1 about PQC roadmap). ETSI TC SET needs to closely follow the publication of these updates and make the appropriate changes to its own documents. ETSI ETSI TR 104 005 V1.2.1 (2026-01) 11 Annex A: Bibliography • ETSI GR QSC 001: "Quantum-Safe Cryptography (QSC); Quantum-safe algorithmic framework". • ETSI GR QSC 003: "Quantum Safe Cryptography; Case Studies and Deployment Scenarios". • ETSI GR QSC 004: "Quantum-Safe Cryptography; Quantum-Safe threat assessment". • ETSI GR QSC 006: "Quantum-Safe Cryptography (QSC); Limits to Quantum Computing applied to symmetric key sizes". • ETSI TR 103 619: "CYBER; Migration strategies and recommendations to Quantum Safe schemes". ETSI ETSI TR 104 005 V1.2.1 (2026-01) 12 History Version Date Status V1.1.1 July 2025 Publication V1.2.1 January 2026 Publication |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 1 Scope | The use of AI to facilitate the use cases may cause AI security and privacy issues specific to the telecom industry. The scope of this proposed work item will be to investigate security and privacy issues related to the use of AI in the telecom industry sector. Harmonisation with 3GPP work in SA1, SA2, and SA3 is anticipated. Key AI use cases in telecom networks are (non-exhaustive list): • Network as a service. • Network optimization. • Network planning and upgrades. • Automating security operations (anomaly detection, planning mitigation and response). This investigation may use but is not limited to the Network Operations Lifecycle Phases methodology. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 2 References | |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 2.1 Normative references | Normative references are not applicable in the present document. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 2.2 Informative references | References are either specific (identified by date of publication and/or edition number or version number) or non-specific. For specific references, only the cited version applies. For non-specific references, the latest version of the referenced document (including any amendments) applies. NOTE: While any hyperlinks included in this clause were valid at the time of publication, ETSI cannot guarantee their long-term validity. The following referenced documents may be useful in implementing an ETSI deliverable or add to the reader's understanding, but are not required for conformance to the present document. [i.1] 3GPP TS 23.228: "Architecture enhancements for 5G System (5GS) to support network data analytics services (Release 18)". [i.2] 3GPP TS 28.104: "Management and orchestration; Management Data Analytics (MDA) (Release 17)". [i.3] 3GPP TR 37.817: "Study on enhancement for data collection for NR and ENDC". [i.4] ETSI TS 138 423: "5G; NG-RAN; Xn Application Protocol (XnAP) (3GPP TS 38.423 version 18.5.0 Release 18)". [i.5] Open Source MANO (OSM) - ETSI. [i.6] Planning for Network Day 0, 1, and 2 Tasks and Stumbling Blocks. [i.7] The difference between day-0, day-1, and day-2 operations. [i.8] Day 0, Day 1, Day 2 Operations: Putting it All Together on Day 2. [i.9] ETSI GR NFV-EVE 022: "Network Functions Virtualisation (NFV) Release 5; Architectural Framework; Report on VNF configuration". [i.10] Pialla, G., Ismail Fawaz, H., Devanne, M. et al.: "Time series adversarial attacks: an investigation of smooth perturbations and defense approaches". Int J Data Sci Anal (2023). ETSI ETSI TR 104 051 V1.1.1 (2025-06) 7 [i.11] Siyang Lu, Mingquan Wang, Dongdong Wang, Xiang Wei, Sizhe Xiao, Zhiwei Wang, Ningning Han, Liqiang Wang: "Black-box attacks against log anomaly detection with adversarial examples", Information Sciences, Volume 619, 2023, Pages 249-262, ISSN 0020-0255. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 3 Definition of terms, symbols and abbreviations | |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 3.1 Terms | |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 3.1.1 Network Operations Lifecycle Phases | The present document adopts a commonly used nomenclature used for describing activities pertaining to the commissioning of a service, namely Day 0 - Day N. This approach and terms have been used in the ETSI Open Source MANO [i.5] project and their application explained in [i.6], [i.7], and [i.8] to describe the onboarding of Containerized Network Functions, with some concrete use cases to be found in [i.9]. For the scope of the present document, the terms have been redefined as below. Day 0 The planning and evaluation phase for a new telecommunications network encompassing all decision-making processes prior to deployment activities. Day 1 The deployment phase for a new network, spanning all activities necessary for the commissioning of infrastructure, systems, and supply chains required for network operations. Day 2 The steady-state operations and maintenance phase for a network, post initial installation, until the final decommissioning of the network. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 3.1.2 NIST AI Attack Taxonomy | AI supply chain: manipulation of training data or AI/ML model or AI/ML supporting software libraries evasion: manipulating data which results in misclassification or no detection poisoning : manipulating training data which results in model learning incorrectly privacy: extracting sensitive information model was trained on prompt injection: submission of malicious prompts to an AI system either directly or through ingestion sources |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 3.1.3 NIST AI Attacker Goals | abuse violation: abuse of a deployed AI/ML model to achieve attacker goals availability breakdown: degradation of AI/ML model performance during deployment integrity violation: erosion of model integrity to elicit incorrect results either through evasion or poisoning privacy compromise: discovery of information pertaining to training data or model characteristics |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 3.2 Symbols | Void. ETSI ETSI TR 104 051 V1.1.1 (2025-06) 8 |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 3.3 Abbreviations | ADRF Analytics Data Repository Function AF Application Function AI Artificial Intelligence AI/ML AF AI/ML Application Function AKMA Authentication and Key Management for Applications AMF Access Management Function AN Access Network AnLF Analytics Logical Function AS Access Stratum CI/CD Continuous Integration and Continuous Delivery CN Core Network CoA Course of Action Cp Control Plane CP Control Plane CPU Central Processing Unit DCCF Data Collection Coordination Function DDoS Distributed Denial of Service E2E End-to-End GBA Generic Bootstrapping Architecture GenAI Generative AI GPU Graphic Processing Unit IaC Infrastructure as Code IN Intelligent Network KPI Key Performance Indicator LLM Large Language Model MANO Management And Orchestration MDA Management Data Analytics MDAS Management Data Analytics Service MDT Minimization of Drive Test MFAF Messaging Framework Adaptor Function ML Machine Learning MTLF Model Training logical function NAS Non-Access Stratum NEF Network Exposure Function NF Network Function NN Neural Network NOC Network Operations Centre NRF Network Repository Function NSACF Network Slice Admission Control Function NWDAF NetWork Data Analytics Function OAM Operations, Administration and Maintenance PCA Principal Component Analysis PCF Policy Control Function PDCP Packet Data Convergence Protocol QoE Quality of Experience QoS Quality of Service RAG Retrieval Augmented Generation RAN Radio Access Network RAT Radio Access Technology RCA Root Cause Analysis RCEF RRC Connection Establishment Failure RLF Radio Link Failure RRC Radio Resource Control SBI Service Based Interface SIP Session Initiation Protocol SLA Service Level Agreement SMART Self-Monitoring, Analysis, and Reporting Technology SMF Session Management Function SOC Security Operations Centre ETSI ETSI TR 104 051 V1.1.1 (2025-06) 9 t-SNE t-distributed Stochastic Neighbour Embedding UDM Unified Data Management UDR Unified Data Repository UE User Equipment UP User Plane |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 4 Convention Description | |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 4.1 Notation | For the purpose of the present document, the following notations apply: <information> stands for the "information" been exchanged or transmitted between different modules or via interfaces. [security component] stands for the "security components" described in clause 7 that has involved in the interactive procedures. (optional) means the description followed by is the alternative that has enhanced effect but more strict requirements compared. It may be referred to specific situations when implementing a certain security component or a security mechanism. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 5 Overview | 5.1 Use of Generative AI vs. traditional AI in telecom providers' networks In many cases, traditional AI (as opposed to Generative AI (GenAI)) might be sufficient in telecom providers' networks. Many of the use cases described in further clauses (e.g. Anomaly detection, Customer churn prediction, Predictive maintenance, Root cause identification, System monitoring, Ticket classification, and routing) may be supported by traditional AI. While GenAI may be suitable for interactive experiences, such as customer care and postmortem assessment of a network error, a lot of other optimization efforts - such as initial detection of errors and root cause analysis - might not need more than traditional ML models. However, when, for example, traditional AI detects a new cyber threat and new firewall rules need to be written to address it or a new threat signature is discovered for future use by traditional AI, GenAI may be employed to generate and deploy these rules and/or signatures. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 5.2 ML functionality in telecom providers' networks | |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 5.2.1 Data collection and preparation mechanisms | In the context of 5G / 3GPP-enabled standards, data collection forms part of the core functionality of the Management Data Analytics (MDA) capability of the network. More generally, the telecom provider should ensure the availability of structured, real-time operational and performance characteristics of their network, such as QoS/QoE metrics, throughput, and measured network latency. Additionally, unstructured data sources such as natural-language trouble tickets generated increasingly represent viable sources of information for AI models, especially in conjunction with the aforementioned structured information. ETSI ETSI TR 104 051 V1.1.1 (2025-06) 10 Depending on the nature of the AI/ML models employed, several pre-processing steps may be required. Structured, quantitative data may require statistical transformations such as normalization to contain their co-domain. In cases where a large number of features (i.e. input parameters) are available, dimensionality reduction techniques such as Principal Component Analysis (PCA), or t-distributed Stochastic Neighbour Embedding (t-SNE) may be employed to reduce the feature space. Natural language data typically undergoes a string of techniques such as normalization (e.g. non-printed character removal or white space normalization), deduplication, and - depending on the use case - removal of private data, pseudonymization, or anonymization for security and privacy compliance. The following steps may include lemmatization, grouping related words into singular concepts (lemmas), and vectorization, which is the conversion of these lemmas to mathematical vector representations. This subsequently allows for sentiment, syntax, and semantic analysis, to determine the premise of a given text. The interfaces provided by Large Language Models (LLMs) typically perform these steps transparently. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 5.2.2 Model engineering and evaluation mechanisms | Model engineering refers to the process of designing ML "pipelines" that convert raw data into actionable inferences by performing a set of transformations. This process is executed based on the insights of domain experts. In the context of a telecommunications network, this translates to the Network Operations Centre (NOC) and Security Operations Centre (SOC) domains. The expert knowledge, along with statistical methods like correlation analysis, is used to create and select a set of features (e.g. mean and variance of some numerical value for a given time window or frequency of words in textual data) from the raw data, that are most meaningful for a given problem that the model aims to solve. Some features may be "synthetic," i.e. created from a combination of existing features. In the training phase, the prepared data is used to obtain model parameters, typically by minimizing a certain error function. The model evaluation process serves to minimize generalization error, i.e. the error exhibited by the model when faced with data set that is different from the training data set. This is typically estimated using validation data sets, reserved within the existing data, and fed to the model post-training. The metrics used to measure this error depend on the nature of the parameter to be predicted. Since telco systems are considered critical infrastructure, a separate test bed may be required for the training and evaluation of any new model. While real-time operational data may be used to train AI models, during the training and evaluation phase, these models are decoupled from downstream systems. For offline models, i.e. those with separate training and implementation phases, evaluation is performed entirely within the established test-bed. Online models, on the other hand, undergo continuous evaluation, both within, and outside the test environment. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 5.2.3 Model deployment mechanisms | |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 5.2.3.1 Introduction | Model deployment is often referred to as model distribution. To take advantage of the operators' Core Network (CN) computing resources, a model can be engineered and evaluated at the core network nodes or AI/ML Application Functions (AI/ML AF) outside of the operator's core network and deployed for execution to the User Equipment (UE) nodes. There are two fundamental ways to distribute information between CN or AF and UE(s). They are utilizing the Control Plane (CP) and User Plane (UP). In addition, a hybrid distribution would utilize both CP and UP. Clauses 5.2.3.2, 5.2.3.3, and 5.2.3.4 discuss possible model distribution use cases. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 5.2.3.2 Model distribution using CP | The main advantage of model distribution using CP is its confidentiality, integrity, and replay protection over the air while protected with NAS security. Another advantage is full operator's control over CP and NAS security. However, a potentially large size of transferred models can adversely affect CP availability for its main purpose of assuring signalling for mobile communications. ETSI ETSI TR 104 051 V1.1.1 (2025-06) 11 C C C © 2020 InterDigital, Inc. All Rights Reserved. 3 UE CN AI/ML AF (R)AN 1. Decision to initiate Model Distribution 2. Selected model forwarded 3. Key or key material derivation from CP (e.g., NAS) security context - . 4. Model confidentiality/integrity/replay protected using CP (e.g., NAS) mechanism 5. Protected model distribution CP 7. Successful access/use of the model within the environmental and contextual constrains 8. Model transfer and access success 6. Model security verification Figure 5.2.3.2-1: Illustration of the CP-based approach to the ML model distribution In Figure 5.2.3.2-1, steps 2 and 8 are realized over the pre-provisioned secure communication channel between the operator's Core Network (CN) and the trusted AI/ML Application Function (AI/ML AF). In step 3, the CN derives model distribution security context from the existing CN security context (e.g. NAS security context) and uses this security context for protection of ML model distribution from the edge of CN through RAN and into the UE. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 5.2.3.3 Model distribution using UP | UP use for model distribution has the advantage of being able to accommodate large ML model sizes. UP might be used with Application Layer security (e.g. end-to-end) and as such can be easily outsourced by an operator. However, being opaque to the operator is one of the main disadvantages. The user QoE is mainly attributable to the serving mobile operator. At the same time, the model that is distributed over UP cannot be controlled by the operator. One of the examples of such discrepancy could be the case when the model that optimizes handovers is distributed from the entity that is not under the control of the operator using the means that are not under the operator's control (i.e. end-to-end confidentiality protected UP). However, when users' QoE is not satisfactory because of slow or failed handovers, it is the serving operator that will be blamed for poor QoE. Various ML model ownership and custodial arrangements are also not fully supported when a model is distributed using UP. Figure 5.2.3.3-1 illustrates the UP-based approach to the ML model distribution. ETSI ETSI TR 104 051 V1.1.1 (2025-06) 12 C UE CN AI/ML AF (R)AN 1. Decision to initiate Model Distribution 2. Request for key/Key material 3. Key or key material derivation with environmental/contextual info bound to the key(s). 4. Key/Key material provided 5. Model confidentiality/integrity/replay protected using received key or key material 6. Protected model distribution UP 8. Successful access/use of the model within the environmental and contextual constrains 7. Model security verification Figure 5.2.3.3-1: Illustration of the UP-based approach to the ML model distribution |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 5.2.3.4 Hybrid model distribution using CP and UP | Hybrid model distribution aims to remediate the CP-only and UP-only modes. Such a mode uses CP to set the cryptographic model distribution control that is shared between the home operator, serving operator, and model owner/custodian. After the CP-assisted framework for model transfer is set, the UP is utilized to distribute the ML model protected by the CP-assisted cryptographic framework. C C UP (CP as an option) C © 2020 InterDigital, Inc. All Rights Reserved. 3 UE CN AI/ML AF (R)AN 1. Decision to initiate Model Distribution 2. Request for key/Key material 3. Key or key material derivation (GBA, AKMA, or other CP-based method - . 4. Key/Key material provided 5. Model confidentiality/integrity/replay protected using received key or key material 6. Protected model distribution CP 8. Successful access/use of the model within the environmental and contextual constrains CP 9. Model transfer and access success 7. Model security verification Figure 5.2.3.4-1: Illustration of the hybrid approach to the ML model distribution ETSI ETSI TR 104 051 V1.1.1 (2025-06) 13 5.2.4 Model operation (i.e. use of the ML model for prediction and inference) mechanisms AI agents are software systems that use AI models along with interfaces to other software, hardware, humans, or other agents to pursue goals and complete tasks on behalf of users. AI agents may show reasoning, planning, and memory and have a level of autonomy to make decisions, learn, and adapt. Such agents may be deployed within numerous functions of a network, for e.g. triggering pre-defined automated workflows to redistribute traffic on prediction of increased congestion. 6 AI/ML use cases for telecom and their impact on AI security |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.1 System monitoring | |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.1.1 Introduction | System monitoring could be seen as a prerequisite to gathering information on the status of the system to perform further investigations as described in the subsequent clauses. The data and information that is monitored may be on different layers in the system, i.e. in the general availability of resources. For example, such information may be Network Functions (NF) or load situations in terms of memory consumption and throughput. The monitoring can also span into protocol states, malformed packets (header values out of range, new header parameters, etc.) protocol failures, and the respective error codes on different protocols such as RRC, PDCP, NAS, SBI, etc., or on application layer in case operator native applications like SIP/IMS media are involved. Further information can be collected on a per NF basis or in a more fine- granular way on a per UE basis, depending on the point of data collection. The 5G system already is designed to some extent with some monitoring capabilities: the Network Data Analytics Function (NWDAF) [i.1] can collect data from various sources i.e. via subscription to events from AMF, SMF, PCF, UDM, NSACF, AF (directly or via NEF) and OAM, from Data Collection Coordination Function (DCCF), from Messaging Framework Adaptor Function (MFAF), from data repositories (e.g. UDR via UDM for subscriber-related information) and ADRF (Analytics Data Repository Function), or NRF for NF-related information. The NWDAF may contain one or two of the Analytics Logical Functions (AnLF) for deriving analytics information and the Model Training Logical Function (MTLF) for training machine learning models. Further, the 5G system comprises the MDA capability [i.2] for mobile networks and services management and orchestration which can consume the analytics provided by the NWDAF. The MDA has system monitoring capabilities and can process and analyse a large number of different data related to network and service events and status. The data monitored by the MDA can be current or historic data, examples are performance measurements, KPIs, Trace data including MDT/RLF/RCEF reports, QoE reports, alarms, configuration data, network analytics data, and service experience data from AFs, etc. The MDA provides its output by the MDAS (Management Data Analytics Service) producer to the consumer NFs, which contains analytics output in terms of statistics or predictions, root cause analysis issues, and, potentially, recommendations for enabling necessary actions for network and service operations. From the Lifecycle Phase perspective, this clause deals with Day 2. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.1.2 Anomaly detection | System monitoring is a prerequisite for Anomaly Detection and for developing the definition of the normal state and behaviour of the system to draw a comparison to events or behaviour that deviate from the normal state. There is some basic Anomaly Detection already defined within the scope of the network analytics and the management data analytics features. The NWDAF can identify a group of UEs or a specific UE with abnormal behaviour, where abnormal is defined as the UE being either hijacked or misused, running malicious applications on the UE. The NWDAF can provide analytics on the following two analytics types based on the exception identity parameters: • Mobility-related analytics: Unexpected UE location, Ping-ponging across neighbouring cells, Unexpected wakeup, Unexpected radio link failures. ETSI ETSI TR 104 051 V1.1.1 (2025-06) 14 • Communication-related analytics: Unexpected long-live/large rate flows, Unexpected wakeup, Suspicion of DDoS attack, Wrong destination address, Too frequent Service Access. Upon detection of abnormal behaviour based on the analytics of the exception identity parameters, the NWDAF informs the consumer e.g. the MDA, about the analytics results. The MDA can utilize AI/ML technologies for the analytics and the MDA function may be deployed per individual MDA capabilities, the relevant ML entities are used for inference. Considering the vast amount of input parameters the MDA can collect from various sources, it can identify current issues in terms of network performance and services. The MDA has the capability of assisted fault management, i.e. it can monitor the status and current behaviour of various network functions and predict potential network failures in advance based on the running trends. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.1.3 Root cause identification | The analytics of the MDA can already contain specific root cause identification of the detected issues. There are already several requirements on the MDA capabilities to provide a root cause identification e.g. network slice throughput issue(s), E2E latency issue, network slice load issues and recommended actions, energy efficiency issue, and mobility performance issues. Such detection and identification can be performed on both, raw operational metrics or text logs, generated by the deployed and instantiated NFs. To achieve this, LLMs or other autocorrelative models may be deployed for log-based detection, streamlined classification, root cause clustering, or regression purposes. A combination of these approaches may be used in an ensemble for greater detection performance. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.1.4 Predictive maintenance | Information from the MDA pertaining to the operational characteristics of the underlying network infrastructure may be used as inputs to NN models trained to predict expected hardware failure times and flag components for pre-emptive replacement or repair. For example, a model may be used to ingest Self-Monitoring, Analysis, and Reporting Technology (SMART) metrics from Hard Disk Drive arrays, monitor their deterioration, and provide proactive drive replacement schedules. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.2 Intelligent networks | |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.2.1 Introduction | Traditionally, in the telecom context, Intelligent Network (IN) allows functionality to be distributed flexibly at a variety of nodes in and outside the network and allows the architecture to be modified to control the services. AI in telecom networking offers several key advantages that are transforming how networks are managed and operated. AI significantly boosts network efficiency by automating routine and complex tasks. This automation leads to faster resolution of issues, more efficient resource allocation, and reduced operational overhead. In turn, high-performance networking is a critical component of the technology infrastructure that enables AI applications to operate efficiently and securely. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.2.2 Ability to self-heal | Given an identified (e.g. by Root Cause Identification subsystem) issue, remediation is typically performed through some combination of manual actions performed by a human, along with the application of infrastructure configuration changes, and occasionally, software code patches. The actions that need to be performed manually by a human in order to rectify the issue may be expressed in the form of a set of remediation steps, a natural language construct. The set of all necessary steps, code patches, and configuration changes required to respond to a given issue may be termed a Course of Action (CoA) for the issue. All elements of a CoA can potentially be generated by an LLM that employs RAG, whose input is the identified issue. If the CoA elements consist entirely of configuration and code changes, it is possible for these changes to be incorporated automatically by the concerned systems, possibly through an intermediate orchestration system. This allows for limited self-healing of the system. Typically, it would be considered belonging to the Day 2 phase. ETSI ETSI TR 104 051 V1.1.1 (2025-06) 15 |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.2.3 Root cause identification | The analytics of the MDA can already contain specific root cause identification of the detected issues. There are already several requirements on the MDA capabilities to provide a root cause identification e.g. network slice throughput issue(s), E2E latency issue, network slice load issues and recommended actions, energy efficiency issue, and mobility performance issues. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.2.4 Predictive maintenance | Information from the MDA pertaining to the operational characteristics of the underlying network infrastructure may be used as inputs to NN models trained to predict expected hardware failure times and flag components for pre-emptive replacement or repair. For example, a model may be used to ingest Self-Monitoring, Analysis, and Reporting Technology (SMART) metrics from Hard Disk Drive arrays, monitor their deterioration, and provide proactive drive replacement schedules. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.2.5 Dynamic optimization | |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.2.5.1 Mobility Optimization | Mobility Optimization is one of the use cases studied in [i.3] and specified in [i.4] and happens in the operational phase, i.e. Day 2. The use case aims to minimize performance loss due to unsuccessful or erroneous mobility management events. Mobility management is expected to guarantee the service-continuity during the mobility by minimizing call drops, Radio Link Failures (RLFs), unnecessary handovers, and ping-pong. In the future, it is expected that handovers will be increasing in numbers as the coverage of a single node decreases and UE mobility gets higher and more active. In addition, for the applications characterized by the stringent QoS requirements such as reliability, latency, etc., the Quality of Experience (QoE) is sensitive to the handover performance, so that mobility management should avoid unsuccessful handovers and reduce the latency during the handover procedures. The RAN AI/ML framework studied in 3GPP TR 37.817 [i.3] and specified in RAN specifications ETSI TS 138 423 [i.4], includes several network entities exchanging AI/ML-related information for the purposes of data collection, data inference, output, and feedback. These network entities are UEs, RAN nodes, and potentially OAM nodes depending on the architecture. The RAN AI/ML framework specifies three use cases including Mobility Optimization for which the UEs and RAN nodes provide input and inference data and the RAN AI/ML framework on RAN and potentially OAM nodes provide output and feedback data to relevant nodes. Note that other use cases include Network Energy Saving and Load Balancing. An OAM and /or NG-RAN node may train a model or perform inference using UE-related information acquired by the RAN node (e.g. UE location information and UE trajectory prediction), and the information obtained from neighbouring RAN nodes (e.g. UE mobility history information). The RAN AI/ML framework includes information transfer procedures from UEs and RAN nodes. UE-related data are annotated with temporary UE identifiers or UE measurement identifiers. The information generated by UE and RAN nodes stays within the 3GPP network domain and is not exposed to third parties. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.2.5.2 Load Balancing | The Load Balancing use case aims to distribute the load evenly among cells and areas of cells, to transfer part of the traffic from congested cells or congested areas of cells, or to offload users from one cell, cell area, carrier, or RAT to improve network performance. This can be done by means of optimization of handover parameters and handover actions. This activity is considered a typical Day 2 operation. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.2.6 Automated network design | The following clause provides an example of an AI/ML pipeline dedicated to automated network design. It utilized the methodology and terminology that is defined in clause 3.1.1. ETSI ETSI TR 104 051 V1.1.1 (2025-06) 16 Network Requirements Analysis and Design tasks typically occur in the Day 0 phase. Requirements Gathering, Strategic Goal Setting, and Outcome Analysis consume a significant proportion of the efforts involved in network design, the output of which is typically captured as documents intended for a wide variety of stakeholders across executive, legal, business, and technical fields. LLMs are expected to be adopted extensively in this stage to perform tasks such as refinement of abstract strategic goals into clear business objectives, and subsequent transformation of these objectives into specialized technical requirements for the set of intended audiences. System architecture specifications constructed using these technical requirements may then be fed to specialized generator or optimizer models to create technical designs for the system. Generative AI techniques can also be used to transform technical requirements into system architecture specifications and initial network layout plans, which can then be optimized using predictive AI/ML techniques. Low-level deployment plans typically involve the use of boilerplate templates requiring a degree of manual entry. Such tasks are likely to be offloaded to multimodal LLMs, i.e. models that can produce multiple forms of output, e.g. text and images, within a single query context. Deployment configuration, considered a Day 1 task, can also potentially be aided through Generative AI. ICT infrastructure deployments increasingly make use of Infrastructure as Code (IaC) and virtualization techniques to accelerate the path to operational readiness. These techniques typically make use of configuration files to describe the intended architecture, topology, and resource allocation of application infrastructure as well as their behaviour. Such configuration files, hitherto written manually, are suitable targets for automation through LLM generation. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.3 Managed telecom services | |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.3.1 Use cases | The following subclauses add detailed descriptions for managed telecom services use cases. These use cases mostly relate to Day 2 in the network lifecycle described in clause 3.1.1. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.3.2 Ticket (e.g. trouble ticket, CR) classification and routing | In many cases, the tickets contain data entered by humans in natural language such as a description of the problem and the steps already taken. While frequently some initial classification of the problem is provided by the person opening the ticket, it can also be marked as 'other' or be imprecise. In any case, the LLM can analyse the problem description text to decide where to route the ticket. For example, which support team should handle it (customer care, technical, etc.) or whether the escalation to the higher tier is already in order. Moreover, after closing the issue, the LLM can prepare a digest of the remedy and publish it in the form of a 'best-known procedure' for more widespread use (e.g. FAQ webpage). Furthermore, the knowledge of how to solve a given problem (or information that is not solvable) can be fed back to the AI model, leading to faster and more precise answers in the future. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.3.3 Customer churn prediction | At a higher level, LLM-based sentiment analysis is also possible with the tickets (see clause 6.3.1) and other sources (e.g. the history of chats clients had with a company chatbot or social media posts). This can be combined with data obtained from network infrastructure measurements such as end-to-end delay or downtimes (see also clause 6.3.4), facilitating trend and changepoint analysis. This combined predictive and generative AI-based study may ultimately give an operator an early warning about possible customer churn. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 6.3.4 Service (e.g. SLA) assurance | Service level assurance utilizes several elements mentioned previously, with monitoring and data collection being a starting point. AI/ML-based analysis of the data can predict, for example, a possible SLA violation (e.g. probability of exceeding an agreed end-to-end delay value within a given time frame is 90 %), analysing the cause of it (delay is not because of the network but due to CPU processing instead of GPU), and taking preventive action before the actual violation occurs (offload the processing from edge node that has no free GPU to a central cloud with free resources). In all of these steps, AI/ML models can be used for prediction, inference, Course-of-Action generation, and execution. ETSI ETSI TR 104 051 V1.1.1 (2025-06) 17 7 Vulnerabilities and mitigation strategies of AI/ML models deployed in telecom use cases |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.1 Attacks on System Monitoring | |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.1.1 Introduction | It is envisioned that the compiled executable AI models will be used as AI Agents attached to network functions or form separate network functions themselves. AI agents that improperly or maliciously function may be potent attack vectors in telecom networks. Their instantiation and use need to be strictly monitored Each use case identified in clause 6 of the present document is represented below with NIST mappings for the principal classification of adversarial AI technique possible and the attackers' goal, along with a brief commentary on the technique, references to implementations of such attacks, and possible steps for remediation or prevention. The goal of the attacker targeting system monitoring can be either to directly disrupt the service or be more indirect like hiding further attacks or even exfiltrating data gathered by the monitoring system to only plan for the subsequent adversarial activities. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.1.2 Anomaly detection | Most NN anomaly detection algorithms operate on time series data. Studies have demonstrated an attack on such models by inducing imperceptible perturbations into an input signal fed to the model [i.10]. This form of attack causes models to misclassify anomalous signals as normal operating behaviour with high confidence, propagating these errors to upstream decision-making systems. Dataset poisoning is also a viable attack strategy as demonstrated by Nguyen et al. Here, by introducing specific patterns of malicious activity into the training dataset as benign, they created models that were effective in their intended operation, but consistently ignored the intended malicious pattern. Log analysis models are also susceptible to perturbation attacks [i.11]. LLMs are further susceptible to indirect prompt injection, where logs (or traces) to be processed may contain malicious code or instructions that can then be executed by downstream systems. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.1.3 Root cause identification | Targeting the MDA system can offer an attacker the opportunity to induce misdiagnosis of a problem root cause, thereby either triggering an erroneous response or distracting response teams to mask a secondary attack. The specific techniques that may be employed by an attacker depend on the RC identification mechanisms used. Models relying on stochastic techniques such as AI/ML models may be poisoned during the training phase, either by means of backdoor injection, or dataset tampering. Prevention of such attacks lies primarily in the training phase, where models will need to be trained with adversarial samples to improve robustness. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.1.4 Predictive maintenance | Similar to the case for stochastic MDA systems, NN models trained for predicting system failures may be compromised by poisoning or signal tampering. Time series models, which are typically employed for such monitoring are also particularly susceptible to perturbation attacks, whereby an attacker introduces specific, smooth perturbations in input signals [i.10] that cause misclassification errors. Such attacks may be used both for generating false positive failure events, that may be used to distract from secondary attacks, and false negatives, that will induce system failure in the future. Training robustness techniques and ensembling may be used to alleviate this risk. ETSI ETSI TR 104 051 V1.1.1 (2025-06) 18 |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.2 Attacks on Intelligent Networks | |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.2.1 Introduction | During the runtime phase, an attack on the AI-powered network control plane can allow an adversary to disrupt the service in several ways. These include limiting the system's ability to correctly execute the reconciliation loop or tricking it into making incorrect decisions regarding resource management. Also, when AI is used in the design phase, the attacker can exfiltrate data related to the planning or try to plant the backdoors that - after the deployment - will allow for executing further attacks. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.2.2 Ability to self-heal | Poisoning of the training data in the case of CoA identification can lead to a set of actions that ostensibly address a given issue in the system but leave unaddressed an exploitable weakness. In the case of CoA execution, the employed LLMs may be induced to generate code with vulnerabilities. This is possible even in the case of unaltered publicly available models and can be exacerbated through training data poisoning to increase the likelihood of vulnerable code being produced for benign prompts. Sandboxing and malicious prompt detection may be used to minimize the effects of exploits inadvertently generated by the LLMs. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.2.3 Dynamic optimization | The time series nature of network performance optimization presents the risk of perturbation attacks similar to those possible on predictive maintenance systems. When applied to the mobility optimization use case, it would then be possible for an attacker to artificially degrade user experience in specific cells by triggering handovers to sub-optimal cells. Load balancing may be disrupted by poisoning attacks that use a specific trigger to force the model to overload specific cells. The mitigation of these attacks follows the strategy outlined for time series models. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.2.4 Automated Network Design | A potential for data extraction exists if an attacker is able to capture prompts given to the model, especially in the case that the model is provided as a public cloud SaaS. By repeating the prompts an attacker can recreate the output, or potentially expose confidential information fed to the model. This attack may be trivially undertaken, with the conclusion that prompts are not secrets. Such attacks can be mitigated to an extent by sanitizing prompts. An attacker may attempt to manipulate the data being ingested by the model in order to induce incorrect results that may be propagated to higher-level systems or models, resulting in suboptimal designs, wasted resources, and potentially lost revenue. An attacker may induce the model to generate insecure or weak code that may be exploited in the future, and have it enter the CI/CD pipeline of the target. This may be achieved by poisoning the datasets that the model was trained on, or by prompt engineering to create functional code with obfuscated vulnerabilities. It is possible to build such poisoned models that ostensibly pass safety tests. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.3 Attacks on Managed Telecom Services | |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.3.0 Introduction | A successful attack on customer-related services can damage the operator's reputation and potentially escalate to both monetary and legal consequences. This can occur if there is a violation of contractual agreements or if a leak of personally identifiable information is discovered. ETSI ETSI TR 104 051 V1.1.1 (2025-06) 19 |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.3.1 Ticket (e.g. trouble ticket, CR) classification and routing | LLM models employed for ticket classification, especially those that are exposed to customer-originated tickets, are at high risk for prompt injection with the aims either of inducing privacy compromise or triggering hallucinatory behaviour. The latter represents a legal risk for the operator as, depending on the jurisdiction, LLM-generated responses presented to a customer are treated as on par with official documentation. Prompt sanitization and the use of RAG techniques for material such as FAQs presented to the customer can largely mitigate these risks. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.3.2 Customer churn prediction | Time series attack techniques such as smooth perturbation may be employed by an attacker to induce churn prediction models to present an overly optimistic estimate, thereby causing revenue loss. The mitigation of these attacks follows the strategy outlined for time series models. |
a65dbfd110e2620b04001bee98bdcd6e | 104 051 | 7.3.3 Service (e.g. SLA) assurance | Due to its compound nature consisting of sets of systems such as RCA, CoA generation, and CoA execution, the possible attacks, and the corresponding mitigation strategies, on SLA assurance systems may be considered as the union of the attacks and mitigations of their constituent systems. ETSI ETSI TR 104 051 V1.1.1 (2025-06) 20 Annex A: Bibliography • ETSI TR 104 221: "Securing Artificial Intelligence (SAI); Problem Statement". • ETSI TR 104 222: "Securing Artificial Intelligence; Mitigation Strategy Report". • 3GPP TR 33.877: "Study on the security aspects of Artificial Intelligence (AI)/Machine Learning (ML) for the Next Generation Radio Access Network (NG-RAN)". • 3GPP TR 33.898: "Study on security and privacy of Artificial Intelligence/Machine Learning (AI/ML)-based services and applications in 5G". • 3GPP TR 33.738: "Study on security aspects of enablers for Network Automation for 5G - phase 3". • ETSI TR 103 305-1: "Cyber Security (CYBER); Critical Security Controls for Effective Cyber Defence; Part 1: The Critical Security Controls". ETSI ETSI TR 104 051 V1.1.1 (2025-06) 21 Annex B: Change history Date Version Information about changes 9-2023 0001 SAI(23)18a016r1 - CR for clauses 2.2, 3.3, and 6.1 of SAI-0014 approved at SAI-18c 9-2023 0002 SAI(23)019008r1 - SAI-0014 - New sub-clause on GenAI vs traditional AI approved with editorial modifications at SAI#19 11-2023 0003 SAI(23)019b001 - Additional content for SAI-0014 approved at SAI#19b 12-2023 0004 Converted to the TR format as DTR/SAI-0011 12-2023 0004 SAI(23)001022 Additional content for clause 5.2 - ML functionality in telecom providers' networks. Approved at SAI#1 (SAI-TC) 01-2024 001 Converted from SAI-0014 (GR) into SAI-0011 (TR) ETSI ETSI TR 104 051 V1.1.1 (2025-06) 22 History Document history V1.1.1 June 2025 Publication |
7bacf70a4dd0b8f943de5e517835b45e | 102 105 | 1 Scope | The present document provides an assessment of the feasibility of using object-orientation in the development of standards, particularly when used in association with Message Sequence Charts (MSC), Specification and Description Language (SDL) defined in ITU-T Recommendations Z.120 [11], Z.100 [9] and Z.105 [10] and the Tree and Tabular Combined Notation (TTCN) defined in ISO/IEC 9646-3 [7] for specifying the behaviour and testing of services and protocols. A number of textual and graphical notations have been defined for object-oriented design purposes. The Guidelines for the Definition of Managed Objects (GDMO), for example, have been used extensively in the specification of international standards for telecommunication network management services. However, it is the universal nature of graphical languages which makes them particularly interesting for standardization applications. Since its introduction in 1994, the Unified Modelling Language (UML) has become one of the most popular and best defined graphical languages for object-oriented design and, for these reasons, this is the only notation considered here. The purpose of this TR is: - to provide a very brief introduction to the UML and the work of the Object Management Group (OMG) in standardizing it (The Universal Modelling Language [12]); - to identify elements of the UML which could have some value if applied to the ETSI standards-making process; - to evaluate the benefits that may be derived from their use. The UML is considered in relation to all types of standard. On completion of this study, a set of guidelines based on a methodological approach to the use of UML in the standards-making process will be developed. These guidelines will assist technical bodies and rapporteurs to make effective use of UML wherever feasible within the process. |
7bacf70a4dd0b8f943de5e517835b45e | 102 105 | 2 References | The following documents contain provisions which, through reference in this text, constitute provisions of the present document. • References are either specific (identified by date of publication, edition number, version number, etc.) or non-specific. • For a specific reference, subsequent revisions do not apply. • For a non-specific reference, the latest version applies. • A non-specific reference to an ETS shall also be taken to refer to later versions published as an EN with the same number. [1] ETS 300 247 (1993): "Business Telecommunications (BT); Open Network Provision (ONP) technical requirements; 2 048 kbit/s digital unstructured leased line (D2048U) Connection characteristics". [2] ETS 300 289 (1994): "Business TeleCommunications (BTC); 64 kbit/s digital unrestricted leased line with octet integrity (D64U); Connection Characteristics". [3] ETS 300 419 (1995): "Business TeleCommunications (BTC); 2 048 kbit/s digital structured leased lines (D2048S); Connection Characteristics". [4] ETS 300 687 (1996): "Business TeleCommunications (BTC); 34 Mbit/s digital leased lines (D34U and D34S); Connection Characteristics". [5] ETS 300 688 (1996): "Business TeleCommunications (BTC); 140 Mbit/s digital leased lines (D140U and D140S); Connection Characteristics". [6] ETS 300 694 (1996): "Private Integrated Services Network (PISN); Cordless Terminal Mobility (CTM); Call handling additional network features; Service description". ETSI ETSI TR 102 105 V1.1.1 (1999-08) 7 [7] ISO/IEC 9646-3 (1999): "Information technology - Open Systems Interconnection - Conformance testing methodology and framework - Part 3: The Tree and Tabular Combined Notation (TTCN)". [8] CCITT Recommendation I.130 (1988): "Method for the characterization of telecommunication services supported by an ISDN and network capabilities of an ISDN". [9] ITU-T Recommendation Z.100 (1993): "Specification and description language (SDL)". [10] ITU-T Recommendation Z.105 (1994): "SDL combined with ASN.1 (SDL/ASN.1)". [11] ITU-T Recommendation Z.120 (1993): "Message Sequence Chart (MSC)". [12] OMG Specifications ad/97-08-02 to 97-08-07 (1997): "Unified Modelling Language", Version 1.1. [13] OMG Specification cf/96-12-03: "Meta Object Facility". [14] OMG Specification ad/98-10-05: "XML Metadata Interchange". [15] OMG Specification formal/98-12-01: "Common Object Request Broker Architecture", Version 2.3. [16] OMG RFP ad/97-12-03 (1997): "Stream-based Model Interchange Format RFP". [17] OMG RFP ad/98-11-01 (1998): "Action Semantics for the UML RFP". [18] OMG RFP ad/99-03-13 (1999): "UML Profile for Scheduling, Performance, and Time RFP". [19] Rumbaugh, Jacobsen & Booch: "The Unified Modelling Language Reference Manual", Addison- Wesley (1999), ISBN 0-201-30998-X. [20] Booch, Rumbaugh & Jacobsen: "The Unified Modelling Language User Guide", Addison-Wesley (1999), ISBN 0-201-57168-4. [21] ETS 300 012-4: "Integrated Services Digital Network (ISDN); Basic User-Network Interface (UNI); Part 4: Conformance test specification for interface IA". |
7bacf70a4dd0b8f943de5e517835b45e | 102 105 | 3 Definitions and abbreviations | |
7bacf70a4dd0b8f943de5e517835b45e | 102 105 | 3.1 Definitions | For the purposes of the present document, the following definition applies: metamodel: a model that defines the language for expressing a model (from The UML Reference Manual [19]) |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.