Content for TR 23.700-84 Word version: 19.0.0
5 Use Cases, Motivations and Key Issues
6 Solutions
7 Overall Evaluation
8 Conclusions
$ Change history
5 Use Cases, Motivations and Key Issues p. 12
5.1 Use Cases p. 12
5.1.0 Guidelines p. 12
5.1.1 Use Case #1: NWDAF-assisted QoS enhancement p. 12
5.1.2 Use Case #2: Enhancements to QoS Determination with NWDAF Assistance p. 12
5.1.3 Use Case #3: NWDAF assistance in device signalling storm prevention and mitigation p. 13
5.1.4 Use Case #4: Motivation and Support for VFL in 5GC p. 13
5.1.5 Use Case #5: NWDAF support for observed service experience analytics based on VFL p. 14
5.1.6 Use Case #6: Analytics-assisted prevention of abnormal NF behaviour causing signalling storm and mitigation of its impact in the network p. 14
5.2 Key Issues p. 15
5.2.0 Mapping of Key Issues to Use Cases p. 15
5.2.1 Key Issue #1: Enhancements to LCS to support Direct AI/ML based Positioning p. 15
5.2.2 Key Issue #2: 5GC Support for Vertical Federated Learning p. 15
5.2.3 Key Issue #3: NWDAF-assisted policy control and QoS enhancement p. 16
5.2.4 Key Issue #4: NWDAF enhancements to support network abnormal behaviours (i.e. signalling storm) mitigation and prevention p. 16
6 Solutions p. 17
6.0 Mapping of Solutions to Key Issues p. 17
6.1 Solution #1: Direct AI/ML based Positioning for case 2b/3b p. 17
6.1.1 Description p. 17
6.1.2 Procedures p. 19
6.1.2.1 Direct AI/ML based Positioning in LMF p. 19
6.1.2.2 Training procedure for Direct AI/ML positioning model p. 20
6.1.2.3 Procedure for ML model training for Direct AI/ML based positioning and data collection p. 21
6.1.3 Impacts on services, entities and interfaces p. 23
6.2 Solution #2: Support for AI/ML Direct Positioning Training, Inference and Data Collection with LMF-side models p. 23
6.2.1 Description p. 23
6.2.2 Procedures p. 24
6.2.2.1 Training procedure for AI/ML Direct Positioning with LMF-side models p. 24
6.2.2.2 Inference procedure for AI/ML Direct Positioning with LMF-side models p. 25
6.2.3 Impacts on services, entities and interfaces p. 26
6.3 Solution #3: Training of the AI/ML positioning model p. 26
6.3.1 Functional Description p. 26
6.3.2 Procedures p. 26
6.3.3 Impacts on existing services, entities and interfaces p. 27
6.4 Solution 4: Data Collection Framework for Direct AI/ML positioning p. 27
6.4.1 Description p. 27
6.4.2 Procedures p. 29
6.4.3 Impacts on services, entities and interfaces p. 30
6.5 Solution #5: LMF selection to support the LMF-sided direct AI/ML positioning p. 30
6.5.1 Description p. 30
6.5.2 Procedures p. 31
6.5.3 Impacts on services, entities and interfaces p. 31
6.6 Solution #6: LMF based ML model training and Inference p. 31
6.6.1 Description p. 31
6.6.2 Procedures p. 31
6.6.3 Impacts on services, entities and interfaces p. 32
6.7 Solution #7 Training of LMF-side Model to determine location p. 32
6.7.1 Description p. 32
6.7.2 Procedures p. 34
6.7.2.1 Training an LMF-side model for NG-RAN assisted positioning p. 34
6.7.2.2 Training an LMF-side model for UE assisted positioning p. 36
6.7.2.3 Selecting UEs for reporting locations based on PEI p. 37
6.7.2.4 Inference of location for NG-RAN assisted positioning p. 38
6.7.2.5 Inference of location for UE assisted positioning p. 39
6.7.2.6 Monitoring accuracy of model for NG-RAN assisted location measurements p. 40
6.7.3 Impacts on services, entities and interfaces p. 40
6.8 Solution #8: MTLF-based model performance monitoring for AI/ML positioning p. 42
6.8.1 Description p. 42
6.8.2 Procedures p. 42
6.8.3 Impacts on services, entities and interfaces p. 43
6.9 Solution #9: new solution for KI#1 support monitoring the performance of AI model p. 43
6.9.1 Description p. 43
6.9.2 Procedures p. 44
6.9.2.1 Performance of AI/ML model monitoring based on PRU p. 44
6.9.2.2 Procedure of PRU information collection for AI/ML model training p. 45
6.9.3 Impacts on services, entities and interfaces p. 46
6.10 Solution #10: Direct AI/ML based positioning with NWDAF assistance p. 46
6.10.1 Description p. 46
6.10.1.1 General p. 46
6.10.1.2 Input data p. 47
6.10.2 Procedures p. 49
6.10.3 Impacts on services, entities and interfaces p. 50
6.11 Solution #11: Data collection procedure for LMF-side model training p. 51
6.11.1 Functional Description p. 51
6.11.2 Procedures p. 51
6.11.3 Impacts on existing services, entities and interfaces p. 52
6.12 Solution #12: new solution for KI#1 support the data collection for AI model training based on authorization p. 52
6.12.1 Description p. 52
6.12.2 Procedures p. 53
6.12.3 Impacts on services, entities and interfaces p. 53
6.13 Solution #13: Sample/Feature alignment and general training procedure for the VFL p. 54
6.13.1 Description p. 54
6.13.2 Procedures p. 55
6.13.2.1 Sample and feature alignment procedure p. 55
6.13.2.2 General training procedure for the VFL between the VFL Active Participant(VFL server) (i.e. AF) and VFL Passive Participant(VFL client) (i.e. NWDAF) p. 58
6.13.2.3 Inferencing procedure for the VFL between the VFL Active Participant (i.e. AF) and VFL Passive Participant (i.e. NWDAF(s)) p. 60
6.13.3 Impacts on services, entities and interfaces p. 61
6.14 Solution #14: General procedure for NWDAF initiated Vertical Federated Learning between NWDAF(s) and AF(s) p. 61
6.14.1 Description p. 61
6.14.2 Procedures p. 62
6.14.3 Impacts on services, entities and interfaces p. 65
6.14.3.1 Training Impact p. 65
6.14.3.2 Inference Impact p. 65
6.15 Solution 15: Support for vertical federated learning: Model Training and Inference p. 66
6.15.1 Description p. 66
6.15.2 Procedures p. 66
6.15.2.1 Sample/Feature alignment p. 66
6.15.2.2 VFL training procedure p. 67
6.15.2.3 Distributed Inference p. 69
6.15.3 Impacts on services, entities and interfaces p. 71
6.16 Solution #16: Support for VFL with NWDAF and AF as Participants p. 71
6.16.1 Description p. 71
6.16.2 Procedures p. 71
6.16.3 Impacts on services, entities and interfaces p. 75
6.17 Solution #17: NF discovery and selection for VFL p. 75
6.17.1 Description p. 75
6.17.2 Procedures p. 76
6.17.2.1 External AF VFL capability report p. 76
6.17.2.2 NWDAF VFL capability report p. 76
6.17.2.3 Client selection process of VFL p. 77
6.17.3 Impacts on existing services, entities and interfaces p. 78
6.18 Solution #18: Vertical Federated Learning between NWDAF and AF p. 78
6.18.1 Description p. 78
6.18.2 Procedures p. 80
6.18.2.1 Discovery and selection of VFL clients p. 80
6.18.2.1.1 Discovery and selection of AF(s) if NWDAF as the VFL server p. 80
6.18.2.1.2 Discovery and selection of NWDAF(s) if AF as the VFL server p. 81
6.18.2.2 VFL training procedure between NWDAF and AF p. 81
6.18.2.2.1 VFL training procedure if NWDAF acts as the VFL server p. 82
6.18.2.2.2 VFL training procedure if AF acts as the VFL server p. 84
6.18.2.3 VFL training procedure among NWDAFs p. 85
6.18.2.4 VFL inference procedure between NWDAF and AF p. 88
6.18.2.4.1 VFL inference procedure if NWDAF acts as the VFL server p. 88
6.18.2.3.2 VFL inference procedure if AF acts as the coordinator p. 89
6.18.2.5 VFL inference procedure among NWDAFs p. 90
6.18.3 Impacts on services, entities and interfaces p. 91
6.19 Solution #19: VFL inference procedure between NWDAF and AF p. 91
6.19.1 Description p. 91
6.19.2 Procedures p. 92
6.19.3 Impacts on Existing Nodes and Functionality p. 92
6.20 Solution #20: Inference procedure for the Vertical Federated Learning between NWDAF(s) and AF(s) p. 93
6.20.1 Description p. 93
6.20.2 Procedures p. 94
6.20.3 Impacts to Services, Entities and Interfaces p. 95
6.21 Solution #21: Vertical Federated Learning for support of Application Layer QoE p. 96
6.21.1 Key Issue mapping p. 96
6.21.2 Description p. 96
6.21.2.1 VFL for Application QoE provisioning p. 96
6.21.2.2 VFL for Network parameters provisioning p. 97
6.21.3 Procedures p. 99
6.21.3.1 Application provisioning for customized QoE p. 99
6.21.3.2 Network parameters provisioning p. 101
6.21.4 Impacts on services, entities and interfaces p. 102
6.22 Solution #22: Vertical Federated learning considering internal NWDAF architecture p. 103
6.22.1 Description p. 103
6.22.2 Procedures p. 105
6.22.2.1 VFL model training and inference involving MTLF, AnLF, ADRF and AF p. 105
6.22.2.2 VFL sample alignment for UE samples p. 108
6.22.3 Impacts on services, entities and interfaces p. 109
6.23 Solution #23: Cross domain VFL involving NWDAF and AF p. 110
6.23.1 Description p. 110
6.23.1.1 Terminology p. 110
6.23.2 Procedures p. 112
6.23.2.1 NWDAF acts as FL Server with VFL capabilities p. 112
6.23.2.2 AF acts as FL Server with VFL capabilities p. 114
6.23.2.3 VFL capabilities within NF profile p. 114
6.23.2.4 Training procedure p. 115
6.23.2.4.1 NWDAF as FL Server with VFL capabilities p. 115
6.23.2.4.2 AF as FL Server with VFL capabilities p. 117
6.23.2.5 Inference procedure p. 119
6.23.3 Impacts on services, entities and interfaces p. 121
6.24 Solution #24: How to support Vertical Federated Learning between NWDAF and AF p. 121
6.24.1 Terminology p. 121
6.24.2 Description p. 122
6.24.3 Procedures p. 122
6.24.4 Impacts on services, entities and interfaces p. 124
6.25 Solution #25: NF Registration and Discovery Enhancement for VFL p. 125
6.25.1 Description p. 125
6.25.2 Procedures p. 126
6.25.3 Impacts on Existing Nodes and Functionality p. 128
6.26 Solution #26: NWDAF-assisted policy control with Recommendation logical function p. 128
6.26.1 Description p. 128
6.26.1.1 Recommendation logical function p. 128
6.26.1.2 Policy control enhancement with the assistance of NWDAF (ReLF) p. 130
6.26.2 Procedures p. 131
6.26.3 Impacts to Services, Entities and Interfaces p. 132
6.27 Solution #27: NWDAF assisted QoS policy generation p. 132
6.27.1 Description p. 132
6.27.2 Procedures p. 133
6.27.3 Impacts on services, entities and interfaces p. 134
6.28 Solution #28: QoS Flow Analytics p. 134
6.28.1 Description p. 134
6.28.2 Procedures p. 135
6.28.3 Impacts on existing services, entities and interfaces p. 139
6.29 Solution #29: How to evaluate NWDAF-assisted policy control and QoS enhancement p. 139
6.29.1 Description p. 139
6.29.2 Procedures p. 140
6.29.3 Impacts on services, entities and interfaces p. 141
6.30 Solution #30: NWDAF-assisted PDU Set assistance information p. 142
6.30.1 Description p. 142
6.30.2 Procedures p. 143
6.30.3 Impacts on services, entities and interfaces p. 143
6.31 Solution #31: PCF as RL Agent and NWDAF as Interpreter p. 143
6.31.1 Key Issue mapping p. 143
6.31.2 Description p. 143
6.31.3 Procedures p. 146
6.31.3.1 PCF as RL Agent p. 146
6.31.4 Impacts on services, entities and interfaces p. 147
6.32 Solution #32: NWDAF-assisted optimized QoS policies determination p. 147
6.32.1 Description p. 147
6.32.2 Procedures p. 148
6.32.3 Impacts on services, entities, and interfaces p. 150
6.33 Solution #33: QoS/policy enhancements assisted by NWDAF p. 150
6.33.1 Description p. 150
6.33.2 Procedures p. 153
6.33.2.1 Procedures for NWDAF assisted QoS and policy determination p. 153
6.33.2.2 Procedures to deploy NWDAF assisted QoS and policy determination during PDU session establishment p. 154
6.33.3 Impacts on services, entities and interfaces p. 155
6.34 Solution #34: BDT Policy Recommendations p. 155
6.34.1 Description p. 155
6.34.2 Procedures p. 156
6.34.3 Impacts on services, entities and interfaces p. 158
6.35 Solution #35: NWDAF-assisted Network Abnormal Behaviour Mitigation and Prevention p. 158
6.35.1 Description p. 158
6.35.2 Procedures p. 159
6.35.2.1 General p. 159
6.35.2.2 Input Data p. 160
6.35.2.3 Output Analytics p. 163
6.35.2.4 Mitigation or Prevention p. 163
6.35.3 Impacts on services, entities and interfaces p. 164
6.36 Solution #36: Registration Signalling Analytics to support detection and prevention of signalling storm p. 164
6.36.1 Description p. 164
6.36.2 Procedures p. 167
6.36.3 Impacts on services, entities and interfaces p. 168
6.37 Solution #37: Signalling storm detection and mitigation based on O&M data p. 168
6.37.1 Description p. 168
6.37.2 Procedures p. 169
6.37.3 Impacts on services, entities and interfaces p. 169
6.38 Solution #38: NWDAF-assisted signalling storm analytics, predictions, prevention and mitigation p. 170
6.38.1 General description p. 170
6.38.1.1 NWDAF-assisted signalling storm detection p. 170
6.38.1.1.1 Input Data p. 171
6.38.1.1.2 Output Data p. 172
6.38.1.1.3 NWDAF-assisted signalling storm analytics Procedure p. 173
6.38.1.2 NWDAF-assisted signalling storm prevention and mitigation p. 174
6.38.2 Impacts on existing services, entities and interfaces p. 177
6.39 Solution #39: NWDAF Analytics based Signalling Strom Prediction p. 177
6.39.1 Description p. 177
6.39.2 Procedures p. 180
6.39.2.1 UE registration signalling storm prevention/mitigation using Control Plane Signalling Abnormality Analytics p. 182
6.39.2.2 Inter-NF Signalling storm prevention/mitigation by SCP using NWDAF Analytics p. 184
6.39.3 Impacts on existing services, entities and interfaces p. 185
7 Overall Evaluation p. 185
8 Conclusions p. 185
8.1 Conclusions for KI#1: Enhancements to LCS to support Direct AI/ML based Positioning p. 185
8.2 Conclusions for KI#2: 5GC Support for Vertical Federated Learning p. 186
8.3 Conclusions for KI#3: NWDAF-assisted policy control and QoS enhancement p. 188
8.4 Conclusions for KI#4: NWDAF-assisted Network Abnormal Behaviour Mitigation and Prevention p. 189
$ Change history p. 191
