Content for TR 23.700-81 Word version: 18.0.0
5 Key Issues p. 15
5.1 Key Issue #1: How to improve correctness of NWDAF analytics p. 15
5.1.1 Description p. 15
5.2 Key Issue #2: NWDAF-assisted application detection p. 16
5.2.1 Description p. 16
5.3 Key Issue #3: Data and analytics exchange in roaming case p. 16
5.3.1 Description p. 16
5.4 Key Issue #4: How to Enhance Data collection and Storage p. 17
5.4.1 Description p. 17
5.5 Key Issue #5: Enhance trained ML Model sharing p. 17
5.5.1 Description p. 17
5.6 Key issue #6: NWDAF-assisted URSP p. 18
5.6.1 Description p. 18
5.7 Key Issue #7: Enhancements on QoS Sustainability analytics p. 18
5.7.1 Description p. 18
5.8 Key Issue #8: Supporting Federated Learning in 5GC p. 18
5.8.1 Description p. 18
5.9 Key Issue #9: Enhancement of NWDAF with finer granularity of location information p. 19
5.9.1 Description p. 19
5.10 Key Issue #10: Interactions with MDAS/MDAF p. 20
5.10.1 Description p. 20
6 Solutions p. 21
6.0 Mapping Solutions to Key Issues p. 21
6.1 Solution #1: Improving Correctness using multiple ML models for an analytics report p. 23
6.1.1 Description p. 23
6.1.2 Procedures p. 24
6.1.2.1 General p. 24
6.1.2.2 Support for describing multiple ML models in provisioning service operations p. 24
6.1.3 Impacts on existing nodes and functionality p. 25
6.2 Solution #2: Improving the Correctness of Service Experience Predictions with Contribution Weights p. 26
6.2.1 Description p. 26
6.2.2 Procedures p. 26
6.2.3 Impacts on existing nodes and functionality p. 26
6.3 Solution #3: Accuracy based NWDAF Analytics Correctness Improvement p. 27
6.3.1 Description p. 27
6.3.2 Procedures p. 28
6.3.3 Impacts on services, entities and interfaces p. 32
6.4 Solution #4: Determining ML model drift for improving analytics accuracy p. 32
6.4.1 Description p. 32
6.4.2 Procedures p. 33
6.4.3 Impacts on services, entities and interfaces p. 35
6.5 Solution #5: Enhancements on ML model provision to improve correctness of NWDAF analytics p. 35
6.5.1 Description p. 35
6.5.2 Procedures p. 36
6.5.2.1 Procedures for AnLF based error monitoring p. 37
6.5.2.2 Procedures for MTLF based error monitoring p. 38
6.5.2.3 Procedures for MTLF to detect dynamics on the analytics p. 39
6.5.3 Impacts on existing nodes and functionality p. 40
6.6 Solution #6: Correctness improvement of NWDAF by determining ML model performance p. 40
6.6.1 Description p. 40
6.6.2 Procedures p. 41
6.6.3 Impacts on existing nodes and functionality p. 42
6.7 Solution #7: Enhancements to NWDAF analytics services p. 42
6.7.1 Description p. 42
6.7.2 Procedures p. 43
6.7.2.1 Analytics consumer rating an NWDAF analytics provided by an NWDAF containing AnLF p. 43
6.7.2.2 NWDAF containing AnLF rating an ML model provided by an NWDAF containing MTLF p. 44
6.7.3 Impacts on services, entities and interfaces p. 45
6.8 Solution #8: NWDAF assisted service type detection p. 45
6.8.1 Description p. 45
6.8.2 Procedures p. 48
6.8.3 Impacts on existing nodes and functionality p. 48
6.9 Solution #9: NWDAF-assisted application detection p. 49
6.9.1 Description p. 49
6.9.2 Input Data p. 50
6.9.3 Output Analytics p. 51
6.9.4 Procedures p. 51
6.9.5 Impacts on existing nodes and functionality p. 52
6.10 Solution #10: Support for Data and Analytics Exchange in Roaming Case p. 53
6.10.1 Description p. 53
6.10.2 Procedures p. 54
6.10.3 Impacts on services, entities and interfaces p. 56
6.11 Solution #11: PDU session management in roaming scenarios using network analytics p. 56
6.11.1 Description p. 56
6.11.2 Procedures p. 57
6.11.3 Impacts on existing nodes and functionality p. 58
6.12 Solution #12: DCCF and MFAF Relocation p. 58
6.12.1 Description p. 58
6.12.2 Procedures p. 59
6.12.3 Impacts on existing nodes and functionality p. 60
6.13 Solution #13: NWDAF MTLF and NWDAF AnLF interoperability support for registration and discovery in 5GC p. 61
6.13.1 Description p. 61
6.13.2 Procedures p. 61
6.14 Solution #14: Enhance trained ML Model sharing via ML Model format p. 62
6.14.1 Description p. 62
6.14.1.1 General p. 62
6.14.1.2 Registration and discovery of NWDAF (containing MTLF) with ML model Format Information p. 63
6.14.3 Impacts on existing nodes and functionality p. 64
6.15 Solution #15: ML model sharing with different AnLF providers p. 64
6.15.1 Description p. 64
6.15.2 Procedures p. 65
6.15.3 Impacts on existing nodes and functionality p. 66
6.16 Solution #16: NWDAF assisted URSP determination p. 66
6.16.1 Description p. 66
6.16.1.1 Input Data p. 69
6.16.1.2 Output Analytics p. 69
6.16.2 Procedures p. 70
6.16.3 Impacts on existing nodes and functionality p. 70
6.17 Solution #17: NSSP in roaming scenarios using network analytics p. 71
6.17.1 Description p. 71
6.17.2 Procedures p. 71
6.17.3 Impacts on existing nodes and functionality p. 72
6.18 Solution #18: Integrity KPI for QoS Sustainability Analytics p. 72
6.18.1 Key Issue mapping p. 72
6.18.2 Description p. 72
6.18.3 Procedures p. 74
6.18.4 Impacts on services, entities and interfaces p. 75
6.18.4.1 Input data p. 75
6.18.4.2 Output analytics p. 75
6.19 Solution #19: Enhanced QoS Sustainability Analytics in the finer granularity area p. 76
6.19.1 Description p. 76
6.19.2 Procedures p. 76
6.19.2.1 General p. 76
6.19.2.2 Input data p. 77
6.19.2.3 Output analytics p. 79
6.19.2.4 Procedures p. 79
6.19.3 Impacts on existing nodes and functionality p. 80
6.20 Solution #20: GTP metrics for QoS Sustainability analytics p. 80
6.20.1 Description p. 80
6.20.2 Input data p. 80
6.20.3 Procedures p. 81
6.20.4 Impacts on services, existing nodes and functionality p. 81
6.21 Solution #21: Federated Learning procedure between different NWDAFs p. 81
6.21.1 Description p. 81
6.21.2 Procedures p. 81
6.21.2.1 NWDAF "FL Capability" Registration p. 81
6.21.2.2 NWDAF Selection for FL operation p. 82
6.21.2.3 FL operation between NWDAFs containing MTLF p. 82
6.21.3 Impacts on existing nodes and functionality p. 84
6.22 Solution #22: Federated learning group creation p. 85
6.22.1 Description p. 85
6.22.2 Procedures p. 85
6.22.3 Impacts on existing nodes and functionality p. 87
6.23 Solution #23: Support of federated learning for model training p. 87
6.23.1 Description p. 87
6.23.2 Procedures p. 88
6.23.3 Impacts on services, entities and interfaces p. 92
6.24 Solution #24: Horizontal Federated Learning among Multiple NWDAFs p. 92
6.24.1 Description p. 92
6.24.2 Procedures p. 93
6.24.2.1 General procedure for Federated Learning among Multiple NWDAF Instances p. 93
6.24.2.2 Procedure for usage of Federated Learning in Abnormal Behaviour p. 95
6.24.3 Nnwdaf services used for model sharing/parameter exchanging p. 96
6.24.3.1 Nnwdaf_MLAggregation Service (Option 1) p. 96
6.24.3.2 Extended Nnwdaf_MLModelProvision Service (Option 2) p. 96
6.24.4 Impacts on existing nodes and functionality p. 97
6.25 Solution #25: Outdoors Advertisement use case with finer granularity location information p. 98
6.25.1 Description p. 98
6.25.2 Input Data p. 98
6.25.3 Output Analytics p. 98
6.25.4 Procedures p. 99
6.25.5 Impacts on existing nodes and functionality p. 100
6.26 Solution #26: Finer granularity of location information based on cell sequence p. 100
6.26.1 Description p. 100
6.26.2 Procedures p. 100
6.26.3 Impacts on existing nodes and functionality p. 101
6.27 Solution #27: Relative Proximity Analytics p. 102
6.27.1 Description p. 102
6.27.2 Procedures p. 104
6.27.3 Impacts on existing nodes and functionality p. 105
6.28 Solution #28: Detect and Improve correctness of NWDAF analytics p. 106
6.28.1 Description p. 106
6.28.2 Procedures p. 108
6.28.3 Impacts on Existing Nodes and Functionality p. 110
6.29 Solution #29: Detection of ML Model degradation and actions p. 110
6.29.1 Description p. 110
6.29.2 Procedures p. 111
6.29.2.1 General p. 111
6.29.2.2 Detection of ML Model Degradation p. 111
6.29.2.3 Act Upon ML Model Degradation p. 111
6.29.2.4 Procedure for the registration of ML model monitoring p. 112
6.29.2.5 Procedure for the detection of ML model degradation p. 112
6.29.2.6 Procedures for ML Model Provisioning p. 114
6.29.3 Impacts on Existing Nodes and Functionality p. 115
6.30 Solution #30: Improve correctness of NWDAF analytics p. 115
6.30.1 Description p. 115
6.30.2 Procedures p. 116
6.30.2.1 Procedures for improving model performance based on analytics correctness evaluation of one candidate model p. 116
6.30.2.2 Procedures for improving model performance based on analytics correctness evaluation of multiple candidate models p. 117
6.30.3 Impacts on existing nodes and functionality p. 118
6.31 Solution #31: Multiple Analytics outputs based NF action decision p. 118
6.31.1 Description p. 118
6.31.2 Procedures p. 118
6.31.3 Impacts on existing nodes and functionality p. 119
6.32 Solution #32: Enhanced ML model provisioning p. 119
6.32.1 Description p. 119
6.32.2 Procedures p. 120
6.32.3 Impacts on Existing Nodes and Functionality p. 122
6.33 Solution #33: Improving correctness of NWDAF analytics by providing correction information p. 122
6.33.1 Description p. 122
6.33.2 Procedures p. 123
6.33.3 Impacts on Existing Nodes and Functionality p. 124
6.34 Solution #34: Enhancing the accuracy of NWDAF Analytics p. 125
6.34.1 Description p. 125
6.34.2 Procedures p. 126
6.34.2.1 AnLF-initiated accuracy monitoring p. 126
6.34.2.2 MTLF-initiated accuracy monitoring p. 127
6.34.3 Impacts on services, entities and interfaces p. 129
6.35 Solution #35: Improve model training and provisioning exploiting sub-areas with similar statistical properties p. 129
6.35.1 Description p. 129
6.35.2 Procedures p. 131
6.35.2.1 Detecting and managing sub-areas using the Area Monitoring analytics service p. 131
6.35.2.2 Usage of sub-areas to improve ML model provisioning p. 133
6.35.3 Impacts on services, entities and interfaces p. 134
6.36 Solution #36: Enhanced provisioning of ML model based on information about how inference will be executed p. 134
6.36.1 Description p. 134
6.36.2 Procedures p. 136
6.36.3 Impacts on services, entities, and interfaces p. 137
6.37 Solution #37: Analytics Exchange in Home routed roaming case p. 137
6.37.1 Description p. 137
6.37.2 Procedures p. 139
6.37.3 Impacts on Existing Nodes and Functionality p. 140
6.38 Solution #38: Interactions between VPLMN and HPLMN for restricted data collection and analytics retrieval p. 140
6.38.1 Description p. 140
6.38.1.1 General p. 140
6.38.1.2 HPLMN collecting data from VPLMN p. 140
6.38.1.3 VPLMN consuming analytics generated by HPLMN p. 141
6.38.2 Procedures p. 142
6.38.2.1 HPLMN collecting data from VPLMN p. 142
6.38.2.2 VPLMN consuming analytics generated by HPLMN p. 143
6.38.3 Impacts on Existing Nodes and Functionality p. 144
6.39 Solution #39: Architecture enhancements to support Data and analytics exchange in roaming case p. 144
6.39.1 Description p. 144
6.39.1.1 General p. 144
6.39.2 Procedure p. 146
6.39.2.1 Procedure for NWDAF in HPLMN to collect data from VPLMN p. 146
6.39.2.2 Procedure for NWDAF consumer in HPLMN to obtain analytics from VPLMN NWDAF p. 148
6.39.3 Impacts on existing nodes and functionality p. 149
6.40 Solution #40: Data and analytics exchange for roaming UEs p. 149
6.40.1 Description p. 149
6.40.2 Procedures p. 150
6.40.3 Impacts on Existing Nodes and Functionality p. 151
6.41 Solution #41: Sending data that is about to be purged p. 152
6.41.1 Description p. 152
6.41.2 Procedures p. 152
6.41.3 Impacts on services, entities and interfaces p. 152
6.42 Solution #42: Storage and retrieval of trained ML models to/from ADRF p. 153
6.42.1 Description p. 153
6.42.2 Procedures p. 153
6.43 Solution #43: ML model storage in ADRF and ML model provision from ADRF p. 156
6.43.1 Description p. 156
6.43.2 Procedures p. 157
6.43.3 Impacts on Existing Nodes and Functionality p. 158
6.44 Solution #44: DCCF Reselection when multiple instances of DCCF exist in a network p. 158
6.44.1 Description p. 158
6.44.2 Procedures p. 158
6.44.3 Impacts on Services, Entities, and Interfaces p. 159
6.45 Solution #45: Managing Impact of Muting on NF Producer p. 160
6.45.1 Description p. 160
6.45.2 Procedures p. 161
6.45.3 Impacts on services, entities and interfaces p. 161
6.46 Solution #46: ADRF / NWDAF Data Storage Management p. 162
6.46.1 Description p. 162
6.46.2 Procedures p. 163
6.46.3 Impacts on services, entities and interfaces p. 165
6.47 Solution #47: Sharing models between NWDAFs p. 165
6.47.1 Description p. 165
6.47.2 Procedures p. 166
6.47.3 Impacts on services, entities and interfaces p. 167
6.48 Solution #48: PCF/UE re-evaluates the URSP rules according to NWDAF's analytics p. 167
6.48.1 Impacts of NWDAF analytics towards URSP rules p. 167
6.48.2 PCF updates or generates URSP rules according to NWDAF's analytics p. 168
6.48.3 Extensions of Analytic ID = "Service Experience" p. 168
6.48.4 Newly introduce Analytic ID = "URSP rules experience" p. 171
6.48.5 Procedures for URSP rules re-evaluation in PCF and UE p. 174
6.48.6 Impacts on services, entities and interfaces p. 175
6.49 Solution #49: NWDAF assisted URSP decision for redundant transmission p. 175
6.49.1 Description p. 175
6.49.2 Procedures p. 177
6.49.3 Impacts on Existing Nodes and Functionality p. 179
6.50 Solution #50: Enhancements on QoS Sustainability analytics p. 179
6.50.1 Description p. 179
6.50.1.1 Input data p. 180
6.50.2 Procedures p. 180
6.50.3 Impacts on Existing Nodes and Functionality p. 181
6.51 Solution #51: Selection, Monitoring, and Maintenance of NWDAF(s) for Federated Learning in 5GC p. 182
6.51.1 Description p. 182
6.51.2 Procedures p. 182
6.51.2.1 Client NWDAF(s) Selection in Federated Learning Preparation Phase p. 182
6.51.2.2 NWDAFs Monitoring and Re-selection in Federated Learning Execution Phase p. 184
6.51.2.3 Dynamic Discovery and Joining of New NWDAF(s) in Federated Learning Execution Phase p. 184
6.51.3 Impacts on services, entities and interfaces p. 186
6.52 Solution #52: FL training update to NWDAF containing AnLF from NWDAF containing MTLF p. 186
6.52.1 Description p. 186
6.52.2 Procedures p. 187
6.53 Solution #53: Horizontal Federated Learning with Multiple NWDAF p. 188
6.53.1 Description p. 188
6.53.2 Procedures p. 189
6.53.2.1 Procedure for registration multiple NWDAF (containing MTLF) with capability of Horizontal Federated Learning p. 189
6.53.2.2 Horizontal Federated Learning Training triggered by MTLF with FL Server capability p. 189
6.53.2.3 Horizontal Federated Learning Training triggered by MTLF only with FL Client capability p. 190
6.53.2.4 Horizontal Federated Learning training procedure p. 191
6.53.3 Impacts on existing nodes and functionality p. 191
6.54 Solution #54: Finer granular location information based on LCS input data p. 192
6.54.1 Description p. 192
6.54.2 Procedures p. 193
6.54.3 Impacts on Existing Nodes and Functionality p. 193
6.55 Solution #55: location information with finer granularity in horizontal and vertical directions p. 193
6.55.1 Description p. 193
6.55.2 Input Data p. 193
6.55.3 Output Data p. 194
6.55.4 Procedures p. 195
6.55.5 Impacts on Existing Nodes and Functionality p. 195
6.56 Solution #56: PSAP resolution with finer granularity of location information p. 196
6.56.1 Description p. 196
6.56.2 Functional description p. 196
6.56.3 Procedures p. 196
6.56.4 Impacts on services, entities and interfaces p. 197
6.57 Solution #57: NWDAF determines granularity when the consumer requests finer granularity location information p. 197
6.57.1 Description p. 197
6.57.2 Input Data p. 198
6.57.3 Output Analytics p. 199
6.57.4 Procedures p. 199
6.57.5 Impacts on Existing Nodes and Functionality p. 200
6.58 Solution #58: Supporting UE mobility analytics with finer granularity than TA/cell p. 200
6.58.1 Description p. 200
6.58.2 Procedures p. 201
6.58.3 Impacts on services, entities and interfaces p. 203
6.59 Solution #59: Enhancement of NWDAF with location accuracy prediction p. 203
6.59.1 Functional Description p. 203
6.59.1.1 General p. 203
6.59.2 Procedure p. 205
6.59.3 Impacts on services, entities and interfaces p. 205
6.60 Solution #60: Interactions with MDAF p. 206
6.60.1 Description p. 206
6.60.1.1 Input data p. 206
6.60.2 Procedures p. 207
6.60.3 Impacts on Existing Nodes and Functionality p. 207
6.61 Solution #61: Improving correctness by retrieval ML model from ADRF p. 207
6.61.1 Description p. 207
6.61.2 Procedures p. 208
6.61.2.1 General p. 208
6.61.3 Impacts on services, entities and interfaces p. 209
6.62 Solution #62: Improving the correctness of analytics based on the provision of context information p. 210
6.62.1 Description p. 210
6.62.2 Procedures p. 210
6.62.3 Impacts on services, entities and interfaces p. 210
6.63 Solution #63: Improving the correctness of NWDAF by rating the quality of the data sources p. 211
6.63.1 Description p. 211
6.63.2 Procedures p. 211
6.63.3 Impacts on services, entities and interfaces p. 213
6.64 Solution #64: Optimize the collection and reporting of network data p. 213
6.64.1 Description p. 213
6.64.2 Procedures p. 214
6.64.2.1 Training data is reported in time periods in model training scenario p. 214
6.64.2.2 Data collection with dynamic frequency in model inference scenario p. 215
6.64.3 Impacts on Existing Nodes and Functionality p. 216
6.65 Solution #65: Optimizing data collection and storage by NWDAF registration in UDM for all Analytics IDs p. 216
6.65.1 Description p. 216
6.65.2 Procedures p. 217
6.65.3 Impacts on services, entities and interfaces p. 217
6.66 Solution #66: DCCF relocation initiated by source DCCF or central DCCF p. 217
6.66.1 Description p. 217
6.66.2 Procedures p. 218
6.66.3 Impacts on Existing Nodes and Functionality p. 219
6.67 Solution #67: Managing analytics input data from ADRF p. 219
6.67.1 Description p. 219
6.67.2 Procedures p. 220
6.67.3 Impacts on services, entities and interfaces p. 221
6.68 Solution #68: Using data synthesis and compression for data storage and transfer p. 221
6.68.1 Description p. 221
6.68.2 Procedures p. 222
6.68.2.1 NWDAF requests ADRF to store data using DSC p. 222
6.68.2.2 NWDAF retrieves data from ADRF using DSC p. 222
6.68.3 Impacts on services, entities and interfaces p. 223
6.69 Solution #69: Model performance guarantee during Federated Learning p. 223
6.69.1 Description p. 223
6.69.2 Procedures p. 224
6.69.3 Impacts on services, entities and interfaces p. 225
6.70 Solution #70: Improved control of location granularity p. 226
6.70.1 Description p. 226
6.70.2 Procedures p. 227
6.70.3 Impacts on Existing Nodes and Functionality p. 227
6.71 Solution #71: Traffic flow statistics use case with finer granularity location information p. 227
6.71.1 Description p. 227
6.71.2 Input Data p. 227
6.71.3 Output Analytics p. 228
6.71.4 Procedures p. 228
6.71.5 Impacts on Existing Nodes and Functionality p. 229
6.72 Solution #72: Use of MDAS analytics for improving AnLF/MTLF analytics accuracy and data collection p. 229
6.72.1 Description p. 229
6.72.2 Procedures p. 231
6.72.2.1 AnLF subscribing from the MDAS/MDAF for fault prediction p. 231
6.72.2.2 MTLF subscribing from the MDAS/MDAF for fault prediction p. 233
6.72.3 Impacts on services, entities and interfaces p. 235
6.73 Solution #73: How NWDAF requests analytics from MDA Management Function p. 235
6.73.1 Description p. 235
6.73.2 Procedures p. 236
6.73.3 Impacts on Existing Nodes and Functionality p. 237
6.74 Solution #74: Supporting NWDAF interactions with MDAF p. 237
6.74.1 Description p. 237
6.74.2 Procedures p. 239
6.74.3 Impacts on services, entities and interfaces p. 241
7 Overall Evaluation p. 241
7.1 Key Issue #1: How to improve correctness of NWDAF analytics p. 241
7.2 Key Issue #2: NWDAF-assisted application detection p. 245
7.3 Key Issue #3: Data and analytics exchange in roaming case p. 245
7.3.1 Solution categorization p. 245
7.3.2 Evaluation of solutions on general architectural enhancements p. 246
7.3.3 Evaluation of solutions on enhancements for specific use cases/functionalities/Analytics IDs p. 250
7.4 Key Issue #4: How to Enhance Data collection and Storage p. 251
7.5 Key Issue #5: Enhance trained ML Model sharing p. 254
7.6 Key Issue #6: NWDAF-assisted URSPs p. 254
7.7 Key Issue #7: Enhancements on QoS Sustainability analytics p. 258
7.8 Key Issue #8: Supporting Federated Learning in 5GC p. 260
7.9 Key Issue #9: Enhancement of NWDAF with finer granularity of location information p. 262
7.10 Key Issue #10: Interactions with MDAS/MDAF p. 267
8 Conclusions p. 268
8.1 Key Issue #1: How to improve correctness of NWDAF p. 268
8.2 Key Issue #2: NWDAF-assisted application detection p. 270
8.3 Key Issue #3: Data and analytics exchange in roaming case p. 271
8.4 Key Issue #4: How to Enhance Data collection and Storage p. 271
8.5 Key Issue #5: Enhance trained ML model sharing p. 272
8.6 Key Issue #6: NWDAF-assisted URSPs p. 273
8.7 Key Issue #7: Enhancements on QoS Sustainability analytics p. 274
8.8 Key Issue #8: Supporting Federated Learning in 5GC p. 274
8.9 Key Issue #9: Enhancement of NWDAF with finer granularity of location information p. 276
8.10 Key Issue #10: Interactions with MDAS/MDAF p. 277
$ Change history p. 278
