R1-2410205
discussion
AI/ML positioning accuracy enhancement
From Fraunhofer
Summary
This Fraunhofer technical document presents 19 proposals and 1 observation for AI/ML positioning accuracy enhancements in NR air interface, covering measurement enhancements, training data collection optimization, and comprehensive model lifecycle management frameworks for Cases 1 and 3a.
Position
Fraunhofer advocates FOR comprehensive AI/ML positioning frameworks that maintain network-centric control (LMF-based functionality management) while supporting flexible measurement approaches and intelligent model lifecycle management. They push FOR complex-valued CIR reporting to preserve information richness, event-based training data collection optimization, and two-stage monitoring processes. They are positioned AGAINST UE-autonomous functionality management without network oversight and advocate FOR balanced approaches that consider both performance and overhead costs in model management decisions.
Key proposals
- Proposal 1 (Sec 2.1): Support both Alternative (a) sample-based measurement and Alternative (b) path-based measurement report
- Proposal 4 (Sec 2.2): Support complex valued sample-based reporting offering lossless reporting of the channel impulse response and supporting future enhancements of the AI/ML model
- Proposal 5 (Sec 3.1): Training data collection can be configured to collect any proportion of unlabeled (data samples with only part A) and labeled (data samples with both parts A and B) data
- Proposal 6 (Sec 3.2): Support opportunistic/event-based training data collection for follow-up model improvement/adaptation triggered by specific input patterns, sub-area challenges, high quality GTLs, or monitoring events
- Proposal 7 (Sec 4.1): At least for Case 1 and Case 3a, functionality management is provided by the network (LMF)
- Proposal 8 (Sec 4.2): At least for Case 1 and Case 3a, functionality monitoring is triggered by the NW (LMF)
- Proposal 11 (Sec 4.3): For model performance monitoring of AI/ML positioning, ensure that Rel-19 supports the necessary signaling mechanisms for Options A-1, A-2, A-3, B-1, and B-2
- Proposal 12 (Sec 4.4): For Case 1/3a, if the UE/gNB model is proprietary to the NW, model management is performed by the UE/gNB. If developed by NW side, management can be performed by either LMF or UE/gNB
- Proposal 13 (Sec 4.5): For both Case 1 and Case 3a consider a two-stage monitoring process: upon identifying potential model performance issues through early metrics, activate broader monitoring metrics to confirm presence and impact
- Proposal 14 (Sec 4.5): The possible actions/decisions include enhanced long-term monitoring, functionality/model activation/switching/fallback, permanent area marking as non-applicable, and AI/ML feature deactivation
- Proposal 15 (Sec 4.5): Support validity indication for AI/ML models including at least information about the existence of ML assisted areas
- Proposal 16 (Sec 4.6): UE/gNB provides expected performance vs cost/overhead information to LMF including model properties, availability, and activation latency
- Proposal 19 (Sec 4.6): In Cases 1 and 3a, the NW configures multiple functionalities at UE/gNB and provides configuration for automatic functionality switching and fallback based on specific parameters or threshold values