R1-2500090
discussion
Discussion on AI/ML for positioning accuracy enhancement
From Huawei
Huawei's prior position on
9.1.2
at
RAN1#119
· AI-synthesized, paraphrased
verify sources →
Argues that ambiguity in sample-based measurements can be avoided by implementation rather than strict specification, proposing flexible determination of selection window parameters by gNBs while capping Nt at 16 and Nt at 64 to protect proprietary channel estimation. Supports enhancing legacy path-based reporting by increasing the number of reported paths to 16 and proposes reusing legacy LoS/NLoS Information IEs for Case 3a. Opposes introducing new positioning methods or implicit signaling for Case 1, insisting on reusing legacy DL-TDOA assistance data explicitly.
Summary
This Huawei contribution addresses open issues for AI/ML-based positioning in NR Rel-19, covering model input/output, training data collection, consistency, monitoring, and lifecycle management. The document contains 28 proposals and 11 observations, primarily arguing for the reuse of legacy signaling mechanisms and opposing the introduction of new complex identifiers or phase-based inputs.
Position
Huawei argues against the necessity of phase information for AI/ML model input, citing that double phase difference cannot mitigate phase errors in NLOS scenarios and that timing/power suffices for performance. They oppose the introduction of an 'associated ID' for TRP location consistency (Alternative 1/2), presenting a technical case that TRP locations are infrequent to change and that UE-side burden from combinatorial model training would be excessive; instead, they support Alternative 3 where Info #7 is not provided. For model output, Huawei proposes reusing legacy IEs for LOS/NLOS indicators and suggests distinguishing Rel-19 timing information via timing quality indicators or a specific Rel-19 type indicator. Regarding monitoring, they propose that label-free monitoring be up to implementation and that for Case 3a, the reporting of measured versus non-measured results can implicitly signal model activation or fallback. Finally, they support functionality-based lifecycle management for UE-side models using legacy capability reporting procedures.
Key proposals
- Proposal 1 (Model Input): Sets the maximum number of Nt' for enhanced path-based measurement in Case 3b to 16.
- Proposal 2 (Model Input): Supports candidate values of k (0 to 5) for timing reporting granularity in enhanced path-based measurements.
- Proposal 4 (Model Input): Proposes reusing legacy reporting of timing or timing/power information for Case 3b, arguing phase information is unnecessary.
- Proposal 5 (Model Output): Proposes reusing the existing 'LoS/NLoS Information' IE format and definition for Case 3a LOS/NLOS indicators.
- Proposal 8 (Model Output): Suggests distinguishing Rel-19 timing information from legacy via timing quality indicator (Alt 1) or a new Rel-19 type indicator (Alt 2).
- Proposal 12 (Model Training): Proposes that labels for Case 3a timing information be in the form of ToF/propagation delay between TRP and PRU.
- Proposal 14 (Model Training): Defines pairing of Part A and Part B for Case 1 data collection based on implementation (same entity) or LMF-delivered timestamps (different entities).
- Proposal 19 (Consistency): Supports Alternative 3, stating Info #7 (TRP locations) should not be provided from LMF to UE, assuming consistency between training and inference.
- Proposal 21 (Model Monitoring): Proposes that label-free monitoring should be up to implementation.
- Proposal 25 (Model Monitoring): Suggests using the request/report of measured vs. non-measured results to imply activation/fallback for Case 3a monitoring.
- Proposal 28 (LCM): Supports functionality-based Lifecycle Management for UE-side models in Case 1/2a, reporting capabilities via legacy procedures.