R1-2409396
discussion
Discussion on AI/ML for positioning accuracy enhancement
From Huawei
Huawei's prior position on
9.1.2
at
RAN1#118bis
· AI-synthesized, paraphrased
verify sources →
Proposes pragmatic implementation-based solutions over rigid standardization, arguing that vendor consistency can resolve ambiguity issues without complex new signaling mechanisms and favoring reuse of legacy procedures.
Summary
This Huawei contribution discusses AI/ML for positioning accuracy enhancement in NR, covering model input, output, training, consistency, monitoring, and lifecycle management. It contains 30 proposals and 17 observations, arguing for implementation flexibility, reuse of legacy mechanisms, and protection of proprietary channel estimation methods.
Position
Huawei argues that ambiguity in sample-based measurements for Case 3b can be avoided by implementation rather than strict specification, proposing that gNBs flexibly determine selection window parameters (Nt, Nt', k) while capping Nt' at 16 and Nt at 64 to protect proprietary channel estimation. They support enhancing legacy path-based reporting by increasing the number of reported paths to 16, citing simulation results showing sub-meter accuracy improvements. For Case 3a, they propose reusing the legacy 'LoS/NLoS Information' IE and distinguishing AI/ML timing outputs via timing quality indicators or new Rel-19 indicators. Regarding Case 1, they oppose introducing new positioning methods or implicit signaling, insisting on reusing legacy DL-TDOA assistance data explicitly. For monitoring, they argue that LMF involvement in Case 1 metric calculation is unnecessary and that Case 3a monitoring should rely on measured/non-measured result reporting to imply activation/fallback.
Key proposals
- Proposal 1 (Model Input): For sample-based measurements in Case 3b, do not pursue tightly specified rules or aligned parameters, as ambiguity can be avoided by implementation.
- Proposal 2 (Model Input): The starting time of the selection window for sample-based measurements should be based on the timing of the first detected path determined by gNB implementation.
- Proposal 3 (Model Input): The gNB can flexibly determine the values of Nt, Nt', and k for reported samples after receiving signaled parameters from the LMF.
- Proposal 4 (Model Input): To protect proprietary channel estimation, the maximum number of Nt' shall not be larger than 16, and the maximum number of Nt shall not be larger than 64.
- Proposal 6 (Model Input): Support the use of legacy mechanisms subject to path-based measurements for Case 3b.
- Proposal 7 (Model Input): Enhance legacy path-based reporting by increasing the number of reported delay and power values from the gNB to LMF (e.g., up to 16 paths).
- Proposal 8 (Model Input): Reuse legacy reporting of timing information or timing and power information for Case 3b, rejecting the necessity of phase information.
- Proposal 10 (Model Output): For Case 3a, the LOS/NLOS indicator can reuse the same format and definition as the existing IE 'LoS/NLoS Information' in 38.455.
- Proposal 13 (Model Output): Distinguish Case 3a timing information from legacy by reusing the timing quality indicator or introducing a Rel-19 indicator for measured vs. non-measured timing.
- Proposal 16 (Model Training): Use legacy UE location quality (LocationUncertainty) for the quality indication of the label for training in direct positioning.
- Proposal 21 (Consistency): For Case 1, reuse legacy DL-TDOA UE-based positioning without introducing a new positioning method.
- Proposal 23 (Consistency): Indicate assistance data for Case 1 in an explicit manner as in legacy, opposing implicit indicators or new associated IDs.
- Proposal 25 (Model Monitoring): For label-based model monitoring in Case 1, do not involve the LMF for metric calculation, as it may lack knowledge of UE-side models.
- Proposal 27 (Model Monitoring): For Case 3a, use the request/report of measured or non-measured results to imply activation/fallback between LMF and gNB.
- Proposal 30 (LCM): Support functionality-based LCM for UE-side models of Case 1 and 2a.