R1-2407654
discussion
Discussion on AI/ML for positioning accuracy enhancement
From Huawei
Summary
This Huawei document presents 25 proposals for AI/ML-based positioning accuracy enhancement in 5G NR, covering model input/output specifications, training procedures, consistency mechanisms, and lifecycle management across different positioning cases (Case 1: UE-based, Case 2: UE-assisted, Case 3: gNB-assisted).
Position
Huawei advocates for pragmatic implementation-based solutions over rigid standardization, arguing that ambiguity issues can be resolved through consistent vendor implementations rather than tight specifications. They strongly oppose complex new signaling mechanisms, pushing instead for reuse of legacy mechanisms (path-based measurements, existing quality indicators, LocationUncertainty for labels). They resist introducing new IDs or complex assistance data, favoring implementation flexibility and proprietary protection (limiting sample count to 16). This positions them against companies likely pushing for more standardized, tightly-specified approaches.
Key proposals
- Proposal 1 (Sec 2.1): For the study of sample-based measurements for Case 3b, we do not need to pursue tightly specified rule or tightly aligned parameters - as long as the rule and parameters could be specified to any extent, that would be helpful for alleviating the ambiguity across vendors
- Proposal 2 (Sec 2.1): For the study of sample-based measurements for Case 3b, the starting time of the selection window is based on the timing of the first detected path
- Proposal 4 (Sec 2.1): The maximum number of Nt' shall not be larger than 16 to protect proprietary channel estimation performance
- Proposal 6 (Sec 2.1): Regarding the type of channel measurements for Case 3b, support the use of legacy mechanism subject to path-based measurements
- Proposal 8 (Sec 2.1): For Case 3b, reuse the legacy reporting of timing information or timing and power information from gNB/UE to LMF. The use of phase information for the measurement reporting would need further justification
- Proposal 10 (Sec 3): For Case 3a, LOS/NLOS indicator can reuse the same format as the existing IE 'LoS/NLoS Information' in 38.455
- Proposal 11 (Sec 3): For Case 3a, to distinguish the reported timing information from the legacy measurement-based timing information, consider Alt.1: Reuse the timing quality indicator or Alt.2: Introduce an indicator to tell that the report is subject to measured timing information or non-measured timing information
- Proposal 13 (Sec 4): For direct positioning, if needed, the UE location quality used in legacy reporting, i.e., LocationUncertainty, can be used for the quality indication of the label for training. There is no need to discuss other options
- Proposal 15 (Sec 4): For Case 3a, the label of the timing information generated from LMF can be in forms of the ToF/propagation delay between the TRP and the PRU
- Proposal 17 (Sec 4): For training data collection of Case 1, no enhancement is needed to provide assistance information (regarding area, PRS, network synchronization error, antenna/beam, etc.) to the UE
- Proposal 18 (Sec 5): For Case 1 and 2a, no need to introduce an associated ID to provide NW-side additional condition for ensuring the consistency between training and inference. Legacy assistance parameters in legacy positioning method can be reused if needed
- Proposal 21 (Sec 6): For model monitoring in Case 1, conclude that Option A-4 is sufficient where the monitoring decision is performed in the same UE-side entity that derives the monitoring metric
- Proposal 22 (Sec 6): For performance monitoring of AI/ML positioning Case 3a, the request/report of a measured or non-measured result can be used to indicate the activation/fallback between LMF and gNB
- Proposal 24 (Sec 6): For model monitoring in Case 3b, no further assistance information or measurement report in addition to inference is required to be sent to the LMF
- Proposal 25 (Sec 7): Support functionality-based LCM for UE-side model of Case 1/2a