R1-2410377
discussion
Discussion on AI/ML for positioning accuracy enhancement
From NTT DOCOMO
Summary
NTT DOCOMO's document provides a comprehensive technical analysis of AI/ML for NR positioning accuracy enhancements, covering data collection, model inference, performance monitoring, and lifecycle management aspects. The document presents 16 detailed proposals and 2 observations addressing specification impacts for Cases 1, 3a, and 3b positioning scenarios.
Position
NTT DOCOMO advocates FOR maximum reuse of existing legacy positioning mechanisms and IEs to minimize specification overhead while enabling AI/ML positioning enhancements. They push FOR sample-based measurements as baseline for better consistency between training and inference, implicit indication of network conditions via associated IDs rather than explicit signaling, and LMF-centric decision making for functionality management. They are AGAINST introducing unnecessary new signaling when existing mechanisms can be extended, and push back against explicit indication of AI/ML-generated measurements without strong justification.
Key proposals
- Proposal 1 (Sec 2.1): For data collection time stamps, existing IEs are reused with SFN level report (NR-TimeStamp) as baseline, using UTC time only when SFN reporting is insufficient
- Proposal 5 (Sec 2.2): Support sample-based measurement for AI/ML positioning in Rel-19, with LMF indicating sample-based vs path-based measurement based on UE capability
- Proposal 8 (Sec 2.2): For AI/ML positioning, phase information report is considered for model input in addition to timing and power information, using Rel-18 measurements as baseline
- Proposal 9 (Sec 2.2): Ensure consistency between training and inference by reusing existing assistance data IEs for DL-TDoA positioning method, with implicit indication of TRP beam/antenna and hardware conditions
- Proposal 11 (Sec 2.3): For AI/ML assisted positioning Case 3a, the necessity to indicate whether intermediate measurement is AI/ML generated needs further justification
- Proposal 12 (Sec 2.3): For Case 3a, LOS/NLOS indicator can reuse existing IE 'LOS/NLOS Information' in 38.455, reported with other measurements from legacy positioning methods
- Proposal 13 (Sec 2.4): For performance monitoring metric calculation of Case 1, both Option A (UE-side) and B (LMF-side) are considered, with both legacy positioning-based and PRU-based ground truth label generation
- Proposal 14 (Sec 2.4): For performance monitoring of Cases 1/3a, LMF is supported for functionality-level decision making
- Proposal 6 (Sec 2.2): For Case 3b sample-based measurement, legacy timing granularity parameters from TS 38.455 are baseline with value ranges (0…5) and (-6…-1), and sample numbers at least {8, 16, 32, …}
- Proposal 15 (Sec 2.4): For Case 1 Option A monitoring, UE performs metric calculation following NW indication including model ID/functionality information and performance metrics/threshold
- Proposal 3 (Sec 2.1): For Case 1 data collection, at least PRS configuration and training data validity area (e.g., AreaID-CellList) are indicated from LMF to UE for training-inference consistency
- Proposal 7 (Sec 2.2): For sample-based measurement definition, the start of Nt samples list is determined by fixed/configured offset relative to reference time
- Proposal 16 (Sec 2.5): For Case 1, RAN1 follows RAN2 agreement procedures for LCM, discussing capabilities to send supported/applicable functionalities to LMF