R1-2410102
discussion
On specification for AI/ML-based positioning accuracy enhancements
From OPPO
Summary
OPPO's technical document presents a comprehensive analysis of AI/ML-based positioning accuracy enhancements for NR Release 19, covering five positioning cases (Case 1, 2a, 2b, 3a, 3b) with 26 detailed proposals and 3 observations addressing measurement enhancement, training/inference consistency, data collection, model inference, performance monitoring, and functionality management.
Position
OPPO advocates for minimal specification impact by leveraging existing protocols (LPP, NRPPa) and leaving implementation details to vendors for flexibility, while strongly opposing phase information reporting due to limited performance gains versus overhead. They push for associated ID mechanisms to ensure training/inference consistency without exposing proprietary network information, and resist specifying performance monitoring mechanisms without ground-truth labels, preferring implementation-based approaches over standardized solutions.
Key proposals
- Proposal 1 (Sec 3.1): For Case 2b/3b time domain channel measurements, existing path-based measurement can be used as LMF-side model input without RAN1 specification impact; if sample-based measurement is supported, use Option A (starting time based on first detected path timing)
- Proposal 2 (Sec 3.1): NOT support reporting based on phase information in addition to timing and power information for R19 AI-based positioning
- Proposal 4 (Sec 3.2): Signal associated ID from network to ensure consistency between AI model training and inference for Case 1 and Case 2a, without disclosing proprietary network information
- Proposal 7 (Sec 3.3): For UE-side data collection (Case 1/2a), reuse legacy LPP signaling with associated ID for consistency, no additional triggering needed, format/content up to UE implementation
- Proposal 11 (Sec 3.3): Rel-19 NOT to specify mechanisms for delivering collected data when training entity differs from data collection entity across all AI positioning cases
- Proposal 12 (Sec 3.4): For Case 1 model inference, signal associated ID for consistency, no need to specify AI model input format, use existing LPP signaling with new indication for AI vs legacy method
- Proposal 17 (Sec 3.5): NOT specify dedicated specification enhancement for functionality/model performance monitoring without ground-truth labels - can be done by implementation
- Proposal 18 (Sec 3.5): For Case 1, NOT specify dedicated specification enhancement for functionality/model performance monitoring - can be done by UE implementation without air interface impact
- Proposal 22 (Sec 3.5): For Case 3a, support NRPPa signaling enhancement to deliver ground-truth labels from LMF to gNB for performance monitoring
- Proposal 24 (Sec 3.6): For Case 1 and Case 2a, functionality/model can be activated/deactivated by LPP signaling, UE can autonomously fallback to legacy operations with indication in reporting
- Proposal 25 (Sec 3.7): Specify UE capability signaling to report supported configurations associated with given functionalities for Case 1 and Case 2a
- Proposal 26 (Sec 3.7): UE can report applicable functionalities by sending ProvideCapabilities message to LMF, triggering capability indication procedure