R1-2409788
discussion
Discussion on Specification Support for AI/ML-based positioning
From Apple
Summary
Apple Inc. presents 45 proposals for Rel-19 AI/ML-based positioning, focusing on specification impacts for sample-based versus path-based measurements, model input types (CIR, PDP, DP), and data collection procedures. The document argues for supporting both measurement types, increasing path support to 128, and defining specific quality indicators and assistance data structures for training, inference, and monitoring across UE, gNB, and LMF entities.
Position
Apple proposes supporting both path-based and sample-based measurement inputs, arguing that sample-based input is a special case of path-based input with equi-spaced timing. They require increasing the number of additional paths (Nt) to 16, 32, 64, or 128 to ensure performance in low-complexity or small-bandwidth scenarios. Apple supports using Channel Impulse Response (CIR) as model input, including phase information, and proposes mitigating phase mismatch through training data compensation. They require specific quality indicators for channel measurements (timing, power, phase) and ground truth labels (0-1 scale for LOS/NLOS). For model monitoring, Apple proposes that the default entity for monitoring is the one hosting the model, supporting specific options (A-2, A-3, B-1) for Case 1 and (A-2, B-1) for Case 3a. They also propose defining AI/ML model assistance data to ensure consistency between training and inference conditions.
Key proposals
- Proposal 1 (Sec 2.1): Support both path-based measurement input and sample-based measurement input.
- Proposal 2 (Sec 2.1): Modify path-based feedback to support sample type feedback by signaling equal spacing size, increasing Nt to 16-128, and supporting absolute/differential RSRPP mapping.
- Proposal 5 (Sec 2.2): Re-use legacy reference time TSubframeRxi for UE channel measurements reported to the LMF.
- Proposal 8 (Sec 2.2): Support using phase information (CIR) in addition to timing and power for model input.
- Proposal 11 (Sec 3): Enhance measurement input reports to LMF for Case 3b to support feedback of CIR, PDP, and DP.
- Proposal 13 (Sec 4.1): Map channel quality indicator to timing, power, and phase quality depending on input type.
- Proposal 17 (Sec 4.3): Specify elements for model training data collection including Part A (measurement/quality/timestamp) and Part B (ground truth/quality/timestamp).
- Proposal 23 (Sec 4.4): Define AI/ML model assistance data to match UE/network conditions between training, inference, and monitoring.
- Proposal 29 (Sec 4.5): Default option for monitoring to occur at the entity with the AI/ML model.
- Proposal 31 (Sec 4.5): Support monitoring options A-2, A-3, and B-1 for Case 1.
- Proposal 37 (Sec 4.5): Support monitoring options A-2 and B-1 for Case 3a.
- Proposal 43 (Sec 5): Use functionality-based LCM for one-sided models without transfer, covering UE capability reporting for scenarios and positioning types.
- Proposal 45 (Sec 5): Include positioning types, measurement capabilities, and site-specific information in UE capability reporting.