R1-2500337
discussion
Specification support for beam management
From vivo
Summary
This document from vivo addresses specification support for AI/ML-based beam management in NR, focusing on consistency issues for UE-side models, performance monitoring procedures, and reference signal configurations. It contains 62 proposals and 6 observations covering topics such as associated ID configuration, Set A/Set B pattern management, overhead reduction techniques like Time Resource Indicators (TRI), and beam failure detection enhancements.
Position
Vivo proposes that the Associated ID be mandatorily configured in both the inference parameter set and the CSI framework to resolve consistency issues between training and inference phases. They require the introduction of a Pattern ID for Set B and a cell indicator for the Associated ID to manage local vs. global ID complexities and handover scenarios. Vivo supports the use of quasi-best Rx beams derived from P3 measurements on a subset of Tx beams to balance performance and overhead. For reporting, they propose using Time Resource Indicators (TRI) and differential quantization with a single reference value to significantly reduce UCI overhead for both UE-side and NW-side models in BM-Case 2. They also advocate for UE-assisted performance monitoring to calculate Beam Accuracy Indicators (BAI) locally, thereby reducing network-side processing and signaling load.
Key proposals
- Proposal 3 (Sec 2.1): For inference, for UE-side model, associated ID should be mandatory configured within both inference parameter set and CSI framework to ensure consistency between training and inference.
- Proposal 6 (Sec 2.2): For inference, for UE-side model, an indicator should be configured within the CSI-ReportConfig to clearly specify whether processing follows Option A (CSI-ReportConfig for inference) or Option B (inference parameter set for applicability only).
- Proposal 7 (Sec 2.3): For inference, for UE-side model, the inference parameter set in Step 3 must include Associated ID, Pattern ID, Set A/B sizes, report content type, BM case indicator, and for Case 2, periodicity and occasion counts.
- Proposal 13 (Sec 3.2): For performance monitoring, for UE-side model BM-Case 1, support associating the inference reference point to the monitoring reference point, where the UE identifies the monitoring reference point first.
- Proposal 23 (Sec 3.4): For monitoring Type 1 Option 2, for BM-Case 1, support reporting the number of correct predictions as the Beam Accuracy Indicator (BAI) result, with field size determined by the number of monitoring occasions.
- Proposal 28 (Sec 4.2): For inference, for UE-side model, support UE to recommend preferred Set B patterns trained during the model training phase to prevent performance degradation from mismatched patterns.
- Proposal 31 (Sec 4.3): For inference, for UE-side model, Set A or resources in Set A should be indicated as a virtual Set or virtual resource, which does not require measurement on these resources for the report.
- Proposal 34 (Sec 4.5): For data collection and inference, for UE-side model, support using a quasi-best Rx beam for Set A and/or Set B measurement, derived from P3 measurement on a small number of Tx beams.
- Proposal 38 (Sec 6.1): For inference, for UE-side model, support reporting predicted L1-RSRPs and corresponding beam information of up to M beams within X dB gap to the largest predicted value, along with the number of reported beams.
- Proposal 45 (Sec 6.4): For inference, for UE-side model, support reporting TRI (Time Resource Indicator) instead of direct predicted beam resource indication with implicit time stamps to reduce overhead in BM-Case 2.
- Proposal 54 (Sec 7.2): For inference, for NW-side model, support reporting L1-RSRPs and corresponding beam information of up to M beams within X dB gap to the largest measured value of L1-RSRP.
- Proposal 58 (Sec 7.4): For data collection and inference, for NW-side model, support reporting TRI instead of direct predicted beam resource indication scheme with implicit time stamp to leverage temporal correlation.
- Proposal 60 (Sec 9): For UE-side model, data collection procedure can be initiated by gNB configuration or UE request signaling.
- Proposal 61 (Sec 10): Support using AI beam prediction for beam failure detection/report enhancement, allowing early identification of beam failure instances based on predicted L1-RSRP thresholds.