R1-2407728
discussion
Discussion on AI/ML for beam management
From China Telecom
Summary
China Telecom presents a comprehensive technical contribution on AI/ML for NR beam management, covering lifecycle management (LCM) aspects including data collection, model inference, and performance monitoring for both network-sided and UE-sided AI/ML models. The document contains 13 detailed proposals addressing BM-Case 1 (spatial beam prediction) and BM-Case 2 (temporal beam prediction).
Position
China Telecom advocates for a comprehensive AI/ML beam management framework that supports both network-sided and UE-sided models with flexible data collection mechanisms. They push FOR beam prediction accuracy KPIs as the primary performance monitoring metric and support hybrid RSRP reporting (predicted vs measured). They push AGAINST using only probability information for performance monitoring and oppose single resource set configuration for Set B in UE-sided models, arguing these approaches lack sufficient accuracy and network utility.
Key proposals
- Proposal 1 (Sec 2.1.1): Support three types of collected data content for NW-sided AI/ML models - M1 L1-RSRPs with beam indication, M2 L1-RSRPs based on beam set measurement, and M3 beam indices based on beam set measurement
- Proposal 2 (Sec 2.1.1): Configure one or two reference signal resource sets for NW-sided AI/ML model data collection, with Set B indication needed when single resource set is used
- Proposal 3 (Sec 2.1.2): For UE-sided models, data collection can be triggered by UE or initiated by network
- Proposal 4 (Sec 2.2.1): For NW-sided model inference, report content for beam information can be L1-RSRP only or L1-RSRP with beam ID
- Proposal 5 (Sec 2.2.1): Maximum number of reported beam information in one report should be configurable by network based on UE capability
- Proposal 6 (Sec 2.2.1): Support timestamp information reporting for Set B measurements in BM-Case 2 for NW-sided models
- Proposal 7 (Sec 2.2.2): Do not support configuring only one resource set for Set B in UE-sided model inference results reporting for BM-Case 1
- Proposal 8 (Sec 2.2.2): Support Option 2 for UE-sided model inference results - report beam information and RSRP of predicted Top-K beams with highest predicted RSRP above proper probability threshold
- Proposal 9 (Sec 2.2.2): Report predicted RSRP for non-measured beams and measured L1-RSRP for measured beams in UE-sided model Top-K results
- Proposal 10 (Sec 2.3.1): Support beam prediction accuracy related KPIs as primary performance monitoring metric
- Proposal 11 (Sec 2.3.1): Prioritize Alternative 2 for UE-assisted performance monitoring - L1-RSRP difference information based on Top-K predicted beam measurements
- Proposal 12 (Sec 2.3.1): Perform model switch or fallback to non-AI/ML method when AI/ML model failure count reaches predefined threshold
- Proposal 13 (Sec 2.4): Confirm working assumption for consistency of associated ID for UE-sided models within cell boundaries