R1-2410174
discussion
Discussion on AI/ML for beam management
From HONOR
Summary
HONOR's technical document presents proposals for AI/ML-based beam management in NR, addressing performance monitoring metrics, inference reporting for both UE-sided and network-sided models, and measurement report optimization. The document contains 13 specific proposals across performance monitoring, model inference, and reporting mechanisms.
Position
HONOR advocates FOR Alt 2 definition of Top-K beam prediction accuracy (measuring if top measured beam is within top-K predicted beams) as more appropriate for scenarios with second-round beam sweeping, supports UE pre-evaluation mechanisms to reduce signaling overhead, and pushes for flexible reporting formats that adapt based on resource set size and number of reported beams. They advocate FOR including probability information in UE-sided model reports despite potential trust issues, arguing it helps gNB estimate prediction reliability.
Key proposals
- Proposal 1 (Sec 2.1): For BM-Case1 and BM-Case2 with UE-sided AI/ML model, support Alt 2 definition of Top-K beam prediction accuracy as percentage of 'Top-1 measured beam is one of Top-K predicted beams' in all evaluation occasions
- Proposal 2 (Sec 2.1): Define resource set for monitoring as full set or subset of Set A, with Top-K measured beams having K highest measured RSRPs and Top-K predicted beams having K highest predicted RSRPs or probability
- Proposal 3 (Sec 2.2): Support UE pre-evaluation of predicted beam quality for Type 1 Option 2 monitoring, allowing gNB to decide whether to trigger full Set A performance monitoring
- Proposal 4 (Sec 3.1): For NW-sided model BM-Case 1, support L1-RSRP reporting of up to M beams within X dB gap to largest measured L1-RSRP if count > N, otherwise report Top N beams, with gNB-configured M, N, X parameters where N<M<resource set size
- Proposal 5 (Sec 3.1): Support both Option 1 (CRI/SSBRI) and Option 2 (bitmap plus one beam index) for beam information reporting, with selection depending on number of reported beams and resource set size
- Proposal 6 (Sec 3.2): For UE-sided model BM-Case1, support Opt 3 including beam information and probability information of predicted Top K beams among Set A
- Proposal 7 (Sec 3.2): Support Option B for RSRP reporting - predicted RSRP if beam not configured for measurement, measured L1-RSRP if beam is configured for measurement
- Proposal 8 (Sec 4): In conclusions, support both Alt 1 and Alt 2 for Top-K beam prediction accuracy with Alt 2 preferred
- Proposal 9 (Sec 4): Resource set for monitoring can be full set or subset of Set A with defined Top-K beam selection criteria
- Proposal 10 (Sec 4): Support UE pre-evaluation capability for Type 1 Option 2 monitoring to reduce signaling overhead
- Proposal 11 (Sec 4): For NW-sided model, support adaptive beam reporting based on X dB gap threshold with minimum N beam guarantee
- Proposal 12 (Sec 4): Support flexible beam information format selection for NW-sided model based on reporting efficiency
- Proposal 13 (Sec 4): For UE-sided model, include probability information (Opt 3) in inference result reports to enable gNB reliability assessment