R1-2410547
discussion
Specification support for AI/ML beam management
From ITL
Summary
ITL's comprehensive technical document presents 25 detailed proposals for AI/ML-based beam management in NR air interface, covering both network-side and UE-side AI/ML models for spatial and temporal beam prediction (BM-Case1 and BM-Case2), along with performance monitoring mechanisms.
Position
ITL advocates FOR flexible beam reporting with up to 256 beams per report, larger quantization steps to reduce overhead, and supporting both network-side and UE-side AI/ML models with unified TCI framework. They push FOR reusing existing CSI reporting frameworks and specifications as baseline while enabling enhanced capabilities. ITL is AGAINST adding unnecessary complexity like probability/confidence information in reports, preferring minimal meaningful information to avoid resource waste.
Key proposals
- Proposal 2 (Sec 2.1): Support data collection content for NW-side model training including L1-RSRPs from Set B beams, L1-RSRPs from Set A beams, beam information for Top K beams among Set A, and timestamps information
- Proposal 4 (Sec 2.2.1): Consider bitmap to indicate reported beams of measurement resource set and one beam index for largest L1-RSRP only when 4<M<size of measurement resource set for NW-sided AI/ML model inference
- Proposal 6 (Sec 2.2.1): Configure max number M (up to 256) of reported beam information in one report based on UE capability signaling, without varying based on other reporting content
- Proposal 8 (Sec 2.2.2): Recommend using larger quantization steps than legacy quantization for L1-RSRP to reduce reporting overhead
- Proposal 10 (Sec 2.2.3): Extend existing TCI direction method to multiple beams with associated timestamp information for future N instances in BM-Case2
- Proposal 13 (Sec 3.2): Consider UE sending request for preferred DL RS configuration and/or DL RS transmission for UE-side model inference data collection
- Proposal 15 (Sec 3.2): Support configuring both Set A and Set B resources as two separate resources for UE-side model inference with FFS on association signaling
- Proposal 16 (Sec 3.2.1): Report beam information for predicted beams using existing CSI reporting framework, with Top-K beams determined based on L1-RSRP and threshold
- Proposal 19 (Sec 3.2.1): For UE-sided BM-Case2, determine earliest time instance with additional offset, each N future time instances consist of at least one slot, with implicit timestamp indication
- Proposal 22 (Sec 4.1): Use L1 signaling to send measurement results to NW for calculation of performance metrics for Type 1 performance monitoring Option 1
- Proposal 24 (Sec 4.1): Support event-triggered UE reporting for UE-sided Type 1 performance monitoring when performance metrics fall below threshold
- Proposal 25 (Sec 4.2): Define new report quantity including 'none' reporting and monitoring decisions (activation/deactivation/fallback) in CSI-ReportConfig for Type 2 performance monitoring