R1-2410736
discussion
FL summary #3 for AI/ML in beam management
From Samsung
Summary
This document is Samsung's FL summary #3 for AI/ML in beam management from RAN1 #119, containing over 40 proposals and observations covering UE-side and NW-side model configurations, performance monitoring, data collection, inference reporting, and beam indication frameworks.
Position
Samsung, as the document moderator, advocates for: (1) merging applicability options to give RAN2 flexibility in container design while supporting both CSI-ReportConfig and inference parameter approaches, (2) comprehensive performance monitoring with beam accuracy indicators and dedicated monitoring configurations, (3) flexible reference time definition based on actual measurement occasions rather than reporting slots, (4) supporting multiple resource sets for temporal beam prediction, and (5) reusing existing CSI framework where possible while allowing necessary enhancements. Samsung pushes against overly restrictive designs that limit implementation flexibility.
Key proposals
- Proposal A (Sec 9): Direction for merging Option 1 and Option 2 for applicability - UE can report applicability via RRCReconfigurationComplete or UAI, with CSI-ReportConfig configuration in Step 3 and optional Step 5
- Proposal 2.1-1a (Sec 2.1): Introduce beam accuracy indicator (BAI) for beam prediction accuracy in UE-assisted performance monitoring, with BAI = Np/N where Np is correct predictions out of N total instances
- Proposal 2.2-1a→b (Sec 2.2): Reuse CSI framework for monitoring configuration - dedicated resource sets and report configuration linked via CSI-ReportConfigID to inference configuration
- Proposal 3.1a (Sec 3.1): Reference time for BM-Case2 based on latest transmission occasion of CSI-RS/SSB resource in Set B for measurement, no later than CSI reference resource
- Proposal 3.2a (Sec 3.2): Support multiple RS sets for Set B in different time instances for BM-Case2, each resource set associated with one time instance
- Proposal 4.1a (Sec 4.1): UE-sided model inference result reporting should include beam information and optionally RSRP of predicted Top K beams from Set A
- Proposal 5.2 (Sec 5.2): For data collection for both NW-sided/UE-sided BM model training, at least L1-RSRPs and/or beam-IDs need to be collected by UE
- Proposal 6.2A (Sec 6.2): Support three types of content for NW-sided model training data collection - all L1-RSRPs, L1-RSRPs plus beam info, or L1-RSRPs plus beam info and RSRPs
- Proposal 5.5a (Sec 5.4): Support Y dB quantization steps (Y=3 and/or 4) larger than legacy for differential L1-RSRP reporting to reduce overhead
- Proposal 7.1 (Sec 7.1): Study extending Rel-17 TCI state activation/indication methods to activate/indicate N TCI states for N future time instances in BM-Case2
- Proposal 8.1 (Sec 8.1): Existing CPU mechanism used as starting point for AI/ML-based CSI processing, with FFS on sharing between legacy and AI/ML reporting
- Proposal B (Sec 9): Support dedicated CSI report configuration for monitoring linked to inference configuration via CSI-ReportConfigID
- Proposal C (Sec 9): Send LS to RAN2 on three content types for NW-sided model training data collection via higher layer signaling
- Proposal D (Sec 9): Associated ID can be configured within CSI framework in CSI-ReportConfig with at least one ID for both Set A and Set B
- Proposal E (Sec 9): For UE-sided model BM-Case1, ranking information of predicted Top K beams conveyed by order of beam information based on model output