R1-2409960
discussion
Discussion on specification support for AI/ML beam management
From Transsion Holdings
Summary
Transsion Holdings presents 13 proposals for AI/ML beam management enhancements in 5G NR, covering data collection, model inference, and performance monitoring for both network-side and UE-side AI/ML models. The document addresses specification impacts across training data collection, inference procedures, and monitoring mechanisms.
Position
Transsion advocates for UE-centric approaches including UE-initiated data collection rather than network-controlled collection, supports separate resource set configurations (Set A and Set B) for better flexibility, and pushes for reduced signaling overhead through larger quantization steps and higher-layer signaling for training data. They favor Option 3 approaches that give UEs more control over reporting applicable inference parameters.
Key proposals
- Proposal 1 (Sec 2.1.1): Support Option 2 for NW-sided model training data - L1-RSRPs from Set A and beam information of Top K from separate Set B resource set
- Proposal 2 (Sec 2.1.1): Support higher-layer signaling for training data collection reporting instead of L1 signaling due to large payload requirements
- Proposal 3 (Sec 2.1.2): Support Option 2 for UE-side model data collection - request from UE for data collection rather than NW-initiated
- Proposal 4 (Sec 2.2.1): Support larger quantization steps than legacy for differential L1-RSRP and/or absolute L1-RSRP to reduce reporting overhead
- Proposal 5 (Sec 2.2.2): For BM Case-2, time instance information of largest RSRP can be reported explicitly over multiple time instances
- Proposal 6 (Sec 2.2.2): Support Option 1 (2nd priority) and Option 3 (1st priority) for reference time determination in BM Case-2
- Proposal 7 (Sec 2.2.2): Support triggering Aperiodic CSI-RS in multiple time instances for Set B in BM-Case 2
- Proposal 8 (Sec 2.3.1): Configuration and reporting of Set A/B measurement results need specification for NW-side model performance monitoring
- Proposal 9 (Sec 2.3.2): Support configuring a window for metric calculation with alternatives based on uplink slot or CSI reference resource
- Proposal 10 (Sec 2.3.2): Support using full Set A for measurement in UE-sided AI/ML model monitoring
- Proposal 11 (Sec 2.3.2): Three alternatives for reporting performance metrics - quantization, statistical values, or specific events
- Proposal 12 (Sec 3): Confirm working assumption that associated ID can be configured within CSI framework for NW-side condition consistency
- Proposal 13 (Sec 4): Support Option 3 for Step 3-5 inference procedures for UE-side model