R1-2500642
discussion
Discussion on Specification Support for Beam Management
From Sony
Summary
Sony presents 18 proposals regarding AI/ML-based beam management for NR, focusing on model inference timing, reporting content, and model monitoring mechanisms. The document supports specific options for reference time configuration in BM-Case 2, advocates for probability-based reporting of Top K beams, and defines event-triggered monitoring and failure detection criteria.
Position
Sony supports Option 3 for the reference time of BM-Case 2 predictions, anchoring it to the latest CSI-RS/SSB transmission occasion in Set B no later than the CSI reference resource to enhance synchronization in fast-fading channels. They propose configuring the duration D_f via RRC for fixed values and L1 signaling for dynamic adjustment. For inference reporting, Sony strongly advocates including probability-related information for the predicted Top K beams to assist network selection. Regarding monitoring, they support Alt 2 and Alt 3 for UE-assisted performance metrics, utilizing L1-RSRP differences between predicted and measured values. Sony further proposes defining specific events to trigger monitoring reports, such as significant variance in multiple prediction results for BM-Case 2, and suggests a model failure detection mechanism based on repeated metric calculations over a time window to avoid ping-pong effects.
Key proposals
- Proposal 1 (AI/ML model inference): Supports Option 3 for BM-Case 2 reference time, based on the latest transmission occasion of CSI-RS/SSB in Set B no later than the CSI reference resource.
- Proposal 2 (AI/ML model inference): Proposes configuring duration D_f via RRC for fixed values and L1 signaling for dynamic adjustment in BM-Case 2.
- Proposal 3 (AI/ML model inference): Supports reporting beam information for predicted Top K beams along with probability-related information.
- Proposal 4 (Model monitoring): Supports configuring a subset from Set A for monitoring and investigating mapping rules between the subset and full set.
- Proposal 5 (Model monitoring): Supports further investigation into scenarios where the monitoring subset does not include Top K predicted beams.
- Proposal 6 (Model monitoring): Supports Alt 2 (L1-RSRP difference based on actual measurement) and Alt 3 (RSRP difference between predicted and measured) for UE-assisted performance monitoring.
- Proposal 7 (Model monitoring): Supports defining events to trigger reporting when measured RSRP differs from predicted RSRP (BM-Case 1) or when prediction results vary significantly (BM-Case 2).
- Proposal 8 (Model monitoring): Calls for further study to reduce measurement overhead in BM-Case 2 by reusing input data from previous inference instances.
- Proposal 9 (Model monitoring): Proposes determining model failure based on performance metrics calculated multiple times over a specific time period, similar to beam failure detection.
- Proposal 10 (Conclusions): Reiterates support for Option 3 reference time configuration for BM-Case 2.
- Proposal 11 (Conclusions): Reiterates support for periodic and dynamic configuration of duration D_f.
- Proposal 12 (Conclusions): Reiterates support for reporting Top K beam information with probability data.
- Proposal 13 (Conclusions): Reiterates support for subset configuration in monitoring.
- Proposal 14 (Conclusions): Reiterates support for investigating subsets excluding Top K beams.
- Proposal 15 (Conclusions): Reiterates support for Alt 2 and Alt 3 in UE-assisted performance monitoring.