R1-2500254
discussion
Discussion on AI/ML for beam management
From China Telecom
China Telecom's prior position on
9.1.1
at
RAN1#119
· AI-synthesized, paraphrased
verify sources →
Advocates for comprehensive support of multiple data collection options for flexibility across different Life Cycle Management purposes, including training, monitoring, and inference. Proposes beam prediction accuracy KPIs as primary performance monitoring metrics and full Set A measurement for accurate UE-assisted performance monitoring. Opposes using only single resource set configurations for Set B in UE-sided models. Deprioritizes link quality metrics for AI/ML model performance monitoring, arguing these do not directly reflect prediction accuracy.
Summary
China Telecom submits 16 proposals addressing AI/ML beam management for NR, covering data collection, model inference, performance monitoring, and consistency mechanisms for both network-side and UE-side models. The document focuses on reducing signaling overhead for NW-sided models, defining specific reporting criteria for UE-sided inference results, and establishing metrics for performance monitoring and fallback procedures.
Position
China Telecom supports collecting L1-RSRP, beam indices, and timestamps for NW-sided models, while proposing payload reduction via bitmap indications and larger quantization steps for differential RSRP. For UE-sided models, they oppose configuring only one resource set for Set B in BM-Case 1 and require that inference reports include Top K beams with predicted RSRP above a certain probability threshold. They support Option 3 for BM-Case 2 reference time, linking it to the latest transmission occasion of CSI-RS/SSB in Set B. For performance monitoring, they prefer Beam prediction accuracy KPIs and support configuring the full Set A for UE-assisted monitoring (Alt 1) or using L1-RSRP differences for Top K predicted beams (Alt 2). Finally, they confirm the working assumption that associated IDs ensure consistency within a cell and propose configuring associated IDs per CSI report configuration.
Key proposals
- Proposal 1 (Data Collection - NW-sided): Supports collecting L1-RSRP, beam indices, and timestamp for NW-sided AI/ML model data collection.
- Proposal 2 (Data Collection - NW-sided): Proposes enhancements for payload size reduction, such as bitmap indication of beam indices and differential RSRP with larger quantization steps.
- Proposal 3 (Data Collection - UE-sided): States that for UE-sided models, data collection can be triggered by the UE or initiated by the network.
- Proposal 4 (Data Collection - UE-sided): Highlights the need to discuss how to indicate the purpose of measurement configuration for UE-sided models.
- Proposal 5 (Model Inference - NW-sided): Proposes that for NW-sided model inference, report content can be L1-RSRP or L1-RSRP and beam ID.
- Proposal 6 (Model Inference - NW-sided): Proposes that the max number of reported beam-related information in one report can be configured by the network based on UE capability.
- Proposal 7 (Model Inference - NW-sided): Supports reporting timestamp information for measurements on Set B for BM-Case 2.
- Proposal 8 (Model Inference - UE-sided): Opposes configuring only one resource set for Set B for UE-sided model inference results report (at least for BM-Case 1).
- Proposal 9 (Model Inference - UE-sided): Proposes that beam information in inference result reports should be CRI/SSBRI of Top K beams with highest predicted RSRP above a certain probability.
- Proposal 10 (Model Inference - UE-sided): Supports Option 3 for reference time of earliest prediction instance in BM-Case 2 (based on latest transmission occasion of CSI-RS/SSB in Set B).
- Proposal 11 (Performance Monitoring): Proposes supporting Beam prediction accuracy related KPIs as the metric for performance monitoring.
- Proposal 12 (Performance Monitoring): Proposes that for UE-assisted performance monitoring (Option 2 Alt 1), the resource set for measurement is configured as the full Set A.
- Proposal 13 (Performance Monitoring): Supports Alt 2 for UE-assisted performance monitoring, using L1-RSRP difference information based on actual measurement of Top K predicted beams.
- Proposal 14 (Performance Monitoring): Proposes performing model switch or fallback to non-AI/ML method if the number of AI/ML model failures reaches a predefined threshold.
- Proposal 16 (Consistency): Proposes that associated ID for UE-sided model is configured per CSI report configuration.