R1-2500050
discussion
Discussion on specification support for AI/ML-based beam management
From FUTUREWEI
FUTUREWEI's prior position on
9.1.1
at
RAN1#119
· AI-synthesized, paraphrased
verify sources →
Advocates for maximizing reuse of existing CSI framework to minimize specification effort. Supports simpler approaches that avoid complex new metrics like probability and confidence information. Opposes supporting multiple alternatives that would increase complexity, specifically opposing Opt 3/4 for inference reporting and other performance monitoring alternatives beyond Alt 1. Favors practical solutions like using RS ID as implicit beam ID rather than defining new beam identifiers.
Summary
Futurewei presents 8 proposals for Rel-19 AI/ML-based beam management, focusing on specification support for performance monitoring, model inference, data collection, and RRC parameter configuration. The document argues for reusing the existing CSI framework to minimize specification effort and overhead, specifically recommending against new metrics for UE-sided models and defining specific reporting contents for network-sided models.
Position
Futurewei proposes reusing the existing CSI framework extensively to reduce specification effort for Rel-19 AI/ML-based beam management. They oppose supporting Opt 3 and Opt 4 for UE-sided model inference reports, arguing that defining probability and confidence metrics is difficult and offers unclear benefits. They require the maximum number of reported beams (M) for network-sided models to be increased from 4 to 8 as a starting point. They support using RS ID as an implicit beam ID indicator and prefer Option B for RSRP reporting, which prioritizes measured L1-RSRP over predicted values when measurements are available. They confirm the working assumption that the associated ID can be configured within the CSI framework to ensure consistency across training and inference.
Key proposals
- Proposal 1 (Performance Monitoring): For UE-assisted performance monitoring (Option 2), support only Alt 1 (Top 1 or Top K beam prediction accuracy) and configure the full Set A for monitoring with a longer period than Set B to reduce overhead.
- Proposal 2 (Network-sided Model Inference): For L1 beam reports, support reporting L1-RSRPs and beam info (CRI/SSBRI) for up to M beams within X dB gap to the largest value, increasing the maximum M from 4 to 8.
- Proposal 3 (UE-sided Model Inference): Do NOT support Opt 3 (probability information) and Opt 4 (confidence information of RSRP) in inference result reports for BM-Case 1, citing difficulty in defining and testing these new metrics.
- Proposal 4 (UE-sided Model Inference): Support Option B for RSRP reporting: report predicted RSRP if the beam is not configured for measurement, but report measured L1-RSRP if the beam is configured for corresponding measurement.
- Proposal 5 (Data Collection): Support a mechanism where the network signals multiple possible DL RS transmission configurations, and the UE reports its supported/preferred ones, rather than open-ended configuration.
- Proposal 6 (Data Collection): Use RS ID as an implicit indication of beam ID and reuse legacy L1-RSRP reporting as much as possible, avoiding new TX beam ID definitions.
- Proposal 7 (Assistance Information): Confirm the working assumption that the associated ID for consistency across training and inference can be configured within the CSI framework.
- Proposal 8 (RRC Parameters): Support specific RRC parameters including separate CSI-ResourceConfigIds for Set A and Set B, configuration of M for NW-sided models, and linking inference and monitoring report configurations via IDs.