R1-2500545
discussion
AI/ML based Beam Management
From Google
Google's prior position on
9.1.1
at
RAN1#119
· AI-synthesized, paraphrased
verify sources →
Advocates for comprehensive AI/ML beam management support including both network-side and UE-side models. Supports flexible reference signal configurations, including AP-CSI-RS and SSB, and enhanced reporting mechanisms with confidence indicators. Proposes separate TCI state pools for AI/ML predictions. Supports practical implementation considerations like CPU resource management and performance monitoring relaxation. Strongly supports Option 1 for inference configuration management to avoid multi-stage RRC configuration latency issues.
Summary
Google presents 26 proposals for ML-based Beam Management in NR, covering beam measurement, reporting, indication, failure recovery, and performance monitoring. The document addresses open issues for both NW-side and UE-side models, specifically focusing on Set A/B reference signal configurations, TCI state handling for predicted beams, and CPU resource allocation for inference.
Position
Google proposes supporting AP-CSI-RS and SSB (including Rel-19 on-demand SSB) as Set B reference signals for both BM-Case 1 and BM-Case 2 to align with current deployments. They require an inference report for BM-Case 1 to consume 1 eCPU and 1 legacy CPU, and propose defining predicted RSRP based on a reference transmission power. For TCI activation, Google supports configuring separate TCI state pools for legacy and predicted beams, and triggering aperiodic CSI-RS to reduce latency for unknown TCI states. They propose reusing the SCell BFR mechanism (SR+MAC CE) for event-triggered performance monitoring reports and support performance monitoring relaxation for low-mobility UEs. Finally, they propose configuring the associated ID per CSI report configuration to ensure consistency between training and inference.
Key proposals
- Proposal 1 (Beam measurement): Support configuring AP-CSI-RS as RS for Set B for BM-Case 2.
- Proposal 2 (Beam measurement): Support configuring SSB (legacy and on-demand) as RS for Set B for both BM-Case 1 and BM-Case 2.
- Proposal 3 (Beam report NW-side): Support reporting a confidence level indicator for L1-RSRP to indicate maximum measurement error.
- Proposal 5 (Beam report NW-side): Support L1-RSRP report retransmission to facilitate NW-side beam prediction for BM-Case 2.
- Proposal 6 (Beam report UE-side): Define predicted RSRP based on reference transmission power and support probability/confidence information reports.
- Proposal 7 (Beam report UE-side): An inference report for BM-Case 1 should consume 1 eCPU and 1 legacy CPU.
- Proposal 9 (TCI activation BM-Case 1): Support triggering aperiodic CSI-RS resources QCLed with the TCI state source to reduce latency for unknown TCI states.
- Proposal 10 (TCI activation BM-Case 2): Configure separate TCI state pools for legacy beam indication and beam prediction.
- Proposal 13 (TCI activation BM-Case 2): Support UE feedback before beam action time for performance validation of predicted beams.
- Proposal 15 (BFR UE-side): For BFR based on UE-side model, report N predicted beam indices via BFRQ if no candidate beam is identified.
- Proposal 18 (Set A RS): Support configuring aperiodic CSI-RS as Set A RS triggered by group-cast DCI.
- Proposal 20 (Performance monitoring): Reuse SCell BFR mechanism (SR+MAC CE) for event-triggered reports of incorrect beam prediction counters.
- Proposal 22 (Performance monitoring): Support performance monitoring relaxation for UE in low mobility and high beam quality states.
- Proposal 24 (Consistency): Support configuring the associated ID per CSI report configuration.