R1-2500555
discussion
Discussion on AIML beam management
From TCL
Summary
TCL proposes integrating AI/ML into NR beam management to simplify the conventional P1/P2/P3 processes into two phases and unify Beam Failure Detection (BFD) and Recovery (BFR) procedures. The document contains 11 proposals and 6 observations focused on reducing overhead through merged Top-K beam sweeping and model monitoring, enhancing TCI framework support for AI/ML, and optimizing report quantization and structure.
Position
TCL proposes simplifying the conventional P1/P2/P3 beam management processes into two phases via AI/ML beam prediction. They argue for merging Top-K beam sweeping measurements with model performance monitoring to reduce signaling overhead, specifically suggesting that Top-K results trigger or calculate monitoring metrics. For Beam Failure Detection and Recovery, TCL proposes studying a unified framework where AI/ML predicts both BF events and candidate beams, potentially replacing legacy BFD with predictive models. Regarding configuration, they propose enhancing the TCI framework with dedicated AI/ML TCI state IDs and new QCL types to support Set A/Set B beam mapping. Finally, they propose specific report enhancements, including unequal quantization step-sizes for RSRP, post-processed RSRP inclusion, and two-stage reporting via PUCCH and PUSCH to handle BM-Case2 overhead.
Key proposals
- Proposal 1 (Sec 2.1): Simplify conventional P1/P2/P3 beam pairing/refinement to two phases using AI/ML beam prediction.
- Proposal 2 (Sec 2.1): Design measurement patterns for spatial domain beam prediction considering both fixed regular and random patterns.
- Proposal 3 (Sec 2.2): Reuse measurement results from Top-K beam sweeping for performance monitoring to reduce overhead.
- Proposal 4 (Sec 3.1): Study using AI/ML based beam prediction to maintain the candidate beam list for Beam Failure Recovery (BFR).
- Proposal 5 (Sec 3.2): Study integrating RS and measurement configuration of BFD and BFR under a unified AI/ML based framework.
- Proposal 6 (Sec 3.3): Study evolving BFD to BF event prediction while using temporal beam prediction to find candidate beams for BFR.
- Proposal 7 (Sec 4.1): Enhance the TCI framework by introducing additional TCI state IDs dedicated for AI/ML BP and new QCL types.
- Proposal 8 (Sec 4.2): Enhance beam management reports by indicating model type and/or using bitmaps to select report quantities.
- Proposal 9 (Sec 4.2): Include L1-RSRP and/or post-processed RSRP (e.g., filtered RSRP) in AI/ML BP reports.
- Proposal 10 (Sec 4.2): Study quantization enhancements for RSRP and a two-stage report mechanism using both PUCCH and PUSCH.
- Proposal 11 (Sec 4.2): Apply grouping or segmentation-based approaches for BM-Case2 reports to reduce overhead, such as reference beam plus bitmap.