R1-2500560
discussion
Discussion on specification support for beam management
From Panasonic
Summary
Panasonic discusses specification support for AI/ML-based beam management in Rel-19, focusing on model training, inference, consistency, and performance monitoring for both NW-sided and UE-sided models. The document contains 13 proposals addressing data collection, measurement windows, beam reporting enhancements, TCI state activation, associated ID applicability, and beam accuracy indicators.
Position
Panasonic supports using a single resource set for NW-sided model training data collection via higher layer signaling, arguing that distinguishing sets is unnecessary since the gNB knows its configuration. They propose extending Rel. 17 TCI state activation signaling to handle K predicted beams for N future time instances in BM-Case 2, offering options for combined or separate TCI state sets. For UE-sided models, they prefer determining the reference time for predictions based on an offset from the latest Set B transmission occasion (Option 3) to simplify UE processing. Crucially, they argue for expanding the applicability of 'associated IDs' to multiple cells to reduce model training complexity, while requiring that these IDs be determined within a NW operator to preserve proprietary vendor information. They also propose defining a Beam Accuracy Indicator (BAI) and a specific monitoring window for UE-assisted performance monitoring to ensure consistent accuracy calculations.
Key proposals
- Proposal 1 (NW-side model training): Support L1-RSRPs measured based on one resource set for both Set A and Set B via higher layer signaling, using legacy differential quantization.
- Proposal 2 (NW-sided model inference): Support configuring a measurement window with the measurement resource set to define the duration for aperiodic CSI measurements.
- Proposal 3 (Beam reporting): Enhance group-based beam reporting to support reporting more than 4 beams in one report for NW-sided models.
- Proposal 4 (Beam indication): Extend Rel. 17 TCI state activation signaling to activate TCI states of K predicted beams for N future time instances in BM-Case 2, considering combined or separate sets.
- Proposal 5 (UE-sided model inference): Support configuring a measurement window with the measurement resource set for UE-sided model inference.
- Proposal 6 (Reference time): Determine the reference time of the earliest predicted result instance based on an offset from the latest transmission occasion of CSI-RS/SSB in Set B (Option 3).
- Proposal 7 (Differentiation): Differentiate predicted and measured beams by introducing prediction-related metrics or distinct resource sets in report configurations.
- Proposal 8 (Set Association): Support mapping/association of beams within Set A and Set B based on QCL relationships to improve prediction performance.
- Proposal 9 (Consistency): Apply the concept of 'associated IDs' across multiple cells to ensure consistency of NW-side additional conditions for UE-sided models.
- Proposal 10 (Proprietary Info): Determine 'associated IDs' within a NW operator (MNO) to preserve proprietary vendor-specific information while enabling multi-cell applicability.
- Proposal 11 (Performance Monitoring): Define a Beam Accuracy Indicator (BAI) as a metric for beam prediction accuracy in UE-assisted performance monitoring.
- Proposal 12 (Monitoring Window): Define a monitoring window for BM-Case 1 and BM-Case 2 to calculate BAI, linking inference results and monitoring instances in the time domain.
- Proposal 13 (NW Monitoring): Enhance group-based beam reporting to support performance monitoring for NW-sided models by allowing reports of more than 4 beams.