R1-2500404
discussion
Specification support for beam management
Summary
Tejas Networks Limited presents 39 proposals and 6 observations addressing AI/ML beam management for NR Air Interface, focusing on consistency mechanisms for UE-sided models, performance monitoring metrics, and reporting configurations for both UE and NW-sided models. The document proposes specific signaling methods for Associated IDs, weighted Beam Accuracy Indicators (BAI), and differential L1-RSRP reporting to optimize overhead and accuracy.
Position
Tejas Networks proposes that the Associated ID for UE-sided models be configured within the CSI-Report Config to ensure consistency between training and inference phases for Set A and Set B resources. They present a technical case for a weighted Beam Accuracy Indicator (BAI) calculation that assigns weights to predicted beams based on their presence in the Top-M measured beams, enhancing monitoring precision. For NW-sided models, they support reporting Differential L1-RSRP with a larger quantization step size and a maximum beam count (M) of 256 to optimize reporting overhead. They propose dynamic control over Set A and Set B measurements and suggest configuring monitoring resources within existing CSI report configurations to reduce signaling complexity. Additionally, they support extending Rel-17 TCI state activation for BM-Case 2 with a configurable number of joint TCI states (N=2 to 4) aligned with UE mobility.
Key proposals
- Proposal 1 (Consistency for additional condition for UE-sided model): Associated ID should be configured as part of the CSI-Report Config and mapped to resources in Set A (predicted beams) and Set B (measured beams).
- Proposal 5 (Consistency for additional condition for UE-sided model): DL Tx beams associated with the same Associated ID should be consistent in transmission properties, such as spatial domain characteristics like beam shape.
- Proposal 6 (Performance monitoring): For Type 1 performance monitoring Option 1 (NW-side), L1 signaling can be used to send measurement results to the NW for metric calculation, without precluding higher layer signaling.
- Proposal 9 (Performance monitoring): Proposes a weighted Beam Accuracy Indicator (BAI) calculation (Option 3c) where weights are assigned to predicted beams based on their accuracy relative to Top-M measured beams.
- Proposal 11 (Performance monitoring): Suggests triggering reports based on Event 1, defined as predicted beam accuracy falling below a threshold accuracy with statistical results in a given window.
- Proposal 13 (Configuration of RS for Set A and Set B for NW sided model): The gNB may require dynamic control over Set A and Set B measurements to minimize overhead while maintaining beam prediction accuracy.
- Proposal 16 (Configuration of RS for Set A and Set B for UE-sided model): For UE-sided model at least for BM Case-1, support configuring only one resource set (Set B) for CSI report generation and inference results reporting.
- Proposal 18 (Measurement report for NW-sided model): For NW-sided model BM-Case 2, support reporting Differential L1-RSRP to reduce overhead by communicating relative changes rather than absolute values.
- Proposal 22 (Inference report at NW sided Model): Considers M = 256 as the starting point for the maximum number of beams reported, as agreed in the study item for NW-sided model inference.
- Proposal 24 (Inference report at NW sided Model): Supports introducing a larger quantization step size for differential L1-RSRP reporting to reduce beam report overhead.
- Proposal 28 (Inference result report for UE-sided model): For UE-sided model BM-Case 1, ranking information of predicted Top K beams is conveyed through the order of reported information, with the first beam having the highest predicted RSRP or probability.
- Proposal 34 (Performance monitoring): For UE-assisted performance monitoring, configure the resource set(s) for monitoring within the existing CSI report configuration used for inference to minimize signaling overhead.
- Proposal 37 (Beam indication): For BM-Case 2, extend Rel-17 TCI state activation methods with a maximum number (N) of joint TCI states, suggesting initial values between 2 and 4.