R1-2410519
discussion
Discussion on specification support for AIML-based beam management
From MediaTek
Summary
MediaTek's technical contribution for RAN1 #119 presents 48 proposals addressing specification support for AI/ML-based beam management in NR, covering associated ID consistency, performance monitoring, Set A/B configuration, training data collection, and responses to RAN2 liaison statements.
Position
MediaTek advocates FOR: (1) flexible associated ID configuration that is not one-to-one mapped with inference configurations to reduce overhead, (2) separate CPU counting for AI/ML vs legacy CSI processing, (3) NW-determined Set B to achieve RS overhead savings, (4) support for aperiodic CSI-RS for BM-Case2, and (5) Option2 for UE applicability reporting. They argue AGAINST: (1) configuring associated ID within CSI Resource/ResourceSet configurations due to overhead concerns, (2) using Alt.4 probability information directly as performance metrics due to model variability, and (3) further enhancements to unified TCI state framework for beam indication citing limited benefits and high standardization effort.
Key proposals
- Proposal 1 (Sec 2.1.1): RAN1 does not consider a one-to-one mapping between the associated ID and inference configurations
- Proposal 6 (Sec 2.2.1): For performance monitoring of the UE-sided model, for Type1 Option1, further study how to facilitate NW to correctly monitor the performance of a UE-sided model under the assumption that UE is not always reporting all the predicted best beams according to the model output
- Proposal 16 (Sec 2.3.1): Set A is assumed to be always configured with physical resources for monitoring/data collection and legacy UE support
- Proposal 18 (Sec 2.3.2): For UE-side AI/ML model, for BM-Case 2, at least for inference, at least for Set B, support Aperiodic (AP) CSI-RS type
- Proposal 22 (Sec 2.3.3): For NW-side model inference BM-Case2, observation window length is configured to UE within CSI-framework with explicit or implicit alternatives
- Proposal 24 (Sec 2.4.1): For UE-sided model, to save RS overhead, Set B is determined by NW from RAN1 perspective
- Proposal 30 (Sec 2.5): For AI/ML-based BM, at this stage, there is no further enhancement needed for beam indication based on unified TCI state framework
- Proposal 32 (Sec 2.6.2): For UE-sided model inference, consider the CPUs are separately counted between legacy CSI reporting and AI/ML-based CSI reporting
- Proposal 35 (Sec 2.6.4): For BM-Case1 and BM-Case2 with a UE-side AI/ML model, consider UE to report various number of Top-K beams in one beam report (K determined by UE, K ≤ nrofReportedRS)
- Proposal 36 (Sec 2.7): For the options for applicability for inference for UE-side model, support Option2
- Proposal 39 (Sec 2.8): For RAN2 LS Q3: the associated ID that is being used to ensure the consistency of NW-side additional condition should not be specific to an inference configuration
- Proposal 42 (Sec 2.8): For RAN2 LS Q4-3: NW-side additional condition should be separated from the inference configuration
- Proposal 9 (Sec 2.2.1): Support beam prediction ranking accuracy as one alternative, which compares the ranking sequence of K beams with the best probabilities among the subset of Set A and the best K beams with the best measurement among the subset of Set A
- Proposal 27 (Sec 2.4.1): To indicate whether a report configuration is for UE side model inference or UE side data collection, NW configures different reportQuantity for each purpose
- Proposal 5 (Sec 2.1.3): For associated ID, considering First part is a PLMN-specific ID, different NW vendors may not use the same ID under one PLMN Second part is a cell/gNB-determined ID