R1-2407796
discussion
Discussion on AI/ML-based beam management
From ZTE
Summary
ZTE presents a comprehensive technical document on AI/ML-based beam management for 5G NR with approximately 35 proposals and 3 observations covering functionality-based lifecycle management, data collection enhancements, model inference mechanisms, performance monitoring, and network-side additional conditions consistency. The document advocates for bitmap-based beam reporting and flexible resource configuration to reduce signaling overhead while maintaining beam prediction accuracy.
Position
ZTE strongly advocates FOR functionality-based lifecycle management over model-ID-based approaches, arguing it reduces complexity and leverages existing UE capability frameworks. They push FOR bitmap-based beam reporting methods to significantly reduce signaling overhead (up to 63.5% reduction) and support flexible quantization with larger step sizes (4 dB vs 2 dB) for differential RSRP reporting. ZTE is AGAINST overly complex associated ID schemes due to signaling overhead and proprietary information disclosure risks, instead favoring performance monitoring-based approaches for consistency.
Key proposals
- Proposal 1 (Sec 1): Functionality-based LCM without specifying any model ID based signaling should be utilized for AI/ML based beam management
- Proposal 2 (Sec 2.1.1): For the contents of collected data from UE to NW, all three options can be supported to serve different LCM operations at the NW side - M1 L1-RSRPs with beam indication, M2 L1-RSRPs based on measurement, and M3 beam indices based on measurement, where M1/M2/M3 can be larger than 4
- Proposal 4 (Sec 2.1.2): One or two RS resource sets for beam measurement can be configured to the UE for training data collection of NW-side AI/ML model, which are dependent on the beam set construction of Set A and Set B
- Proposal 8 (Sec 2.1.3.2): Support to specify the bitmap based method for the reporting of beam information given the fact that it can greatly reduce the reporting overhead by 8.3%, 37.5%, 53.8%, and 63.5% when half of measurements are reported and the number of beams is 8, 16, 32, and 64 respectively
- Proposal 11 (Sec 2.1.4): At least support L1 signaling for NW-side data collection irrespective of the purpose of data collection (e.g., model training, model inference, and performance monitoring)
- Proposal 14 (Sec 3.1.2): Support enhancements to report information about measurements of multiple past time instances in one reporting instance for BM-Case2, including indication of timestamp information, reference beam, and common beam information
- Proposal 16 (Sec 3.2): For UE-sided model at least for BM Case-1, CSI-ReportConfig is used for configuration of inference results reporting with two alternatives - one CSI-ResourceConfigId for Set B or two CSI-ResourceConfigIds for Set A and Set B separately
- Proposal 20 (Sec 3.4.1): For UE-sided model, support reporting of RSRP in Option 2 and probability information in Option 3 for beam selection at the NW side, where predicted Top-K beams are the K beams with the highest predicted RSRP or probability information
- Proposal 25 (Sec 3.6): For BM-Case2 (both UE-sided and NW-sided model), support to extend the Rel-17 TCI state activation/indication signalling methods to activate/indicate N TCI states corresponding to N future time instances
- Proposal 27 (Sec 4.1): Support beam prediction accuracy related KPIs (Alt 1 and Alt 2) as the primary performance metric for performance monitoring of UE-sided AI/ML model
- Proposal 29 (Sec 4.2.2): For Option 2 (UE-assisted performance monitoring), further consider monitoring report initiated by NW and monitoring report initiated by UE, where report content can be a calculated performance metric or threshold indication
- Proposal 32 (Sec 4.3): Model/functionality failure detection should be based on monitoring results of several consecutive times within a predefined monitoring window
- Proposal 34 (Sec 5): Confirm the working assumption that the associated ID at least can be configured within CSI framework
- Proposal 36 (Sec 5): For UE sided model in beam management, UE may assume the similar properties of a DL Tx beam set/list associated with the same associated ID
- Proposal 9 (Sec 2.1.3.1): At least for NW sided model, support differential L1-RSRP reporting with larger quantization step size applicable to the differential L1-RSRP