R1-2410347
discussion
AI/ML for Beam Management
From Meta
Summary
Meta presents a comprehensive technical document on AI/ML for beam management in 5G NR, covering configuration aspects, performance monitoring, and beam indication enhancements. The document contains 14 formal proposals addressing both UE-sided and network-sided AI/ML models for beam management use cases.
Position
Meta advocates for flexible AI/ML beam management implementation with simplified configuration approaches (avoiding explicit CSI framework requirements for set A), comprehensive performance monitoring including both Type 1 and Type 2 options with event-driven mechanisms, and enhanced TCI framework extensions for predicted beam indication. They push for UE-assisted monitoring capabilities and fallback mechanisms while supporting both spatial and temporal beam prediction use cases.
Key proposals
- Proposal 1 (Sec 2.1.1): For configuration of set A for UE sided models, support that set A is not explicitly configured using CSIReportConfig
- Proposal 2 (Sec 2.1.2): The associated ID should imply that the spatial filters used of the transmission of the DL reference signals for measurement during data collection for model training are identically mapped to the reference signals for measurement/reporting during inference phase
- Proposal 3 (Sec 2.2.1): For BM Case 2 measurement report, multiple time instances with a configured number of beams per time instance should be supported and the absolute value of time instances can be derived by NW based on periodicity of measurement RS transmissions
- Proposal 4 (Sec 2.2.1): For training data collection for NW sided model, support L1-RSRP reporting for a configured set of beams using higher layers
- Proposal 5 (Sec 2.3.1.1): For UE side model with monitoring Type-1, Option-1, measurement report with L1-RSRP and RS index is sufficient and other contents may not be needed
- Proposal 6 (Sec 2.3.1.1): For UE sided model with monitoring Type-1, Option-2, Alt-2/3 (RSRP difference reporting) should be supported as configurable reporting contents but confidence information and/or probability information may be optional based on UE model capabilities
- Proposal 7 (Sec 2.3.1.1): For UE sided model with monitoring Type-1, Option-2, consider supporting UE report of fallback to non-AI/ML methods
- Proposal 8 (Sec 2.3.1.2): Support Type 2 performance monitoring of UE sided models with reporting for AI/ML model switching/activation and indication of non-AI/ML fallback
- Proposal 9 (Sec 2.3.1.3): For UE sided model with Type-1 performance monitoring, support event triggered monitoring where the gNB configures one or more events to the UE
- Proposal 10 (Sec 2.3.1.3): For UE sided model with Type-1 performance monitoring, support at least Events 1,2 and 3
- Proposal 11 (Sec 2.3.1.3): For UE sided model with Type-2 performance monitoring, support event driven indication of fallback or model switching/activation
- Proposal 12 (Sec 2.3.2): Consider UE assisted performance monitoring for NW sided models
- Proposal 13 (Sec 2.4): Consider enhancements to the unified TCI framework for indication of predicted beams which are not part of activated TCI states
- Proposal 14 (Sec 2.1.2): For reporting RSRP difference information, the baseline can be for the current indicated beam where UE can measure and predict the RSRP