R1-2409625
discussion
Discussion on AIML for beam management
From Spreadtrum
Spreadtrum's prior position on
9.1.1
at
RAN1#118bis
· AI-synthesized, paraphrased
verify sources →
Advocates for UE-initiated data collection for UE-side models, supports up to 16 beams reporting per instance, and favors reusing existing CSI frameworks to minimize specification impact while opposing larger quantization steps to maintain accuracy.
Summary
Spreadtrum presents 15 proposals and 6 observations regarding AI/ML for NR Beam Management, focusing on UE-side and NW-side model configurations, inference reporting, and performance monitoring. The document argues for specific signaling frameworks to ensure consistency between training and inference, particularly emphasizing the use of associated IDs and dedicated monitoring configurations.
Position
Spreadtrum supports Option 1 or Option 2 for UE functionality determination, requiring the associated ID to be configured for both training and inference to guarantee consistency. They oppose configuring only Set B resources for UE-side inference and argue that larger quantization steps should not be considered for NW-side model inference. For BM-Case 2, they propose reusing the TCI indication framework and supporting implicit time reporting, while allowing adaptive selection of reported beam counts. They require the associated ID to be configured in CSI-ReportConfig and oppose Alt.4 for performance monitoring, instead supporting dedicated monitoring configurations and Option 1 for training-inference consistency.
Key proposals
- Proposal 1 (Sec Discussion): Supports Option 1 or Option 2 for information provided to the UE to decide applicable functionality, where the NW configures inference-related configurations in Step 3.
- Proposal 2 (Sec Discussion): Requires the associated ID to be configured for both data collection (training) and inference to ensure the UE can determine applicable functionality.
- Proposal 3 (Sec AI/ML model data collection): Opposes configuring only the resource set for Set B for UE-side model inference results reporting, as it limits application scenarios.
- Proposal 4 (Sec NW side model): Supports only CRI/SSBRI of a measurement resource set for reported beam information in NW-side models to minimize standardization complexity.
- Proposal 5 (Sec NW side model): Supports using both L1 and high-layer signaling for data collection for training in NW-side models.
- Proposal 6 (Sec NW side model): Opposes considering larger quantization steps for reported L1-RSRP at least for model inference in BM-Case1 and BM-Case2.
- Proposal 7 (Sec AI/ML model inference): Proposes reusing the TCI indication framework for BM-Case2, allowing beams from multiple time instances to be indicated via multiple beam indications.
- Proposal 8 (Sec NW-side model): States that reporting multiple past time instances in one reporting instance for BM-Case2 is not needed for NW-side models.
- Proposal 9 (Sec UE-side model): Supports implicit reporting of time information for BM-Case2, where the order of reported beams corresponds to predicted time instances.
- Proposal 10 (Sec UE-side model): Proposes that the number of reported beams for BM-Case2 can be adaptively selected by the UE for each N instance.
- Proposal 11 (Sec UE-side model): Requires the associated ID to be configured in CSI-ReportConfig.
- Proposal 12 (Sec UE-side model): States that for UE-side models, ranking information of predicted Top K beams is conveyed by the order of beam information, with the method of obtaining ranking belonging to UE implementation.
- Proposal 13 (Sec AI/ML model monitoring): Opposes Alt.4 (probability information) for Option 2 (UE-assisted performance monitoring) due to reliability concerns and model limitations.
- Proposal 14 (Sec AI/ML model monitoring): Supports configuring resource sets for monitoring in a dedicated report configuration for Type 1 Option 2 of UE-side model monitoring.
- Proposal 15 (Sec AI/ML model monitoring): Proposes considering Option 1 (based on associated ID) to ensure consistency between training and inference for UE-side models.