R1-2409877
discussion
Discussion on AI/ML for beam management
From Xiaomi
Summary
This Xiaomi contribution discusses specification impacts of AI/ML for beam management in NR, covering both spatial (BM Case 1) and temporal (BM Case 2) beam prediction for UE-side and NW-side models. The document contains 47 proposals and 1 observation across functionality identification, data collection, signaling mechanisms, and performance monitoring.
Position
Xiaomi advocates FOR comprehensive AI/ML beam management support with both spatial and temporal prediction capabilities, flexible functionality identification based on beam set relationships, optional NW-side additional conditions, and dual benchmark approaches for performance monitoring. They push FOR practical implementation aspects like separate reporting of legacy and AI-based CSI reports, cell group specific associated IDs, and support for both explicit and implicit beam set associations. They are AGAINST overly complex mandatory additional conditions that could increase signaling overhead.
Key proposals
- Proposal 2-1 (Sec 2.1): BM Case 1 and BM Case 2 can be considered as different conditions for different functionalities
- Proposal 2-8 (Sec 2.2): Answer to Question 4-1: The NW-side additional condition is optional
- Proposal 2-13 (Sec 2.3): Confirm the WA and support to introduce associated ID within CSI framework per CSI-reportconfig, per Resourceconfig or per resource set
- Proposal 3-1 (Sec 3.1): For data collection of NW-side AI/ML model training, support to define a time window or a data size for each report with more than one data sample
- Proposal 3-7 (Sec 3.2): Both explicit and implicit association between set B and set A can be supported for data collection for UE-side AI/ML model training
- Proposal 4-1 (Sec 4.1): Support to report the predicted L1-RSRP if the beam is not configured for corresponding measurement, and report the measured L1-RSRP if the beam is configured for corresponding measurement for UE-side model inference
- Proposal 4-5 (Sec 4.2): For UE-sided model inference, support to configure one resource set with resources in more than one measurement time instance for configuration of set B in BM Case 2
- Proposal 5-1 (Sec 5.1): Both of the following two Benchmark/reference for performance comparison should be supported: Alt.1: The best beam(s) obtained by measuring beams of a set indicated by gNB and Alt.4: Measurements of the predicted best beam(s) corresponding to model output
- Proposal 5-4 (Sec 5.2): For Type 1 performance monitoring of UE-side AI/ML model, both NW-side initiated and UE-side initiated performance monitoring can be supported
- Proposal 5-10 (Sec 5.3): For performance monitoring of network-side AI/ML model, support to report measurement results of set B and set A separately
- Proposal 5-12 (Sec 5.4): Confirm the necessity of assessment/monitoring of inactive models/functionalities
- Proposal 2-5 (Sec 2.1): Define different ranges of number of beams in set B and/or set A as different conditions for different functionalities
- Proposal 4-10 (Sec 4.2): Support following two TCI state indication mechanism for TCI state indication of more than one predicted time instance
- Proposal 5-2 (Sec 5.1): In addition to beam prediction accuracy, Support the L1-RSRP difference evaluated by comparing measured RSRP and predicted RSRP performance metrics for performance monitoring
- Proposal 3-4 (Sec 3.1): Exchange the UE-side additional condition such as Rx beam assumption and UE speed during the procedure of data collection for NW-side AI/ML model training