R1-2410354
discussion
AI/ML for beam management
From Ericsson
Summary
Ericsson presents a comprehensive technical document on AI/ML for beam management in NR air interface, containing 20 proposals and 3 observations. The document addresses both UE-sided and NW-sided AI/ML models for beam management enhancement, covering data collection, inference reporting, performance monitoring, and overhead reduction mechanisms.
Position
Ericsson advocates for a comprehensive AI/ML framework that leverages existing CSI mechanisms with minimal specification impact while maximizing functionality. They strongly push FOR: (1) expanding aperiodic CSI-RS resources from 16 to 64 beams to enable practical AI/ML deployment, (2) uncertainty quantification in UE predictions to enable trustworthy AI/ML, (3) adaptive Top-K beam selection based on model confidence, and (4) extensive overhead reduction mechanisms for NW-sided models. They push AGAINST complex new signaling frameworks, favoring reuse of existing CSI infrastructure with targeted enhancements.
Key proposals
- Proposal 1 (Sec 2): For UE-sided or NW-sided data collection, at least support UE measurement of a resourceSet with 64 NZP CSI-RS resources in the aperiodic reporting configuration
- Proposal 2 (Sec 3.1): For UE-sided model, regarding configuration of set A/B, support alternative 3 (two CSI-ResourceConfigIds configured for Set A and Set B separately)
- Proposal 3 (Sec 3.5): For UE-sided model inference, enable NW to specify set A beam subset restriction similar to codebook subset restriction (CBSR) that is specified for CSI feedback
- Proposal 4 (Sec 3.3): For UE-sided model for BM-Case 2, for inference results report, support 'Option 1: Based on the uplink slot for the report' with slot n+δ reference time configuration
- Proposal 8 (Sec 3.6.1.2): For UE-sided model inference, support option 3 (probability related information of predicted Top K beam(s)) and option 4 (confidence interval for RSRP prediction)
- Proposal 9 (Sec 3.6.1.3): For UE-sided model inference, support that value of K could be adaptive and based on the UE-sided model output
- Proposal 10 (Sec 3.6.2): For UE-side AI/ML model inference, for BM-Case2, support that UE can update reported inference results of N future time instances after such report
- Proposal 11 (Sec 3.7.1): For UE-sided model performance metric reporting, support alternative 3 (RSRP difference information between predicted and measured L1-RSRP) with both statistical and per sample metrics
- Proposal 13 (Sec 3.7.1): For UE-assisted performance metric calculation, support an event where the UE can early indicate if functionality is not working properly, e.g. prediction accuracy below 50%
- Proposal 15 (Sec 4.1): For NW-sided model, conclude there is no critical specification impact for configuring UE with set A or set B beams
- Proposal 16 (Sec 4.2): For NW-sided model, regarding max number of reported beam related information in one report, use 256 beams as a starting point
- Proposal 17 (Sec 4.3.1.1): For NW-sided model inference, support NW configuration for UEs to pre-process set B beams to reduce reporting overhead via reporting only beams within X dB of strongest beam or at most N strongest beams
- Proposal 18 (Sec 4.3.1.2): For NW-sided model inference, support methods for UEs to compress the set B temporal domain measurement results to reduce reporting overhead
- Proposal 19 (Sec 4.3.2): For NW-sided data collection, RAN1 studies possible omission/selection of collected data including avoiding duplicated samples and certain events
- Proposal 20 (Sec 4.3.2): For NW-sided data collection, conclude that it is up to RAN2 on whether RRC/MDT procedures should be supported