R1-2409741
discussion
Specification support for beam management
From Intel
Summary
Intel presents a comprehensive set of proposals for AI/ML-aided beam management in NR Rel-19, covering configuration, data collection, inference, and performance monitoring for both network-side and UE-side models. The document contains approximately 35 distinct proposals aimed at leveraging existing CSI frameworks while introducing specific mechanisms for consistency, monitoring triggers, and model failure indication.
Position
Intel proposes that NW-sided models utilize implicit Set A/B configuration via legacy CSI frameworks to minimize signaling overhead, whereas UE-sided models require explicit configuration of Sets A and B, potentially using a new IE for Set A resources. They require the prediction window reference time for BM-Case 2 to be tied to the UL report slot and mandate differentiation between measured and predicted RSRP in UE reports via an additional bit. Intel supports multiple performance monitoring metrics (Alt. 1, 2, 4) and proposes UE-event-triggered monitoring to reduce overhead, including specific events for beam mismatch and RSRP degradation. They require an Associated ID mechanism for training-inference consistency, limited to the serving cell, and argue that performance monitoring-based consistency checks should be deprioritized. Finally, they propose that NW-side additional conditions be treated as optional and functionally separate from inference configuration in response to RAN2 inquiries.
Key proposals
- Proposal (Sec 2.1.1): For NW-sided models, implicit configuration of Set A/B using existing CSI-ResourceConfig and CSI-ReportConfig is sufficient; multiple resource sets per L1 report are unnecessary.
- Proposal (Sec 2.2.1): For UE-sided models, explicit configuration of Set A and Set B is required, supporting either separate CSI-ResourceConfigIds or a new IE for Set A resources derived from TCI states.
- Proposal (Sec 2.2.2): For BM-Case 2 UE-sided models, the prediction window reference time should be defined relative to the UL slot for the report, with window length subject to UE capability.
- Proposal (Sec 3.1.1): For NW-sided training data collection, L1-RSRPs from the union of Sets A and B (transparent to UE) should be reported, with beam information for Top-K beams determined implicitly.
- Proposal (Sec 3.1.1): Data collection containers should depend on purpose: L1 reporting for inference/monitoring, and RRC/MAC-CE for training data.
- Proposal (Sec 3.2.2): For UE-sided inference reporting, an additional bit per beam is needed to differentiate between measured L1-RSRP and predicted RSRP.
- Proposal (Sec 3.2.3): RAN1 should consider beam indication for predicted beams whose TCI states are not part of the MAC-CE activated TCI state list to avoid latency.
- Proposal (Sec 4.1): Support Alt. 1 (prediction accuracy), Alt. 2 (link quality), and Alt. 4 (RSRP difference) as model monitoring metrics.
- Proposal (Sec 4.2): For UE-sided monitoring, support UE-event-triggered reporting based on events such as Top-K beam mismatch, RSRP threshold breaches, or RSRP difference thresholds.
- Proposal (Sec 4.3): Support indication of AI/ML model failure if the RSRP difference between measured and predicted values falls below a threshold on configured resources.
- Proposal (Sec 5.1): For consistency between training and inference, an Associated ID should be assigned to Set A/B configurations, limited to the serving cell scope.
- Proposal (Sec 5.1): Performance monitoring-based approaches for ensuring consistency should be deprioritized in favor of Associated ID mechanisms.
- Proposal (Sec 6): RAN1 should respond to RAN2 LS regarding applicable functionality reporting, clarifying that NW-side additional conditions (Associated IDs) can be optional and separate from inference configuration.