R1-2409499
discussion
Discussion on specification support for beam management
From CMCC
Summary
CMCC presents 49 proposals and 2 observations regarding the specification impacts of AI/ML-based beam management for NR, covering data collection, inference, and monitoring for both NW-side and UE-side models. The document addresses configuration of Set A and Set B, reporting content for training and inference, Top-K beam sweeping procedures, and performance monitoring mechanisms including KPI definitions and resource set configurations.
Position
CMCC proposes that L1 signaling be supported for NW-sided training data collection and that Top-K beam sweeping be supported for NW-side inference to enhance prediction accuracy. They require that for UE-side models, the overall CPU be separately counted between legacy CSI reporting and AI/ML-based CSI reporting, while being shared among AI/ML features. CMCC prefers Option 1 for BM-Case 2 differential RSRP reporting, which includes CRI of top-K predicted beams per instance. They support dedicated resource sets and report configurations for UE-side model monitoring (Type 1 Option 2), linking the monitoring configuration to an inference report configuration via CSI-ReportConfig ID. Additionally, they propose that the granularity of UE capability reporting for AI/ML be at the sub-use case level, including details on Set A/B size, RS type, and model outputs.
Key proposals
- Proposal 1 (Sec 2.1.1): L1 signaling can be supported for NW-sided training data collection.
- Proposal 7 (Sec 2.1.2): For NW-side model data collection, UE capability for max RS per resource set can be enhanced, or Set A can contain multiple resource sets in one CSI-ResourceConfig.
- Proposal 10 (Sec 2.2.1): For UE-side model training, UE can report or request gNB for the preferred Set B pattern.
- Proposal 16 (Sec 3.1.1): Top K beam sweeping procedure is proposed to be supported for NW-side inference to increase prediction accuracy.
- Proposal 24 (Sec 3.2.2): For UE-side model inference, configuration can use one CSI-ResourceConfigId for both Set A and Set B, or two separate IDs, with association indicated by RS ID or bitmap.
- Proposal 28 (Sec 3.2.4): For UE-side BM-Case 2 inference, N future time instances can be reported in one report, with reference time based on UL slot for report + offset.
- Proposal 31 (Sec 3.2.5): For BM-Case 2 differential RSRP reporting, Option 1 is preferred, including CRI of top-K predicted beams per instance and index of beam with largest RSRP over multiple instances.
- Proposal 36 (Sec 4.1): NW-side monitoring of NW-side AI/ML model can be supported, with KPI of performance monitoring left up to gNB.
- Proposal 41 (Sec 4.2): For UE-side Type 1 monitoring Option 2, beam prediction accuracy is defined as the percentage of 'Top-1 genie-aided beam of monitoring set is one of the Top-K predicted beams', calculated as statistical results over N instances or time period T.
- Proposal 45 (Sec 4.3): For UE-side monitoring Type 1 Option 2, dedicated resource set(s) and report configuration for monitoring are configured in a dedicated CSI report configuration used for monitoring.
- Proposal 48 (Sec 5): For UE-side model, overall CPU is separately counted between legacy CSI reporting and AI/ML-based CSI reporting, but shared among AI/ML features.
- Proposal 49 (Sec 6): Granularity of UE capability for AI/ML can be sub-use cases (BM Case 1 or 2) with additional info on Set A/B size, RS type, time instances, and model outputs.