R1-2409569
discussion
Specification Support for AI/ML for Beam Management
From Kyocera
Kyocera's prior position on
9.1.1
at
RAN1#118bis
· AI-synthesized, paraphrased
verify sources →
Advocates for a comprehensive AI/ML beam management framework leveraging existing CSI infrastructure with minimal new specifications, strongly supporting virtual Set A configuration for UE-side models and flexible Set B definition through new information elements.
Summary
Kyocera presents a comprehensive framework for AI/ML-assisted beam management in NR Rel-19, covering configuration, inference reporting, consistency, and performance monitoring for both NW-sided and UE-sided models. The document contains 35 proposals and 4 observations, focusing on defining Set A/B configurations, introducing associated IDs for consistency, and detailing specific metrics for Type 1 performance monitoring.
Position
Kyocera proposes that for UE-side AI/ML models, Set A be virtually configured for reference mapping while Set B is explicitly configured for measurements, requiring a new IE to associate beams with these sets. They require the introduction of an 'associated ID' within the CSI framework (nzp-CSI-RS-ResourceSet/csi-SSB-ResourceSet) to ensure consistency of network-side additional conditions, arguing that this ID should be limited to small geographical areas or single cells to avoid revealing proprietary vendor information. For inference reporting, they support 4-bit probability precision and propose studying confidence intervals defined as statistical ranges for predicted RSRP. Regarding performance monitoring, Kyocera supports Type 1 Option 2 metrics including beam prediction accuracy and RSRP differences, but deprioritizes probability-based metrics (Alt 4) due to the lack of ground truth. They propose a 'BeamCorrespondence' IE to link monitoring resources to beam indices and argue that Type 2 monitoring and NW-side monitoring should be left to vendor implementation without specification impact.
Key proposals
- Proposal 1 (Sec 2.1): For UE-side AI/ML models, Set A is virtually configured as a reference for mapping model outputs to CRIs, while Set B is explicitly configured for UE measurements.
- Proposal 2 (Sec 2.2): An additional Information Element (IE) is required within CSI-ResourceConfig or a new AI/ML configuration to associate beams with Sets A and B.
- Proposal 3 (Sec 2.3): Configuration of Sets A/B imposes gNB scheduling restrictions but requires no additional UE capability, classifying it as an implementation issue.
- Proposal 4 (Sec 2.4): For NW-side training, study using the CSI framework with new IEs like 'CSI-AIML-ResourceSet' and 'AIML-info-RSRP' to collect inputs and labels, reporting all Set B measurements.
- Proposal 5 (Sec 3.1): Support inference report options 1-3 for UE-side models; further study Option 4 (confidence information) considering quantization effects and overhead.
- Proposal 6 (Sec 3.2): Define confidence information as the range within which predicted RSRP falls a certain percentage of the time if the same input sample is reused.
- Proposal 9 (Sec 3.4): Support 4-bit width for probability information reporting, representing one decimal precision, with rounding left to UE implementation.
- Proposal 10 (Sec 3.5): Support CRI-based vector reporting or Bitmap reporting for ranking top-K beams, with FFS on overhead feasibility.
- Proposal 11 (Sec 4.1): Ensure consistency for UE-side models by assuming consistent coverage, beam properties, set sizes, and resource indexing for the same associated ID.
- Proposal 12 (Sec 4.2): Introduce the associated ID as a new IE within RRC messaging of nzp-CSI-RS-ResourceSet and csi-SSB-ResourceSet.
- Proposal 15 (Sec 5.1): For Type 1 Option 1 monitoring, report model outputs (top-K beams, predicted RSRP, probability, confidence) and performance measurements (CRI/SSBRI, L1-RSRP).
- Proposal 17 (Sec 5.2): Support Alt1 (beam prediction accuracy), Alt2 (L1-RSRP difference), and Alt3 (RSRP difference) for Type 1 Option 2 monitoring, deprioritizing Alt4.
- Proposal 22 (Sec 5.2): Introduce a new IE 'BeamCorrespondence' within 'NZP-CSI-RS-Resource' to link performance monitoring resources to beam indices when using dedicated monitoring reports.
- Proposal 24 (Sec 5.3): Type 2 performance monitoring has no specification impact and should be left to UE vendor implementation.
- Proposal 25 (Sec 5.4): NW-side AI/ML model performance monitoring has no specification impact and should be left to NW vendor implementation.