R1-2409501
discussion
Discussion on AI/ML for CSI prediction
From CMCC
Summary
CMCC discusses specification impacts for AI/ML-based CSI prediction in Rel-19, focusing on ensuring consistency between training and inference phases. The document presents 8 proposals covering consistency mechanisms via associated IDs and performance monitoring, data collection reuse from beam management, and adaptation of Rel-18 CSI parameters.
Position
CMCC proposes studying two options for ensuring consistency of NW-side additional conditions across training and inference for AI-based CSI prediction: one based on associated ID and another based on performance monitoring. They propose combining these solutions, where the associated ID preliminarily guarantees consistency without exposing proprietary information, while performance monitoring handles residual issues. CMCC requires that if associated ID is supported, the UE assumes consistency of NW-side additional conditions with the same ID at least within a cell, and that the ID is configured within the CSI framework. For performance monitoring, they propose using intermediate KPIs as a starting point for Type 3 monitoring and present two alternatives for Type 1 monitoring involving UE-calculated metrics or UE-recommended LCM decisions. Finally, CMCC proposes reusing data collection mechanisms from AI/ML beam management and adapting Rel-18 CSI parameters, such as measurement and reporting windows, for AI/ML-enabled CSI prediction.
Key proposals
- Proposal 1 (Sec 2.1): Study two options for consistency of NW-side additional conditions across training and inference: Option 1 based on associated ID and Option 2 based on performance monitoring.
- Proposal 2 (Sec 2.1): Propose that the associated ID based solution can be used together with performance monitoring to guarantee consistency.
- Proposal 3 (Sec 2.1): If associated ID is supported, the UE assumes NW-side additional conditions with the same associated ID are consistent at least within a cell.
- Proposal 4 (Sec 2.1): If associated ID is supported, it can at least be configured within the CSI framework.
- Proposal 5 (Sec 2.2): For Type 3 performance monitoring, intermediate KPI (e.g., SGCS, NMSE) can be considered as a starting point for defining the performance metric.
- Proposal 6 (Sec 2.2): For Type 1 performance monitoring, propose two alternatives: Alt1 where UE calculates metric/comparison outcome based on NW-configured threshold, or Alt2 where UE recommends LCM decision to NW.
- Proposal 7 (Sec 2.3): For data collection of AI/ML based CSI prediction, the data collection mechanism of AI/ML based beam management can be reused as much as possible.
- Proposal 8 (Sec 2.4): For CSI prediction, CSI related parameters in Rel-18 (e.g., measurement window, reporting window) can be reused with modification to adapt to AI/ML-enabled CSI prediction.