R1-2500319
discussion
Discussion on CSI Prediction
From TCL
Summary
TCL presents 13 proposals regarding AI-based CSI prediction using UE-side models, covering data collection, model inference, and performance monitoring. The document argues for UE-initiated data collection for training, NW-indicated collection for inference and monitoring, and the reuse of legacy Rel-18 Doppler codebook mechanisms.
Position
TCL proposes that data collection for model training should be requested by the UE, citing that the UE possesses more model information than the network, while data collection for inference and performance monitoring should be indicated by the NW. They argue for reusing the legacy feedback mechanism for codebook type 'TypeII-Doppler-r18' and supporting dynamic updates to the number of predicted CSIs based on performance deterioration. For performance monitoring, TCL prefers Type 3 as the baseline over Type 2 due to lower reporting overhead, and supports real-time monitoring with periodic, semi-persistent, aperiodic, and event-driven reporting. They further propose that CSI processing criteria and timelines require further discussion for different model operation stages.
Key proposals
- Proposal 1 (Data collection): For CSI prediction using the UE-side model, the data collection for model training should be requested by the UE.
- Proposal 2 (Data collection): For CSI prediction using the UE-side model, the data collection for model inference should be indicated by the NW.
- Proposal 3 (Data collection): For CSI prediction using the UE-side model, the data collection for performance monitoring should be indicated by the NW.
- Proposal 4 (Model inference): For CSI prediction using UE-side model, dynamically update the number of predicted CSI should be supported.
- Proposal 5 (Model inference): For CSI prediction using the UE-side model, the legacy feedback mechanism for the codebook type “TypeII-Doppler-r18” should be reused.
- Proposal 6 (Model inference): For CSI prediction using the UE-side model, the predicted CSI support periodic, semi-persistent, and aperiodic reporting.
- Proposal 7 (Model inference): For CSI prediction using UE-side model, CSI processing criteria and timeline should be further discussed for different stages of model operation.
- Proposal 8 (Performance monitoring): For CSI prediction using the UE-side model, Type 3 should be used as a baseline, and the definition and configuration of performance metrics need further discussion.
- Proposal 9 (Performance monitoring): For CSI prediction using the UE-side model, the periodic or semi-persistent CSI-RS resource can be used for both model inference and performance monitoring simultaneously.
- Proposal 10 (Performance monitoring): For CSI prediction using the UE-side model, additional aperiodic CSI-RS resources should be configured to support performance monitoring.
- Proposal 11 (Performance monitoring): For CSI prediction using the UE-side model, support real-time monitoring of inference result.
- Proposal 12 (Performance monitoring): For CSI prediction using the UE-side model, support periodic, semi-persistent, and aperiodic reporting the performance metric(s), performance monitoring output, or ground truth CSI(s), if supported.
- Proposal 13 (Performance monitoring): For performance monitoring, support UE reporting of performance metric(s) based on event-driven mechanisms.