R1-2409853
discussion
Discussion on specification support for CSI prediction
From NEC
Summary
NEC's contribution discusses specification support for AI/ML-based CSI prediction with UE-sided models, focusing on consistency between training and inference phases and normative work preparation. The document contains 14 proposals and 1 observation covering consistency mechanisms, performance monitoring, model management, and CSI reporting enhancements.
Position
NEC advocates for reusing existing associated ID mechanisms from beam management cases for training-inference consistency, emphasizing UE-side performance monitoring (Type 1 and 3) over network-side monitoring to minimize overhead while maintaining control. They push for NW-controlled AI/ML-based CSI reporting to preserve legacy principles and support discontinuous CSI-RS reception mechanisms, positioning against pure UE-autonomous reporting that could lead to uncontrolled behavior across different UE implementations.
Key proposals
- Proposal 1 (Sec 2): For the consistence of NW-side additional condition across training and inference for CSI prediction with UE-sided model, reuse the associated ID introduced in BM-Case 1/2 (if the consistence is necessary)
- Proposal 2 (Sec 2): For the consistence of NW-side additional condition across training and inference for CSI prediction with UE-sided model, performance monitoring can be used (irrespective of whether the consistence is necessary)
- Proposal 3 (Sec 3.1): For data collection for CSI prediction using UE-side model, at least the CSI measurement period in the observation window and the CSI prediction period in the prediction window need to be provided from UE to NW
- Proposal 4 (Sec 3.2): For performance monitoring, the definition of performance metric should focus on Type 1 and 3 performance monitoring (i.e., calculation of performance metric is performed at UE side)
- Proposal 5 (Sec 3.2): For Type 1 and 3 performance monitoring, the performance metric should be intermediate KPI (e.g., SGCS, NMSE), or input/output data distribution related metric (e.g., similarity, divergence, distance)
- Proposal 6 (Sec 3.2): For Type 1 performance monitoring, the performance monitoring output should comprise states indicating whether current and candidate AI/ML models meet NW-provided thresholds
- Proposal 8 (Sec 3.2): Potential spec impact includes determination and reporting of performance monitoring output for Type 1, quantization and reporting of metrics for Type 2, and association between predicted and ground-truth CSI reporting for Type 3
- Proposal 10 (Sec 3.2): Study how to refine the performance monitoring procedure when the target timing of predicted CSI is not aligned with the timing of available ground-CSI truth
- Proposal 11 (Sec 3.3): For model switching, selection and update, if input/output parameters change, UE must report the changes to NW
- Proposal 12 (Sec 3.3): AI/ML-based CSI prediction and reporting should be performed under NW configurations
- Proposal 13 (Sec 4): For periodic CSI report, study the mechanism of discontinuous CSI-RS reception
- Proposal 14 (Sec 4): For CSI prediction, the CSI reporting periodicity may be updated autonomously upon reaching a significant point of variation determined by time, location or distance