R1-2410202
discussion
Discussion on AIML CSI compression
From TCL
Summary
TCL presents a comprehensive technical contribution on AI/ML CSI compression for NR air interface, addressing resource configuration, priority rules, UE capabilities, collaborative training, and overhead reduction. The document contains 7 proposals and 5 observations covering specification impacts for CSI compression model inference.
Position
TCL advocates FOR practical AI/ML CSI compression solutions that maintain compatibility with legacy CSI frameworks while introducing necessary AI/ML-specific enhancements. They push AGAINST overly complex collaborative training options (Options 1-2) that require full model parameter exchange, instead favoring structured approaches (Options 3-5) that balance standardization needs with implementation flexibility. TCL emphasizes the need for dedicated AI/ML priority rules and UE capability descriptions while promoting efficient overhead reduction mechanisms.
Key proposals
- Proposal 1 (Sec 2.1): Design AI/ML-specific CSI-RS resource and CSI reporting configuration compatible with traditional CSI reporting, including AI/ML-specific CSI-RS resource configuration, dedicated fields in CSI-ReportConfig IE, and dedicated report quantities for AI
- Proposal 2 (Sec 2.2): Define AI/ML-specific priority for CSI reporting in relation to CSI compression compared with traditional CSI priority rules
- Proposal 3 (Sec 2.2): Redefine priority rules when UE supports both AI/ML and non-AI/ML CSI reporting, considering different types of CSI reporting
- Proposal 4 (Sec 2.3): Study how to describe UE capabilities to implement AI/ML models for inference on CSI compression and calculations
- Proposal 5 (Sec 2.4): Option 1 and 2 for collaborative training should not be standardized, at least out of scope for R19
- Proposal 6 (Sec 2.4): RAN1 should down select among options 3, 4 and 5 considering if unified model format or structure is shared between NW and UE side models
- Proposal 7 (Sec 2.5): RAN1 should consider whether/how to transmit raw CSI for monitoring purpose, with FFS on overhead reduction scheme based on spectral or temporal processing