R1-2410249
discussion
Discussion on AI/ML for CSI compression
From Panasonic
Summary
Panasonic's technical document presents their comprehensive view on AI/ML-based CSI compression for 5G NR, containing 5 proposals and 29 observations covering inter-vendor collaboration approaches, temporal domain aspects, and specification impacts. The document advocates for combining Direction A (parameter/dataset sharing) with Direction C (fully standardized reference models) for Release 19 implementation.
Position
Panasonic strongly advocates FOR the combination of Direction A (parameter/dataset sharing enabling UE-side offline engineering) with Direction C (fully standardized reference models) as the optimal approach for Release 19, positioning this as more practical than Direction B which requires extensive standardization of UE-side conditions. They push AGAINST standalone approaches, arguing that Direction C alone provides limited performance while Direction A alone lacks reliability without standardized model structures. They specifically favor using RAN4's proposed CNN model parameters as the starting point for standardized reference models.
Key proposals
- Proposal 1 (Sec 2.1): {Target CSI} in Option 3a-1 should be target CSI exchanged from the NW-side
- Proposal 2 (Sec 2.3): For Direction C, RAN4 proposed model structure and/or model parameters for the feasibility study of testing options could be starting point
- Proposal 3 (Sec 2.3): Direction C can be used to define some of 'minimum performance' in RAN4. Further extension on top of Direction C (e.g., the combination with Direction A or B) can be considered in RAN1
- Proposal 4 (Sec 2.5): RAN1 consider the combination of Direction A and Direction C in Rel.19
- Proposal 5 (Sec 3.2): Rank adaptation handling should be studied for handling rank > 1