R1-2410572
discussion
Discussion on AI/ML for CSI compression
From CEWiT
Summary
CEWiT contributes technical analysis on AI/ML-based CSI compression for NR air interface, addressing data collection, model monitoring, and inference aspects. The document contains 14 proposals and 4 observations covering spatial-temporal-frequency CSI compression implementations.
Position
CEWiT advocates FOR network-side monitoring as the primary approach for AI/ML CSI compression, supporting basis components transmission method over high-resolution codebook methods for reduced payload. They push FOR Option-1 CQI determination (not based on CSI reconstruction output) as the starting point while arguing AGAINST Option-1c. They strongly support reusing existing legacy signaling methods and procedures rather than creating entirely new mechanisms.
Key proposals
- Proposal-1 (Sec 2.1): In case of Case-3 and Case-4 based CSI compression, study the effects of having a separate prediction module versus compression plus prediction module at the UE side
- Proposal-3 (Sec 2.2): For NW sided data collection, specify the CSI configuration, parameter combinations and periodicity for data collection
- Proposal-6 (Sec 2.2.2): For UE sided data collection, existing procedures can be reused for data collection
- Proposal-7 (Sec 2.3.1): In the case of selection of number of basis components, the knee point can be a good indicator
- Proposal-10 (Sec 2.3.1): For model monitoring of AI/ML based CSI compression, prioritize NW sided monitoring
- Proposal-11 (Sec 2.4.1): For AI/ML based CSI compression, further study configurations and related aspects for various model options
- Proposal-13 (Sec 2.4.2): For CQI determination, consider Option-1 (CQI is NOT calculated based on the output of CSI reconstruction part from the realistic channel estimation) to be the starting point
- Proposal-2 (Sec 2.1): Study methods to model the absence of past CSI in the case of rank adaptation in Case-3 and Case-4 based CSI compression
- Proposal-4 (Sec 2.2.1): For NW sided data collection, additional information e.g. SINR (in terms of CQI) on top of ground truth CSI
- Proposal-8 (Sec 2.3.1): For model monitoring of AI/ML based CSI compression, reuse necessary signalling for transmitting the basis using legacy methods
- Proposal-9 (Sec 2.3.1): For model monitoring of AI/ML based CSI compression, prioritize transmission of basis components based monitoring technique
- Proposal-12 (Sec 2.4.1): In the case of dynamic change in channel conditions, consider signalling and appropriate indicators for UE report for model switching procedure
- Proposal-5 (Sec 2.2.1): Consider channel parameter as a part of dataset-ID for data collection
- Proposal-14 (Sec 2.4.2): For CQI determination, Option 1c can be deprioritised