R1-2410379
discussion
Discussion on AI/ML for CSI compression
From NTT DOCOMO
Summary
NTT DOCOMO proposes a combined approach using both Direction A and Direction C for AI/ML-based CSI compression in 5G NR, advocating for eT2-like CSI monitoring schemes and performance optimization methods. The document contains 2 formal proposals and 3 observations addressing inter-vendor collaboration and performance monitoring challenges.
Position
NTT DOCOMO advocates FOR a hybrid approach combining Direction A and Direction C for AI/ML CSI compression, emphasizing that Direction C provides essential baseline performance and interoperability while Direction A enables advanced field adaptation. They push FOR eT2-like CSI monitoring over SRS-based approaches and advocate FOR studying direct KPI estimation for UE-side monitoring, positioning themselves as pragmatic supporters of standardized reference models that can serve as foundations for more complex inter-vendor collaboration schemes.
Key proposals
- Proposal 1 (Sec 3): Concluded in RAN1 that Directions A and C are needed for the normative work, with Direction C providing practical deployment options with minimized complexity and Direction A used for better field data adaptation
- Proposal 2 (Sec 4): For NW-side monitoring, support the scheme based on the eT2-like CSI report from the UE; for UE-side monitoring, further study the reliability of direct KPI estimation
- Observation 1 (Sec 3): There are performance-improving spaces for Direction C (inter-vendor collaboration Option 1) through reference model selection and adaptive layers
- Observation 2 (Sec 3): Direction C can be helpful to be considered together with Direction A/B, as it can alleviate complexity/overhead issues of Option 3a-1/4-1/3b
- Observation 3 (Sec 4): For NW-side monitoring, eT2-like CSI-based approach has better accuracy than SRS-based one; for UE-side monitoring, direct KPI estimation can outperform others regarding overhead, latency, or complexity