R1-2600133
discussion
6GR CSI: Aspects of Downlink-based CSI Acquisition
From InterDigital
Summary
This document from InterDigital presents 21 Observations and 21 corresponding Proposals on downlink-based CSI acquisition for 6G, covering both non-AI/ML and AI/ML enhancements. The proposals address support for up to 256 CSI-RS ports, CSI-RS overhead reduction via sparse patterns, a unified scalable codebook, decoupled CSI triggering/reporting, UE-initiated CSI reporting, a flexible CPU framework, hybrid beamforming enhancements, and AI/ML-based CSI-RS overhead reduction and CSI compression.
Position
InterDigital proposes expanding the maximum number of CSI-RS antenna ports to 256 for 6GR to improve spectral efficiency, reliability, and coverage. They propose studying CSI-RS pattern adaptation based on channel sparsity order in delay/frequency and/or spatial domain as an alternative to simple CSI-RS density reduction, which they argue is inefficient for overhead reduction. InterDigital requires a single unified DL codebook that is scalable and forward-compatible, supporting both low-resolution (Type-I) and high-resolution (Type-II) CSI across sTRP, mTRP, multi-panel, and high Doppler scenarios. They oppose consideration of AI/ML-based JSCC/JSCM (sub-cases A and B) with two-sided models in Rel-20 due to significant inter-vendor training complexity involving UL propagation channel and equalization blocks, and they propose waiting for Rel-20 5GA SRS fusion work to conclude before pursuing sub-case D studies. InterDigital supports studying UE-initiated CSI reporting with UE determination of reporting quantities, decoupled triggering from timing, and a flexible CPU framework where processing units scale with the CSI processing window.
Key proposals
- Proposal 1 (Sec 2.1): Support increasing the number of CSI-RS ports, e.g., up to 256 to enhance spectral efficiency, reliability, and coverage.
- Proposal 3 (Sec 2.2): Study CSI-RS pattern adaption according to the UE channel properties, e.g., according to the sparsity order of the channel in delay/frequency and/or spatial domain.
- Proposal 6 (Sec 2.2): Study aggregation of both homogenous and heterogenous CSI-RS resources in a CSI-RS resource set to support a larger number of CSI-RS antenna ports.
- Proposal 7 (Sec 2.3): Support a single unified DL codebook as the design target for 6G, where the unified codebook is scalable and forward compatible and supports both low- and high-resolution CSI for different deployment scenarios.
- Proposal 8 (Sec 2.3): To support a simplified codebook operation, study methods to reduce UE complexity and reporting overhead by using UE-specific set of SD bases vectors for selection and reporting.
- Proposal 11 (Sec 2.4.1): Support a CSI reporting framework where the triggering of the CSI report is decoupled from the timing indication of the CSI reporting resource.
- Proposal 13 (Sec 2.4.2): Study UE-initiated CSI reporting and determination of reporting quantities.
- Proposal 15 (Sec 2.4.3): Study a CSI CPU framework where the number of CPUs for processing a CSI request is determined according to the CSI processing window associated with the request.
- Proposal 16 (Sec 2.5): Study CRI-based HBF with low overhead and complexity for 6GR.
- Proposal 17 (Sec 2.6): Study CSI reporting to support CSI determination per frequency segment for HBF.
- Proposal 18 (Sec 3.1): Study performance and complexity of CSI-RS overhead reduction Sub-Case A, including FDR, SDR and hybrid FDR-SDR approaches.
- Proposal 19 (Sec 3.2.1): AI/ML-based CSI compression and feedback sub-cases involving two-sided models should not be considered in Rel-20 due to their significant complexity for the inter-vendor training collaboration, lack of clarity on suitable baseline(s) for 6GR, and to avoid parallel work.
- Proposal 20 (Sec 3.2.2): For AI/ML-based CSI compression sub-case C, RAN1 to further identify the possible variants for further studies that at least include usage of one-sided models at the NW-side for determining the codebook(s) and limiting the scope of learned representation of the CSI feedback.
- Proposal 21 (Sec 3.2.2): Studies on AI/ML-based CSI compression sub-case D with SRS fusion for CSI reconstruction based on CSI feedback should not be pursued until the ongoing work on AI/ML-based CSI compression as part of Rel-20 5GA is completed.