R1-2601839
discussion
On downlink-based CSI acquisition in 6GR
From Nokia
Nokia's prior position on
10.5.3.1
at
RAN1#124
· AI-synthesized, paraphrased
verify sources →
Proposes establishing 5G Rel-20 CSI-RS design up to 128 APs as the baseline for 6GR CSI acquisition and strives for reusing existing 5G CSI-RS design principles for new 6GR antenna array configurations, including studying new frequency domain densities of 1/16 and 1/32. Presents technical case for a unified codebook design that delivers accurate PMI representation across the full range of feedback overhead for both SU-MIMO and MU-MIMO operations, while studying codebook adaptation aspects for different sub-array patterns. Questions the necessity of periodic CSI reporting given semi-persistent reporting flexibility, proposes studying the feasibility of PUSCH-only CSI reporting to simplify UCI handling, and advocates for studying UE-initiated/event-based CSI reporting as a new type. In the AI/ML domain, observes that JSCC/JSCM primarily improves low-SNR performance by mitigating the cliff effect and that one-sided JSCM performs close to two-sided, while proposing to study generalization, scalability, and fine-tuning for CSI prediction under Rel-18-defined cases.
Summary
This Nokia Tdoc discusses the evolution of the downlink CSI acquisition framework for 6GR, presenting 40 proposals and 29 observations aimed at enhancing flexibility, scalability, and energy efficiency. Key themes include CSI-RS design for large antenna arrays (up to 256 ports), a streamlined CSI framework addressing energy savings and large bandwidths, support for AI/ML-based CSI prediction and compression, and a simplified signaling and processing framework.
Position
Nokia proposes a comprehensive restructuring of the 6GR DL CSI acquisition framework, using the 5G NR framework as a baseline but mandating studies to address its identified shortcomings. The company proposes studying CSI-RS designs supporting up to 256 antenna ports through new CDM group sizes, block-pattern and comb-pattern based configurations, and flexible density adaptation in frequency and spatial domains. It questions the necessity of periodic CSI reporting and proposes studying PUSCH-only CSI reporting to simplify UCI handling, and supports decoupling CSI measurement triggering from reporting to alleviate PUSCH scheduling restrictions. Nokia proposes adopting UE-initiated/event-based CSI reporting, proposing studies on use cases where reporting occurs only upon an associated event or the UE selects a reporting configuration, and presents technical arguments for simplifying the legacy CPU occupation model by evolving the CSI processing framework. It advocates for a single specified codebook design for PMI compression and proposes studying AI/ML-based CSI compression under practical constraints, including constrained QAM modulation symbols for JSCM, while presenting initial simulation results on single-sided JSCM performance.
Key proposals
- Proposal 1 (Sec 2): 6GR discussions on CSI acquisition framework to consider at least the following guiding design principles: Minimize overhead and save energy while at least maintaining, and preferably increasing, the system performance relative to 5G NR; Tackle CSI aging; Support UEs with different capabilities; Enable flexible CSI reporting; Enable early triggering/activation.
- Proposal 3 (Sec 3): Study block pattern based multi-CSI-RS resource aggregation configurations with new CDM types(s), different antenna port densities and number of resources up to 256-APs.
- Proposal 9 (Sec 3): 6GR should strive for a common CSI-RS design framework that enables the co-existence of different modes (e.g. classical and AI/ML) of CSI acquisition in 6GR.
- Proposal 10 (Sec 4.1): 6GR CSI acquisition framework to accommodate coexistence of UEs with different capabilities (e.g., in terms of maximum number of CSI-RS ports, or support of interpolation/prediction).
- Proposal 14 (Sec 4.2): Study how to evolve the Rel-18/Rel-19 CSI framework for energy saving in 6GR to at least address the shortcomings of legacy framework, such as by focusing on reducing reporting and higher-layer configuration overhead, considering RS (such as CSI-RS) adaptation, targeting jointly network and UE energy/power savings, and studying leveraging SRS for the purpose of spatial adaptation.
- Proposal 17 (Sec 4.3): Study enhancements to frequency-domain CSI resolution aimed at the large signal bandwidths expected for 6GR.
- Proposal 19 (Sec 4.4): Study limiting CSI reporting to PUSCH in 6GR.
- Proposal 21 (Sec 4.6): Support decoupling triggering CSI measurement from triggering CSI reporting, to avoid unnecessary limitations on data scheduling/transmission.
- Proposal 22 (Sec 4.7): Study how to evolve and simplify the CSI processing framework in 6GR.
- Proposal 24 (Sec 4.8): Study adopting UE-initiated/event-based CSI reporting as a new CSI reporting type in 6GR, and discuss its scope in terms of applicable scenarios and associated events/conditions.
- Proposal 27 (Sec 4.9): Study how DMRS/PDSCH resources can be leveraged for channel/interference measurements, including which CSI quantities can be derived, to mainly increase CSI accuracy and reduce CSI-RS overhead.
- Proposal 30 (Sec 5.1): For the frequency and/or spatial domain CSI prediction, study generalization and scalability with a baseline of the three cases defined in Rel-18.
- Proposal 33 (Sec 5.2): Study enhanced CSI time-domain prediction by jointly leveraging multiple CSI-RS resource types to form a larger observation window for channel evolution tracking.
- Proposal 34 (Sec 6.1): In 6GR, study a single specified codebook design for PMI compression to target the best trade-off between UPT and feedback overhead over a wide range of feedback overheads, for both SU-MIMO and MU-MIMO operations.
- Proposal 38 (Sec 6.2): The study should focus on evaluating JSCC/M (including single‑sided JSCM) under practical constraints, such as constrained modulation symbols, and should evaluate the robustness and recovery behavior under bursty channel errors due to the absence of channel coding in JSCC/M, and other relevant non‑idealities.