R1-2601814 discussion

Discussion on aspects of downlink-based CSI acquisition for 6GR

From Spreadtrum
Status: not treated
WI: FS_6G_Radio
Agenda: 10.5.3.1
Release: Rel-20
Source: 3gpp.org ↗
Spreadtrum's prior position on 10.5.3.1 at RAN1#124 · AI-synthesized, paraphrased
verify sources →
Proposes a unified CSI acquisition framework design that associates a CSI report configuration with a channel measurement configuration and/or an interference measurement configuration, supporting multiple scenarios including early CSI for candidate cells. Proposes studying both NW-controlled and UE-initiated CSI reporting, and considers PUSCH-only based CSI reporting for all time domain behaviors in 6GR. For CSI-RS design, supports up to 128 ports per single CSI-RS resource and proposes PN sequence with UE-specific seed with component RE pattern aggregation. On AI/ML, presents technical case showing JSCC/JSCM performance gains over SSCC especially at low SNR but proposes postponing two-side model work until NR standardization completes. Prioritizes low overhead CSI-RS for 6GR Day 1 AI use case, presenting evaluation results showing less than 5% SGCS loss with 87% frequency domain and 75% spatial domain RS overhead reduction, while noting spatial domain generalization across deployment scenarios needs further study due to approximately 28% SGCS loss.

Summary

Spreadtrum's R1-2601814 presents 21 proposals and 12 observations on 6G downlink CSI acquisition, covering a unified CSI framework, basic CSI-RS design supporting up to 128 ports, and AI/ML-based CSI compression and low-overhead CSI-RS. The document recommends postponing two-side AI/ML models (JSCC/JSCM) until NR work completes while prioritizing low-overhead CSI-RS as a Day-1 6G AI use case.

Position

Spreadtrum proposes a unified CSI acquisition framework spanning sTRP, mTRP, mTRP calibration, and early CSI scenarios, preferring to study inclusion of candidate cell early CSI rather than creating separate frameworks. They oppose L2-based (MAC-CE) CSI reporting as a Day-1 6G priority, arguing it introduces substantial latency from MAC PDU construction and HARQ retransmission, and incurs significant overhead from MAC sub-headers and CRC compared to L1 UCI. For CSI-RS design, Spreadtrum supports up to 128 ports as a starting point (opposing 256 ports due to UE measurement complexity), requires using a single CSI-RS resource to contain all ports rather than resource aggregation, and proposes sparse frequency/spatial domain density combined with component RE pattern aggregation for overhead reduction. On AI/ML, Spreadtrum presents a case for postponing JSCC/JSCM standardization until NR two-side model work completes, citing unresolved inter-vendor issues, while prioritizing low-overhead CSI-RS reconstruction as the Day-1 6G AI use case based on simulation results showing less than 5% SGCS loss with 87% frequency domain and 75% spatial domain RS overhead reduction.

Key proposals

Your notes

Private to your account