ZTE · 10.5.3.1

ZTE proposes a decoupled CSI acquisition framework where measurement and report configurations, as well as their triggering, are separated to natively support features like AI/ML-based beam management and spatial-domain CSI-RS overhead reduction, and opposes moving CSI containers to L2 signaling, requiring L1 (UCI) to carry CSI reports due to latency, reliability, and overhead concerns. For CSI compression, ZTE proposes a unified fixed codebook based on the Rel-16 eType-II structure with layer-specific SD basis selection when L ≤ 2 and layer-common SD basis selection when L > 2, supplemented by both Type-I-like and Type-II-like downloadable codebooks to cover broader scenarios like multi-panel UEs and near-field. ZTE presents a technical case against prioritizing JSCC/JSCCM for AI-powered CSI compression, arguing that performance gains diminish to ~2.5% SGCS at typical SINR ranges and requiring further justification on NW-side complexity, reliability, PAPR, generality, and scalability, while supporting SSCC with a UE-sided linear matrix and NW-sided AI model. ZTE proposes early CSI acquisition during initial access and inter-cell-cluster handover to enable mTRP CJT for small/medium-sized data packets that dominate >90% of real-field traffic, with early CSI reports transmittable before or after handover command reception. For CSI-RS, ZTE proposes supporting up to 256 ports in a single resource with cross-RB/slot mapping, and a hybrid high-density + low-density CSI-RS transmission pattern with UE-side AI/non-AI channel prediction to reduce overhead for ultra-massive MIMO.