FUTUREWEI · 10.5.3.1

Futurewei proposes studying structured mapping between CSI-RS ports and REs in spatial, frequency, and time domains to facilitate CSI-RS overhead reduction, demonstrating that a beam-domain-first transformation approach achieves approximately 1.5–3 dB better performance at 0 dB SNR compared to frequency-interpolation-first methods for 256 ports at 1/8 RE/RB/port density. They propose considering AI/ML-based approaches for CSI prediction from sparse CSI-RS, presenting results showing AI/ML consistently outperforms linear interpolation baselines in low SNR regimes, with gains more significant when sparsity is introduced in the spatial domain. For hybrid antenna architectures at UMB, they propose studying simultaneous multi-beam codebook designs enabling fast full CSI acquisition, arguing that for 2048 elements with 128 RF chains, full CSI port sweeping can be completed in 16 OFDM symbols versus 2048 symbols using a DFT codebook. They propose studying a unified codebook design where the network indicates the CSI-RS precoding matrix and CSI reporting basis/dictionary matrix to the UE, enabling full CSI acquisition for hybrid architectures, higher-resolution reporting, and incorporating existing 5G NR codebooks as special cases. For interference, they propose enhancements for reporting interference statistics (standard deviation/variance, maximum, minimum, percentile values) and prospective DL interference probing and reporting based on NZP CSI-RS based IMR measurements.

Futurewei proposes adopting the 5G NR MIMO/RS/CSI framework as the baseline for 6G without fundamental paradigm shifts, focusing evolutionary enhancements on UMB support. They propose studying hybrid antenna architectures for UMB base stations with large antenna element counts (e.g., >512 elements, specifically 2048-element combinations with 32-256 TXRUs at 7/15 GHz) and present technical case for simultaneous multi-beam codebooks that enable fast full CSI acquisition in N/K_s orthogonal beams per OFDM symbol. For CSI-RS overhead reduction, they propose structured port-to-RE mapping exploiting beam-domain channel sparsity and a joint channel estimation algorithm based on IDFT, contrasting with AI/ML-based approaches requiring both UE-sided and NW-sided model evaluation. They propose CSI framework simplification through streamlined CSI-ReportConfig/CSI-ResourceConfig settings and require proper conventional non-AI/ML baselines be identified for all AI/ML evaluations. They propose adopting 5G NR two-sided AI/ML CSI compression and further studying JSCC and downloadable codebook approaches while deferring joint source-channel-modulation coding due to higher ecosystem impact.