Diffusion-Based Channel Inpainting for Robust SRS-Based CSI Acquisition
Year: 2026
Type: article
Abstract: Accurate Channel State Information (CSI) is essential for reliable multiuser Multiple-Input Multiple-Output (MIMO) and millimeter-wave communication. In 5G NR, uplink Sounding Reference Signals (SRS) are sparsely allocated, requiring channel reconstruction from partial observations. Prior masked-autoencoder (MAE) methods improve reconstruction but degrade under interference and non-Gaussian noise. We propose a diffusion-based inpainting framework that incorporates system-model knowledge via likelihood-guided inference, enabling robust reconstruction under arbitrary masking without retraining. On Clustered Delay Line (CDL) channels, the method achieves up to 14 dB Normalized Mean Squared Error (NMSE) improvement over UNet and MAE baselines under distribution shift, while maintaining competitive performance under matched conditions.
Source: IEEE Communications Letters
Authors: Usman Akram, Fan Zhang, Yang Li, Haris Vikalo
Institution: The University of Texas at Austin
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,645
OpenAlex ID: W7166887042