Dual-path residual attention network for efficient channel estimation in RIS-assisted communication systems

Abstract

Channel estimation of reconfigurable intelligent surface-aided multi-user communication (RIS-MUC) systems is one of the key tasks for expanding network coverage and improving signal transmission quality. However, such a system typically involves cascaded channels with complex statistical distributions, making channel estimation more challenging. Existing channel estimation methods face the dual challenges of high pilot overhead and limited estimation accuracy. To address the above problems, this paper proposes an efficient channel estimation framework that integrates deep learning and two-timescale channel estimation to minimize pilot overhead and improve estimation accuracy. First, this paper models the channel estimation problem as a denoising problem. Then, a denoising neural network based on the convolutional neural network (CNN) and residual structures is designed, which is named the dual-path residual attention network (DPRAN). The network leverages parallel residual structures and spatial attention mechanisms to extract spatial features from the noisy channel matrix for channel recovery. Experimental results reveal that the proposed method can achieve higher channel estimation accuracy under different channel conditions and system configurations.