Channel Estimation for Double-RIS-Assisted MIMO Communication Systems Based on Deep Learning

Abstract

This letter addresses the channel estimation (CE) problem in a double-reconfigurable intelligent surface (RIS)-assisted multiple-input multiple-output (MIMO) wireless communication system. Existing solutions often introduce additional control overhead due to frequent updates of RIS phases and amplitudes. To address this, we adopt a phase shift conversion strategy that eliminates the need for such switching by designing two orthogonal phase shift modes to extract the double-reflection channel. We then develop a first-order fast iterative shrinkage-thresholding algorithm (FISTA) for channel pre-estimation, and further refine it by means of a residual convolutional hybrid attention network (RE-CHAN) which leverages parallel residual blocks with convolutional kernels of different sizes and multi-head attention to enhance feature learning capability. Simulations show that FISTA achieves higher pre-estimation accuracy than existing methods, while the RE-CHAN model outperforms state-of-the-art models in the normalized mean squared error (NMSE) metric.