Abstract:  A channel estimation algorithm based on deep compressed sensing is proposed to solve the problem of high cost and limited accuracy of channel estimation pilots in reconfigurable intelligent surface (RIS) assisted multi-user communication system. To reduce the pilot cost of the traditional orthogonal matching pursuit (OMP) algorithm, an improved OMP algorithm is proposed by using the unique double-structured sparse property of cascaded channels to obtain rough estimation of the cascaded channel. In order to further improve the accuracy of channel estimation, a deep learning model is designed, which regards the coarsely estimated channel matrix as a low-resolution image, and uses the multi-convolutional network to learn the implicit noise features to the maximum extent. Finally, the multi-convolutional network structure based on residual connection (RMCN) is proposed. By using the spatial characteristics and additivity of noise, the influence of noise on the channel matrix is eliminated, and a cascade channel matrix with high resolution is output to complete the channel estimation. The simulation results show that compared with the traditional OMP scheme, the normalized mean square error of the proposed RMCN-OMP algorithm is reduced by about 2.5 dB while reducing pilot overhead and achieving higher estimation accuracy.

Key words: reconfigurable intelligent surface, channel estimation, deep compression sensing , multi-convolutional network