Channel Estimation for Intelligent Reflecting Surface Assisted Massive MIMO Systems—A Deep Learning Approach

Abstract

The benefits of intelligent reflecting surfaces (IRS) assisted massive multiple-input multiple-output systems are based on the accurate acquisition of channel state information but at the cost of the high pilot overhead. In this work, the sparse structure in angular domain is revealed, and then the uplink cascaded channel estimation is converted into a compressive sensing (CS) problem. However, the angle of arrival and angle of departure are fundamentally continuous values, and thus they usually do not fall precisely on the discrete grids, resulting in grid mismatch. In this case, the typical CS solution usually yields the compromised reconstruction performance. To address this issue, we proposed a deep learning-based approach with the traditional orthogonal matching pursuit followed by the residual network to improve the performance. Furthermore, a straightforward network structure is proposed to reduce computational complexity. Simulation results demonstrate that the proposed solutions achieve a better estimation performance and require lower pilot overhead compared with the state-of-the-art ones.