mmWave RIS-Assisted SIMO Channel Estimation Based on Global Attention Residual Network

Reconfigurable intelligent surface (RIS) is promising for enhancing millimeter wave signal coverage. However, traditional channel estimation (CE) methods have high complexity and pilot overhead due to RIS's passive nature and a large number of unit cells. Recently, deep learning (DL) has shown the potential in improving communication system performance. This letter proposes a DL-based scheme for estimating the cascaded channel in a RIS-assisted communication system. The proposed scheme utilizes the global attention residual network, which considers multi-channel information fusion on the channel feature matrices to improve CE matrix accuracy. Simulation results demonstrate that the proposed scheme significantly improves CE accuracy and has good generalization performance.