Deep Learning Enabled Channel Estimation for RIS-Aided Wireless Systems

Abstract

Channel Estimation is one of the essential tasks to realize a reconfigurable intelligent surface (RIS)-aided orthogonal frequency division multiplexing (OFDM) communication system. Compared with conventional systems, the RIS introduces a cascaded channel with high dimension and sophisticated statistics. In this case, it is infeasible to derive the optimal minimum mean square error (MMSE) estimator. Additionally, the analytical channel estimators, e.g., the least square (LS) estimator and the linear minimum mean square error (LMMSE) estimator are computational costly and imprecise for practical RIS-aided systems. To address these challenge problems and accurately estimate the channel in an RIS-aided OFDM system, we model the channel estimation as a super-resolution (SR) and image restoration (IR) problem to recover the channel matrix from estimated channel at pilot positions. A convolutional neural network based on super-resolution convolutional neural network (SRCNN) and denoising convolutional neural network (DnCNN), named SRDnNet, is then proposed. The simulation results show that the performance of the proposed SRDnNet outperforms the state-of-the-art deep learning-based estimation methods and the LMMSE estimator.