Model-Driven Channel Estimation for OFDM Systems Based on Image Super-Resolution Network

Reliable channel estimation is a crucial task for orthogonal frequency division multiplexing (OFDM) systems to achieve high data rate. In this paper, a deep learning-based channel estimation method that combined with image super-resolution (SR) and convolutional neural network (CNN) is proposed. Using the idea of model-driven approach, the network is initialized by the least square estimation and then trained offline to extract valid features of two-dimensional channel response matrices for high accuracy channel estimates. The results show that the proposed method significantly outperforms the linear minimum mean squared error (LMMSE) estimator in mean square error (MSE) and has potential in spectrum saving.