Deep Learning Based CS Channel Estimation in MIMO System

In high-speed mobile communication scenarios, pilot-based channel estimation algorithms need more pilots in order to track channel changes, greatly reducing the efficiency of system. In order to save the number of pilots, we propose a novel compressive sensing channel estimation algorithm based on deep learning (DLCS). The proposed scheme takes a two-stage estimation process: 1) a deep learning model to obtain the sparsity of the channels in the Doppler domain and 2) the compressive sensing reconstruction process to estimate the channel state information. Simulation results showed that the proposed DLCS algorithm can work well with fewer pilots and its performance is better than that of traditional channel estimations in high-speed scenarios.