Deep Learning-Based Channel Estimation Method for MIMO Systems in Spatially Correlated Channels

Abstract

This paper proposes a deep learning-based channel estimation method for multiple-input multiple-output (MIMO) systems in spatially correlated channels. To reduce the pilot overhead of pilot symbol-assisted channel estimation, the proposed method utilizes fewer pilot symbols than the number of transmit antennas. Firstly, based on pilot symbols, the estimated partial MIMO channel matrix, consisting of the partial coefficients of the MIMO channel matrix, is obtained by the linear minimum mean square error algorithm. After that, a deep neural network uses the estimated partial MIMO channel matrix as an input and we have the predicted MIMO channel matrix, that corresponds to the channel state information not transmitting pilot symbols. Finally, by aggregating the estimated partial MIMO channel matrix and the predicted MIMO channel matrix, the proposed method can acquire the reconstructed MIMO channel matrix. In simulation results, to show the validity of the proposed method, various performances of the proposed and conventional channel estimation methods were evaluated. Simulation results show that even though the proposed method does not send the pilot symbols for all transmit antennas, it can achieve almost the same bit error rate and improved throughput performances compared with the conventional channel estimation method.

Keywords

• Channel estimation method
• deep learning
• MIMO system
• spatially correlated channel