Underwater Acoustic Channel Estimation Based on Sparsity-Aware Deep Neural Networks

The estimation of the underwater acoustic channel (UAC) is a difficult problem in underwater acoustic orthogonal frequency division multiplexing (UA-OFDM) systems due to the detrimental characteristics of the UAC, including severe multipath fading, Doppler spread, and large transmission delay, etc. To overcome the problems and improve the performance of UAC estimation, a deep-learning-based approach utilizing a sparsity-aware deep neural network (DNN) emulating the sparse recovery algorithm of approximate message passing (AMP) is proposed for UAC estimation in this paper. The proposed method called Sparsity-Aware-DNN-based UAC Estimation (SAD-UACE) decomposes the conventional iterative sparse recovery algorithm of AMP into several differently parameterized layers of a DNN to learn the inherent sparse structure of the UAC, so that the accuracy of estimation can be improved, especially in severe conditions of underwater acoustic communications. Simulation results show that the proposed SAD-UACE method can significantly improve the accuracy and spectral efficiency of UAC estimation, compared with other state-of-the-art methods, especially in severe conditions of low SNR or insufficient pilots.