Channel Replay Aided Neural Network Equalizer for Underwater Acoustic Communications

Underwater acoustic (UWA) communication is challenging due to various factors, such as extensive multipath propagation, strong Doppler effect, and so on. Channel equalization plays a key role in recovering transmitted symbols from received signals. Deep learning based channel equalization has recently showed great potentials. However, the training of a deep neural network for channel equalization is challenging due to limited number of pilot symbols in practical systems. To overcome this delimma, we propose a channel replay aided neural network equalizer (NNE). Attribute to the channel replay technique, theoretically infinite channel realizations can be obtained for generating sufficient training data. At-sea experimental results verified the feasibility of the proposed NNE for UWA communications and also showed its performance advantage over traditional adaptive equalization.