Doppler Shift Compensation Using an LSTM-based Deep Neural Network in Underwater Acoustic Communication Systems

In underwater acoustic (UA) communication, Doppler effect is much larger than that of the terrestrial radio communication as propagation speed of the acoustic wave in water is far slower than that of the electromagnetic wave in the air. Frequency offset and Doppler effect severely degrade the performance of orthogonal frequency-division multiplexing (OFDM) systems. This paper presents a Doppler shift mitigation approach in UA OFDM communication systems incorporated with a deep neural network (DNN) based receiver. The regression-based DNN is incorporated with a long short-term memory (LSTM) layer, which has an improved feature extraction capability compared with the commonly used neural network methods. The DNN is trained by simulation data with Doppler shifts and is used to predict the transmitted bits. The results show that the proposed LSTM-based DNN receiver achieves better performance than the traditional receiver, which is implemented with the least-squares (LS) channel estimator, and the commonly used neural network methods.