Intelligent Receiver Design for Underwater Acoustic OFDM Communication Based on LSTM Networks*

The environment of underwater acoustic channel presents more intricate and demanding compared to the terrestrial radio frequency (RF) radio channel. Underwater acoustic channel estimation is a key signal processing module to cope with the channel multipath effect and Doppler effect. Intelligent underwater acoustic channel estimation based on Deep-learning neural network (DNN), convolutional neural network (CNN) and other deep learning networks is one of the development trends of underwater acoustic communication. Long-short-term memory (LSTM) networks, one of recurrent neural networks (RNN) deep learning network, exhibit particular suitability for learning from sequential data. For this reason, this research paper introduces a novel architecture for an underwater acoustic orthogonal frequency-division multiplexing (OFDM) communication system, based on the LSTM network. The LSTM network is used to replace the channel estimation, channel equalization, and signal decoding in the traditional underwater acoustic OFDM communication. Simulations are performed to analyze the effects of training signal-to-noise ratio and cyclic prefix on channel estimation performance. The communication performance of the designed LSTM channel estimator is compared with DNN channel estimator, least square (LS) channel estimator and least minimum mean square error (LMMSE) channel estimator based on underwater acoustic Millica channel and Norway Oslo Fjord channel (NOF). The results indicated that the designed LSTM channel estimator performs better.