Underwater Acoustic OFDM Receiver Using a Regression-based Deep Neural Network

The orthogonal frequency-division multiplexing (OFDM) technology is a promising technology for many scenarios in underwater acoustic (UA) communications. This paper presents the design of a UA OFDM communication system, which explains the design of a traditional transmitter and a modified receiver integrated with a deep neural network (DNN). The DNN is proposed to replace the channel estimation, channel equalization and demodulation in the traditional receiver design and recover transmitted bits directly. The regression-based DNN consists of a long short-term memory (LSTM) layer. The training stage of the DNN can be either offline or online. During the testing stage, the trained network is used to recover online transmitted data directly. The offline training method is performed with maximum possible channel scenarios with a large data set. Meanwhile, the online training uses a small data set with short training time. The designed regression-based DNN receiver achieves a better performance compared to previously developed DNN receivers and the traditional receiver which is implemented with the least-squares (LS) estimator.