DNN based signal detection for underwater acoustic FBMC communications

The traditional filter bank multi-carrier (FBMC) communication receiver can only recover the transmitted symbols after channel estimation and equalization, thus the system complexity is high and the channel estimation accuracy is poor. In this paper, deep neural network (DNN) is integrated into underwater acoustic multi-carrier communication, and a DNN based signal detection method for underwater acoustic FBMC communications (DNN-FBMC) is proposed. In the training stage, the DNN parameters are improved by a large number of data iteration, hyper-parameter selection and the optimization algorithm, so that the expected estimation effect is achieved. The trained DNN model is used to replace the channel estimation and equilibrium modules and others in the receiver of traditional FBMC communication system. Meantime, the channel state information is learned adaptively and the inherent imaginary interference is avoided. In the test stage, the frequency domain sequence is considered as the input of network to predict the sending binary sequence directly. The simulation results show that the proposed DNN-FBMC signal detection method has better BER performance than traditional channel estimation algorithm.