Research On Physical Layer Technology Of Underwater Acoustic OFDM Based On Deep Learning

Orthogonal Frequency Division Multiplexing(OFDM)is a multi-carrier transmission technology with high spectrum utilization and strong anti-multipath capability.OFDM has been widely used in underwater acoustic communication.Deep learning technology has attracted much attention because of its excellent performance in the fields of computer vision and natural language processing.Its strong ability of feature extraction and fitting has provided a chance for underwater acoustic OFDM communication.This thesis conducts an in-depth study on the channel coding,modulation,channel estimation,demodulation and channel decoding for the underwater acoustic OFDM system,and tries to use the deep learning technology to improve the performance of the underwater acoustic communication system.Firstly,according to the physical characteristics of underwater acoustic channel,this thesis proposes an underwater acoustic OFDM channel estimation scheme based on Enhanced Super-Resolution Generative Adversarial Networks(ESRGAN).In this scheme,the idea of image super-resolution is introduced into the channel estimation task,and the generative adversarial networks and Residual in Residual Dense Block are used for channel estimation.This scheme utilizes the error between underwater acoustic channel features to optimize the loss function of ESRGAN,improves the system’s anti-noise ability.Simulation results show that the proposed method has better channel estimation accuracy compared with existing methods.Secondly,according to the characteristics of underwater acoustic communication,this thesis designs an underwater acoustic OFDM using autoencoder based on Convolutional Neural Network(CNN).In this scheme,the coding layer of autoencoder is regarded as the transmitter and the decoding layer as the receiver.CNN is used to reduce training parameters,accelerate convergence.Autoencoder realizes the functions of modulation and demodulation.Then,bidirectional LSTM is added into the structure of CNN-based autoencoder with the help of the idea of convolutional codes.The bidirectional LSTM introduces time correlation for the transmitted information.Autoencoder realizes the functions of channel coding and channel decoding.Simulation results show that this scheme reduces the influence of the poor channel state on the performance of underwater acoustic communication,and improves the reliability of the system.Finally,this thesis combines the proposed the underwater acoustic OFDM channel estimation scheme based on ESRGAN with underwater acoustic OFDM communication scheme based on autoencoder,jointly optimizes the entire communication process in an end-to-end manner,and realizes a deep learning-based underwater acoustic OFDM transmission scheme.This scheme ensures the optimal overall communication performance.The neural network is used to complete the process of channel coding,modulation,channel estimation,demodulation and channel decoding.Simulation results show that this scheme has better performance than the underwater acoustic OFDM physical layer transmission scheme which optimizes each individual module separately.