Research On Underwater Acoustic Multi-Carrier Communication Technology Based On Deep Neural Network

Underwater acoustic communication plays an important role in oceanographic data acquisition,environmental monitoring,and offshore petroleum exploration,etc.In comparison to the wireless channel,the underwater acoustic channel is one of the most complex channels that have the characteristics of severe multipath effect,Doppler frequency shift,and large volumes of ambient noise.With the development of deep neural network technology,utilizing the deep neural network in underwater acoustic communication has become a feasible solution.Autoencoder(AE)is a deep neural network(DNN)model whose structure is similar to that of the communication system.To improve the system’s performance,an autoencoder is attempted to apply to the underwater acoustic communication system.Aiming at the characteristics of large transmission delay and serious Doppler spread of underwater acoustic channel,an asymmetric autoencoder is proposed.At the transmitter,the autoencoder has relatively fewer hidden layers,thereby effectively avoiding overfitting caused by local optimization during the training process.At the receiver,it has a relatively larger number of hidden layers and neurons,which can more accurately match the complex underwater acoustic channels.Moreover,to solve the problem of severe multipath,an Attention-asymmetric AE model is designed.The use of the Attention model can efficiently sift through a large amount of information,thereby significantly reducing the number of underwater acoustic channels.Therefore,only the channel with the largest diameter is retained,while smaller ones are filtered out,so as to greatly reduce the influence of the multipath effect on the received signal.Simulation and the lake experiments demonstrate that the Attention-asymmetric AE-based underwater acoustic communication system can still achieve a performance close to the training level when the testing environments significantly differ from the training environments.And the performance is better than the autoencoder system based on the symmetric structure of the literature,which verified the Attentionasymmetric AE-based underwater acoustic communication system has high environmental adaptability and strong generalization ability.