Multi-task recognition of modulation types and arrival directions of underwater acoustic signals based on convolutional neural networks

Abstract Nowadays, underwater acoustic communication reconnaissance technology is of great importance in military, marine science, and resource exploration fields. The fields involve military reconnaissance and intelligence collection, ocean monitoring and security, marine scientific research, and resource exploration. This article mainly focuses on the identification of modulation types and emission angles contained in underwater acoustic signals. Convolutional Neural Networks (CNNs) are used for deep learning to train, validate, and obtain final results on signal data. This network achieves the independent tasks of automatic modulation classification and Direction of Arrival (DOA) estimation of radio signals simultaneously. The model achieves efficient identification and filtering by designing different filter sizes and sizes, as well as different Y-shaped connection structures. It includes one input block and two output blocks, and the two output blocks correspond to the outputs of modulation classification and DOA estimation. At the same time, based on this, the parameters of the network layer are modified to achieve the generalization ability of the neural network. The training and validation data of the network consists of simulation data from underwater acoustic communication emissions and real data from the Songhua Lake experiment in Jilin Province, China, to test the network recognition efficiency.