DOA estimation based on CNN for underwater acoustic array

Abstract The direction of arrival (DOA) estimation of space signals is a basic problems in array signal processing, which is also one of the tasks in many fields such as radar arrays and sonar arrays. In array signal processing, the most commonly used array covariance matrix is a complex matrix. Since traditional neural networks can only deal with real numbers, they cannot handle real and imaginary numbers at the same time, so the input of the neural network is very limited. Based on the application of convolutional neural network (CNN) in RGB three-channel image processing, this paper proposes to use two-channel including real and imaginary covariance matrices as the input signal of the CNN in order to estimate direction of underwater acoustic signal. After modelling and simulation, compared with the traditional MUSIC algorithm, CNN algorithm has higher accuracy and shorter estimation time in small SNR environment. Therefore, the method proposed in this paper can effectively identify the incoming wave direction of the unknown signal in water after training.