Two-Stage Learning Model-Based Angle Diversity Method for Underwater Acoustic Array

Abstract

The diversity combining technique performs well in the inhibition of multipath, it has been widely used in underwater acoustic (UWA) array signal processing. However, the underwater noise can seriously affect the processing results of the diversity. The conventional filtering algorithms cannot deal with the nonlinear components of underwater radiation noise and have a poor processing effect on complex signals. This study proposes a novel underwater array angle diversity method based on a two-stage model to overcome the problem. A noise-reduction model with a deep convolutional neural network (DCNN) as the backbone network for deep residual learning by preprocessing complex-type data on the received and reference noise signals in the first stage. In the second stage, a modified weighted delay summation beamformer group model is constructed. This model adjusts the weights of each channel by a gradient descent criterion. The desired angle estimates and delay information are then obtained. Finally, the delayed combining of the signals of each path is completed by the combining strategy. Simulation test results show that the proposed algorithm has a lower bit error rate (BER) for diverse received signals. On-lake tests further verify the effectiveness of the algorithm.