An Underwater Velocity-Independent DOA Estimation Method Based on Cascaded Neural Network

Abstract

The underwater environment introduces uncertainty into the acoustic velocity, which affects the performance of traditional direction of arrival (DOA) estimation methods. This research proposes a cascaded neural network based underwater DOA estimate approach to address this issue. In this method, the cascade neural network is composed of a velocity regressor and a velocity classifier. To determine the estimated value of acoustic velocity, the velocity classifier first breaks down the input data into distinct velocity domains. It then regulates the velocity regression process. Then, the array steering matrix predicted by the blind source separation algorithm is utilized to determine the angle, and the acoustic velocity is modiffed by the cascaded neural network. Eventually, it is possible to derive the DOA estimation value under the calculated acoustic velocity. The suggested method has a high estimation accuracy especially when the acousitc velocity is unknown, as shown by the simulation results.