Deep Learning for DOA Estimation Using a Vector Hydrophone

Source azimuth can be estimated via the complex sound intensity method based on a single vector hydrophone, which exploits the physical properties of acoustic pressure and particle velocity components. Deep learning has been successfully used to estimate source depth and distance via training and prediction; herein it is adapted to estimate the source azimuth by analyzing the sound fields provided by a single vector hydrophone. This study compares the estimation results of the two abovementioned approaches that applied to the data due to a moving ship circling around a moored vector hydrophone. Preliminary results demonstrate that these two approaches generally agree; however their accuracies are unsatisfactory because of a low signal-to-noise ratio and strong interference. To increase the estimation accuracy, herein, cross-spectral processing of input data is used for deep learning. The proposed deep-learning method with appropriate input data can effectively suppress interference and accurately estimate source azimuth.