An Array Calibration Method based on Beam Pattern and Neural Network in Underwater USBL System

In underwater localization scenarios, especially for the ultra short baseline (USBL) system, array calibration is essential. The localization performance may decrease rapidly when the array coordinate deviation exists. The main drawbacks of conventional calibration methods are the fact that they usually require a large amount of measurements, so the time-saving calibration methods are demanded in practical application scenarios. Motivated by the fact that neural networks' efficiency in learning nonlinear relationship, we propose the calibration method based on beam pattern and neural network, which can achieve one measurement in calibration procedure. Differential beam pattern model is used to generate the training data. Simultaneously, the array errors are modeled in polar coordinates form, corresponding radius and angle deviations for each array element. The performance of calibration is evaluated based on the estimated accuracy of sound source's direction of arrival (DoA) after calibration. Both the simulation and experiment results have verified our proposed method, ledding to an average 46.5% error reduction in the DoA estimation during the experiments when compared with traditional method.