Acoustical Direction Finding using a Bayesian Regularized Multilayer Perceptron Artificial Neural Networks on a Tri-Axial Velocity Sensor

Abstract

A two-dimensional direction-of-arrival estimation scheme based on Bayesian-regularized (BR) Multilayer Perceptron (MLP) artificial neural network (ANN) is developed around a unit acoustic vector sensor (AVS). The AVS basically consists of three collocated and orthogonally oriented velocity sensors, hence, senses acoustic waves in the three Cartesian directions while offering portability in size and simplicity in its array manifold. It is shown that the Bayesian regularized Multilayered Perceptron neural network performs well in terms of estimation’s root-mean-square error even when tested with data of different signal-to-noise ratio (SNR) after training. This is useful as it accounts for unexpected changes of received data SNR during field operation. The proposed system is ideal for applications in mobile systems such as robots for search-and-rescue operations or soldiers in the battle field to estimate the source of a sniper fire.