The artificial neural networks application in direction finding

<jats:p>Modern methods of digital signal processing provide super resolution under conditions of high noise levels and multipath propagation, but all these methods require significant computational resources. There are models that propose using an artificial neural network for a rough estimation of the bearing, followed by its refinement. The paper describes a developed and implemented laboratory prototype for measuring the azimuth angle to a sound source. The device uses an artificial neural network to estimate the bearing. An array of four microphones is used to capture acoustic signals. A microcontroller with built in software handles preliminary data processing and control of peripheral devices.A convolutional neural network with one dimensional layers and eight output neurons performs the acoustic signal analysis. The network was trained on a dataset of 8364 signals and validated on 2 092 signals. The developed prototype has confirmed the feasibility of using neural networks to solve the bearing estimation problem. The prediction accuracy for the azimuth angle is greater than 0.88.</jats:p>