Finding Direction of Arrival using Uniform Circular Array of Microphones and Recurrent Neural Network

Estimating a sound source’s direction of arrival (DOA) has become a significant and crucial area of interest. It has gained recognition in various applications, such as telecommunications, robotics, and defense systems. In this paper, we propose estimating the DOA of a sound source using a uniform circular array (UCA) of microphones and a specific type of neural network called a recurrent neural network (RNN). The RNN was trained for multi-class DOA estimation problem, where 1400 noisy signal vector realizations are used for each class. We assessed the accuracy of RNN-based DOA estimation for UCA compared to a uniform linear array (ULA). Additionally, a comparison between UCA and ULA has been made regarding the DOA estimate accuracy of the Delay and Sum (DAS) beamforming technique. We concluded that DOA estimation via UCA, outperforms different techniques for DOA estimation using ULA.