Direction of Arrival Estimation with Uniform Linear Array based on Recurrent Neural Network

This paper proposes the direction of arrival (DOA) estimation of an acoustic source based on a recurrent neural network (RNN) with a uniform linear array (ULA). The RNN has been trained using a large number of independent realizations of microphones signal of the ULA. The authors have compared the DOA estimation accuracy of RNN model against the standard algorithm i.e. delay and sum (DAS) beamformer and it has been observed that RNN based DOA estimation outperforms the DAS beamformer.