Direction Of Arrival Estimation In Autocorrelation Domain With Deep Learning and Autoregressive Modelling

In this paper, a deep learning model based on autocorrelation domain is proposed for direction of arrival estimation in uniform linear arrays. Firstly, the covariance matrix of received signal is constructed and spatial autocorrelation vector is calculated by averaging the diagonals of covariance matrix. The autoregressive model coefficients for calculated spatial autocorrelation is estimated by proposed deep convolutional neural network and corresponding direction of arrival is calculated. Simulation results demonstrate that the proposed method shows better performance than the classical method where autoregressive model coefficients are calculated with Yule-Walker based approach. Furthermore, the proposed model requires less learnable parameters than neural network models created for direcion of arrival estimation in literature.