A Novel Three-Dimensional Direction-of-Arrival Estimation Approach Using a Deep Convolutional Neural Network

This article aims to constitute a noteworthy contribution to the domain of direction-of-arrival (DoA) estimation through the application of deep learning algorithms. We approach the DoA estimation challenge as a binary classification task, employing a novel grid in the output layer and a deep convolutional neural network (DCNN) as the classifier. The input of the DCNN is the correlation matrix of signals received by a $4 \times 4$ uniformly spaced patch antenna array. The proposed model's performance is evaluated based on its capacity to predict angles of arrival from any direction in a three-dimensional space, encompassing azimuth angles within the interval $[0^\circ, 360^\circ)$ and polar angles within $[0^\circ, 60^\circ ]$ . We aim to optimize the utilization of spatial information and create a robust, precise, and efficient DoA estimator. To address this, we conduct comprehensive testing in diverse scenarios, encompassing the simultaneous reception of multiple signals across a wide range of signal-to-noise ratio values. Both mean absolute error and root mean squared error are calculated to assess the performance of the DCNN. Rigorous comparison with conventional and state-of-the-art endeavors emphasizes the proposed model's efficacy.