Direction of Arrival Estimation Applied to Antenna Arrays using Convolutional Neural Networks

In this paper, an effort is made to solve the direction of arrival (DoA) estimation problem by constructing a convolutional neural network (CNN) architecture, which estimates the angles of arrival of the incoming source signals received by a uniform linear array (ULA) antenna. The input of the CNN is the sampled correlation matrix of the signals, while the the output is a pool of the highest probabilities of the network's estimated values. The problem is modeled as a multi-label classification task, meaning that the space of angles is divided into a grid of multiple classes. To model the problem in this way, we assume that we cannot have two or more signals coming from the same angle. This also allows us to further increase the quality of our predictions, meaning that we can set an a priori minimum distance between each given output. In this way we can filter out duplicate outputs and have the desired result.