Direction of Arrival Estimation by Using Deep Autoencoder Network
Ekin Nurbaş, Emrah Onat · 2020
In this paper, an unique and innovative method which uses problem deep ecoder networks is proposed for the solution of arrival angle estimation problem. Also details of the proposed method are explained. Although there are various direction finding methods in the literatüre, no mature studies performing sirection finding using an autoencoder have been found. an encoder have been found. In the study shown in [1], the autoencoders designed for different spatial regions were used to filter the received signal into spatial subregions before the deep neural network making direction prediction. In the proposed study, direction estimation is made only by using deep ecoder networks over the reconstruction errors that result from encoding and recoding of signals reaching multiple antenna arrays. For every azimuth direction in the region of interest, there exists a unique otuencoder which has been trained by signals with constant SNR. Proposed direction finding method that looks at the reconstruction errors produced by these autoencoders for the impinging signal has been compared with the currently known and widely used MUSIC algorithm and the performances of both algorithms have been examined in detail.