DOA Estimation in MIMO Radars via Deep Learning

The Direction of Arrival (DOA) estimation is an active research area in array signal processing. Conventional DOA estimation methods require high computational complexity for the multiple-input and multiple-output (MIMO) radars which require the use virtual data vector. In addition, while most conventional methods perform well in high signal-to-noise ratio (SNR) environments, the results in low SNR conditions are not satisfactory. To address these problems, this paper introduces an architecture composed of denoising convolutional autoencoders (DCAE) and convolutional neural networks (CNN) named as DCAE-CNN architecture. The DCAE is used to restore the data prior to DOA estimation, and CNN is employed to estimate the angle of arrival by mapping the restored data to the corresponding angles. Compared to the conventional MUSIC algorithm, experimental results of the proposed DCAE-CNN scheme demonstrate more satisfactory performance in terms of accuracy in low SNR circumstances and reduce the computation time considerably which makes it's use possible for in real-time applications.