Combining U-Net Auto-encoder and MUSIC Algorithm for Improving DOA Estimation Accuracy under Defects of Antenna Array

Abstract

Direction of arrival (DOA) estimation plays a crucial role in radio signal surveillance and reconnaissance systems because it provides spatial information to localize radiated signal sources. Conventional DOA estimation algorithms, such as multiple signal classification (MUSIC) and estimation of signal parameters via rotational invariant technique (ESPRIT), are very sensitive to defects of antenna arrays that reduce the accuracy of estimated DOA in real applications. To mitigate this issue, an auto-encoder based on U-Net is proposed to transfer the imperfect covariance matrix to a new one; then, the MUSIC algorithm is applied to the new covariance matrix to estimate the DOAs of incoming signals. The proposed approach is investigated through simulation for a uniform linear array of eight elements with an inter-element space of half-wavelength. The simulation results indicate that our proposed method achieves a good performance in terms of DOA estimation accuracy. In comparison, the proposed model has outperformed the other models, such as conventional MUSIC, ESPRIT, and two other deep neural networks.