Enhanced angle-of-arrival estimation via convolutional neural network-based MUSIC algorithm

Direction of Arrival (DoA) is an important concept in radio communications, and the MUSIC algorithm is an important algorithm for solving the DoA problems. Some studies have shown that reducing the noise in the signal matrix can increase the accuracy of MUSIC algorithm. This paper proposes a deep learning framework to that enables this work. Simulations show that the proposed framework has superior estimation accuracy when SNR is low or targets are close, and the RMSE is estimated to be reduced by up to 30 degrees.