Performance Analysis of DOA Estimation of Two Targets Using Deep Learning
Yuya Kase, Toshihiko Nishimura, Takeo Ohgane, Yasutaka Ogawa, Daisuke Kitayama, Yoshihisa Kishiyama · 2019
Direction of arrival (DOA) estimation of wireless signals is demanded in many situations. In addition to classical methods such as MUSIC and ESPRIT, non-linear algorithms such as compressed sensing has been very common recently. Deep learning or machine learning is also known as a non-linear algorithm and now applied to various fields. Generally, DOA estimation using deep learning is classified as on-grid estimation. Thus, the accuracy may be degraded when the DOA is on the boundary. In this paper, the performance of DOA estimation using deep learning is compared with one of MUSIC which is off grid estimation. The simulation results show that deep learning based estimation performs less well than MUSIC due to the grid boundary problem. When the allowable estimation error is relaxed, however, it is found that the success rate of DOA estimation surpass one of MUSIC.