Accuracy Improvement in DOA Estimation with Deep Learning

Direction of arrival (DOA) estimation wireless signals is demanded in many applications. In addition to classical methods such as MUSIC and ESPRIT, non-linear algorithms compressed sensing have become common subjects study recently. Deep learning or machine also known a algorithm has been applied various fields. Generally, DOA using deep classified on-grid estimation. A major problem that the accuracy may be degraded when near boundary. To reduce errors, we propose method combining two DNNs whose grids are offset by one half grid size. Simulation results show our proposal outperforms which typical off-grid method. Furthermore, it shown DNN specially trained for close case achieves very high compared with MUSIC.