DOA Estimation Using Deep Neural Network with Regression

In this paper, we propose a direction-of-arrival (DOA) estimation method, which uses deep neural network (DNN) with regression to learn the mapping from the array covariance matrix to the DOAs of multiple source signals. Simulation results demonstrate that the proposed method is efficient in both training and testing stages over the existing DNN based classification method. In particular, the training time and the testing time spent by the proposed method are about 9 times and 2 times less than those by the DNN based classification methods, respectively. Furthermore, it performs significantly better than the classical multiple signal classification method (MUSIC) in terms of efficiency. On the other hand, the root mean square error (RMSE) of DOA estimates by the proposed method is much higher than that by the MUSIC method and approaches to the CRB in harsh conditions such as low SNRs and/or closely spaced source signals.