A Deep Network Solution for DOA Estimation of Adjacent Signals with Uncertain Number

This paper introduces a deep neural network algorithm to estimate the direction of arrival (DOA) of nearby signals under unknown source number. Firstly, the covariance matrix is processed by a multi-channel convolutional network to extract the features of each potential signal, and then the on-grid estimates of independent components are obtained. Subsequently, by combining these extracted features and the spectral peak angles from each channel, a deep fully connected network calculates the continuous DOA estimates, meanwhile the number of signals is estimated simultaneously. Finally, the simulations have been carried out to validate the performance of the proposed algorithm with respect to generalization ability in DOA estimation. Moreover, the proposed algorithm could also offer superior resolution than that of the classic MUSIC algorithm, especially for the low signal-to-noise ratio, few snapshots, and adjacent signals.