Deep Learning-aided Direction-of-Arrival Estimation by Exploiting Difference Coarray

In this paper, a deep learning-aided direction of arrival (DOA) estimation method is proposed. Via the theory of difference coarray, we first transform the array covariance matrix into virtual array-element domain, which formulates an equivalent single-snapshot array observation under a virtual array with much larger aperture. Then, we view the above single-snapshot DOA estimation as a sparse signal reconstruction in noise under the framework of compressive sensing, and construct a deep convolution network to learn the inverse transformation of this sparse spatial spectrum through large training data. Compared with some exist DOA estimation methods, the proposed one can not only achieve highly time-efficient DOA estimation than other compressive sensing-based methods, but also manifest better estimation accuracy than other deep-learning-based methods. Simulation results illustrate the advantages of the proposed method.