Robust Direction-of-Arrival Estimation via Sparse Representation and Deep Residual Convolutional Network for Co-Prime Arrays

Co-prime arrays can achieve higher degrees-of-freedom (DOF) with far fewer sensors. State-of-the-art mod-el-driven DOA estimation methods for co-prime arrays face challenges in practical applications because of pre-assumption dependencies. In this paper, we propose a spatial spectrum recovery method based on deep residual convolutional network (DRCN) for effective DOA estimation with a co-prime array. First, a pseudo spectrum is constructed via the observation vector and the extended array manifold matrix of the virtual array. Then, a deep learning framework with residual blocks is proposed to directly learn the mapping from the pseudo spectrum to the super resolution spectrum. The learning-based method enhances the generalization of untrained scenarios and robustness to non-ideal conditions, e.g., small angle separations, small snapshots, low SNRs and imperfect arrays, which makes up for the defects of previous model-driven methods. Simulations are carried out to validate the superior performance of the proposed method, particularly when the deviation of the array manifold matrix is significant.