A gridless DOA estimation algorithm based on unsupervised deep learning

A gridless direction-of-arrival (DOA) estimation algorithm with unsupervised deep learning is proposed to eliminate the influence of the spatial angle discretization on the estimation results and the deep neural network's reliance on labels in the training dataset. This algorithm is inspired by the atomic norm minimization (ANM) algorithm and unsupervised deep learning, and a loss function is designed to solve the ANM problem in the unsupervised deep learning framework. The proposed algorithm can address the grid mismatch issue and be more robust to the training dataset, since it no longer depends on the spatial angle discrete grid and labels in the training dataset. Simulation results show that the proposed algorithm is state of the art when compared with the other advanced algorithms currently in use.(c) 2022 Elsevier Inc. All rights reserved.