DeepDOA: A Novel Deep Learning-Based Method for DOA Superresolution in a Single Snapshot

In this letter, a deep learning-based framework for direction of arrival (DOA) estimation is developed. We first show that the spatial spectrum representation can be formulated as a convex problem. Then, a deep feedforward neural network called DeepDOA is trained to optimize the program. In contrast to traditional spectrum estimation algorithms with computationally complex iterations or matrix inversion, the proposed deep learning-based framework could efficiently obtain DOA estimates in near real-time. Moreover, the improvement in spatial resolution is converted to constructing the expected smooth spectrum with part of the output vector compared to generating the spatial spectrum directly. Simulation results have demonstrated the superiority of the proposed method in both DOA estimation precision and computation efficiency.