A Robust Deep Learning-Based DOA Super-Resolution Estimation Method With Single Snapshot

At present, deep learning (DL) has attracted widespread attention and has been widely applied in different areas of wireless communication, and signal processing. Due to its excellent feature extraction capability, it has been extensively used in the direction of arrival (DOA) estimation. In this paper, a robust network for super-resolution single-snapshot DOA estimation called DeepDOA is proposed. Different from other network-based DOA algorithms, DeepDOA takes the normalized real and imaginary parts of the received signal as inputs. Meanwhile, the network output is a vector instead of a spatial spectrum, and the corresponding spatial spectrum can be obtained by a simple transformation, which makes the spectrum smoother. Moreover, a sparse constraint is first introduced into the loss function, which improves the accuracy of the algorithm. Numerical simulations are conducted to verify its excellent performance. Finally, a frequency-modulated continuous wave (FMCW) radar location system is built to validate the proposed algorithm.