Phase error immunized DOA estimation using deep learning enhanced self-calibration

Direction of arrival (DOA) estimation is a fundamental problem in array signal processing with wide applications in wireless communications. However, practical sensor arrays inevitably suffer from phase errors that significantly degrade the performance of conventional DOA estimation algorithms. This paper first proposes a deep learning (DL)-based DOA estimation framework, which achieves phase-error-immunized DOA estimation by integrating signal feature extraction and the fully-connected deep neural network (called PIFNN). The PIFNN is a lightweight and robust solution for coarse DOA estimates. In order to further improve the accuracy, we propose PIFNN-WF, which uses the above coarse DOA estimates as initial values and refines them using the WF self-calibration algorithm. The proposed PIFNN-WF method achieves an accurate DOA estimation as well as immunity to array phase error. Simulation results demonstrate that both PIFNN and PIFNN-WF methods perform regardless of array phase errors, where the PIFNN dramatically reduces the runtime. On the other hand, the PIFNN-WF method maintains superior estimation accuracy across different phase errors, achieves faster execution speed compared to conventional self-calibration methods, and exhibits better generalization capability compared to other deep learning algorithms.