Robust DOA Estimation Method Based on Self-Supervised Learning and Adversarial Augmentation

Direction-of-arrival (DOA) estimation in complex electromagnetic environments plays a critical role in target detection and localization. However, achieving accurate estimation remains a challenge under adverse and non-ideal conditions. We propose a novel self-supervised learning and adversarial augmentation complex-valued convolutional neural network (SACVCNN) for DOA estimation. The method leverages self-supervised pretraining with adversarial data augmentation to improve the robustness of spatial feature extraction, while the use of complex-valued convolutional layers preserves both amplitude and phase information for more comprehensive signal modeling. Simulation results demonstrate that SACVCNN achieves superior accuracy and robustness compared to model-driven and data-driven methods, especially under low-SNR and limited snapshots.