Deep MVDR: A High Resolution DoA Estimation Method against Array Gain-Phase Error

Direction of Arrival (DoA) estimation is a critical issue in array signal processing, and several classical methods such as Minimum Variance Distortionless Response (MVDR) and Multiple Signal Classification (MUSIC), have been proposed to solve this problem and have achieved remarkable results. However, these DoA estimation methods suffer performance degradation under model mismatches such as gain-phase errors. Deep learning (DL) methods have strong capability to learn the mapping from received data to true angles directly, achieving notable performance in model mismatches scenarios, but their performance tends to degrade in different SNRs or snapshots scenarios. To address these issues, a high-resolution DoA estimation method named Deep MVDR is proposed in this paper, which consists of covariance-enhanced autoencoder, DoA estimator and super resolution modules. These modules can utilize the feature extraction capabilities of neural networks to improve performance and refine spatial spectrum. Simulation results demonstrate the correctness and effectiveness of the proposed method across various target numbers, SNRs and snapshots with gain-phase errors, and show that Deep MVDR outperforms other existing methods in terms of spatial resolution.