Data Augmentation for DOA Estimation Under Limited Snapshots Scenarios

This paper applies data augmentation techniques for direction-of-arrival (DOA) estimation under conditions of limited snapshots. A novel data augmentation method, the Information Extraction-Generative Hybrid Architecture (IE-GHA), is proposed to effectively exploit the source direction information inherently present in snapshots. This method constructs a hybrid architecture combining Variational Autoencoder (VAE) and Generative Adversarial Network (GAN), integrating their respective strengths in feature learning and data generation. The prior information of the array manifold is introduced at the decoder input, while a reinforcement learning-based reward mechanism is designed at the output, using the deviation between generated data and theoretical phase differences as the reward signal. Simulation results indicate that IE-GHA obviously increases DOA estimation accuracy compared to the three existed methods. Furthermore, experiments using real-world data further validate the effectiveness of the proposed method, indicating superior estimation performance in practical scenarios.