Direction of Arrival Estimation Using Hybrid CNN-GRU

This paper presents a novel hybrid deep-learning network for accurate direction of arrival (DOA) estimation. The proposed method leverages the columns of the covariance matrix to construct a model-assisted deep-learning framework, integrating the strengths of the Gated Recurrent Units (GRU) and the Convolutional Neural Networks (CNN). This hybrid architecture utilizes selected elements from the received signal's covariance matrix as input features, while the corresponding angles serve as labels for training. Numerical results demonstrate that the proposed method outperforms existing deep learning approaches, showcasing superior performance across various conditions.