A Gridless DOA Estimation Method Based on the Spatial Partitioning and Deep-Learning Network

Existing deep-learning-based DOA estimation methods predominantly rely on grid searches over the entire spatial domain, which often suffer from low estimation accuracy. To reduce complexity and improve the performance, this paper proposes a novel DOA estimation method based on the spatial partitioning. By dividing the whole spatial domain into multiple small-scale sectors, a convolutional neural network model is designed to implement both the source detection and covariance matrix reconstruction. Thus, the DOA estimation is transformed into a task of source detection and reconstruction of the covariance matrix within localized spatial sectors. The proposed method can also achieve continuous DOA estimation for slowly moving sources. Experimental results demonstrate that the proposed method achieves higher estimation accuracy with lower model complexity compared to existing deep learning-based methods.