Residual Neural Network with Attention Mechanism for Direction-of-Arrival Estimation in Low SNR

In this paper, we propose an improved residual neural network (ResNet) based on an attention mechanism for estimating the direction of arrival (DOA) of linear arrays. The real, imaginary, and phase components of the covariance matrix are used as inputs to an auto-encoder (AE), whose output is then introduced into a cascaded ResNet for DOA estimation. To enhance estimation accuracy, an attention mechanism is integrated into the ResNet, effectively assigning greater weight to the most relevant features. Experimental results demonstrate the proposed method's superior DOA estimation performance compared to existing methods, especially in low SNR.