A Robust Two-Dimensional DOA Estimation Approach Based on Convolutional Attention Network

The direction of arrival (DOA) estimation of the signal is an important task in radio signal positioning. Various methods have been investigated to cope with the DOA task. However, since the imperfect interference factors are often present in practical antenna arrays, the performance of DOA estimation is often significantly degraded. Besides, few methods deal with the DOA estimation for signals of multiple frequencies. In this paper, we consider the problem of two-dimensional DOA estimation in the presence of imperfect factors, and propose a novel approach where the convolutional attention network is used for DOA estimation. The frequency information is introduced as a token added to the network, which improves the network robustness while taking into account the case of the signal of multiple frequencies. Besides, we extend the mean square error (MSE) to the design of a new loss function for training to improve the accuracy of the model. The advantages of the proposed DOA estimation scheme are demonstrated through numerical experiments.