DOA Estimation Based on Complex-Valued Channel Attention Networks

In complex electromagnetic environments, more accurate estimation of direction can be performed using deep learning techniques. Previous methods have the problem of forcibly decoupling the complex representation of the received signal into independent real components, which leads to the loss of the intrinsic properties of the signal model and thus affects the accuracy. In addition, methods using only convolutional neural networks or deep neural networks cannot make good use of the dual-channel characteristics of signal. To address the above problems, a DOA estimation method based on Complex-Valued Channel Attention Neural (CCAN) network is constructed in this paper. The features of the received covariance matrix are extracted by complex-valued convolution, and the result is obtained after CCAN training. We incorporate the channel attention mechanism, which can be better adapted to the dual-channel complex-valued convolution and the channel characteristics of the signal. The experimental results show that the estimation accuracy of the CCAN method is high in target-unknown and low signal-to-noise (SNR) ratio scenarios.