Array DOA Estimation Based on Complex-valued Convolutional Network

Aiming at the problem that the existing deep learning methods are difficult to effectively use the complex-valued phase information of array echo signals, a uniform linear array direction of arrival (DOA) estimation method based on complex-valued convolutional network is proposed in this paper to improve the accuracy of DOA estimation and enhance the adaptability of multi-source parameter estimation under the condition of low signal-to-noise ratio. This method uses the Hermitian characteristics of the actual array output signal covariance matrix, takes its upper triangular data as the input of the complex-valued network, takes the corresponding noiseless data as the label, learns the upper triangle of the signal ideal covariance matrix, and then reconstructs the ideal matrix combined with its Hermitian and Toeplitz characteristics. Finally, subspace algorithm is used to estimate DOA. Simulation results show that compared with the traditional subspace class and real-valued convolutional network algorithm, this algorithm has higher estimation accuracy under low signal-to-noise ratio.