DOA Estimation Based on CNN in L-type Mutual Matrix

Direction of arrival (DOA) estimation of space signal is a basic problem in array signal processing, and it is also one of the important tasks in radar, sonar and many other fields. Traditional algorithms such as MUSIC have poor real-time performance and low accuracy, and the complexity of 2D-MUSIC is greatly increased. Therefore, convolutional neural network(CNN) is introduced to DOA estimation in recent ten years, and the weights of CNN are shared, which greatly reduces the training complexity. Therefore, CNN is used as the basic framework of DOA estimation. In order to improve the performance of CNN DOA estimation, the following two improved methods are proposed. (1) Sparse array can reduce the cost and system complexity. To solve the problem of angle ambiguity in sparse array, coprime array would be combined with CNN. (2) In order to solve the problem that CNN feature information is not fully used, Attention mechanism is introduced in CNN for DOA estimation, which establishes Attention-CNN algorithm, and different weights are assigned to each channel. Simulation results show that the DOA estimation based on Attention-CNN can improve the estimation accuracy and reduce the computational complexity, which performs better than MUSIC and other traditional algorithms.