深度学习在DOA估计中的应用综述与展望OA

A systematic review is presented in this paper on the application of deep learning in direction-of-arrival(DOA)estima-tion for array signals.Based on the selected literatures on deep learning-based DOA estimation,the model performance is charac-terized by three categories of evaluation metrics:efficiency-based,error-based and classification-based indicators.The comprehen-sive advantages of deep learning models in terms of efficiency,accuracy and robustness are verified by means of quantitative com-oparison.Furthermore,the three key influencing factors,namely the number of sources,signal-to-noise ratio and snapshot num-ber,are focused and their regulatory rules on model performance are deeply analyzed,revealing the variation characteristics of model performance under different scenarios.This review systematically integrates existing research achievements,clearly presents the technical progress and performance differences of deep learning in DOA estimation,and puts forward application suggestions based on the core advantages of novel neural networks and hybrid architectures,which is of great significance for promoting the the-oretical innovation of deep learning methods in DOA estimation under complex scenarios.