Study On Super-resolution DOA Estimation Methods Based On Deep Learning For MIMO Radar

Direction of arrival(DOA)estimation of multiple input and multiple output(MIMO)radar has important research value and application prospect in fields such as autonomous driving and military detection.The performance of existing model-driven DOA estimation methods depends on the match between the pre-assumed mathematical model and the observed signal of the array.With the increasing expansion and complexity of application scenarios,model-driven methods often suffer from mismatch between the model and the signal,leading to a decline in DOA estimation performance.Deep learningbased DOA estimation methods can adaptively improve the structure and parameters according to the data,and have strong generalization ability,which can effectively improve the performance of DOA estimation in complex scenarios.This thesis focuses on the research of DOA estimation based on deep learning,especially solving the performance degradation caused by array imperfections and single snapshot sampling in complex scenarios for MIMO radar.The main contents are as follows:1.The equivalent virtual array mechanism of MIMO radar are researched,and the signal model of MIMO radar after matched filtering at the receiving array is derived.In addition,the effects of array imperfections and number of snapshots on DOA estimation performance for MIMO radar are analyzed.2.A DOA estimation method for MIMO radar based on auto-encoder and convolutional neural network is proposed.By transforming the DOA estimation problem into a multi-classification problem and utilizing the data reconstruction ability of autoencoder and the adaptive feature extraction ability of convolutional neural network,the robustness of DOA estimation in the cases of array mutual coupling,gain and phase inconsistence and sensor position error is improved.3.A DOA estimation method for MIMO radar based on deep unfolded network and auto-encoder is proposed.The DOA estimation problem is transformed into a sparse recovery problem,and a neural network is constructed based on the iterative shrinkage thresholding algorithm(ISTA)to improve the interpretability of the neural network model.Additionally,an auto-encoder is introduced to achieve data-driven denoising,which improves the anti-jamming ability of DOA estimation in the case of single snapshot sampling.All the methods proposed in this thesis have been verified by simulation experiments.The experimental results demonstrate that the proposed methods are superior to the existing model-driven methods in terms of robustness and anti-jamming,effectively improving DOA estimation performance for MIMO radar in the cases of array imperfections and single snapshot sampling.