STAP Technique For Airborne Radar Target Detection Based On Bayesian Sparse Variational Autoencoding

Target detection of airborne radar is an important branch of airborne radar task.Due to the large intensity and wide distribution of clutter caused by the down-looking environment of airborne radar,and the space-time coupling of clutter,it is difficult to suppress the clutter.space-time adaptive processing(STAP)technology provides an effective way to suppress airborne radar clutter.The filtering performance of STAP depends on whether the estimation of the clutter plus noise covariance matrix of the cell under test is accurate or not,and the training samples used to estimate the clutter plus noise covariance matrix are at least twice the degree of freedom of the system in theory.However,because of the complexity of the actual clutter scene,the clutter is non-homogeneous or non-stationary,which makes the training sample seriously insufficient.In order to solve the degraded performance of STAP due to the insufficient training samples,a series of improved methods have appeared.Among them,the sparse recovery STAP algorithm based on probability model has the advantages of no need to set hyperparameters and weak sensitivity to atomic correlation.The filtering performance under insufficient training samples is better than other STAP algorithms.The sparse Bayesian learning STAP algorithm has the best performance under non-homogeneous scene,but it still has many problems,such as poor robustness for clutter environment with high noise power,slow convergence speed and high computational complexity.In the first chapter,the research background,development history and current situation of STAP technology are introduced.The sparse recovery STAP technology is introduced by analyzing the small sample problem in the non-homogeneous clutter environment of airborne radar.In the second chapter,the STAP signal model,the solution process of STAP optimal weight vector and the filtering process are described in detail.On this basis,the theoretical basis and implementation mode of sparse STAP are introduced,and the existing sparse recovery-based STAP methods are listed.At the end of the second chapter,the evaluation criteria of STAP filtering performance and detection performance are given.The third chapter mainly introduces the generation model of the STAP method based on Bayesian Sparse Variational Auto-Encoding network,By analyzing the existing sparse recovery STAP algorithm based on probability model,the modeling idea of Bayesian hierarchical model is expounded.Then,the model construction and parameter updating formula under Single Measurement Vector and Multiple Measurement Vector,as well as the whole filter detection process are introduced.Finally,the robustness of Bayesian hierarchical sparse model in noisy clutter environment and the performance advantages in non-homogeneous environment are analyzed through experiments.In the fourth chapter,the encoder of the STAP method based on Bayesian sparse variable auto-coding is introduced.Firstly,by analyzing the problems of slow convergence and high computational complexity of the existing STAP method and the theoretical basis and implementation form of the variational auto-encoding network,the modeling idea of the Bayesian Sparse Variational Auto-Encoding network is described.Secondly,the network structure,sampling process and loss function of Bayesian sparse Variational Auto-Encoding are introduced in detail.Finally,the advantages of Bayesian Sparse Variational Auto-Encoding network in convergence speed and computational complexity are compared and analyzed by experiments.The robustness of Bayesian Sparse Variational Auto-Encoding model on different data sets is analyzed by comparing the detection performance and speed performance of Mountaintop and MCARM data sets.In chapter 5,the main contents of this paper are summarized,and the existing problems such as mesh mismatch are analyzed.In the fifth chapter,we summarize the main contents of this paper and look forward to the future work.