Research On Learning-Based Space Time Adaptive Processing

Space time adaptive processing(STAP)is a kind of clutter suppression techniques that use simultaneously spatial and temporal information to detect relatively slow moving targets.However,it is difficult to satisfy the condition that the traditional STAP method requires a lot of secondary data,when it estimates the interference covariance matrix.What is more,after estimating the covariance,we should reverse it,and the process will cause a large amount of computation,which greatly improves the requirement for the hardware devices.In order to solve the problem,the researchers use the reduced rank STAP to reduce the number of secondary data and the computation.However,it should be noted that no single algorithm is superior under all circumstances.For example,while reduced rank STAP algorithms might enhance the performance of STAP with limited secondary data,they can also result in loss in adaptivity in one of the spatial and temporal dimensions.Having this in mind,under the condition that the secondary data is limited,this thesis has done the following work:1.Four signal models including the moving target,clutter,jamming,noise are studied.Besides,for the traditional STAP,the experimental results show that the method can correctly detect the moving targets under the Reed-Mallett-Brennan(RMB)rule,and when the secondary data does not meet the rule,the method cannot correctly detect the moving targets.2.The basic principle of sparse Bayesian learning(SBL)is studied.According to the clutter sparseness in the spatial frequency and normalized Doppler frequency,we combine the SBL with the STAP.The method can correctly detect the moving targets with a small amount secondary data.In addition,because of the group sparsity of the clutter profile in the angle-Doppler domain across nearby range cells,A STAP method based on group SBL(GBSL)is proposed.The experimental results show that the proposed method can correctly detect the moving target with a lower computation and a fewer secondary data.3.The traditional STAP methods based on pattern classification are studied,including least square method,binomial method.Moreover,a STAP method based on Adaboost is proposed.After theoretical analysis and simulation experiments,it is found that the STAP method based on least square require higher energy of the moving target,and the STAP method based on binomial is relatively low.For the STAP method based on Adaboost,it use the least square model as the weak learning model,which makes the least square method can detect correctly moving targets,and its output performance is better than the STAP method based on binomial.At last,the STAP method based on classification is a compromise between the STAP method based on GSBL and traditional STAP method in terms of the number of secondary data.