Robust Sparseness Space-time Adaptive Processing Algorithms For Airborne Radar

Ground motive target detection(GMTD)is one of the major tasks of airborne radar.Since the airborne radar is down-looking and platform is motive,the received signal contains strong clutter which has extended Doppler frequencies,and the motive target is always completely covered by clutter.Space-time adaptive processing(STAP)can obtain the statistical information of clutter from training data adaptively.Hence,clutter can be suppressed and the output signal-to-clutter-plus-noise(SCNRout)is improved,which is helpful to the target detection.Thus,STAP method has been applied in GMTD of airborne radar extensively.The traditional STAP method requires a great deal of independent and identically distributed(IID)training samples to estimate the clutter distribution,and has high computational load.However,there are many non-ideal conditions in the real scenarios,such as radar parameter errors,array geometry structure errors and fast varying clutter environments.Hence,there are insufficiency training samples to obtain the desired STAP performance.Especially,when the training samples are few,non-homogeneous or non-stationary,performance of the traditional STAP method degrades severely,and the target cannot be detected at worst.This thesis devotes to the STAP method when the training samples are few,nonhomogeneous or non-stationary,which always lead to the actual clutter information estimation error and target self-nulling phenomenon.The proposed algorithms exploit the low rank property of clutter subspace and the sparse distribution of clutter spectrum in the angle-Doppler plane.The STAP methods based on sparseness for airborne radar are studied,which can improve the clutter suppression performance,motive target detection performance and robustness of STAP in the complex electromagnetic environments.The major problems of traditional STAP method for airborne radar are introduced firstly,and then the characteristics and shortcomings of existing STAP methods are summarized.The sparseness of space-time signal is analyzed intuitively and theoretically.Next,the STAP algorithms based on sparse filter,knowledge-aided sparse recovery of clutter,jointly sparse recovery of multiple snapshots and S transformation are proposed,respectively.A brief summary about the major algorithms is given as follows.(1)The sparse filter algorithms are discussed firstly,which can speed the convergence rate and reduce the computational burden.The generalized sidelobe canceler(GSC)architecture is discussed firstly.Sparseness of GSC filter is analyzed,and a sparse constraint is further imposed to the minimum variance criterion.Hence,a novel cost function is obtained.The recursive least squares(RLS)and least mean square(LMS)methods are used to update the filter weight,respectively,which can reduce computational load.Since the l1-norm constraint is imposed in the novel cost function,the filter weight needs a regularization parameter.The traditional sparse filter always sets the regularization parameter as a constant,but performance of filter is not optimum.To cope with this problem,a novel method which can obtain the most suitable regularization parameter adaptively is proposed.The optimal regularization parameter is calculated via the minimum variance criterion in each iteration,which can improve the convergence performance and SCNRout.Furthermore,a novel sparse filter for the direct form processor(DFP)architecture is proposed.A sparse constraint is imposed to the conventional linearly constrained minimum variance criterion.For solving this novel optimization problem,a RLS algorithm is proposed.This algorithm obtains the minimum output power by joint iterative updating on regularization parameter and filter weight adaptively.The theoretical analysis validate that the adaptive regularization parameter can improve the SCNRout performance.(2)A robust STAP algorithm based on knowledge-aided sparse recovery is proposed.STAP for airborne radar employs training snapshots to estimate a clutter covariance matrix(CCM).However,the resulting estimated covariance matrix can be corrupted by an outlier,i.e.target-like signal,which leads to a further target self-nulling phenomenon.When these outliers are dense,STAP performance degradation is most severe.To cope with this problem,a novel robust STAP algorithm based on knowledge-aided sparse recovery is proposed,which can eliminate the influence of dense outliers on target detection.This algorithm exploits the property that the clutter components of side-looking airborne radar are distributed along the clutter ridge,which is used to distinguish clutter components and outliers.Each snapshot is decomposed by the sparse recovery method,and then the similar degree function is defined to recognize and select clutter components via a threshold.Subsequently,the CCM is estimated by the select clutter components.Therefore,this algorithm can select appropriate coefficients and space-time steering vectors to assess clutter accurately.Furthermore,since the clutter distribution is range-dependent,the non-side-looking array radar cannot obtain clutter statistical information by using the snapshots from adjacent range cells directly.Hence,a direct data domain(D3)algorithm based on knowledge-aided is proposed.The traditional D3 method must know the prior knowledge about target,which is not accurate in practice.To cope with this problem,the proposed algorithm decomposes the snapshot under detection via the sparse recovery.And then,the clutter components are eliminated from the coefficient vector via the prior knowledge about clutter distribution.The remainder elements in the coefficient vector are used to judge whether the target is present or not.Hence,this algorithm can separate the target from clutter directly,improving the target detection performance.The space-time sub aperture smoothing is not required,so the proposed algorithm can be applied in any array radar in theory.(3)Focused on the target self-nulling phenomenon caused by the outliers in training snapshots,a novel robust STAP algorithm based on joint sparse recovery is proposed.This algorithm is applied in side-looking airborne radar.When the sparse recovery is high resolution,the algorithm exploits the characteristic that distribution and correlation between clutter and outliers are different among multiple snapshots.The l2-norm of all coefficient vectors of sparse recovery is employed to highlight the positions of clutter components,which is helpful to select the most suitable sparse recovery coefficients to estimate the clutter spectrum.And then,Gaussian weight is exploited to suppress the influence of outliers.Hence,outliers are eliminated in the CCM estimation,and the target self-nulling phenomenon can be overcome efficiently.Moreover,an iterative adaptive approach(IAA)with sparse constraint is proposed for STAP of airborne radar.The conventional IAA method has weak slow target detection performance and high computational burden.To cope with these shortcomings,we impose the sparse constraint to the weighted least-squares model.The novel expression of space-time spectrum estimate is derived,and adaptive regularization parameter is discussed.The proposed algorithm exploits the sparseness of space-time spectrum to reduce the computational burden via a threshold.Since the fixed dictionary always causes mismatch between the real space-time steering vectors of clutter and the given vectors in dictionary,this algorithm utilizes dictionary self-calibration to correct the space-time steering vectors of clutter.Hence,the clutter spectrum estimation is more accurate,and convergence rate is improved.(4)A robust STAP algorithm based on S transformation is proposed.When the training snapshots contain outliers but the prior knowledge about clutter is not known,outliers may cause target self-nulling phenomenon.Hence,outliers in training snapshots must be eliminated before the CCM estimation.Generally,each snapshot contains clutter which is random distribution with some law,while outliers only appear in some snapshots.Hence,S transformation spectrum distribution of outliers is different from that of clutter in the time-frequency domain.The signals from all range cells in every array element and coherent pulse comprise fast time sequences,which are expanded in time-frequency matrixes by S transformation.According to the theoretical derivation,the time-frequency distribution of clutter is random completely,while that of outliers is isolate in time domain and concentrate in high frequency domain.Therefore,the frequency bands in which outliers are concentrate are cut off from the S transformation matrixes.The CCM is estimated by the remainder parts of S transformation matrixes.This process can restrain the influence of outliers on the adaptive filter weight vector,so the proposed algorithm is robust for outliers.The STAP algorithms proposed in this dissertation,which based on sparse filter,knowledge-aided sparse recovery,jointly sparse recovery of multiple snapshots and S transformation,are advantageous in target detection.These algorithms can be applied in the non-homogeneous and few snapshots scenarios,which are useful in theory and application.