【作者】 阳召成；

【导师】 黎湘；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2013， 博士

【摘要】 空时自适应处理(Space-Time Adaptive Processing, STAP)技术可实现对机载/星载雷达中强地/海杂波的有效抑制，显著改善机载/星载雷达对空/地运动目标的检测性能。传统STAP方法通常要求有两倍于系统空时自由度的独立同分布的训练样本才能使得最优性能损失低于3dB。并且传统STAP方法的计算复杂度高，需要大量的存储单元。然而在实际应用中，受限于雷达系统参数、阵列几何结构、系统内部非理想因素、变化的杂波环境以及当前器件发展水平和计算能力，上述条件常常无法满足。本文以机载雷达为背景，围绕杂波抑制以及运动目标检测技术展开研究，探索了空时回波信号中杂波的稀疏性本质特性，针对STAP中稀疏性先验，开展了一系列降低独立同分布训练样本数、适合复杂多变环境的稳健STAP方法研究。本文首先阐述了STAP技术面临的难点问题，归纳总结了现有STAP方法和稀疏恢复算法的发展历史和研究现状，探讨了基于稀疏性STAP方法的优势和存在的关键科学问题，为后续研究提供了基础。第二章研究了STAP中存在的稀疏性约束，建立了基于稀疏性的STAP理论框架。一方面分析了杂波空时功率谱稀疏性的内在机理，指出了杂波空时功率谱稀疏度与杂波秩的相互关系；另一方面，从系统自由度和杂波自由度方面分析了空时滤波器中的“稀疏性”。在上述基础上，以稀疏滤波器、空时功率谱稀疏性和阵列流形知识三条主线构建了基于稀疏性的STAP理论框架，并分别推导或介绍了上述理论框架下稀疏滤波器STAP基本原理，基于空时功率谱稀疏恢复STAP基本原理，基于阵列流形知识的STAP基本原理以及基于阵列流形知识和空时功率谱稀疏恢复STAP基本原理。第三章研究了基于稀疏滤波器的STAP技术。考虑到空时滤波器中的“稀疏性”，针对现有自适应迭代空时滤波器收敛速度慢的问题，提出了基于稀疏滤波器的一系列STAP方法，包括：基于L1范数约束的采样矩阵求逆STAP方法，基于L1范数约束的在线序贯下降STAP法，基于L1范数约束的递归最小二乘STAP方法以及基于L1范数约束的传统/改进共轭梯度STAP方法。详细分析了基于L1范数约束空时滤波器算法中涉及的正则化参数设置问题，针对正则化参数设置难的问题，提出了两种基于开关选择机制的自适应正则化参数选择方法。在给定一定先验知识的条件下，这两种算法都能有效实现正则化参数的自适应挑选。第四章研究了基于阵列流形知识的STAP技术。严格推导了基于阵列流形知识的STAP技术的基本原理，为基于阵列流形知识的STAP技术提供了理论基础。针对阵列流形先验知识不确定性问题，提出了基于阵列流形知识和低秩特性的STAP方法和基于阵列流形知识特征分析的STAP方法。针对最小均方方法中求逆运算复杂度高的问题，利用杂波子空间只依赖于杂波空时导向矢量的特点，提出了基于Gram-Schmidt正交化的子空间计算方法。针对基于阵列流形知识和低秩特性的全维STAP方法难以满足实时性处理强的雷达系统，提出了次优的基于阵列流形知识和低秩特性的降维STAP方法。这些算法具有极快的收敛性且对运动平台速度测量误差、偏航角测量误差以及杂波内部运动比较稳健。第五章研究了基于空时功率谱稀疏恢复的多训练样本STAP技术。针对直接进行中值滤波器的非参数化自适应迭代算法计算复杂度高和对弱目标检测难的问题，提出了利用多训练样本的基于自适应软判决门限的非参数化自适应迭代算法。该算法不仅计算复杂度低，而且能实现对更慢速度和更低SNR运动目标的有效检测。针对现有稀疏恢复算法计算复杂度高和正则化参数设置难的问题，提出了基于复数域同伦的空时功率谱稀疏恢复STAP算法。研究表明相比现有复数域同伦算法和FOCUSS算法，该算法参数设置更为简单，计算复杂度也更低。第六章研究了基于空时功率谱稀疏恢复的直接数据域STAP技术。基于加权L1范数方法相比L1范数更能逼近L0范数的事实，提出了直接数据域的基于Capon谱L1范数加权的空时功率谱稀疏恢复STAP方法和基于Fourier谱L1范数加权的空时功率谱稀疏恢复STAP方法。该方法能获得比不加权方法更高的输出SINR。针对直接数据域稀疏恢复STAP只利用单个样本进行空时功率谱估计精度不高的问题，提出了基于空时平滑思想的空时功率谱稀疏恢复STAP方法。该方法利用平滑处理获得了多个空时子快拍样本，由多个空时子快拍样本稀疏恢复的空时功率谱相比单个样本稀疏恢复的空时功率谱估计精度更高，所获得的输出SINR更高。第七章研究了基于阵列流形知识与空时功率谱稀疏恢复的STAP技术。该技术利用阵列流形知识获得了缩小的空时导向字典，提出了采用正交匹配追踪和同伦的基于阵列流形知识与空时功率谱稀疏恢复的STAP方法，通过在缩小的空时导向字典内稀疏恢复杂波的空时功率谱获得了更优的功率谱估计结果和更低的计算复杂度。探讨和分析了基于阵列流形知识和空时功率谱稀疏恢复STAP技术中涉及的字典设计和杂波阵列流形选择问题。第八章对全文进行总结，并指出进一步可能的研究方向。本文研究成果对稀疏表示/稀疏重构/压缩感知等理论发展，对系统辨识、运动平台下杂波抑制、运动目标检测等应用均具有一定的理论指导意义和工程应用价值。更多还原

【Abstract】 Space-timeadaptiveprocessing(STAP)caneffectivelysuppressthestrongground/seaclutter and improve the moving target detection performance for airborne/spaceborneradar systems. Traditional STAP algorithms require about twice the degrees of freedom(DOFs) of the independent and identically distributed (IID) training snapshots to yieldan average performance loss of roughly3dB, have a high computational complexity andneed a lot of memory elements. However, in real applications, because of the limits ofradarsystemparameters,arraygeometrystructure,internalsystemnon-idealeffects,vary-ing clutter environments and the computational capacity, it is usually hard to satisfy theabove conditions. In this dissertation, we focus on the clutter suppression and movingtarget detection in airborne radar applications. By exploiting the intrinsic clutter sparsityof the space-time returns, we design a series of robust STAP algorithms for the complexand varying clutter environments with lower requirements of IID training snapshots.This dissertation first discusses the difficulties in STAP, elaborates the developmenthistory and current work of the STAP algorithms and sparse recovery algorithms, andanalyzes the advantages and key problems of the sparsity-based STAP algorithms, whichprovides a basis for the following researches.The second chapter studies the sparsity in the STAP and builds the theory frameworkfor the sparsity-based STAP algorithms. On one hand, it analyzes the intrinsic sparsityof the space-time power spectrum and discusses the relationship between the clutter spar-sity and clutter rank. On the other hand, it details the sparsity of the space-time filtersfrom the points of the system DOFs and the clutter DOFs. Then it builds the theoryframework for the sparsity-based STAP algorithms based on sparse filters, space-timepower spectrum sparsity and array geometry knowledge, followed by the derivations orintroductionsoftheprinciplesofthesparse-filter-basedSTAPtechniques, thespace-time-power-specturm-sparsity-based STAP techniques, the array geometry knowledge-aided(KA) STAP techniques and the space-time-power-specturm-sparsity-based STAP tech-niques exploiting prior knowledge of array geometry.The third chapter studies the sparse-filter-based STAP techniques. Considering thesparsity in the space-time filters, this chapter proposes a series of L1-regularized STAP al-gorithmstoovercometheslowconvergenceoftheiterativeadaptivespace-timefilters,in- cluding the L1-norm-based sample matrix inversion STAP algorithm, the L1-norm-basedonline coordinate gradient STAP algorithm, the L1-norm-based recursive least-squaresSTAP algorithm and the L1-norm-based conjugate gradient STAP algorithms. It also de-tails the setting of the regularization parameters in the proposed algorithms and developstwo approaches to adaptively select the regularization parameters’ value. Under condi-tions of certain prior knowledge, these two approaches can effectively select the regular-ization parameters’ value.The fourth chapter studies the array geometry KA-STAP techniques. It first strictlyderives the principle of the array geometry KA-STAP techniques which provides a basisfor this technique. For the uncertainty of the array geometry prior knowledge, we pro-pose the KA-STAP algorithm that exploit the low-rank dominant clutter and the arraygeometry properties (LRGP) and the eigenanalysis-based STAP algorithm. For the highcomputational complexity of the least-squares method, we present an efficient subspacecomputationmethodbasedonGram-Schmidtorthogonalizationbyexploitingthefactthatthe clutter subspace is only determined by the space-time steering vectors. For practicalapplications, a reduced-dimension version of the proposed LRGP-KA-STAP algorithmis also developed. It illustrates that a fast convergence and a robustness to the platformvelocity measured errors, yaw angle measured errors and the inner clutter motion of theproposed algorithms are obtained.The fifth chapter studies the space-time-power-specturm-sparsity-based STAP tech-niques with multiple training snapshots. To improve the performance of the recentlydeveloped weighted least-squares-based iterative adaptive approach (IAA) in STAP forweak or slow targets detection, we propose a novel IAA scheme to adaptively suppressthe ground clutter by using the training data. It shows that the proposed IAA scheme out-performs the conventional IAA scheme over weak or slow targets detection but a lowercomputational complexity. Since it is hard to set the regularization parameters of the cur-rentsparserecoveryalgorithms,weproposeanovelspace-time-power-specturm-sparsity-basedSTAPalgorithmbasedonthecomplex-valuedhomotopytechnique. Comparedwiththe current complex homotopy sparse recovery algorithm and the focal under-determinedsystem solution algorithm, the proposed algorithm provides excellent detection perfor-mance with lower computational complexity and easier parameter settings.Thesixthchapterstudiesthedirectdatadomainspace-time-power-specturm-sparsity-based STAP techniques. We first develop novel space-time-power-specturm-sparsity- basedSTAPalgorithmsthatutilizetheCaponspectrumandtheFourierspectrumweightedl1-norm penalty to further enforce the sparsity and approximate the original l0-norm. Thisweighted methods exhibit better performance than the non-weighted methods. Since theestimated clutter covariance matrix is not stable for the current direct data domain space-time-power-specturm-sparsity-based STAP algorithm, we propose a novel direct data do-main space-time-power-specturm-sparsity-based STAP algorithm utilizing subaperture s-moothing techniques. It uses only the snapshot in the cell under test to generate multiplesub-snapshots and jointly recovers the clutter spectrum using all generated multiple sub-snapshots, which can reduce the variance of the estimated clutter spectrum resulting inimproved signal-to-interference-plus-noise-ratio (SINR).The seventh chapter studies the space-time-power-specturm-sparsity-based STAPtechniques exploiting prior knowledge of array geometry. By using the prior knowledgeof array geometry, we propose orthogonal matching pursuit and homotopy KA-STAP al-gorithms which reduce the dimension of the space-time steering dictionary. Because theproposed algorithms recover the clutter power spectrum in a reduced-dimension space-time steering dictionary, it results in a lower computational complexity and a better spec-trum estimate. Furthermore, the details of the selection of potential clutter array manifoldvectors according to prior knowledge are also discussed for the proposed algorithms.Theeighthchaptermakesasummaryofthedissertation,whileseveralopenproblemsfor the sparsity-based STAP algorithms are proposed.In conclusion, the studies and results in this dissertation provide useful theory di-rections and valuable engineering applications for sparse representation/reconstruction,compressive sensing, system identification, clutter suppression and moving target detec-tion.更多还原