【作者】 范西昆；

【导师】 王永良；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2006， 博士

【摘要】 机载雷达空时二维自适应处理(STAP)算法,以其优良的杂波和干扰抑制性能引起雷达界的广泛重视,三十多年来一直是雷达信号处理研究领域的一个热点问题。本文结合实际工程背景,研究了适合工程实现的STAP算法,以及算法的实时实现问题。为了进一步优化现有部分自适应STAP滤波器的结构,深入研究了局域杂波自由度问题。分析了降维变换对干扰协方差矩阵杂波自由度的影响。给出了局域构成与局域杂波自由度之间的函数表达式,并予以证明。通过这一理论成果,我们从局域构成方式就可以知道局域杂波自由度,进而就可以得出局域STAP处理器所需的可调控权数。分析得出现有的降维STAP处理器,特别是只做时域(空域)一维域变换的STAP处理器,存在进一步降低系统自由度的可行性。分别利用仿真数据和实测数据验证了理论研究成果的正确性。针对现有局域STAP处理器的不足,提出了一种对杂波自由度和干扰环境变化更加稳健的局域STAP处理器—FDSP (Flexible DOF STAP Processor)。该处理器根据外部干扰环境的变化情况,依据准则自适应地调整系统自由度。它不仅降低了非主杂波区杂波抑制的计算量,并且可以有效抑制有源干扰。给出了该处理器的实现方法和权值递推求解的快速算法。利用实测数据证明了该处理器的有效性。通过我国某新型机载相控阵雷达的实测数据,比较研究了几种数据独立的样本选取方法的性能,得出分段的方法在这类方法中具有优势。并对分段方法中的最优的分段数目、段内的样本选取方式以及对消形式等问题进行了研究。比较了分段方法和功率选取方法的性能,得出功率选取法更能够发掘输入数据中的信息,对于孤立的强杂波点的消除具有明显的优势。针对功率选取法中强目标信号自相消效应明显这一不足,提出了改进的功率选取法,该方法通过功率-相位联合选取样本,避免了在训练样本集中包含含有目标的样本,明显地去除了强目标的自相消效应。用实测数据比较研究了两种工程中常用的空时二维权值计算算法的性能。实验结果表明这两种算法均具有良好的杂波、干扰抑制性能,传统的SMI算法在杂波抑制性能上较QRD-SMI算法有一定的优势,而QRD-SMI算法对“目标消除效应”较SMI算法有更强的稳健性。综合比较算法性能、数值特性以及可并行实现性,QRD-SMI更适合在STAP的工程实现时采用。针对多DSP并行处理系统,研究了机载雷达空时自适应处理算法的实时实现。在分析了部分自适应STAP算法内在并行性的基础上,提出了一种任务级的STAP并行处理算法。给出了该算法的任务划分、执行模型、任务映射策略以及性能评价函数。该算法在不同的计算阶段之间进行数据重映射。通过数据实验证明这种基于多DSP并行处理系统的STAP并行算法具有较高的实时性能。更多还原

【Abstract】 The Space-Time Adaptive Processing (STAP) algorithm that is well-known in the area of multi-channel Airborne Early Warning (AEW) radar has outstanding performance of clutter and jamming suppression. The practical STAP algorithms and its real time implementation problems are researched in this dissertation.To optimize existing reduced dimension STAP processor, local clutter DOFs (degree of freedom) problem is further researched. The impact of reduced dimension transformation on interference covariance matrix is analyzed. The functional relationship between the constitution of local processing domain and local clutter DOFs is presented. By this research finding, the number of adaptive weights can be known from the number of local clutter DOFs which can be calculated from the constitution of local processing domain. Then, the conclusion can be drawn that the system DOFs of current reduced dimension STAP processor can be feasibly reduced more by reduced-rank methods. The validity of the research is proved by simulated data and measurement data processing results.A robust localized space-time adaptive processor, which is called FDSP (Flexible DOF STAP Processor), is presented. The processor adjusts system DOFs adaptively with variable interference environment. Compared with the fixed DOF STAP processor, FDSP reduces large computational complexity and improves the performance of clutter and jamming suppression. The implementation of FDSP is illustrated. The validity of FDSP is proved by measurement data processing results.The data independent sample selection strategies are surveyed by measurement data of Chinese new type AEW radar, which indicates that range segment method have the best results. Then, the comparative study on PST (Power Selected Training) and range segment method is given, which shows that PST has better performance on strong clutter discretes rejection because it exploits more information inherent in the data. Many times, however, targets are also present whose inclusion in the STAP weight training results in significant target self-cancellation as well as degradation in clutter mitigation performance. An ameliorated PST training method is presented that excises targets from the training set based on a phase measurement for each potential STAP training sample. The resulting training method based on both phase and power selection critera is shown to offer significant performance gains on experimental data.The test results of two practical STAP weights computational algorithms are presented, which shows that both algorithms have satisfactory performance on resisting clutter and jamming, and traditional SMI algorithm has better performance on clutter rejection, but QRD-SMI is more robust to the effect of target self-nulling. Considering algorithm performance, numerical characteristic and parallelism, QRD-SMI is more suitable for real time implementation of STAP.STAP algorithms require very high computing power and hardly implements. A detail analysis of computation steps of partially-adaptive STAP algorithm is present, which indicates that there is a natural, inherent parallelism in STAP algorithm. A parallel algorithm based on multi-DSP system of partially-adaptive STAP is presented. The execution model, task mapping strategy and performance evaluation functions of this algorithm is also illustrated. Data remapping is used between successive computation steps of STAP. The effectiveness of the implementation is demonstrated with experimental results.更多还原