【作者】 范梅梅；

【导师】 李磊； 黎湘；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2012， 博士

【摘要】 认知雷达通过自适应、动态地调整发射信号使其与目标及环境相适应，因而可以大幅提升目标识别性能。在认知雷达研究的诸多问题中，自适应波形设计是认知雷达提高目标识别性能的关键问题。论文首先针对现有最优波形设计方法目标函数不完善、不统一的问题，分别研究将待识别目标冲激响应建模为确定性信号和随机信号模型的最优波形设计方法，建立不同准则目标函数之间的区别与联系；针对目前对最优波形特征随环境变化规律认识不足、难以对各波形性能分析与评估提供依据的问题，研究目标与杂波、噪声之间的关系对最优波形的影响。然后针对面向多目标、动目标的自适应波形设计技术空缺的现状，研究面向多目标和动目标场景的自适应波形设计方法。最后针对固定调制方式调制参数可变的宽带信号波形研究信号参数与目标识别性能之间的关系。论文主要内容概括如下：第一章阐述了论文的研究背景及意义，总结了认知雷达系统的关键技术，论述了针对目标识别的自适应波形设计技术研究的意义，分析了该技术的研究现状及存在的问题，最后介绍了本文的主要工作与组织结构。第二章研究基于确定性信号检测理论的最优波形设计方法。首先，分析了雷达目标识别问题与信号检测、最优波形设计之间的联系，建立了目标识别最优波形设计的确定性信号模型。然后，分别基于Neyman-Pearson准则和最小错误概率准则推导了噪声环境下最优波形设计的目标函数，提出了杂波环境下，以检测性能最优为准则的波形设计方法，克服了传统发射-接收联合最优化技术迭代寻优难以收敛到全局最优值的缺点；在上述研究基础上，将噪声和杂波环境下针对确定性目标识别的波形设计目标函数统一到确定性信号检测理论的框架内。最后，通过仿真实验分析不同环境下最优波形的特点，研究目标能量谱密度与噪声和杂波功率谱密度间的关系对最优波形的影响，并深入分析产生该影响的原因。第三章研究基于随机信号估计方法的最优波形设计问题。首先，建立了随机目标的信号模型，将待识别目标冲激响应视为具有一定先验知识的随机信号。在此基础上，研究了基于线性贝叶斯理论和基于信息论的两类波形最优化设计方法，提出了基于局部SNR最大准则的目标函数，充实了基于线性贝叶斯理论的最优波形设计方法；然后，给出了杂波环境下基于线性最小均方误差估计（LMMSE）、局部SNR和互信息准则的目标函数，并针对噪声和杂波环境，建立了各准则下目标函数的相互关系，揭示了各目标函数的联系，并分析了不同准则最优波形特点及其与环境适应的能力。之后，通过仿真实验研究目标与环境的相互关系对最优波形的影响。论文第二章和第三章针对不同类型（确定、随机）目标识别的波形最优化技术进行了梳理和统一，分析讨论了不同环境下最优波形的特点，为后续最优波形设计问题中目标函数的选择和波形性能的评估奠定了理论基础。第四章针对现有波形自适应方法仅面向单个目标、不适用于多目标识别的问题，研究针对多目标识别的波形自适应技术。首先，建立多目标识别的信号模型，定义了识别性能评判准则。在此基础上，提出了以各目标与观测信号间的互信息线性加权和（WLS-MI）为目标函数的最优波形设计方法，实现对目标数目较少时的最优波形设计。针对基于WLS-MI的方法求解最优波形效率低的问题，从目标冲激响应角度，提出了基于多目标冲激响应线性加权求和（WLS-TIR）及基于多目标自相关矩阵线性加权求和（WLS-ACM）两种最优波形设计方法；从发射波形角度，提出了对各目标的差异最大波形线性加权求和（WLS-ST-D）、对各目标的互信息最大波形加权求和（WLS-ST-MI）以及对各假设的差异最大波形加权求和（WLS-SH-D）三种波形设计方法，以上方法克服了目标数目较多时基于WLS-MI的方法无法获得最优波形的解析表示、需要从高次方程多个根中搜索选择确定最优波形的困难。在以上六种基于线性加权求和最优波形设计算法的基础上，提出了一种权值计算方法及波形自适应方法。最后，通过仿真实验，分析比较了本章所提的各种最优波形设计方法相比于传统固定发射的宽带信号的性能改善，验证了本章所提方法的有效性。第五章针对现有目标识别波形自适应方法仅面向静态目标、对动目标识别的性能下降及其初始误判问题，研究针对动目标识别的波形自适应技术。首先，针对目标运动方向与与雷达视线不一致和一致两种场景分别分析了目标运动对现有波形自适应方法识别性能的影响。然后，针对目标运动方向与雷达视线不一致的场景，提出了利用最小二乘支持向量机预测目标姿态并用于更新最优波形的波形自适应方法；针对目标运动方向与雷达视线一致的场景，提出了基于决策层融合的波形自适应方法。最后，通过仿真实验验证了本章提出的两种方法的有效性，并研究了基于融合的波形自适应方法相对于仅利用宽带信息的波形自适应方法的识别性能改善因子随信号能量的变化关系。第六章针对调制方式不变调制参数可变的波形参数选择问题，研究波形参数与识别性能之间的关系。首先，阐述了适用于具有复杂结构及运动特性目标的信号模型，给出了面向目标识别的序贯假设检验模型。然后，在噪声环境下，以平均观测脉冲次数作为评判目标识别性能的标准，分别针对各次观测独立同分布和指数相关两种情况，推导了平均观测脉冲次数和发射信号的关系，建立了目标函数关于发射信号的表达式。之后，基于杂波环境统计特性，以Kullback-Leibler信息数作为评判标准，推导了Kullback-Leibler信息数与发射信号的关系，建立了杂波环境下目标函数关于发射信号的表达式。最后，针对特定目标，通过仿真实验分析了宽带信号的带宽、载频、脉宽等参数与识别性能的关系。本章研究从理论上为波形参数的选择提供了依据。第七章总结了论文的主要工作和创新点，对下一步的研究进行了展望。更多还原

【Abstract】 Cognitive radar can improve target recognition performance dramatically byadaptively and dynamically optimizing the waveforms according to the current targetand environment. Adaptive waveform design is a critical problem in the research ofcognitive reader. This dissertation analyzes the relation of the object functions for bothdeterministic signal and stochastic signal to overcome the lack of a general descriptionof the present object functions. Then, we address the impact of the relation of targetand environment to the waveform behavior for its importance in performanceevaluation. After that we study the problem of adaptive waveform design for multipletargets and moving target. At last, we investigate the relationship of recognitionperformance and waveform parameters with fixed modulation but variableparameter.The main scientific contributions of this dissertation are summarized asfollows:In Chapter One, the background and significance of this research is introduced, andits key technologies are summarized. After that, the importance and scientific value aswell as the technology development of the adaptive waveform design for targetrecognition are expatiated. At last, the main contributions of this dissertation aresummed up.In Chapter Two, the optimal waveform design methods based on deterministicsignals detection theory are addressed. First, the relation of radar target recognitionbetween signal detection and waveform design is analyzed, and the model fordeterminant target is introduced. Then, the object function for known determinanttarget in noise based on Neyman-Pearson criterion and minimum probability of errorcriterion are given; A new method for target recognition in clutter based on thedetection performance under Neyman-Pearson theory is proposed, which can obtainthe analytical result of the optimal waveform, and overcomes the disadvantage oftraditional transmit-receive optimization methods, which is neither guaranteed toconverge nor to produce the optimal signal; then object functions for knowndeterminant target recognition in the present articles are analyzed and added into thesystem of object function based on detection theory. At the last part of this chapter, thebehavior of the optimal waveforms is analyzed by simulations, and the impact ofrelation between target and environment on the waveform behavior and its causes arestudied intensively.In Chapter Three, the optimal waveform design methods based on stochasticsignals estimation are addressed. First of all, the stochastic signal model is present,which treats the target to be recognized as a stochastic process. After that, the optimalwaveform design methods based on linear Bayesian estimation theory and information theory are discussed, and a new object function of local SNR is proposed which enrichthe methods of optimal waveform based on linear Bayesian estimation theory. Based onthe results of waveform design in noise, the object functions in clutter based on LinearMinimum Mean Square Error (LMMSE), local SNR and Mutual Information (MI) aregiven. Then the relationship of the object functions both in noise and clutter areestablished, and the behavior of the optimal waveforms is analyzed. Then the impacts ofenvironment on the waveforms obtained by different criteria are studied by large mountof simulations. At last, the method of waveform synthesis is introduced simply. Thischapter and Chapter Two make an intensive collection and analysis on the presentwaveform design methods, and together with the analysis on the waveform behavioraccording to the environment, they will provide a valuable theoretic and technicalsupport for the selection of object function and performance evaluation of optimalwaveforms.In Chapter Four, adaptive waveform design methods for multiple targetsrecognition are studied on the realization that the adaptive waveform design method forsingle target is unsuitable for multiple targets. First of all, the signal model of multipletargets is present, and the measure of recognition goodness is defined, then the problemsof waveform for multiple targets recognition are pointed out. After that, six kinds ofwaveform design methods are proposed from different stages of signal processing. Thefirst one is based on the Weighted Linear Sum of MI (WLS-MI) between theobservation and each target impulse response, which performs well when target numberis less than3. Then to overcome the difficulty of WLS-MI when dealing with moretargets, two methods are proposed based on WLS of Target Impulse Response(WLS-TIR) and the method based on WLS of Autocorrelation Matrix (WLS-ACM)founded in the processing of target impulse responses. Besides, three methods are putforward founded in the processing of optimal waveform, two of which are obtained byWLS of the optimal waveforms for each target, and are called WLS-ST-D andWLS-ST-MI, respectively, according to the kind of criteria of optimal waveform whichare difference maximum and MI maximum, and the left one is to obtain the waveformfor multiple targets by WLS of the optimal signal for each hypothesis based ondifference maximum (WLS-SH-D). Then the algorithm of weight calculation isresearched, based on which an adaptation mechanism for multiple targets recognition isproposed. The simulation results show the performance improvement of the proposedwaveform design methods and the adaptation mechanism relative to the traditionalwaveform.In Chapter Five, the adaptive methods for moving target are researched for thepoor performance of existing methods aims at static target when used for moving target.Two moving scenarios are considered. The first one is the assumption that the target ismoving in a direction different from the radar line of sight, and the second one is the assumption that the target moves parallel to the radar line of sight. The impact of targetmotion on the performance of adaptive waveform is analyzed intensively. After that,two methods of waveform adaptation are proposed aims at the two scenarios. For thefirst scenario, an adaptive waveform method based on the aspect prediction by LeastSquare Support Vector Machine (LSSVM) is proposed; and to improve the recognitionperformance in the second scenario, a method based on information fusion on decisionlevel is proposed, which makes use of the recognition results of both wideband andnarrow band signals. The simulations verify the validity of the proposed method.Besides, the recognition performance improvements of the sensor fusion methodrelative to the wideband signal only method are compared under different energyrestriction. The conclusion about the impact of target to noise ratio on optimalwaveform obtained in the third chapter and the theory of sequential hypothesis test areused to explain the recognition improvements variations.In Chapter Six, the relationship of recognition performance and waveformparameters with fixed modulation but variable parameter is studied. The signal modelwhich is suitable for target with complex structure and moving behavior is depicted, andthe sequential hypothesis test model is present. The average number of observations isdefined as measure of recognition performance in noise, and the relationship of averagenumber of observations for noise of IID (Independent and Identically Distributed) andexponential correlations is studied respectively, then the object functions about thetransmitted signal for both cases are present. After that, KLIN (Kullback-Leiblerinformation numbers) is defined as measure of recognition performance in clutter, andthe relationship of KLIN and transmitted signals are establish based on the analysis ofthe statistical feature of the clutter, and the object function about the transmitted signalis depicted. At the last part, for some given targets, the relationship of the recognitionperformance and parameters (bandwidth, carry frequency and pulse width) fortraditional signals are analyzed by simulation. The research of this chapter providestheoretical references for parameter selection for recognition of given target.In Chapter Seven, the main contributions and innovative work of this dissertationare concluded. And the potential problems and future work to be further researched arepointed out.更多还原