【作者】 许丹；

【导师】 周一宇；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2008， 博士

【摘要】 辐射源指纹识别,又称特定辐射源识别(Specific Emitter Identification;SEI),或辐射源个体识别,是指对接收的电磁信号进行特征测量,并根据已有的先验信息确定产生信号的辐射源个体的过程。随着雷达、通信系统参数捷变性能的提高,基于常规参数测量的信号识别技术已经很难满足电子侦察的作战要求,而基于信号无意调制等特征的辐射源指纹识别技术成为当前电子侦察的重要发展方向。从目前辐射源指纹识别的研究来看,主要存在两个方面的问题:一是缺乏指纹的机理研究,使得所采用的指纹很容易受信号有意调制的影响而失效;二是指纹的“可测性”存在不足,往往由于背景噪声、接收机非线性、多径干扰等因素使得分类性能难以达到理想的精度。本文针对上述两个问题展开了系统深入的研究,主要研究内容如下:针对自激振荡类型的发射机,从典型射频振荡器(磁控管)的通用等效电路模型出发,提出了一种模型化的辐射源指纹分析方法。该模型化方法扩展了传统的特征定义,集成信号瞬时频率、瞬时带宽的相关信息,能反映器件的多方面的物理特性。在工作电压存在波纹(波纹可变)时,该方法能弥补传统上升沿、下降沿时延测量方法的局限性,可在脉冲中部提取出特征量。本章接着研究了一种基于粒子仿真的指纹分析方法,并将其用于对磁控管自由振荡和受控振荡的物理过程的模拟,验证了前文的结论。针对主振放大式发射机,提出了一系列利用辐射源末级功放的非线性特性进行指纹识别的方法。(一)在功放激励信号为变功率的单音正弦信号时,提出了用近似的谐波约束特征作为指纹,或由吴文俊方法得到约简的指纹特征。(二)在激励信号为有意调制变化的窄带信号时,提出一种多通道相关指纹识别(Multi-Channel Correlation Fingerprinting;MCCF)方法。该方法首先由功率放大器的泰勒级数模型导出窄带输出信号的载频分量和倍频分量表达式,然后利用两分量的关联性,将载频分量作为放大器激励信号的近似,代入倍频分量的表达式中,用最小二乘方法估计出指纹特征量。在此基础上本文分析了MCCF的指纹特征的可观测条件和估计的CRLB。该方法定义的指纹特征与放大器的级数模型有关,与激励信号的形式无关,因此是发射机固有的。依据本方法对长沙地区的调频广播电台进行了指纹识别实测实验,在倍频分量功率比载频分量小60dB到80dB的典型条件下,对四个电台的发射机进行了有效的分类。(三)在激励信号为波形可变的宽带信号时,提出了一种基于子空间比较的指纹识别方法。给出了两种Volterra系统的MIMO建模方法,基于这两种MIMO模型可得到相应的参数子空间。为了避免Volterra模型存在的维数爆炸问题,直接对参数子空间进行比较来实现功放的指纹分类。对本方法的“独立性”和“可测性”进行了分析。数值仿真实验对本方法进行了验证。针对连续波信号中的功放无意调制识别问题,提出了一种基于交叉关联积分的特定辐射源识别方法CCI-SEI。交叉关联积分算法用于对两个信号的重构矢量作高精度比较。在比较之前,采用重采样的方式使得重构矢量分布更连续,同时给出了一种统计测试方法和理论分析方法用于选择合适的相空间重构参数。该方法借鉴了美国海军研究室T.L.Carroll的工作,因此本章介绍了他提出的相空间差分方法,然后分析了相空间差分方法在低信噪比条件下的不足,并引出了本文的CCI-SEI方法。本文的实验结果验证了CCI-SEI的性能优势。将核主元分析应用到辐射源指纹识别中,并提出了两种核函数。提出的瞬时频率RBF核函数及其高精度估计方法可用于对瞬时频率差别很小的调频信号进行特征分析。本文对估计方法的CRLB进行了分析,解释了性能改善的根本原因。提出的基于Frobenius内积的非对称子空间核函数AF-SKF可用于对脉冲组的特征提取,该核函数的定义、性质和计算方法的推导和证明都在文中给出。同时推导结果也指出:Wolf等人提出的对称子空间核函数WS-SKF实际上是本文AF-SKF的一个特例。在这之后,为了解决KPCA在数据量较大、特征发生漂移等情况下的实用性问题,给出了一种核主元更新的SVDU-KPCA算法,并用数值仿真对该算法进行了验证。提出了一种适用于多径环境下SEI的频域广义观测模型。结合信号分类的假设检验理论,研究了三种鲁棒检验方法:信号相依污染度模型法、随机信号鲁棒检验方法和弱随机信号的渐近鲁棒检验法。其中,前两种方法都是基于Huber的混合模型,适用于多径效应影响高于加性噪声的情况。弱随机信号的渐近鲁棒检验器的推导实际上证明了在多径影响明显小于零均值加性高斯白噪声的情况下,渐近鲁棒检验器退化为直接似然比检验器。更多还原

【Abstract】 Specific Emitter Identification (SEI) refers to designating the unique transmitter of a given signal, using only external feature measurements, by comparing those features with a library. With the improving ability of the parameter agileness in radar and communication systems, signal identification utilizing conventional parameter measurement techniques is difficult to satisfy the demand of ELectronic INTelligence (ELINT) systems. Specific Emitter Identification utilizing the signal’s unintentional modulation then become an important trend of electronic intelligence development.The existing research on SEI has two main problems. One is the absent of research on the mechanism of fingerprints. As a result, the obtained features are easy to fail when the intentional modulation changes. The other problem is the insufficient of the feature measurability. The background noise, nonlinearity of the receiver, multipath effect may all decrease the accuracy of classification.This paper is aiming to solve the above two problems. The content of the dissertation is as follows.Based on the equivalent circuit model of RF self-excitation device, such as magnetron, a SEI method for self-oscillatory transmitter is proposed. This model-based method defines a set of fingerprints, which expand the traditional features, and integrate the relationship of the signal’s instantaneous frequency and instantaneous band. The method can extract features from the middle of the pulse,so it can cover the shortage of the traditional feature extraction methods only using the rising and trailing edge of the pulse. A Particle In Cell (PIC) simulation method for fingerprints analysis is also studied. The fingerprints of the previous model is proved by the simulation of the free and the controlled excitation process.A series of SEI method utilizing the power amplifier(PA)’s nonlinear property are developed. (1)When the driven signal is sine signal with changeable power (in different observation), an approximate harmonic constraint feature is obtained. A more general method based on Wu method is also presented. (2) When the driven signal is narrow band with changeable intentional modulation, a Multi-Channel Correlation Fingerprinting(MCCF) method is proposed. From amplifier’s Taylor series model, the expressions of the Carrier Component (CC) and the Harmonic Component (HC) of the output signal is derived. Then a least square algorithm is deduced by substituting the CC into the HC as an approximation of the driven signal. The observation condition and estimation CRLB of MCCF are provided. The fingerprints of MCCF depend on the Taylor series model, and are independent of the input signals, so they are inherent. The experiment of four FM broadcast emitters in Changsha area shows that, the MCCF method works well, even when the HC is 60dB to 80dB smaller than the CC.(3) When the driven signal is wide band with changeable intentional modulation, a subspace comparison based fingerprinting method is proposed. Two MIMO models of PA’s Volterra series model are given, based on which the parameter subspace can be derived. To avoid the dimension explosion of the Volterra series model, different PA systems is identified by comparing the subspace directly. The independent and measurable properties of the method are given, and the principle of the method is verified by the simulation experiments.A Cross-Correlation Integral SEI method(CCI-SEI) is developed to identify the amplifier’s unintentional modulation of the continuous signal. The CCI algorithm is used to compare the reconstructed vectors of the two signal with high degree of accuracy in the phase space. Before the comparison, a re-sample step is taken to make the reconstructed vectors’ distribution more smoothly. Also a statistical test method and a theoretical analysis method are developed to choose appropriate reconstruct parameters. CCI-SEI borrows the ideas from the work of T.L.Carroll, so Carroll’s Phase Space Difference (PSD) method is introduced, the shortage of PSD method in low SNR is analyzed too.The experiment results show that CCI-SEI has better performance than the PSD method.Studies the utility of Kernel Principle Component Analysis (KPCA) method in SEI application. Two kernel functions are developed to extract fingerprints. The Instantaneous Frequency RBF kernel(IF-RBF) and its estimator can be used in the feature extraction of frequency modulation signals with Tiny Instantaneous Frequency Difference (TIFD). This kernel uses the IFD of the two signals, so the IFD estimator’s CRLB is derived, which can explain the reason of the improvement. Another proposed kernel function is Asymmetric Frobenius Subspace Kernel Function (AF-SKF), which can be used in the feature extraction of pulse groups. The rationality of the definition, the property and the calculation method are given. The derivation also shows that, the symmetric subspace kernel function WS-SKF is just a special case of AF-SKF. After the discussion of kernel function, a SVD Updating based KPCA algorithm (SVDU-KPCA) is derived. The simulation experiment proofs it.A Frequency Domain Generalized Observation Model (FDGOM) for SEI in multipath environment is proposed. Combined with the hypothesis testing theory of signal classification, three robust hypothesis testing methods are studied: Signal Dependent Contamination Parameter (SDCP) method, random signal robust hypothesis testing method and the weak random signal asymptotic robust testing method. The first two methods base on the Huber’s contamination model, and are suit for the case that the interference of multipath is bigger than the interference of noise. The derivation of asymptotic robust testing shows that asymptotic testing method will degenerate into a simple likelihood ratio testing method when the interference of noise is obviously bigger than the multipath.更多还原