【作者】 唐智灵；

【导师】 杨小牛；

【作者基本信息】 西安电子科技大学 ， 信息与通信工程， 2013， 博士

【摘要】 在通信对抗领域，通信侦察的两个基本任务是定位和识别。定位技术作为阵列信号处理的一个主要研究方向，已得到了广泛应用。而对通信辐射源识别的研究则相对薄弱。如果说定位解决了“在哪里”的问题，那么识别要解决“是什么”的问题，两者只有有机结合才能达到有效侦察。随着定位技术的实际应用，通信目标识别的需求也将变得越来越迫切。在通信侦察中运用细微特征分析识别重要的通信辐射源个体目标，掌握使用者的身份和性质，并通过监视跟踪对敌方的战术、战略动向做出预测，有利于在复杂的信息战环境下掌握军事行动的主动权。本论文以通信辐射源稳态信号为研究对象，通过研究信号的预处理、模式分割以及特征提取方法，获取通信信号的个体特征。通过MATLAB软件建立了无意调制信号验证模型以产生仿真信号用于验证模式分割和特征提取算法。并用这些算法从实际采集的调频手持机信号中提取了个体特征量，最后用成熟的ECOC分类识别器对个体特征量进行了分类识别，证明这些算法对通信信号个体识别是有效的。取得的研究成果为：1.对振荡器产生的无意调制进行了研究。根据振荡器噪声的统计特性,在MATLAB中建立了振荡器噪声源模型，仿真分析了单独存在闪烁噪声以及闪烁噪声混合高斯白噪声情况下的时域波形和频域特性。同时还分析了从健伍手持对讲机功放输出的未调射频信号解调恢复的无意调制信号，表明信号的功率谱密度具有幂律的特点，呈现出了闪烁噪声的特点。2.对射频功率放大器产生的无意调制进行了研究。根据射频功放的非线性分析理论，在MATLAB软件中建立了记忆多项式功放模型，并用这个模型仿真了QPSK、WCDMA两种信号的放大，分析计算了因功放引起的信号失真量ACPR（相邻信道功率比）。此外，还对真实射频采样数据的ACPR值进行了计算和分析，结果表明信号带宽越宽，ACPR值越低。从时域上看，射频功率放大器的非线性对信号的影响引起信号幅度以及相位相对于原信号的变化，并且这类变化反映了功率放大器的特征，可以用于通信辐射源的分类识别。3.对信号预处理算法进行了研究。改进了Wornell与Oppenheim提出的噪声分离方法，使其能够用于混合无意调制信号与噪声的分离。首先使用多目标进化优化算法求解非线性方程，得到分形信号的参数估计，然后使用这些估计参数构建分形滤波器，对小波系数进行滤波处理，最后将处理以后的小波系数进行逆变换，得到混合的分形信号。仿真对不同的混合成分、不同信噪比、不同的数据长度的情况进行了分析比较，证明在一定的信噪比条件下，采用足够长度的采样数据进行处理，恢复的混合分形信号具有较小的均方根误差。4.对三种特征提取方法进行了研究。（1）基于多重分形维的特征提取方法：将一维信号变换为二维信号矩阵后得到信号的纹理特性。利用这一特性，通过计算数据阵列的分形维谱而获得了特征矢量。算法仿真表明在提取特征之前进行去噪处理能够获得稳定的信号特征量；（2）基于顺序统计的方法：根据在弱非线性系统中，窄带功率放大器的输入-输出是单调函数的原理，将接收信号做白化处理以后进行顺序统计，通过最小二乘法对顺序统计结果做线性回归而估计出特征参数作为射频信号个体特征；（3）基于高阶累积张量的特征提取方法：推导了接收信号的3阶、4阶累积量与电台个体特征的关系，提出了一种将4阶累积量视为3阶张量、用Kernel PCA方法提取个体特征量的方法。最后用ECOC分类器将这些特征量进行了分类，证明这些方法都是有效的。本文进一步扩展了对通信辐射源个体特征识别的研究，探索和验证了无意调制中提取特征量的新方法。这些方法能用于非合作条件下识别通信辐射源的身份，使识别通信辐射源的手段更为丰富。更多还原

【Abstract】 In the field of communications countermeasure, there are two tasks known as positionand recognition. As a main research direction, position technology been used widely. But theresearch on recognition technology is not enough relatively. Position can solve the problem of“where”, whereas recognition will resolve the problem of “what”. A more effectivecommunications reconnaissance must be the combination of these two technologies. Now theresearch on recognition technology is more urgent because position has been widely applied.The military initiative position would be taken in the complicated electronic warfareenvironment, if an important communications radiation source is recognized with its tinyfeatures and its identity is known.In this paper, pre-processing, pattern separation and feature extraction forcommunications steady signals have been studied. With MATLAB, Models of unintentionalmodulation signals are built to generate simulating signals used to verified algorithms ofpattern separation and feature extraction. Then these algorithms were used to extract thefeatures of10Kenwood FM tow-way radio signals, which were labeled by ECOC multiclassfier. The results showed these algorithms can extract identity feature. The author’s majorcontributions are outlined as follows:1. Unintentional modulation introduced by oscillator has been studied. According to thestatistic characteristic of oscillator noise, a noise model was built with MATLAB. Generationsof not only flicker noise, but also flicker noise plus Gaussian white noise were simulated forresearch on time waveform and frequency spectrum. Then real signals acquired from aKenwood tow-way radio were analysised, which show its spectrum is obeyed power lawwhich indicated there was flicker noise.2. Unintentional modulation induced by RF power amplifier has been studied. Accordingto the nonlinear theory of RF power amplifier, a memory polynomial model was constructedand simulated for QPSK and WCDMA signals in MATLAB. ACPR values of simulating datawere computed. It can be seen that the wider the bandwidth was, the lower the ACPR was.The effects caused by amplifier on signals were variations of amplitude and phase. Becausethese variations carried information of power amplifier, so they can be used to recognizecommunications radiate source.3. Algorithm of pre-processing signals has been studied. The algorithm of noiseseperation proposed by Wornell and Oppenheim was improved for filtering hybridunintentional modulation signals from noise combined signals. Multi objects optimizationalgorithm was used to solve nonlinear equations to acquire parameters of fractal signals. Then these parameters were used to construct a fractal filter. After wavelet coefficients processed bythe fractal filter and were transformed to time domain, white noises were removed fromsignals. Cases of different SNR, data length were simulated that indicated recovered signalshad low RMS error if length of data was enough.4. Three feature extraction methods have been studied.(1) Fractal method changed1-Ddata into2-D to expose texture of signals. Then fractal dimension spectrum of2-D data arraycan be computed and used as feature vectors. The simulation showed that nearly samefeatures can be acquired if data were pre-processed before features were extracted.(2) Orderstatistic method sorted data that have been whited. Then processed data were a line which wasthe monotone function about power amplifier input-output. Features can be extracted withfunction fitting.(3) Research on high order cumulants tensor method derived the relationbetween3,4order cumulants and identity features. By kernel PCA method, features can beextracted from3order tensors transformed from4order cumulants. At last, ECOC classifierwas used to recognize these features. Results declared that these methods can extract featureseffectivly.This paper improved the research on the recogniztion of communications radiate source.Methods of extracting features from steady communications signals were explored andverified. These methods may be used to recognize identity of communications radiate sourcein the non-cooperative communication environment and enrich current available methods.更多还原