【作者】 陆满君；

【导师】 司锡才；

【作者基本信息】 哈尔滨工程大学 ， 通信与信息系统， 2010， 博士

【摘要】 通信辐射源个体识别是通信侦察领域新的研究方向,其定义为对接收到的通信电台发射信号进行特征测量,并根据已有的先验信息确定产生信号的通信电台个体的过程。通信辐射源个体识别根据各电台硬件差异在发射信号上表现出来的区别于其它个体的无意调制特征来判断信号来自哪个电台,进而实现电台跟踪和确认。通信辐射源个体识别在无线电安全通信、军事通信对抗和民用无线电监测等领域具有十分重要的意义,受到国内外相关研究人员的关注。针对通信辐射源个体识别研究中迫切需要解决的关键问题,本文提出和推导了一系列具有理论和实用价值的算法,并且通过实际电台数据和计算机仿真实验验证了算法的优良性能。归纳起来,本文所做的工作主要包括以下几个方面：1、在通信信号参数估计方面,从瞬时频率的概率密度函数、信噪比估计和调制样式识别三个方面进行分析。瞬时频率的概率密度函数方面,研究了载波信号瞬时频率的概率密度函数模型,在分析相位误差以及相位误差近似函数的基础上,提出了一种瞬时频率概率密度的近似密度函数,该密度函数表达式具有很好的近似效果,且该公式的表示形式简单,公式中的参数也具有明确的物理含义。并且通过假设检验的方法验证了该公式的正确性。随后分析了不同采样速率及信噪比对瞬时频率概率密度函数的影响。最后通过蒙特卡罗仿真验证了本文提出的瞬时频率的概率密度函数的正确性。在通信信号信噪比估计方面,分析了PSK信号和QAM信号的高阶统计矩与信噪比之间的关系,分析了不同统计矩对信噪比估计性能的差异,为信噪比估计选取最佳的统计量提供一种简单的方法。分析表明对于PSK信号,一阶二阶统计矩的信噪比估计方法具有最佳的信噪比估计性能；而对于QAM信号高阶统计矩具有较好的估计性能,并且推导了具有解析式的高阶统计量的信噪比估计方法,仿真结果表明该方法具有较小的均方误差和较低的复杂度。在调制样式识别方面,本文从线性统计矩和方向数据统计矩两方面分别’对常见的数字通信信号的调制样式识别进行了分析研究。线性统计矩方面利用四阶累积不变量特征和信号的谱峰特征相结合的方法来实现调制样式的自动分类。在方向数据统计矩方面,根据瞬时频率和瞬时相位的概率密度函数来提取信号的三角矩特征,实验结果表明方向数据统计方法能够实现调制样式的自动识别,该方法由于需要计算余弦函数值,因此计算量较大,在实际使用中需要借助迭代等快速算法来实现。2、在暂态特征分析方面,从三个不同的角度进行了个体特征分析研究。首先根据包络信号的非线性特征提出了一种积分包络的暂态信号特征提取方法。对于包络相似的暂态信号经积分包络变换后,信号的非线性特征就会体现在信号的积分包络上,通过PCA提取信号的主分量特征就可以对信号进行分类,实验结果表明该方面具有较好的分类效果和抗噪能力。其次从小波分析方法入手利用离散小波变换提取不含调制信息的暂态包络特性,用每一层小波变换的能量作为特征,用遗传算法来选择分类效果最好的几个特征,对实验采集的暂态信号进行识别。最后进一步研究暂态信号的非线性特征提出了一种基于分类的自适应时频分析方法。该方法通过分类信号的模糊函数自适应的选择高斯径向核函数的参数,使信号的时频分布具有最大的可分离度,该方法在暂态信号识别上效果最好,同时该方法在训练时所需的时间也是最长的,通过梯度上升迭代算法可以降低训练过程中的计算量,减少训练时间。3、在稳态特征分析方面,通过分析码元持续期间信号瞬时频率的特征,提取了能够有效区分通信电台的个体特征。为衡量所选特征的分类能力,根据评价特征集分离性能的指标-可分离指数,择优选择相关特征作为电台的个体特征,为后续分析提供了一个分析工具。根据码元保持时间内信号瞬时频率的特征,提出了一种基于分类的分数阶Fourier变换方法,通过联合优化分数阶Fourier变换阶数和距离测度的方法来选择最优的分数阶Fourier变换阶数和距离测度。4、在跳频网台信号分选方面,分析了跳频网台细微特征的来源,其中主要分析了跳频网台频率合成器的相位噪声,相位噪声是影响短期频率稳定度的主要原因。在跳频电台频率跳变时刻提取信号的暂态特征,由于跳频信号的周期性使得暂态信号的获取变得容易,而通过观测多个跳频脉冲的暂态信号特征也使暂态特征分析更加可靠,因此利用暂态信号分析来进行跳频电台的个体识别具有较好效果,同时通过跳频电台的个体识别的方法来进行网台信号分选。更多还原

【Abstract】 Communication emitter identification is a new issue in the field of communication reconnaissance in recent years, which is defined as designating the unique transmitter of a given signal, using only external feature measurements, by comparing those features with a library of clusters and selecting the cluster that best matches the feature measurements. With the features reflected on signal by the difference of the transmitter hardware, this issue focuses on seeking the source of the received signal, so the transmitter tracking and confirming can be realized. Communication emitter identification is paid much attention, which has very important significance in secure communication via wireless network, communication countermeasure and radio monitoring.It is urgent to solve the key problems of communication emitter identification. A series of theoretically and practically valuable algorithms are proposed, and the good performances of them are verified by simulation experiments with real data. The main contributions of this dissertation are concluded as follows:1. The communication parameter estimation is researched from three aspect: probability density distribution, SNR (signal to noise ratio) estimation, modulation recognition. In aspect of the probability density distribution problem, a probability distribution model of instantaneous frequency is researched for carrier signal. Based on analysis of the phase error distribution and approximate distribution, a approximate distribution functions are proposed, which has a good approximation effect, a simple form, and distinct physical meaning of parameters. The influence of the probability density distribution is analyzed in the case of different sampling rate or different SNR. Simulation results verify the probability distribution model of instantaneous frequency.In aspect of the SNR estimation, the relationship between SNR and high-order moment is analyzed for PSK and QAM signals. And a simple method is proposed to select the optimum moments by analyzing the SNR estimation performance with respect to different moments. The analysis shows the first-and second-order moments SNR estimator with the best performance for PSK signal, but the high-order moments SNR estimator has better performance for QAM signal. High-order moments SNR estimation method with Analytic expression is derived, which has low MSE (mean square error) and low complexity.In aspect of modulation recognition, linear statistics and directional data statistics are discussed respectively to recognize the modulation types of digital communication signal. For linear statistics, fourth-order cumulant and spectral peak features are used to recognize the modulation types. For directional data statistics, triangular moments are extraction based on the probability density distribution of instantaneous frequency and instantaneous phase. The results show that linear statistics and directional data statistics both can be used to recognize the modulation types. But directional data statistics requires a large amount of calculation for the cosine function, so the method is realized by means of iterative algorithm.2. For the difference reflected on signal by the difference of hardware under non-steady state, the transient characteristics are analyzed from three different angles. First, according to nonlinear characteristics of signal, the feature extraction method based on integral envelope is proposed. The transformation of integral envelope reflects the nonlinear characteristics in the signal envelope, and then the classification features are extracted by PCA (Principle Component Analysis). The results show that the method has good classification and anti-noise performance. Second, wavelet transform is used to extract features from the transmitters. The most discriminatory features are selected form a large number of wavelet transform features by genetic algorithms. Experiment results show that the method achieves good accuracy recognition rate in terms of a little features. Third, classification dependent adaptive time-frequency representation is proposed to research the nonlinear characteristics of transient signal. The method adaptively selects the parameter of Gaussian radial kernel function based on the ambiguity function. The method makes the separation degree maximum of the time-frequency distribution. The experimental results show that the method with the best performance, but the method needs long training time. Gradient ascend iterative algorithm is used to reduce the computational complexity and training time.3. For steady-state characteristics, the individual features are extracted by analyzing the characteristics of instantaneous frequency during the symbol hold time. A separation exponent is proposed to evaluate the separation performance of characteristic set in order to measure the classification ability of selective characteristics. This is a useful tool for selecting the correlative characteristics. And then classification dependent fractional Fourier transform is proposed to select the optimal order of fractional Fourier transform and distance measure by joint optimization the two parameters.4. For the frequency-hopping signal sorting, the frequency synthesizer phase noise of frequency-hopping radio is analyzed for its significant source of fine feature. The phase noise is the main reason of short-term frequency stability. The transient characteristics are extracted at frequency hopping time. it comes easy for the periodicity of frequency hopping signal. And frequency-hopping signal sorting is realized by the individual identification of frequency hopping signal.更多还原