【作者】 张志民；

【导师】 皇甫堪；

【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】 在现代信息化战争中，通信信号，特别是数字通信信号，其调制方式的自动识别具有极其重要的意义。信噪比是衡量截获信号质量的重要指标。本文以数字通信信号为主，在前人工作的基础上，结合某调制识别接收机设计的应用需求，对调制识别算法进行研究。首先提出两种信噪比估计算法；然后分别针对窄带、宽带截获信号提出两种调制方式自动识别算法；最后在调制识别接收机中实现这些算法。所做的主要工作包括：1、提出两种信噪比估计算法。第一种算法基于接收信号的高阶矩，这是一种半盲信噪比估计算法。在这种算法中，接收信号被建模成经高斯复噪声污染的确定性信号。通过研究接收信号的模平方序列、模四次方序列的平均各态历经性，推导出信号功率、噪声功率、信噪比估计方差的克拉美-劳限。对非恒包络调制信号，通过按数字调制类型符号周期分段的方法，将问题转化为估计恒包络信号的信噪比。该算法直接对接收机输出的中频信号进行处理，无需知道信号的调制类型和实现载波同步，对符号数、过采样数都具有稳健性。仿真结果表明，在符号数为500，过采样数为100，信噪比在-5～25dB范围内时，算法的平均估计标准差小于0.18dB。第二种算法基于特征分解和子空间概念。该算法无需知道接收信号的调制类型及符号速率等先验知识，不需要实现载波同步，能在较大信噪比范围内对截获信号的信噪比进行精确的估计。计算机仿真表明，当符号数为2500，自相关矩阵维数为50，信噪比范围为5～20dB时，算法的平均估计偏差小于0.1dB。在同等条件下，算法的执行速度快于第一种信噪比估计算法。2、针对窄带调制类型，提出一种基于判决理论的调制识别算法。首先，提出离散随机序列的ZA变换的概念。针对窄带信号，提出一种不需要进行希尔伯特变换的瞬时幅度提取算法。它直接以观测样本为基础，无需进行FFT和IFFT，执行速度快，为实时运行创造了条件。算法对码元同步没有要求，对信噪比、符号个数及载波频率失谐都具有稳健性。仿真表明，在-20～32dB的信噪比范围内，该算法的最大误差不超过0.09。其次，对于基带采用矩形脉冲成型的数字调制类型，提出一种不需要进行去卷绕处理的瞬时相位、瞬时频率提取算法。与传统的去卷绕方法相比，该方法的主要优点是耗时少、计算速度快。仿真表明，当符号数为10000时，使用这种方法提取瞬时相位比用传统方法快约50毫秒，可用于实时系统中。再次，以瞬时幅度、瞬时相位、瞬时频率为基础，借助于序列的ZA变换，提出12个分类特征参数。这些参数的特点在于计算简单，用常规信号处理技术即可提取；且所需先验知识少，噪声抑制能力强，耗时少，可用于实时系统。最后，利用上述12个分类特征参数，提出一种基于判决理论的窄带通信信号调制方式自动识别算法。计算机仿真结果表明，在载波频率已知，数字调制成型脉冲为矩形，信噪比为3dB，样本数等于4×104时，算法的平均识别率超过98%，平均识别时间少于50毫秒。3、针对宽带调制类型，提出一种基于接收信号的谱相干函数和四阶、八阶循环累积量的调制方式识别算法。首先利用半盲信噪比估计算法估计接收信号的信噪比，然后采用最近邻准则对信号分类。在这种识别算法中，利用了谱相干函数对加性噪声不敏感、对信道影响不敏感的优点。算法无需知道载波频率、载波相位和符号速率等先验知识。分别在高斯信道和多径衰落信道的情况下，通过仿真评估了算法的性能。仿真结果表明，对于高斯信道，在信噪比为10dB、样本数为4096时，识别算法的平均识别率高于97%；对于多径衰落信道，在路径数不超过4、信噪比为10dB、样本数为4096时，识别算法的平均识别率均高于90%。4、实现了调制识别接收机。接收机的前端部分采用基于带通采样的宽带中频结构，实现各种调制信号的采集。接收机具有实时、事后处理及离线分析等三种工作模式。利用10种数字、模拟调制类型分别对两种识别算法进行现场实验，证明了它们的有效性。本文提出的通信信号调制方式自动识别算法，己在某新型号通信对抗设备中得到应用，效果良好，为该设备的定型起到了重要作用。理论分析、仿真结果和现场试验均表明：本文提出的诸多新算法，不仅在理论上具有一定的创新性，而且具有重要的实用价值。更多还原

【Abstract】 In modern information-based warfare, automatic modulation recognition ofcommunication signals is extremely important, especially for digital communicationsignals. The signal-to-noise ratio (SNR) is an important index of the quality of theintercepted signals. In order to design a modulation recognition receiver based on theprevious work, the modulation recognition algorithms of communication signals areinvestigated in this thesis, which are mainly on the digital communication signals. TwoSNR estimation algorithms are proposed, and two automatic modulation recognitionalgorithms of narrowband and broad band intercepted signals are given. Furthermore,the proposed algorithms are implemented in the modulation recognition receiver. Themain work of this thesis could be summarized as follows:1. Two SNR estimation algorithms have been proposed.The first algorithm is based on the higher order moments (HOM) of the receivedsignal, which is a semi-blind SNR estimation algorithm. In this algorithm, the receivedsignal is modeled as a deterministic signal corrupted by complex Gaussian noise. Withthe mean-ergodicity properties of the squared and quartic modulus of the sequence, theCramer–Rao bounds (CRBs) of the signal power, noise power as well as the SNRestimation variances are derived. For the non-constant envelope modulated signal, thesignal is required to be divided into a series of successive segments according to thedigital modulation symbol period, resulting in estimating the SNR of a constantenvelope signal. The intermediate frequency (IF) signal at the output of the receiver, isdirectly processed without the knowledge of the modulation type or realizing carriersynchronization. The proposed algorithm is robust to the symbol number and theoversample factor. Simulation result shows that the average estimation standarddeviation is lower than0.18dB (500symbols, oversample factor is100, SNR scopefrom-5dB to25dB).The second algorithm is based on the concept of eigenvector decomposition andsubspace. The proposed approach could accurately estimate a wide range of SNR valuesof the intercepted signal without any a priori knowledge, e.g., the modulation type, thesymbol rate of the received signal, more the carrier synchronization is not required.Computer simulations confirm that the average estimation bias is less than0.1dB (2500symbols, true SNR from5dB to20dB) when the dimension of the autocorrelationmatrix is50. Under the same simulation conditions, the proposed approach performsfaster than the algorithm given above.2. A modulation recognition algorithm based on the decision theory is proposed forthe narrowband modulation types.Firstly, the concept of zero-amplitude (ZA) transformation of a discrete random sequence is given. For narrowband signals, an instantaneous amplitude extractionalgorithm is presented without Hilbert transformation. Directly based on the observedsamples, the proposed algorithm does not need to perform FFT (Fast FourierTransformation, FFT) or IFFT (Inverse Fast Fourier Transformation, IFFT), whichresults in being real time implemented. The symbol synchronization is not necessaryand the algorithm is robust to SNR offset, symbol number and carrier frequencymismatch. Simulation results show that the maximum error is less than0.09when theSNR varies from-20dB to32dB.Secondly, for the digitally modulated types with rectangular baseband pulseshaping, an instantaneous phase and frequency extraction method is proposed whichdoes not need to perform phase unwrapping. Compared to the traditional method whichuses phase unwrapping, this novel approach performs much faster. Simulation indicatesthat under the constraint of10000symbols, the novel method is50milliseconds fasterthan the traditional method, thus it can be implemented in the real time environment.Thirdly, with the knowledge of the instantaneous amplitude, the instantaneousphase, the instantaneous frequency, as well as the ZA transformation of a sequence,12classification feature parameters can be extracted. All of these parameters can becalculated relatively easily by utilizing the conventional signal processing tools. Thisless time-cost procedure requires less a priori knowledge and is powerful in noisesuppression. As a result, this proposed method could also be implemented in real timesystem.Fourthly, by utilizing the12classification feature parameters mentioned above, anautomatic modulation recognition algorithm based on the decision theory is given forthe narrowband communication signals. Computer simulation shows that the averagerecognition rate of the algorithm is over98%(with known carrier frequency andrectangular digital pulse shaping, SNR is3dB,4×104samples). More, the averageclassification time is less than50milliseconds.3. For broad band modulation types, a modulation recognition algorithm ispresented which is based on the spectral coherence function (SCF), the fourth-andeighth-order cyclic cumulants (CCs) of the received signal. First, the semi-blind SNRestimation algorithm is applied so as to evaluate the SNR of the received signal, then aminimum distance criterion is employed to classify the signal. This recognitionalgorithm has the advantage of SCF’s insensitive nature to additive noise and to channeleffects. It does not require a priori knowledge such as carrier frequency, carrier phaseand symbol rate etc. To evaluate its performance, simulations have been performed inGaussian and multipath fading channels, respectively. It demonstrates that the averagerecognition rate of the recognition algorithm in Gaussian channels is above97%(withSNR is10dB and4096samples), and the rate is above90%in multipath fadingchannels with not more than4paths (SNR is10dB and4096samples). 4. A modulation recognition receiver is implemented, the front end of whichemploys broad band IF structure based on band pass sampling to collect data of allkinds of modulated signals. The receiver is operated in three modes, i.e. the real mode,the post processing mode and the off-line analysis mode. In order to evaluate theperformances of the two proposed recognition algorithms, field tests have been carriedout on10digital and analog modulation types. The experiment results have confirmedtheir validity.The proposed automatic modulation recognition algorithms of communicationsignals have been implemented in a novel communication countermeasure equipment.Its excellent performance promotes the finalization of the equipment. Theoreticalanalysis, simulation results and the field tests confirm that the proposed algorithmspresented are not only creative in theory but have great potential.更多还原