【作者】 雷恒恒；

【导师】 赵拥军；

【作者基本信息】 解放军信息工程大学 ， 军事情报学， 2010， 硕士

【摘要】 随着新体制雷达的迅猛发展,空间电磁环境的日益复杂,雷达信号脉内调制方式识别和特征分析变得愈加困难。本文首先以小波包变换和瞬时频率分析为基础,研究了两种雷达信号调制方式识别的算法,然后从雷达信号调制参数和雷达信号包络特征两个方面,研究了雷达信号脉内特征分析的相关算法。主要工作如下:1.研究了基于小波包的雷达信号脉内调制方式识别算法。首先深入研究小波包变换的基本理论,利用最优小波基技术,提取了含有信息量大的小波包重构系数。针对小波包重构系数维数过高和如何构造识别特征的问题,给出了两种特征构造的方法:第一、计算小波包重构系数的能量,以能量的统计特征为识别特征;第二、对小波包重构系数进行奇异值分解,并取前6个大的奇异值做为识别特征。详细介绍了BP网络模型的参数设置问题,并建立了相应的BP分类器。最后通过实验分析了基于两种构造特征的识别算法的性能。2.研究了基于瞬时频率的雷达信号调制方式识别算法。针对在低信噪比条件下瞬时频率统计特征稳定性差而导致识别率低的问题,给出了一种基于瞬时频率图的雷达信号调制方式识别的方法。在深入分析雷达信号瞬时频率特性的基础上,给出了一种将瞬时频率转化为二值图像的方法,提取了瞬时频率二值图像的矩特征,并进行了归一化处理,消除了采样频率和信号长度对特征的影响。建立了两种级联分类器并设计了相应的BP神经网络分类器,最终通过仿真实验分析了该方法的性能。3.研究了线性调频信号参数的贝叶斯估计方法和相关接收提取相位编码子码宽度的方法。推导了一种线性调频信号参数的贝叶斯估计模型,并将其推广到正弦信号载频的贝叶斯估计中。通过MCMC方法对贝叶斯估计进行优化计算,解决了贝叶斯估计方法计算量大的问题;研究了一种结合随机游走抽样和独立马尔科夫抽样的混合抽样方法,提高了MCMC算法的收敛速度;利用相关接收法提取子码宽度,采用平滑,子码宽度微调和实虚部结果相平均等方法,提高了子码宽度的估计精度。4.研究了雷达信号包络特征分析的相关算法。深入研究了希尔伯特变换提取包络的算法,针对其抗噪性能差,提取的包络误差大等缺点,研究了复小波变换提取雷达信号的包络算法。对提取的包络进行平滑处理,减小噪声对包络的影响。提取了复小波变换后几层的包络特征,并使用PCA技术对包络特征进行了优化处理,减小了包络特征的维数并增加了包络特征的稳定性。更多还原

【Abstract】 Radar signal intra-pulse modulation recognition and analysis have encountered challenges, as new system radars have rapidly developed and electromagnetism environment in space was more and more complex. First, two algorithms of modulation recognition is researched based on wavelet packet and BP neural network. Then, intra-pulse character of radar signal is researched from radar signal modulation parameter extraction and radar signal inter-pulse analysis. My main work is summarized as follow:1. Radar signal intra-pulse modulation recognition algorthms based on wavelet packet is studied. Wavelet packet translation is studied deeply, using optimization algorithm of wavelet packet decomposed, reconstruct coefficients which contained large information was extracted. To solve the mesh of the wavelet packet reconstruct coefficients too high and how to construct the recognition feature, two method are proposed: First one, the energy of wavelet packet reconstruct coefficient is calculated, statistical features of energy are recognition feature; Another one, feature selecting based on SVD from Reconstruct coefficients of wavelet packet is proposed, six big singular values was recognition feature. How to setup parameter of BP model is introduced, and establish the BP classifier. The performance of two kinds of features is analyzed by simulating.2. Radar signal intra-pulse modulation recognition algorthms based on instantaneous frequency is studied. To solve the lower recognition rate issue, a radar signal intra-pulse modulation recognition algorthms based on instantaneous frequency image is proposed. The instantaneous frequency was studied deeply, a method of transforming the instantaneous frequency into binary image is proposed. The moment feature of the binary image is extracted, and normalizing the moment feature eliminates the affect of sampling frequency and the length of signal. Two classifiers of and a BP classifier is established, finally the performance of this method is analyzed by simulating.3. Bayesian parameter estimation algorithm of LFM signal and code width of PSK signal based on Correlation receive algorithm is studied. The model of Bayesian estimation is studied for estimating the modulation parameter of LFM signal and the frequency of sine signal. The MCMC algorithm is introduced to calculate the Bayesian estimation for eliminating the computation burden. To improve the convergence speed of MCMC, a hybrid sampling between Rand Walks and Independent Markov Chain is proposed. Correlation receive algorithm is studied for estimating the width of code. The precision of code width is improved by smoothing, fine-tuning the width of code and averaging the real part’result and imaginary part’result. 4. The algorithm of envelop feature analysis of radar signal is studied. The performance of Hilbert transform and complex wavelet transform for extracting envelop is studied deeply. In order to eliminating the affect of noise, complex wavelet transform is used to extract envelop. To eliminate the affect of noisy, envelop is smoothed. Finally the envelop feature is extracted from the last layers of complex wavelet transform, and the PCA method is used to eliminate the redundancy of the envelop feature and enhance the stability of the envelop feature.更多还原