﻿ 基于修正的Rife和SVM的辐射源特征提取和识别
 文章快速检索 高级检索

Emitter feature extraction and recognition based on the modified Rife and SVM
ZHANG Chunjie, GONG Zailan , REN Lili
College of Information and Communication, Harbin Engineering University, Harbin 150001, China
Abstract:For a single sine signal with mixed phase noise, the modified Rife algorithm has higher frequency measurement accuracy. This paper proposed a recognition method for emitter individuals based on the modified Rife and support vector machine (SVM). First the characteristics of the frequency spectrum of a frequency oscillator were analyzed, then the basic principle of the modified Rife algorithm and the classification thoughts for SVM was expounded. Two precise parameters of carrier frequency and frequency offset which are also the two vectors for SVM were got through the modified Rife algorithm. Classifiers were used to identify different sources of emitter individuals. Finally, emitter feature extraction and recognition research were done for the actually measured data. The computer simulation results proved effectiveness of the algorithm presented in this paper.
Key words: feature extraction     emitter recognition     phase noise     frequency offset     modified Rife algorithm     support vector machine

1 修正的Rife算法和SVM基本原理 1.1 Rife算法基本原理

Rife算法根据2个相邻谱线最大幅度和次大幅度的比值对离散傅里叶变换(discrete Fourier transform,DFT)频谱进行插值，又称为双线幅度法。

1)0≤－m0fs/N≤Δf/3；

2)－Δf/3≤－m0fs/N≤0。

1.2 SVM分类器

y(x)=ATx+b

M arg in=2/||A||

G(xi,xj)=〈Γ(xi)·Γ(xj)〉

2 特征提取及识别步骤

 图 1 利用修正的Rife算法进行特征提取流程

1)确定判别函数；

2)确定分类面方程；

3)找到支持向量；

4)确定分类间隔；

5)求出最优分类面；

6)核函数的确定。

3 仿真实验

3.1 相位噪声的产生

 个体1 个体2 个体3 频偏/kHz L(f)/( dBc·Hz-1) 频偏/kHz L(f)/( dBc·Hz-1) 频偏/kHz L(f)/( dBc·Hz-1) 1 -84 6 -70 3 -72 10 -105 60 -83 30 -103 100 -108 600 -97 300 -100 1 000 -114 6 000 -90 3 000 -99 10 000 -132 60 000 -112 30 000 -122
 图 2 3个个体的相位噪声功率谱密度
3.2 频率漂移估计仿真结果

 图 3 频偏量随信噪比的变换情况

 图 4 频偏量随信号频率变化情况

 kHz 个体 -9 dB -3 dB 3 dB 10 dB 13 dB 15 dB 20 dB 30 dB 个体1 766.856 8 297.079 8 153.188 3 63.060 2 40.426 4 37.802 6 18.512 2 6.560 7 个体2 872.121 8 413.741 5 277.781 5 268.069 9 237.909 6 236.430 8 232.957 5 214.732 3 个体3 769.725 6 303.767 0 163.065 3 81.642 0 63.462 4 68.309 2 54.446 6 49.317 3 无相噪信号 766.863 4 296.578 5 152.832 7 63.159 9 40.147 1 38.026 8 18.260 2 5.856 6 CRB 545.969 3 273.632 9 137.141 3 61.258 8 43.367 9 34.448 3 19.371 7 6.125 9

 kHz 个体 480 MHz 500 MHz 520 MHz 540 MHz 560 MHz 580 MHz 600 MHz 个体1 61.684 0 62.723 5 61.853 7 62.058 8 61.330 0 61.908 3 65.635 0 个体2 251.326 3 260.675 1 254.629 9 256.775 8 242.142 7 253.866 7 254.974 4 个体3 76.358 1 81.756 5 82.103 1 79.622 5 82.287 6 85.567 0 85.648 8 无相噪信号 61.259 6 62.719 5 61.794 6 61.298 0 61.262 9 61.938 7 65.905 5 CRB 61.258 8 61.258 8 61.258 8 61.258 8 61.258 8 61.258 8 61.258 8

3.3 SVM完成对实测数据分类

 样本数 训练样本 测试样本 f0/MHz 频偏/kHz Tf/MHz Tb/kHz 1 502.378 8 4.1447 502.380 0 11.729 0 2 502.380 0 13.404 0 502.379 1 6.581 2 3 502.379 4 6.701 0 502.397 1 7.764 0 4 502.3797 11.704 0 502.380 0 11.729 0 5 502.378 8 6.259 7 502.378 8 5.704 8 6 502.379 1 5.716 3 502.379 1 6.853 0 7 502.378 0 12.185 0 502.379 1 6.319 7 8 502.378 8 5.554 2 502.379 1 6.701 4

2个辐射源的训练样本和测试样本的分类结果如图 5所示。仿真结果表明，这种基于SVM的二元分类的方法，识别率为100%。由此可见，二元分类中，对于小样本空间的同型号、同标称频率的辐射源个体的识别问题，这种算法不仅简单，而且有很好的识别效果。

 图 5 基于SVM的2个辐射源的分类结果

 图 6 基于SVM的3个辐射源训练样本分类结果

4 结束语

 [1] 柴恒, 孙毓富. 基于无意调制特征的雷达个体识别[J]. 舰船电子对抗, 2011, 34(4): 1-5. [2] CONNING M, POTGIETER F. Analysis of measured radar data for specific emitter identification[C]// IEEE Radar Conference. Washington DC, USA, 2010: 35-38. [3] 李娜. 雷达信号有意调制分析及指纹特征提取[D]. 哈尔滨: 哈尔滨工程大学, 2013: 8-10. [4] 田波, 龙良将. 基于时频原子方法的雷达辐射源信号无意调制特征提取[J]. 电子元器件应用, 2010, 12(5): 44-47. [5] 任黎丽. 辐射源指纹识别与细微特征提取方法研究[D]. 哈尔滨: 哈尔滨工程大学, 2012: 26-31. [6] VAPNIK V N. The nature of statistical learning theory[M]. New York: Springer-Verlag, 1995: 5-13. [7] RUEDA L, OOMMEN B J. On optimal pairwise linear classifiers for normal distributions: the 3-dimensional case[J]. Pattern Recognition, 2003, 36(1): 13-23. [8] PERNKOPF F. Bayesian network classifiers versus selective k-NN classifier[J]. Pattern Recognition, 2005, 38(l): l-10. [9] 陈娟. 雷达辐射源特征提取与个体识别[D]. 西安: 西安电子科技大学, 2013: 4-5. [10] SHOTTON J, SHARP T. Real-time human pose recognition in parts from single depth images[J]. Communications of the ACM, 2013, 56(1): 116-124. [11] CHEN T, JIN W, LI J. Feature extraction using surrounding-line integral bispectrum for radar emitter signal[C]// IEEE International Joint Conference on Neural Networks. Hong Kong, China, 2008: 294-298.

#### 文章信息

ZHANG Chunjie，GONG Zailan，REN Lili

Emitter feature extraction and recognition based on the modified Rife and SVM

Applied Science and Technology, 2015, (03): 7-12.
DOI:10.3969/j.issn.1009-671X.201403021