基于时频特征提取和残差神经网络的雷达信号识别

doi:10.12305/j.issn.1001-506X.2021.04.08

摘要/Abstract

摘要：

针对低信噪比(signal to noise ratio, SNR)下雷达信号脉内调制类型识别率较低的问题, 提出了基于时频特征提取和残差神经网络的雷达信号识别算法。时频特征提取首先通过分数阶傅里叶变换对信号进行Chirp基分解, 按照Chirp基载频与调频率的不同组合对信号划分类别, 并设置对应的分类特征参数。然后, 计算信号的伪Wigner-Ville时频分布并提取Zernike矩。上述特征参数组成信号特征矢量, 使用残差神经网络分类器实现雷达信号识别。仿真结果表明, 在SNR=-2 dB时识别准确率能达到93%以上, 同时鲁棒性验证良好, 算法复杂度能够满足现实要求。

关键词: 雷达信号识别, 分数阶傅里叶变换, Chirp基分解, Zernike矩, 残差神经网络

Abstract:

Aiming at the problem of low recognition rate of radar signal pulse modulation type under low signal to noise ratio(SNR), a radar signal recognition algorithm based on time-frequency feature extraction and residual neural network is proposed. The time-frequency feature extraction firstly performs chirp-based decomposition of the signal through the fractional Fourier transform, classifies the signal according to different combinations of chirp-based carrier frequency and frequency modulation, and sets the corresponding classification feature parameters. Then, the pseudo Wigner-Ville time-frequency distribution of the signal is calculated and Zernike moments is extracted. The above-mentioned characteristic parameters form a signal characteristic vector, and a residual neural network classifier is used to realize radar signal recognition. Simulation results show that the recognition accuracy can reach more than 93% when SNR is under -2 dB. At the same time, the robustness is well verified, and the algorithm complexity can meet the actual requirements.

Key words: radar signal identification, fractional Fourier transform, Chirp-based decomposition, Zernike moment, residual neural network

中图分类号:

TN911

谢存祥, 张立民, 钟兆根. 基于时频特征提取和残差神经网络的雷达信号识别[J]. 系统工程与电子技术, 2021, 43(4): 917-926.

Cunxiang XIE, Limin ZHANG, Zhaogen ZHONG. Radar signal recognition based on time-frequency feature extraction and residual neural network[J]. Systems Engineering and Electronics, 2021, 43(4): 917-926.

图/表 16

表1

图1

表2

图2

图3

图4

图5

表3

图6

图7

图8

表4

图9

表5

图10

图11

参考文献 29

1	KISHORE T R , RAO K D . Automatic intrapulse modulation classification of advanced LPI radar waveforms[J]. IEEE Trans.on Aerospace Electronic & Systems, 2017, 53 (2): 901- 914.
2	IGLESIAS V , GRAJAL J , ROYER P , et al. Real-time low-complexity automatic modulation classifier for pulsed radar signals[J]. IEEE Trans.on Aerospace and Electronic Systems, 2015, 51 (1): 108- 126. doi: 10.1109/TAES.2014.130183
3	黄颖坤, 金炜东, 葛鹏, 等. 基于多尺度信息熵的雷达辐射源信号识别[J]. 电子与信息学报, 2019, 41 (5): 1084- 1091.
	HUANG Y K , JIN W D , GE P , et al. Radar emitter signal identification based on multi-scale information entropy[J]. Journal of Electronics & Information Technology, 2019, 41 (5): 1084- 1091.
4	LIU Y J , XIAO P , WU H C , et al. LPI radar signal detection based on radial integration of Choi-Williams time-frequency image[J]. Journal of Systems Engineering and Electronics, 2015, 26 (5): 973- 981. doi: 10.1109/JSEE.2015.00106
5	高敬鹏, 孔维宇, 刘佳琪, 等. 基于时频分析的自适应PCA辐射源调制识别[J]. 应用科技, 2018, 45 (5): 33- 37.
	GAO J P , KONG W Y , LIU J Q , et al. Research on emitter modulation recognition of the adaptive PCA based on time-frequency analysis[J]. Applied Science and Technology, 2018, 45 (5): 33- 37.
6	DUDCZYK J . A method of feature selection in the aspect of specific identification of radar signals[J]. Bulletin of the Polish Academy of Sciences Technical Sciences, 2017, 65 (1): 113- 119. doi: 10.1515/bpasts-2017-0014
7	GUO Q , NAN P L , WAN J . Signal classification method based on data mining for multi-mode radar[J]. Journal of Systems Engineering and Electronics, 2016, 27 (5): 1010- 1017. doi: 10.21629/JSEE.2016.05.09
8	GAO J P , LU Y , QI J W , et al. A radar signal recognition system based on non-negative matrix factorization network and improved artificial bee colony algorithm[J]. IEEE Access, 2019, 7, 117612- 117626. doi: 10.1109/ACCESS.2019.2936669
9	刘鲁涛, 戴亮军, 陈涛. 基于频谱复杂度的雷达信号调制方式识别[J]. 哈尔滨工程大学学报, 2018, 39 (6): 1081- 1086.
	LIU L T , DAI L J , CHEN T . Radar signal modulation recognition based on spectrum complexity[J]. Journal of Harbin Engineering University, 2018, 39 (6): 1081- 1086.
10	ZHANG Y X , HONG R J , PAN P P , et al. Frequency-domain range sidelobe correction in stretch processing for wideband LFM radars[J]. IEEE Trans.on Aerospace Electronics System, 2017, 53 (1): 111- 121. doi: 10.1109/TAES.2017.2649278
11	黄宇, 刘锋, 王泽众, 等. 基于FRFT的雷达信号chirp基稀疏特征提取及分选[J]. 航空学报, 2013, 34 (2): 393- 400.
	HUANG Y , LIU F , WANG Z Z , et al. Chirp function sparse feature extraction and sorting of radar signals based on FRFT[J]. Acta Aeronautica et Astronautica Sinica, 2013, 34 (2): 393- 400.
12	曲志昱, 毛校洁, 侯长波. 基于奇异值熵和分形维数的雷达信号识别[J]. 系统工程与子技术, 2018, 40 (2): 303- 307.
	QU Z Y , MAO X J , HOU C B . Radar signal recognition based on singular value entropy and fractal dimension[J]. Systems Engineering and Electronics, 2018, 40 (2): 303- 307.
13	ZHANG M , LIU L , DIAO M . LPI radar waveform recognition based on time-frequency distribution[J]. Sensors, 2016, 16 (10): 1682. doi: 10.3390/s16101682
14	PENG S L , JIANG H Y , WANG H X , et al. Modulation classification based on signal constellation diagrams and deep learning[J]. IEEE Trans.on Neural Networks and Learning Systems, 2019, 30 (3): 718- 727. doi: 10.1109/TNNLS.2018.2850703
15	邓兵, 王旭, 陶然, 等. 基于分数阶傅里叶变换的线性调频脉冲时延估计特性分析[J]. 兵工学报, 2012, 33 (6): 764- 768.
	DENG B , WANG X , TAO R , et al. Performance analysis of time delay estimation for linear frequency-modulated pulse based on fractional fourier transform[J]. Acta Armamentarii, 2012, 33 (6): 764- 768.
16	赵兴浩, 邓兵, 陶然. 分数阶傅里叶变换数值计算中的量纲归一化[J]. 北京理工大学学报, 2005, 25 (4): 361- 364.
	ZHAO X H , DENG B , TAO R . Dimensional normalization in the digital computation of the fractional Fourier transform[J]. Transactions of Beijing Institute of Technology, 2005, 25 (4): 361- 364.
17	LYU Q S , QIN H L . A joint method based on time-frequency distribution to detect time-varying interference for GNSS receivers with a single antenna[J]. SenSors, 2019, 19, 1946. doi: 10.3390/s19081946
18	ZHANG X W , ZUO L , HUANG W Z , et al. Efficient method for the field-programmable gate arrays calculation of Wigner-Ville distribution[J]. IEEE Signal Processing Letters, 2019, 13 (6): 589- 595.
19	陈雕, 张登福, 雍霄驹, 等. 雷达信号时频分析的特征提取[J]. 计算机应用, 2012, 32 (7): 2063- 2065, 2069.
	CHEN D , ZHANG D F , YONG X J , et al. Feature extraction in time-frequency analysis of radar signal sorting[J]. Journal of Computer Applications, 2012, 32 (7): 2063- 2065, 2069.
20	FAROKHI A S , SHAMSUDDIN S M , FLUSSER J , et al. Rotation and noise invariant near-infrared face recognition by means of Zernike moments and spectral regression discriminant analysis[J]. Journal of Electronic Imaging, 2013, 22 (1): 013030. doi: 10.1117/1.JEI.22.1.013030
21	LECUN Y , BENGIO Y , HINTON G . Deep learning[J]. Nature, 2015, 521 (7553): 436. doi: 10.1038/nature14539
22	O'SHEA T , HOYDIS J . An introduction to deep learning for the physical layer[J]. IEEE Trans.on Cognitive Communications and Networking, 2017, 3 (4): 563- 575. doi: 10.1109/TCCN.2017.2758370
23	PENG S L , JIANG H Y , WANG H X , et al. Modulation classification based on signal constellation diagrams and deep learning[J]. IEEE Trans.on Neural Networks and Learning Systems, 2019, 30, 718- 727. doi: 10.1109/TNNLS.2018.2850703
24	ISHITSUKA K , ISO S , ONISHI K , et al. Object detection in ground-penetrating radar images using a deep convolutional neural network and image set preparation by migration[J]. International Journal of Geophysics, 2018, 2018, 9365184.
25	HAN L , SUN J Z , ZHANG W . Convolutional neural network for convective storm nowcasting using 3-D Doppler weather radar data[J]. IEEE Trans.on Geoence and Remote Sensing, 2020, 58 (2): 1487- 1495. doi: 10.1109/TGRS.2019.2948070
26	LAN R S , ZOU H Z , PANG C , et al. Image denoising via deep residual convolutional neural networks[J]. Signal, Image and Video Processing, 2019, 15 (9): 1- 8. doi: 10.1007/s11760-019-01537-x
27	YAN H Y , XU X . End-to-end video subtitle recognition via a deep residual neural network[J]. Pattern Recognition Letters, 2020, 131 (3): 368- 375.
28	YUAN Q Q , ZHANG Q , LI J , et al. Hyperspectral image denoising employing a spatial-spectral deep residual convolutional neural network[J]. IEEE Trans.on Geoence and Remote Sensing, 2019, 57 (2): 1205- 1218. doi: 10.1109/TGRS.2018.2865197
29	HUANG H , PU C Y , LI Y , et al. Adaptive residual convolutional neural network for hyperspectral image classification[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2020, 13, 2520- 2531. doi: 10.1109/JSTARS.2020.2995445

信号	数学形式
CW	$s\left( t \right) = A{{\rm{e}}^{{\rm{j}}2{\rm{ \mathsf{ π} }}{f_0}t}}$
LFM	$s\left( t \right) = A{{\rm{e}}^{{\rm{j}}2{\rm{ \mathsf{ π} }}({f_0}t + \frac{1}{2}\mu {t^2})}}$
EQFM	$s\left( t \right) = A{{\rm{e}}^{\rm{j}}}\left( {^{2{\rm{ \mathsf{ π} }}{f_0}t + {\rm{ \mathsf{ π} }}k{{(t - \frac{T}{2})}^3}}} \right)$
SFM	$s\left( t \right) = A{{\rm{e}}^{{\rm{j}}[2{\rm{ \mathsf{ π} }}{f_0}t + {m_f}{\rm{ sin }}(2{\rm{ \mathsf{ π} }}{f_m}t)]}}$
FMCW	$s\left( t \right) = \left\{ \begin{array}{l}A{{\rm{e}}^{{\rm{j}}2{\rm{ \mathsf{ π} }}({f_0}t + \frac{1}{2}\mu {t^2})}}, t \in \left[ { - T/2, 0} \right]\\A{{\rm{e}}^{{\rm{j}}2{\rm{ \mathsf{ π} }}({f_0}t - \frac{1}{2}\mu {t^2})}}, t \in \left[ {0, T/2} \right]\end{array} \right.$
FSK	$s\left( t \right) = \mathop \sum \limits_{i = 1}^N A{{\rm{e}}^{{\rm{j}}2{\rm{ \mathsf{ π} }}{f_i}t}}, {f_i} \in \{ {f_1}, {f_2}, \ldots , {f_N}\} $
BPSK	$s\left( t \right) = \mathop \sum \limits_{i = 1}^N A{{\rm{e}}^{{\rm{j}}(2{\rm{ \mathsf{ π} }}{f_0}t + {\rm{ \mathsf{ π} }}{C_n})}}$
FPSK	$s\left( t \right) = A{{\rm{e}}^{{\rm{j}}[2{\rm{ \mathsf{ π} }}{f_0}t + {\varphi _k} + {\theta _0}]}}$

信号分类	μ	f	调制类型
第1类	0	唯一	CW BPSK
第2类	0	不唯一	FSK
第3类	不等于0，唯一	唯一	LFM
第4类	不等于0，唯一	不唯一	Frank码
第5类	不等于0，不唯一	唯一	FMCW
第6类	不等于0，不唯一	不唯一	EQFM SFM

卷积层层数	SNR/dB	识别率/%
卷积层层数	SNR/dB	CNN	残差神经网络
4	-4	66.750 0	85.862 5
8	-4	81.125 0	86.125 0
12	-4	44.250 0	86.000 0
16	-4	24.875 0	86.250 0
4	2	96.875 0	98.875 0
8	2	70.812 5	99.187 5
12	2	58.625 0	99.307 5
16	2	25.000 0	99.312 5

训练集SNR/dB	测试集SNR/dB	总体识别率/%
-6	-2	89.437 5
-2	4	99.812 5
2	-2	91.375 0
6	0	94.187 5

信号类型	信号参数	取值
CW	载频/MHz	30~75
LFM	载频/MHz	30~75
LFM	调频率/(MHz·μs^-1)	15~25
EQFM	载频/MHz	30~75
EQFM	调制系数k	10~25
SFM	载频/MHz	30~75
	调制系数k	10~25
	调制频率m_f/MHz	0.15~0.35
FMCW	载频/MHz	30~75
FMCW	调频率/(MHz·μs^-1)	15~25
FSK	两频率点选取范围/MHz	30~75
BPSK	载频/MHz	30~75
BPSK	7位、11位、13位barker码	-
FPSK	载频/MHz	30~75
FPSK	9点、16点、25点Frank码	-