Please wait a minute...
浙江大学学报(工学版)  2020, Vol. 54 Issue (10): 1945-1954    DOI: 10.3785/j.issn.1008-973X.2020.10.011
计算机技术     
基于时频特征的卷积神经网络跳频调制识别
李红光(),郭英*(),眭萍,齐子森
空军工程大学 信息与导航学院,陕西 西安 710077
Frequency hopping modulation recognition of convolutional neural network based on time-frequency characteristics
Hong-guang LI(),Ying GUO*(),Ping SUI,Zi-sen QI
Information and Navigation College, Air Force Engineering University, Xi’an 710077, China
 全文: PDF(1630 KB)   HTML
摘要:

针对人工设计特征表征能力不足,提取难度大的问题,提出基于卷积神经网络(CNN)的跳频信号调制方式识别系统. 该系统通过训练学习跳频信号时频图特征,将调制方式识别问题转化为图像识别问题. 采用组合时频变换方法对跳频信号进行时频变换得到二维时频图;经过自适应维纳滤波算法滤除背景噪声,提高系统抗噪性;采用连通域检测和双线性插值算法提取跳频信号每跳时频图,对时频图大小进行重置调整;将已处理的时频图输入到设计的11层卷积神经网络中进行训练学习,通过在输出层增加Softmax分类器,实现跳频调制方式分类识别. 仿真结果表明,该系统在信噪比为–4 dB条件下,对跳频信号BPSK、QPSK、8PSK、SDPSK、QASK、16QAM、32QAM和GMSK共8种调制方式的平均识别率达到92.54%.

关键词: 跳频信号调制识别卷积神经网络(CNN)时频变换卷积层    
Abstract:

A frequency hopping signal modulation recognition system based on convolutional neural network (CNN) was proposed aiming at the problem of insufficient artificial feature representation ability and difficulty in extraction. The modulation recognition problem was transformed into an image recognition problem by learning the time-frequency characteristics of the frequency hopping signal. The two-dimensional time-frequency diagram was obtained by using time-frequency transform, and the adaptive Wiener filter was used to remove the background noise of the time-frequency diagram to improve system noise immunity. The connected-domain detection algorithm and the bilinear interpolation algorithm were applied to extract the time-frequency diagram of each hop and reset the size. The processed time-frequency map was input into the designed 11-layer convolutional neural network for training and learning. The classification and recognition of frequency hopping modulation were realized by adding a Softmax classifier in the output layer. The simulation results showed that the average recognition rate of the frequency hopping signals BPSK, QPSK, 8PSK, SDPSK, QASK, 16QAM, 32QAM and GMSK was 92.54% when the signal-to-noise ratio was –4 dB.

Key words: frequency hopping signal    modulation recognition    convolutional neural network (CNN)    time-frequency transform    convolution layer
收稿日期: 2019-09-23 出版日期: 2020-10-28
CLC:  TP 391  
基金资助: 国家自然科学基金资助项目(61601500);军队研究生资助项目(JY2018C169)
通讯作者: 郭英     E-mail: toumingwings@163.com;kdydsp@163.com
作者简介: 李红光(1986—),男,博士,从事信息对抗理论的研究. orcid.org/0000-0002-3466-4396. E-mail: toumingwings@163.com
服务  
把本文推荐给朋友
加入引用管理器
E-mail Alert
作者相关文章  
李红光
郭英
眭萍
齐子森

引用本文:

李红光,郭英,眭萍,齐子森. 基于时频特征的卷积神经网络跳频调制识别[J]. 浙江大学学报(工学版), 2020, 54(10): 1945-1954.

Hong-guang LI,Ying GUO,Ping SUI,Zi-sen QI. Frequency hopping modulation recognition of convolutional neural network based on time-frequency characteristics. Journal of ZheJiang University (Engineering Science), 2020, 54(10): 1945-1954.

链接本文:

http://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2020.10.011        http://www.zjujournals.com/eng/CN/Y2020/V54/I10/1945

图 1  基于CNN的跳频调制方式识别流程
图 2  4种时频变换方法的FH信号时频图
图 3  
图 3  5种跳频调制信号单跳时域和时频域特征图
图 4  CNN训练识别基本结构
图 5  4种典型的CNN调制识别结构
图 6  4种CNN结构的调制识别率和损失值的训练结果
CNN结构 训练集 验证集 测试集
Rpsr / % Lloss Rpsr / % Lloss Rpsr / % Lloss
CNN_1 100 0.0076 80.05 0.0892 79.59 0.0928
CNN_2 100 0.0045 81.92 0.0812 81.21 0.0873
CNN_3 100 0.0027 85.06 0.0553 84.83 0.0576
CNN_4 100 0.0032 84.69 0.0574 84.14 0.0603
表 1  不同样本集下的4种CNN调制识别训练结果
${K_{\rm l}}$ r ${R_{\rm{c}}}$/kB $V$/(幅·s?1 ${R_{{\rm{psr}}}}$/%
$7 \times 7$ 2 2269.476 2.92 91.32
$5 \times 5$ 2 1585.653 5.24 87.39
$3 \times 3$ 2 357.248 8.23 83.46
$3 \times 3$ 4 357.248 8.23 86.75
$3 \times 3$ 6 357.248 8.23 90.14
$3 \times 3$ 8 357.248 8.23 91.18
表 2  6种参数组合的CNN调制识别训练结果
图 7  基于CNN的跳频调制方式识别结构
p/% 训练集 验证集 测试集
Rpsr / % Lloss Rpsr / % Lloss Rpsr / % Lloss
0 100 0.006 2 91.88 0.085 3 90.05 0.087 4
10 100 0.013 7 87.92 0.184 7 87.42 0.188 5
20 100 0.006 9 89.24 0.093 8 88.21 0.100 4
30 100 0.007 5 88.13 0.096 6 87.98 0.099 5
40 99.93 0.062 1 90.75 0.088 2 90.62 0.110 6
50 100 0.003 8 92.81 0.087 1 92.63 0.095 4
60 100 0.004 3 89.46 0.109 2 88.34 0.101 3
70 99.96 0.076 7 88.25 0.114 7 87.83 0.132 8
80 100 0.006 6 91.33 0.087 1 91.35 0.099 6
90 100 0.009 2 89.75 0.093 6 89.43 0.101 2
表 3  不同Dropout比例的CNN调制识别训练结果
图 8  不同信噪比下的FH调制识别率
输出 Rpsr /%
输入BPSK 输入QPSK 输入SDPSK 输入8PSK 输入QASK 输入16QAM 输入32QAM 输入GMSK
BPSK 89.2 1.3 6.8 1.8 0 1.6 0 1.5
QPSK 2.1 94.2 1.3 3.1 0.5 0 1.8 1.3
SDPSK 8.1 1.4 91.1 1.7 0 0 0 2.1
8PSK 0.6 2.3 0.8 92.8 0 0 0 0.5
QASK 0 0.8 0 0.6 94.6 2.5 1.4 0
16QAM 0 0 0 0 3.2 91.7 4.7 0
32QAM 0 0 0 0 1.7 4.2 92.1 0
GMSK 0 0 0 0 0 0 0 94.6
表 4  信噪比为−4 dB时8种FH信号调制方式的识别结果
1 HAN L B, GAO F F, LI Z, et al Low complexity automatic modulation classification based on order-statistics[J]. IEEE Transactions on Wireless Communications, 2017, 16 (1): 400- 411
doi: 10.1109/TWC.2016.2623716
2 DOBTE O A, ONER M, RAJAN S, et al. Cyclostationarity-based robust algorithms for QAM signal identification[J]. IEEE Communications Letters, 2012, 16 (1): 12- 15
doi: 10.1109/LCOMM.2011.112311.112006
3 SU W Feature space analysis of modulation classification using very high-order statistics[J]. IEEE Communications Letters, 2013, 17 (9): 1688- 1691
doi: 10.1109/LCOMM.2013.080613.130070
4 吴斌, 袁亚博, 汪勃 基于记忆因子的连续相位调制信号最大似然调制识别[J]. 电子与信息学报, 2016, 38 (10): 2546- 2552
WU Bin, YUAN Ya-bo, WANG Bo Maximum likelihood modulation recognition for continuous phase modulation signals using memory factor[J]. Journal of Electronics and Information Technology, 2016, 38 (10): 2546- 2552
5 WANG F G, DOBRE O A, CHAN C, et al. Novel fold-based kolmogorov-smirnov modulation classifier[J]. IEEE Signal Processing Letters, 2016, 23 (7): 1003- 1007
doi: 10.1109/LSP.2016.2572666
6 赵雄文, 郭春霞, 李景春 基于高阶累积量和循环谱的信号调制方式混合识别算法[J]. 电子与信息学报, 2016, 38 (3): 674- 680
ZHAO Xiong-wen, GUO Chun-xia, LI Jing-chun Mixed recognition algorithm for signal modulation schemes by high-order cumulants and cyclic spectrum[J]. Journal of Electronics and Information Technology, 2016, 38 (3): 674- 680
7 XIE L J, WAN Q Cyclic feature based modulation recognition using compressive sensing[J]. IEEE Wireless Communications Letters, 2017, 6 (3): 402- 405
doi: 10.1109/LWC.2017.2697853
8 ZIEGLER J L, ARN R T, CHAMBERS W, et al. Modulation recognition with GNU radio, keras, and hackrf [C]// IEEE International Symposium on Dynamic Spectrum Access Networks. Piscataway: IEEE, 2017: 1-3.
9 LI X Q, HUANG Z T, WANG F H, et al Towards convolutional neural networks on pulse repetition interval modulation recognition[J]. IEEE Communications Letters, 2018, 22 (11): 2286- 2289
doi: 10.1109/LCOMM.2018.2864725
10 QU Z Y, MAO X J, DENG Z A Radar signal intra-pulse modulation recognition based on convolutional neural network[J]. IEEE Access, 2018, 6 (1): 43874- 43884
11 RAJENDRAN S, MEERT W, GIUSTINIANO D, et al Deep learning models for wireless signal classification with distributed low-cost spectrum sensors[J]. IEEE Transactions on Cognitive Communications and Networking, 2018, 4 (3): 433- 445
doi: 10.1109/TCCN.2018.2835460
12 ALI A, FAN Yang-yu K-sparse autoencoder based automatic modulation classification with low complexity[J]. IEEE Communications Letters, 2017, 21 (10): 2162- 2165
doi: 10.1109/LCOMM.2017.2717821
13 LI R D, LI L Z, YANG S Y, et al Robust automated VHF modulation recognition based on deep convolutional neural networks[J]. IEEE Communications Letters, 2018, 22 (5): 946- 949
doi: 10.1109/LCOMM.2018.2809732
14 郭立民, 寇韵涵, 陈涛 基于栈式稀疏自编码器的低信噪比下低截获概率雷达信号调制类型识别[J]. 电子与信息学报, 2018, 40 (4): 875- 881
GUO Li-min, KOU Yun-han, CHEN Tao Low probability of intercept radar signal recognition based on stacked sparse auto-encoder[J]. Journal of Electronics and Information Technology, 2018, 40 (4): 875- 881
15 KINGMA D, BA J. Adam: a method for stochastic optimization [C]// The 3rd International Conference on Learning Representations. San Diego: IEEE, 2015: 1-15.
16 张坤峰, 郭英, 齐子森, 等 基于稀疏贝叶斯重构的多跳频信号参数估计[J]. 华中科技大学学报: 自然科学版, 2017, 45 (1): 97- 102
ZHANG Kun-feng, GUO Ying, QI Zi-sen, et al Parameter estimation of multi-hop frequency signals based on sparse Bayesian reconstruction[J]. Journal of Huazhong University of Science and Technology: Natural Science Edition, 2017, 45 (1): 97- 102
17 ZHANG M, LIU L T, DIAO M LPI radar waveform recognition based on time-frequency distribution[J]. Sensors, 2016, 16 (10): 1682- 1702
doi: 10.3390/s16101682
18 STANKOVIC S, DJUROVIC I, PITAS I Watermarking in the space/spatial-frequency domain using two-dimensional Radon-Wigner distribution[J]. IEEE Transactions on Image Processing, 2001, 10 (4): 650- 658
doi: 10.1109/83.913599
19 DENG L, YU D Deep learning: methods and applications[J]. Foundations and Trends in Signal Processing, 2014, 7 (3): 197- 387
20 HE K M, ZHANG X Y, REN S Q, et al. Delving deep into rectifiers: surpassing human-level performance on imagenet classification [C]// 2015 IEEE International Conference on Computer Vision. Santiago, Chile: IEEE, 2015: 1026-1034.
21 ZHANG M, DIAO M, GUO L M Convolutional neural networks for automatic cognitive radio waveform recognition[J]. IEEE Access, 2017, 5 (1): 11074- 11082
22 HAO W, BIE R, GUO J, et al Optimized CNN based image recognition through target region selection[J]. Optik-International Journal for Light and Electron Optics, 2018, 156 (1): 772- 777
23 FU L, FENG Y, ELKAMIL T, et al Image recognition method of multi-cluster kiwifruit in field based on convolutional neural networks[J]. Transactions of the Chinese Society of Agricultural Engineering, 2018, 34 (1): 205- 211
[1] 陈巧红,陈翊,李文书,贾宇波. 多尺度SE-Xception服装图像分类[J]. 浙江大学学报(工学版), 2020, 54(9): 1727-1735.
[2] 张彦楠,黄小红,马严,丛群. 基于深度学习的录音文本分类方法[J]. 浙江大学学报(工学版), 2020, 54(7): 1264-1271.
[3] 杨萍,王丹,康子健,李童,付利华,余悦任. 基于模式识别和集成CNN-LSTM的阵发性房颤预测模型[J]. 浙江大学学报(工学版), 2020, 54(5): 1039-1048.
[4] 贾子钰,林友芳,张宏钧,王晶. 基于深度卷积神经网络的睡眠分期模型[J]. 浙江大学学报(工学版), 2020, 54(10): 1899-1905.
[5] 叶刚,李毅波,马逐曦,成杰. 基于ViBe的端到端铝带表面缺陷检测识别方法[J]. 浙江大学学报(工学版), 2020, 54(10): 1906-1914.
[6] 吕艳,张萌,姜吴昊,倪益华,钱小鸿. 采用卷积神经网络的老年人跌倒检测系统设计[J]. 浙江大学学报(工学版), 2019, 53(6): 1130-1138.
[7] 赫贵然,李奇,冯华君,徐之海,陈跃庭. 基于CNN特征提取的双焦相机连续数字变焦[J]. 浙江大学学报(工学版), 2019, 53(6): 1182-1189.
[8] 董月,冯华君,徐之海,陈跃庭,李奇. Attention Res-Unet: 一种高效阴影检测算法[J]. 浙江大学学报(工学版), 2019, 53(2): 373-381.
[9] 王敬昌,陈岭,余珊珊,蒋晨书,吴勇. 基于门控循环单元的多因素感知短期游客人数预测模型[J]. 浙江大学学报(工学版), 2019, 53(12): 2357-2364.
[10] 李浩,赵文杰,韩波. 基于滤波器裁剪的卷积神经网络加速算法[J]. 浙江大学学报(工学版), 2019, 53(10): 1994-2002.
[11] 张林, 程华, 房一泉. 基于卷积神经网络的链接表示及预测方法[J]. 浙江大学学报(工学版), 2018, 52(3): 552-559.
[12] 王卫星, 孙守迁, 李超, 唐智川. 基于卷积神经网络的脑电信号上肢运动意图识别[J]. 浙江大学学报(工学版), 2017, 51(7): 1381-1389.
[13] 唐有宝, 卜巍, 邬向前. 多层次MSER自然场景文本检测[J]. 浙江大学学报(工学版), 2016, 50(6): 1134-1140.