基于图像深度学习的无线电信号识别

doi:10.11959/j.issn.1000-436x.2019167

摘要/Abstract

摘要：

提出了一种利用图像深度学习解决无线电信号识别问题的技术思路。首先把无线电信号具象化为一张二维图片，将无线电信号识别问题转化为图像识别领域的目标检测问题；进而充分利用人工智能在图像识别领域的先进成果，提高无线电信号识别的智能化水平和复杂电磁环境下的识别能力。基于该思路，提出了一种基于图像深度学习的无线电信号识别算法——RadioImageDet 算法。实验结果表明，所提算法能有效识别无线电信号的波形类型和时/频坐标，在实地采集的12种、4 740个样本的数据集中，识别准确率达到86.04%，mAP值达到77.72，检测时间在中等配置的台式计算机上仅需33 ms，充分验证了所提思路的可行性和所提算法的有效性。

关键词: 无线电信号识别, 深度学习, 射频机器学习, 卷积神经网络, 图像目标检测

Abstract:

A technical idea was innovatively proposed that uses image deep learning to solve the problem of radio signal recognition.First,the radio signal was transformed into a two-dimensional picture,and the radio signal recognition problem was transformed into the object detection problem in the field of image recognition.Then,the advanced achievements about image recognition were used to improve the intelligence and ability of radio signal recognition in complex electromagnetic environment.Based on the proposed idea,a novel radio signal recognition algorithm named RadioImageDet was proposed.The experimental results show that the algorithm can effectively identify the waveform types and time/frequency coordinates of radio signals.After training and testing on the self-collected data set with 12 types and 4 740 samples,the accuracy reaches 86.04% and the mAP value reaches 77.72,while the detection time is only 33 ms on the medium configured desktop computer.

Key words: radio signal recognition, deep learning, radio frequency machine learning, convolutional neural network, image object detection

中图分类号:

TN971

周鑫,何晓新,郑昌文. 基于图像深度学习的无线电信号识别[J]. 通信学报, 2019, 40(7): 114-125.

Xin ZHOU,Xiaoxin HE,Changwen ZHENG. Radio signal recognition based on image deep learning[J]. Journal on Communications, 2019, 40(7): 114-125.

图/表 15

图1

表1

图2

图3

图4

表2

图5

表3

图6

图7

图8

图9

图10

表4

图11

参考文献 37

[1]	DOBRE O A . Signal identification for emerging intelligent radios:classical problems and new challenges[J]. IEEE Instrumentation ＆Measurement Magazine, 2015,18(2): 11-18.
[2]	YAO Y , HUANG Z . Communicational signals’ modulation recognition (Chinese with English abstract)[J]. Communications Technology, 2003,6(1): 41-50.
[3]	AZZOUZ E E , NANDI A K . Automatic identification of digital modulation types[J]. Signal Processing, 1995,47(1): 55-69.
[4]	LOPATKA J , PEDZISZ M . Automatic modulation classification using statistical moments and a fuzzy classifier[C]// International Conference on Signal Processing Proceedings. 20003(1): 1500-1506.
[5]	NANDI A K , AZZOUZ E E . Algorithms for automatic modulation recognition of communication signals[J]. IEEE Transactions on Communications, 1998,46(4): 431-436.
[6]	REICHERT J , . Automatic classification of communication signals using higher order statistics[C]// IEEE International Conference on Acoustics,Speech,and Signal Processing. IEEE, 1992: 221-224.
[7]	SPOONER C M , MODY A N , CHUAN G J . Modulation recognition using second-and higher-order cyclostationarity[C]// IEEE International Symposium on Dynamic Spectrum Access Networks. IEEE, 2017: 1-3.
[8]	SOLIMA N S S , HSUE S Z . Signal classification using statisticalmoments[J]. IEEE Transactions on Communications, 1992,40(5): 908-916.
[9]	YAMASHITA Y , OCHIAI H . A classification of OFDM signals with or without DFT precoding based on high-order moment[C]// IEEE International Symposium on Dynamic Spectrum Access Networks. IEEE, 2017: 1-2.
[10]	CHEN J , KUO Y , LI J ,et al. Review of automatic communication signals recognition (Chinese with English abstract)[J]. Journal of Circuits and Systems, 2005,10(5): 102-109.
[11]	PARNAS D L . The real risks of artificial intelligence[J]. Communications of the ACM, 2017,60(10): 27-31.
[12]	WONG M L D , NANDI A K . Automatic digital modulation recognition using spectral and statistical features with multi-layer perceptron[C]// Signal Processing and its Applications. 2001: 390-393.
[13]	ARULAMPALAM G , RAMAKONAR V , BOUZERDOUM A ,et al. Classification of digital modulated schemes using neural networks[C]// International Symposium on Signal Processing and its Applications. 1999: 649-652.
[14]	TRIANTAFYLLAKIS K , SURLIGAS M , VARDAKI S G . Phasma:an automatic modulation classification system based on random forest[C]// IEEE International Symposium on Dynamic Spectrum Access Networks. IEEE , 2017: 1-3.
[15]	O’SHEA T , CORGAN J , CLANCY T C . Convolutional radio modulation recognition networks[C]// International Conference on Engineering Applications of Neural Networks. 2016: 213-226.
[16]	O’SHEA T , CORGAN J , CLANCY T C . Unsupervised representation learning of structured radio communication signals[C]// First International Workshop on Sensing,Processing and Learning for Intelligent Machines. 2016: 1-5.
[17]	O’SHEA T , WEST N , VONDAL M ,et al. Semi-supervised radio signal identification[C]// International Conference on Advanced Communication Technology. 2017: 33-38.
[18]	O’SHEA T , WEST N . Radio machine learning dataset generation with GNU radio[C]// Proceedings of the GNU Radio Conference. 2016: 1-6.
[19]	ZIEGLER J L , ARN R T , CHAMBERS W . Modulation recognition with GNU radio,Keras and HackRF[C]// IEEE International Symposium on Dynamic Spectrum Access Networks. IEEE, 2017: 1-3.
[20]	KARRA K , KUZDEBA S , PETERSEN J . Modulation recognition using hierarchical deep neural networks[C]// IEEE International Symposium on Dynamic Spectrum Access Networks. IEEE, 2017: 1-3.
[21]	WEST N , O'SHEA T , . Deep architectures for modulation recognition[C]// IEEE International Symposium on Dynamic Spectrum Access Networks. IEEE, 2017: 1-6.
[22]	WEST N , HARWELL K , MCCALL B . DFT signal detection and channelization with a deep neural network modulation classifier[C]// IEEE International Symposium on Dynamic Spectrum Access Networks. IEEE, 2017: 1-3.
[23]	ARUMUGAM K , KADAMPOT I , TAHMASBI M ,et al. Modulation recognition using side information and hybrid learning[C]// IEEE International Symposium on Dynamic Spectrum Access Networks. IEEE, 2017: 1-2.
[24]	HUANG L , YU J , SHEN Z ,et al. Radio individual identification via stable communication signals based on subordinate component analysis[C]// IET International Radar Conference. 2016,1, 1-4.
[25]	YUAN Y , HUANG Z , WANG F ,et al. Radio specific emitter identification based on nonlinear characteristics of signal[C]// IEEE International Black Sea Conference on Communications and Networking. IEEE, 2015: 77-81.
[26]	AHMAD K , SHRESTA G , MEIER U ,et al. Neuro-fuzzy signal class IFIER (NFSC) for standard wireless technologies[C]// International Symposium on Wireless Communication Systems. 2010: 616-620.
[27]	SCHMIDT M , BLOCK D , MEIER U . Wireless interference identification with convolutional neural networks[J]. IEEE 15th International Conference on Industrial Informatics, 2017: 180-185.
[28]	ZHANG M , DIAO M , GUO L . Convolutional neural networks for automatic cognitive radio waveform recognition[J]. IEEE Access, 2017,5(1): 11074-11082.
[29]	GIRSHICK R , DONAHUE J , DARRELL ,et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 20141: 580-587.
[30]	GIRSHICK R , . Fast R-CNN[C]// IEEE International Conference on Computer Vision. IEEE, 2015: 1440-1448.
[31]	REN S , HE K , GIRSHICK R ,et al. Faster R-CNN:towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017,39(6): 1137-1149.
[32]	LIN T , DOLLAR P , GIRSHICK R ,et al. Feature pyramid networks for object detection[C]// IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2017: 936-944.
[33]	REDMON J , FARHADI A . YOLO9000:better,faster,stronger[C]// IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2017: 6517-6525.
[34]	LIU W , ANGUELOV D , ERHAN D ,et al. SSD:single shot multibox detector[C]// European Conference on Computer Vision. 2016: 21-37.
[35]	DANEV B , CAPKUN S . Transient-based identification of wireless sensor nodes[C]// International Conference on Information Processing in Sensor Networks. 2009: 25-36.
[36]	SZEGEDY C , LIU W , JIA Y ,et al. Going deeper with convolutions[C]// IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2015: 1-9.
[37]	HE K , ZHANG X , REN S ,et al. Deep residual learning for image recognition[C]// IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2016: 770-778.

网络层类型	输入尺寸	参数	激活函数
conv层	128×2	3×1卷积核256个特征映射dropout 50%	ReLU
conv层	256×126×2	3×2卷积核80个特征映射dropout 50%	ReLU
dense层	9 920×1	256个神经元dropout 50%	ReLU
dense层	256×1	11个神经元	softmax

层名称	输入尺寸	尺寸/步进	卷积核数
conv1	512×512	3×3 / 1×1	32
maxpool1	512×512	2×2 / 2×2	—
conv2	256×256	3×3 / 1×1	64
maxpool2	256×256	2×2 / 2×2	—
conv3	128×128	3×3 / 1×1	128
conv4	128×128	1×1 / 1×1	64
conv5	128×128	3×3 / 1×1	128
maxpool3	128×128	2×2 / 2×2	—
conv6	64×64	3×3 / 1×1	256
conv7	64×64	1×1 / 1×1	128
conv8	64×64	3×3 / 1×1	256
maxpool4	64×64	2×2 / 2×2	—
conv9	32×32	3×3 / 1×1	512
conv10	32×32	1×1 / 1×1	256
conv11	32×32	3×3 / 1×1	512
conv12	32×32	1×1 / 1×1	256
conv13	32×32	3×3 / 1×1	512
maxpool5	32×32	2×2 / 2×2	—
conv14	16×16	3×3 / 1×1	1 024
conv15	16×16	1×1 / 1×1	512
conv16	16×16	3×3 / 1×1	1 024
conv17	16×16	1×1 / 1×1	512
conv18	16×16	3×3 / 1×1	1 024
conv19	16×16	3×3 / 1×1	1 024
conv20	16×16	3×3 / 1×1	1 024
conv21	32×32	1×1 / 1×1	64
reshape1	32×32	2×2	256
concatenate1	16×16	—	1 280
conv22	16×16	3×3 / 1×1	1 024
conv23	16×16	1×1 / 1×1	85

频率范围/MHz	信号名称
870～885	Tele2G_DL_CDMA
935～955	Mobi2G_DL_GSM900
955～960	Uni2G_DL_GSM900
1 805～1 820	Mobi2G_DL_GSM1800
1 840～1 860	Uni4G_DL_FDD1800
1 860～1 875	Tele4G_DL_FDD1800
1 885～1 905	Mobi4G_TDLTE_20M
1 905～1 915	Mobi4G_TDLTE_10M
2 010～2 025	Mobi3G_TDSCDMA
2 110～2 130	Tele4G_DL_FDD2100
2 130～2 135	Uni3G_DL_WCDMA
2 135～2 140	Uni3G_DL_WCDMA
2 140～2 155	Uni4G_DL_FDD2100
2 575～2 595	Mobi4G_TDLTE_20M
2 595～2 615	Mobi4G_TDLTE_20M
2 615～2 635	Mobi4G_TDLTE_20M

序号	信号名称	精准率	召回率	AP值
1	Tele2G_DL_CDMA	100%	98%	97.67
2	Mobi2G_DL_GSM900	100%	95%	95.35
3	Uni2G_DL_GSM900	100%	92%	92.31
4	Mobi2G_DL_GSM1800	100%	93%	93.02
5	Uni3G_DL_WCDMA	100%	100%	100
6	Mobi3G_TDSCDMA	30%	62%	45.76
7	Uni4G_DL_FDD1800	100%	26%	26.02
8	Uni4G_DL_FDD2100	94%	60%	58.50
9	Tele4G_DL_FDD1800	72%	98%	85.09
10	Tele4G_DL_FDD2100	54%	99%	55.02
11	Mobi4G_TDLTE_20M	94%	84%	83.91
12	Mobi4G_TDLTE_10M	93%	100%	99.92
平均	mAP	—	—	77.72