跳频信号盲检测的半监督干扰对消方法

doi:10.12305/j.issn.1001-506X.2023.07.35

摘要/Abstract

摘要：

实际跳频信号所处的电磁环境较为复杂且难以预料，这给基于仿真数据训练的检测算法带来困扰。针对这一问题，提出一种名为半监督干扰对消的方法。该方法首先以暹罗嵌套Unet为主干, 引入图注意力机制和集成通道注意力模块, 得到干扰对消网络，并用成对的跳频信号时频图以及对应的标签对其进行预训练，使其获得干扰对消及检测信号的能力。然后，将没有标签、干扰更为复杂的时频图输入到干扰对消网络，得到低熵预测，作为伪标签。同时，对这些没有标签的时频图进行强增强，得到变形时频图。训练网络使得变形时频图的检测结果与伪标签具有一致性，从而强化网络在没有标签的数据上的泛化能力。仿真结果表明，所提方法可以在复杂干扰下实现参数估计和盲检测，并利用无标签数据增强网络性能。

关键词: 跳频检测, 干扰对消, 注意力机制, 半监督学习

Abstract:

The real electromagnetic environment for real hopping frequency is complex and unpredictable, which poses a problem to the detection algorithm based on simulation data training. To address this problem, a method called semi-supervised interference cancellation is proposed. The method firstly introduces a graph attention mechanism and an ensemble channel attention module with Siamese nested Unet backbone to obtain an interference cancellation network, and pretrains it with paired spectrograms of hopping signals and corresponding labels to obtain the ability of interference cancellation and signal detection. Secondly, input the unlabeled spectrograms with more complex interference to the interference cancellation network to obtain low-entropy predictions as pseudo labels. Meanwhile, the unlabeled spectrograms are also strongly enhanced to obtain the distorted spectrograms. The network is trained so that the detection results of the distorted spectrograms are consistent with the pseudo-label, thus strengthening the generalization ability of the network on the unlabeled data. The simulation results show that the proposed method can achieve parameter estimation and blind detection under complex interference and enhance the network performance with unlabeled data.

Key words: frequency hopping detection, interference cancelation, attention mechanism, semi-supervised learning

中图分类号:

TN911.7

邓喆, 雷菁, 孙承哲. 跳频信号盲检测的半监督干扰对消方法[J]. 系统工程与电子技术, 2023, 45(7): 2236-2248.

Zhe DENG, Jing LEI, Chengzhe SUN. Semi-supervised interference cancellation method for frequency hopping signal blind detection[J]. Systems Engineering and Electronics, 2023, 45(7): 2236-2248.

图/表 22

图1

图2

图3

图4

表1

图5

图6

图7

表2

表3

图8

图9

图10

图11

图12

表4

图13

表5

表6

表7

图14

表8

参考文献 26

1	LYU J F, QU W. Application of the wavelet rearrangement algorithm in the detection of noncooperative frequency hopping signals[C]//Proc. of the IEEE 11th International Conference on Signal Processing, 2012: 263-266.
2	SIROTIYA M, BANERJEE A. Detection and estimation of frequency hopping signals using wavelet transform[C]//Proc. of the 2nd UK-India-IDRC International Workshop on Cognitive Wireless Systems, 2010.
3	赵宏伟, 李勇. 一种基于时频分布的跳频信号检测方法的研究[J]. 信息安全与通信保密, 2006, 4 (6): 98-99, 102
	ZHAO H W , LI Y . The research of FH signal detection based on time-frequency distribution[J]. China Information Security, 2006, 4 (6): 98-99, 102
4	陈利虎, 张尔扬. 一种多分量跳频信号参数盲估计方法[J]. 信号处理, 2009, 25 (2): 194- 198.
	CHEN L H , ZHANG E Y . A new method for blind parameter estimation of multicomponent frequency-hopping signals[J]. Signal Processing, 2009, 25 (2): 194- 198.
5	XU M K, PING X J, LI T Y, et al. A new time-frequency spectrogram analysis of FH signals by image enhancement and mathematical morphology[C]//Proc. of the 4th International Conference on Image and Graphics, 2007: 610-615.
6	LUO S G, LUO L Y. Adaptive detection of an unknown FH signal based on image features[C]//Proc. of the 5th International Conference on Wireless Communications, Networking and Mobile Computing, 2009.
7	王曼颖, 龚晓峰, 雒瑞森, 等. 基于自适应形态学的跳频信号参数联合盲估计[J]. 系统工程与电子技术, 2021, 43 (5): 1398- 1405.
	WANG M Y , GONG X F , LUO R S , et al. Joint blind parameter estimation of frequency hopping signal based on adaptive morphology[J]. Systems Engineering and Electronics, 2021, 43 (5): 1398- 1405.
8	张盛魁, 姚志成, 何岷, 等. 改进时频脊线的跳频参数盲估计算法[J]. 系统工程与电子技术, 2019, 41 (12): 2885- 2890.
	ZHANG S K , YAO Z C , HE M , et al. Blind estimation algorithm for frequency hopping parameters of improved time-frequency ridge[J]. Systems Engineering and Electronics, 2019, 41 (12): 2885- 2890.
9	BOUSIAS ALEXAKIS E , ARMENAKIS C . Evaluation of unet and unet++ architecturs in high resolution image change detection applications[J]. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2020, 43, 1507- 1514.
10	CHEN L C, ZHU Y K, PAPANDREOU G, et al. Encoder-decoder with atrous separable convolution for semantic image segmentation[C]//Proc. of the Computer Vision-European Conference on Computer Vision, 2018.
11	DAUDT R C, LE S B, BOULCH A. Fully convolutional Siamese networks for change detection[C]//Proc. of the 25th IEEE International Conference on Image Processing, 2018: 4063-4067.
12	ZI W J , XIONG W , CHEN H , et al. SGA-Net: self-constructing graph attention neural network for semantic segmentation of remote sensing images[J]. Remote Sensing, 2021, 13 (21): 4201.
13	ZHANG C X , YUE P , TAPETE D , et al. A deeply supervised image fusion network for change detection in high resolution bi-temporal remote sensing images[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2020, 166, 183- 200.
14	TARVAINEN A, VALPOLA H. Mean teachers are better role models: weight-averaged consistency targets improve semi-supervised deep learning results[C]//Proc. of the 31st Annual Conference on Neural Information Processing Systems, Neural Information Processing Systems Foundation, 2017: 1196-1205.
15	SOHN K, BERTHELOT D, CARLINI N, et al. Fixmatch: simplifying semi-supervised learning with consistency and confidence[C]//Proc. of the 34th Conference on Neural Information Processing Systems, Neural Information Processing Systems Foundation, 2020: 596-608.
16	PENG D F , BRUZZONE L , ZHANG Y J , et al. SemiCDNet: a semisupervised convolutional neural network for change detection in high resolution remote-sensing images[J]. IEEE Trans.on Geoscience and Remote Sensing, 2020, 59, 5891- 5906.
17	FANG S , LI K Y , SHAO J Y , et al. SNUNet-CD: a densely connected Siamese network for change detection of VHR images[J]. IEEE Geoscience and Remote Sensing Letters, 2022, 19, 1- 5.
18	SUN C Z , WU J J , CHEN H , et al. SemiSANet: a semi-supervised high-resolution remote sensing image change detection model using Siamese networks with graph attention[J]. Remote Sensing, 2022, 14 (12): 2801.
19	CUBUK E D, ZOPH B, SHLENS J, et al. Practical automated data augmentation with a reduced search space[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2020: 702-703.
20	HARALICK M R , SHAPIRO L G . Computer and robot vision[M]. Massachusetts: Addison-Wesley Publishing Company, 1992.
21	OESTGES C , CLERCKX B . MIMO wireless communications: from real-world propagation to space-time code design[M]. New York: Academic Press, 2010.
22	CSURKA G, LARLUS D, PERRONNIN F. What is a good evaluation measure for semantic segmentation[C]//Proc. of the 24th British Machine Vision Conference, 2013.
23	韩文草, 孙闽红, 王之腾, 等. 基于DeepLabV3与GAN的雷达时频混叠多信号智能检测与分离[J]. 信号处理, 2022, 38 (5): 1065- 1074.
	HAN W C , SUN M H , WANG Z T , et al. Intelligent detection and separation of radar time frequency aliasing multi-signal based on DeepLabV3 and GAN[J]. Journal of Signal Processing, 2022, 38 (5): 1065- 1074.
24	张珊, 吴瑛, 陈秋华. 基于CASH-CFAR的跳频信号检测[J]. 计算机工程与设计, 2010, 31 (21): 4697- 4700.
	ZHANG S , WU Y , CHEN Q H . Frequency hopping signal detection based on CASH-CFAR[J]. Computer Engineering and Design, 2010, 31 (21): 4697- 4700.
25	齐昶, 王斌, 颜羡卿. 短波环境下跳频信号检测[J]. 雷达科学与技术, 2011, 9 (4): 335- 340.
	QI C , WANG B , YAN X Q . Frequency hopping signal detection in short wave environment[J]. Radar Science and Techno-logy, 2011, 9 (4): 335- 340.
26	IANDOLA F N, MOSKEWICZ M W, ASHRAF K, et al. SqueezeNet: Alex Netlevel accuracy with 50x fewer parameters and < 1MB model size[EB/OL]. [2022-06-13]. https://arxiv.53yu.com/abs/1602.07360.

名称	描述	幅度
翻转	将图像水平或者垂直对调	-
调换	将两个输入调换位置	-
平移/%	将图像沿着某轴平移	[-0.2, 0.2]
遮挡/%	用方块遮挡住图像的一部分	[0.2, 0.4]
裁剪/%	沿着某轴裁剪图像	[-0.2, 0.2]

参数	取值
有标签数据总数/对	400
测试数据集大小/对	1 452
有标签数据信噪比/dB	0~10
有标签数据信干比/dB	0~10
无标签、测试数据信噪比/dB	-10~0
无标签、测试数据信干比/dB	-10~0
跳频频率集大小/点	64
符号速率/(sym/s)	5×10⁵
采样率/Hz	6.144×10⁷
信号帧长/s	4×10^-2
最大多普勒频移/Hz	100
多径数量	3
莱斯信道K因子	4
多径延迟/s	[0 0.9 1.7]×10^-5
短时傅里叶变换窗	256点凯撒窗
时频图大小/点	[256 256]

参数	取值
优化器	AdamW
初始学习率	1×10^-3
学习率衰减	1×10^-2
训练轮次	100.0
无监督损失加权系数w	2.0
预测熵门限值λ_e	0.3

网络	F1	IoU	漏警率	虚警率
CA-CFAR	—	—	0.199 7	0.067 6
DeepLabv3+	0.735 0	0.593 0	0.030 1	0.033 7
UNet	0.653 0	0.617 0	0.026 4	0.014 5
ICnet	0.969 0	0.891 0	0.056 7	4.64×10^-4

比例	F1	交并比	漏警率	虚警率
全监督	0.951 0	0.841 0	0.107 8	7.258×10^-4
1∶1	0.956 0	0.856 0	0.085 6	4.734×10^-4
1∶2	0.967 0	0.889 0	0.058 6	4.821×10^-4
1∶4	0.969 0	0.891 0	0.056 7	4.64×10^-4