基于自监督对比学习的信号调制识别算法

doi:10.12305/j.issn.1001-506X.2023.04.29

摘要/Abstract

摘要：

近年来, 基于深度学习的信号调制识别技术发展迅速, 但大多数解决方案以监督学习方法为主, 需要大量带标签样本。考虑到信号数据样本分析标注难度大、成本高, 提出了一种通过自监督对比学习利用大量无标签样本进行模型预训练、基于预训练特征值提取及利用少量带标签样本进行调制识别训练的学习方法, 可大幅降低训练所需的带标签样本数量。在RadioML2018.01A上的试验表明, 所提方法仅需1%的带标签数据就能达到全数据集上与监督学习模型相当的性能, 且带标签数据减少至0.1%时, 对信噪比大于等于8 dB的24类调制信号的识别准确率仍然能达到93%以上。

关键词: 深度学习, 调制识别, 自监督对比学习, 模型预训练

Abstract:

The technique of signal modulation identification based on deep learning has made significant progress in recent years, but most of the solutions are based on supervised learning methods which require a large number of labeled samples. It is well-known that labeling on signal data samples is difficult and costly. Therefore, a semi-supervised learning method is proposed, which is pre-trained through self-supervised contrast learning method with a large number of unlabeled samples, and the modulation identification network based on the pre-trained feature extraction network can be trained to convergence by a small number of labeled samples. In this way, the dependence on labeled samples can be reduced significantly. Experiments on RadioML2018.01A show that the proposed algorithm used 1% of the labeled samples can almost achieve the same identification performance as the supervised learning algorithm with the use of all of the labeled samples. Meanwhile, when trained on only 0.1% of the labeled samples, the identification model on 24 modulation types still achieved the accuracy of above 93% under the condition that the signal-to-noise ratio is equal to and over 8 dB.

Key words: deep learning, modulation identification, self-supervised contrast learning, model pretraining

中图分类号:

TN911.7

陈洋, 廖灿辉, 张锟, 刘建, 王鹏举. 基于自监督对比学习的信号调制识别算法[J]. 系统工程与电子技术, 2023, 45(4): 1200-1206.

Yang CHEN, Canhui LIAO, Kun ZHANG, Jian LIU, Pengju WANG. A signal modulation indentification algorithm based on self-supervised contrast learning[J]. Systems Engineering and Electronics, 2023, 45(4): 1200-1206.

图/表 10

图1

图2

表1

图3

图4

表2

表3

图5

图6

图7

参考文献 19

20	邵凯, 朱苗苗, 王光宇. 基于生成对抗与卷积神经网络的调制识别方法[J]. 系统工程与电子技术, 2022, 44 (3): 1036- 1043.
	SHAO K , ZHU M M , WANG G Y . Modulation recognition method based on generative adversarial network[J]. Systems Engineering and Electronics, 2022, 44 (3): 1036- 1043.
21	OORD A, LI Y, VINYALS O. Representation learning with contrastive predictive coding[EB/OJ]. [2021-10-10]. https://arxiv.org/abs/1807.03748.
22	HE K M, FAN H Q, WU Y X, et al. Momentum contrast for unsupervised visual representation learning[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 9729-9738.
23	CHEN X L, FAN H Q, GIRSHICK R, et al. Improved baselines with momentum contrastive learning[EB/OJ]. [2021-10-10]. https://arxiv.org/abs/2003.04297.
24	CHEN X L, XIE S N, HE K. An empirical study of training self-supervised visual transformers[C]//Proc. of the IEEE/CVF Conference on Computer Vision, 2021: 9640-9649.
25	CHEN T, KORNBLITH S, NOROUZI M, et al. A simple framework for contrastive learning of visual representations[C]//Proc. of the 37th IEEE International Conference on Machine Learning, 2020: 1597-1607.
26	CHEN T , KORNBLITH S , SWERSKY K , et al. Big self-supervised models are strong semi-supervised learners[J]. Advances in neural information processing systems, 2020, 33, 22243- 22255.
27	NAIR V, HINTON G E. Rectified linear units improve restricted Boltzmann machines[C]//Proc. of the 27th IEEE International Conference on Machine Learning, 2010: 807-814.
28	LIN M, CHEN Q, YAN S C. Network in network[C]//Proc. of the International Conference on Learning Representations, 2014.
29	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.
30	SZEGEDY C, WEI L, JIA Y, et al. Going deeper with convolutions[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2015.
1	NANDI A K , AZZOUZ E E . Algorithms for automatic modulation recognition of communication signals[J]. IEEE Trans.on Communications, 1998, 46 (4): 431- 436. doi: 10.1109/26.664294
2	AZZOUZ E E , NANDI A K . Automatic identification of digital modulation types[J]. Signal Processing, 1995, 47 (1): 55- 69. doi: 10.1016/0165-1684(95)00099-2
3	HO K C , PROKOPIW W , CHAN Y T . Modulation identification of digital signals by the wavelet transform[J]. IEE Proceedings-Radar, Sonar and Navigation, 2002, 147 (4): 169- 176.
4	SWAMI A , SADLER B M . Hierarchical digital modulation classification using cumulants[J]. IEEE Trans.on Communications, 2000, 48 (3): 416- 429. doi: 10.1109/26.837045
5	SHERME A E . A novel method for automatic modulation recognition[J]. Applied Soft Computing, 2012, 12 (1): 453- 461. doi: 10.1016/j.asoc.2011.08.025
6	GARDNER W A. Cyclostationarity in communications and signal processing[R]. Yountville CA: Statistical Signal Processing Inc, 1994.
7	FEHSKE A, GAEDDERT J D, REED J H. A new approach to signal classification using spectral correlation and neural networks[C]//Proc. of the IEEE International Symposium on New Frontiers in Dynamic Spectrum Access Networks, 2005: 144-150.
8	BOUTTE D , SANTHANAM B . A hybrid ICA-SVM approach to continuous phase modulation recognition[J]. IEEE Signal Processing Letters, 2009, 16 (5): 402- 405. doi: 10.1109/LSP.2009.2016444
9	王建新, 宋辉. 基于星座图的数字调制方式识别[J]. 通信学报, 2004, 25 (6): 166- 173.
	WANG J X , SONG H . Digital modulation recognition based on constellation diagram[J]. Journal of Communications, 2004, 25 (6): 166- 173.
10	QUINLAN J R . Induction of decision trees[J]. Machine Learning, 1986, 1 (1): 81- 106.
11	CORTES C , VAPNIK V . Support-vector networks[J]. Machine Learning, 1995, 20 (3): 273- 297.
12	HEARST M A , DUMAIS S T , OSUNA E , et al. Support vector machines[J]. IEEE Intelligent Systems and their applications, 1998, 13 (4): 18- 28. doi: 10.1109/5254.708428
13	LIPPMANN R . An introduction to computing with neural nets[J]. IEEE ASSP Magazine, 1987, 4 (2): 4- 22. doi: 10.1109/MASSP.1987.1165576
14	O'SHEA T J, CORGAN J, CLANCY T C. Convolutional radio modulation recognition networks[C]//Proc. of the International Conference on Engineering Applications of Neural Networks, 2016: 213-226.
15	WEST N E, O'SHEA T. Deep architectures for modulation recognition[C]//Proc. of the IEEE International Symposium on Dynamic Spectrum Access Networks, 2017.
16	O'SHEA T J , ROY T , CLANCY T C . Over-the-air deep learning based radio signal classification[J]. IEEE Journal of Selected Topics in Signal Processing, 2018, 12 (1): 168- 179. doi: 10.1109/JSTSP.2018.2797022
17	查雄, 彭华, 秦鑫, 等. 基于多端卷积神经网络的调制识别方法[J]. 通信学报, 2019, 40 (11): 30- 37.
	ZHA X , PENG H , QIN X , et al. Modulation recognition method based on multi-inputs convolution neural network[J]. Journal of Communications, 2019, 40 (11): 30- 37.
18	查雄, 彭华, 秦鑫, 等. 基于循环神经网络的卫星幅相信号调制识别与解调算法[J]. 电子学报, 2019, 47 (11): 2443- 2448.
	ZHA X , PENG H , QIN X , et al. Satellite amplitude-phase signals modulation identification and demodulation algorithm based on the cyclic neural network[J]. Acta Electronica Sinica, 2019, 47 (11): 2443- 2448.
19	周鑫, 何晓新, 郑文昌. 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.
	ZHOU X , HE X X , ZHENG W C . Radio signal recognition based on image deep learning[J]. Journal of Communications, 2019, 40 (7): 114- 125.

模板名称	输出维度
信号输入层	1 024×2
数据增强层	256×2×1
残差栈1	128×2×64
残差栈2	64×2×64
残差栈3	32×2×64
残差栈4	16×2×64
残差栈5	8×2×64
残差栈6	4×2×64
全局平均池化	64
全连接层1	64
全连接层2	64
全连接层3	64

信号属性	数据集
调制类型	OOK, 4ASK, 8ASK, BPSK, QPSK, 8PSK, 16PSK, 32PSK, 16APSK, 32APSK, 64APSK, 128APSK, 16QAM, 32QAM, 64QAM, 128QAM, 256QAM, AM-SSB-WC, AM-SSB-SC, OQPSK, AM-DSB-WC, AM-DSB-SC, FM, GMSK
信噪比	0∶2∶30 dB
采样点数	1 024×2
数据量	4 096×24×16=1 572 864

方法名称	模型描述	层数	参数量(×10³)
CNN监督方法	7层一维卷积层(卷积核大小为7, 通道数为64)	9	260
CNN监督方法	2层全连接层(输出维度分别为128, 24)	9	260
Res-Net监督方法	6层一维残差栈(卷积核大小为5, 通道数为32)	39	327
Res-Net监督方法	3层全连接层(输出维度分别为128, 128, 24)	39	327
CLDNN监督方法	3层一维卷积层(卷积核大小为8, 通道数为64)	7	300
CLDNN监督方法	2层长短时记忆网络 (通道数为64)	7	300
本文方法	2层全连接层(输出维度分别为128, 24)	37	620
	6层二维残差栈(卷积核大小为3×2, 通道数为64)
	1层全连接层(输出维度为24)

[1]	汪锐, 张天骐, 安泽亮, 王雪怡, 方竹. 基于联合特征参数和一维CNN的MIMO-OFDM系统调制识别算法[J]. 系统工程与电子技术, 2023, 45(3): 902-912.
[2]	张晔, 侯毅, 欧阳克威, 周石琳. 单变量序列数据分类方法综述[J]. 系统工程与电子技术, 2023, 45(2): 313-335.
[3]	邵正途, 许登荣, 徐文利, 王晗中. 基于LSTM和残差网络的雷达有源干扰识别[J]. 系统工程与电子技术, 2023, 45(2): 416-423.
[4]	龚佩, 李天昀, 章昕亮, 寸陈韬. 利用联合特征参数的卫星单-混信号调制识别[J]. 系统工程与电子技术, 2023, 45(2): 589-596.
[5]	李瑞泽, 张双辉, 刘永祥. 基于卷积ADMM网络的高效结构化稀疏ISAR成像方法[J]. 系统工程与电子技术, 2023, 45(1): 56-70.
[6]	韩啸, 陈世文, 陈蒙, 杨锦程. 基于互易点学习的LPI信号开集识别[J]. 系统工程与电子技术, 2022, 44(9): 2752-2759.
[7]	张立民, 谭凯文, 闫文君, 张聿远. 基于多级跳线残差网络的雷达辐射源识别[J]. 系统工程与电子技术, 2022, 44(7): 2148-2156.
[8]	康颖, 赵治华, 吴灏, 李亚星, 孟进. 基于Deep SVDD的通信信号异常检测方法[J]. 系统工程与电子技术, 2022, 44(7): 2319-2328.
[9]	金国栋, 薛远亮, 谭力宁, 许剑锟. 基于孪生神经网络的目标跟踪算法进展研究[J]. 系统工程与电子技术, 2022, 44(6): 1805-1822.
[10]	秦博伟, 蒋磊, 许华, 齐子森. 基于残差生成对抗网络的调制识别算法[J]. 系统工程与电子技术, 2022, 44(6): 2019-2026.
[11]	赵晓枫, 徐叶斌, 吴飞, 牛家辉, 蔡伟, 张志利. 基于全局感知机制的地面红外目标检测方法[J]. 系统工程与电子技术, 2022, 44(5): 1461-1467.
[12]	邹虹, 白陈阳, 何鹏, 崔亚平, 王汝言, 吴大鹏. 基于分布式深度学习的边缘服务放置策略[J]. 系统工程与电子技术, 2022, 44(5): 1728-1737.
[13]	陈冬, 句彦伟. 基于语义分割实现的SAR图像舰船目标检测[J]. 系统工程与电子技术, 2022, 44(4): 1195-1201.
[14]	孙晶明, 虞盛康, 孙俊. 基于深度学习的HRRP识别姿态敏感性分析[J]. 系统工程与电子技术, 2022, 44(3): 802-807.
[15]	邵凯, 朱苗苗, 王光宇. 基于生成对抗与卷积神经网络的调制识别方法[J]. 系统工程与电子技术, 2022, 44(3): 1036-1043.