卷积神经网络卫星信号自动调制识别算法

崔天舒; 崔凯; 黄永辉; 赵文杰; 安军社

doi:10.13700/j.bh.1001-5965.2020.0711

卷积神经网络卫星信号自动调制识别算法

doi: 10.13700/j.bh.1001-5965.2020.0711

崔天舒^{1, 2},
崔凯^{1, 2},
黄永辉^{1, 2, ,},
赵文杰¹,
安军社^{1, 2}

1.
中国科学院国家空间科学中心复杂航天系统电子信息技术重点实验室, 北京 100190
2.
中国科学院大学计算机科学与技术学院，北京 100049

基金项目:

中国科学院复杂航天系统电子信息技术重点实验室自主部署基金 Y42613A32S

详细信息

通讯作者:
黄永辉, E-mail: yonghui@nssc.ac.cn

中图分类号: V47;TN927⁺.21
计量
- 文章访问数: 458
- HTML全文浏览量: 206
- PDF下载量: 48
- 被引次数: 0
出版历程
- 收稿日期: 2020-12-24
- 录用日期: 2021-03-12
- 网络出版日期: 2022-06-20
- 整期出版日期: 2022-06-20

Convolutional neural network based algorithm for automatic modulation recognition of satellite signals

CUI Tianshu^{1, 2},
CUI Kai^{1, 2},
HUANG Yonghui^{1, 2
, ,},
ZHAO Wenjie¹,
AN Junshe^{1, 2}

1.
Key Laboratory of Electronics and Information Technology for Space Systems, National Space Science Center, Chinese Academy of Sciences, Beijing 100190, China
2.
School of Computer Science and Technology, University of Chinese Academy of Sciences, Beijing 100049, China

Funds:

Laboratory Fund of Key Laboratory of Electronics and Information Technology for Space Systems, CAS Y42613A32S

More Information

Corresponding author: HUANG Yonghui, E-mail: yonghui@nssc.ac.cn

摘要

摘要:
自动调制识别是空间认知通信系统的关键技术，有助于实现自适应信号解调。深度神经网络虽然具有特征提取能力强的优势，但也存在参数众多、计算量大的问题，难以实现空间在轨应用。针对以上问题，提出了一种轻量化、高性能的卷积神经网络结构。网络先提取信号的同相正交相关特征，再提取时域特征，最后提取各通道特征均值进行分类。对11种调制方式分类的实验结果表明：当信噪比高于0 dB时，平均识别准确率能达到86.94%，较传统的高阶累积量的方法提高了31.54%;与目前高识别准确率的深度神经网络模型相比，仅使用不到10%的模型参数，在树莓派4B上计算速度平均提高了20倍。
- 认知通信系统 /
- 调制方式识别 /
- 同相正交相关特征 /
- 卷积神经网络 /
- 深度学习
Abstract:
Automatic modulation recognition is a key technology for spatial cognitive communication system, which helps to realize adaptive signal demodulation. Although the deep neural network has the advantage of strong feature extraction, it suffers from the problems of numerous parameters and large amount of calculation, and thus is difficult to be implemented in in-orbit applications. To mitigate these problems, we propose a lightweight, high-performance convolutional neural network structure. The network first extracts the in-phase and quadrature features of the signal, then the time domain features, and finally the mean value of each channel feature for classification. The experimental results of the classification of 11 modulation methods show that when the signal-to-noise ratio is higher than 0 dB, the average recognition accuracy can reach 86.94%, which is 31.54% higher than that of traditional cumulant methods. Compared with the current deep neural network model with high recognition accuracy, the network proposed uses only less than 10% of model parameters, and increases the calculation speed by an average of 20 times on Raspberry Pi 4B.
- cognitive communication system /
- modulation recognition /
- in-phase and quadrature correlation feature /
- convolutional neural network /
- deep learning

HTML全文

图 1 卫星智能接收系统

Figure 1. Intelligent receiver system for satellite

下载: 全尺寸图片幻灯片

图 2 IQCNet网络结构

Figure 2. IQCNet network structure

下载: 全尺寸图片幻灯片

图 3 实验总体流程

Figure 3. Flowchart of overall experiment process

下载: 全尺寸图片幻灯片

图 4 实验详细流程

Figure 4. Flowchart of experimental procedures

下载: 全尺寸图片幻灯片

图 5 16个卷积核时不同深度的网络结构识别准确率比较

Figure 5. Comparison of recognition accuracy between different depths of network structure under 16 convolution kernels

下载: 全尺寸图片幻灯片

图 6 24个卷积核时不同深度的网络结构识别准确率比较

Figure 6. Comparison of recognition accuracy between different depths of network structure under 24 convolution kernels

下载: 全尺寸图片幻灯片

图 7 32个卷积核时不同深度的网络结构识别准确率比较

Figure 7. Comparison of recognition accuracy between different depths of network structure under 32 convolution kernels

下载: 全尺寸图片幻灯片

图 8 与其他方法的识别准确率比较

Figure 8. Comparison of recognition accuracy with other methods

下载: 全尺寸图片幻灯片

图 9 IQCNet混淆矩阵(信噪比为0 dB, 识别准确率为81.84%)

Figure 9. IQCNet confusion matrix (SNR=0 dB, Accuracy=81.84%)

下载: 全尺寸图片幻灯片

图 10 IQCNet混淆矩阵(信噪比为6 dB, 识别准确率为87.62%)

Figure 10. IQCNet confusion matrix(SNR=6 dB, Accuracy=87.62%)

下载: 全尺寸图片幻灯片

表 1 IQCNet(2, 32)网络结构

Table 1. IQCNet(2, 32) network structure

层名称	输入尺寸	尺寸/步进	卷积核数量
Conv2d-1	128×2	1×2/1×1	32
BatchNorm2d-1	128×1		32
MaxPool2d-1	128×1	2×1/2×1	32
Conv2d-2	64×1	1×3/1×1	32
BatchNorm2d-2	64×1		32
MaxPool2d-2	64×1	2×1/2×1	32
AdaptiveAvgPool2d	64×1		32
Linear-1	16

下载: 导出CSV

表 2 IQCNet-N(2, 32)网络结构

Table 2. IQCNet-N(2, 32) network structure

层名称	输入尺寸	尺寸/步进	卷积核数量
Conv2d-1	128×2	1×2/1×1	32
BatchNorm2d-1	128×2		32
MaxPool2d-1	128×2	2×1/2×1	32
Conv2d-2	64×2	1×3/1×1	32
BatchNorm2d-2	64×2		32
MaxPool2d-2	64×2	2×1/2×1	32
AdaptiveAvgPool2d	64×2		32
Linear-1	16

下载: 导出CSV

表 3 对比网络参数

Table 3. Parameters of networks for comparision

网络名称	卷积层数	卷积核尺寸	通道数	LSTM层数	LSTM单元数
CNN2	2	(1, 3), (2, 3)	256, 80	0	0
CLDNN	3	(1, 8)	50, 50, 50	1	50
CNN_LSTM	2	(1, 3), (2, 3)	128, 32	1	128

下载: 导出CSV

表 4 实验设备

Table 4. Experimental equipment

测试设备	CPU	GPU	内存/GB
PC	i9-7920X	RTX 2080Ti	64
Jetson Nano	Cortex-A57	128个CUDA核	4
树莓派4B	Cortex-A72	无	4

下载: 导出CSV

表 5 平均识别准确率

Table 5. Average recognition accuracy

网络名称	平均识别准确率/%
网络名称	信噪比-20~18 dB	信噪比0~18 dB
IQCNet-N(3, -)	47.70	73.47
IQCNet-N(4, -)	50.33	76.49
IQCNet-N(5, -)	53.49	80.14
IQCNet(3, -)	54.36	82.36
IQCNet(4, -)	58.37	86.72
IQCNet(5, -)	59.27	86.82
IQCNet-N(-, 16)	50.07	75.90
IQCNet-N(-, 24)	51.75	78.38
IQCNet-N(-, 32)	49.70	75.82
IQCNet(-, 16)	57.68	85.66
IQCNet(-, 24)	56.87	84.94
IQCNet(-, 32)	57.45	85.30
IQCNet-N	50.51	76.70
IQCNet	57.33	85.30

下载: 导出CSV

表 6 不同方法平均识别准确率

Table 6. Average recognition accuracy of different methods

方法	平均识别准确率/%
方法	信噪比-20~18 dB	信噪比0~18 dB
Cumulants+KNN	34.54	55.40
CNN2	56.82	78.86
CLDNN	56.84	82.92
CNN_LSTM	60.19	86.89
IQCNet(4, 24)	58.15	86.94

下载: 导出CSV

表 7 网络参数与计算时间

Table 7. Network parameters and compute time

网络名称	网络参数/个	RTX 2080Ti训练时间/s	树莓派4B推理时间/s	Jetson Nano推理时间/s
CNN2	5 369 947	823	3 035	5 243
CLDNN	71 311	4 523	3 794	1 585
CNN__LSTM	108 971	2 673	3 564	990
IQCNet	5 963	171	181	473

下载: 导出CSV

参考文献(24)

[1]	VIDAL O, VERELST G, LACAN J, et al. Next generation high throughput satellite system[C]//2012 IEEE First AESS European Conference on Satellite Telecommunications. Piscataway: IEEE Press, 2012: 1-7.
[2]	SMITH A, EVANS M, DOWNEY J. Modulation classification of satellite communication signals using cumulants and neural networks[C]//2017 Cognitive Communications for Aerospace Applications Workshop. Piscataway: IEEE Press, 2017: 1-8.
[3]	SILLS J A. Maximum-likelihood modulation classification for PSK/QAM[C]//IEEE Military Communications. Piscataway: IEEE Press, 1999: 217-220.
[4]	WEI W, MENDEL J M. Maximum-likelihood classification for digital amplitude-phase modulations[J]. IEEE Transactions on Communications, 2000, 48(2): 189-193. doi: 10.1109/26.823550
[5]	HASSAN K, DAYOUB I, HAMOUDA W, et al. Automatic modulation recognition using wavelet transform and neural network[C]//2009 9th International Conference on Intelligent Transport Systems Telecommunications. Piscataway: IEEE Press, 2009: 234-238.
[6]	DONG S L, LI Z P, ZHAO L F. A modulation recognition algorithm based on cyclic spectrum and SVM classification[C]//2020 IEEE 4th Information Technology, Networking, Electronic and Automation Control Conference. Piscataway: IEEE Press, 2020: 2123-2127.
[7]	FLOHBERGER M, GAPPMAIR W, KOUDELKA O. Modulation classifier for signals used in satellite communications[C]//2010 5th Advanced Satellite Multimedia Systems Conference and the 11th Signal Processing for Space Communications Workshop. Piscataway: IEEE Press, 2010: 198-202.
[8]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2016: 770-778.
[9]	ARISOY E, SETHY A, RAMABHADRAN B, et al. Bidirectional recurrent neural network language models for automatic speech recognition[C]//2015 IEEE International Conference on Acoustics, Speech and Signal Processing. Piscataway: IEEE Press, 2015: 5421-5425.
[10]	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. https://www.cnki.com.cn/Article/CJFDTOTAL-TXXB201907012.htm ZHOU X, HE X X, ZHENG C W. Radio signal recognition based on image deep learning[J]. Journal on Communications, 2019, 40(7): 114-125(in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-TXXB201907012.htm
[11]	KARRA K, KUZDEBA S, PETERSEN J. Modulation recognition using hierarchical deep neural networks[C]//2017 IEEE International Symposium on Dynamic Spectrum Access Networks. Piscataway: IEEE Press, 2017: 1-3.
[12]	PENG S L, JIANG H Y, WANG H X, et al. Modulation recognition using hierarchical deep neural networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2019, 30(3): 718-727. doi: 10.1109/TNNLS.2018.2850703
[13]	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. doi: 10.11959/j.issn.1000-436x.2019206 ZHA X, PENG H, QIN X, et al. Modulation recognition method based on multi-inputs convolution neural network[J]. Journal on Communications, 2019, 40(11): 30-37(in Chinese). doi: 10.11959/j.issn.1000-436x.2019206
[14]	ZHANG M, DIAO M, GUO L M. Convolutional neural networks for automatic cognitive radio waveform recognition[J]. IEEE Access, 2017, 5: 11074-11082. doi: 10.1109/ACCESS.2017.2716191
[15]	O'SHEA T J, CORGAN J, CLANCY T C. Convolutional radio modulation recognition networks[C]//International Conference on Engineering Applications of Neural Networks, 2016: 213-226.
[16]	WEST N E, O'SHEA T. Deep architectures for modulation recognition[C]//2017 IEEE International Symposium on Dynamic Spectrum Access Networks. Piscataway: IEEE Press, 2017: 1-6.
[17]	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
[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. doi: 10.3969/j.issn.0372-2112.2019.11.029 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(in Chinese). doi: 10.3969/j.issn.0372-2112.2019.11.029
[19]	YAO T Y, CHAI Y, WANG S, et al. Radio signal automatic modulation classification based on deep learning and expert features[C]//2020 IEEE 4th Information Technology, Networking, Electronic and Automation Control Conference. Piscataway: IEEE Press, 2020: 1225-1230.
[20]	WU Y L, LI X J, FANG J. A deep learning approach for modulation recognition via exploiting temporal correlations[C]//2018 IEEE 19th International Workshop on Signal Processing Advances in Wireless Communications. Piscataway: IEEE Press, 2018: 1-5.
[21]	KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90. doi: 10.1145/3065386
[22]	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL]. (2015-04-10)[2020-12-01]. https://arxiv.org/abs/1409.1556.
[23]	IOFFE S, SZEGEDY C. Batch normalization: Accelerating deep network training by reducing internal covariate shift[C]//International Conference on Machine Learning. New York: ACM, 2015: 448-456.
[24]	SAINATH T N, VINYALS O, SENIOR A, et al. Convolutional, long short-term memory, fully connected deep neural networks[C]//2015 IEEE International Conference on Acoustics, Speech and Signal Processing. Piscataway: IEEE Press, 2015: 4580-4584.