Modulation Recognition Algorithm Based on Truncated Migration and Parallel ResNet

doi:10.16182/j.issn1004731x.joss.21-0282

Abstract

Abstract:

A truncated migration data preprocessing algorithm is proposed for the problem of limited time series characteristics of the signal extracted by convolutional neural network. The distance unit at one end of the sampling matrix is truncated, migrated to the other end to form a new matrix, allowing the convolutional neural network to extract more sampling points and compare more symbolic information.An improved parallel ResNet is proposed, which focuses on features in both horizontal and vertical directions simultaneously by two parallel branches. The results show that the algorithm has an accuracy rate of about 10% higher than that of ordinary convolutional networks. When the signal-to-noise ratio is 14 dB, the improved network has an accuracy rate of 93.78% and when the signal-to-noise ratio is greater than 0 dB, the accuracy rate is above 91%.

Key words: convolutional neural network, truncated migration, data preprocessing, parallel ResNet

CLC Number:

TP183
TP391

Yecai Guo, Qingwei Wang. Modulation Recognition Algorithm Based on Truncated Migration and Parallel ResNet[J]. Journal of System Simulation, 2022, 34(09): 2009-2018.

Figures/Tables 15

Fig. 1

Fig. 2

Fig. 3

Fig. 4

Fig. 5

Fig. 6

Table 1

Table 2

Table 3

Fig. 7

Table 4

Fig. 8

Table 5

Fig. 9

Fig. 10

References 20

[1]	Zhu D, Mathews V J, Detienne D H. A Likelihood-Based Algorithm for Blind Identification of QAM and PSK Signals[J]. IEEE Transactions on Wireless Communications (S1536-1276), 2018, 17(5): 3417-3430.
[2]	Chen W, Xie Z, Ma L, et al. A Faster Maximum Likelihood Modulation Classification in Flat Fading Non-gaussian Channels[J]. IEEE Communications Letters (S1089-7798), 2019, 23(3): 454-457.
[3]	Ramezani-Kebrya A, Kim I M, Kim D I, et al. Likelihood-Based Modulation Classification for Multiple-antenna Receiver[J]. IEEE Transactions on Communications (S0090-6778), 2013, 61(9): 3816-3829.
[4]	Teng C F, Liao C C, Chen C H, et al. Polar Feature Based Deep Architectures for Automatic Modulation Classification Considering Channel Fading[C]// 2018 IEEE Global Conference on Signal and Information Processing (GlobalSIP). Anaheim, California, USA:IEEE, 2018: 554-558.
[5]	Zhang Z, Zhong H, Liu Y. Modulation Classification in Multipath Fading Channels Using Sixth-order Cumulants and Stacked Convolutional Auto-encoders[J]. Iet Communications (S1751-8628), 2017, 11(6): 910-915.
[6]	Wang Y, Liu M, Yang J, et al. Data-driven Deep Learning for Automatic Modulation Recognition in Cognitive Radios[J]. IEEE Transactions on Vehicular Technology (S0018-9545), 2019, 68(4): 4074-4077.
[7]	O'Shea T J, Corgan J, Clancy T C. Convolutional Radio Modulation Recognition Networks[C]// International Conference on Engineering Applications of Neural Networks. Aberdeen, Scotland, UK: Springer, 2016: 213-226.
[8]	Liu X, Yang D, Gamal A E. Deep Neural Network Architectures for Modulation Classification[C]// 2017 51st Asilomar Conference on Signals, Systems, and Computers. Pacific Grove, California, USA: IEEE, 2017: 915-919.
[9]	Mossad S, ElNainay M, Torki M. Deep Convolutional Neural Network with Multi-task Learning Scheme for Modulations Recognition[C]// 2019 15th International Wireless Communications & Mobile Computing Conference (IWCMC). Tangier, Morocco: IEEE, 2019: 1644-1649.
[10]	Huang L, Zhang Y, Pan W, et al. Visualizing Deep Learning-based Radio Modulation Classifier[J]. IEEE Transactions on Cognitive Communications and Networking (S2332-7731), 2020, 7(1): 47-58.
[11]	Wang J, Wang W, Luo F, et al. Modulation Classification Based on Denoising Autoencoder and Convolutional Neural Network with GNU Radio[J]. The Journal of Engineering (S2051-3305), 2019(19): 6188-6191.
[12]	Huang L, Pan W, Zhang Y, et al. Data Augmentation for Deep Learning-based Radio Modulation Classification[J].IEEE Access (S2169-3536), 2019, 8: 1498-1506.
[13]	Zhang Z, Wang C, Gan C, et al. Automatic Modulation Classification Using Convolutional Neural Network with Features Fusion of SPWVD and BJD[J]. IEEE Transactions on Signal & Information Processing Over Networks (S2373-776X), 2019, 5(3): 469-478.
[14]	张清. 机器学习在通信信号调制识别中的应用研究[D].南京: 南京邮电大学, 2020.
	Zhang Qing. Research on Applications of Machine Learning in Modulation Recognition of Communication Signals[D]. Nanjing: Nanjing University of Posts and Telecommunications, 2020.
[15]	Kumar Y, Sheoran M, Jajoo G, et al. Automatic Modulation Classification Based on Constellation Density using Deep Learning[J]. IEEE Communications Letters (S1558-2558), 2020, 24(6): 1275-1278.
[16]	Luo W, Li Y, Urtasun R, et al. Understanding the Effective Receptive Field in Deep Convolutional Neural Networks[C]//30th International Conference on Neural Information Processing Systems (NIPS'16).Barcelona: Curran Associates Inc, 2017: 4905-4913.
[17]	Emam A, Shalaby M, Aboelazm M A, et al. A Comparative Study between CNN, LSTM, and CLDNN Models in The Context of Radio Modulation Classification[C]// 2020 12th International Conference on Electrical Engineering (ICEENG). Cairo, Egypt:IEEE, 2020: 190-195.
[18]	West N E, O'Shea T J. Deep Architectures for Modulation Recognition[C]// 2017 IEEE International Symposium on Dynamic Spectrum Access Networks(DySPAN). Baltimore, Maryland, USA: IEEE, 2017: 1-6.
[19]	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 (S2332-7731), 2018, 4(3): 433-445.
[20]	Xu J, Luo C, Parr G, et al. A Spatiotemporal Multi-channel Learning Framework for Automatic Modulation Recognition[J]. IEEE Wireless Communication Letters (S2162-2337), 2020, 9(10): 1629-1632.

训练集∶测试集	平均准确率/%	验证损失值	迭代一次
5∶5	61.81	1.08	177
6∶4	62.40	1.06	205
7∶3	63.48	1.03	228
8∶2	63.67	1.02	236

耐心值	平均识别率/%	最终损失值	最终验证损失值	迭代次数
5	62.19	1.04	1.07	36
10	63.08	1.01	1.05	58
15	63.48	1.00	1.03	64
20	63.14	0.98	1.07	93

超参数	最大识别准确率/%	平均识别准确率/%	输入形状	一次迭代时间/s
S=1	91.78	61.50	(32,128,2)	116
S=2	91.69	61.46	(22,128,2)	86
S=3	91.91	61.53	(18,128,2)	60
S=4	91.35	61.03	(16,128,2)	55
S=5	90.12	60.22	(14,128,2)	53
S=6	88.67	58.60	(12,128,2)	50

模型	最高准确率	平均准确率
ResNet	76.87	50.86
CLDNN	82.44	55.12
Inception	74.70	50.06
LSTM	86.07	55.87
MCLDNN	91.64	60.88
PResNet	93.78	63.48

模型	FLOPs	参数个数
PResNet	1.71G	31 719 179
ResNet152^[13]	22.6G	60 419 944
VGGNet^[14]	18.2G	93 337 594
ResNet50^[15]	7.76G	25 636 712
Inception-ResNet-V2^[15]	26.4G	55 873 736