DP-DRCnet卷积神经网络信号调制识别算法

doi:10.12382/bgxb.2021.0620

摘要/Abstract

摘要：

卷积神经网络在降低系统网络开销的同时,如何保证较高的信号调制识别准确率是目前面临的重要问题。提出一种轻量级卷积神经网络。该网络分为两路,并行提取信号的自相关和互相关特征,之后两路特征进行合并,实现不同调制方式的分类识别;该网络采用控制模型中卷积层的输入数据维度及卷积核数量的方案,实现对网络模型开销的控制。通过对多种不同的调制方式进行识别验证。实验结果表明:在信噪比为-6~12dB条件下,其平均识别准确率可达到86.5%;与传统卷积神经网络相比,计算量降低了94.44%;与常规轻量级卷积神经网络相比,计算量降低了67.6%,该网络性能优于现有的基于轻量级卷积神经网络的调制方式识别方法。

关键词: 调制识别, 卷积神经网络, 特征提取, 深度学习

Abstract:

How to ensure higher signal modulation recognition accuracy while reducing system network overhead is a important problem currently faced by the convolutional neural networks.To this end, a lightweight convolutional neural network is proposed.This networkis split into two paths to parallelly extract auto-correlation and cross-correlation features of signal.Then. features from these two paths are combined so that the network can ultimately achieve classification and recognition with different modulation modes. In addition, the overhead of the network is controlled by adopting the scheme of controlling the input data dimension of the convolution layer and the number of convolution cores in the model.The recognition verification of different modulation modes is performed.The experimental result shows that: the average recognition accuracy reaches 86.5% when the signal-to-noise ratio is in the range of -6~12dB;compared with the conventional convolutional neural network, the computational load is reduced by 94.44%; compared with the regular lightweight convolutional neural network, the computational load is reduced by 67.6%.The performance of the proposed network is better than the existing modulation recognition methods based on lightweight convolutional neural network.

Key words: modulation recognition, convolution neural network, featureextraction, deep learning

中图分类号:

TN97

王洋, 冯永新, 宋碧雪, 田秉禾. DP-DRCnet卷积神经网络信号调制识别算法[J]. 兵工学报, 2023, 44(2): 545-555.

WANG Yang, FENG Yongxin, SONG Bixue, TIAN Binghe. A Modulation Recognition Algorithm of DP-DRCnet Convolutional Neural Network[J]. Acta Armamentarii, 2023, 44(2): 545-555.

图/表 26

图1 卷积层(1×8)特征提取可视化过程[11]

Fig.1 Visualization process of the extraction of convolution layers (1×8) features[11]

图2 卷积层(2×1) (1×3)特征提取可视化过程[18]

Fig.2 Visualization process of the extraction of convolution layers (2×1) (1×3)features[18]

图3 双路卷积层(1×3)、(2×1)特征提取

Fig.3 Feature extraction of two-path convolution layers (1×3),(2×1)

图4 DP-DRCnet网络结构

Fig.4 Structure of the DP-DPCnet network

图5 DP-DRCnet网络特征提取算法

Fig.5 Feature extraction algorithm of the DP-DPCnet network

图6 DP-DRCnet网络上下路创建

Fig.6 Add/Drop creation of the DP-DRCnet network

图7 自适应池化层输出维度

Fig.7 Output dimension of adaptive pooling layer

表1 DP-DRCnet(2,2,32)网络模型参数

Table 1 Model parameters of DP-DRCnet(2,2,32)

层名称	输入数据维度	填充维度	步长	卷积核维度	通道数
conv_u1	2×128	0×0	1×1	2×1	32
avg_u1	1×126				32
conv_u2	2×32	0×0	1×1	2×1	32
avg_u2	1×31				32
conv_d1	2×128	0×0	1×1	1×3	32
avg_d1	2×124				32
conv_d2	2×32	0×0	1×1	1×3	32
avg_d2	2×29				32
pl_a	2×16				64
fc_main	1024

表2 DP-DRCnetdown(2,2,32)网络模型参数

Table 2 Model parameters of DP-DRCnetd(2,2,32)

层名称	输入数据维度	填充维度	步长	卷积核维度
conv_u1	2×128	0×0	1×1	1×3
avg_u1	2×124
conv_u2	2×32	0×0	1×1	1×3
avg_u2	2×29
conv_d1	2×128	0×0	1×1	1×3
avg_d1	2×124
conv_d2	2×32	0×0	1×1	1×3
avg_d2	2×29
pl_a	2×16
fc_main	1024

表3 DP-DRCnetup(2,2,32)网络模型参数

Table 3 Model parameters of DP- DRCnetu(2,2,32)

层名称	输入数据维度	填充维度	步长	卷积核维度
conv_u1	2×128	0×0	1×1	2×1
avg_u1	1×126
conv_u2	2×32	0×0	1×1	2×1
avg_u2	1×31
conv_d1	2×128	0×0	1×1	2×1
avg_d1	1×126
conv_d2	2×32	0×0	1×1	2×1
avg_d2	1×31
pl_a	2×16
fc_main	1024

表4 对比网络的模型参数

Table 4 Model parameters of the network for comparison

网络名称	卷积层数	卷积核维度	LSTM 层数	全连接层数	通道数
CLDNN	3	1×8	1		32/64/128
CNN_LSTM	2	1×3,2×3	1		64/128
IQCNet	5	2×1,1×3		1	32

图8 模型训练流程

Fig.8 Model training process

图9 不同深度识别度对比

Fig.9 Comparsion of recognition at different depths

图10 相同深度不同数量卷积核识别度对比

Fig.10 Comparison of recognition of convolution kernels with different numbers butthe same depth

图11 F1-score对比(SNR=0)

Fig.11 F1 score comparison (SNR=0)

图12 F1-score对比(SNR=12)

Fig.12 F1 score comparison (SNR=12)

表5 识别率对比

Table 5 Comparison of recognition rate%

网络名称	-6~ 0/dB	2~ 6/dB	8~ 12/dB	平均识别率
DP-DRCnet(2,2,32)	69.24	82.91	87.68	79.94
DP-DRCnet(3,3,32)	73.3	88.45	91.55	84.43
DP-DRCnet(4,4,32)	76.59	90.2	92.71	86.5
DP-DRCnet_u(2,2,32)	62.26	74.12	75.62	70.67
DP- DRCnet_u(3,3,32)	69.09	83.67	87.76	80.17
DP- DRCnet_u(4,4,32)	73.95	84.79	87.3	82.01
DP- DRCnet_d(2,2,32)	66.95	83.7	85.06	78.57
DP- DRCnet_d(3,3,32)	68.56	85.88	89.39	81.28
DP- DRCnet_d(4,4,32)	67.05	85.64	88.62	80.44

图13 识别率对比

Fig.13 Comparison of recognition rate

表6 识别率对比

Table 6 Comparison of recognition rate%

网络名称	-6~ 0/dB	2~ 6/dB	8~ 12/dB	平均识别率
DP-DRCnet(4,4,32)	76.59	90.2	92.71	86.5
CLDNN	65.95	82.82	84.67	77.81
CNN_LSTM	71.87	86.07	88.48	82.14
IQCnet(5,32)	68.01	87.65	88.62	81.4

表7 网络参数与计算量

Table 7 Network parameters and computational load

对比参数	CLDNN	CNN_LSTM	IQCNet(5,24)	DP-DRCnet(4,4,32)
网络参数量	9.81×10⁴	7.8×10⁵	2.39×10⁴	1.9×10⁴
单样本计算量	1.88×10⁷	9.15×10⁶	1.57×10⁶	5.09×10⁵
训练时间/测试时间	3121/867	1209/410	156/87	97/42

图14 F1-score对比(SNR=0)

Fig.14 F1 score comparison(SNR=0)

图15 F1-score得分对比(SNR=12)

Fig.15 F1 score comparison (SNR=12)

图16 DP-DRCnet(4,4,32)识别混淆图(SNR=0)

Fig.16 Recognition confusion map of DP-DRCnet(4,4,32)(SNR=0)

图17 DP-DRCnet(4,4,32)识别混淆图(SNR=12)

Fig.17 Recognition confussion map of DP-DRCnet(4,4,32)(SNR=12)

图18 IQNet(5,32)识别混淆图(SNR=0)

Fig.18 Recognition confusion map of IQNet(5,32) (SNR=0)

图19 IQNet(5,32)识别混淆图(SNR=12)

Fig.19 Recognition confusion map of IQNet(5,32)(SNR=12)

参考文献 21

[1]	赫彬, 苏洪涛. 认知雷达抗干扰中的博弈论分析综述[J]. 电子与信息学报, 2021, 43(5):1199-1211.
	HE B, SHU H T. A review of game theory analysis in cognitive radar anti-jamming[J]. Journal of Electronics & Information Technology, 2021, 43(5):1199-1211. (in Chinese)
[2]	HUANG S, YAO Y Y, YAN X H, et al. Cumulant based maximum likelihood classification for overlapped signals[J]. IEEE Electronics Letters, 2016, 52(21):1761-1763. doi: 10.1049/ell2.v52.21 URL
[3]	SHI Q H, KARASAWA Y. Noncoherent maximum likelihood classification of quadrature amplitude modulation constellations: simplification,analysis,and extension[J]. IEEE Transactions on Wireless Communications, 2011, 10(4):1312-1322. doi: 10.1109/TWC.2011.030311.101490 URL
[4]	张笑宇, 冯永新, 钱博. 一种分数域数字信号调制方式识别方法[J]. 弹箭与制导学报, 2021, 41(1):13-17.
	ZHANG X Y, FENG Y X, QIAN B. A modulation recognition method for digital signals in fractional domain[J]. Journal of Projectiles,Rockets,Missiles and Guidance, 2021, 41(1):13-17. (in Chinese)
[5]	DONG S L, LI Z P, ZHAO L F. A Modulation recognition algorithm based on cyclic spectrum and SVM classification[C]//Proceedings of the 2020 IEEE 4th Information Technology,Networking,Electronic and Automation Control Conference (ITNEC).Chongqing, China:IEEE, 2020: 2123-2127.
[6]	LIANG Y, XIANG X, SUN Y, et al. Novel modulation recognition for WFRFT-based system using 4th-order cumulants[J]. IEEE Access, 2019, 7(1): 86018-86025. doi: 10.1109/Access.6287639 URL
[7]	PENG S L, SUN S J, YAO Y D. A survey of modulation classification using deep learning: signal representation and data preprocessing[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 33(2):7020-7038. doi: 10.1109/TNNLS.2021.3085433 URL
[8]	LÜ H J, ZHOU X, HUO J H, et al. Joint OSNR monitoring and modulation format identification on signal amplitude histograms using convolutional neural network[J]. IEEE Optical Fiber Technology, 2021, 61:102455.
[9]	张思成, 林云, 涂涯, 等. 基于轻量级深度神经网络的电磁信号调制识别技术[J]. 通信学报, 2020, 403(11):16-25.
	ZHANG S C, LIN Y, TU Y, et al. Electromagnetic signal modulation recognition technology based on lightweight deep neural network[J]. Journal on Communications, 2020, 403(11):16-25. (in Chinese)
[10]	刘明骞, 郑诗斐, 李兵兵. 基于深度学习的MPSK信号调制识别[J]. 国防科技大学学报, 2019, 41(5):153-158.
	LIU M Q, ZHENG S F, LI B B. Modulation recognition of MPSK signals based on deep learning[J]. Journal of National University of Defense Technology, 2019, 41(5):153-158. (in Chinese)
[11]	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, 2017,2017, 99:1-1.
[12]	KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[J]. IEEE Advances in Neural Information Processing Systems, 2012: 1097-1105.
[13]	LIN R D, REN W J, SUN X, et al. A hybrid neural network for fast automatic modulation classification[J]. IEEE Access, 2020, 8:130314-130322. doi: 10.1109/Access.6287639 URL
[14]	张天骐, 范聪聪, 葛宛营, 等. 基于ICA和特征提取的MIMO信号调制识别算法[J]. 电子与信息学报, 2020, 42(9):2208-2215.
	ZHANG T Q, FAN C C, GE W Y, et al. MIMO signal modulation recognition algorithm based on ICA and feature extraction[J]. IEEE Journal of Electronics & Information Technology, 2020, 42(9):2208-2215. (in Chinese)
[15]	江伟华, 童峰, 王彬, 等. 采用主分量分析的非合作水声通信信号调制识别[J]. 兵工学报, 2016, 37(9):1670-1676. doi: 10.3969/j.issn.1000-1093.2016.09.017
	JIANG W H, TONG F, WANG B, et al. Modulation recogntion method of non-cooperation underwater acoustic communcation signals using principal component analysis[J]. Acta Armamentarii, 2016, 37(9): 1670-1676. (in Chinese)
[16]	安泽亮, 张天骐, 马宝泽, 等. 基于一维CNN的多入多出OSTBC信号协作调制识别[J]. 通信学报, 2021, 42(7):84-94. doi: 10.11959/j.issn.1000-436x.2021142
	AN Z L, ZHANG T Q, MA B Z, et al. Cooperative modulation recognition based on one-dimensional convolutional neural network for MIMO-OSTBC signal[J]. Journal on Communications, 2021, 42(7):84-94. (in Chinese) doi: 10.11959/j.issn.1000-436x.2021142
[17]	WEST N E, O'SHEA T J. Deep architectures for modulation recognition[C]// Proceedings of IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN). Baltimore,MD,US:IEEE, 2017:1-6.
[18]	崔天舒, 崔凯, 黄永辉, 等. 卷积神经网络卫星信号自动调制识别算法[J]. 北京航空航天大学学报, 2022, 48(6):986-994.
	CUI T S, CUI K, HUANG Y H, et al. Convolutional neural network satellite signal automatic modulation recognition algorithm[J]. Journal of Beijing University of Aeronautics and Astronautics, 2022, 48(6):986-994. (in Chinese)
[19]	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL]. IEEE Computer Science. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.740.6937&rep=rep1&type=pdf,2014-11-18/2021-09-05.
[20]	SZEGEDY C, LIU W, JIA Y, et al. Going deeper with convolutions[EB/OL]. IEEE Computer Society, https://ieeexplore.ieee.org/document/7298594,2015-07-12/2021-09-05.
[21]	HE K M, ZHANG X, REN S Q, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2014, 37(9):1904-1916.