基于深度学习的复合神经网络机场信号检测框架

侯进; 吕志良; 徐茂; 吴佩军; 刘雨灵; 张笑语; 陈曾

doi:10.3969/j.issn.0258-2724.20180164

基于深度学习的复合神经网络机场信号检测框架

doi: 10.3969/j.issn.0258-2724.20180164

详细信息

作者简介:
侯进（1969—），女，副教授，博士，研究方向为计算机图形学、机器学习、深度学习、人工智能，E-mail：jhou@swjtu.edu.cn

中图分类号: V221.3
计量
- 文章访问数: 413
- HTML全文浏览量: 191
- PDF下载量: 25
- 被引次数: 0
出版历程
- 收稿日期: 2018-03-10
- 修回日期: 2018-07-02
- 网络出版日期: 2018-07-08
- 刊出日期: 2019-08-01

Combined Neural Networks Based on Deep Learning for Signal Detection in Aeronautical Communications

摘要

摘要: 针对信号调制识别对复杂通信环境缺乏适应性与精度不足的问题，提出一种基于深度学习的多特征复合神经网络框架. 该框架首先使用前端卷积神经网络检测信号载波特征，再对前端初筛选信号执行预处理将其转换为信号时频图，最后设计了后端轻量化卷积神经网络,检测信号时频特征. 基于TensorFlow平台的复合神经网络对机场真实信号检测精度达到99.23%，实验表明该方法可有效应用于实时机场信号检测.
- 调制识别 /
- 卷积神经网络 /
- 深度学习 /
- 载波特征 /
- 时频特征
Abstract: In order to increase the generality and accuracy of radio modulation recognition in complex radio propagation environment, a multiple feature combined convolutional network system based on deep learning is proposed. Carrier features were detected with front convolutional network in the first stage. Then, the signal filtered by the front CNN was converted into spectrograms with the proposed pre-process method. Finally, the lightweight backend convolutional network was designed to extract the time-frequency features of spectrograms. The networks, which run on TensorFlow, achieved 99.23% accuracy with real airport communication signals. The experiment indicates that the proposed networks could be applied in real-time airport radio detection.
- modulation recognition /
- convolutional neural network /
- deep learning /
- carrier features /
- time-frequency features

HTML全文

图 1 系统框架

Figure 1. System framework

下载: 全尺寸图片幻灯片

图 2 前端CNN权值层

Figure 2. Weight layers of front CNN

下载: 全尺寸图片幻灯片

图 3 信号预处理过程

Figure 3. Signal data pre-process

下载: 全尺寸图片幻灯片

图 4 后端原型CNN构件

Figure 4. Prototypical backend CNN components

下载: 全尺寸图片幻灯片

图 5 拆解卷积核

Figure 5. Convolutional kernel deconstruction

下载: 全尺寸图片幻灯片

图 6 残差连接

Figure 6. Residual connection

下载: 全尺寸图片幻灯片

图 7 前端CNN与决策树混淆矩阵

Figure 7. Confusion matrices of front CNN and decision-tree

下载: 全尺寸图片幻灯片

图 8 前端CNN与文献[8]验证精度曲线

Figure 8. Validation curves of front CNN and Ref. [8]

下载: 全尺寸图片幻灯片

图 9 后端CNN第三层卷积提取特征

Figure 9. Extracted features of the third convolutional layer of backend CNN

下载: 全尺寸图片幻灯片

图 10 后端CNN验证精度曲线

Figure 10. Validation curves of backend CNN

下载: 全尺寸图片幻灯片

图 11 后端与V2 CNN混淆矩阵

Figure 11. Confusion matrices of backend CNN and V2 CNN

下载: 全尺寸图片幻灯片

图 12 后端CNN单通道灰度训练

Figure 12. Single-channel grey images for backend CNN

下载: 全尺寸图片幻灯片

图 13 各项试验的精度与耗时

Figure 13. Accuracy and consumed time of experiments

下载: 全尺寸图片幻灯片

表 1 决策树与前端CNN对比结果

Table 1. Comparison between decision-tree and front CNN

方法识别准确率/% 耗时/s

前端 CNN 90.1 68.5
决策树 81.3 33.8

下载: 导出CSV

表 2 前端CNN与文献[8]CNN对比结果

Table 2. Comparison of front CNN and that in Ref. [8]

方法识别准确率/% 耗时/s CNN参数量

前端 CNN 90.1 68.5 707 756
文献[8] CNN 89.3 112.7 2 828 628

下载: 导出CSV

表 3 后端CNN与V2 CNN对比结果

Table 3. Comparison of backend CNN and V2 CNN

方法识别准确率/% 耗时/s CNN参数量

后端 CNN 99.19 1 864.6 685 380
V2 CNN 99.96 2 343.3 58 837 142

下载: 导出CSV

表 4 基于复合CNN的实验

Table 4. Combined CNN experiment

方法识别准确率/% 耗时/s

复合 CNN 99.23 411

下载: 导出CSV

参考文献(20)

温欣. 基于决策树的调制模式及GNU Radio模块实现[D]. 哈尔滨: 哈尔滨工业大学, 2010

LI Shiping, CHEN Fangchao, WANG Long. Modulation recognition algorithm of digital signal based on support vector machine[J]. Control and Decision Conference (CCDC), 2012, 229(5): 3326-3330.

AHN W H, NAH S P, SEO B S. Automatic classification of digitally modulated signals based on k-nearest neighbor[J]. Lecture Notes in Electrical Engineering, 2015, 329(1): 63-69.

WONG M L D, TING S K, NANDI A K. Naïve Bayes classification of adaptive broadband wireless modulation types with higher order cumulants[C]//International Conference on Signal Processing and Communication Systems. Gold Coast: [s.n.], 2009, 5(2): 1-5

XU J L, SU Wei, ZHOU Mengchu. Distributed automatic modulation classification with multiple Sensors[J]. IEEE Sensor Journal, 2010, 10(11): 1779-1785. doi: 10.1109/JSEN.2010.2049487

HELMY M O, ZAKI F W. Identification of linear dimensional digital modulation schemes via clustering algorithms[C]//International Conference on Computer Engineering & Systems. Cairo: [s.n.], 2009, 5(1): 358-390

HARING L, CHEN Y, CZYLWIK A. Automatic modulation classification methods for wireless OFDM system in TDD mode[J]. IEEE Transaction on Communications, 2010, 58(9): 2480-2485. doi: 10.1109/TCOMM.2010.080310.090228

O’SHEA T J, CORGAN J, CLANCY T C. Convolutional radio modulation recognition networks[C]//International Conference on Engineering Applications of Neural Networks. Aberdeen: [s.n.], 2016, 6(1): 213-226

O’SHEA T J, WEST N. Radio machine learning dataset generation with gnu radio[C]//Proceedings of the GNU Radio Conference. Boulder: [s.n.], 2016, 1(1): 69-74

WEST N E, O’SHEA T J. Deep architectures for modulation recognition[C]//IEEE International Symposium on Dynamic Spectrum Access Networks. Baltimore: [s.n.], 2017: 1-6

LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324. doi: 10.1109/5.726791

SIMONYAN K, ZISSERMAN A. Very deep convoluiotnal networks for large-scale image recognititon[DB/OL]. [2018-02-22]. https://arxiv.org/abs/1409.1556

HUBEL D H, WIESEL T N. Receptive fields,binocular inter-action and functional architecture in the cat's visual cortex[J]. The Journal of Physiology, 1962, 160(1): 106-154. doi: 10.1113/jphysiol.1962.sp006837

FUKUSHIMA, K NEOCOGNITRON. A self organizing neural network model for a mechanism of pattern recognition un-affected by shift in position[J]. Biological Cybernetics, 1980, 36(4): 193-202. doi: 10.1007/BF00344251

IOFFE S, SZEGEDY C. Batch normalization: accelerating deep network training by reducing internal covariate shift[C]//International Conference on Machine Learning. Lille: [s.n.], 2015, 33(1): 448-456

SZEGEDY C, VANHOUCKE V, IOFFE S, et al. Rethinking the inception architecture for computer vision[C]//IEEE Conference on Computer Vision & Pattern Recognition. Las Vegas: IEEE, 2016, 26(1): 2818-2826

HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//IEEE Conference on Computer Vision & Pattern Recognition. Las Vegas: IEEE, 2016, 26(1): 770-778

LIN M, CHEN Q, YAN S. Network in network[DB/OL]. [2018-02-22]. https://arxiv.org/abs/1312.4400

KRIZHEVSKY A, SUTSKEVER I, HINTON G. Imagenet classification with deep convolutional neural networks[C]//International Conference on Neural Information Processing Systems. Lake Tahoe: [s.n.], 2012, 60(2): 1097-1105

SZEGEDY C, IOFFE S, VANHOUCKE V. Inception-v4, inception-resnet and the impact of residual connections on learning[C]// Thirty-First AAAI Conference on Artificial Intelligence. San Francisco: [s.n.], 2017: 936-940

施引文献

附加材料(0)

访问统计

点击查看大图

图(13) / 表(4)

计量

文章访问数: 413
HTML全文浏览量: 191
PDF下载量: 25
被引次数: 0

方法	识别准确率/%	耗时/s
前端 CNN	90.1	68.5
决策树	81.3	33.8

方法	识别准确率/%	耗时/s	CNN参数量
前端 CNN	90.1	68.5	707 756
文献[8] CNN	89.3	112.7	2 828 628

方法	识别准确率/%	耗时/s	CNN参数量
后端 CNN	99.19	1 864.6	685 380
V2 CNN	99.96	2 343.3	58 837 142

基于深度学习的复合神经网络机场信号检测框架

doi: 10.3969/j.issn.0258-2724.20180164

作者简介: 侯进（1969—），女，副教授，博士，研究方向为计算机图形学、机器学习、深度学习、人工智能，E-mail：jhou@swjtu.edu.cn

计量

出版历程

Combined Neural Networks Based on Deep Learning for Signal Detection in Aeronautical Communications

计量

出版历程

目录

作者简介:
侯进（1969—），女，副教授，博士，研究方向为计算机图形学、机器学习、深度学习、人工智能，E-mail：jhou@swjtu.edu.cn