引用本文
  • 张凤荔,周志远,王瑞锦,黄鑫,韩英军.基于生成对抗网络的无线电数据增扩与分类[J].信息安全学报,已采用    [点击复制]
  • ZHANG Fengli,ZHOU Zhiyuan,WANG Ruijin,HUANG xin,and HAN yingjun.Radio Data Augmentation with GAN for AutomaticModulation Classification[J].Journal of Cyber Security,Accept   [点击复制]
【打印本页】 【在线阅读全文】【下载PDF全文】 查看/发表评论下载PDF阅读器关闭

过刊浏览    高级检索

本文已被:浏览 78次   下载 0  
基于生成对抗网络的无线电数据增扩与分类
张凤荔1, 周志远1, 王瑞锦2, 黄鑫1, 韩英军3
0
(1.电子科技大学信息与软件工程学院;2.电子科技大学信息与软件工程学院 成都中国;3.四川中烟工业有限责任公司成都卷烟厂 成都中国)
摘要:
相较于传统的无线电数据特征提取方法,深度学习具有高效灵活的特点,其在调制数据识别领域有极大的潜力。然而,高质量标记数据的可用性在很大程度上限制了深度学习模型性能。为解决收集无线电调制数据代价昂贵且困难以及调制数据数量影响调制分类模型性能的现实问题,本文提出了基于生成对抗网络(Generative Adversarial Networks,GAN)的数据增扩方法,实现对原始无线电调制数据的增强扩充,基于无线电数据采样方式与传统的无线电调制数据增强特点,设计了能深刻捕获无线电数据空域特征与时序特征的无线电数据分类模型AMCST(Automatic modulation classification based Spatial and Temporal feature),并用一个公共数据集验证了调制增扩方法的数据增扩性能与调制分类模型的分类性能。实验结果表明,基于GAN的数据增扩方法可以极大的提高基于空域特征与时序特征的调制分类模型AMCST在调制分类方面的精度。
关键词:  无线电调制数据增扩  无线电调制数据分类  生成对抗网络  卷积神经网络  循环神经网络  
DOI:
投稿时间:2021-12-22修订日期:2022-05-27
基金项目:国家自然科学基金(61802033,61472064,61602096);四川省区域创新合作项目(2020YFQ0018);四川省科技计划重点研发项目(2020YFG0475,2018GZ0087,2019YJ0543);博士后(2018M643453);广东省国家重点实验室项目(2017B030314131);网络与数据安全四川省重点实验室开放课题(NDSMS201606)资助.
Radio Data Augmentation with GAN for AutomaticModulation Classification
Abstract:
Compared with traditional radio data feature extraction methods, deep learning has the characteristics of high efficiency and flexibility, and it has great potential in the field of modulation data recognition. However, the availability of high-quality labeled data greatly limits the performance of deep learning models. In order to solve the problem that it is expensive and difficult to collect radio modulation data and the amount of modulation data affects the performance of the modulation classification model, this paper proposes a data augmentation method based on Generative Adversarial Networks (GAN) to realize the enhancement and expansion of the original radio modulation data. In the light of radio data sampling method and traditional radio modulation data enhancement characteristics, designed an automatic modulation classification model based Spatial and Temporal feature (AMCST) that can deeply capture the spatial characteristics and temporal characteristics of radio data, and verify the data augmentation performance of augmentation method and the classification performance of modulation classification model with a public data set. The experimental results show that the GAN-based data augmentation method can greatly improve the accuracy of modulation classification based on the spatial and temporal characteristics AMCST.
Key words:  radio modulation data enhancement  radio modulation data classification  generative adversarial networks  convolutional neural networks  recurrent neural networks