首页期刊简介编委会征稿启事出版道德声明审稿流程读者订阅论文查重联系我们English
引用本文
  • 幸晨杰,王良刚.应用深度神经网络和集成学习的电台个体识别[J].电讯技术,2021,61(9): - .    [点击复制]
  • XING Chenjie,WANG Lianggang.Radio individual recognition based on deep neural network and ensemble learning[J].,2021,61(9): - .   [点击复制]
【打印本页】 【下载PDF全文】 查看/发表评论下载PDF阅读器关闭

←前一篇|后一篇→

过刊浏览    高级检索

本文已被:浏览 1171次   下载 31 本文二维码信息
码上扫一扫!
应用深度神经网络和集成学习的电台个体识别
幸晨杰,王良刚
0
(中国西南电子技术研究所,成都 610036)
摘要:
提出了一种基于深度神经网络的个体智能识别方法,可用于电台个体分类识别。该方法构建集成多子网络的一维深度卷积模型,以电台时序信号作为模型输入,进行电台个体分类。利用深度神经网络自动特征化的能力,该方法从时序信号中自动获取个体特征,从而以端到端的形式实现从电台信号识别电台个体。该方法能够免去基于专家知识的特征提取工作,自动提取的个体深度特征还有助于区分传统特征无法区分的高度相似电台个体。实验证明,该方法能有效降低模型调参设计难度,能减轻单一网络带来的特征提取识别过拟合问题,能提高电台个体识别算法的泛化能力与鲁棒性。在信噪比12 dB的条件下,对10类电台8PSK调制信号进行特征提取与识别,整体正确率91.83%,平均正确率为89.12%;对MSK调制信号进行特征提取与识别,平均分类精度为89.1%。
关键词:  电台个体识别  深度神经网络  集成学习  特征提取
DOI:
基金项目:
Radio individual recognition based on deep neural network and ensemble learning
XING Chenjie,WANG Lianggang
(Southwest China Institute of Electronic Technology,Chengdu 610036,China)
Abstract:
A radio individual recognition method based on deep neural network model is proposed.This one-dimensional deep neural network model consists of multiple sub-networks.Radio signal time series are input to this model,and the radio corresponding to each input signal is thereafter recognized.This method utilizes the capability of automatically extracting data features by deep neural networks,to extract fingerprint features from radio signal time series,thus recognizing radio individuals from signal series in an end-to-end manner.Consequently,this method can save feature extraction labor which relies on expert knowledge,and can automatically extract deep features to help recognize very similar radio individuals,which can hardly be distinguished through very similar conventional features.Experiments suggest this method is effective in simplifying parameter adjustment in model optimizing,alleviating over-fitting by single network,and improving generalization and robustness of the radio recognition algorithm.Under the circumstance of 12 dB signal-to-noise ratio(SNR),8PSK modulated signals from 10 radio individuals are classified,where overall accuracy is 91.83% and mean accuracy is 89.12%.Under the same SNR,MSK modulated signals are classified with 89.1% overall accuracy.
Key words:  radio individual identification  deep neural network  ensemble learning  feature extraction
安全联盟站长平台