摘要
文章针对现实环境下无线电通信环境复杂,通信方式复杂多变,且人工提取无线电特征工作量巨大又存在不稳定因素,提出基于神经网络学习的无线电特征提取方法。用深度卷积神经网络代替传统的特征提取算法对无线电时域信号进行编码学习;然后通过深度学习特征矢量的相似度,实现同类调制信号相同特征的自动匹配;最后,利用训练数据的类别标签信息自上而下对无线电特征进行微调,得到无线电深度学习表示向量以此训练全连接分类器实现无线电调制分类。实验结果表明,本方法能够有效克服传统线性分类方法的缺点,有效地提升无线电调制方式分类性能。
This paper proposes a radio feature extraction method based on neural network learning in view of the fact that the radio communication environment is complicated and the communication method is complex and changeable. Using deep convolutional neural network instead of traditional feature extraction algorithm to encode the radio time domain signal coding learning. Then through the similarity of deep learning feature vector, to achieve the same feature of the same kind of modulation signal automatic matching. Finally, the use of training data class label information the radio characteristics are fine-tuned from top to bottom, and radio deep learning representation vectors are obtained to train the fully-connected classifier for radio modulation classification. Experimental results show that this method can effectively overcome the shortcomings of traditional linear classification methods and effectively improve the classification performance of radio modulation methods.
作者
何学智
林林
黄自力
解金豹
He Xuezhi;Lin Lin;Huang Zili;Xie Jinbao(Fujian Newland Computer Co., Ltd., Fuzhou 350015, China;Hefei University of Technology, Hefei 230009, China)
出处
《无线互联科技》
2018年第11期20-24,共5页
Wireless Internet Technology
基金
装备预研基金
项目名称:基于深度学习的无线电信号特征提取技术研究
项目编号:6140137050207
福建省区域科技发展项目
项目名称:基于深度学习的嵌入式视觉感知技术研究和应用
项目编号:2018H4004
中央引导地方科技发展专项
项目名称:工业物联网感知识别技术创新平台
项目编号:2017L3016
关键词
无线电
特征提取
深度学习
非线性映射
卷积神经网络
radio
feature extraction
deep learning
nonlinear mapping
convolutional neural network
