基于注意力残差网络的Wi-Fi设备的射频指纹识别  

Radio frequency fingerprint identification of Wi-Fi device based on attention residual network

作　　者：张凯 郭剑黎 胡军星 任俊霞 谭磊 ZHANG Kai;GUO Jianli;HU Junxing;REN Junxia;TAN Lei(State Grid Henan Electric Power Company,Zhengzhou 450000;Henan Jiuyu Tenglong Information Engineering Co.,Ltd.,Zhengzhou 450052)

机构地区：[1]国网河南省电力公司,郑州450000 [2]河南九域腾龙信息工程有限公司,郑州450052

出　　处：《南京信息工程大学学报（自然科学版）》2022年第3期324-330,共7页Journal of Nanjing University of Information Science & Technology（Natural Science Edition）

基　　金：国家自然科学基金(61673067)。

摘　　要：由于无线介质的开放性,传统的基于安全协议的无线网络安全存在隐患,基于物理层的射频指纹(RFF)识别,具有特征难以伪造的优点,能有效提高无线网络的安全性.针对多场景、多设备识别任务,构建了基于注意力残差卷积神经网络的射频指纹识别方法.实验采集构建了完备的数据集,数据集包含32个Wi-Fi模块,覆盖802.11b标准的2.4 GHz模块.对比结果表明:该方法在32个Wi-Fi模块的识别中达到90%的识别精度,高于传统算法86%的识别率和卷积神经网络方法的89%的识别率;不同采样率的数据集在2 dB时均可以达到90%以上的识别精度,最终在信噪比(SNR)大于20 dB时,识别精度可以达到96%.The openness of wireless media has been a security threat for traditional wireless network based on security protocol.While the Radio Frequency Fingerprint(RFF)identification is based on physical layer security,and considering the RFF is impossible to forge,the RFF identification can effectively improve the security of wireless network.Aiming at the multi-scene and multi-device identification,an RFF identification approach is constructed based on attention residual convolution neural network.The dataset contains 32 Wi-Fi modules,covering the 2.4 GHz module of 802.11b standard.The comparison results show that the recognition accuracy of the proposed approach is 90%for the 32 Wi-Fi modules,higher than that of traditional algorithm(86%)and convolutional neural network approach(89%);the recognition accuracy can be higher than 90%on the dataset with different sampling rates when the SNR is greater than 2 dB,which can reach as high as 96%when the SNR is greater than 20 dB.

关 键 词：射频指纹 设备识别 注意力残差网络 物理层安全 通信安全 

分 类 号：TN918[电子电信—通信与信息系统]