| 摘要: |
| 近年来，利用射频指纹（Radio Frequency Fingerprint，RFF）技术对设备进行识别认证，构建保密通信系统成为研究的热点。相比于传统的认证体制，射频指纹利用设备本身的硬件特性进行识别，具有更高的安全性。与其他射频技术相比，Wi-Fi信号频谱更宽，应用更加广泛，但也更容易受室内多径干扰，造成对Wi-Fi射频指纹识别率下降的问题。针对这一问题，本文提出一种基于功率谱特征的Wi-Fi射频指纹提取方法，通过计算其信号帧中短导码符号和长导码符号的功率谱比值，并以此比值作为射频指纹特征。本文采用了27台Wi-Fi路由器进行实验，在室内场景中模拟的四个不受外界干扰的相对静止情形以及简单的移动环境中采集数据，运用随机森林模型进行训练和测试，识别率能达到93.3%。理论分析和实验结果表明，本文方法能够较好地抵抗多径效应和加性噪声对射频指纹的影响，即使设备在相对移动的情况下，提取的射频指纹信息也具有较好的稳健性。因此，本文所提的功率谱特征方法在物理层设备认证和身份识别领域具有一定的应用价值。 |
| 关键词: 物理层安全 射频指纹 设备识别 软件无线电 随机森林 |
| DOI：10.19363/J.cnki.cn10-1380/tn.2021.03.01 |
| 投稿时间：2020-03-27修订日期：2020-04-27 |
| 基金项目:本课题得到江苏省产业前瞻与关键核心技术—竞争项目（No.BE2019109），网络通信与安全紫金山实验室，移动通信国家重点实验室自主研究经费（No.2020B05）资助。 |
|
| Power Spectrum Based Wi-Fi RF Fingerprint Extraction Method |
| CHEN Tianshu,HU Aiqun,JIANG Yu |
| School of Cyber Science and Engineering, Southeast University, Nanjing 211189, China;School of Information Science and Engineering, Southeast University, Nanjing 210096, China;The Purple Mountain Laboratories for Network and Communication Security, Nanjing 211111, China;School of Cyber Science and Engineering, Southeast University, Nanjing 211189, China;The Purple Mountain Laboratories for Network and Communication Security, Nanjing 211111, China |
| Abstract: |
| In recent years, using Radio Frequency Fingerprint (RFF) technology to realize the identification and authentication of devices and build a secure communication system has become a research hotspot. Compared with the traditional authentication systems, RFF identification has higher security by utilizing the hardware characteristics of the devices themselves. Compared to other radio frequency technologies, Wi-Fi signals have broader spectrum and wider application. However, they are also more susceptible to indoor multipath interference, resulting in the decline of the recognition accuracy of Wi-Fi devices. To solve this problem, a novel Wi-Fi RFF extraction method based on power spectrum characteristics is proposed in this paper, in which the power spectrum ratio of short training symbols (STS) and long training symbols (LTS) in the signal frame is calculated as fingerprint feature. In the experiment, the frame data of 27 routers were collected in four relatively static situations without external disturbance and simple moving states simulated in indoor scene, and then they were trained and tested by using random forest model. The recognition rate can reach 93.3%. The theoretical analysis and experimental results demonstrate that this method can better resist the influences of multipath and additive noises, and the extracted RFF characteristics still have good robustness even when the devices are in motion. Therefore, the power spectrum based method presented in this paper has a certain application value in the field of physical layer device authentication and identity recognition. |
| Key words: physical layer security radio frequency fingerprint device identification software defined radio random forest |
