| With the rapid development of Internet of Things technology,common authentication technologies,such as authentication based on device MAC address,security certificate,identity authentication instruction,etc.,are easy to counterfeit and lost.Using physical characteristics of wireless signals as the device radio frequency fingerprint(RFF)can realize the classification and identification of devices,which is of great significance to the Io T physical layer security research.This article is aimed at LoRa devices.The modulation technology of LoRa signal and its physical frame structure are analyzed.Three RFF extraction methods are put forward and the effectiveness and stability of these methods is verified through experiments on 8 devices.A LoRa identification system based on physical layer RFF is designed and implemented,and its effectiveness working as a physical layer security system through experiments is tested.The main work of this article is as follows:1.The modulation technology and physical frame structure of LoRa signal are analyzed,as well as its basic symbol(Chirp symbol)in time and frequency domain.At the same time,the signal synchronization algorithm?the concept of IQ tracing figure and the signal demodulation method are given for subsequent RFF extraction and system design.2.The influence of carrier frequency offset on the signal is studied,and a coarse frequency offset estimation using the average of the preamble frequency is proposed.After the coarse frequency offset compensation,the precise frequency offset compensation based on differential calculation is carried out.The overall frequency offset characteristic consists of the coarse and precise frequency offset.Experiments show that even in the case of low signal-to-noise ratio(7.5d B),the recognition accuracy reaches more than 90%.3.In order to increase the RFF dimension,signal characteristics related to IQ offset are studied.The error amplitude vector and the overall signal IQ offset are proposed as two RFF characteristics by comparing the received signal with local standard signal.Experiments show that these two characteristics have high requirements on the signal-to-noise ratio of signal.When the SNR is 30 d B,the recognition accuracy can only achieve approximately 70%.4.Because of the linear frequency modulation characteristic,signal has rich characteristic components in frequency domain,therefore the cross-power spectral density characteristic is studied.This characteristic can be obtained by calculating the preamble correlation coefficient of the received signal and the local standard signal at each frequency point.And experiments indicate that the highdimensional feature has better performance after dimensionality reduction.When the proportion of feature after dimensionality reduction to original characteristic is 99%,the recognition accuracy is close to 90%;when the SNR is 0d B,the recognition accuracy is approximately 70%,which has better performance than the above two RFF features.5.In order to obtain the most suitable classification algorithm for the system,multiple algorithms are analyzed,and three RFF characteristics are tested under three classification algorithms:linear discriminant analysis(LDA),linear support vector machine(SVM)and Gaussian SVM.Finally it is determined that the mixed features of the three characteristics and LDA can be used in system model training.In addition,because the dimension of the mixed features is too high,it is necessary to reduce its dimension first and then perform the classification.6.A LoRa identification system based on physical layer RFF is designed and implemented.Its functional integrity and the recognition accuracy in multiple scenarios are tested through experiments.In three experimental scenarios: indoor scene 1 with short distance and line-of-sight communication,indoor scene 2 with long distance and no line-of-sight communication,and scene 3 with long distance and wall between transceivers,the accuracy is up to 99%?94% and 83% respectively. |
PDF Full Text Request