WiFi Device Identification Technology Based On Radio Frequency Fingerprint

As the era of the Internet of Everything kicks off,the topic of secure connectivity between devices is becoming more and more important.Some scholars have proposed the concept of endogenous security to solve the problem in today’s complex interconnected networks.However,in wireless connectivity,the open nature of electromagnetic signals makes this connection process highly vulnerable to various attacks.Therefore,the identification and authentication process between devices should have a security solution based on the wireless side in addition to a security solution based on the network side.RF fingerprinting technology provides a solution to this problem by providing physical layer identification from the wireless side and improving the security of the network.In this paper,we will study the device identification problem in the WiFi device connection process.And we design a WiFi device security identification framework using a device identification scheme based on RF fingerprint technology.The main research content is as follows.Firstly,this paper studies the basic theory related to RF fingerprinting.The mechanism of RF fingerprint generation is analyzed,the mathematical process of each device of the transmitter is studied,and its influence on RF fingerprint and the mathematical model of RF fingerprint of radiation source devices is constructed.A WiFi device data acquisition platform is built,and the RF signals of 298 WiFi devices are collected in a variety of channel environments to construct the actual measurement dataset.Secondly,this paper studies the RF fingerprinting technique of the feature engineering method.The power spectral density analysis and bispectral analysis of signals are introduced,and the improved spectral analysis of signal zero-padding is introduced,based on which the improved RF fingerprint features based on signal zero-padding are proposed,i.e.,ZP-PSD and ZP-BS features.The experimental analysis is carried out through the measured data set to evaluate the RF fingerprint features proposed in this paper from several angles,and the effectiveness of the features proposed in this paper is proved by comparing and analyzing the noise immunity performance,environmental adaptability,and feature capacity with the existing methods.And,the phenomenon that increasing the sampling rate of the signal at the receiver side can improve the recognition accuracy of the system to a certain extent is found through experiments.Finally,this paper studies the deep learning method for RF fingerprinting.The convolutional neural network and its computation process are introduced,the general structure of the deep residual network is introduced,the complex convolutional layer and complex fully connected layer is introduced to build the complex residual network,and the network structure is adapted to plural RF signal processing.Combining feature engineering with deep learning,ZP-PSD features are used as the input of the complex residual network to build a deep learning model based on spectral analysis features and improve the recognition performance of the deep learning model.Experimental analysis is performed through real measurement data sets to evaluate the RF fingerprint features proposed in this paper from multiple perspectives,and the effectiveness of the system model proposed in this paper is demonstrated by comparing and analyzing the noise immunity performance,environmental adaptability,and user capacity with feature engineering methods.