Research On Wireless Device Security Authentication Based On RF Fingerprint Identification Technology

With the rapid development of 5G technology,wireless network communication technology has derived rich application scenarios in various fields such as national defense and military,livelihood services and medical education,etc.The growth of application demand is accompanied by the demand for fast and secure access authentication of massive wireless communication terminals.Device authentication is the security foundation of wireless network communication,and wireless communication is more vulnerable to malicious attacks due to its openness.Facing increasingly serious security threats and challenges,a reliable and lightweight secure authentication method is urgently needed.The traditional security authentication access method based on cryptography theory has so far suffered from a series of problems such as key leakage and third-party attacks.And wireless communication devices due to its hardware tolerance generated by the RF fingerprint is the inherent characteristics of the device,with the physical characteristics of difficult to copy.The security authentication of wireless devices based on RF fingerprint technology expects to solve the problem of secure authentication access of wireless communication by extracting the hardware difference of wireless devices as the unique hardware identification of the devices.In this thesis,we study and analyze different RF fingerprint features,including those extracted based on a priori knowledge and data reduction means,design the security authentication test process,compare the performance of different security authentication algorithms,and build a targeted neural network for authentication.Firstly,this thesis investigates the RF fingerprint features based on orthogonal frequency division multiplexed signals,derives and extracts the RF fingerprint features of signals by using a priori knowledge in the communication field and data downscaling means,analyzes their advantages and disadvantages as RF fingerprint features for different features,and performs joint estimation simulation for frequency bias and timing error.On this basis,the separability of various RF fingerprints is quantified and analyzed for each kind of RF fingerprint.In order to visually verify the differentiation degree of RF fingerprint features,differentiation degree tests are performed using optimized classifiers based on different features to verify their separation degree,classification accuracy,and noise robustness,in which the error vector and leading correlation noise are more robust,the frequency bias and timing error are less robust,Self-encoder extracted features have better noise immunity.Then,this thesis introduces a variety of security authentication algorithms including traditional machine learning domain algorithms as well as deep single classification networks combined with neural networks.Based on the frame structure characteristics of the communication signal,optimizes the design of the network model to enhance the network’s extraction capability for RF fingerprint features.On this basis,different security authentication models are tested based on RF fingerprint features to verify their authentication accuracy and noise immunity,and to compare the applicability and advantages and disadvantages of different models in RF fingerprint-based authentication scenarios.Finally,this thesis builds an RF fingerprint demonstration and verification platform based on the universal software radio platform and its supporting software laboratory virtual instrument engineering platform,and provides a detailed introduction of the platform’s modular design,display interface,and implementation principles.Performance tests of the network model proposed in this thesis are conducted based on the data collected by the platform,and its authentication performance based on actual signals is verified.