Research On RF Fingerprint Spoofing Techniques And Countermeasures

With the development of wireless network technology,various Internet of Things applications have gradually been introduced into people's lives in all aspects.The security of information interaction in wireless networks has become increasingly important.Due to the openness of the wireless communication medium,normal communication of legitimate users may be interfered or disrupted by malicious users through eavesdropping and deception.Under this circumstance information security of legitimate users is threatened.The current device identity authentication mechanism based on high-level encryption is facing more severe challenges under the pressure of malicious users eavesdropping and increasing computing power.Because of radio frequency(RF)fingerprint's stability and uniqueness,it is considered to be an effective device identification method,which can effectively improve the security of the wireless access stage of the network.Further research on RF fingerprint-based device identification mechanism is still needed.The extraction process of RF fingerprints usually uses the signal pre-processing technology,which may cause a certain degree of information loss.Besides,there is insufficient research on the anti-camouflage of RF fingerprints,and the device's RF fingerprints are at risk of being disguised by malicious users.In view of the above shortcomings,two aspects of the device identification mechanism based on RF fingerprint are studied.A device identification method based on deep neural network is proposed.This method uses samples which are from the original sampling of wireless signal to directly learn and extract the RF fingerprint features.Deep neural network is used in order to learn the features in the RF signal for device identification.Experiment results prove that the device identification method based on deep neural network has better identification ability than traditional machine learning technologies.The security risks that malicious users may camouflage legitimate users' RF fingerprints are studied.First,the fingerprint spoofing method based on passive eavesdropping is analyzed.Because the malicious user's own device also has some certain defects,it is difficult to simply emulate the RF fingerprint by analyzing the characteristics of the eavesdropped legitimate user signal.In this regard,a method of RF fingerprint spoofing attack based on multi-user collaboration is proposed.In the process of fingerprint spoofing,a cooperative attacker is introduced to act as an observer.This observer compares the difference between the RF fingerprints of the attacker and the legitimate user.The comparison result is fed back to the attacker,which is used to help the attacker to improve the fingerprint spoofing method.Cooperative attackers and observer cooperate to continuously improve the quality of fingerprint camouflage and realize fingerprint spoofing attacks.This process is modeled as a generative adversarial network,enabling fingerprint spoofing through machine learning.Experimental results show that this method can achieve good fingerprint spoofing in the noisy environment and can also deceive fingerprint identifiers with different machine learning methods.Furthermore,the fingerprint identifier is trained by using the high-quality adversarial samples generated during the training of the generated adversarial network,which effectively improves the performance of the fingerprint identifier.The simulation results indicate that the fingerprint identifier enhanced by the adversarial sample training can effectively discover the spoofed fingerprint,which significantly enhances the anti-spoofing ability of the device identification system based on the RF fingerprint.