Research On Physical Layer Authentication Technology Based On Machine Learning

With the rapid development of the Fifth Generation(5G)mobile communication technology,more and more information needs to be transmitted through free space,and the broadcast characteristics of wireless channels bring opportunities for malicious attacks.Access authentication,as the first step to ensure the security of data transmission,is an important means to ensure the security of communication network.With the rapid increase of terminal devices and resource-limited application scenarios,the traditional cryptography authentication mechanism exposes many shortcomings such as complex key management and high computational overhead.Physical layer authentication can provide an effective authentication method for wireless network by using the imitability of received signal strength(RSS),radio frequency fingerprint(RF),channel state information(CSI)and other physical layer attributes.At the same time,machine learning algorithm can well analyze the physical layer attributes,improve the accuracy of physical layer authentication,and greatly promote the application of physical layer authentication in the actual scene.However,the current research mainly based on the quasi-static communication scenarios and the attack model is assumed to be relatively ideal,lacking of universality in practical application.In the dynamic communication scenario and when the physical characteristics of the attacker is similar to legitimate users,it faces the problem of low accuracy of certification,which can not meet the requirements of authentication accuracy in practical application.In view of these problems,this thesis mainly completes the following research contents:(1)Research on lightweight cross-layer authentication technology for dynamic scenariosAiming at the degradation of authentication performance caused by the rapid change of physical layer attributes in dynamic communication scenarios,a lightweight cross-layer authentication scheme was proposed by combining physical layer authentication with upper layer authentication.Firstly,a lightweight physical layer authentication scheme based on artificial neural network is proposed,which can quickly update model parameters.Furthermore,in order to improve the stability of authentication performance,an efficient cross-layer authentication mechanism is proposed.By introducing the upper-layer authentication to supervise and guide the physical layer authentication model parameter updating,a reasonable balance between authentication efficiency and reliability can be achieved.Simulation results show that the proposed scheme can significantly improve the authentication performance and stability of dynamic communication scenarios.Compared with the traditional physical layer authentication scheme,the delay cost can be reduced by about 25%.(2)Research on close attacker detection technology based on deep learningIn order to solve the problem of low detection accuracy caused by the difficulty in distinguishing the physical layer attributes of the attacker close to the legitimate user,a deep learning-based close attacker detection mechanism was proposed to realize virtual multi-point authentication by flexible sampling of UAV.Furthermore,in order to determine the geographical location of the attacker,an attacker data location algorithm is proposed,which can accurately locate the sample data of the attacker and then determine its geographical location,and provide effective guidance for the human intervention of the attacker.Simulation results show that this scheme can significantly improve the detection accuracy of close range attackers,and the detection rate can be more than 99.99%when the appropriate flight altitude and flight trajectory are selected.(3)Research on physical layer two-fold authentication technology based on time series modelAiming at the problem of low detection accuracy caused by the difficulty in distinguishing the physical layer attributes of the close attacker from the legitimate user,a time series model based physical layer two-fold authentication mechanism was proposed by fully mining the higher-order features of physical layer attribute changes and combining the lower-order features with the higher-order features.Specifically,the two-fold authentication mechanism consists of a lightweight authentication module and a time series model based authentication module.The lightweight authentication module is used to quickly detect most common attackers and determine their attack targets,while the time series model based authentication module is used to accurately identify the attackers.The simulation results show that the proposed scheme can significantly improve the detection accuracy of close attackers whose physical layer attribute value is still within the range of legal threshold value but the change trend is abnormal,and the detection rate is more than 98%.