Research On Intelligent Physical Layer Security Authentication Technologies

Physical layer security provides a paradigm shift for the security of fifth generation and beyond wireless networks because of its low computational complexity and low overhead.However,due to the complex and dynamic na-ture of the wireless environment,trad.itional physical layer authentication has three weaknesses:low reliability of using single feature,difficulty in learn-ing time-varvying attributes,and privacy leakage.This dissertation introduces intelligence into physical layer security to improve reliability and accuracy of authentication.The specific contributions and results are summarized as fol-lows:Considering that the incomplete channel estimation may cause the low reli-ability of conventional physical layer authentication scheme,this thesis has ex-panded from the perspective of channel feature extraction and classifier.Differ-ent from most existing authentication techniques that rely on the single channel feature,the proposed scheme utilizes two-dimensional features and relation-ships to achieve enhancement.The intrinsic relationship between channel fea-ture dimension and security performance is studied.Specifically,an algorithm based on Karhunen-Loeve transform is designed to obtain the optimal channel characteristics.Furthermore,a Gaussian Mixture Model(GMM)is proposed to detect spoofing attackers.Probabilistic models of different transmitters are used to cluster signal data.Training data for a spoofer operating through an un-known channel,a pseudo adversaiy model is developed to enhance the spoofing detection performance.Monte Carlo simulations are used to evaluate the de-tection performance of the GMM-based physical layer authentication scheme.The results show that the probability of detecting a spoofer is higher than that obtained using similar approaches.Due to the time-varying environment,those one-time physical layer au-thentication methods may be limited in detecting illegitimate users.In this the-sis,an intelligent learning algorithm is proposed based on a convolutional neural network.Specifically,a PCA-based feature reconstruction matrix is designed to track time-varyina physical attributes.By utilizing convolutional neural net-work as an intelligent authentication algorithm,the proposed scheme concur-rently integrates"blind feature" extraction with spoofing attacker authentica-tion,hence,leading to effective physical layer security.Convolutional neural networks map channel characteristics to deeper dimensions.Moreover,this thesis prototypes the proposed scheme on the USRP platform and verify its re-liability in a conference room.Experimental results show that the convolutional neural network can tackle the challenges of physical layer authentication under the interference conditions,which indicates its adaptability to a time-varying environment.Although multi-dimensional channel characteristics and intelligent algo-rithms may achieve better authentication performance,accurate learning and mining of Alice during the authentication process may lead to leakage of its private information.Therefore,the privacy protection for Alice is very impor-tant.This thesis studies secure communications in multiuser wireless networks where full-duplex(FD)jamming operates to enhance physical layer security.The considered multiuser system is equipped with FD legitimate receivers in contrast to conventional frameworks where a half-duplex(HD)receiver is at hand.An alternative solution is to exploit the FD capability of the receiver to transmit jamming signals against the eavesdropper.In the ideal case where full-duplex self-interference is completely eliminated,the best user selection scheme is exploited to strengthen the secure performance of the system.This thesis first analyzes the optimal power allocation strategy with the knowledge of instantaneous channel state information(CSI)of every channel.Then the se-curity level of the proposed scheme is investigated through the ergodic secrecy rate(SR)and secrecy outage probability with only the statistical knowledge of CSI.Simulation results show that the application of user selection and FD communication lead to a significant improvement.In the actual scenario where full-duplex self-interference can not be completely suppressed,the impact of self-interference and channel interference on physical layer security is investi-gated.Moreover,this thesis applies a reinforcement learning algorithm,called Q-learning,to model the interaction between the source and multiple jamming users.Simulation results verify the effectiveness of the proposed scheme.