Research On Physical Layer Security Technology Based On Wireless Channel Characteristics And Intelligent Algorithms

Wireless communication technology faces various security issues due to the openness of its transmission medium and its broadcast characteristics.For example,EA,JA,and SA.In addition,the network is highly heterogeneous in wireless communication,and its dynamic topology makes it difficult for a traditional cryptographic-based security system to perform unified security encryption or security management.The terminal in wireless communication is developing toward miniaturization and low power consumption.The traditional cryptography-based security encryption system or security encryption algo-rithm is difficult to perform on the terminal,or obtain the security performance it deserves.The physical layer security technology utilizes the physical layer channel information to process the transmitted signal or implement a corresponding security policy,and has the characteristics of light weight and high security.Therefore,physical layer security tech-nology has been the focus and research of the academic community.This dissertation studies the physical layer security authentication based on wireless channel characteristics of wireless communication,and proposes a multi-user lightweight authentication mechanism in wireless communication scenarios.By combining the deep learning algorithm,the problem of relying on the threshold of the traditional authenti-cation method is solved,which not only accomplishes the threshold freedom,but also achieves the light weight authentication of multiple users.The practical multi-user wire-less communication scenario is simulated by USRP,and the performance of the proposed multi-user authentication algorithms are verified and analyzed.In addition,this disser-tation also studies the physical layer security transmission mechanism based on channel prediction.A variety of channel prediction algorithms based on backpropagation neural networks are proposed.Detailed research content and innovations are as follows:?1?The security threat in multi-user wireless communication is studied.A PLA scheme based on DNN is proposed.This scheme uses CSI to detect spoofing attacks in wireless networks,enhancing the security of wireless communication.A specific PLA mechanism and a PLA scenario including a legal node,a eavesdropping node,and a ma-licious node are given.Then,three multi-user authentication algorithms under different gradient descent algorithms for DNN are proposed,which can accelerate the training of deep neural networks,achieving less computational overhead and lower energy consump-tion.In addition,by deducing the maximum likelihood function of the multi-user authen-tication method,it explains why cross entropy is chosen as the cost function,and its vec-torization cost function is also given.Using a mini batch scheme and L₂regularization to improve training accuracy and avoid overfitting,respectively.In addition,simulation and experimental results show that the PLA method based on DNN can effectively distinguish multiple legitimate nodes,malicious nodes and attack nodes.Finally,by comparing with the traditional hypothesis test method,our proposed methods have better performance.?2?Aiming at the requirement of dynamic environment for environment adaptive authentication,an adaptive multi-user physical layer authentication scheme based on DL is proposed to better adapt to the dynamic authentication environment.An adaptive multi-user authentication algorithm based on DNN,an adaptive multi-user authentication algo-rithm based on CNN and an adaptive multi-user authentication algorithm based on SCPN are given.In addition,accelerated gradient algorithms,mini-batch techniques,and L₂reg-ularization are also used to speed up neural network training and prevent network over-fitting.Through digital simulation and the emulation with USRP in a EC scenario of practical factory environment,the authentication performance of the proposed algorithm is analyzed and evaluated,and the practical significance of the proposed algorithms have been analyzed and discussed.?3?The training of PLA based on machine learning algorithms requires a large num-ber of training samples,which makes the collection of training process time-consuming and difficult to collect large amounts of data in some scenarios.Therefore,we propose three data enhancement algorithms to generate a large amount of training data by using a small number of signal samples collected,speed up the training of the authentication model,and improve the authentication rate.By combining the deep neural network with the data enhancement method,the performance of the deep learning based multi-user PLA scheme is improved and the training speed is accelerated even with fewer training samples.This chapter performs simulation and analysis under the NIST data set.The multi-user PLA algorithm based on DNN is verified in the practical industrial IoT environment,and the three data enhancement algorithms are performed.The algorithms were verified and the results show that the proposed methods have strong practicality and effectiveness.?4?Finally,the secure coding transmission strategy and the design of beamforming based on channel prediction in time-varying channel environment are studied.A channel prediction scheme based on BPNN is proposed,and a single-time channel prediction algo-rithm and multi-time channel prediction algorithm are given.Then the Rayleigh channel impulse response under Jakes model is predicted,and its performance is analyzed.We analysis the performance comparing with the traditional channel prediction algorithm and the performance of different neural network hidden layer numbers,respectively.Further-more,a secure coding transmission strategy and a beamforming design scheme based on channel prediction are proposed.The physical layer security authentication methods proposed in this dissertation are partially simulated in the edge computing scenario.Because the edge server is physically close to the terminal users in the edge computing scenario,it is convenient to extract the physical channel feature parameters of the terminal users,and the edge computing server can provide powerful computing resources for completing the training of the deep neural network.For the terminal nodes,the PLA methods add almost no additional communi-cation overhead or computational cost,which are lightweight authentication methods for edge computing scenarios.The simulated wireless channel characteristic data is obtained under the practical environment,which provides an important reference for the physical layer security from theoretical analysis to practical application.