Research On Key Technology Of Physical Layer Security In Intelligent Communication

With the commercial expanding of the fifth generation mobile communication,three main scenes of application in 5G which are enhanced mobile broad band,massive machine type communication and ultra-reliability low latency communication no longer satisfy the demands of human society.The 6th generation mobile communication which is destined to transcend 5G?s key performance indicators will be developed towards a new direction of global coverage,all spectra,full application and strong security.In order to improve the quality of function and service in 6G,artificial intelligent mobile communication system become the inevitable trend of 6G.Considering AI's marvelous ability of parallel computing and end-to-end optimization,the combination of AI and physical layer algorithm has gradually shown the marvelous performance in dealing with the booming increase of communication traffic and complex modeling of wireless environment.However,as the development of intelligent communication,the blindspot of AI-based models and algorithms forces 6G to face a new round of challenges on the way to achieve strong security.Based on the application of AI in physical layer of intelligent communication,this article aims at the potential secure problem to present a new guidance of physical layer secure algorithms in 6G.The major research contents are as follows:First,this thesis deeply invetigates the key technology of conventional physical layer security,and analyze the potential threats of AI based physical layer.Based on the architecture of physical layer,the main approaches of attack on traditional physical layer will be listed.In the meantime,the design ideas and implementation of traditional physical layer secure algorithms will be reviewed.Furthermore,the typical AI-based algorithms will be briefly presented,in order to bring in the concept of the intelligent communication.However,AI based models and algorithms have endured the threats of adversarial attack in computer vision.Based on the discussion of adversarial attack in computer vision,the current situation of the potential threats that adversarial attack caused in physical layer will be sketched.Further consideration about adversarial attack on physical layer in mobile communication will be made.Then,this thesis focuses on the possibility of malicious attack that the CSI feedbak may have in massive MIMO system and studies the design and implementation of adversatial attack on CSI feedback in massive MIMO communication system.On the basis of further understanding of CSI feedback in massive MIMO communication system,the main approaches of CSI feedback will be introduced where the deep learning based algorithm will serve as primary content.Furthermore,tracing back to the birth and growth of adversarial attack on image reconstruction,the adversarial attack on DL-based CSI feedback will be presented.Based on the characteristic of adversarial attack in mobile communication,the design of adversarial attack make use of the specific mathematic model of bias layer and implement the attack during data transmission.The details of attack,testing program and simulation data will be given later.Jamming attack will be chose to launch alongside adversarial attack on the CSI feedback network in different environments and compression rates.The influence on the performance of CSI feedback network under different attack will be compared and analyzed.Based on the simulation results,adversarial attack shows destructive influence on CSI feedback comparing to jamming attack which arouse further consideration on physical layer security in intelligent communication system.Finally,this thesis focuses on the difference of robust performance that several BF schemes hold in massive MIMO system and compares the robust performance against adversarial attack of different BF methods with no knowledge of CSI in massive MIMO communication system.Based on perfect CSI,the BF design in massive MIMO communication system will be explained in detail.And developments of BF design approaches will be exhibited especially the DL based CSI feedback and BF design joint network.Schemes of adversarial attack on DL based CSI feedback and BF design joint network,DL based CSI feedback and eigen-beamforming network and DL based CSI feedback and BF design separated network will be given.Schemes of attack against different model will be slightly altered based on the characteristic of models.After detailed introduction of simulation schemes and data source,the numerical results will be analyzed in detail.As shown in simulation results,the DL based CSI feedback and BF design joint network performs the best robust performance under adversarial attack while the attack has the most expensive cost.