Research On Physical Layer Authentication Technology For Resource-constrained Devices

With the vigorous development of wireless communication technology,the 5th Generation Mobile Communication System(5G)will become an indispensable information exchange platform for human society.The influence of wireless network has developed from personal communication business to all walks of life in the whole society.As an important part of 5G wireless network,resource-constrained devices such as IoT terminals and UAV have received more attention from industry and academia because of their wide application range,simple deployment,and low cost.However,due to factors such as small memory,weak computing power,and limited power,it is difficult for resource-constrained devices to carry the computing overhead caused by traditional security protocols.The security threats faced by resource-constrained devices become more and more severe,which has become a short board that restricts the overall security of 5G networks.In recent years,physical layer authentication has received extensive attention in wireless network security.The physical layer authentication makes full use of the specificity of the wireless channel to identify the identity of the access signal,and its security does not depend on the computational complexity,which has become an important development direction of lightweight security in recent years.In this thesis,based on federated learning,transfer learning and other algorithms,lightweight physical layer authentication technology for resource-constrained devices is studied from the perspectives of multi-attribute joint judgment and collaborative training,and focuses on the difficulties faced by the deployment of physical layer authentication technology in 5G resource-constrained scenarios.The specific research results are summarized as follows:(1)PUF authentication technology enhanced by multipath channelAiming at the problem that authentication based on Physically Unclonable Function(PUF)is vulnerable to modeling attacks,this thesis proposes a physical layer authentication scheme combining PUF and channel attributes.The multipath channel attributes are used as random factors to design a dual mapping algorithm to randomize the fixed mapping relationship of PUF challenge response pairs(Challenge and Response Pairs,CRPs).Due to the unique spatiotemporal properties of multipath channels,it is difficult to be predict and simulate,which effectively solves the security problem that is easy to be modeled by machine learning when some CRPs are leaked in PUF authentication.At the same time,because of channel reciprocity,legitimate users can be authenticated correctly.Simulations show that the proposed authentication mechanism outperforms the traditional PUF authentication scheme in resisting modeling attack based on machine learning.Compared with traditional authentication schemes based on channel attributes,the miss detection rate of the proposed scheme is reduced about 3.6 times,which proves the effectiveness of the scheme.(2)Collaborative physical layer authentication based on federated learningTo solve the problem that resource-constrained devices is difficult to support the physical layer authentication training algorithm based on machine learning independently,this thesis proposes a collaborative physical layer authentication mechanism based on federated learning.A distributed authentication framework is designed to assign the machine learning training tasks of authentication to trusted collaborator nodes.Then,the models trained by the collaborator nodes are aggregated through the federated aggregation algorithm to obtain a global classifier for physical layer authentication.Since the training task is independently completed,the central node does not need to undertake the high-complexity training task independently,which relieves the computational pressure.At the same time,each collaborator node does not need data interaction,thus ensuring privacy.Simulations show that when resisting spoofing attacks,the missed detection rate and false positive rate of this mechanism are both less than 5×10-3,which proves that the proposed cooperative authentication scheme can release the computational pressure of resource-constrained nodes without losing performance.(3)Fast physical layer authentication handover based on transfer learningIn view of the computational overhead and communication delay caused by repeated train the authentication model in the process of wireless network handover,this thesis designs a fast physical layer authentication mechanism based on transfer learning.Source network uses the transfer learning algorithm to transfer the previously physical layer authentication model to the target network when wireless network handover.The target network uses the model trained by source network as the initial value of training to continue model training,and obtains a new model for the physical layer authentication of the target network.Through this method of transfer the authentication model,the target network only uses the model transfered by the source network for a small amount of training,which greatly saves the computational cost and reduces the delay caused by the training process.The simulation results show that the training convergence rate of the proposed mechanism is improved by more than 80%compared with the traditional physical layer authentication hadover scheme,while the false alarm rate and miss detection rate are consistent with the traditional schemes in resisting spoofing attacks.