Research On Machine Learning Based Physical Layer Authentication

With the development of society,communication has been closely linked with many aspects of the world.And communication security is the basic element to ensure the successful operation of related applications,especially when it comes to the use of wireless communication to transmit confidential information.Due to the broadcast characteristics of the wireless channel,wireless communication system is more vulnerable to sundry attacks.Before the 5G was proposed,the commonly used security authentication schemes were mostly encryption technologies built on top of the physical layer.With the explosive growth of network access equipment and the continuous improvement of computing power,traditional security authentication schemes are gradually being challenged.In the while,the design of various related algorithms has become increasingly complex.The machine learning algorithm can efficiently implement the correlation analysis and feature extraction of massive data.And through channel estimation during communication,we can easily get enough channel data.Therefore,it is feasible to apply machine learning to the physical layer security authentication at the theoretical level.In order to efficiently identify and extract the characteristics of channel physical information,this paper focuses on the design of physical layer security authentication scheme based on machine learning.The research contents of this paper include the following aspects:(1)The traditional communication security solutions are being chanllaged because of many factors such as the changes of communication scenarios and the improvement of computing power.In this case,the physical layer-based security are a highly potential research direction.This paper briefly summarizes the existing physical layer security solutions.Due to the insufficiency and limitations of existing research work,a physical layer security authentication scheme combining machine learning algorithms is proposed.This scheme does not require special assumptions about the channel model or channel variation,nor does it require any prior knowledge about the communication system,and is entirely an adaptive security authentication scheme built on the channel observation data.(2)Among the many machine learning algorithms,this paper designs physical layer security authentication schemes based on kNN algorithm,logistic regression algorithm,SVM and CNN.First of all,according to the physical layer security authentication model,the authentication process is briefly discussed.Then,through the analysis of the wireless channel,the algorithm of using the wireless channel CSI for physical layer security authentication is proposed.In order to realize the effective utilization of CSI data,the data obtained by simulation are specially preprocessed according to the characteristics of different machine learning algorithms.The effectiveness of combining machine learning algorithms with physical layer security authentication was demonstrated by simulation under different parameter settings.After analysis,we found that the performance of the secure authentication schemes based on machine learning algorithm are greatly improved compared with the scheme based on NP test..Specifically,the physical layer security authentication scheme based on kNN has the lowest time complexity,and the physical layer security authentication scheme based on CNN has the best performance.(3)Considering that machine learning algorithms,especially deep learning,often require a large number of labeled samples.However,the labeling work can be quite expensive in the presence of an attacker who can play tricks.In order to realize the fully utilization of channel samples,this paper also proposes a physical layer security authentication algorithm based on semi-supervised learning.In the case where there are abundant channel observations but only a small part of them are labeled.Generating pseudo labels for the unlabeled channel records through using the proposed semi-supervised k-means algorithm greatly expand the CNN training sample set.Monte Carlo simulation proves that the physical layer security authentication scheme based on semi-supervised learning algorithm has excellent performance under the condition of limited sample labeling.