Anonymous Network Traffic Feature Camouflaging Via Generative Adversarial Nets

With the close integration of the Internet and modern life,people's demand for network security continues to increase,bringing about the prosperity of various anonymous network technologies.On the other hand,due to various reasons,there are different levels of network censorship in almost areas of society.The development of anonymous technology,especially the wide application of various encryption technologies,makes content-based network censorship more difficult.Therefore,network censors currently use traffic analysis technology to conduct attacks: by analyzing the statistical characteristics of traffic metadata,classifying protocols or contents of unknown traffic,and intercepting sensitive traffic.In order to resist traffic analysis,researchers have proposed various defensive measures.The basic idea is to erase the statistical characteristics of the carried traffic by means of traffic deformation,and disguise the traffic into other traffic.However,most of the existing defense systems lack dynamics: they are designed to be no longer changed,and lack the flexibility to cope with changes in the network environment,making them easier to detect by evolving network censorship technologies.Therefore,the main problem studied in this paper is to design a dynamic traffic camouflage technology.In this paper,a comprehensive investigation of related technologies is carried out,and a threat model of network censorship based on traffic analysis is proposed.The main characteristics of the attackers are analyzed in detail.A new privacy model is proposed.Under the threat model,the indistinguishablity based on classification attack is used as a measure of the effectiveness of the defense system.This paper proposes a new traffic masquerading technology that allows the defense system to dynamically masquerade any target traffic.This study used the Generative Adversarial Nets(GAN)model to learn the characteristics of target network traffic and to help generate traffic that is indistinguishable from the target traffic.Based on the natural flexibility of Generative Adversarial Nets,it only needs the target samples for training,and samples with the same distribution as the target can be automatically generated.Therefore,the system can flexibly change the target traffic,change the disguised traffic mode,and achieve strong dynamics.Finally,in order to assess the effectiveness of the system,this paper tests it on a large traffic data set.The results show that the technology we have proposed performs well on all attacking methods.And by horizontal comparison with similar methods,the system has dynamic advantage over similar defense methods.As far as we know,this is the first study to use the Generative Adversarial Nets for traffic feature camouflaging.Under the latest traffic analysis attack technologies,the effectiveness of this system is significantly improved.