Design And Analysis Of Network Traffic Camouflage Scheme Based On Generative Adversarial Network

Network is an important infrastructure in today’s information society.With the rapid development of Internet technology,network has become an important tool for people to realize information transmission and resource sharing.At the same time,network security and user privacy protection have also become important issues of general concern.Covert communication technology,represented by anonymous communication technology,has attracted extensive attention because it can hide users’ identity and sensitive information.Anonymous communication technologies represented by cryptography,Tor(the Onion Router)network and Mixnet network are mainly realized by using the idea of confusion encryption.Although it realizes the protection of source anonymity and sensitive information,it can not hide the fact that covert communication exists.For example,the traffic characteristics of nodes in the network are different from normal traffic,and easy to identify exclusive protocols and fixed network facilities are used.This kind of covert communication technology is difficult to apply in the more rigorous review environment such as military and intelligence.It is easy for the enemy or reviewer to find the existence of covert communication and destroy it directly.At the same time,the cost of node deployment also affects the large-scale deployment of these schemes.In order to hide the covert traffic better,people began to study the traffic camouflage technology based on machine learning in recent years.In this kind of technology,people study more about how to bypass the intrusion detection system based on machine learning,and how to realize traffic camouflage and prevent traffic analysis from the perspective of covert communication.In2014,Good Fellow proposed the generative adversarial network(GAN)technology,which is designed based on game theory and deep learning technology.It has been successfully applied to medical image recognition,face recognition and other fields and has made remarkable achievements.Due to the dynamic nature of its generated samples,it also has broad application prospects in the field of network security.The application of generative adversarial network makes it possible to analyze the dynamic defense traffic and get rid of the white box hypothesis,which provides an important direction for the further research of traffic camouflage.We have conducted in-depth research on this and achieved the following results:Firstly,a covert communication system based on generative adversarial network is proposed.The technical scheme avoids the white box assumption in the traditional defense scheme and is designed to avoid the traffic analysis attack.The core of the framework is the camouflage traffic generation model,which takes the flow feature vector as the input and the camouflage traffic feature vector as the output.This paper presents the design scheme of each component module of the covert communication system,including the frequency table of node conventional application,the camouflage traffic generation model base based on GAN,the traffic analysis and avoidance algorithm based on GAN,the multi application detection classifier,and the packet mechanism of covert communication.The camouflage traffic characteristics are generated by the camouflage traffic generation algorithm,and a variety of classifier models are used for adversarial training to detect its indistinguishability.The technical scheme of this paper can effectively avoid the traffic detection methods based on feature and machine learning technology,so that the normal traffic and camouflage traffic sent by the source can not be distinguished,so as to realize the existence hiding of the source.Secondly,the specific implementation scheme of covert communication traffic camouflage algorithm based on generation adversarial network is given.It includes: network traffic generation technology,traffic preprocessing,GAN’s generator and discriminator,and the specific implementation and flow of traffic camouflage algorithm.Finally,the designed scheme of covert communication technology is simulated by using Python language and tensorflow framework.The simulation includes two parts: nonadversarial training experiment and adversarial training experiment.The experimental data of non-adversarial training show that our covert communication technology scheme performs well in traditional machine learning evaluation indexes such as accuracy.The experimental data after countermeasure training show that the scheme realizes the indistinguishability between normal traffic and camouflage traffic.In addition,compared with the normal communication process,the hidden communication technology scheme proposed in this paper pays less delay cost and lower cost of communication node deployment.