Research On Security Communication Technology Based On End-to-End Learning

The transmission channel in wireless communication system has strong openness,which makes wireless communication have greater security risks.Communication security is the basis of information security,so the security of wireless communication has attracted more and more attention.The traditional high-level encryption technology needs strong computing power as support.The continuous improvement of computer computing power makes this method no longer reliable.In contrast,physical layer security technology is a secure communication method based on information theory,which can play a greater advantage in wireless applications.With the continuous technical change and application expansion,artificial intelligence technology is gradually introduced into the bottom signal processing,and is breaking through the traditional classical communication theoretical framework.More consensus is that the introduction of artificial intelligence in physical layer communication can establish a new paradigm for the underlying signal processing and communication mechanism.Under the guidance of this consensus,researchers in the field of wireless communication proposed a communication system based on end-to-end learning.The system has achieved good results in terms of reliable communication in the physical layer.This achievement also makes scholars see the possibility of end-to-end learning communication system in solving physical layer secure communication.The system began to become a new perspective for physical layer security design.From this point of view,we decided to study the security problems in end-to-end communication system through simulation.The main research work is summarized as follows:(1)Traditional physical layer security gives the theoretical goal of security capacity in various channel scenarios,and end-to-end learning communication is a new method used to achieve this goal in recent years.The first research work in this paper is to design a security constraint function with the goal of approximating the security capacity in an end-to-end learning communication system based on degenerate eavesdropping channel.The results show that the encoder under this method can adaptively generate symbol mapping with high security performance,and the security rate of the system is close to the safety capacity.(2)To further optimize this physical layer security technique implemented with end-to-end learning to deal with more complex communication scenarios(such as non-degenerate eavesdropping channel scenarios),we decided to introduce artificial uncertainty.In the classic physical layer security technology,artificial noise and cooperative interference are effective ways to introduce artificial uncertainty,but they are complicated to implement and difficult to integrate into the end-to-end learning communication system.Therefore,in the research of the fourth chapter,we fully combine artificial noise technology with neural network to build artificial noise generation network.The network can learn the method of constructing artificial noise through self-learning and can be well applied to the end-to-end learning communication system.(3)In order to achieve a security solution that does not interfere with legitimate receivers,typical cooperative interference techniques need to consider many factors(such as the type of interference signal,interference cancellation mechanism,etc.),and this process is very cumbersome and complicated.In the research of Chapter 5,we build a cooperative interference generation network,which does not need to consider the type of input signal,and can achieve high security under the premise that legitimate receivers are not interfered.