Research On Secure Communication At The Physical Layer Under Multi-agent Attacks

With the rapid development of wireless communication network,the demand for the transmission performance of wireless communication is becoming higher and higher.Due to the openness and mobility of wireless communication network,it is vulnerable to the threat of attackers compared with the wired network.At present,some intelligent programmable radio devices are widely used by attackers to attack the transmission of information.These attackers can choose their attack type flexibly,for example,attackers can choose to eavesdrop the information when they are close to the transmitter,or choose to send jamming or spoofing signals to block the transmission when they are close to the receiver.Moreover,attackers can use their learning ability to learn that how to maximize their own interests by choosing different types of attacks based on the changing channel environment.In this way,the attackers can attack the legitimate user effectively,which results in a serious information leakage problem.In order to solve the above security problems and improve the secrecy performance of wireless communication under some uncertain attack types,game theory and reinforcement learning have been applied to the research of physical-layer secure communication.However,the current researches still have some key problems that are not taken into account,for example: 1)How to protect the information security under the attack of an attacker and improve the confidentiality of the system with imperfect channel estimation;2)How to achieve an effective policy and analyze the impact on the system under statistical channel state information;3)How to suppress the attack rate of smart attackers efficiently and safeguard the system security under multiple smart attackers.This article will focus on the above key issues to study in-depth.Research methods and innovative results are briefly described as follows:Firstly,we consider a communication system model,consisting of a signal transmitter,a receiver,and a smart attacker.For this model,the transmitter sends the message to the receiver,and the attacker chooses an attack mode to block the transmission.The actions between the transmitter and the intelligent attacker are formulated as a zero-sum game,and the Nash equilibrium with imperfect channel estimation is derived.To solve the combinedimpact of attacker's smart attacks and imperfect channel estimation,we propose a power control algorithm based on Q learning to obtain the adaptive power strategy for the transmitter.The experiment results show that the proposed power control strategy can improve the secrecy performance and reduce the attack rate under channel estimation error.Secondly,based on the above model,we analyze the system secrecy performance with the statistical channel state information of the attacker in-depth,and the formula of channel secrecy capacity based on the statistical channel state information is given.Moreover,we derive the Nash equilibrium.The experiment results show that the influence of system secrecy performance based on the statistical channel information can be reflected by the attacker's attack cost and average channel gain.By observing the system secrecy capacity based on the instantaneous channel state information,we can find that the system secrecy capacity based the statistical channel state information has performance loss in terms of security.Finally,in view of the problem of how to reduce the attack rate of the attackers and realize the secure communication transmission under multiple attackers,a communication model which consists of one transmitter,one receiver and multiple intelligent attackers is considered.The actions between the transmitter and multiple intelligent attackers are formulated as a stochastic game,and the transmitter can choose its own transmit power flexibly.The attackers do not cooperate with each other,and each attacker chooses the action to make their own best benefits.The experiment results show that the new algorithm based on multiagent attacks can improve the system secrecy data rate and ensure the security of communication transmission to a certain extent.