Reinforcement Learning-Based Intelligent Reflecting Surface Assisted Communications Against Smart Attackers

Wireless communications are vulnerable to cyber attackers, which now have the flexibility to choose their type of attack. In this paper, combined with intelligent reflect surface (IRS), we jointly optimize base station beamforming and IRS reflected beamforming to counter smart attackers, thereby improving system security. Considering that attackers can flexibly choose their attack methods, such as jamming or eavesdropping, we make the base station intelligent by using reinforcement learning, which can predict the attack methods of attackers and choose whether to add artificial noise into the transmitted signals. At the same time, the interaction between the base station and the smart attackers are established as a non-cooperative game, the Nash equilibrium of the game is derived. Based on this, the base station anti-smart attackers strategy based on Deep Q-learning (DQN) is proposed, which can restrain the attack of the attacker to improve the security of the system. It can be verified from the simulation results that the proposed anti-smart attackers strategy can effectively enhance the secrecy rate of the wireless communication system, resist the attacker’s attack, and intelligently transmit artificial noise to improve system security.