Reinforcement Learning Based Radio Resource Allocation For Physical Layer Security In Internet Of Vehicles

With the development of intelligent connected vehicles and mobile communications,vehicular communications and internet of vehicle are more and more common in industry and life.And information security is critical.However,the computing power of the eavesdropping devices and the number of communication devices are increasing rapidly,and the traditional encryption algorithm has some limitations.Physical layer security is a complementary technology of traditional encryption schemes,and it is a research hotspot in recent years.Secure resource allocation is one of the branches of physical layer security,which is based on the point of view of optimization and signal processing.It uses resource scheduling to intentionally expand the difference between legitimate channels and eavesdropping channels to improve the security performance.Traditional secure resource allocation algorithms are usually modeled as an optimization problem,and all users are allocated spectrum and power under the scheduling of a computing center.The environment of the internet of vehicle is changing rapidly,some problems such as delays are difficult to build an exact mathematical model.The objective of optimization,such as secrecy capacity,is usually nonconvex due to its logarithmic subtraction property,so it is difficult to obtain the analytical solution.At the same time,vehicular communications require extremely low latency,but a centralized computing center will generate additional latency and overhead,which make it difficult to meet the latency requirements of vehicular communications.Therefore,this paper proposes two secure resource allocation methods based on reinforcement learning and has achieved the following innovative results:Aiming at the problem of delay optimization and the difficulty to obtain the analytic solution in traditional secure resource allocation algorithm,a centralized resource allocation method based on deep reinforcement learning is proposed.In this method,the base station is designed as a deep reinforcement learning agent.By designing appropriate states,actions,and rewards,the delay of V2 V communication and the system secrecy rate of the V2 I link are optimized at the same time.Simulation shows that this method achieves excellent system performance.Aiming at the extra delay and overhead caused by the computing center in the centralized resource allocation algorithm,a distributed resource allocation method based on multi-agent reinforcement learning is proposed.In this method,the vehicular transmitter of each V2 V link is designed as an agent.Each agent in the system automatically selects spectrum and power based on its local observation.This method solves the instability of the environment in a multi-agent system,and enables distributed agents to cooperate to improve system performance by designing unique states,actions and rewards.This method avoids the extra delay and overhead,and obtains a system performance close to the single agent method.This method ensures the confidentiality and delay requirements of vehicular communications.