Research On Spectrum Decision Algorithm Based On Reinforcement Learning

With the emergence of various new wireless communication technologies,the limited spectrum resources become more scarce.According to the relevant research data,spectrum resources are not effectively utilized in time or space dimension,and some spectrum resources are wasted.Cognitive radio can reasonably solve the problem of lack of resources and low utilization rate by giving cognitive users the right to access licensed frequency bands.As the core of spectrum decision-making in cognitive radio,cognitive engine can solve the problem of parameter adaptation in cognitive wireless communication by combining artificial intelligences.Traditional algorithms have a significant effect on cognitive environments that are known or have little change.However,the cognitive spectrum environment is usually dynamic,and cognitive users are not necessarily able to obtain all the information.It is necessary to study a decisionmaking scheme with stronger environmental adaptability.In order to explore the intelligent spectrum decision-making of cognitive users in dynamic environment,this thesis establishes a cognitive engine model based on reinforcement learning algorithm.Firstly,considering the underlying spectrum sharing scenarios of primary users and cognitive users,the problem of cognitive user intelligent power control is studied simply by using the reinforcement learning algorithm.Secondly,for the problem of the cognitive users which easily jammed in cognitive environment,a novel model based on the interaction between a cognitive user with frequency hopping and a smart jammer was investigated.The model uses the trial-anderror and feedback learning mechanism of reinforcement learning to make the dynamic interaction among primary users,cognitive users and intelligent jammers,and gets adaptive optimization strategy during the interaction process.Considering the problem of channel selection and power allocation,this thesis designs the energy efficiency function as the evaluation standard,and proposes a cognitive anti-interference decision algorithm based on improved reinforcement learning.Simulation results show that the proposed algorithm can converge faster and the adaptive strategy can effectively improve the SU's performance against smart jammers which is about 11% higher than the traditional strategy.