Dynamic Power Control Method Based On Deep Reinforcement Learning

5G technology has been widely used as wireless communication technology has become a newly developing tendency.More and more devices get access to Internet through wireless technology,and Everything in Internet based on 5G technology is possible to reach.However,the rapid growth of wireless communication services led to a dramatic increase in the demand for spectrum.Transmit power is an important wireless communication resource,and an effective power control method not only can improve the utilization rate of spectrum resources,but also improve the quality of service(Quality of Service,Qo S)of users.DRL has seen unprecedented growth,and it not only has the powerful perception ability of deep learning,but also has the decision-making ability of reinforcement learning,and it can overcome the problem that traditional reinforcement learning cannot deal with high-dimensional continuous state and action space as well.Therefore,this research considers using deep reinforcement learning method to solve the dynamic power control in spectrum.In this study,to ensure the quality of communication service for secondary users in the cognitive wireless network and reduce the waste of resource caused by the unreasonable transmit power,the relevant research is carried out.The objectives of this study are:Firstly,a dynamic spectrum resource allocation model for secondary user overlapping access to primary user channel is established,and a double DQN dynamic power control method based on SumTree sampling is proposed.The selection of the target Q value action and the calculation of the target Q value are decoupled in the algorithm network structure.The use of SumTree sampling not only can ensure the priority Class transfer also guarantee the lowest priority of non-zero probability sampling and preventing overfitting of the system.The method can enable the secondary users to choose a lower transmit power to complete the power control task through continuous learning in a dynamically changing environment.Moreover,to further improve the effect of dynamic power control,a dueling DQN dynamic power control method based on SumTree sampling with high reward and penalty is proposed.The dueling DQN algorithm is used in the network structure of the algorithm which can evaluate the value of the state and action,finally,the sum is re-evaluated.SumTree sampling is added to the process of playback about experience in dueling DQN to ensure the possibility that all experience samples are extracted.At the same time,a reward function based on high reward and punishment is designed.According to the spectrum access situation of the secondary user,the most successful action of power control of the secondary user is given a high reward value,and the action of the most unsuccessful power control of the secondary user is given a high penalty value.The final simulation results show that the two methods proposed in this research can improve the total throughput upper limit of the secondary user,and the success rate of the secondary user power control for solving the dynamic power control problem of multiple users in cognitive wireless networks are effective.The power consumption of the secondary user is reduced,and the purpose of reducing energy consumption is achieved.