AI Enhanced Adaptive Link Strategies For Simultaneous Wireless Information And Power Transfer Networks

Among the emerging technologies of the Internet of things,simultaneous wireless information and power transfer technology has received considerable attention because it can extend the life of energy constrained devices.Adaptive modulation,power control and other link control strategies can improve the throughput,reliability and other performance under different channel and other environmental conditions.Therefore,it is necessary to introduce these adaptive link control technologies in the new communication network scenario of SWIPT network.In the past research on SWIPT,most of the research on link layer focused on resource allocation and so on,few people considered adaptive link control technology.This thesis focuses on this and studies the application of adaptive link control strategy in SWIPT scenario.In order to meet the requirements of speed,reliability and delay,this thesis will focus on adaptive link control strategy in SWIPT scenario to improve the average spectrum efficiency of network transmission and ensure BER and energy transmission requirements.Therefore,this thesis studies the influence of link control strategy and energy transmission on various performances,obtains the corresponding mathematical expressions of various performances,and establishes an applicable optimization model.Two modeling methods are used,convex optimization and Markov decision-making process,and corresponding algorithms are designed for both,genetic algorithm and deep reinforcement learning algorithm.By comparing the advantages and disadvantages of the two models,it is found that the Markov decision model is more suitable for the modeling of adaptive link strategy,and the deep reinforcement learning algorithm based on neural network is more suitable for solving the optimization model.In addition,in order to improve the practical application value of link control strategy.This thesis starts from two aspects,one is to solve the practicability of link control strategy under different business requirements.The designed adaptive link control strategy can realize autonomous relearning according to different services.Therefore,we proposes two techniques(imitation learning and prior experience generation)to speed up the convergence of the existing deep reinforcement learning algorithm dqn,which greatly improves the convergence speed of the algorithm.So the algorithm can re converge in the changing environment in a shorter time and achieve high performance,which is suitable for the actual scenarios with different business requirements and variable business requirements.On the other hand,the point-to-point link control strategy is extended to the multiuser transmission scenario,and a multi-user scheduling scheme which can cooperate with the point-to-point adaptive link strategy is designed.At the same time,in order to deal with the problem that environment change and time delay make the scheduling strategy invalid,a multi-user scheduling strategy considering the future environment change is designed by using deep reinforcement learning technology.Using the multi-user diversity characteristics,the energy transmission efficiency,overall transmission time performance and fairness are optimized.