Research On IoT Coverage Enhancement And Resource Management For 5G System

The Internet of Things technology has completely changed the wireless sensor network,and it has realized the universal communication of internet of all things in modern era.Unfortunately,with the increasing of IoT devices,the existing IoT technology is limited by the scale of the spectrum.Massive users can not reuse the licensed frequency band,while in the scene of resource-rich unlicensed frequency band,there is no suitable technology that can be fully used by massive IoT devices.Therefore,exploring a high-coverage IoT system based on unlicensed spectrum is an effective solution to access massive IoT users.With the complexity of the IoT application scenario,it is difficult to allocate resource with massive IoT devices accessing.The fifth-generation mobile communication system(5G)has been proposed to improve the system capacity while focusing on the quality of user experience(QoE).The resource allocation algorithm in the system before 5G is to improve the user's QoE from the perspective of improving the accuracy and quality of service.It is obviously that the user's QoE can not be fully guaranteed by such a solution.In addition,with more diverse of user requests,the traditional resource allocation algorithm can not flexibly cope with more complex scenarios.Therefore,the research of radio resource allocation based on QoE is of great significance.This thesis focuses on two aspects of massive IoT devices accessing,including the design of a narrowband IoT communication system which can enhance coverage and reduce energy consumption from the physical layer technology in unlicensed bands.This thesis proposes a new deep reinforcement learning method,which satisfies interference constraints and maximizes the QoE.The main contents and innovations of this thesis are as follows.First,the traditional Narrowband Internet of Things(NB-IoT)is a Frequency Division Duplex(FDD)system which is proposed by the Third Generation Partnership Projec,t(3GPP).Unlike NB-IoT,this thesis designs a time division duplex(TDD)system which is based on Frequency Division Duplex(FDMA)in unlicensed frequency bands.Furthermore,the downlink design details of this system are also described.Secondly,several schemes of transmitting and receiving algorithms for downlink data channel and synchronous channel are devised,and the performance of these links is verified by comparison with other different methods,which further improves the coverage performance.Thirdly,for massive Internet of Things scenarios,a deep reinforcement learning based centralized resource allocation algorithm is suggested to satisfy the interference constraint of the link.Finally,different from the traditional resource allocation algorithm aiming at optimizing quality of service(QoS)of a communication system,this thesis focuses on maximizing QoE which is measured by Mean Opinion Score(MOS).Compared with the traditional reinforcement learning algorithm,this paper uses an actor-critic(AC)reinforcement learning framework to overcome the problem to allocate continuous power which the traditional reinforcement learning algorithm can not achieve.Simulation results show that the proposed algorithm has stable convergence and better QoE level under different conditions.