Research On QoS Modeling And Optimization For 5G/B5G

Driven by the explosive data flow of mobile users and new quality of service requirements,the communication industry is experiencing a new evolution of network infrastructure intensive.With the increasing density and variety of access points(AP),the network benefits from the near end transmission and the spatial reuse of system resources,thus introducing a new ultra-dense network(UDNs).Due to the ubiquitous AP sharing limited available resources in ultra-dense networks,the demand for efficient resource allocation and optimal control schemes becomes more urgent.However,in large-scale ultra-dense networks,the significant increase of computational complexity and energy consumption is a challenging problem.At the same time,in the future network,delay will be an important performance index,and also an important part of user QoS.The plan is reduced to 10ms level or even lower,and the queuing delay is the key factor affecting the delay.Therefore,an optimization scheme based on delay QoS guarantee needs to be proposed.Most of the existing work does not consider the queuing delay of the system.Therefore,this paper will focus on this problem,and put forward a series of feasible solutions,as follows:1.Research on 5G/B5G oriented super dense network and its key technologies,and investigate some common resource allocation and optimal control methods.Then,according to the MEC scenario,an efficient way of information interaction between MEC server and core network is proposed.At the same time,the QoS index on the data path is monitored,which can timely and efficiently feed back the changes of users' QoS requirements,timely adjust the user service strategy,and improve the user experience.This solution can be used as an information interaction tool between core network and MEC server,and provides an effective tool for obtaining timely information in ultra-dense network optimization.2.Based on the effective capacity model,the queuing delay in wireless communication system is analyzed and the interference model is built.Because it is difficult to analyze the non independent and co distributed autocorrelation channel,most research work assumes that the channel is independent and co distributed.So this paper first models the QoS of the independent and co distributed channel,and gives the effective capacity model of the channel.Based on the effective capacity model,a new computational offloading scheme is proposed,which is applied to the ultra-dense network optimization scheme.However,it is unrealistic to consider the independent co distribution channel in practice.Therefore,this paper studies the effective capacity of the autocorrelation Rayleigh fading channel,and gives the discrete expression.3.Based on the Markov decision process,the classical algorithms of reinforcement learning and deep reinforcement learning are compared.Considering the huge state and action space,DQN and DDPG are selected as the optimization methods in this paper.Combined with the advanced technologies of 5G/B5G network,such as MEC,network slicing,etc.,a power control scheme based on effective capacity and deep reinforcement learning is proposed.The results show that the method adopted in this paper has certain advantages in the overall effective capacity,delay and energy consumption of the network.To sum up,through detailed analysis and research,effective capacity and deep reinforcement learning are successfully applied to 5G/B5G oriented ultra-dense network optimization scheme,which combines a variety of key technologies,solves the neglect of queuing delay in traditional optimization scheme,further improves network performance,and improves user service quality.