Research On Resource Management And Control Technology Of Mobile Fronthaul Network Based On Reinforcement Learning

With the development of communication technology,exponential traffic growth,demand for high-speed wireless data communication,and continuous deployment of emerging services and applications have put considerable pressure on access networks.Cloud Radio Access Network(C-RAN)solutions capacity,access and data rate issues.Mobile edge computing(MEC),which provides computing capabilities at the edge of cellular networks,emerges as the times require to meet the needs of services for ultra-low latency.The fronthaul structure of the fusion of MEC and C-RAN has attracted much attention.Reasonable resource allocation can efficiently utilize resources,meet user needs and reduce energy consumption.However,due to the tidal effect of the network,the diversification of application scenarios and user needs,the resource management and control of the fronthaul network integrating MEC is becoming more and more difficult.Deep Reinforcement Learning(DRL)has the ability of perception and decision-making.It can obtain the best strategy through real-time interaction with the environment,and can be used as an enabling technology for mobile fronthaul network resource management and control.This paper mainly focuses on the research on the resource management and control technology of mobile fronthaul network based on reinforcement learning.The main research work and innovations are as follows:(1)The current access network integrating MEC and C-RAN usually places the MEC server in the baseband unit(BBU)pool.In order to reduce the pressure on the fronthaul network,this paper proposes a C-RAN architecture that integrates multi-layer edge computing,which can sink the computing power to the remote radio head(RRH)to process some services.The simulation results show that the service pressure of the fronthaul link can be reduced by 22.04%.Aiming at the problem of mobile user service diversity,a user task offloading and resource allocation scheme is proposed,which can jointly optimize offloading decision,communication resource allocation and computing resource allocation,considering delay constraints and resource constraints according to network conditions and service requests.A model for saving task computing time under the unit power consumption of the system user is established.Using deep reinforcement learning algorithm,algorithm training is adaptively performed according to environmental changes,and a reasonable offloading decision and resource allocation scheme are obtained.The simulation results show that this scheme can significantly increase the amount of offloaded tasks,actively offload tasks that require a lot of computing resources.In the offloading tasks,tasks with low latency requirements are preferentially processed in the RRH.(2)Aiming at the problem of routing and spectrum allocation of optical network between BBU pools during load migration between BBU pools,a solution using reinforcement learning is proposed.Using the DRL perception capability,it analyzes the service migration request and the network environment between BBU pools to select the optimal routing and spectrum allocation strategy.Calculate the spectrum free resources based on the change value of information entropy to improve the agent’s ability to select the free spectrum slots.The results show that the blocking rate of the proposed scheme is better than the first-fit algorithm based on the shortest path.The free spectrum slots selection scheme based on the change value of information entropy can reduce the probability of service blocking.