Deep Learning-based Resource Allocation For 5G Fronthaul Networks

The vision and demand of fifth generation mobile communication network(5G),is meeting the demands of explosive growth of future mobile traffic data growth,huge amounts of equipment connections,sprung up of all kinds of new business and application scenarios,depth of fusion with the industry at the same time,meet the diverse needs of the vertical industry terminal connected,so as to realize the real "Everything Connected"[1].From traditional closed and inelastic network infrastructure to an open,scalable and flexible ecosystem,there are various kinds of services and different service characteristics of network access are not identical.These requirements bring new challenges to the design of 5G network,including network function,architecture,resources and so on.These challenges require 5G network services to be flexible,adaptive and intelligent.Therefore,5G optical access network is in urgent need of improving its flexibility and intelligence in network design to cope with the challenges of large-scale traffic and multi-service scenarios.This paper aims at the situation that 5G fronthaul network is faced with a large amount of access traffic data and a variety of traffic services.We study the resources allocation algorithm of 5G fronthaul network,and study deeply the network traffic processing mechanism and network resources allocation algorithm under the condition of elastic optical network.Under the condition of comprehensive consideration of traffic service types and 5G fronthaul network resources,a traffic processing mechanism based on deep learning algorithm and a resource scheduling algorithm based on priority determination are proposed.It has realized the innovation and breakthrough in 5G fronthaul network resources allocation.The work and innovation of this paper are as follows:First of all,a traffic prediction mechanism based on traffic classification is proposed to solve the problem of large-scale complex traffic data,which is caused by the diversified business scenarios and multi-device access in 5G network.This paper proposes traffic classification and traffic forecast in 5G access network based on deep learning.Through network traffic combing the fronthaul network can response to traffic request according to the different needs of different traffic types,and forming a mapping relationship between business types and traffic data.At the same time,through the traffic flow prediction for future time,the network can better satisfy the 5G network requests:high reliability and low latency.Through the experimental simulation,the deep learning-based algorithm we used can reduce the response delay in traffic processing and improve the prediction accuracy.Secondly,a resource allocation algorithm based on traffic request priority determination is proposed to meet the demand of resource allocation intelligent in 5G fronthaul network scenarios.We study resource allocation algorithms and a variety of neural network algorithms under the condition of elastic optical network.With the combing result of large-scale historical traffic data,and according to the traffic prediction results and traffic priority determination,the traffic requests with higher priority are allocated resources.5G fronthaul network can realize the intelligent allocation of elastic resources according to the different demands of different demands of traffic services,and realize the demand of intelligent and high utilization rate of the network limited resources allocation.At the same time,we put forward the evaluation mechanism of resource allocation algorithm,which considers the balance judgment between system resource allocation anXd user request resource.On the basis of satisfying QoS,an efficient utilization of 5G fronthaul network resources is realized.Through experimental simulation,the proposed intelligent resource allocation algorithm based on traffic request priority can reduce the network blocking rate and improve the resources utilization rate.