Research On Resource Management Technologies In Mobile Edge Intelligent Systems

The advancement of mobile communication technology boosts the rapid development of mobile applications and multimedia services,generating tremendous mobile data traffic at network edge,which poses unprecedented traffic loads on wireless networks.Therefore,these is a strong demand to push artificial intelligence(AI)to the network edge such that the potential of the big data can be fully unleashed.Besides,increasing user requirements,including low transmission delay,high data rate,and so forth,bring significant challenges to communication networks.To cope with these enormous traffic demands and challenges,mobile edge computing(MEC)is emerged as a promising architecture to reduce the transmission delay and bandwidth waste,by offloading storage and computing resources to the edge of networks.The confluence of AI and MEC gives birth to a brand-new inter-discipline,edge intelligence(EI).However,research on EI is currently in its early stage,and there are still some shortcomings in terms of AI model deployment,intelligent MEC and so forth.To this end,this paper focuses on how to provide better solutions to multi-node cooperative edge caching and resource allocation with the aid of AI technologies.Specifically,the main contributions and innovation points include the following three aspects:Firstly,this paper focuses on a multi-service scenario in the MEC systems,where the MEC server can provide three multimedia services including live streaming,buffered streaming and low latency enhanced mobile broadband applications for edge users simultaneously.To satisfy various quality of service(QoS)requirements for different multimedia applications,the 5G QoS model is applied to flexibly schedule the limited radio resource.Then,this paper formulates a long-term cumulative average QoS maximization problem,and adopts a deep reinforcement learning method to make decisions to dynamically allocate radio resource.Compared with round-robin and priority-based scheduling algorithms,the simulation results validate that the proposed algorithm outperforms other resource scheduling algorithms.Secondly,this paper addresses a cooperative caching and content delivery scenario in fog radio access networks.Considering limited caching storage in each fog-computing-based access point(F-AP),diverse user preferences,unpredictable user mobility and time-varying channel states,a long-term cumulative multi-user average transmission delay minimization problem is formulated,and then a deep reinforcement learning based delay-aware cache update policy is proposed.The proposed cache update policy will decide to replace the stored contents in F-APs with the proper contents at each time slot.Compared with first in first out,least recently used and least frequently used caching policies,simulation results illustrate that the proposed caching policy yields better average hit ratio and lower average transmission delay.Thirdly,this paper investigates a multi-node cooperative edge caching and multi-service content delivery scenario in mobile EI systems.To enhance edge caching capacity,a multi-node cooperative service architecture is proposed.In the proposed architecture,apart from fixed edge service nodes,mobile edge service nodes equipped with storage and communication capability are also utilized to cache contents and transmit data.A long-term cumulative average QoS maximization problem is formulated to cope with the movement characteristics of mobile nodes,time-varying channel gains and interference,random content requests.Notably,this paper proposes a multi-node hybrid caching policy,including a centralized popular content placement policy and a distributed popular content placement policy.Besides,a service node selection policy is proposed to choose the appropriate node with the most service time to avoid service interruptions.Thereafter,a deep reinforcement learning based spectrum resource allocation algorithm is designed to flexibly schedule the downlink spectrum.Finally,the simulation results validate that the proposed algorithms achieve tradeoff between performance and complexity in comparison with other benchmark algorithms.