Latency-Minimized Resource Allocation Optimization In Fog Radio Access Networks

The ever growing popularity of the Internet of Things(IoT),mobile Internet,and the intelligent applications pose ultra-low latency demand for next generation wireless networks.As the matter of fact the rigorous low la-tency demand of the fifth generation(5G)mobile wireless networks and be-yond 5G cannot be mitigated by exploiting the traditional cellular networks and obsolete handcrafted approaches.Although the cloud radio access net-work(C-RAN)has been considered as one of the key components of 5G that can enhance energy efficiency and spectrum efficiency,the constrained fronthaul delay leads to degradation of the overall system performance.Hereafter,the fog radio access networks(F-RANs)have been revealed as one of the groundbreaking techniques to support the services of the IoT by leveraging edge caching and edge computing.Although F-RANs are promising to support these enabling technologies,delay performance is still not satisfactory and should be further optimized.Additionally,many fundamental challenges remain in F-RANs,including resource allocation and resource management require investigation.The main contributions of this dissertation are summarized as follows:In the first contribution,this thesis utilizes the joint distributed com-puting approach,distributed content delivery scheme,and optimal trans-mission rate.Due to constrained fronthaul,it is not trivial to fetch all the requested contents of IoT users from the remotely located cloud server-s.Therefore,this work mainly focuses on alleviating the heavy burden on fronthaul and achieving ultra-low latency by enhancing the support of content and computing service requests in the vicinity of the UE.A mixed-integer nonlinear programming problem(MINLP)is formulated,and to solve the problem joint distributed computing scheme and the distributed content sharing scheme are proposed with the greedy algorithm.Further-more,the weighted minimum mean square error(WMMSE)approach is employed to address the nonconvex sum-rate optimization problem.The key idea is that a set of F-APs form the cluster and share their computation and contents resources to satisfy the user demand.The simulations result-s show that the proposed scheme outperforms and achieved lower latency compared to other approaches.To enhance the performance of the system and minimize delay,in the second contribution of this dissertation,we have adopted deep rein-forcement learning(DRL)scheme.In the proposed downlink F-RAN ar-chitecture,a latency optimization problem is formulated,and to solve the problem,a DRL based joint proactive cache placement,and power allo-cation strategy is proposed in this chapter.A distributed content caching approach is immensely considered where a number of F-APs cache the popular contents without knowing the specific user request.The F-APs are connected through wired connections that have a very high data rate and can share the content with each other with the minimum latency.The content serving model is designed based on the availability of the contents.Each user can be adaptively satisfied their demand either at the edge or at the cloud computing tier.The numerical results depict that the DRL based proactive content caching and power allocation approach achieved high-er performance compared to other benchmarks.This is because the DRL based distributed content caching scheme enhances the caching capacity at the edge,which leads to lower delay in the system.In the context of tackling the rigorous computation demand of numer-ous IoT devices,the efficient joint computation offloading and resource al-location scheme is studied in the third contribution of this thesis.Generally,in F-RANs,to satisfy the massive computation demand,the user equipment offloads its task to the remote cloud and generates intolerable delay.These computation tasks also create a heavy burden on the constrained fronthaul.To address this bottleneck,a DRL based joint mode selection and resource allocation approach is studied in F-RANs.First,the controller makes an ac-tion for selecting a proper mode to execute the computation task and,based on the selected mode,allocates an optimal amount of resources.To achieve computation efficiency,the system splits the task into small subtasks and processes with several F-APs at the edge.