Game-theoretic Appoaches To Resource Allocation In Fog Radio Access Networks

To meet the stringent and diverse requirements in future wireless networks,fog radio access networks(F-RANs)are proposed and have attracted a lot of attention from both the academia and industry.To fully unleash the potential of F-RANs,resource allocation plays a key role.On the other hand,compared to traditional resource allocation approaches,game theory based resource allocation is more efficient,which is a hot topic in the academia.Hence,in this dissertation,novel game-theoretic approaches to resource allocation in F-RANs are proposed to overcome the issue of high computing burden and the requirement on global information.Firstly,focusing on an uplink F-RAN,non-cooperative game based communication mode selection and radio resource allocation for user equipments(UEs)is studied.Then,a hierarchical game based radio resource allocation for network slicing in F-RANs is presented.At last,radio and cache resource allocation is investigated by combining two different game models.The main contents and contributions are summarized below.1.Non-cooperative Game Based Mode Section and Radio Resource Allocation of UEsConsidering the joint optimization of UE communication mode and subchannel allocation in a centralized way can put heavy burden on the cloud,this dissertation proposes to use non-cooperative game to perform distributed optimization,which fully utilizes the resource management capabilities of UEs in F-RANs.First,since subchannels are equally shared by UEs,the communication mode and subchannel selection problem is modeled as a non-cooperative game among them.Then,under pre-determined UE association and power control schemes,a multi-agent reinforcement learning(MARL)based mode selection and subchannel allocation algorithm is developed for the game.To realize UE association and power control with low complexity,UE association based on matching theory and distributed power control based on hierarchical game are further proposed.The properties of all the proposals are discussed and it is proved that the MARL based algorithm is guaranteed to converge with the convergence point approaching the Nash equilibrium of the formulated non-cooperative game.Via simulation,the convergence of the MARL based mode selection and subchannel allocation algorithm is verified.Meanwhile,it can achieve near optimal performance with proper parameter setting and outperforms another distributed approach.In summary,this study provides a useful idea for distributed resource management at UEs,which is one of the key char-acteristics of F-RANs.2.Hierarchical Game Based Radio Resource Allocation for Network Slicing in F-RANsTo avoid the huge burden on the global resource manager(GRM)in-curred by executing centralized algorithms and collecting global informa-tion and meanwhile achieve slice customization,this dissertation studies hierarchical game based radio resource allocation for network slicing in F-RANs.First,under an architecture consisting of a GRM and the local resource managers(LRMs)of slices with heterogenous performance met-rics,resource allocation for network slicing is modeled as a hierarchical game between them,and the equilibrium of the game is defined with its existence and uniqueness discussed.Afterwards,given the NP-hardness of LRMs' problems and the discreteness of the GRM's strategy,under the assumption of rational resource managers,an exhaustive search based algo-rithm is proposed to reach equilibrium in a distributed way.Further,facing the high complexity of exhaustive search,local-optimal resource allocation algorithms are designed for resource managers with bounded rationality.Simulation results reveal that a performance tradeoff between slices exists and the local-optimal proposals can reach competitive performance compared to optimal solutions and other baselines.This study is critical to guarantee the heterogenous performance of F-RAN slices and can be extended to the allocation of multi-dimensional resources among them.3.Composite Game Based Joint Radio and Cache Resource AllocationWhen radio resource and cache resource allocation are correlated in F-RANs,it is essential to study their joint optimization to achieve better performance.Considering radio resource allocation and cache resource allocation are conducted on different timescales and rely on different information,to alleviate the burden on the resource manager in the cloud,joint radio and cache resource allocation is modeled as a composite game involving the resource manager and fog access points(FAPs).In the upper level,the resource manager allocates cache resource to maximize the difference between the benefit brought by long-term system throughput and the caching cost,while the radio resource allocation behaviors of FAPs are captured by a coalitional game whose aim is to optimize system short-term throughput.To solve the formulated composite game,a distributed algorithm is designed based on FAP preference to reach stable coalition formation.Then,facing the challenges incurred by no closed form expression and discrete variables,single-agent reinforcement learning(SARL)and MARL based caching algorithms are proposed for the resource manager,and their convergence and optimality are rigourously studied.By simulation,it is shown that the proposed coalitional algorithm achieves a considerable improvement in short-term throughout compared to the non-cooperative case,owing to the better utilization of cache resource.Moreover,MARL based caching outperforms various baselines.The framework of this study can guide the design of approaches to other mixed-timescale resource allocation problems in F-RANs.