Offloading and content caching in 5G heterogeneous networks: a game-theoretic perspective

The continuous evolution of wireless networks leads to the next generation of wireless network development, dubbed the fifth-generation (5G) mobile wireless network. The forthcoming 5G mobile networks will support diverse mobile data traffic with a one millisecond transmission delay along with reduced energy consumption. In fact every aspect of ongoing 5G research is pointing towards a much better quality-of-service (QoS) provision than its predecessor technology, the currently operating fourth-generation (4G) wireless network. It has been widely claimed that 5G will finally provide the technological edge and the necessary infrastructure to support the so-called internet-of-things (IoT). Thus efficient spectrum usage or spectrum management for 5G is of paramount importance to 5G fruition. While venturing the possibility of accommodating new spectrum seems audacious at this point, it will be the spectrum sharing technologies such as heterogeneous networks (HetNets) at the heart of initial 5G deployment. Het-Nets are becoming increasingly common and rely heavily on the spectrum reuse concept to provide services such as offloading and content-caching. Indeed recent literature puts much emphasis on offloading and content-caching as promising 5G technologies and they will play a vital role in futuristic distributed resource management. Firstly, this thesis focuses on developing a novel offloading approach for 5G dense HetNets. Secondly, this thesis attempts to devise a novel small-cell oriented content-caching approach. The main contribution of this thesis can be summarised in two parts. In the first part, a novel pricing strategy is derived which enforces the low data rate macro users to associate with small-cells. The objective was to increase the spectrum utilisation of small-cells, while preventing cross-tier interference to macro users who are located close to small-cells. Furthermore the proposed pricing algorithm acknowledges one of the shortcomings of the traditional received-signal strength (RSS) based user-association where a user always selects the strongest base station, i.e. the macrocell and thus it becomes overloaded in the process. The pricing algorithm achieves our design objectives as stated above. An evolutionary game-theoretic analysis is provided to show, how the proposed pricing strategy can influence users to select small-cells. Another important aspect of our proposed pricing technique is that a macrocell can control its population share (PS) precisely. By tuning the rate-threshold in the pricing algorithm, a macrocell can have an optimum population share. The evolutionary game captures the essence of the proposed pricing scheme. The mathematical background is provided to show that a unique solution exists. In addition, it is proved that the price-based network selection strategy is evolutionary stable. In the second part, this thesis turns its focus on developing a novel caching strategy that utilises small-cells. Existing caching strategies mainly rely on small cells for caching smaller contents. For retrieving larger contents, existing algorithms either download them from cloud storage or redirect an individual's request to a macrocell. Both approaches result in a higher delay, which motivates us to propose a caching strategy that enures more effcient utilisation of small-cells. The first caching model considers the cache-association strategies of users, which is modelled by a binary decision variable. Unlike existing work, this thesis formulates the small-cell based caching problem into a Nash bargaining game (NBG) and utilises the cooperative nature of an NBG to devise a unique, fair and optima caching strategy. However, the optimisation problem formulation which retains the Nash's axiomatic conditions, turns out to be an integer programming problem. Therefore constraint relaxations are applied to find a centralised solution, which can be used as a lower bound to derive the optimal solution. Afterwards Lagrangian relaxation is used to provide a distributed caching solution. In addition, a low complexity heuristic solution is proposed as an alternative solution to the centralised and distributed solution. By applying the concepts of two-dimensional coordinate geometry it is proved that the bandwidth allocation mechanism of the heuristic algorithm has the same convergence property as Newton's method. Further investigation suggests that the bandwidth allocation mechanism approaches convergence in an iterative manner, therefore it exhibits a much desired property of distributed implementation and parallel computation. This thesis concludes by extending the previously proposed caching model including a cache-placement decision and providing appropriate solution methodology.The new NBG model involves a bilinear term and thus becomes a non-convex integer programming problem. The first step towards solving the problem is to apply a relaxation technique which will convexify the non-convex NBG by relaxing the bilinear term. McCormick relaxation is a well-known method used for convexifying an optimisation model involving bilinear term, which is used in conjunction with Lagrange relaxation to find a feasible solution.