Probabilistic Cache Policy Design for Cellular Networks with Stochastic Geometry Analysis

The annual data traffic of mobile cellular networks is growing explosively, which has led to backhaul link congestion and latency during data reception from cellular networks. For the fifth generation (5G) of cellular systems, faster transmission speeds, lower latency, and higher spectral efficiency than the previous generations are required. Edge caching promisingly fulfills these requirements by alleviating the unprecedented data congestion and traffic escalation issues of cellular networks. Edge caching is a technique to proactively store the potentially preferable files at the edge of the network (e.g. base stations or user equipment). To achieve an appropriate cache strategy, the two phases of cache placement and cache delivery need to be addressed and optimized. Moreover, a cache policy can be designed based on a static or dynamic framework. For the former, only one shot of network operation is considered, while for the latter, the dynamics of network operation are taken into account. This thesis aims to design optimal static and dynamic caching policies based on the considered model of network operation. For the static caching, this thesis considers the multipoint multicast transmission scheme with a probabilistic cache placement. Building on stochastic geometry, the outage probability is analyzed as network performance to design a static cache strategy. As such, a constrained optimization problem is formulated considering resource and cache allocation parameters, and two algorithms are devised to numerically solve it. Simulation results show that the usage of multipoint multicast is a promising and competitive approach compared to the singlepoint scheme from the literature. This thesis also proposes a hybrid scheme combining the multiantenna single-point unicast and multipoint multicast components to simultaneously leverage the advantages of these schemes for a static cache strategy. To find the hybrid cache solution, a timevarying optimization problem is formulated considering cache and resource allocation parameters as well as content assignment between those two different components. Simulation results indicate the superiority of the proposed hybrid scheme from the spectral efficiency perspective. For the dynamic caching, the dynamics of the user requests in a cellular network are formulated based on a Markov decision process. As such, a reinforcement learning algorithm is exploited to devise a dynamic cache strategy. Simulation results show significant improvements brought by proposed dynamic caching from the quality-of-service and power consumption point-of-view.