چكيده به لاتين
With the advent of new multimedia applications, content delivery traffic to wireless cellular networks users has increased dramatically. Transferring content from mainstream servers to mobile users consumes a significant part of bandwidth from the network backhaul, which can lead to a long latency experience for the users. To meet these challenges, the idea of temporarily storing (caching) users' popular content on equipment at the edge of the network (such as small base stations or SBSs) has been proposed. In fact, bringing content closer to users by effectively placing them on the edge also reduces the delay in receiving content by them.
In this thesis, the issue of locating a library of content files stored in SBSs is raised as an optimization problem with the aim of minimizing the average latency of mobile users. In the content placement problem, parameters such as content popularity, wireless channel, and backhaul status are the inputs to the problem, and the calculation of the optimal location configuration depends on the values of these parameters. These parameters are random in nature and their instantaneous values over time form a random process. Unlike previous works based on offline optimization (based on complete instantaneous information) or model-based (based on the probabilistic model of the system), in this thesis, a model-based approach based on game theory and multi-agent learning is adopted. In the proposed approach, the calculation of the content placement policy is one of the the SBSs duties to decide automatically, based on local information about the content they should store. The issue of content placement in this context is modeled by using the "potential game" formalism between SBSs, in which each SBS seeks to store content that minimizes the average latency of users within its coverage area. To calculate the Nash equilibrium, two algorithms are proposed that are based on multi-agent learning procedures for potential games. The first algorithm in multi-agent joint action learning space and the second algorithm in multi-agent independent action learning space of SBSs by equilibrium learning, are able to bring the SBS content placement strategy profile to Nash equilibrium, especially in circumstances of larger-scale scenarios and even in the absence of observability of other SBS decisions. Therefore, the focus of this thesis is on the second method.
The simulation results show that the proposed method, which is based on game theory, is able to approach the result of centralized algorithm with an approximate margin of about 36.3% with accurate instantaneous data. Also, the performance of the proposed method in different scenarios has been tested and analyzed in terms of changes in the statistical parameter of content popularity, number of SBS, and mobile user population.
