Caching techniques have been extensively studied and deployed as powerful solutions to performance improvement in a wide variety of computer systems. Motivated by new technologies and challenges emerging from prospective cellular architectures, this thesis proposes the design and analysis of novel caching techniques targeting the improvement of mobile users' quality of experience. We are particularly attentive to small cell networks enabled with Coordinated Multi-Point (CoMP) Joint Transmissions (JT) technology First, we study the scenario where content placement is performed by a centralized intelligence aware of previously estimated files popularities and the whole network topology. The best content placement is obtained from solving an optimization problem that we approximate by an efficient greedy algorithm. This solution depends on strict assumptions and may fail to capture short-scale content's popularity variability. However, it is useful to determine performance bounds and to provide insights on the problem's inherent trade-offs. Then, we introduce a dynamic framework, where each cache individually updates its content on-the-fly as a response to arriving requests based on pre-defined caching policies. The proposed caching policies define a set of probabilistic rules that take into account the overall performance gain of any cache update operation. Our first policy achieves implicit coordination between caches and asymptotically converge to the optimal cache configuration under stationary request sequences. We also study the case where requests are non-stationary and provide a policy that provides satisfactory practical results. Finally, we present a set of experiments based on numerical simulations designed to capture intrinsic attributes of real small cell networks. The empirical results confirm the asymptotic convergence to an optimal solution of our first policy. We observe that both proposed policies achieve desirable performance levels when exposed to either stationary or non-stationary request sequences. Furthermore, our policies outperform other state-of-the-art policies in all tested scenarios.
