With the explosion in the number of connected wireless devices, and the distinct services they provide, the next generation of wireless networks, 5G and beyond 5G, are expected to provide a wide range of heterogeneous services with very different requirements; ranging from high data rates to ultra-low latency. Thus, the current wireless network architectures are undergoing a deep transformation to adapt themselves to the future network requirements, with special emphasis on the network edge.

A significant portion of current network traffic is content-oriented with a high number of repetitions, a small set of popular multimedia contents are requested by many users. We can exploit the repetitive nature of such demands through caching; storing popular contents at the network edge can alleviate network congestion and reduce content delivery latency. To this purpose, in this dissertation a cache-aided wireless interference network, in which both the edge nodes (ENs) and the user devices are equipped with cache memories is studied. Each user requests one file from a library of $N$ popular files. The goal is to design the cache contents without the knowledge of the particular user demands, such that all possible demand combinations can be satisfied reliably over the interference channel. Novel transmission schemes are presented for both centralized and decentralized caching architectures. Following, we extend the study to the fog radio access network (F-RAN), in which KT ENs connected to a cloud server via orthogonal fronthaul links, serve K_R users through a wireless Gaussian interference channel. In this scenario, the ENs can fetch content unavailable in their caches using the fronthaul links.

Finally, the management and orchestration (MANO) of resources by virtual network functions (VNFs) is studied in this dissertation. Caching and other future 5G technologies rely on the wireless network edge to provide the necessary infrastructure and network functions (NFs) for them to work. Traditionally, the deployment of new NFs has been done through the acquisition and installation of proprietary hardware running licensed software. To overcome this limitation, and to obtain a more flexible network architecture, network function virtualization (NFV) has been proposed. With virtualization, NFs are run as standalone software instances that can be decoupled from the underlying hardware, such that VNFs can be deployed in a shared pool of resources and migrated among hardware entities. This technology entails the management and orchestration of the underlying infrastructure to optimize network's performance. Current threshold-based policies inefficiently over-provision network resources and under-utilize available hardware, incurring a high costs for network operators, and consequently, the users. In this dissertation, a MANO algorithm for VNFs allowing a central unit to learn to autonomously re-configure resources (processing and storage) is presented. The proposed algorithm learns to deploy new VNF instances, or offload them to the cloud, depending on the network conditions, the available pool of resources, and the VNF requirements, with the goal of minimizing a cost function that takes into account the economic cost as well as the latency and the quality-of-service (QoS) experienced by the users.