Robust Fronthaul in Wireless Networks: A Caching and Traffic Prediction Approach

The accelerated advancement of fifth-generation (5G) mobile communications has significantly increased the number of users accessing wireless networks, a trend that is expected to further escalate with the advent of sixth-generation (6G) technologies. To handle the massive traffic demands and relieve pressure on cellular networks,caching has gained attention that not only reduces network latency but also alleviates congestion on the fronthaul link. The goal of this thesis is to establish a foundation for a robust fronthaul link in a cloud radio access network (C-RAN). The first solution is the implementation of a hybrid millimeter-wave/ microwave fronthaul link. This hybrid approach leverages the high capacity of millimeter-wave links while maintaining reliability, as it allows for a switch to lower frequencies in the event of blockage. The second solution is dynamic caching, which can alleviate the load on the fronthaul link by storing content at the network’s edge. In the first part of this thesis, our goal is to address user requests in a C-RAN with a hybrid fronthaul link. By modeling the environment as a hidden mode Markov decision process (HM-MDP), we aim to minimize the long-term network cost through dynamic caching and fetching, along with choosing the link frequency using dynamic programming (DP). In our model, we assume a non-stationary but known content request pattern. In the second part of the thesis, we lean toward a more realistic way of modeling the environment, which is non-stationary and unknown. To deal with this unknown environment, we develop an online, multivariate, non-parametric change point detector(CPD) to detect mode changes. Operating under the assumption of an HM-MDP and with the aid of reinforcement learning, we optimize caching and fetching decisions to minimize the total network cost. Finally, we explore the caching optimization problem using a real-world dataset, with a focus on addressing dynamic content creations and requests. Our approach entails designing a predictive caching strategy, in which we learn about upcomingrequests and their life cycles. Subsequently, by integrating this knowledge into a partially observable Markov decision process solver, we aim to maximize cache hits within the network.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.