Optimizing Mobile Backhaul Using Machine Learning

Abstract

The thesis focuses on the analysis of current limitations of the mobile backhaul solutions technology when applied to 5G technology. The fast growth in connected devices along with the introduction of 5G technology is expected to cause a challenge for efficient and reliable network resource allocation. Moreover, massive deployment of Internet of Things and connected devices to the Internet may cause a serious risk to the network security if they are not handled properly. To solve those challenges, the Mobile Back haul (MB) infrastructure must increase capacity, improve reliability, availability and security.

Software Defined Networks (SDN) and Machine Learning (ML) techniques were used on top of the basic IP routing to measure and estimate the available resources in the network and apply Traffic Engineering (TE) logic to reallocate available resources to newly added slices. The experiment was performed in a virtual environment using Mininet simulator tool and other opensource software and ML algorithms.

In this thesis, a system was developed to measure the existing resources in the mobile backhaul and redistribute dynamically to different network slices either existing or new slices to make sure that each slice requirements are met.

The thesis includes an early prototype of the Mobile Backhaul Orchestrator (MBO) that will be simulated to confirm it can effectively allocate resources to new slices while maintaining existing slices, and that it can contain the traffic within a slice during peaks without affecting traffic in other slices.