NetLabeller: Architecture with data extraction and labelling framework for beyond 5G networks

Abstract

The next generation of network capabilities coupled with artificial intelligence (AI) can provide innovative solutions for network control and self-optimisation. Network control demands a detailed knowledge of the network components to enforce the correct control rules. To this end, an immense number of metrics related to devices, flows, network rules, etc. can be used to describe the state of the network and to gain insights about which rule to enforce depending on the context. However, selection of the most relevant metrics often proves challenging and there is no readily available tool that can facilitate the dataset extraction and labelling for AI model training. This research work therefore first develops an analysis of the most relevant metrics in terms of network control to create a training dataset for future AI development purposes. It then presents a new architecture to allow the extraction of these metrics from a 5G network with a novel dataset visualisation and labelling tool to help perform the exploratory analysis and the labelling process of the resultant dataset. It is expected that the proposed architecture and its associated tools would significantly speed up the training process, which is crucial for the data-driven approach in developing AI-based network control capabilities.