Research Projects

The initial 5G rollouts in many countries are just beginning, and operators must devise intelligent ways to manage their complex 5G infrastructure, particularly radio access networks (RAN). The 5G standards propose eight functional splits for mapping various radio functions onto RAN components, with one chosen based on user traffic and network conditions. Functional splitting allows decentralization of radio functions, achieving cost savings through resource pooling. Virtualized radio access networks (vRANs) provide flexibility in assigning processing capacity to the network by using servers to host disparate RAN functions. However, managing vRANs efficiently is a big headache for operators. With the amount of data emanating from 5G networks, operators are looking towards data-driven management approaches powered by Artificial Intelligence (AI) and Machine Learning (ML). However, these approaches suffer from problems such as longer training times and slow inference time, and need to be validated due to potential impact on service-level agreements (SLA). To address these issues, this proposal aims to create a network digital twin (NDT) for a virtualized 5G RAN that interacts with the network and captures data used to train AI/ML models running within the NDT. The proposed NDT leverages advancements in ML, specifically agent-based learning, which can train itself based on past events in the network and make decisions for current conditions while considering future demands. The developed NDT aims to help operators planning to deploy vRAN and give them confidence in using ML-based network management solutions.