Digital Twin assisted AI for sustainable Radio Access Networks

This CD laboratory aims to create a foundation for using artificial intelligence (AI) based learning and training methods in wireless networks in various scenarios, with the benefits of efficiency, sustainability, and reliability. For this purpose, we develop so-called “digital twins” (DT), representing enormously different environments such as trains, industrial sites, and dynamic environments, along with the corresponding wireless access and user populations.

Digital Train Twin
Digital Twin Assisted AI for sustainable RAN

Mobile networks have been an integral part of everyday life and work for some time and are increasingly used in industrial environments. Accordingly, the requirements for these networks can be highly diverse depending on the user, situation, and application.

Train passengers want to work, communicate via phone or the internet, or use digital entertainment in a fully occupied train. Speed, design of the railway carriage (metal and window materials that hinder the propagation of the signal), and environment (when it travels through rural areas with low network coverage) pose significant challenges for performant and reliable mobile coverage. On the other hand, digitizing industrial processes requires an extremely high level of reliability in networking machines in real time. These requirements demand more and more information from the mobile network, making it act as a sensor.

The first generations of mobile networks could hardly consider differentiated user requirements during operation. It was only with the 5G standard that functions in the wireless network were implemented, allowing for dynamic resource optimization and the so-called network slicing. The challenge now is to develop methods for realizing dynamic optimization. In this CD laboratory, the automated optimization is implemented through a model-based agent. This agent must always consider the current state of the wireless interface as well as a prediction of future load profiles. The successful implementation is thus based on digital representations of environments, and this is where the aforementioned digital twins come into play.

The titular digital twins, which form the basis of data-driven AI management, are to be understood to represent a physical process or object in a natural environment as a virtual object. They enable validation, simulation, or representation of its current or future status. One possible solution for building them is using traditionally purely data-driven machine learning (ML) methods, which, however, require enormous resources due to their need for training data.

By developing the various aspects of the respective network along with the environment and users in the form of interactive digital twins of railway networks, industrial sites, and dynamic environments that interact with each other, laboratory leader Philipp Svoboda and his team can use ML in a resource-efficient manner in various scenarios to improve wireless connectivity.