Generative Network Performance Prediction with Network Digital Twin
Bilgehan Erman, Catello Di Martino
Abstract
Following the successful deployments of the 5G networks, the vision for the 6G era is rapidly forming. Among the expectations of the 6G vision are the creation of digital twin, which is a fusion of the physical and digital worlds but with additional dimensions of reality allowing us to make analytic observations over the full lifecycle of systems from design to realtime operations. Despite the success of the 5G network deployments, several open problems will persist in designing, delivering, and maintaining private wireless networks for autonomous industrial settings. In these dynamic environments with changing use cases and SLA profiles, immediate detection of performance problems becomes critical. The existing instrumentation methods proven useful for public networks are not as effective for maintaining these private industrial networks, where use-case-dependent SLA compliance prediction becomes a necessity. The described solution builds on an existing Network Digital Twin platform used for continuous testing and SLA management of mission-critical wireless industrial private networks. The solution first builds an association between the field measurements and the digital representations of the environment by making use of Deep Neural Network models. Then in subsequent steps, Generative Adversarial Network methods are used to create what-if scenarios with alternate properties and layout configurations to examine potential concerns, and to explore options toward better SLA compliance outcomes.