Group-based link modeling for wireless digital twins: Towards accurate network performance prediction

Abstract

Wireless networks are increasingly deployed in diverse domains, from best-effort object tracking to real-time control in smart factories. Yet, their performance strongly depends on configuration choices, especially at the MAC level. Thus, a single homogeneous configuration is often suboptimal due to the heterogeneous nature of individual links. We argue that Digital Twins (DTs) are a promising enabler for autonomous networks, capable of adapting configurations dynamically to prevailing conditions. However, global modeling approaches in DTs make it difficult to capture link-level variability and to accurately model the impact of configuration changes on performance. In this work, we propose a link-oriented prediction model designed to serve as a cornerstone for future wireless Digital Twins. Our model estimates the Packet Reception Rate (Packet Reception Rate (PRR)) under different MAC configurations, capturing the unique characteristics of each communication link. To improve scalability in large deployments, we explore a clustering-based approach, where predictive models are trained per group of similar links rather than per individual link. Our experimental evaluation shows that these data-driven methods effectively capture link heterogeneity while offering robust prediction accuracy and enhanced generalization capabilities.