A Tiers-of-Knowledge Framework for 6G Digital Twinning

Abstract

Radio access networks (RAN) are facing a major transformation driven by the need for site-specific automation and customization in the 6G era. This transformation is enabled by the use of geo-spatial digital twins (GSDT) together with artificial intelligence (AI). This paper provides an examination of how these technologies integrate to revolutionize RAN performance, arguing that AI and digital twins form a mutually beneficial cycle: AI is essential for creating and maintaining high-fidelity twins via techniques like neural ray tracing and semantic fusion, while the twins provide the synthetic data necessary to train robust AI models for network management. We study the operational and planning impact of this transformation on RAN and propose a novel Tiers of Knowledge framework, classifying network prior knowledge from Tier 0 (no prior knowledge) to Tier 4 (site-specific predictive knowledge) and a theoretical Oracle bound. This taxonomy moves beyond binary definitions of twinning to systematically map the depth of environmental awareness to the feasibility of specific downstream tasks—ranging from static coverage planning at lower tiers to fully proactive, AI-driven beam tracking and mobility management at higher tiers. Within this framework, we study the benefits of digital twins in critical use cases such as handover optimization, beam management, and positioning. Finally, we analyze the economic and operational trade-offs of twinning, demonstrating that higher visual fidelity does not always equate to better network performance, and lay down practical considerations for future research in 6G digital twinning. We complement this taxonomy with a mathematical grounding that casts each tier as a Bayesian estimation problem with progressively richer side information, and with system-level quantification of digital twin-enabled improvements to example RAN functions. We study digital twin synchronization as a use-case-dependent latency budget analysis that gates dynamic twinning. We further formalize the joint cost surface, establishing utility and cost of digital twin-enabled RAN as a function of fidelity and tier of knowledge, and characterize the bounded sufficiency point at which additional fidelity ceases to be worth its cost.