Future communication systems such as 6G networks are evolving toward highly distributed, autonomous, and heterogeneous infrastructures integrating cloud-edge continuum architectures, Open RAN (O-RAN), massive IoT deployments, edge computing, and highly dynamic wireless environments.

These systems are expected to support demanding services such as mission-critical communications, industrial automation, autonomous mobility, and immersive applications, operating under highly dynamic traffic conditions, frequent topology changes, fluctuating resource availability, and stringent latency and reliability requirements.
Managing such systems through risk-free configuration, optimization, and evolution operations is becoming increasingly challenging. This is particularly true when performing real-time network optimization, operational what-if analysis, network troubleshooting, or planning network upgrades and extensions.

To address these challenges, recent research initiatives have investigated the application of the Digital Twin paradigm to communication networks, commonly referred to as Network Digital Twins (NDTs).

An NDT is a virtual representation of a communication network that remains sufficiently aligned with the physical infrastructure to reproduce its operational state and behavior, support predictive analysis, and evaluate hypothetical scenarios before applying decisions to the real system.

However, maintaining an accurate and temporally consistent NDT in large-scale and highly dynamic networks remains a major challenge.

Current NDTs predominantly rely on explicit synchronization mechanisms to maintain fidelity between the physical and virtual systems. Although recent works have introduced AI-assisted prediction mechanisms to reduce synchronization overhead, these approaches do not fully address the problem of dynamically adapting the fidelity of the NDT according to predictive uncertainty, information value, network dynamics, and operational requirements. Adaptive fidelity can be interpreted as a multi-resolution representation mechanism, where the NDT dynamically adjusts its observation granularity, synchronization overhead, and reconstruction accuracy according to information value, predictive uncertainty, network dynamics, and available resources. The main objective of this PhD thesis is to design, develop, and validate an Adaptive Fidelity Management framework enabling scalable and resource-efficient Network Digital Twins for future communication systems.