Subject Ph. D.

Modern wireless networks operate in highly dynamic environments, characterized by constant variations in radio conditions, traffic patterns, and node mobility. These dynamics make any attempt to model and predict network behavior accurately particularly challenging. While simulations remain an essential tool for performance design and evaluation, traditional models often fail to capture the temporal and contextual variability of real-world environments.

To overcome these limitations, Digital Twins (DTs) offer an innovative approach [2,3]: they act as virtual replicas of physical systems that remain synchronized in real time through continuous data exchange. Applied to wireless networks, this concept paves the way for adaptive management, where the network becomes aware of its own state, capable of self- configuration and proactive performance optimization.

Recent studies on Network Digital Twins (NDTs) have enabled dynamic digital representations of networks. Two main methodological families currently dominate the field [4]:

1. Physics-based approaches, which rely on analytical models or high-fidelity simulations (e.g., ray tracing [5]) to achieve remarkable realism, albeit at a prohibitive computational cost. Conversely, simpler models such as the Log- Distance Model provide greater computational efficiency at the expense of accuracy [6].

2. Data-driven approaches, which leverage machine learning to model network behavior from real-world measurements (e.g., [7]). Although promising, these approaches still face three major limitations: (i) they often rely on idealized and extensive telemetry assumptions (e.g., [3]), (ii) they tend to model the network as a whole, overlooking link-level heterogeneity [1], and (iii) they struggle to maintain accuracy in highly variable contexts.

Within this landscape, the thesis distinguishes itself through a clear focus on measurement frugality and multi- level scalable modeling. Rather than accumulating massive datasets, it aims to intelligently select and exploit the most informative measurements to preserve an accurate network representation at minimal cost. It will also introduce a hierarchical modeling framework capable of dynamically adapting to topology and load changes, while ensuring responsiveness suitable for real-time applications. More precisely, the research will be structured around two main axes:

1. Frugal telemetry for traffic characterization: the thesis will propose adaptive telemetry mechanisms combining passive observation and lightweight active probing to achieve an optimal trade-off between accuracy, data collection cost, and update latency.

2. Scalable and fine-grained modeling of wireless links: the thesis will develop a hybrid approach based on variable-granularity learning, where wireless links are dynamically grouped according to their statistical and topological similarities. This will enable the sharing of predictive models while preserving local behavioral fidelity. The approach will aim to balance accuracy, scalability, and responsiveness.

This work will constitute a technological and experimental research endeavor with strong scientific depth, positioned at the intersection of network modeling, intelligent telemetry, and machine learning applied to wireless networks. The expected contributions could support future developments in 6G networks, massive IoT systems, and autonomous communication infrastructures.

References : - [1] Samir Si-Mohammed and Fabrice Theoleyre. "Per Link Data-driven Network Replication Towards Self- Adaptive Digital Twins". In: IEEE MSWiM. 2025. - [2] Emmanuelle Abisset-Chavanne et al. "A Digital Twin use cases classification and definition framework based on Industrial feedback". In: Computers in Industry 161 (2024), p. 104113. - [3] Mehdi Kherbache et al. "Constructing a Network Digital Twin through formal modeling: Tackling the virtual-real mapping challenge in IIoT networks". In: Internet of Things 24 (2023), p. 101000. - [4] Adil Rasheed, Omer San, and Trond Kvamsdal. "Digital twin: Values, challenges and enablers from a modeling perspective". In: IEEE access 8 (2020), pp. 21980-22012. - [5] Andrew S Glassner. An introduction to ray tracing. Morgan Kaufmann, 1989. - [6] Eduardo Nuno Almeida et al. "Machine Learning Based Propagation Loss Module for Enabling Digital Twins of Wireless Networks in ns-3". In: Proceedings of the 2022 Workshop on ns-3 (Best paper). 2022, pp. 17-24. - [7] Miquel Ferriol-Galmés et al. "Building a digital twin for network optimization using graph neural networks". In: Computer Networks, 217 (2022), p. 109329.