SAURON: A Zero-Shot Network Digital Twin for 5G Traffic and Threat Generation

As modern 5G networks evolve, network management has become increasingly complex, making reliable operations using only raw telemetry impractical. Network digital twins (NDT) are emerging as a solution to model complex scenarios without compromising the real network. Unfortunately, historical datasets are limited, and in most cases private, restricting data-driven network management. We present SAURON (Synthetic Autoregressive Unified Replica for Operator Networks), a zero-
shot NDT that synthesizes statistically faithful 5G packet traces without the need for retraining. Built on Chronos-2, a pre-trained autoregressive transformer, SAURON augments its conditional mean forecast with a spectral diversity step that injects controlled phase noise, followed by quantile mapping that anchors every feature’s marginal distribution to the empirical reference and enforces physical network constraints. Then, SAURON reconstructs the packet trace, preserving microscopic burstiness, to generate synthetic traces indistinguishable from real ones. We evaluate SAURON on real 5G captures, spanning from a DNS-flooding attack to a nominal multi-user baseline. Results shows that SAURON almost reach the theoretical lower bound of residual distributional distances, outperforming literature solutions.