Improving TSN Time Synchronization via Bayesian Optimization: Experimental Evaluation in Wired and Wireless Networks

The goal of Time-Sensitive Networking (TSN) standards is to give conventional Ethernet and Wi-Fi networks deterministic communication capabilities. High-precision time synchronization, made possible by the IEEE 802.1AS Generalized Precision Time Protocol (gPTP), is a basic requirement for these deterministic services. Deploying gPTP, on widely used soft-real-time systems, such as Linux on generic x86 hardware, is difficult. Because of variations in operating system latency, network media properties and hardware oscillator stability, standard "out-of-the-box" configurations frequently fail to meet strict industrial requirements. This thesis presents a new automated approach for PTP servo parameter using Bayesian Optimization (BO). The research looks into how well the open-source LinuxPTP stack synchronizes across several Intel NUC devices and transmission media. The experimental results for wired Ethernet networks clearly show that the optimal configuration for servo parameters is strongly hardware-dependent. Most importantly, it has been shown that processor power management states (C-states) are one of the major sources of synchronization jitter and are as critical as optimization algorithms. In the Wireless TSN domain, the research highlights that the impact of non-Gaussian packet delay variation is extremely significant. The optimization framework shows that it can identify conservative PI gains that outperform both the default settings and linear‑regression‑based adaptation methods, achieving up to a 22% reduction in the synchronization‑offset RMSE. Moreover, an examination of wireless frequency bands reveals that path stability (multipath effects) is more significant than spectrum congestion, for synchronization accuracy. The research presents a clear and effective methodology for achieving robust synchronization accuracy across different platforms in the microsecond domain, filling the gap between theoretical TSN standards and practical noise-resilient implementations.