Enabling Predictive Scheduling in Wireless TSN via Physics-Aware REMs

Abstract

Integrating Time-Sensitive Networking (TSN) with wireless systems presents significant challenges, particularly in industrial environments that demand mobility support and operate under rapidly varying wireless channel conditions. While wired TSN scheduling mechanisms are well established, effective solutions for Wireless TSN remain limited, largely due to their reliance on reactive adaptation strategies. In this work, we leverage Radio Environmental Maps (REMs) to shift the handling of channel variability from real-time traffic scheduling to fast, predictive channel modeling, enabling the pre-computation of communication schedules based on established wired TSN principles within a co-designed communication-computing system. We propose a physics-aware residual training approach that reduces REM training delay by nearly 40% while maintaining high prediction accuracy. The resulting predictive knowledge improves coordination between applications and the communication network, supporting decisions such as mobility route planning toward guaranteed communication determinism.