Ilmu Komputer & AI editorial
Open AccessOA2026
DeepNC: A Fast GNN-based Pre-Verification Surrogate for TSN Configuration
A Graph Neural Network surrogate that fuses Network Calculus principles to accelerate TSN configuration-space exploration by over two orders of magnitude.
Jiayi Zhu; Jing Lin; Zelong Tian; Feng He; Luxi Zhao· 2026· DOI 10.48550/arXiv.2607.24398
The core problem
Time-Sensitive Networking (TSN) provides deterministic communication for safety-critical domains, where schedulability guarantees are essential. Formal verification methods such as Network Calculus (NC) are the cornerstone for these guarantees. However, during automated configuration-space exploration, repeated schedulability analysis consumes over 90% of the total configuration time, becoming the primary bottleneck for large-scale TSN configurations. This work introduces DeepNC, a novel pre-verification surrogate module that pioneers the structural fusion of NC principles into a Graph Neural Network (GNN) for TSN configuration-space exploration. Rather than replacing formal verification, DeepNC acts as a high-speed pre-verification filter, reserving computationally expensive formal verification only for promising candidates. The authors are Jiayi Zhu, Jing Lin, Zelong Tian, Feng He, and Luxi Zhao.
Innovation
Extensive evaluations demonstrate that DeepNC significantly improves worst-case delay prediction accuracy over state-of-the-art learning-based methods. Specifically, it increases the average by 55.8% and reduces the average MAPE by 65.3%. More importantly, its high-fidelity regression substantially reduces the number of formal verification calls during configuration-space exploration by 93.25%, while accelerating NC-based verification by more than two orders of magnitude. These results indicate that DeepNC can effectively filter out unpromising configurations, leading to dramatic time savings in the configuration process. The reduction in verification calls directly translates to faster exploration, enabling large-scale TSN configurations that were previously infeasible.
Time-Sensitive Networking (TSN) provides deterministic communication for safety-critical domains, where schedulability guarantees are essential. Formal verification methods such as Network Calculus (NC) are the cornerstone for these guarantees. However, during automated configuration-space exploration, repeated schedulability analysis consumes over 90% of the total configuration time, becoming the primary bottleneck for large-scale TSN configurations. This work introduces DeepNC, a novel pre-verification surrogate module that pioneers the structural fusion of NC principles into a Graph Neural Network (GNN) for TSN configuration-space exploration. Rather than replacing formal verification, DeepNC acts as a high-speed pre-verification filter, reserving computationally expensive formal verification only for promising candidates. The authors are Jiayi Zhu, Jing Lin, Zelong Tian, Feng He, and Luxi Zhao.
DeepNC is designed as a surrogate model that predicts worst-case delays for TSN configurations. It structurally incorporates Network Calculus principles into a Graph Neural Network architecture. The TSN configuration is represented as a graph, where nodes correspond to network elements (e.g., switches, links, flows) and edges represent connections or dependencies. The GNN processes this graph to produce a delay prediction. The model is trained to perform high-fidelity regression, enabling it to act as a pre-verification filter: only configurations predicted to be promising are passed to the formal NC verification. This approach reduces the number of expensive formal verification calls. The architecture can be conceptualized as follows:
Why it matters
The key innovation of DeepNC lies in its structural fusion of Network Calculus principles into a GNN, which allows it to learn accurate delay predictions while respecting domain constraints. By acting as a pre-verification filter, DeepNC addresses the bottleneck of repeated schedulability analysis, which consumes over 90% of configuration time. The 93.25% reduction in formal verification calls and the two-orders-of-magnitude speedup demonstrate the practical impact of this approach. The improvement in prediction accuracy (55.8% increase in and 65.3% reduction in MAPE) further validates the effectiveness of the NC-informed GNN architecture. This work paves the way for scalable TSN configuration in safety-critical domains, where deterministic communication is paramount. Future work may explore extending DeepNC to other formal verification methods and network types.
Who should read this
CS practitioners and researchers
Opening member content…