Toward a Time-Bounded Solution for Real-Time Latency Prediction in Dynamic 5G Communication

Abstract

The growing adoption of safety-critical applications in domains such as Industry 4.0 and autonomous driving places strict latency requirements on Fifth Generation (5G) wireless networks. Existing approaches typically rely on latency measurements, detecting violations only after they occur, or on pre-planned network configurations, which are impractical in highly dynamic environments such as mobile device communication. In this work, we propose a proactive analytical approach for predicting Worst-Case Latencies (WCLs) in 5G networks before violations occur. Our method derives upper bounds on latency analytically, ensuring both explainability and computational efficiency. Additionally, we introduce the concept of a Validity Interval (VI), which quantifies how long a latency prediction remains valid. We extensively evaluate our solution in synthetic and realistic simulations to define the impact of key parameters of our new model on global pessimism, computational overhead, and accuracy. Our results show that with complete knowledge of future channel conditions, our analytical model can consistently upper-bound the simulated worst-case latencies. Even with limited knowledge about future channel conditions, using state-of-the-art forecasting methods, our approach still bounds 99.8% of actual latencies, demonstrating its robustness. With this, our works demonstrate that future 5G networks can enable safe and reliable real-time applications even in highly dynamic environments.