Proactive Resource Management for Predictive 5G Uplink Slicing

Abstract

The 5th generation of mobile communication networks (5G) introduced the concept of network slicing for enabling multiple, diverging service types by providing virtually independent communications within one physical network. While Ultra-Reliable Low Latency Communication (URLLC) aims to provide latency guarantees below 5 ms for mission-critical applications such as Smart Grid as well as Industry 4.0, Enhanced Mobile Broadband (eMBB) focuses mainly on high data rates. Thus, since different Key Performance Indicators (KPIs) such as latency, data rate or time-criticality need to be considered, the allocation of resources between corresponding slices is challenging. This work, therefore, aims to reduce uplink latency for URLLC transmissions by deploying Proactive Grants to minimize the impact of time-consuming scheduling requests and any negative influence on other slices. Resources are allocated proactively by base stations utilizing Machine Learning (ML) models trained on real-world measurements. An experimental evaluation via an Software-Defined Radio (SDR)-based physical testbed demonstrates delay reductions towards the mission-critical threshold while simultaneously increasing spectral efficiency. Compared to Round Robin (RR)-based slicing, latency decreases by 49 %, while maintaining a high throughput of 98 % in the eMBB slice.