Empirical performance analysis and ML-based modeling of 5G non-standalone networks
Konstantinos Kousias,Mohammad Rajiullah,Giuseppe Caso,Ozgu Alay,Anna Brunstrom,Usman Ali,Luca De Nardis,Marco Neri,Maria-Gabriella Di Benedetto
DOI: https://doi.org/10.1016/j.comnet.2024.110207
IF: 5.493
2024-01-24
Computer Networks
Abstract:Fifth Generation (5G) networks are becoming the norm in the global telecommunications industry, and Mobile Network Operators (MNOs) are currently deploying 5G alongside their existing Fourth Generation (4G) networks. In this paper, we present results and insights from our large-scale measurement study on commercial 5G Non Standalone (NSA) deployments in a European country. We leverage the collected dataset, which covers two MNOs in Rome, Italy, to study network deployment and radio coverage aspects, and explore the performance of two use cases related to enhanced Mobile Broadband (eMBB) and Ultra-Reliable Low Latency Communication (URLLC). We further leverage a machine learning (ML)-based approach to model the Dual Connectivity (DC) feature enabled by 5G NSA. Our data-driven analysis shows that 5G NSA can provide higher downlink throughput and slightly lower latency compared to 4G. However, performance is influenced by several factors, including propagation conditions, system configurations, and handovers, ultimately highlighting the need for further system optimization. Moreover, by casting the DC modeling problem into a classification problem, we compare four supervised ML algorithms and show that a high model accuracy (up to 99%) can be achieved, in particular, when several radio coverage indicators from both access networks are used as input. Finally, we conduct analyses towards aiding the explainability of the ML models.
computer science, information systems,telecommunications,engineering, electrical & electronic, hardware & architecture