Predictive QoS for cellular connected UAV payload communication

Abstract

Unmanned aerial vehicles (UAVs), or drones, are revolutionizing industries due to their versatility, affordability and applicability. Reliable communication links are essential for UAV operations, especially for beyond visual line of sight scenarios where drones are flown beyond the operator’s line of sight. Cellular networks, particularly in the context of 5G and beyond, offer potential solutions to meet the data-intensive demands of UAV applications. This study explores the feasibility of predictive quality of service for forecasting uplink (UL) throughput quality of service (QoS) parameter in UAV payload communication links. Comprehensive field tests were conducted to ensure accurate real-world results, as simulations may not fully capture real-world complexities. Field trial measurements were conducted in a sub-urban area to evaluate drone performance at various altitudes and bands. This sheds light on potential challenges and trade-offs for cellular-connected drones and their coexistence with terrestrial users. Drones flying at high altitudes often experience line of sight propagation, causing them to undergo frequent handovers between multiple base stations. Field trials demonstrated that drones connected to a 700 MHz signal encountered minimal interference and no handovers. Conversely, drones connected to the 3500 MHz frequency band faced multiple handovers, highlighting the complexities of UAV-cellular integration and emphasizing the significance of frequency band selection in drone applications. By harnessing machine learning (ML) models and comparative analysis of centralized and federated learning methods, this research investigates ML model performances in forecasting UL throughput based on prediction accuracy. The findings emphasize the importance of diverse training data and highlight the impact of frequency bands on UAV communication. These insights lay the groundwork for addressing UAV communication complexities and advancing the integration of machine learning and network dynamics for improving UAV operations.