Machine-Learning Identification of Airborne UAV-UEs Based on LTE Radio Measurements

The overall cellular network performance can be optimized for both ground and aerial users, if different treatment is given for the two user classes. Airborne UAVs experience
different radio conditions that terrestrial users due to clearance in the radio path, which leads to strong desired signal reception, but at the same time increases the interference. Based on this, one can for instance use different interference coordination techniques for aerial users as for terrestrial user and/or use specific mobility settings for each class. This paper compares three different classification algorithms, which use standard LTE measurements from the UE as input, for detecting the presence of airborne users in the network. The algorithms are evaluated based on measurements done with mobile phones attached under a flying drone and on a car. Results are discussed showing the advantages and drawbacks for each option regarding different use cases, and the compromise between specificity and sensibility. For the collected data results show reliability close to 99% in most cases and also discuss how waiting for the final decision can even improve this accuracy to values close to 100%.