Machine Learning-Based Field Data Analysis and Modeling for Drone Communications

In recent years, unmanned aerial vehicle (UAV), also called a drone, is getting more and more important in many emerging technology areas. For communication area, the drone also takes an important role in lots of significant topics like emergency communications, device-to-device (D2D) communications, and the Internet of Things (IoT). One of the important drone applications is to collect and share data among drones and other aircraft, which is useful for drone control so that dangerous conditions can be avoided. In particular, the drone control and safety guarantees are difficult to attain, especially, when drones fly beyond the line of sight (BLOS). For this reason, we develop a drone location information sharing system using the 920-MHz band. We use this system to do a long distance propagation field experiment for model establishment. Unfortunately, the current data collection for model establishment work needs a great effort and time to do experiments to collect a huge number of data for data analysis so that a suitable model can be established. Therefore, in this paper, we propose a novel method, which is based on machine learning approach, to data analysis and model establishment for drone communications, so that the effort and cost for establishing model can be reduced and a model, which captures more details about the drone communications, can be obtained. The results of this paper validate that the proposed method can indeed establish a more complicated model with less effort. Specifically, from the distribution of the training error, it can be known that there are over 80% training errors with intensity less than 5, which ensures the error performance of the proposed method.