Channel Characterization for Aircraft Integrated Antennas via Machine Learning

Abstract

Wireless communication with unmanned aerial vehicles (UAVs) will become an integral part of future wireless communication systems. In UAVs real-time data transfer is very critical because UAVs need to be controlled from the ground and their data is also transferred in real time. However, aircraft antennas are prone to airframe shadowing. Aircraft surfaces, on which antennas are placed, obscure the main line-of-sight path. In addition, losses used in link budget analyses show great variability in real time. Therefore, observed end-to-end channels, including all impairments, are in general quite different from theoretical calculations in practice. In this work an end-to-end channel link budget, including all effects, is modelled by applying machine learning methods on measured data obtained during past flights. It is observed that ensemble bagged trees (EBT) and exponential Gaussian process regression (GPR) provide the two best results. Pre-processing data and utilizing raw data are also compared. EBT and exponential GPR can predict the amount of end-to-end losses with 7.49% and 8.07% sensitivity respectively using processed data. When raw data is used as input to the EBT method, it can predict the amount of end-to-end loss with a sensitivity of 2.79%, while a theoretical prediction error is 21.9%.