Aerial-DRaGon: Machine Learning-based Channel Modeling for Airspace Communication Networks

The planning and reliable operation of air-space networks strongly relies on accurate radio propagation modeling. In previous work, we demonstrated that the machine learning-driven DRaGon method outperforms channel prediction for both traditional empirical models as well as raytracing. With this paper, we introduce this new generation of data-driven channel models to aerial networks in which communication nodes move at different heights above the ground. For accurate and time-critical predictions, the proposed Aerial-DRaGon model considers a vast range of features, from communication-specific to geospatial and meteorological. In a comprehensive performance evaluation, the results for Aerial-DRaGon - using several lightweight machine learning algorithms - are compared against real-world measurements from diverse environments. There, it is found that gradient-boosted tree learners yield the best prediction accuracy. The best-performing regressors are further compared with a diverse set of state-of-the-art channel models for both terrestrial as well as aerial networks. As Aerial-DRaGon outperforms the latter, it proves itself as a promising channel prediction candidate for future air-space-networks applications like communication-aware trajectory planning.