LOS Classification of UAV-to-Ground Links in Built-Up Areas

The use of UAVs for various applications is a rapidly growing research field nowadays. Knowledge of the wireless UAV-to-ground propagation channel is crucial for designing an efficient communication system and for evaluating its performance. The presence of LOS is essential for radio network planning and RF coverage prediction. Built-up areas contain a mixture of LOS and NLOS conditions due to buildings shadowing, states which cannot be easily distinguishable. Hence ray-tracing simulations were performed to model the UAV trajectory, the site-specific urban environment and terrain. In this contribution we develop a method to identify the LOS and the NLOS conditions based on the NB and WB channel statistical parameters: received power, K-factor, mean ToA and delay spread, and their combinations. Population classifications using a single feature and multiple features were investigated. We found that a classification between LOS and NLOS populations based on a single feature leads to poor to moderate performances depending on the feature. However, combing a few features improved the classification performance. Variances of KNN, decision tree and SVM classifiers were trained based on all features, resulting in good true positive and true negative rates of 87% and 75%, respectively.