LoS Probability Prediction for A2G mmWave Communications by Using Ray-tracer Under Virtual Urban Scenarios

Line-of-sight (LoS) probability is of great significance for reliable air-to-ground (A2G) millimeter wave (mmWave) communication. In this paper, a machine learning (ML)-based LoS probability prediction method for A2G mmWave communication is proposed. A fully connected neural network optimized by Bayesian optimization is utilized to predict LoS probability accurately with the type of scene, ground distance and vertical height between unmanned aerial vehicles (UAVs) and ground receivers as the inputs. For the difficulty of measured data acquisition, massive channel data is generated by ray-tracing (RT) simulation under the virtual scenarios reconstructed by statistical building characteristics. Simulation results show that the proposed method is consistent well with the RT simulation and verified the good performance of LoS probability prediction in four virtual urban scenarios by comparison with other existing models.