UAV-to-ground path loss prediction based on synesthesia of machines

In this paper, a real-time unmanned aerial vehicle (UAV)-to-ground path loss prediction model via intelligent multi-modal sensing-communication integration is developed in the operational mode of Synesthesia of Machine (SoM). In the modeling process, a dataset including multi-modal sensing and communication data is constructed in AirSim and Wireless InSite to support the exploration of the non-linear mapping relationship between physical environment and electromagnetic space. To explore the mapping relationship between the environmental features extracted from multi-modal sensing image data in physical environment and path loss in electromagnetic space, a convolution neural network (CNN) is constructed and trained. Therefore, based on the dataset, the real-time path loss prediction in the UAV-to-ground scenario is achieved. Simulation results show that the prediction average mean square error (MSE) of the proposed model is 6.4641 × 10⁻⁵ in the test set. The accuracy and utility of the proposed model are validated by comparing the prediction results of the model and ray-tracing (RT)-based results.