Multimodal Fusion-Enabled Scenario-Adaptive Real-Time Channel Prediction for UAV Communications

Abstract

Accurate modeling and prediction of UAV channels remain a major challenge in complex urban macrocell environments. In this paper, a channel impulse response (CIR) prediction method for UAV communications is first proposed, based on image data, three-dimensional (3D) point cloud data, global positioning system (GPS) data, and communication settings. It can support both sub-6 GHz and millimeter-wave frequency bands. A multimodal communication scenario is built according to the 3rd Generation Partnership Project (3GPP) urban macrocell standard, thereby aligning with the philosophy of standardized model. This scenario allows the collection of both electromagnetic propagation data and physical sensing data at the same time. A multimodal spatial perception model is then developed. It fuses 3D spatial point cloud information with electromagnetic parameters. Through this fusion, a mapping from complex physical scenes to UAV channel is established. To improve prediction generality, a CIR prediction network with trajectory generalization ability is designed. This network enables accurate channel prediction along multiple UAV flights, and multiple frequency bands. A dynamic output mechanism is further introduced, enabling the model to adaptively predict a variable count of multipath components in real-time. Experimental results show that the proposed method can predict the CIR in complex urban environments. These results demonstrate that the proposed method provides a solution for UAV channel modeling in urban environments.