A Novel Neural Point Field Framework for End-to-End Wireless Channel Modeling

In this paper, we present a novel neural wireless channel modeling methodology by leveraging a point-cloud-based neural network. During the training phase, the proposed method learns the interaction between wireless transceivers and their environment with wave propagation properties. After training, our work serves as a neural forward model for rapidly predicting the wireless channel for new transceiver locations. As such, it establishes an end-to-end pipeline for network planning and deployment optimization. A noteworthy aspect of our approach lies in the intriguing point-cloud-based neural framework, allowing seamless generalization to large-scale scenes.