Point-Cloud Transmission Oriented 3D Semantic Communication for Accurate Scene Generation-Enabled Digital Twin

Rui Xu, Yuanhang He, Bei Chen, Wenkai Huang, Boyu Han, Jianhua Li

Abstract: Accurate 3D scene generation is a cornerstone of digital twin (DT) systems. Among existing approaches, point cloud-based methods stand out for their superior accuracy and robustness. However, the wireless transmission of massive point cloud data, collected by edge devices and LiDAR sensors, introduces significant challenges, including high network throughput demands and the impact of unreliable wireless channels. To address these issues, we propose a novel point-cloud transmission oriented 3D Semantic communication (PCT-3DSC) scheme. The proposed PCT-3DSC integrates an efficient semantic encoder, which eliminates redundant information to reduce transmission overhead, an adaptive channel encoder/decoder to ensure reliable communication under challenging wireless conditions, and an accurate 3D scene generator for reconstructing high-fidelity 3D scenes. Experimental results demonstrate that our scheme achieves fast 3D object generation within 5 seconds while boosting the resolution to 512, significantly outperforming existing state-of-the-art methods in terms of generation quality and robustness.

External IDs:dblp:conf/infocom/XuHCHHL25