Transformer-Based Semantic Communication System for 3D Point Cloud Transmission

This paper introduces the Transformer-Based Semantic Communication System (TSCS), a novel framework for efficient and robust 3D point cloud transmission. By leveraging transformers to extract semantic features and reconstruct point clouds, TSCS addresses key limitations of traditional octree-based methods, such as inefficient feature extraction and susceptibility to the cliff and leveling effects. The cliff effect refers to a sharp decline in performance below a certain signal-to-noise ratio (SNR), while the leveling effect occurs when further increases in SNR yield diminishing returns due to inherent system constraints. Experimental results demonstrate that TSCS achieves superior reconstruction quality, outperforming traditional methods in terms of PSNR D1 (point-to-point) and PSNR D2 (point-to-plane) metrics. These findings highlight the potential of TSCS for realworld applications, including autonomous driving, augmented reality, and robotics. Future work will focus on scaling TSCS to handle large point clouds and integrating multi-modal data to further enhance its capabilities.