A Semantic Communication System for Point Cloud

Abstract

Point cloud, as a 3D representation, finds wide applications in domains such as autonomous driving, virtual reality (VR), and augmented reality (AR). However, traditional communication systems are not well-suited for large-scale point cloud data transmissions, as such systems operate at the bit level without leveraging the inherent semantic information of the point cloud. This paper introduces a point cloud-based semantic communication system (PCSC) that leverages AI-based encoding techniques to extract semantic information from the point cloud. Furthermore, joint source-channel coding (JSCC) technology is employed to overcome noise channel distortion and address the “cliff effect” prevalent in traditional communication methods. Additionally, the proposed system achieves controllable coding rates without requiring extensive network fine-tuning. By analyzing the significance of the encoded semantic vector, the method discards semantically-unimportant information, enhancing transmission efficiency. Moreover, the PCSC is integrated with the recently proposed model division multiple access (MDMA) technology, resulting in a multi-user point cloud MDMA transmission system (M-PCSC). Experimental results demonstrate that the proposed method surpasses traditional approaches by 10 dB under the PSNR D1 and PSNR D2 metrics while effectively mitigating the “cliff effect” in transmission.