VISTA: Video Transmission over A Semantic Communication Approach
Chengsi Liang, Xiangyi Deng, Yao Sun, Runze Cheng, Le Xia, Dusit Niyato, Muhammad Ali Imran
Abstract
Video transmission over ultra-reliable and low-latency communication (URLLC) is a promising trend to support various multimedia services. However, in view of the rapid surge in video content demand along with a stringent resolution requirement, there is an unprecedented burden on wireless networks with limited yet precious bandwidth resources. In this paper, we propose a VIdeo transmission framework over Semantic communicaTion Approach (VISTA), where semantics rather than all bits of a video should be transmitted with the aim of reducing bandwidth consumption while keeping a high visual perception. Specifically, the semantic segmentation module in VISTA is first developed to classify and encode the dynamic and static segments in source video separately. Next, the semantic location graphs (SLGs) are built to describe semantics and location relations among the detected dynamic objects. Through the joint source-channel coding (JSCC) module which is adaptive to different channel conditions, the encoded semantic features and SLGs are transmitted over wireless channel. Finally, the video is recovered at the receiver end based on the distorted semantic features and SLGs with the assistance of frame interpolation module. Simulation results demonstrate that VISTA outperforms two benchmarks in terms of required bandwidth reduction and robustness against channel noise, further satisfying the requirements on URLLC.