Robust Semantic Communication Driven by Knowledge Graph
Linsheng Hu, Yihao Li, Hao Zhang, Lu Yuan, Fuhui Zhou, Qihui Wu
Abstract
Semantic communications could improve the transmission efficiency significantly by extracting the semantic information. In contrast to traditional communication systems, semantic communication systems are additionally impacted by semantic noise because of the semantic, in addition to the physical noise present in the wireless communication environment. Semantic noise is a type of noise that leads to decoding errors and misunderstandings of semantic information, which leads to misdirection between the transmitted and received semantic symbols. In this paper, we present a knowledge graph enabled robust semantic communication system to avoid semantic noise from influencing semantic communication systems. In order to improve the robustness of semantic communication systems, we design a knowledge graph shared by both the transmitter and the receiver for encoded symbols representation. The transmitter only needs to transmit the indices of these symbols in the knowledge graph. A knowledge graph completion module is designed at the receiver to restore the semantic. Compared to baseline models for text transmission that merely take physical noise into account, the proposed scheme yields a remarkable improvement in avoiding semantic noise under various signal-to-noise ratios.