Cooperative Semantic Communication With On-Demand Semantic Forwarding

In this paper, a deep learning-based cooperative semantic communication system is proposed on relay channels. In order to enhance the reliability and adaptability of the system to varying channel conditions, an on-demand semantic forwarding framework is established, where the relay attempts to recover and re-transmit the source semantic information, as required by the destination. To be specific, for determining whether a semantic forwarding is needed from the relay, a semantic similarity check $(\textit {SSC})$ is proposed, based on which the degree of semantic information recovery at the destination can be accurately estimated. On the other hand, in order to effectively merge the semantic information received through different paths, a semantic combining $(\textit {SeC})$ method is proposed by combining the semantic features abstracted from both the direct link and the relay link. For achieving a desirable performance trade-off between the degree of semantic information recovery and the transmit energy consumption, a new metric of semantic energy efficiency (SEE) is proposed. Simulation results verify the performance gains achieved by the proposed cooperative semantic communication system with on-demand semantic forwarding, as compared to the state-of-the-art schemes employing separate source-channel coding in low-to-medium signal-to-noise ratio (SNR) regimes. Furthermore, as compared to the case with always-forwarding, almost the same performance is achieved by the proposed on-demand forwarding, but with lower energy consumption.