Content-Aware Semantic Communication for Goal-Oriented Wireless Communications

While semantic communications can remarkably reduce the amount of the data transmission traffic without missing critical information, its applications to the efficient goal-oriented wireless communications remain limited. In this paper, we proposed the Content-Aware Semantic Communication (CA-SC) framework, which fuses both global and task-related semantic information via the semantic decoder, thus achieving the goal-oriented wireless communications. In particular, the CA-SC is based on the attention map to locate the task-related semantic information for supporting an adaptive rate allocation scheme. In order to further improve the coding rate as well as reconstructing intended data, we formulate a rate-distortion optimization problem as the loss function for the CA-SC framework, which aims to jointly optimize the semantic codec and the channel codec. Numerical results show that the CA-SC scheme can achieve better performance compared with existing semantic codec scheme and the traditional codec scheme in the image reconstruction task and the object detection task.