Generative Semantic Communication: Diffusion Models Beyond Bit Recovery

Semantic communications have the potential to play a key role in next-generation AI-native communication systems, especially when combined with the expressivity power of generative models. In this paper, we focus on image transmission and present a novel generative-based semantic communication framework, whose core is a generative model operating at the receiver side. This model regenerates images suitable for downstream applications such as detection, reconstruction, and positioning of semantically relevant objects in the scene observed from the sensors present at the sender side. We devise the encoding rule to transmit only what is strictly relevant to trigger the generative model to fulfill the scope of the transmission. Furthermore, we propose a training strategy to make the generative model robust against additive noise due to propagation through the communication channel. We prove, through an in-depth assessment of multiple scenarios, that our method outperforms existing solutions in generating high-quality images with preserved semantic information even in cases where the received conditioning content is significantly degraded or compressed. More specifically, our results show that objects, locations, and depths are still recognizable even in the presence of highly noisy conditions of the communication channel or at very low bits per pixel.