Design and Implementation of Dual-task Picture Information Semantic Communication Architecture Based on Generative Artificial Intelligence

With the rapid development of Generative AI technology, especially the rise of large-scale language models and diffusion models, the field of semantic communication is facing new opportunities and challenges. Traditional semantic communication systems have some limitations in conveying semantic information in a dynamic environment, but generative artificial intelligence (AI) is expected to break through this bottleneck with its ability of context understanding and data generation. This paper proposes GenAI-SC, a generative semantic communication architecture that utilizes large-scale pre-training models for semantic information extraction and adopts a diffusion model as its core. The architecture extracts and compresses semantic information through large-scale models and utilizes diffusion models to achieve high-quality image recovery, which is especially suitable for complex communication environments. To verify the effectiveness of this architecture, this paper design two task modes, plain text diffusion image generation and text-image joint diffusion generation, to flexibly cope with different transmission tasks. The experimental results show that GenAI-SC performs well under different channel conditions and inference steps, especially in the low signal-to-noise ratio environment, showing that GenAI-SC has good multitasking capability and communication robustness. This paper provides an innovative path for the design and implementation of semantic communication system.