A Semantic Communication System Based on Vector Quantization and Generative Model
Yongyi Miao, Jun Hong Yan, Youfang Wang, Zhongdang Li, Die Hu · 2024
Most existing semantic communication systems directly map source data to channel symbols, resulting in constellation points that may appear at arbitrary positions within the constellation, which is somewhat inconsistent with contemporary digital communication system design. In this paper, we propose a high-performance deep learning-based quantized vector (VQ) semantic communication (SC) system, named VQ-DeepSC-E, where only limited indices need to be transmitted at the transmitter to reconstruct images at the receiver, significantly reducing communication costs and aligning more closely with digital communication systems. Compared to existing VQ semantic communication systems, the proposed method utilizes generative adversarial network (GAN) to extract image features, enhances feature indexing performance and reduces the generation of artifacts in reconstructed images. The simulation results indicate that the proposed method achieves superior Structural Similarity Index (SSIM) and Learned Perceptual Image Patch Similarity (LPIPS) scores compared to the existing VQ-DeepSC method.