VimSC: Robust Underwater Acoustic Image Semantic Communication Based on Vision Mamba Model

Underwater acoustic communication is challenged by strong multipath effects, significant path loss, and limited available bandwidth, which result in low transmission efficiency and poor communication quality. By leveraging artificial intelligence, semantic communication has demonstrated improved performance in channels with low signal-to-noise ratios and restricted bandwidth, showing great potential for future underwater acoustic communication applications. This paper proposes an image semantic communication model based on Vision Mamba called VimSC, which utilizes Vision Mamba as the backbone network and incorporates OFDM extension to mitigate the multipath fading effects inherent in underwater acoustic channels. A channel state information (CSI) feedback precoding network is implemented to adjust the semantic symbols before transmission. Channel estimation and equalization are conducted at the receiver to facilitate accurate reconstruction of the received image. Experimental results demonstrate that VimSC significantly outperforms traditional coding and decoding methods in underwater acoustic communication, exhibiting better adaptability to adverse channel conditions and enhanced robustness. Compared to the DeepJSCC model, VimSC shows a notable improvement in the quality of image information recovery at the receiver.