MVSC: Mamba Vision Based Semantic Communication for Image Transmission With SNR Estimation

Abstract

This letter proposes a novel semantic communication approach named Mamba Vision-based Semantic Communication (MVSC) for image transmission with integrated Signal-to-Noise Ratio (SNR) estimation. Unlike prior works that assume the SNR of the received signal is known and input a predetermined SNR value into a deep learning (DL) network, MVSC introduces an implicit SNR estimation module, allowing the network to infer channel conditions for SNR adaptation. To further improve performance, we propose the MVSC4, a joint-optimized of MVSC, which is trained using a multi-task learning strategy that simultaneously optimizes image reconstruction, SNR estimation, signal denoising, and image classification. This joint optimization enhances the network’s robustness to varying SNR conditions, particularly in low-SNR environments. Comparative experiments on CIFAR-10 and Kodak datasets demonstrate that MVSC4 outperforms both CNN-based and Transformer-based methods in terms of Peak Signal-to-Noise Ratio (PSNR) and Multiscale Structural Similarity (MS-SSIM). The results demonstrate the effectiveness and robustness of the proposed approach.