Enhancing Image Transmission Efficiency: Semantic Communications with Variational Autoencoders

This paper investigates a Variational Autoencoder (VAE)-based end-to-end semantic communication framework tailored for image transmission. The proposed network is rigorously evaluated on the MNIST dataset under various channel conditions, including Rayleigh, Rician, and Nakagami-m fading channels. Performance metrics such as Peak Signal-to-Noise Ratio (PSNR) and Learned Perceptual Image Patch Similarity (LPIPS) are employed to assess the system's ability to preserve both pixel-level fidelity and perceptual quality. Additionally, the impact of the VAE's latent dimensionality on communication efficacy is studied. Finally, a comprehensive assessment of the system's computational complexity and latency is presented.