Improving Image Transmission with VAE-based Semantic Communication Using HQAM

Abstract

Semantic communication subverts classical communication techniques by changing the focus on the preservation of meaning rather than on correctly transmitting logical bits. This is achieved by implementing a semantic encoder/decoder pair that operates at a higher level of abstraction and takes into account the nature of the source and destination, as well as their shared knowledge. One approach for the implementation of semantic encoder/decoder pairs is the use of artificial neural networks, and in particular, variational autoencoders. In this work, we propose the use of hierarchical quadrature amplitude modulation to improve the perceived quality of data in a semantic communication link using a variational autoencoder. We show that the use of the hierarchical modulation consistently improves the performance of the semantic communication link in terms of the SSIM under noisy channel scenarios, even achieving the overcoming of noise for signal-to-noise ratio values above 10 dB.