Nonlinear transform coding for semantic communications

Abstract

The modular design and limited processing mechanism of traditional communication systems limit the continuous improvement of end-to-end data transmission capability.For this reason, a new nonlinear transform coding framework for semantic communications was proposed.First, an end-to-end rate distortion optimization criterion for semantic communication was derived based on variational theory.Based on this, a nonlinear transform was designed to extract the compact representation of source data in the semantic latent space, and variable-rate nonlinear joint source-channel coding was implemented through the guidance of variational entropy model.Experiments show that semantic nonlinear transform coding can significantly improve the end-to-end data transmission performance and robustness, and is one of the key technologies to catalyze future semantic communications.

Keywords

• semantic communications;nonlinear transform;nonlinear coding;variational entropy modeling;rate-distortion optimization