Topology-Preserving Deep Joint Source-Channel Coding for Semantic Communication

A team of researchers has introduced TopoJSCC, a novel AI framework designed to revolutionize semantic communication for wireless vision applications. The system addresses a critical flaw in existing Deep Joint Source-Channel Coding (DeepJSCC) schemes, which primarily optimize for pixel-by-pixel accuracy but fail to protect the global structural relationships and connectivity within an image—information vital for tasks like autonomous navigation. TopoJSCC integrates concepts from topological data analysis, specifically persistent homology, directly into its end-to-end neural network training. It enforces consistency by mathematically penalizing differences in the 'cubical persistence diagrams' of original and reconstructed images, and the 'Vietoris–Rips persistence' of latent features before and after transmission through a noisy channel.

This topology-aware approach requires no additional side information and focuses on preserving the essential shape and structure of data, not just individual pixels. In practical experiments, TopoJSCC demonstrated superior performance in challenging low signal-to-noise ratio (SNR) and constrained bandwidth environments, showing measurable improvements in both topological preservation metrics and traditional peak signal-to-noise ratio (PSNR). The work, detailed in the arXiv preprint 2603.17126, represents a significant shift from pixel-fidelity to semantic-fidelity in AI-driven communication, aiming to create more robust and reliable data pipelines for real-time, mission-critical systems.