JSCGC: Joint Source-Channel-Generation Coding for Wireless Generative Communications

2026-06-12 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Wireless Communication Systems, Information Theory · Depth: Expert, extended

Summary

Joint Source-Channel-Generation Coding (JSCGC) introduces a novel generative communication paradigm, replacing traditional decoders with generative models at the receiver. This system treats the received signal as a condition to guide sampling from a learned conditional distribution, shifting the objective from distortion minimization to controlled generation for mutual information maximization under perceptual constraints. JSCGC employs a unified joint training and efficient stochastic sampling framework, implemented for latent-space image transmission using a Mamba-based encoder, a latent flow matching model (Z-Image), and a communication-aware adapter. Experiments on the Kodak dataset demonstrate that JSCGC consistently improves feature-based, semantic-level, and distributional quality across various channel conditions. For instance, under AWGN at 5 dB SNR, it reduces LPIPS and FID to 79.42% and 53.68% of DiffJSCC baselines, respectively, while boosting the CLIP score by 11%. Notably, its error behavior manifests as semantic inconsistency rather than perceptual distortion.

Key takeaway

For AI Scientists and Machine Learning Engineers developing wireless communication systems, consider adopting the Joint Source-Channel-Generation Coding (JSCGC) paradigm. This approach moves beyond traditional distortion-minimizing reconstruction, offering superior perceptual quality and semantic consistency, especially in low SNR environments. You should prioritize end-to-end joint optimization of encoding and generation, and evaluate systems using perceptual and semantic metrics rather than pixel-level distortion. This shift enables systems that produce realistic outputs even when semantic fidelity to the source is reduced.

Key insights

Generative communication shifts from distortion-based reconstruction to controlled content generation guided by received signals.

Principles

• Maximize mutual information under perceptual constraints, not explicit distortion.
• Jointly optimize encoding, transmission, and generative decoding end-to-end.
• Communication errors shift from perceptual distortion to semantic inconsistency.

Method

JSCGC uses a unified joint training framework for encoder and generative model, optimizing a variational objective. It employs an efficient conditional sampling strategy via probability flow ODEs.

In practice

• Implement latent-space image transmission with Mamba-based encoders.
• Use communication-aware adapters to inject signals into generative models.
• Prioritize perceptual metrics (LPIPS, CLIP Score) over PSNR for generative systems.

Topics

• Generative Communications
• Joint Source-Channel Coding
• Diffusion Models
• Image Transmission
• Perceptual Quality
• Wireless Communication Systems

Best for: Computer Vision Engineer, AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• JSCGC: Joint Source-Channel-Generation Coding for Wireless Generative Communications — JSCGC redefines communication from distortion minimization to controlled generation for mutual information maximization.
• Dual-Constrained Diffusion Image Compression for Operational Rate-Distortion-Perception Optimization — DCIC navigates the full RDP surface by integrating diffusion with dual distortion and idempotence constraints for flexible image compression.
• UniVL: Unified Vision-Language Embedding for Spatially Grounded Contextual Image Generation — UniVL unifies spatial and semantic image generation conditioning by rendering text into masks, eliminating separate text encoders.
• Generative AI improves a wireless vision system that sees through obstructions — Generative AI significantly improves wireless vision systems for accurate 3D object and indoor scene reconstruction through obstructions.
• Where's the raccoon with the ham radio? (ChatGPT Images 2.0) — ChatGPT Images 2.0 significantly advances image generation, particularly in complex scene composition and accurate text rendering.
• Adapting Diffusion Language Models for Lossless Pixel-Level Image Transmission — DDM-SSCC adapts diffusion language models for lossless pixel-level image transmission, achieving superior exact recovery via novel coding and denoising strategies.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.