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

2026-06-10

AI Summary

Traditional communication systems rely on generic distortion metrics that poorly capture human visual perception, often yielding blurry or unrealistic reconstructed images. This work proposes a joint source-channel-generative coding framework that, for the first time, integrates a generative model into the wireless communication decoder. By conditioning the generative sampling process on the received signal and operating in a latent space, the method abandons pixel-level reconstruction in favor of optimizing mutual information under perceptual constraints, thereby achieving semantic-level fidelity. Theoretical analysis confirms the validity of the learning and inference procedures, while experiments demonstrate that the proposed approach substantially improves reconstruction quality at the feature, semantic, and distributional levels across diverse channel conditions. Notably, reconstruction errors shift from perceptual distortion to controllable semantic inconsistencies.

Abstract

Conventional communication systems, including both separation-based coding and learning-based joint source-channel coding (JSCC), are typically designed under Shannon's rate-distortion theory. However, relying on generic distortion metrics fails to capture complex human visual perception, often resulting in blurred or unrealistic reconstructions. In this paper, we propose Joint Source-Channel-Generation Coding (JSCGC), a generative communication paradigm that replaces the conventional decoder with a generative model at the receiver. The received signal is treated as a condition that controls the sampling process into the learned conditional distribution, reformulating communication from deterministic reconstruction for distortion minimization to controlled generation for mutual information maximization under perceptual constraints. Based on this formulation, we develop a unified joint training and efficient stochastic sampling framework, and provide theoretical analysis of its effectiveness in both learning and inference stages. Extensive experiments on latent-space image transmission demonstrate that the JSCGC consistently improves feature-based, semantic-level, and distributional quality across diverse channel conditions, while exhibiting a distinct error behavior characterized by semantic inconsistency rather than distortion.

Problem

• generative communication
• perceptual quality
• joint source-channel coding
• semantic inconsistency
• distortion metrics

Innovation

• Joint Source-Channel-Generation Coding
• Generative Communication
• Perceptual Quality
• Conditional Generation
• Mutual Information Maximization