JSCGC: Joint Source-Channel-Generation Coding for Wireless Generative Communications
Researchers have introduced Joint Source-Channel-Generation Coding (JSCGC), a novel communication paradigm that leverages generative models at the receiver. This approach moves beyond traditional rate-distortion theory by reformulating communication as controlled generation, aiming to maximize mutual information under perceptual constraints rather than solely minimizing distortion. Experiments show JSCGC enhances feature-based, semantic, and distributional quality in image transmission, exhibiting semantic inconsistency as a distinct error characteristic. AI
IMPACT Introduces a new generative communication paradigm that could improve data transmission quality and efficiency.