Twin Fused Semantic Communication for Image Transmission With Equalization

Abstract

This letter investigates a dual-source semantic communication setting in which semantically correlated visual information is transmitted over heterogeneous channels and jointly decoded at the receiver. Built upon existing DeepJSCC architectures, this work focuses on improving joint semantic decoding under heterogeneous channel conditions by combining an efficient encoder design with receiver-side semantic equalization. An encoder design incorporating expanded bottleneck downsampling (EBD) is adopted to improve representation efficiency and reduce transmitter-side complexity, providing the primary performance gain. In addition, a receiver-side spatial–channel equalization (SCE) module is introduced to mitigate channel-imbalance effects and yield a consistent additional improvement during joint decoding. Experiments demonstrate improved PSNR and MS-SSIM under heterogeneous SNR conditions, validating the complementary roles of encoder efficiency and receiver-side equalization in robust joint semantic decoding.