RAMO Lab to Present at ICASSP 2026 in Barcelona

[January 22, 2026]

The paper "Hybrid Semantic-Complementary Transmission for High-Fidelity Image Reconstruction," authored by Hyelin Nam, Jihong Park, Jinho Choi, and Seong-Lyun Kim, has been accepted for presentation at ICASSP 2026 in Barcelona, Spain. This work has also been extended for a journal submission to IEEE Transactions on Cognitive Communications and Networking (TCCN).

Research Overview: Hybrid Semantic Communication (HSC)

Traditional Semantic Communication (SC) utilizes neural networks to transmit semantic representations (SR), but often fails to reconstruct fine, image-specific details because the models are trained on broad distributions. To address this, RAMO Lab proposes Hybrid Semantic Communication (HSC).

• Complementary Representation (CR): HSC supplements SR with a flexible "Complementary Representation" (CR). While the SR load is fixed by the neural network structure, the CR load can be adjusted to capture residual image-specific details.

• High-Fidelity Reconstruction: By deriving an optimal closed-form expression for the CR, the framework allows for precise control over reconstruction fidelity.

• Performance: Simulation results on the MNIST dataset show that HSC substantially enhances fidelity compared to baseline SC. As shown in the lab's findings, adding CR reduces error much more effectively than increasing SR alone, reaching an MSE of zero when the rank reaches the image dimension.

The results demonstrate that as the CR rank increases, reconstructed images progressively recover finer details that are typically lost in pure semantic transmission.