Abstract
This letter presents a conditional entropy-constrained multi-stage vector quantization (CEC-MSVQ) framework for semantic communication (SC). The proposed method integrates multi-stage VQ (MSVQ) for fine-grained rate control and entropy-constrained VQ (ECVQ) for improved rate–distortion efficiency. By modeling stage-wise quantization outputs as a Markovian sequence with conditional entropy modeling, CEC-MSVQ jointly trains semantic encoder–decoder networks and VQ codebooks while explicitly optimizing the rate–distortion objective. During inference, multi-rate transmission is supported by selectively activating VQ modules. Simulation results show that CEC-MSVQ achieves superior task performance over existing VQ-based SC, confirming the effectiveness of the proposed conditional entropy modeling.
Funding Information
  • a Grant of Korean ARPA-H Project through the Korea Health Industry Development Institute
  • Ministry of Health & Welfare, Republic of Korea (RS-2025-25454431)
  • National Research Foundation of Korea
  • Korean government (RS-2024-00453301)

This publication has 11 references indexed in Scilit: