Reproduce “MOC-RVQ: Multilevel Codebook-Assisted Digital Generative Semantic Communication” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yingbin Zhou, Yaping Sun, Guanying Chen, Xiaodong Xu, Hao Chen, Binhong Huang, Shuguang Cui, Ping Zhang
The MOC-RVQ system introduces a multilevel codebook-assisted digital generative semantic communication framework that resolves the incompatibility between vector quantization and digital modulation, enhancing robustness against channel noise. It achieves performance comparable to analog JSCC with one-sixth the bandwidth, outperforming traditional codecs in noisy conditions.
Key insights
- Designing VQ codebooks with an 8-state per-head structure allows direct compatibility with standard digital modulation (e.g., 64-QAM) while multilevel residual quantization maintains high feature representation quality.
- Heuristic reordering of codebooks (CR), especially with Gray code mapping, minimizes the semantic distance between code vectors assigned to numerically adjacent indices, improving robustness against channel errors.
- A dedicated Noise Reduction Block (NRB) combined with feature requantization effectively restores noisy semantic features by leveraging a pre-trained high-quality semantic knowledge base, even under adverse channel conditions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.