Toward Universal Semantic Communication via Matchable Semantic Subspace Transmission.

Area of Science:

• Computer Science
• Artificial Intelligence
• Communication Engineering

Background:

• Existing communication methods struggle to balance task performance and bandwidth efficiency.
• Current systems often rely on fixed vocabularies, limiting their adaptability to new categories and open scenarios.

Purpose of the Study:

• To propose Universal Semantic Communication (UniSC), an open-vocabulary framework for efficient and robust semantic communication.
• To enable reliable task execution under stringent bandwidth and channel constraints with arbitrary semantic categories.

Main Methods:

• UniSC formulates transmission as a Matchable Semantic Subspace Transmission (MSST) problem, projecting images into a compact, robust semantic subspace.
• Key components include a Visual Semantic Engine (VSE), Semantic Squeeze Network (SSN), Noise-Adaptive Semantic Re-expansion (NASR), and a Vision-Language Model (VLM)-based Decoder.
• A two-stage training strategy optimizes cross-modal alignment and transmission robustness.

Main Results:

• UniSC demonstrates strong generalization capabilities for arbitrary text-defined semantic categories.
• The framework achieves state-of-the-art performance in semantic segmentation benchmarks, particularly under harsh channel conditions (low SNR, extreme compression).
• UniSC outperforms existing methods in both low-signal-to-noise ratio and extreme-compression scenarios.

Conclusions:

• UniSC offers a universal semantic communication solution that overcomes limitations of fixed vocabularies and bit-level reconstruction.
• The proposed MSST approach ensures cross-modal matchability and robustness against channel noise.
• UniSC represents a significant advancement in semantic communication, enabling efficient and adaptable task execution in diverse environments.