Adaptive Dual-Path Framework for Covert Semantic Communication
A recent preprint on arXiv (2605.03423) has unveiled an innovative adaptive dual-path system aimed at covert semantic communication (SemCom). Unlike conventional methods that hide messages using power-domain signal superposition, this new framework cleverly embeds hidden information within specific task features using intrinsic semantic encoding. It consists of two encoding paths: an Explicit path for general tasks and a Stego path that blends public and hidden data through contrastive representation alignment. Moreover, it includes a Gumbel-Softmax mechanism for selecting paths adaptively, ensuring network blocks function based on the task's requirements. This framework also incorporates multi-objective optimization to maintain accurate semantic understanding and reliable covert transmission.
Key facts
- Proposes adaptive dual-path framework for covert semantic communication
- Embeds covert data within task-specific features via semantic-level intrinsic encoding
- Dual encoding paths: Explicit path for public tasks, Stego path for joint encoding
- Uses contrastive representation alignment for Stego path
- Gumbel-Softmax mechanism for adaptive path selection
- Multi-objective optimization for semantic understanding and covert transmission
- Published on arXiv with ID 2605.03423
- Differs from conventional power-domain signal superposition methods
Entities
Institutions
- arXiv