Controllable Steganography for Robust and Efficient Semantic Communication Systems

As a significant direction in next-generation communication technology, semantic communication systems aim to achieve efficient transmission of semantic information. In high-threat environments, protecting the concealment of semantic information is crucial, as traditional encryption methods can easily attract the attention of attackers. Steganography, as an effective concealment technique, faces challenges in coding efficiency that hinder its applicability in semantic communication systems. To address these issues, this article proposes a controllable steganography method suitable for semantic communication systems, aiming to balance concealment, efficiency, and resource constraints. First, a steganography framework based on redundancy- free semantic representation is constructed. By directly embedding secret semantics within cover semantics of the same dimension, bandwidth utilization is enhanced. Second, a controllable encoding strategy driven by a conditional semantic generator is designed, utilizing dynamic sampling to generate diverse cover semantics and enhance resistance to reverse inference attacks. Next, a lightweight reversible semantic mapping component is developed to achieve high-fidelity and low-latency mapping between secret semantics and steganographic semantics, meeting the requirements of low-power devices. Finally, a comprehensive security evaluation system covering resistance to detection attacks and reverse inference attacks is established, validating the superior concealment and security of the proposed method in various threat environments.