针对无线语义保密的 VENENA：误导性视觉加密框架

窃听是长期以来威胁无线通信安全和隐私的挑战，因为 difficult to detect and costly to prevent。随着网络向第六代 (6G) 演进以及语义通信在下一代无线系统中日益 central，securing semantic information transmission 成为一个 critical challenge。虽然经典物理层安全 (PLS) focus on passive security，但最近提出的 physical layer deception (PLD) 概念 offer semantic encryption measure to actively deceive eavesdroppers。然而，既有 PLD 研究主要是 information-theoretical 和 link-level oriented，lack considerations of system-level design and practical implementation。在本 work 中，我们提出 Visual ENcryption for Eavesdropping NegAtion (VENENA)，一个人工智能驱动的框架 for secure image-based communication。VENENA 通过对其进行视觉编码来保护 confidential messages，同时主动误导窃听者：合法接收者使用人工智能 (AI)-based classifiers 提取 true message semantics，而拦截者 only perceive falsified content。该框架传输两个叠加的 image components with different power levels - 一个 high-power decoy image 和 一个 low-power correction mask，ensure only authorized receivers with favorable channel conditions can reconstruct the true message。实验验证表明，合法用户的准确率超过 93%，而窃听者的成功率仅为 52% 即使在 system design is fully known 的情况下，验证了 VENENA 为 6G semantic communication active defense capability。