Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Privacy-Preserving Semantic Communication over Wiretap Channels with Learnable Differential Privacy

Weixuan Chen, Qianqian Yang, Shuo Shao, Shunpu Tang, Zhiguo Shi, Shui Yu

TL;DR

This work tackles privacy in semantic communication over wiretap channels by introducing a learnable differential privacy (DP) mechanism that perturbs private semantic latent codes. It leverages GAN inversion (Semantic StyleGAN) to obtain disentangled semantic representations and employs NN-based DP protection/deprotection modules guided by adversarial training to resemble genuine DP noise while remaining invertible for the legitimate receiver. By pre-sharing private indices and enabling tunable privacy budgets, the approach achieves explicit security control without key exchange, significantly degrading Eve's reconstruction while preserving Bob's task performance. Experimental results on CelebAMask-HQ show superior LPIPS and FPPSR outcomes compared with both direct transmission and traditional DP protection across varying budgets and channel conditions, demonstrating robust, practical security for SemCom in comparable-SNR wiretap settings.

Abstract

While semantic communication (SemCom) improves transmission efficiency by focusing on task-relevant information, it also raises critical privacy concerns. Many existing secure SemCom approaches rely on restrictive or impractical assumptions, such as favorable channel conditions for the legitimate user or prior knowledge of the eavesdropper's model. To address these limitations, this paper proposes a novel secure SemCom framework for image transmission over wiretap channels, leveraging differential privacy (DP) to provide approximate privacy guarantees. Specifically, our approach first extracts disentangled semantic representations from source images using generative adversarial network (GAN) inversion method, and then selectively perturbs private semantic representations with approximate DP noise. Distinct from conventional DP-based protection methods, we introduce DP noise with learnable pattern, instead of traditional white Gaussian or Laplace noise, achieved through adversarial training of neural networks (NNs). This design mitigates the inherent non-invertibility of DP while effectively protecting private information. Moreover, it enables explicitly controllable security levels by adjusting the privacy budget according to specific security requirements, which is not achieved in most existing secure SemCom approaches. Experimental results demonstrate that, compared with the previous DP-based method and direct transmission, the proposed method significantly degrades the reconstruction quality for the eavesdropper, while introducing only slight degradation in task performance. Under comparable security levels, our approach achieves an LPIPS advantage of 0.06-0.29 and an FPPSR advantage of 0.10-0.86 for the legitimate user compared with the previous DP-based method.

Privacy-Preserving Semantic Communication over Wiretap Channels with Learnable Differential Privacy

TL;DR

This work tackles privacy in semantic communication over wiretap channels by introducing a learnable differential privacy (DP) mechanism that perturbs private semantic latent codes. It leverages GAN inversion (Semantic StyleGAN) to obtain disentangled semantic representations and employs NN-based DP protection/deprotection modules guided by adversarial training to resemble genuine DP noise while remaining invertible for the legitimate receiver. By pre-sharing private indices and enabling tunable privacy budgets, the approach achieves explicit security control without key exchange, significantly degrading Eve's reconstruction while preserving Bob's task performance. Experimental results on CelebAMask-HQ show superior LPIPS and FPPSR outcomes compared with both direct transmission and traditional DP protection across varying budgets and channel conditions, demonstrating robust, practical security for SemCom in comparable-SNR wiretap settings.

Abstract

While semantic communication (SemCom) improves transmission efficiency by focusing on task-relevant information, it also raises critical privacy concerns. Many existing secure SemCom approaches rely on restrictive or impractical assumptions, such as favorable channel conditions for the legitimate user or prior knowledge of the eavesdropper's model. To address these limitations, this paper proposes a novel secure SemCom framework for image transmission over wiretap channels, leveraging differential privacy (DP) to provide approximate privacy guarantees. Specifically, our approach first extracts disentangled semantic representations from source images using generative adversarial network (GAN) inversion method, and then selectively perturbs private semantic representations with approximate DP noise. Distinct from conventional DP-based protection methods, we introduce DP noise with learnable pattern, instead of traditional white Gaussian or Laplace noise, achieved through adversarial training of neural networks (NNs). This design mitigates the inherent non-invertibility of DP while effectively protecting private information. Moreover, it enables explicitly controllable security levels by adjusting the privacy budget according to specific security requirements, which is not achieved in most existing secure SemCom approaches. Experimental results demonstrate that, compared with the previous DP-based method and direct transmission, the proposed method significantly degrades the reconstruction quality for the eavesdropper, while introducing only slight degradation in task performance. Under comparable security levels, our approach achieves an LPIPS advantage of 0.06-0.29 and an FPPSR advantage of 0.10-0.86 for the legitimate user compared with the previous DP-based method.
Paper Structure (39 sections, 25 equations, 16 figures)

This paper contains 39 sections, 25 equations, 16 figures.

Table of Contents

  1. Introduction
  2. Backgrounds
  3. Related Works and Motivation
  4. Adversarial training-based methods
  5. Encryption-based methods
  6. Physical layer-based methods
  7. Contributions
  8. System Model
  9. Proposed Method
  10. System Overview
  11. System Overview
  12. Transmitter
  13. Bob's Receiver
  14. Eve's Receiver
  15. Overall Architecture
  16. ...and 24 more sections

Figures (16)

  • Figure 1: The framework of our proposed secure SemCom system.
  • Figure 2: The framework of the Semantic StyleGAN generator.
  • Figure 3: Illustration of the adversarial training process for the DP protection module.
  • Figure 4: The network architecture of the proposed NN-based DP protection module.
  • Figure 5: The LPIPS performance of the Proposed System (Basic Eavesdropper) under different privacy budgets $\epsilon$, where the channel SNR is set to 20 dB.
  • ...and 11 more figures