Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Diffusion-enabled Secure Semantic Communication Against Eavesdropping

Boxiang He, Zihan Chen, Fanggang Wang, Shilian Wang, Zhijin Qin, Tony Q. S. Quek

TL;DR

The paper addresses semantic eavesdropping in DL-based semantic communication by introducing diffusion-enabled paired pluggable modules that embed artificial noise (AN) or adversarial perturbations into a forward diffusion process and recover detailed semantic information at the legitimate receiver through a diffusion-based reverse process. It analyzes two practical scenarios: unknown Eve knowledge (AGN-based encryption with a DDPM-based decryption and DRL-driven AN-power allocation) and known Eve knowledge (ARN-based encryption with DDPM decryption), achieving secure recovery without retraining the core transceiver. The key contributions include a triple-objective optimization framework balancing communication quality, privacy leakage, and covertness; a diffusion-model design for the forward and reverse processes; and extensive experiments on MNIST, CIFAR-10, and Fashion-MNIST showing near-zero leakage and high-quality semantic recovery across unknown and known Eve knowledge, with efficient DRL-based adaptation. The results indicate practical, scalable secure semantic communication with limited channel-input distortion and substantial resistance to eavesdropping in realistic wireless settings.

Abstract

In this paper, AN is introduced into semantic communication systems for the first time to prevent semantic eavesdropping. However, the introduction of AN also poses challenges for the legitimate receiver in extracting semantic information. Recently, denoising diffusion probabilistic models (DDPM) have demonstrated their powerful capabilities in generating multimedia content. Here, the paired pluggable modules are carefully designed using DDPM. Specifically, the pluggable encryption module generates AN and adds it to the output of the semantic transmitter, while the pluggable decryption module before semantic receiver uses DDPM to generate the detailed semantic information by removing both AN and the channel noise. In the scenario where the transmitter lacks eavesdropper's knowledge, the artificial Gaussian noise (AGN) is used as AN. We first model a power allocation optimization problem to determine the power of AGN, in which the objective is to minimize the weighted sum of data reconstruction error of legal link, the mutual information of illegal link, and the channel input distortion. Then, a deep reinforcement learning framework using deep deterministic policy gradient is proposed to solve the optimization problem. In the scenario where the transmitter is aware of the eavesdropper's knowledge, we propose an AN generation method based on adversarial residual networks (ARN). Unlike the previous scenario, the mutual information term in the objective function is replaced by the confidence of eavesdropper correctly retrieving private information. The adversarial residual network is then trained to minimize the modified objective function. The simulation results show that the diffusion-enabled pluggable encryption module prevents semantic eavesdropping while the pluggable decryption module achieves the high-quality semantic communication.

Diffusion-enabled Secure Semantic Communication Against Eavesdropping

TL;DR

The paper addresses semantic eavesdropping in DL-based semantic communication by introducing diffusion-enabled paired pluggable modules that embed artificial noise (AN) or adversarial perturbations into a forward diffusion process and recover detailed semantic information at the legitimate receiver through a diffusion-based reverse process. It analyzes two practical scenarios: unknown Eve knowledge (AGN-based encryption with a DDPM-based decryption and DRL-driven AN-power allocation) and known Eve knowledge (ARN-based encryption with DDPM decryption), achieving secure recovery without retraining the core transceiver. The key contributions include a triple-objective optimization framework balancing communication quality, privacy leakage, and covertness; a diffusion-model design for the forward and reverse processes; and extensive experiments on MNIST, CIFAR-10, and Fashion-MNIST showing near-zero leakage and high-quality semantic recovery across unknown and known Eve knowledge, with efficient DRL-based adaptation. The results indicate practical, scalable secure semantic communication with limited channel-input distortion and substantial resistance to eavesdropping in realistic wireless settings.

Abstract

In this paper, AN is introduced into semantic communication systems for the first time to prevent semantic eavesdropping. However, the introduction of AN also poses challenges for the legitimate receiver in extracting semantic information. Recently, denoising diffusion probabilistic models (DDPM) have demonstrated their powerful capabilities in generating multimedia content. Here, the paired pluggable modules are carefully designed using DDPM. Specifically, the pluggable encryption module generates AN and adds it to the output of the semantic transmitter, while the pluggable decryption module before semantic receiver uses DDPM to generate the detailed semantic information by removing both AN and the channel noise. In the scenario where the transmitter lacks eavesdropper's knowledge, the artificial Gaussian noise (AGN) is used as AN. We first model a power allocation optimization problem to determine the power of AGN, in which the objective is to minimize the weighted sum of data reconstruction error of legal link, the mutual information of illegal link, and the channel input distortion. Then, a deep reinforcement learning framework using deep deterministic policy gradient is proposed to solve the optimization problem. In the scenario where the transmitter is aware of the eavesdropper's knowledge, we propose an AN generation method based on adversarial residual networks (ARN). Unlike the previous scenario, the mutual information term in the objective function is replaced by the confidence of eavesdropper correctly retrieving private information. The adversarial residual network is then trained to minimize the modified objective function. The simulation results show that the diffusion-enabled pluggable encryption module prevents semantic eavesdropping while the pluggable decryption module achieves the high-quality semantic communication.
Paper Structure (16 sections, 1 theorem, 53 equations, 11 figures)

This paper contains 16 sections, 1 theorem, 53 equations, 11 figures.

Table of Contents

  1. Introduction
  2. General Semantic Communication System
  3. Security-aware Semantic Communication Framework via Paired Pluggable Modules
  4. Diffusion-enabled Secure Paired Pluggable Modules Design
  5. Problem Formulation
  6. Secure Paired Pluggable Modules Design for Alice having no Eve's knowledge
  7. Design of $f_{\text{Alice}}$
  8. Design of $f_{\text{Bob}}$
  9. Power allocation
  10. Secure Paired Pluggable Modules Design for Alice having Eve's knowledge
  11. Performance Analysis
  12. Experiment Setup
  13. Performance Evaluation for Unknown Eve's Knowledge
  14. Performance Evaluation for Known Eve's Knowledge
  15. Performance Evaluation On Multiple Datasets
  16. ...and 1 more sections

Key Result

Theorem 1

The upper bound of MI $I(\bm{x}_{\text{Alice}};\bm{y}_{\text{Eve}})$ can be given by

Figures (11)

  • Figure 1: Semantic communication system, where "JSC" denotes joint source channel.
  • Figure 2: Security-aware semantic communication framework via the paired pluggable modules, where the semantic transmitter consists of semantic encoder and joint source channel encoder, and the semantic receiver includes semantic decoder and joint source channel decoder. When Alice transmits semantically private information, the paired modules are inserted to prevent eavesdropping; conversely, when transmitting public information, the modules can be unplugged.
  • Figure 3: Pluggable paired modules for Alice having no Eve's knowledge, where (a) is the pluggable encryption module via AGN, and (b) is the pluggable decryption module via DDPM.
  • Figure 4: Training procedure for ARN in the case of Alice having Eve's knowledge. Specifically, the input data to DNN block consists of the output of the semantic transmitter. The stochastic gradient algorithm is used to minimize the loss function in problem \ref{['case2_SSC_problem_trans_replace_fur']} to obtain the optimal ARN parameters.
  • Figure 5: The communication MSE, the privacy leakage MI, and the covertness MSE are evaluated with respect to (w.r.t.) the SNR of Alice-Eve, where the SNR of Alice-Bob link is fixed at $5$ dB. The results reveal that the designed paired pluggable modules achieve excellent communication MSE while keeping the MI for privacy leakage close to zero.
  • ...and 6 more figures

Theorems & Definitions (4)

  • Remark 1
  • Remark 2
  • Theorem 1
  • Remark 3