Privacy-Preserving Framework: Joint Sensitive Semantic Partitioning and Channel Estimation-Based Channel Selection

Abstract

Ensuring the security of users’ sensitive semantic information has become a critical challenge in modern semantic communication systems. Identifying highly sensitive semantic content and selecting secure transmission environments significantly enhances overall network robustness. To address this issue, this study proposes a privacy-preserving semantic communication framework based on sensitive-semantic partitioning and channel estimation. The framework first employs a Quantum Approximate Optimization Algorithm (QAOA) based semantic partitioning method at the transmitter to extract and categorize highly sensitive semantic units. For legitimate users, a Correlation based Semantic Channel Estimation (CSCE) scheme is then applied, utilizing antenna space correlation, while an Eavesdropper Channel Simulation Generative Adversarial Network (ECS-GAN) predicts potential eavesdropping channels. Finally, a fuzzy logic controller combines the sensitivity of selected semantic units with the secure channel capacity derived from channel estimation, producing a security membership score that determines the safe transmission channel. This process ensures privacy protection within semantic communication systems.