Security-aware Semantic-driven ISAC via Paired Adversarial Residual Networks

2025-09-25

AI Summary

To address the challenge of simultaneously ensuring sensing and communication performance while defending against eavesdropping attacks in Integrated Sensing and Communication (ISAC) systems, this paper proposes SS-ISAC—a secure semantic-driven ISAC framework. Methodologically, it introduces paired, plug-and-play trainable Adversarial Residual Networks (ARNs) as lightweight encryption/decryption modules, enabling end-to-end privacy protection without hardware modifications. A joint optimization loss function is further designed to balance adversarial robustness, communication rate, sensing accuracy, and privacy leakage risk. Simulation results demonstrate that SS-ISAC maintains baseline communication and radar performance—achieving bit error rate <10⁻³ and range resolution ≤0.5 m—while reducing the eavesdropper’s information extraction accuracy to near-random guessing (≈12.5%), thereby significantly enhancing physical-layer security.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Search and Optimization: Adversarial Search

Application Category

• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Assisted, interactive, and conversational search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

This paper proposes a novel and flexible security-aware semantic-driven integrated sensing and communication (ISAC) framework, namely security semantic ISAC (SS-ISAC). Inspired by the positive impact of the adversarial attack, a pair of pluggable encryption and decryption modules is designed in the proposed SS-ISAC framework. The encryption module is installed after the semantic transmitter, adopting a trainable adversarial residual network (ARN) to create the adversarial attack. Correspondingly, the decryption module before the semantic receiver utilizes another trainable ARN to mitigate the adversarial attack and noise. These two modules can be flexibly assembled considering the system security demands, without drastically modifying the hardware infrastructure. To ensure the sensing and communication (SAC) performance while preventing the eavesdropping threat, the above ARNs are jointly optimized by minimizing a carefully designed loss function that relates to the adversarial attack power, SAC performance, as well as the privacy leakage risk. Simulation results validate the effectiveness of the proposed SS-ISAC framework in terms of both SAC and eavesdropping prevention performance.

Problem

• Developing secure semantic-driven integrated sensing and communication framework
• Preventing eavesdropping threats while maintaining sensing and communication performance
• Using adversarial residual networks for flexible encryption and decryption modules

Innovation

• Paired adversarial residual networks for encryption
• Flexible pluggable modules for security adaptation
• Joint optimization balancing attack power and performance