自适应迭代步长与信道效应融合的调制识别对抗样本生成策略  

Modulation recognition adversarial samples generation strategy integrating adaptive iteration step size and channel effects

作　　者：杜习雨 龙吟 DU Xiyu;LONG Yin(School of Computer Science and Technology,Southwest University of Science and Technology,Mianyang 621010,China)

机构地区：[1]西南科技大学计算机科学与技术学院,四川绵阳621010

出　　处：《微电子学与计算机》2025年第4期70-77,共8页Microelectronics & Computer

基　　金：国家自然科学基金(62101467)。

摘　　要：在现代通信系统中,为了应对潜在的恶意窃听者或侦察者,对抗样本的出现提供了一种创新思路,以确保合法通信的安全传输。然而,现有研究在生成对抗样本时未充分考虑在传输过程中信道效应对其攻击效果的影响,同时在生成对抗样本的过程中采用固定的扰动步长,这限制了攻击策略的灵活性和对抗样本的攻击性。为了解决上述问题,提出了一种新的调制识别对抗样本生成方法。该方法充分考虑了信道效应,并引入了自适应迭代步长的机制。具体来说,考虑了对抗样本在传输时从发送端到窃听端受到的信道特性的影响,保证了对抗样本在传输过程中的稳定性。在此基础上,进一步提出一种自适应机制,解决迭代过程中不能根据模型敏感度动态调整扰动幅度的问题。实验结果表明,与现有方法相比,该方法在攻击成功率上有显著提高,验证了其在安全通信方面的实用价值。特别是在拟合真实通信环境时,该方法展现出了更强的鲁棒性和适应性。The emergence of adversarial samples in modern communication systems provides an innovative approach to ensure the secure transmission of legitimate communications,by addressing potential malicious eavesdroppers or scouts.However,existing research has not fully considered the influence of channel effects on attack effectiveness during the generation of adversarial examples.Additionally,a fixed perturbation step size is used in the process of generating adversarial examples,limiting the flexibility and aggressiveness of the attack strategy.To address these issues,a new method for generating adversarial examples for modulation recognition is proposed.This method fully considers channel effects and introduces an adaptive iteration step mechanism.Specifically,it takes into account the influence of channel characteristics on adversarial samples during transmission from sender to eavesdropper,ensuring stability during transmission.Furthermore,an adaptive mechanism is introduced to dynamically adjust perturbation amplitude according to model sensitivity during iterative processes.Experimental results demonstrate that compared with existing methods,this proposed approach significantly improves attack success rates and exhibits stronger robustness and adaptability when applied in real communication environments.

关 键 词：调制识别 对抗样本 信道 自适应机制 

分 类 号：TN91[电子电信—通信与信息系统]