基于PCA和SVM的电磁模板分析攻击

Electromagnetic Template Analysis with PCA and SVM

在分析密码芯片电磁辐射数据相关性的基础上,提出了一种基于主成分分析(PCA)技术和多分类支持向量机(SVM)的模板分析密码旁路攻击方法。将密码设备运行时采集到的泄漏的电磁信号经过PCA处理之后作为特征向量,其对应的密钥作为类别,用已知密钥情况下获得的样本训练多分类SVM,用训练好的多分类SVM对未知密钥的电磁信号进行分类,并根据分类结果推测密钥值。实验表明,在用相同多个主成分和训练样本的条件下,SVM的分类效果好于大多数文献上使用的Bayes判别的分类效果。

Based on the introduction of the relationship between data being operated in a cipher device and the electromagnetic （EM） emission from it, a novel side channel crypto-analysis, electromagnetic template analysis with principle component analysis （PCA） and support vector machine （SVM） is proposed. In this method, eigenvectors were firstly extracted with PCA from EM signals captured while cipher device was executing, and then were used to train a multi-classify SVM combined with known secret keys as classes label, after that a sampie of EM signal with unknown secret key was classified with the trained multi-classify SVM, finally the secret key was deduced with the re suit of the SVM. It was confirmed with experiments that with the same number of principle components and training samples, the correct rate of SVM was higher than Bayes distinguish method which was widely used in published literatures.