独立成分分析在电磁攻击中的应用

Independent component analysis applied in electromagnetic attack

电磁泄漏曲线的对齐与有效点的选取是信息安全的重要研究方向.针对曲线过偏移的问题,提出了一种新的曲线对齐方法——双峰式相关对齐法.在有效抑制曲线过偏移的同时,实现了曲线的精确对齐通过独立成分分析（Independent Component Analysis,ICA）方法实现了有效点的选取.通过对电磁泄露曲线求得未知的源信号,由源信号作为特征点进行分类分析.分别采用ICA、主成分分析（Principal Components Analysis,PCA）、PCA-ICA、ICA-PCA四种方法对数据进行了降维处理.通过支持向量机（Support Vector Machine,SVM）对降维后的数据进行分类对比,最终得出：在10～100维范围内,PCA-ICA的分类效果最佳,ICA其次,而ICA-PCA的效果最差;在100～900维的范围内,PCA与ICA-PCA分类效果随着维度的增加几乎呈直线趋势增加.

Alignment and valid points selection of electromagnetic leakage curves are important re- search direction of information security. For the problem of curve excessive shifted, a new method of curve alignment is proposed, whose name is bimodal correlation alignment method. The method effectively con- trols the curve excessive shift, and also achieves a precise alignment result. It is proposed to use independ- ent component analysis （ICA） to select valid points. By the method of ICA, unknown source signals are obtained from electromagnetic leakage curves for classification analysis. ICA, principal components analy- sis（PCA）, ICA-PCA, and PCA-ICA are adopted for dimensionality reduction. The classification result is got by support vector machine （ SVM ） . It shows that success rate of ICA and PCA - ICA grows rapidly with the increase of the dimensions. PCA-ICA is the best, ICA is better, whice the worst is ICA-PCA, in the dimensionality 10 to 100. In the dimensionality 100 to 900, the success rate grows mostly like a line with the increase of the dimension.