基于回声状态网络的功耗曲线特征提取

ESN-based power trace feature extraction

在模板(Template Attacks,TA)攻击的研究中,如何利用功耗曲线信息,合理选择有效点,增强匹配效果是改进模板攻击的一个重要方向.文中分析了目前有关功耗曲线主要特征提取方法的优缺点,并提出了一种基于回声状态网络(Echo State Network,ESN)的功耗曲线特征提取方法.该方法针对ESN分类方法中的储备池参数选择问题,以时间预测序列精度为标准,采用网格法进行参数空间的优化搜索,并利用神经网络以数据样本形式作为定量知识自行处理的能力,对粗略对齐下的功耗曲线的特征提取能力进行了测试和评估.实验结果表明,基于ESN功耗曲线特征提取方法在功耗曲线数量相同条件下,通过合理选择内核参数,能够降低模板攻击对功耗曲线预处理技术的依赖,提高正确密钥的分类精度.

In the study of template attacks （TA）, the method of choosing valid point from power traces and improvement of the template attack becomes an important direc- tion. This paper analyzes the advantages and disadvantages of the current power trace feature extraction methods. Meanwhile, it presents a new power trace feature extraction which is based on echo state net-works（ESN）. In order to better choose the reservoir parameters in the echo state network classification process, a grid method is used to op-timize the search of the parameter space, with the precision of time series prediction as the standard in this paper. Since a neural network can use data samples as quantitative knowledge to conduct the automatic process, the feature extraction capability for power trace roughly aligned is tested and evaluated. The experiment result shows that, with the same amount of power traces, when the core parameters are appropriately chosen, ESN-based power trace feature extraction can reduce the dependence on pretreatment technologies in template attacks, thus increase the precision of classification of the key.