Research On Power Analysis Methods Based On Ridge Regression And Data Rearrangement

Side channel analysis allows cryptographic algorithms to be analyzed by side information such as power consumption and time during the operations of the cryptographic device,which has the advantages of high attack efficiency and low implementation cost,etc.Among them,the most common method is power analysis.Traditional methods have many disadvantages,such as the low correlation between the leakage model and power traces,the large number of power traces,etc.In recent years,machine learning techniques have been introduced into power analysis to enhance the effectiveness of leakage models and optimize the feature extraction,etc.At the same time,various data pre-processing methods have been widely used to reduce data noise and improve attack efficiency.In this thesis,regression techniques are used to optimize leakage models to improve the accuracy of power analysis,and a data rearrangement method based on correlation ranking is proposed for data pre-processing,to improve the efficiency of power analysis.The specific work is as follows.1.A power analysis method based on mean-ridge regression is proposed.Ridge regression is introduced to construct a leakage model with a better fit.Firstly,the intermediate values are filtered in combination with the degree of the model to reduce the scale of the data.Then,the weight coefficients and optimal ridge parameters are calculated to construct a dynamic initial model.Finally,the output of the model is averaged to reduce the noise and obtain a leakage model based on mean-ridge regression.For PRESENT,the analysis results show that the mean-ridge regression-based power analysis expands the correlation coefficient corresponding to the correct key and reduces the number of power traces needed for analysis in comparison with the existing pre-processing methods.2.A data rearrangement method based on correlation ranking is proposed.Firstly,the mean power trace is calculated and used as the "benchmark".After that,the correlation between each power trace and the "benchmark" is quantified.Finally,the data is re-ranked in descending order of the correlation.The results of the power analysis based on the rearranged power traces show that,for PRESENT,the data rearrangement improves the accuracy of the power analysis based on mean-ridge regression,which is better than the existing pre-processing methods.For SPEEDY,the data rearrangement-based power analysis significantly reduces the number of power traces required for the analysis and provides new results for the safety evaluation of SPEEDY.3.The second-order power analysis based on data rearrangement is designed and implemented.Firstly,the input of the first-order power analysis of AES is preprocessed based on data rearrangement.And the comparison experiment and superposition experiment are carried out between date rearrangement and existing pre-processing methods.Then,the first-order protection of AES is executed by performing exclusive OR between the intermediate values and random masks.Finally,the data rearrangement is applied to the second-order power analysis.The experimental results show that the first-order power analysis based on data rearrangement has a more obvious improvement in the accuracy and other indicators,and the superposition with other pre-processing methods also has an optimization effect.The method reduces the number of power traces of the analysis when combined with the second-order power analysis,verifying that it is still effective when applied to the high-order power analysis.