Power analysis attack using neural networks with wavelet transform as pre-processor

This work proposes a novel methodology to perform power analysis attack on secure system by using wavelet transform as a pre-processor followed by machine learning technique. The proposed methodology uses known plain text attack. The power supply current traces from the cryptographic device are obtained by varying the atmospheric temperature. Then the current traces are pre-processed by using wavelet transform, data normalization and principal component analysis (PCA). The featured data samples selected by the pre-processor are then used to train the neural network. Through supervised learning algorithm and wavelet pre-processing, we are able to achieve around 25% improvement in guessing the secret key when compared to existing method of machine learning alone.