Acoustic Side Channel Attack for Device Identification using Deep Learning Models

Abstract

Side-channel attacks are easy to execute and very hard to detect because of their passive nature. If the side channels can be used to decode the device operation, the same can be used to identify the physical property difference of the devices. We investigated the possibility of differentiating the devices based on acoustic side-channel attacks. The Mel Frequency Cepstral Coefficients (MFCC) acoustic features are extracted from the audio samples recorded from different computing devices including embedded modules, laptops, and PCs. A comparative analysis of classification accuracy for the various machine learning algorithms in terms of Precision and Recall is also presented. Our results show that CNN and LSTM give the desired results with better accuracy among the different classification models considered.