A Hierarchical Approach for Multiple Periodicity Detection in Software Code Analysis

This paper introduces an end-to-end processing method for multiple periodicity signal detection and analysis with particular application in software analysis using analog side channels. The probabilistic distributions of signal blocks are estimated with kernel density estimation. The corresponding kernel bandwidths, which are optimally found in a data-driven manner, are used to detect change points. After separating the signal into parts with different behaviors, average magnitude difference function is leveraged iteratively to find the smallest periodic signal sections. To illustrate efficiency of the proposed method, we use EM side-channel signals collected from real-life applications to successfully detect multiple existing periodicities.