Gaining Forensic Insights through Electromagnetic Emanations: A Methodology and Detailed Analysis

Abstract The forensic analysis of smart devices plays an important role in prosecutions around the world. However, digital forensic investigators continue to face challenges in devising effective techniques to access data stored on smart devices. Electromagnetic side-channel analysis (EM-SCA) offers a promising avenue for developing tools to investigate these devices. However, existing EM-SCA-based Machine Learning (ML) models are tightly coupled to the specific processors on whose data they are trained, requiring the cross-device portability. The direct application of pre-existing EM-SCA-based ML models on new devices yields poor results — making cross-device model portability an unproductive endeavour. This research focuses on inductive, feature extraction, and fine-tuned transfer learning techniques to enhance results and develop more generalised EM-SCA-based ML models. The results demonstrate a significant improvement during the transfer learning process, reaching up to 99%, thus validating the generalisability of the methodology and its potential to serve as a common technique for efficiently investigating smart devices using electromagnetic side channels.