From Lab to Crime Scene: Applying EM Side-Channel Analysis with Cross-Device Portability in Real-World Forensics
Year: 2025
Type: preprint
Abstract: Smart devices have become crucial in forensic investigations, often serving as key pieces of evidence or victims in criminal cases. With the increasing presence of smart devices at crime scenes, forensic analysis has become essential. While numerous tools and techniques exist for investigating smart devices, many require tampering or manipulation of these devices during analysis. To overcome this limitation, electromagnetic side-channel analysis (EM-SCA) combined with machine learning (ML) has been recently proposed as a non-invasive forensic technique. However, existing EM-SCA-based ML models are typically developed for individual devices, limiting their generalisability. To address this, cross-device portability of EM-SCA-based ML models has been applied within devices of the same manufacturer. However, this application has primarily been conducted in controlled environments, where researchers handle the devices directly and EM traces are collected using near-field antennas. In real-world scenarios, investigators may not have direct access to the devices, necessitating alternative data collection methods. This study evaluates the applicability of EM-SCA-based ML models in practical forensic investigations by collecting EM traces using a far-field antenna at various distances from the iPhone 13 and iPhone 14 Pro. Experimental results show that data collected using far-field antenna produce similar performance to near-field antennas. By applying fine-tuned feature extraction across different distances, the model achieves an accuracy of 99%. Despite these advancements, EM-SCA-based ML models are not yet widely adopted by forensic investigators, primarily due to the lack of accessibility to specialised equipment and knowledge. To bridge this gap, we propose a real-world forensic investigation framework that outlines how forensic experts can effectively apply cross-device portability of EM-SCA-based ML models in actual criminal cases. This study contributes to the field of digital forensics by providing a practical approach for utilising non-invasive side-channel analysis in real-life forensic investigations.
Source: SSRN Electronic Journal
DOI: 10.2139/ssrn.6144507
Institution: University of Vavuniya
Language: English
Topic: Digital and Cyber Forensics
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4413986935