Integrating Advanced Signal Analysis and Deep Learning in TEMPEST Techniques

This work explores electromagnetic (EM) leakage from transition-minimized differential signal (TMDS)-based external interfaces, focusing on information leakage from video display units (VDUs) due to common mode noise caused by asynchronous differential signal. This work addresses a methodology to improve these EM leakage signals via deep learning. In this process, we provide a solution for addressing data diversity issues that are directly related to the performance of deep learning models, and suggest that applications can be extended to ambient noise and stripe noise cancellation, etc. Consequently, this work extends the scenario of side-channel attack and provides insights for the development of defense strategies.