Multiband Pixel Colour Classification from HDMI Emissions

Abstract

We demonstrate methods to enhance the reconstruction of displayed information from the compromising emanations of HDMI or DVI video cables. Using a software-defined radio receiver, we acquire multiple recordings of such emissions for the same displayed image, at adjacent, overlapping reception bands. We first perform frequency alignment and coherent periodic averaging on each of these recordings individually. We then mutually align the resulting frames such that we can extract colour-identifying features for each displayed pixel across multiple reception bands. These features then go into a clustering algorithm to classify the signals emitted by different TMDS symbols. Finally, we build a graph data structure of the most common transitions between such symbols, and identify loops in this graph as candidates of pixel colours that cycle through multiple symbols due to the DC-balancing algorithm applied by the TMDS encoding. This can enhance the readability of eavesdropped text with some colour combinations, as we demonstrate for signals recorded at 12 metres distance.