Your WiFi Is Leaking: Ignoring Encryption, Using Histograms to Remotely Detect Skype Traffic

This paper presents a remote, undetectable, high accuracy mechanism to infer Skype voice traffic on WiFi networks with a success rate of ?97% and only a ?3% false positive rate. In spite of any encryption scheme employed, we infer user activity by exploiting a variety of frame size and interarrival time distributions. We demonstrate ways to use these efficiently and optimise the Random Forest classifier generated. The final product is an efficient classifier that we believe can be implemented at very low cost on portable, commodity hardware. Given its design and the side-channel data used, these methods should easily generalise to other encrypted communication methods such as 4G LTE. With longer range wireless communications becoming more prevalent, and increased commercial interest in tracking and analysing publicly broadcast wireless data, this paper highlights a plausible threat to users' private activities.