Mobile Traffic Anonymization Through Probabilistic Distribution

Current implementations of mobile apps offer limited security assurances against traffic analysis. Encryption is not effective in hiding particular patterns within packets that can be used as side-channel information to classify specific apps. What we need is an anonymity system that ensures strong security with acceptable computational overhead and latency for interactive app usage. Recently, we undertook this problem by mutating an app traffic that we try to defend so that it resembles the traffic of another app. Our goal in this paper is to develop a simpler and more scalable system to anonymize mobile app packet traffic without the need of another app's model traffic. This could happen using probabilistic distribution of packet sizes. We propose first a scheme that regenerates statistical modeling of app packet lengths. Then, we present a privacy preserving technique that implements defense against traffic analysis and confines the incurred overhead by mutation of packet lengths of the incoming traffic to the regenerated ones. Experiments show that using this confusion technique, we are able to reduce 91.1% classification accuracy to 0.9% with only 12.51% overhead. On the other hand, we demonstrate a second technique for anonymization where we thwart traffic analysis of mobile app traffic by modifying its packet sizes probability distribution to another dissimilar distribution.