Detecting Corruptive Noise Rounds for Statistical Disclosure Attacks

The Statistical Disclosure Attack (SDA) is an efficient de-anonymization technique to measure the vulnerability of anonymous communication systems. The attack is presented as a signal detection problem, aiming to distinguish the signal (i.e., communication partners of the targeted user) from noise (i.e., other users in the system). To better filter out background noise, several noise estimation methods have been proposed in previous studies that utilize statistics from rounds in which the targeted user (Alice) does not send any message (called noise rounds). Since these rounds are independent from Alice, including some of them in the SDA calculations distracts the current noise estimation from the background noise present in Alice's signal. We call these rounds as corruptive noise rounds. Including corruptive noise rounds in the attack calculations reduces the accuracy of the attack, and regaining the previous accuracy may require additional observations. This issue leads to incorrect vulnerability measurement of systems. In this paper, we propose a method to detect and exclude corruptive noise rounds from attack calculations. Our results show that the communication partners of Alice are identified in earlier observations (i.e., shorter time) than the previous approaches, following the proposed corruptive noise round exclusion method.