This project conducts a dialectic study of location privacy in wireless networks -- a ``good cop'' vs ``bad cop'' exploration of the technical limits and abilities for surveillance in common wireless data networking. The work combines the current state of knowledge in privacy with probabilistic regression. Current location monitoring in wireless networks is integrally related to determining location from radio frequency information; current techniques to enhance privacy depend on the ability to fool observers by varying that RF information without affecting the ability to use the underlying network. This RF information is imprecise and subject to considerable environmental noise. Alternatively, privacy-enhancing mechanisms vary aspects of the media access layer, such as unique keys used to identify stations, in an effort to cloak an individual station and enhance privacy.
To date, there has been little investigation in to the issue of unintended disclosure of private information in wireless networks. For example, current monitoring techniques can identify the unique brand of cellphones being used in a region by characteristics of the electronic components used in different brands of cellphones. Normally, such sophisticated analysis requires expensive equipment. As more bandwidth is unregulated, the ability to monitor such bandwidths becomes an intrinsic aspect of the underlying technology. Software defined radio systems have the need and ability to use a number of frequency ranges and to exercise control over the MAC and PHY layers of the wireless medium, and emerging hybrid reconfigurable systems will accelerate the wide spread use of such devices. As such capable handsets become more commonplace, their very programmability will allow a broader range of increasingly sophisticated surveillance.
By exploiting a combination of machine learning algorithms and embedded systems, the research funded by this proposal seeks to verify or refute the privacy enhancing aspects of current research. Not only will this help determine the technological limits to existing and proposed privacy mechanisms, but it may help set guidance for policies in such domains. Moreover, the contrapuntal study that attempts to violate location privacy through advanced statistical machine learning may indicate that wireless privacy can not be enforced by technology, and must be dictated by policy. This surveillance technology may also be interesting in its own right, allowing more accurate tracking of parties that have noisy location sensors. The study uses probabilistic regression models to estimate trajectories from radio data collected by a software radio. The algorithms are sought that are simple enough to run on the hybrid reconfigurable logic embedded in the software radio, allowing diverse and realistic experimentation and evaluation.
