Building information-theoretic confidentiality and traffic privacy into wireless networks

This dissertation studies how information-theoretically secure mechanisms for confidentiality and data-traffic privacy can be incorporated into existing and emerging wireless systems. The dissertation consists of three parts. In the first two parts, we study how certain properties of wireless channels can be employed to enhance confidentiality services that have traditionally been the responsibility of higher layers. We first explore the use of the wireless medium for the extraction of secret keys at the two ends of a wireless link, wherein the transceivers at either end are separated by a rich multipath scattering environment. We build a low complexity algorithm that allows two wireless devices to extract a common sequence of random secret bits, by repeatedly probing and estimating a time-varying channel between themselves. Further, we report on an implementation and evaluation of our algorithm on a modified 802.11a system. Next, we study the problem of securely pairing wireless devices in proximity of one another by establishing a shared secret key using a public source of RF transmissions. We employ measurement data to characterize the rate at which bits can be extracted and explore the simultaneous use of multiple transmitters to increase rate. Finally, we study the case when the public transmitter itself is under the arbitrary control of an adversary and we demonstrate a method that can allow successful key-extraction even with such an active adversary. In the final part of this dissertation, we introduce the problem of an unintended information-leakage channel in data traffic consisting of varying packet sizes. Packet sizes convey semantic information that can be related to their con- tent, which can be used as a fingerprint for classification. We formally study the packet-size side channel and explore obfuscation approaches to prevent infor- mation leakage, while considering padded dummy traffic and delay as bounded resources. We show that randomized algorithms for obfuscation can be studied as well known information-theoretic constructs, such as discrete channels with and without memory, and often lead to efficiently solvable constrained optimization problems.Ph.D.Includes bibliographical referencesIncludes vitaby Suhas Mathu