The fundamental objective of sensor networks underpinning a variety of applications
is the collection of reliable information from the surrounding environment.
The correctness of the collected data is especially important in applications involving
societal welfare and safety, in which the acquired information may be utilized by
end-users for decision-making. The distributed nature of sensor networks and their
deployment in unattended and potentially hostile environments, however, renders this
collection task challenging for both scalar and visual data.
In this work we propose and address the twin problem of carrying out and defending
against a stealthy attack on the information gathered by a sensor network at
the physical sensing layer as perpetrated by a competing hostile network. A stealthy
attack in this context is an intelligent attempt to disinform a sensor network in a
manner that mitigates attack discovery. In comparison with previous sensor network
security studies, we explicitly model the attack scenario as an active competition between
two networks where difficulties arise from the pervasive nature of the attack,
the possibility of tampering during data acquisition prior to encryption, and the lack
of prior knowledge regarding the characteristics of the attack.
We examine the problem from the perspective of both the hostile and the legitimate
network. The interaction between the networks is modeled as a game where
a stealth utility is derived and shown to be consistent for both players in the case of stealthy direct attacks and stealthy cross attacks. Based on the stealth utility,
the optimal attack and defense strategies are obtained for each network. For the
legitimate network, minimization of the attacker’s stealth results in the possibility of
attack detection through established paradigms and the ability to mitigate the power
of the attack. For the hostile network, maximization of the stealth utility translates
into the optimal attack avoidance. This attack avoidance does not require active
communication among the hostile nodes but rather relies on a level of coordination
which we quantify. We demonstrate the significance and effectiveness of the solution
for sensor networks acquiring scalar and multidimensional data such as surveillance
sequences and relate the results to existing image sensor networks. Finally we discuss
the implications of these results for achieving secure event acquisition in unattended
environments.
Identifer | oai:union.ndltd.org:tamu.edu/oai:repository.tamu.edu:1969.1/ETD-TAMU-2946 |
Date | 15 May 2009 |
Creators | Czarlinska, Aleksandra |
Contributors | Kundur, Deepa |
Source Sets | Texas A and M University |
Language | en_US |
Detected Language | English |
Type | Book, Thesis, Electronic Dissertation, text |
Format | electronic, application/pdf, born digital |
Page generated in 0.0019 seconds