Ensuring data integrity in sensor-based networked systems.

摘要

One of the greatest challenges yet to be surmounted in computer science and engineering is that of the design, implementation, modeling, optimization, and effective use of distributed networked embedded systems; this is particularly true of distributed sensor networks. This dissertation addresses one of the canonical problems in sensor networks---data integrity. The goal of the data integrity approach is to lessen imperfections in the data via noise reduction, fault detection and correction, and recovery of missing data.; In sensor networks, a multitude of distributed sensing and computing nodes are embedded in the physical environment. This integration of sensing and computation will enable applications such as environmental surveillance (e.g. water and soil contamination control), battlefield planning, automated manufacturing, and traffic monitoring, which in turn promise to change the way we live and interact with our environment. Standing in the way of these advances is the fact that sensor networks often collect data with noise, faults, and missing samples. This is due to: the large scale and distributed nature of sensor networks: their heterogeneous node structure; cost and power constraints; operation in unpredictable, unconditioned and often harsh environments, the presence of lossy wireless links and the inherent unreliability of sensors.; This dissertation develops a systematic and consistent approach to maximizing and ensuring data integrity of distributed, recorded, and collected sensor measurements in a sensor network. This approach employs the principle of separation of concerns, to distinguish between different types of inconsistencies and to identify two essential phases of the data integrity problem. In the first phase, the data integrity approach identifies and builds models to represent sensors, noise, faults, and physical phenomena. In the second phase, the data integrity approach introduces procedures and algorithms that exploit constraints imposed by the models to assure data integrity. The presented data integrity approach observes three types of inconsistencies in the measured data, streams: (i) noisy readings, (ii) faulty (wrong) readings, and (iii) missing measurements that can be caused by either accidental; malicious, or intentional (e.g. interrupting measurement to save power) sources. Such classification identifies a broad spectrum of complex integrity problems, all of which can be addressed using the generic two phase approach. (Abstract shortened by UMI.)