SELF ORGANIZING WIRELESS SENSOR NETWORKS
Blankenship, Gilmer L
MetadataShow full item record
This dissertation is concerned with the properties of self-organizing network systems, where a large number of distributed sensor nodes with limited sensing, processing and communication capability organize themselves into a cooperative network without any centralized control or management. Due to the distributed nature of the management and lack of global information for in-node decision making, sensor management in such networks is a complicated task. The dynamics of such networks are characterized by constraints and uncertainty, and the presence of disturbances that significantly affect aggregate system behavior. In this dissertation we examine several important topics in the management of self-organizing wireless sensor networks. The first topic is a statistical analysis to determine the minimum requirements for the deployment phase of a random sensor network to achieve a desired degree of coverage and connectivity. The second topic focuses on the development of a viable online sensor management methodology in the absence of global information. We consider consensus based sensor data fusion as a motivating problem to demonstrate the capability of the sensor management algorithms. The approach that has been widely investigated in the literature for this problem is the fusion of information from all the sensors. It does not involve active control of the sensors as part of the algorithm. Our approach is to control the operations of the nodes involved in the consensus process by associating costs with each node to emphasize those with highest payoff. This approach provides a practical, low complexity algorithm that allows the nodes to optimize their operations despite the lack of global information. In the third topic we have studied sensor networks that include "leaders," "followers," and "disrupters." The diffusion of information in a network where there are conflicting strategies is investigated through simulations. These results can be used to develop algorithms to manage the roles in the network in order to optimize the diffusion of information as well as protect the network against disruption.