Self-organization and resource management in distributed sensor networks

With recent advances in communication and sensing technologies, distributed sensor networks (DSNs) are becoming more viable for civilian and military applications. A DSN typically consists of a sensor node layer having a large number of small, low power, low cost sensors connected as an ad-hoc network and communicating in a peer-to-peer fashion. The decision node layers of the DSN are usually hierarchically organized to collect the data from its sensor node layer, extract important and relevant features, and arrive at increasingly more abstract decisions. Self-organization of the sensor node layer to best achieve a specified sensing task and resource management at the decision node layer levels are two critically important research issues for bringing such DSNs closer to application.In this work, a task-oriented self-organization algorithm that enables the formation of a sensor group for a task announced to the sensor system is proposed. It sequentially selects the best-matched sensors via the iterative use of a leader election algorithm and a residual task calculation algorithm. To improve the associated communication overhead, the sensor node location information obtained from a localization algorithm is used in task broadcasting thus allowing the implementation of the algorithm to be confined to a dynamically maintained contributor group. Sensor localization aspect is based on a refinement of an algorithm that utilizes only the neighborhood information of each sensor node corresponding to each of its preset radio transmission power levels.Resource management within the decision node layer must account for, and be robust against, various types of time-varying delays and nonlinearities that are inherent in a communication network setting. This is especially the case when the DSN is deployed in a highly dynamic environment. Resource management must also be carried out so that nodes generating data/features perceived to be more important receive a higher proportion of the limited resources available at each decision node. A virtual queuing framework is proposed to address these concerns. Its effectiveness is demonstrated via the design of Hinfinity-norm based feedback controllers.Network calculus notions are refined and generalized to account for the time-varying link delays, and with the virtual queuing framework in place, these newly developed notions are used to arrive at feedback controllers that enable set point tracking of node buffers and rate control of incoming data sources