文摘
Energy efficiency has long been a key issue in designing and deploying successful wireless sensor network (WSN) applications because most wireless sensor nodes are generally powered by batteries and replacing/recharging them is expensive or even prohibitive in many situations. Systematic management of WSN data in energy efficient ways forms the core problem this thesis intends to address. In this thesis, I addressed this problem from physical, network, and application layers. On physical layer, a model-driven approach was designed to enable adaptive sensor scheduling that achieves significant energy saving while statistically guaranteeing sampling quality requirements. On networking layer, a potential field based anycasting framework, leveraging the analogy between communication networks and electric networks, was designed to enable scalable and robust many-to-many data routing. This routing framework allows global optimization of custom objectives with completely distributed and localized computations. This flexibility allows this framework to be used in many other routing scenarios such as opportunistic routing and multicasting in wireless mesh networks. On the application layer, a probabilistic stream relational algebra was designed to capture the intrinsic inter- and intra-stream correlations in data samples coming out of WSNs. This new data stream model extends from the classical relational algebra thus had expressing power to model many data management applications. The newly introduced re-sample operator in this model allows readily incorporation of quality of service and energy efficient data stream representations. These new techniques would greatly help improve energy efficiencies in managing data streams in WSNs.