Towards adaptive sensor networks

Sammanfattning: Wireless sensor networks consist of many small embedded devices that are equipped with sensors and a wireless communication unit. These devices, or sensor nodes, are typically low cost, resource constrained and battery-powered. Sensor network applications include environmental monitoring, industrial condition monitoring, building surveillance, and intelligent homes.Sensor network applications today are often developed either using standard software components which enables simpler development but leads to far from optimal performance, or software customized for the specific application which complicates development, software updates, and software reuse.We suggest that logic is separated from configuration and other information, for instance, network statistics and estimated power consumption. Software components publish their configuration and other information using a generalized programming abstraction. Configuration policies are separate modules that react on changes and reconfigure the sensor node as needed. These configuration policies are responsible for coordinating the configuration between the components and optimize the sensor network towards the application objectives.One of our contributions is that we demonstrate the need for self-monitoring and self-configuration based on experiences from two deployed sensor networks. Our main contribution is that we devise a configuration architecture that solves the problem of cross-layer optimization for sensor network applications without tight coupling between components, thus enabling standard and easily replaceable components to be used. The configuration architecture makes it possible to reach the same level of performance as specialized cross-layer optimizations but without adding application-specific knowledge to the components.