Paradigms for Summarizing Data in Sensor Networks

Remote monitoring and response are intended to be a major application for sensor networks. The low cost and untethered nature of microsensors makes it possible to deploy them over large geographical areas and close to phenomena of interest, enabling unprecedented spatial and temporal measurement resolution. However, our ability to “view” (access) the remote phenomenon through this “virtual telescope” is limited by the severe resource constraints of the sensor networks. Indeed, the “information highway” of sensor network is more like a “dirt road” than the “autobahn” we have come to expect from the wired Internet—the sensor nodes themselves are highly constrained in energy (limited by an onboard battery), and the low power radio and the wireless medium provides limited and highly variable and bandwidth (due to radio interference and multihop communication). Thus, significant software and algorithmic advances are needed to efficiently utilize the resources of the sensor network: what to measure, when to measure, and how best to fuse those remote measurements into compact summaries that can be communicated inexpensively across the network. Reducing the amount of data communicated through a sensor network is of paramount importance because wireless communication is by far the most energy-costly operation in the sensor system. By some estimates, the energy required to transmit one bit can be 100 to 1000 times larger than the energy spent in processing a bit. Thus, a natural idea is to utilize the on-board processing power to “summarize” the data at each sensor node, and fuse data from multiple sensors as it is routed through the network. Such “in-network aggregation” appears to be so fundamental to efficient use of sensor networks, that this functionality is already built into the sensor database systems such as TinyDB [6] and Cougar [1, 2].