View article

Most existing work in aggregation and querying for sensor network data has focused on the use of standard statistical operations (average, median, etc) to reduce the quantity of transmitted data within a network These operations have been carried out without considering the nature of the actual data in the network. In this paper, we show how aggregation without correlation awareness will result in variable (often high) levels of error in the end results, and reformulate the nature of the aggregation problem in terms of information loss v.s. packet rate reduction. We instead propose Foxtrot - using phase space data representation combined with novel aggregation methods to limit errors to application-specific ranges. Foxtrot reduces data rates without significant information loss. We demonstrate that Foxtrot is able to achieve these goals, both using simulation methods and a TinyOS implementation.