Maximizing the accuracy of sound based tracking via a low-cost network of reconfigurable embedded nodes

This paper presents an approach for optimizing the accuracy of data models produced based on data sampled through a network of embedded sensors. The method considers three orthogonal facets defining model precision: minimizing the sampling error of the individual embedded nodes, sampling sufficient data from distributed areas to correctly represent the phenomenon of interest, and meeting the timing delays that guarantee the timeliness of data. The three objectives are achieved by dynamically reconfiguring the architecture of the embedded nodes, and dynamically selecting the data transfer paths to the decision making nodes. Sound based trajectory tracking is used as a case study for the proposed approach.