Supplemental Material for TKDE - 2011 - 02 - 0065

We distinguish two types of topology changes. Changes in the underlying network topology do not affect our suppression and redundancy schemes, because we operate at the level of suppression edges, which we do not assume to be direct communication edges. Changes in the suppression topology can be handled in a straightforward manner. For each suppression edge i → j that we want to change or create, the base station first learns a new suppression model from the past values of the readings involved (there is no need for any pilot runs because we have already been collecting data). Then, we send the new model to both i and j, which start tracking the suppression state and the transmission history map anew for the suppression edge. Note that changing the topology of cascaded suppression likely involves multiple suppression edges, because we must ensure the property that for each sensor k, there is a path consisting of suppression edges involving k’s reading to the base station, and the error bounds along the path sum up to no more than . Hence, frequent changes to the suppression topology may be costly. In practice, however, for many environmental sensing applications, the sensors are carefully designed and deployed to mitigate the need for frequent topology changes. In particular, cluster heads and those with power-hungry sensors carry more battery capacity, in anticipation of imbalanced energy use. This design effectively avoids topology changes due to load imbalance.