Self-Calibration of Sensor Networks

We consider the problem of locating and orienting a network of unattended sensors that has been deployed in a scene with unknown sensor locations and orientation angles, when no “anchor” nodes are present. Many localization problems assume that some nodes have known locations and propagate location information about other nodes using triangulation procedures. In our formulation, we do not require such anchor nodes, but instead assume prior probability density function for the nominal locations of a subset of the nodes. These nominal locations typically have high uncertainty, on the order of tens of meters. The self-calibration solution is obtained in two steps. Relative sensor locations are estimated using noisy time-of-arrival and direction-of-arrival measurements of calibration source signals in the scene, and absolute location calibration is obtained by incorporating prior nominal location knowledge. We consider a Bayes approach to the calibration problem and compute accuracy bounds on the calibration procedure. A maximum a posteriori estimation algorithm is shown to achieve the accuracy bound. Experiments using both synthetic data and field measurements are presented.