LARGE-SCALE SENSOR NETWORK LOCALIZATION VIA RIGID SUBNETWORK REGISTRATION

We focus on algorithms for sensor network localization (SNL) that proceed by dividing the whole network into smaller subnetworks, localize them in parallel using some fast and accurate algorithm, and finally register the localized subnetworks in a global coordinate system. These divide-and-conquer algorithms are used to leverage existing high-precision SNL algorithms (e.g., algorithms based on convex optimization) to large-scale networks, which could otherwise only be applied to small-to-medium sized networks. The contribution of this paper concerns the final registration phase. We consider a least-squares formulation of the registration problem with and without anchor constraints, and demonstrate how this otherwise non-convex problem can be relaxed into a tractable semidefinite program. Simulation results for large-scale SNL demonstrate that the registration algorithm (together with an accurate localization scheme) offers a good tradeoff between run time and accuracy