Distributed multidimensional scaling with relative error-based neighborhood selection for node localization in sensor networks

This paper introduces a distributed weighted-multidimensional scaling algorithm that adaptively emphasizes the lowest relative error within the sensor networks. Each node adaptively chooses a neighborhood of sensors within 1 or 2 hops based on their relative errors and network density, then updates its position estimate by minimizing a local cost function and last passes this update to its neighboring sensors. For received signal-strength (RSS) based range measurements, we demonstrate performance via extensive simulation that location estimates are better than dwMDS(G), especially with rough initial global coordinates and in anisotropic topologies.