Adaptive neighborhoods for manifold learning-based sensor localization

Connectivity measurements, i.e., whether or not two sensors can communicate, can be used to calculate localization in networks of inexpensive wireless sensors. We show that a Laplacian eigenmaps-based algorithm, combined with an adaptive neighbor weighting method, can provide an accurate, low complexity solution. Laplacian eigenmaps is a manifold learning method which optimizes using eigen-decomposition, thus is non-iterative and finds the global optimum. Comparatively, the new localization method is less computationally complex than multi-dimensional scaling (MDS), and we show via simulation that it has lower variance.