Sensor Network Localization by Augmented Dual Embedding

In this work we propose an anchor-based sensor networks localization scheme that utilizes a dual spectral embedding. The input noisy distance measurements are first embedded by Diffusion embedding and then by Isomap. This allows to better estimate the intrinsic network geometry and derive improved adaptive bases, that are used to estimate the global localization via $L₁ regression. We then introduce the Augmented Dual Embedding by computationally augmenting the set of measured distances and computing the dual embedding. This significantly improves the scheme's robustness and accuracy. We also propose a straightforward approach to preprocessing the noisy distances via the triangle inequality. The proposed scheme is experimentally shown to outperform contemporary state-of-the-art localization schemes.