iHRNL: Iterative Hessian-based manifold regularization mechanism for localization in WSN

In this paper, we propose an iterative Hessian regularization technique for node localization in wireless sensor network (WSN). The technique is based on the observation that received signal strength indicator (RSSI)-based node localization problem using a few location-aware anchor nodes, and a large number of location-unaware non-anchor nodes can be modeled as a manifold regularization-based regression problem. A signal transmitted from a node attenuates with distance. However, this attenuation is not uniform due to noise and other detrimental channel conditions. This apparently embeds the nodes in an unknown high-dimensional space. The proposed technique assumes that these nodes are the data points lying on a high-dimensional manifold and the locations of these nodes can be obtained accurately through an iterative Hessian regularized regression. In Hessian regularization, temporary location values are assigned to each non-anchor node, which are further refined in the subsequent iterations. This is followed by localized Procrustes analysis to offset the affine transformation on temporary location values. To validate the proposed technique, we deployed TelosB motes to form a WSN. The beacons broadcast from nodes were used for finding the RSSI distance estimates followed by the localization process. We observed that our proposed technique is able to localize the sensor nodes accurately and outperformed the baseline supervised learning, Hessian and iterative Laplacian regularization methods with an increase in accuracy of around $$70\%$$ over the baseline supervised method.