WSN Localization Using Similarity-Based Ricci Flow Embedding

Abstract

In low-cost Wireless Sensor Networks (WSN), node localization in a noisy environment is a significant problem as additional hardware like GPS is avoided due to high cost. The pairwise distances between sensor nodes are estimated using the Received Signal Strength Indicator (RSSI) because of the low computational complexity and cost-effectiveness. However, the RSSI distance estimates are rough and fluctuate owing to various environmental conditions like noise, interference, and fading. This results in non-Euclidean pairwise distance estimates. To compute the actual inter-node distances, we consider the manifold assumption that the nodes embedded in the low dimensional space apparently lie in an unknown high dimensional manifold. The inter-node distances can be estimated by reducing the dimension through manifold learning. This work proposes a localization technique based on Similarity-Based Ricci Flow Embedding (SBRFE), an unsupervised manifold learning method that reduces the non-Euclidean artefacts of a distance matrix and converts it into a Euclidean one. Each inter-node distance is considered independently, and the corresponding sectional curvature is updated iteratively till all the distances become Euclidean. Experiment results show that the SBRFE algorithm outperforms all the conventional manifold learning algorithms and can localize all the sensor nodes with an increased accuracy up to 94.55%.