Patch-Based LLE With Selective Neighborhood for Node Localization

Received signal strength indicator (RSSI) gives a rough initial measure of the inter-node distances at low cost without the need for additional equipment or complexity. However, the intensity of RSSI varies over different regions of sensing field due to various environmental factors. This necessitates the need for a mechanism to process RSSI locally to degrade the effect of noise. In this paper, a patch-based locally linear embedding (PLLE) is employed, in which nodes in patches of two-hop neighborhood are localized followed by stitching of these localized patches. Neighborhood selection is used to determine an optimal neighborhood of a node for embedding. The neighborhood selection mechanism increases the accuracy of the PLLE. Experimental and simulation results show that the PLLE is able to localize the nodes with acceptable accuracy in a noisy environment. Results also indicate that the PLLE is able to localize sensor nodes more accurately as compared with native centralized LLE and other existing manifold learning techniques.