Variants of multidimensional scaling for node localization

Recently multidimensional scaling (MDS) has been successfully applied to the problem of node localization in ad-hoc networks, such as wireless sensor networks. The MDS-MAP method uses MDS to compute a local, relative map at each node from the distance or proximity information of its neighboring nodes. Based on the local maps and the locations of a set of anchor nodes with known locations, the absolute positions of unknown nodes in the network can be computed. We investigate several variants of MDS and their effects on the accuracy of localization in wireless sensor networks. We compare metric scaling and non-metric scaling methods, each with several different optimization criteria. Simulation results show that different optimization models of metric scaling achieve comparable localization accuracy and non-metric scaling achieves more accurate results than metric scaling for sparse networks at the expense of higher computational cost.