Experimental study of localization in sensor networks and design of adaptive localization algorithms

Abstract

[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] We design an experimental framework to evaluate localization methods for sensor networks. We use this framework to evaluate three existing localization approaches: Ad-hoc Positioning System (APS), Multidimensional Scaling (MDS), and Semidefinite Programming (SDP). Using this evaluation, we study the effect of several factors that affect the performance of localization. Through experimental study of the three selected localization methods, we present one possible combination of APS and MDS into one relative localization method that we refer to as: Simple Hybrid Absolute Relative Positioning (SHARP). Our proposed method performs better than both APS and MDS if both the localization accuracy and the energy consumption are considered. We further investigate different approaches to design of localization methods that are adaptive to network properties. This adaptation can be at the network level or at the partition level. In both cases, off-line training is used to decide which localization methods perform the best under what network properties. In the first case, dynamic discovery of network properties is assumed. Then, the training results are used to decide which algorithm to use. In the second case, training is done on network partitions (maps). The network is divided into local maps, where each map is supposed to have different set of network properties. Every map runs its own localization method. Finally, local maps are merged to form a global map, and anchors are then used to estimate the absolute positions.