Author(s): Tomic, Slaviša
Date: 2017
Persistent ID: http://hdl.handle.net/10362/21537
Origin: Repositório Institucional da UNL
Project/scholarship: info:eu-repo/grantAgreement/FCT/OE/SFRH%2FBD%2F91126%2F2012/PT;
Subject(s): Target localization; Target tracking; Wireless sensor network; Received signal strength (RSS); Angle of arrival (AoA); Convex optimization; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática
This thesis addresses the target localization problem in wireless sensor networks (WSNs) by employing statistical modeling and convex relaxation techniques. The first and the second part of the thesis focus on received signal strength (RSS)- and RSS-angle of arrival (AoA)-based target localization problem, respectively. Both non-cooperative and cooperative WSNs are investigated and various settings of the localization problem are of interest (e.g. known and unknown target transmit power, perfectly and imperfectly known path loss exponent). For all cases, maximum likelihood (ML) estimation problem is first formulated. The general idea is to tightly approximate the ML estimator by another one whose global solution is a close representation of the ML solution, but is easily obtained due to greater smoothness of the derived objective function. By applying certain relaxations, the solution to the derived estimator is readily obtained through general-purpose solvers. Both centralized (assumes existence of a central node that collects all measurements and carries out all necessary processing for network mapping) and distributed (each target determines its own location by iteratively solving a local representation of the derived estimator) algorithms are described. More specifically, in the case of centralized RSS-based localization, second-order cone programming (SOCP) and semidefinite programming (SDP) estimators are derived by applying SOCP and SDP relaxation techniques in non-cooperative and cooperative WSNs, respectively. It is also shown that the derived SOCP estimator can be extended for distributed implementation in cooperative WSNs. In the second part of the thesis, derivation procedure of a weighted least squares (WLS) estimator by converting the centralized non-cooperative RSS-AoA localization problem into a generalized trust region sub-problem (GTRS) framework, and an SDP estimator by applying SDP relaxations to the centralized cooperative RSS-AoA localization problem are described. Furthermore, a distributed SOCP estimator is developed, and an extension of the centralized WLS estimator for non-cooperative WSNs to distributed conduction in cooperative WSNs is also presented. The third part of the thesis is committed to RSS-AoA-based target tracking problem. Both cases of target tracking with fixed/static anchors and mobile sensors are investigated. First, the non-linear measurement model is linearized by applying Cartesian to polar coordinates conversion. Prior information extracted from target transition model is then added to the derived model, and by following maximum a posteriori (MAP) criterion, a MAP algorithm is developed. Similarly, by taking advantage of the derived model and the prior knowledge, Kalman filter (KF) algorithm is designed. Moreover, by allowing sensor mobility, a simple navigation routine for sensors’ movement management is described, which significantly enhances the estimation accuracy of the presented algorithms even for a reduced number of sensors. The described algorithms are assessed and validated through simulation results and real indoor measurements.
