Source Localization based on Received Signal Strength in Wireless Sensor Networks

Abstract

The fifth and future sixth generation of wireless technology will inevitably enable the proliferation of more wireless sensor networks (WSNs) consisting of a few, tens, hundreds, and even thousands of sensor nodes. WSNs are envisioned to play key roles in future industrial revolutions such as the fourth one. In this context, the localization of sensor nodes became of great importance due to the fact that the information gather by any sensor is fundamentally linked with a location regardless of an event. This results in a close dependence between the obtained information and the location of the sensor nodes.

This thesis addresses the centralized target localization problem in WSNs mainly based on metaheuristic algorithms. Contrary to most of the state-of-the-art works, in this thesis, we propose algorithms based on the differential evolution (DE) that do not approximate the maximum likelihood function nor require initial guessing points. The proposed algorithms carefully combine four techniques: The DE, opposition based learning (OBL), adaptive redirection, and anchoring for a better estimation of the unknown position of the target nodes in WSNs. The OBL increases the chances of starting the proposed algorithms with better individuals. The adaptive redirection guides the individuals that are outside of the solution space towards the global minima of the localization problem. The anchoring also guides the individuals through the evolution processes by incorporating the connectivity information, which is inferred from the interaction between sensor nodes.

The thesis focuses on noncooperative scenarios under simplistic and realistic assumptions, where the transmit power and/or path loss exponent are considered known and unknown, respectively. Although this thesis focuses mainly on noncooperative scenarios, it also deals with the cooperative scenario case. Additionally, it also analyzes the effects of more detailed (complex) path loss models on the localization accuracy. The thesis thoroughly compares the proposed algorithms in terms of localization accuracy, resilience to the distribution of sensor nodes, and computational complexity with nonconventional algorithms based on metaheuristic techniques and widely-known conventional algorithms based on the linear least squares, semidefinite programming, second-order cone programming, and unscented transformation. The proposed algorithms in this thesis are assessed and validated through simulation results and, more importantly, by using real indoor measurements.