A localization algorithm with learning-based distances for wireless Ad-Hoc sensor network

Abstract

This thesis describes a new localization algorithm for multi-hop ad-hoc sensor networks. Existing localization algorithms, which use either range-free or range-based information, have the problem that they can work appropriately only in their preferred sensor network environments with either high or low distance measurement error. The working environment, however, may change unpredictably from time to time or from applications to applications, making the existing algorithm unsatisfactory for applications that require localization. To solve the problem, we propose a supervised loaming-based approach in which we define a distance function and a learning-based distance localization algorithm. Through simulation, we show that by more efficiently using available information, our algorithm, which has competitive overheads, produces more accurate positions than existing works. This thesis also introduces an extension of the localization algorithm for an Evolvable Sensor Network. We first examine the requirements that a localization algorithm must satisfy to work efficiently and effectively in the network. An extension of the localization algorithm is then proposed. The simulation results show that it works appropriately.