A Bayesian approach to relative node localization in Wireless Sensor Networks

Wireless sensors find many applications in, including but not limited to, wildlife habitat monitoring, defect monitoring in assembly lines, source localization, intrusion detection, etc. Many of these tasks require knowledge of the sensor locations. Often, we have information about only a few sensors (nodes) and the remaining locations have to be estimated. The process of estimating the node locations (referred to as blindfolded nodes) relative to the nodes (reference nodes), whose exact location is known a priori, is known as relative node localization. In this class project, we develop a Bayesian framework to accomplish this task. We consider the node localization based on received signal strength (RSS) measurements. In this, we follow part of the work Neal Patwari, et al [1]. Ours is a general relative location method. However, throughout, we will illustrate our method with a particular data set, that accompanying the electronic version of Patwari, et al’s paper. The data were collected at a Florida office of Motorola. Sensors, each with a transmitter and a receiver, were placed about the office space. For each pair of sensors, say the ith sensor and the jth sensor, the data consists of the strength of the signal received by the ith sensor from the jth sensor. By convention, a sensor cannot observe itself. We further assume the received signal strength is symmetrical. In other words, the strength of the signal received by the ith sensor from the jth sensor equals that received by the jth from the ith. The data set also includes the actual locations of the sensors, information we will make use of below in evaluating our models.