Robust estimation of a reception region from location fingerprints

Abstract

A method for fitting an ellipse-shaped reception region to a set of location-stamped radio signal reception reports, or location fingerprints, is presented. Reports are modelled as having a multivariate Student distribution. The method is less sensitive to outliers than existing smallest-enclosing ellipse and Normal-distribution based methods. A Gibbs sampling algorithm and an Expectation-Maximisation algorithm to compute ellipse parameters are presented.