Statistical Trilateration with Skew-t Errors

Abstract

In the problem of determining a target's location using radio signal time-of-flight to reference nodes with known locations, measurement errors can be skewed because of multipath effects. In this paper, range errors are modelled using the skew-t distribution. An Expectation-Maximisation (EM) algorithm for computing the unknown location is presented, and its accuracy is compared with a descending Gauss-Newton algorithm by simulations. The EM algorithm improves the positioning accuracy significantly. Furthermore, it is shown how to fit the parameters of a skew-t distribution to training data using a Gibbs sampler.