No-calibration localisation for indoor wireless sensor networks

In this paper, we propose and evaluate an algorithm and a few variations to exclude aberrant landmarks before performing a maximum likelihood estimator in order to evaluate a mobile node position in a 1 hop indoor wireless network more accurately. We exclude data coming from landmarks based on a global likelihood to alleviate the effects of multipath propagation and replace this data with a constant bias. We compare the cases when one, two or three landmarks are excluded with the classical unbiased maximum likelihood estimator over three different testbeds. We then evaluate a heuristic approach that does not need to examine all possible subsets of landmarks but selects excluded landmarks based on a threshold on the bias. Results indicate that, depending on the testbed characteristics, one strategy or the other can lead to better results, but in most cases the performance is improved w.r.t. classical maximum likelihood estimation.