Position estimation under model misspecification

Abstract

When time-based radio range measurements between network nodes are perturbed by a line-of-sight blocking obstacle, position estimation accuracy degrades significantly. The perturbation is caused by multipath propagation or excess delays since waves travel at slower speed while piercing the obstacles. Degradation of positioning accuracy results from position estimators not being aware of the line-of-sight blocking obstacle, i.e. The estimator assumes line-of-sight, but reality is non-lineof-sight. Hence the assumed model of the estimator differs from reality. In such a scenario, the model is said to be misspecified or mismatched. To assess the performance of position estimators operating under model misspecification, we employ the misspecified Cramér-Rao bound. We use the misspecified Cramér-Rao bound to predict the performance of a maximum likelihood position estimator under model mismatch. We numerically show a large increase of the mean squared error when the maximum likelihood estimator is misspecified. Based on our findings, we emphasize the importance of non-line-of-sight identification for designing position estimators.