Cubature-based Kalman filters for positioning

We review a family of nonlinear filtering methods that includes unscented filters and cubature Kalman filters. These methods approximate the integrals occurring in the Bayesian formulation of the filtering problem by a sum of weighted integrand evaluations calculated at prescribed nodes. In addition to methods from the literature we introduce a new spherical-radial integration rule based filter. The filters are compared using an extensive set of positioning benchmarks including real and simulated data from GPS and mobile phone base stations. It is found that in tested scenarios no particular filter in this family is clearly superior.