Multiple sensor estimation using a new fifth-degree cubature information filter

In this paper, a new class of cubature information filters (CIF) is proposed for multiple sensor estimation. This new CIF generalizes the conventional third-degree CIF to attain higher estimation accuracy using a class of higher-degree cubature integration rules including the fifth-degree Mysovskikh's spherical rule and the arbitrary degree radial rule. The statistical linear error propagation method is utilized to incorporate the high-degree cubature rule into the extended information filtering framework such that more accurate estimation can be achieved than the extended information filter. It also outperforms the unscented information filter as well as the particle filter. In addition, the high-degree CIF maintains close performance to the Gauss-Hermite information filter but uses significantly fewer quadrature points. As a result, the curse of the dimensionality problem existing in the tensor product-based Gauss-Hermite information filter can be greatly alleviated. Besides the improved estimation accuracy and computation efficiency, the high-degree CIF also exhibits the desirable robustness under unknown noise statistics. The proposed CIF is compared with other information filters via the state estimation of a two-phase permanent magnet synchronous motor and a target tracking problem, and demonstrates the best performance.