Two-step measurement update for extended Kalman filtering

The nonlinear filtering for a class of discrete-time stochastic dynamic systems whose measurement equations contain linear (or universal linearizable) components and nonlinear components which are mutually statistical independent is investigated. A two-step measurement update is proposed for the filtering of the systems. The first-step update is a linear (or universal linearization) measurement correction which introduces an intermediate estimate, while the second-step nonlinear linearization update produces the final posterior estimate based on the first-step estimate. Since the first measurement correction is a linear or universal linearization update, it provides an accurate linearization reference point for the second nonlinear measurement update. Two simulation examples show superiority of the new estimation method.