Square-root unscented Kalman filter for inertial navigation system alignment

Covariance matrix of state gradually loses its positive definition, thus bringing about computational divergence in some recursive processes in unscented Kalman filters (UKF). The covariance matrix of state is here decomposed and guaranteed nonnegative while its square-root used in the computation of the filter. The paper presents the square-root unscented Kalman filter (SRUKF) to estimate the psi-angle in the inertial navigation system having large misalignment on stationary and moved-base. Monte Carlo simulation results showed that SRUKF and UKF are basically uniform in their precision of filtering and rate of convergence, but SRUKF has better performance in numerical stability than UKF.