Estimation of the Covariance Parameters in Time-Discrete Linear Systems with Applications to Adaptive Filtering

摘要

The Kalman filter sequentially generates the minimum variance estimate of the state of a linear dynamic system. This estimate is a function of the covariance parameters of the dynamic system model, which implies that these be known a priori. Unfortunately some or all of these covariance parameters are often unknown in engineering applications. Two methods of estimating the unknown covariance parameters are examined in the dissertation. The first method is to compute the maximum likelihood estimates of the unknown covariance parameters from the measurement residuals generated by a sub-optimal sequential filter. The second method is to estimate the states and unknown covariance parameters from the measurements simultaneously. (Author)