On Performance Measures for Approximative Parameter Estimation
Gustaf Hendeby, Fredrik Gustafsson · KTH Publication Database DiVA (KTH Royal Institute of Technology) · 2004
The Kalman filter computes the minimum variance state estimate as a linear function of measurements in the case of a linear model with Gaussian noise processes. There are plenty of examples of non-linear estimators that outperform the Kalman filter when the noise processes deviate from Gaussianity, for instance in target tracking with occasionally maneuvering targets. Here we present, in a preliminary study, a detailed analysis of the well-known parameter estimation problem. This time with Gaussian mixture measurement noise. We compute the discrepancy of the best linear unbiased estimator BLUE and the Cramer-Rao lower bound, and based on this conclude when computationally intensive Kalman filter banks or particle filters may be used to improve performance.