Author: I. Bilik
Publisher:
ISBN: 9789533070001
Category :
Languages : en
Pages :
Book Description
Two new recursive filters, named as GMKF and NL-GMKF, for linear and nonlinear, nonGaussian problems were presented in this chapter. The GMKF algorithm consists of the GSF followed by an efficient model order reduction method. The GSF provides a rigorous solution for state vector estimation in a linear DSS model with GMM-distributed system and measurement noises and it generalizes the original KF to GMMs. Practical implementation of the optimal GSF is limited due to the exponential model order growth. The GMKF solves this problem via an efficient model order reduction method. The problem of exponential growth of the model order was solved via the mixture PDF estimation at each step using the greedy EM algorithm. It was shown that greedy EM-based order reduction scheme does not.