System identification for cascaded filter modeling

摘要

Design of cascaded kalman filters require adequate knowledge of the first filter f or: (1) propoer modeling in the second(integrating) filter or, at least, (2) correct error mean) analysis. System identification can provide the technology for estimating the possibbly unknown first filter model from simulated or real test data. The fidelity of the estimated model may vary depending on the model structure and observability of the model parameters from the test data. Low order generic models (eg. 1st order Markov processes) of the trajectory errors provide simple average error models for easy inclusion in the integration filter. However, specific situations such as time varying PDOP are missed. System identification can then be used to fit generic models to real world or simulation test data that a ccounts for the propoer time correlation and PDOP. The resulting modl provides enough fidelity for credible coverariance analysis design studies. Simulation examples are used to illustrate the approach.