Minimax design of Kalman-like filters in the presence of large parameter uncertainties.

An attempt is made to guide the designer by means of a systematic procedure which will yield a filter that performs in an acceptable fashion when there is some uncertainty in the various system parameters. Very general restrictions are placed upon these uncertainties, and it is shown that they are applicable in most practical situations. The minimax approach results in a unique fixed filter design which places a least upper bound on a given sensitivity measure over the assumed range of uncertain parameters. The resulting filters are identical in form to the Kalman filter, provide nearly optimal performance over the entire range of uncertain statistics, and are independent of the actual noise statistics. Examples are given.