00071 §e586b9e4c806e780db8de67c99d012

70

Hembree & Zimmer

To test the DBAF, two time senes of fifty values each, with a measurement enor of 0.25 and system errors of 0.32 (conesponding to a "tnie" gain of 0.3) and .41 (conesponding to a "tnie" gain of 0.7), were randomly generated and processed by the filter. A plot of the process data, measured data, and the parallel filter estimate for the case of the "true" gain being eÄ…ual to 0.3 is given in FigurÄ™ 8. It is seen that even though the measurement is Ä…uite noisy, the adaptive estimate is relatively good. A plot of the individual filter weights is shown in FigurÄ™ 9. NotÄ™ that the weight for the "conect" filter goes to 1.0 as the weights for the other filters go to zero.

A similar situation is seen to occur for the data conesponding to a "true" gain of 0.7 as shown in Figures 10 and 11. Notice that the convergence to the conect model is much stronger for this case than for the case conesponding to a true gain of 0.3. The reasons for this are examined in the following sections.

FigurÄ™ 8. Process Data, Measurement Data, and Estimate (true gain = 0.3).