The system of the Kalman filtering is defined by
(20) |
The minimum variance is sequentially calculated.
(21) | |||
(22) |
(23) |
= | (24) | ||
= | (25) |
(26) |
In this paper, we reconsider how to estimate Ck,0 and Dk,0 from the observed component Xk(t) using Kalman filter. The estimation duration is [Th-1-Th]. It is then decomposed into discrete time , , where the sampling period is and fs is the sampling frequency.
First, Bk(t) and are estimated the Kalman filtering with the parameters in Eqs. (18) and (19) as shown in Table 2. By performing the Kalman filtering according to Eqs. (18) and (19), we obtain the minimal-variance estimated value and the covariance matrix at discrete time tm. Therefore, we obtain the estimated and .
Next, we obtain the estimated and from Eqs. (4) and (6) from the above parameters.
Finally, Ck,0(t) and Dk,0(t) are estimated with our previous Kalman filtering [Unoki and Akagi1999b] with the parameters in Eqs. (18) and (19) as shown in Table 2. Note that let Ck(t) and Dk(t) be and , respectively. By Performing the Kalman filtering according to Eqs. (18) and (19), we obtain the minimal-variance estimated value and the covariance matrix at discrete time tm. As a result, the estimated and , and the estimated errors Pk(t) and Qk(t) are determined by and , , and , respectively.