next up previous
Next: Determination for the estimated Up: Algorithm implementation Previous: Grouping constraints

Separation block

In the separation block, the instantaneous amplitude Ak(t) and the instantaneous input phase $\theta_{1k}(t)$ are estimated from Sk(t) and $\phi _k(t)$ for the concurrent time-frequency region. This is done by the following steps, which optimize Ck,1(t) and Dk,0.

Step. 1
Let Dk,0 in Eq. (10) be any value within $-\pi/2 \leq D_{k,0} \leq \pi/2$.
Step. 2
Using the Kalman filter, determined the estimated region, which is $\hat{C}_{k,0}(t)-P_k(t) \leq C_{k,1}(t) \leq \hat{C}_{k,0}(t)+P_k(t)$, where $\hat{C}_{k,0}(t)$ is the estimated value and Pk(t) is the estimated error.
Step. 3
Select candidates of Ck,1(t) using the spline interpolation in the estimated error region $\hat{C}_{k,0}-P_k(t) \leq C_{k,1}(t) \leq \hat{C}_{k,0}(t)+P_k(t)$.
Step. 4
Determine Ck,1(t) using the correlation between the instantaneous amplitudes Ak(t).
Step. 5
Repeating Steps.1 to 4, determine Dk,0 using the correlation between the instantaneous amplitudes Ak(t)
Step. 6
Determine $\theta_k(t)$ from $\hat{C}_{k,1}(t)$ and determine $\theta_{1k}(t)$ from $\hat{D}_{k,0}$. Then, determine $\theta_{2k}(t)=\theta_k(t)+\theta_{1k}(t)$.
Step. 7
Determine Ak(t) and Bk(t) from Eqs. (7) and (8), respectively.



 

Masashi Unoki
2000-11-07