Estimation of C_k,0(t) and D_k,0(t)

  

Figure: Signal processing of a separation block.

We previously set the statistical parameters (mean and variance) for A_k(t) and $\theta_{1k}(t)$ with ad-hoc values without considering the distribution of the power of noise. As a result, the estimation of C_k,0(t) and D_k,0(t) was influenced by the power of the noise components that passed through the channels.

Since we need to know the statistical parameters of A_k(t) and $\theta_{1k}(t)$, we improved the estimation of C_k,0(t) and D_k,0(t) by using Kalman filtering as follows (See Appendix A about details):

1.

estimate B_k(t) and $\theta_{2k}(t)$ with Kalman filtering,

2.

estimate A_k(t) and $\theta_{1k}(t)$ from Eqs. (4) and (6),

3.

calculate mean and deviation of A_k(t) and $\theta_{1k}(t)$, and

4.

estimate C_k,0(t) and D_k,0(t) with the previous method [Unoki and Akagi1999b].