Considerations

The above results show the advantage of the proposed method. However, when the SNR of f(t) is 20 dB, the improvements in segregation accuracy using the proposed model, shown in Figs. , , and , were smaller. We suspect that the order of R for the polynomial approximation ( C_k,R(t), D_k,R(t), and E_0,R(t)) affects the improvement in segregation accuracy obtainable using the proposed model.

The results of simulations 1 and 2 showed that the difference in segregation accuracy improvement depended on the two types of noise. We suspect that the construction of the constant Q filterbank used here affected the segregation accuracy depending on the noise type. Since it has a constant Q on any channel (the same filter shape within all channels), the power of the components for which pink noise passed through the channels would be distributed approximately equally, while the power of the components for which white noise passed through the channels would be concentrated at higher frequencies. On the other hand, the harmonic components of the desired signal should not be satisfied precisely at a higher frequency, while they should be satisfied precisely at a lower frequency. Therefore, we consider that the interplay between the above factors reduces the improvement in segregation accuracy obtained by using the proposed model, as shown in Figs. (b) and (b).