Dr. Braileanu compares his A-EWD design technique with our WS technique (which he calls CID) and cites the relative errors of the two techniques as a function of frequency. The error plot he presents in his Fig. 8, which we have redrawn as Fig. 1, illustrates that the two techniques exhibit similar small errors in the filter passband but differ in the filter stopband. The magnitude of the errors for both techniques is on the order of 10~((-2)) to 10~((-7)). In many practical applications, the energy content in the stopband of a system contributes an insignificant level to the time signal passing through the filter. Small errors in insignificant contributions represent third- and fourth-order effects and are of little practical importance. In reality, analog filters built with real components have a tolerance on the order of one part in 100, and the spectral errors resulting from component tolerance spread far outweigh the simulation differences reported by Dr. Braileanu. Further, comparing the size of low-level errors in the frequency-domain description of the different filter design methods begs the real question: What effect do the errors cause in the fidelity of the time-domain signals? In reality, the time-domain response of different filter implementations is remarkably tolerant of small changes in the filter's frequency response. This tolerance is the result of the temporal averaging inherent in the convolution process.
展开▼