In contrast to exact integral forms or general-purpose filter designs
for fractional differintegrals, we will develop them in *closed form*
as *exponentially distributed pole-zero pairs*. (On a log scale, the
poles and zeros are *uniformly* spaced.) It appears that such
filters approach the Chebyshev optimal approximation (in terms of
log-log *slope error*) for any
as the pole-zero density
and span along the negative real axis increase.

Any needed integer part for can be trivially provided using zeros or poles at/near the origin of the complex plane.

The proposed filter structure is furthermore robust for *time-varying*
, because the poles are fixed, and only the zeros need to slide
left or right along the negative real axis of the
plane as
is changed. On a log scale, the spacing of the zeros does not
change as they are slid left and right.

The fractional order sets the spacing of the zeros array relative to the poles array along the negative-real axis. For , the zeros lie on the midpoints between the poles. For , the array of zeros slides to right so as to cancel all of the poles, leaving the trivial filter , as desired. At , all poles are canceled except the first to the left along the negative real axis, leaving a normal integrator , as desired. At , all poles are canceled except the last, and one zero is exposed near , yielding a normal ``leaky differentiator with high-frequency leveling'' for frequencies interior to the interval . In the limit as the number of poles goes to infinity, and as , we obtain the ideal differentiator , as desired.

Download spectilt.pdf

[Comment on this page via email]

Copyright ©

Center for Computer Research in Music and Acoustics (CCRMA), Stanford University