Filters Just Aren't Perfect; Temperature Tuning - Alesis Andromeda A6 Tips And Tricks Manual

21.4 filters just aren't perfect:

Michael Caloroso wrote:
Many moons ago I took my favorite CD and pumped it through my Micromoog filter with it wide open
and quite frankly it sounded like shit too.
Scholastically speaking, most if not all synthesizer filters are Chebyshev class filters. Chebyshev
filters can provide a steep and consistent cutoff slope, but at the expense of flat response in the
passband (unfiltered frequencies).
If you want the flattest response possible in the passband, then you design a Butterworth filter but
you sacrifice steep cutoff slopes and consistency of the poles in the frequency domain. Crossovers
are designed with Butterworth filters so that the frequency response in the passband is as flat as
possible.
In the real (not virtual) electrical engineering world, you can't have both, no such design exists. This
has been studied for almost a hundred years.
Excerpts from my BSEE course Control Systems 482: the problem is further complicated when
variable resonance is desired. Every electrical circuit has a set of conditions where undesired
oscillation can occur, and when you implement controlled feedback for resonance you're raising the
risk of that. We're not talking audio band oscillation, we're talking oscillation way above that. The
downfalls of ultrasonic oscillation are decreased headroom, loss of function, and intermodulation
distortion that shows up in the audio band, among other foibles. Anytime any capacitive or inductive
effect appears, whether in an actual device or in circuit/IC substrate traces, you're asking for trouble
in unwanted oscillation and instability. Not every filter design in theory works on a real circuit board.
It also happens to be easier to achieve a controllable resonance with Chebyshev filters than is does
with Butterworth filters, and it takes less feedback (and less power) to implement resonance (and
self-oscillation) in a Chebyshev filter.
If you were to plot the passband frequency response of any synthesizer filter, be it Moog, Oberheim,
CEM, SSM, Roland, ARP, Korg, et al, you would find that *all* of them would have peaks and dips
in the frequency response, maybe 3dB or more. We're not after audiophile super-flat accurate
frequency response here - that's missing the whole point of a musical instrument.
A violin sounds interesting to the ear because of its timbre, and the spectral analysis (frequency
domain) of a violin shows that it does not fit any scholastic equation - it's imperfect. Yet it sounds
pleasing to the ear. The resonance of the violin body has a direct impact on its timbre.
Musically speaking, the goal of synthesizer filters is to COLOR THE SOUND, so deviation in the
passband is a side benefit of Chebyshev filters. The passband also varies with different resonance
levels. Different Chebyshev filter designs, IE Moog and Oberheim, add their own "color" in the
passband. It's imperfect on an audiophile engineering spec but our ears tell us that it sounds
pleasing. Plainly speaking, a Butterworth just doesn't color the sound enough to sound interseting.
So why does running the filters in series result in a better sound? You're probably hearing the dips/
peaks of FILTER1 being offset by the peaks/dips of FILTER2.

21.5 temperature tuning

Jeff wrote: