Non-linear techniques with filters

[Ian Fritz]

Hi Grant --

The effect you see is probably due to the overcompensation of the tempco
(3500 vs. 3300 ppm/K). Getting the sign right here is a bit confusing, and
it actually depends on which "flavor" of converter you have. If you use the
design where the control voltage is applied to the base of the reference-leg
transistor and the base of the output-leg transistor is grounded, then
overcompensation causes too high a frequency at high frequencies as T
increases. This is because the base voltage must go more *negative* to
produce a frequency increase. This is where the sign confusion arises.

The base bulk resistance does have a temperature dependence, but this
shouldn't have too large an effect, since the base-resistance error
correction is quite small.

I suggest trying to put a metal film resistor in series with the tempco to
get a composite with the right coefficient. There is quite a bit of
variation in the actual tempco, so its not likely to be precisely 3500 ppm.
To start with, you might try something around 170 ohms.

You might also want to have a look at the VCO notes at my site, where I have
discussed my efforts at reducing drift,  including getting the correct
tempco behavior.

  Ian

  http://home.earthlink.net/~ijfritz/index.htm



----- Original Message -----
From: "Grant Richter" <grichter at asapnet.net>
To: <synth-diy at node12b53.a2000.nl>
Sent: Monday, January 15, 2001 5:03 AM
Subject: Non-linear techniques with filters


> Now that Terry and Juergen have brilliantly dissected the mechanism at
work
> in the transistor ladder, it suggests another class of filter which
> deliberately exploits this perceptual effect.
>
> That is introduction of a non-linear network before a linear filter with a
> complimentary (or otherwise) transformation on the output. As I understand
> it, the non-linearities are then incompletely cancelled at the crossover
> point.
>
> The easiest to achieve would be log-filter-expo or the inverse
> expo-filter-log. I believe Terry told me the transistor ladder is of the
> expo-filter-log type.
>
> Certain of the Electronotes articles suggest the log-filter-expo could
> produce interesting results since if H(t) = X(t) * Y(t) then expo H(t) =
log
> X(t) + log Y(t) so the excitation and formant are combined additively.
>
> Please excuse my crude and possibly erroneous mathematics.
>
>
>