[sdiy] YuSynth vco tracking

[cheater cheater]

You said your CV is best behaved below 0V. If I were you, I'd do it this way:

1. set up some frequency at 0V
2. make sure the frequency at -1V is 1/2 of that at 0V (never mind if
this makes the freq at 0V shift)
3. make sure the frequency at -2V is 1/4 of that at 0V, and frequency
at -1V is 1/2 at 0V

4. ... etc until -6V

5. set up the correct frequency at 0V by using a trimmer before the
expo. The voltages below should shift pitch correctly.

6. set up the correct frequency above 0V using the HF trimmer. At this
point I think you shouldn't have to touch the scale and offset
trimmers. Maybe the offset trimmer a *bit* if the HF influences the 0V
frequency.

D.

On Mon, Apr 19, 2010 at 21:19, David G. Dixon <dixon at interchange.ubc.ca> wrote:
>> Two points, two trimmer adjustments
>
> ...I should probably mention that, when I do this, typical tracking below CV
> = 0V is roughly as follows (absolute and relative errors in parentheses):
>
>  0V: 1000     ( 0   ,  0.00%)
> -1V:  500.3   (+0.3 , +0.06%)
> -2V:  250.2   (+0.2 , +0.08%)
> -3V:  125.0   ( 0.0 ,  0.00%)
> -4V:   62.37  (-0.13, -0.21%)
> -5V:   31.05  (-0.20, -0.64%)
>
> Hence, the LF tracking is actually a sigmoid rather than a line.

A sigmoid's first derivative doesn't change signs.

The series of the relative errors in fact looks very much like a
parabola. Can you try to plot this in 0.1V intervals? If it's a
parabola, then it might be that there's some simple electronic way to
eliminate that. I think this is worth pursuing.

http://www.wolframalpha.com/input/?i=-0.64%2C+-0.21%2C+0.00%2C+0.08%2C+0.06%2C+0.00

And for a parabolic fit see here:

http://www.wolframalpha.com/input/?i=Fit[{{-5%2C+-0.64}%2C+{-4%2C+-0.21}%2C+{-3%2C+0.00}%2C+{-2%2C+0.08}%2C+{-1%2C+0.06}%2C+{0%2C+0.00}}%2C+{1%2Cx%2Cx^2}%2C+x]

Obviously, this fit is very limited by the density and range of the
sample series; it's very much possible that in general the error has a
better fit by a third-degree polynomial, which would explain why the
value for -1V is less than it would be on a parabola and for 0V is
more than it would be on a parabola. See here:

http://www.wolframalpha.com/input/?i=Fit[{{-5%2C+-0.64}%2C+{-4%2C+-0.21}%2C+{-3%2C+0.00}%2C+{-2%2C+0.08}%2C+{-1%2C+0.06}%2C+{0%2C+0.00}}%2C+{1%2Cx%2Cx^2%2Cx^3}%2C+x]

Still this could be simple to fix.

Maybe it's an even higher order polynomial, but again, better
observations need to be performed for us to be able to tell.

> Of course, the error at the lowest octave is audible.  Perhaps it would be
> better to start at 31.25 or 62.5 and perfect the octave there.  However,
> this would send 500 closer to 501.  I can improve things if I drop the
> frequency at CV = 0V back to 500Hz (change a resistor), but then the HF
> (above 4kHz) goes flat.  Of course, the frequency errors up there are not
> really audible; only the beating of one VCO against another, which requires
> tracking precision from one VCO to another rather than accuracy.

You're assuming here that the musician will only be using your VCOs.
VCOs that are precise but arrive at this precision in another way will
make your VCO sound bad. Furthermore, other instruments will do the
same. For example samplers, live instruments, etc. To mention an
extreme, in an orchestra a difference this big is of course not
acceptable.

> Do other VCOs lose tracking accuracy at low frequencies?  (I suppose I could
> improve things by using opamps with lower offsets, but I'm not sure that is
> really the problem; I think it is an artifact of 2164).

Losing accuracy at low freq makes your VCO 'fat' :-) That's a loss
that's not necessarily desirable, but it's IMO not as bad as loss of
frequency in the higher tones. Bad tuning in bass is generally less
unmusical IMO, bad tuning in high frequencies can really hurt my ears.

Cheers,
D.