quantizer question

[Ingo Debus]

Fraser, Colin J wrote:
> > But i really don't want to start any war at DAC field.
> 
> No war intended, but I hate to be told something won't work, when I know
> from experience that it does...  :-)

I use a Doepfer "analog output interface" for cv generation. There's a
AD558J 8 bit DAC in it, the datasheet says it has +/- 1/2 LSB max
relative accuracy ("Relative Accuracy is defined as the deviation of the
code transition points from the ideal transfer point on a straight line
from the zero to the full scale of the device", quoted from datasheet).
I use 2 LSBs per semitone for VCO frequency control. This sounds ok to
my ears for a range of 5 or 6 octaves. At high frequencies the VCOs seem
not to be at the correct pitch, but this is probably a problem of the
VCOs itself (CEM3340, on Doepfer "Voice Cards"). If there were an
audible DAC error, I'd expect a few incorrect sounding semitone
intervals while most other semitone intervals would sound right.
But maybe a person with better trained ears would hear those incorrect
intervals.

Seems strange, since the DAC inaccuracy could introduce a maximum error
of 1/4 semitone to a semitone interval, which would surely be audible
(even to me ;-)). But it doesn't. The DAC seems to be much better than
the datasheet says.

> While we're on the subject, given that you might want to generate pitch
> bend / glide etc on a DAC output cv, does anyone know what DAC
> resolution is required versus frequency range to fool a human ear into
> hearing a continuously variable pitch instead of stepping ?

I have another input on the VCOs for pitch bend. The sensitivity is +/-
15 semitones for the entire DAC range (could have used +/- 12 semitones
instead, but I wanted the pitch bend range setable in semitone steps and
it was easier to implement the software this way). It sounds ok to my
ears.

Ingo