DAC "error"

[Harry Bissell]

Ive been quiet all day Dammit....

1) Are you doing a keyboard... or midi conversion...

2) Can we assume a keyboard, and pitch bend is done with a wheel etc...

3) Lets make the keyboard 63 notes (5 octave)

4) so we need a 6 bit converter (good for 64 notes...)

5) an 8 bit DAC with 1/4 lsb. error would have 1/16th of a semitone error
(since Data bit 5 is the LSB (and the spec is for DB7)  (data bits 0-7).

6) one semitone = 100 cents

7) so the error is about 6 cents... Good enough for most people. Not a problem
in a monosynth... maybe weak in a poly.

HOWEVER Paul is correct in pointing out that that error could show up in
obnoxious forms like every other note being sharp or flat.

The REV 2. Prophet V used a discrete 7 bit DAC with resistors matched to .01%

The Resistor Divider method may be better than you think... even 1% resistors
from the same manufacturing lot are usually closer than 1%... and the errors
do not add.. .they usually (statistically) subtract... One that is a little
high is often compensated by one that is a little low...  Depends on how fussy
you are...

The PRO One uses an 8 bit DAC, with accurate octave switching... and I've
never had a
problem...

:^) Harry

Any Y'know Paul... I've never accused you you trying to get away with the
cheaper components.... :^)

Paul Schreiber wrote:

> > Do you mean that 10.4mV is the resolution error?  The differential
> > linearity error would be some fraction of that depending on the DAC.
>
> That's the resolution error.
>
> >
> > Assuming you mean 10.4mV as a resolution error, I believe this spec is
> > important if one is trying to accurately hit *any arbitrary* voltage
> > accurately, not if one is *only* interested in hitting multiples of
> > 83.3mV accurately, starting at 0V (as with simple MIDI/CV converters).
>
> For this argument, assuming 1V/Oct control for VCOs. The pitches are on
> 83.3mv steps.
>
> >
> > However I agree that even with this "step voltage selection" method a
> > 1/4 LSB 8 bit converter would not have sufficient differential linearity
> > to meet your pitch accuracy specification.
>
> Amen, brother! But if 1LSB *is a step*, then 1/4LSB is 25% error! You need
> 1/100th LSB.
>
> Paul S.