RE: [sdiy] DAC selection in MIDI-CV Part #1

[jhaible at debitel.net]

>I've found that the hand-matched metal film resistor R-2R ladder gives good
>results on monophonic keyboard interfaces. Start with a bunch of 2% or 1%
>and narrow it down with a DVM to as close as you can get. Six bits is all
>I've ever used (5-octave keyboards).

Right. I did the same thing, and it works ok. (If we have the time to
select resistors. It's fun if you do it once; not so much fun if you have
to do it all the time. Outright expensive if you have somebody else
doing it, i.e. if you buy 0.1% resistors. A real good DAC is less expensive then.)

There's also the issue of the switch resistance. I've used a 4050 for the
clone of the OB-4-voice key assigner. You must perform a lot of
tricks to get the accuracy. Connect several gates in parallel for the
higher bits (to decrease the error from the driver output resistance),
and/or use a rather high resistance for the R2/2R ladder. 
But higher impedance also means slow settling time (assuming
constant stray capacitance), so you might get in trouble if you
want to demux a lot of channels in an acceptable time.





>Also I've used an octal 8-channel 8-bit linear DAC (AD7228 I believe) driven
[...]
>Or maybe I picked a really good specimen for the DAC? 

Most likely, yes. In fact, *most* specimens should be better than worst case
data sheet errors. But what if you run into one that is not good enough?
As we're building this for a hobby, it's ok to just be lucky. (I guess the few
who accidetially bought a real bad specimen are the few people who
swearcheap  8 bit DACs are not enough.)

If you are a manufacurer, IMO you cannot rely on this. Guess you
bought hundreds of cheap DACs which all are fine (because they
perform like *typical* specs, i.e. much better than guaranteed),
make a product, sell the product to happy customers, and then
you order a new batch of DACs, and you get specimens that
are just inside the worst case specs. Major disaster. Especially
if you have just soldered them in, and sent your Midi2CV boxes
out to your customers.

Testing is a very important factor in semiconductor pricing.
If you can test for yourself (selecting 0.1% resistors, or
just try and use (or in rare exception, discard) a cheap DAC),
you can save a lot of money. (You invest your own time,
which can be fun.) If you don't want to test for yourself
(because you plan to sell hundreds of items), your best
solution is just like Paul described.



>Anyway not to belittle any efforts to obtain subjectively perfect pitch
>accuracy on anybody's DAC - especially on a product offered as a kit or
>finished unit. In that case the math and accuracy are much appreciated and
>even demanded. Lab quality to me is a big selling point. It's just that the
>garage workbench hacks can certainly produce useable machines, good enough
>for rock and roll.

Yes - because some of our garage workbench hacks are better than lab quality. (;->)
That's big advantage of building single quantity stuff. We're in love with what
we achieved, we love to adjust all these these trimpots to perfection.
In Production, however, you curse every extra trimpot you have to adjust.

JH.