[sdiy] Consider this DAC

[Magnus Danielson]

Antti Huovilainen wrote:
> On Sat, 13 Mar 2010, Paul Perry wrote:
> 
>> The thing about a DC cou[[led DAC that gives 16 bit accuracy...
>> that's going to be at least 4 bits better than the drift and noise floor
>> of the rest of your system.
>> Unless you are a very good analog engineer (certainly, far better than 
>> I.)
> 
> Surely achieving < 1mV (for max 5V signal) noise floor and drift can't 
> be quite that hard?
> 
> Plus of course there's the thing where you want higher resolution than 
> accuracy to avoid stepping even if you don't particularly care if your 
> filter cutoff is off by a 2 cents.

I was thinking about that. A standard 20 or 24 bit DAC isn't that 
expensive and a pair of resistors and a suitable op-amp for boosting the 
swing will work well enough for much less money.

If it is necessary to get some precision it should not be too hard to do 
a self-calibration system by adding a comparator per output. Just hook 
one input to the reference and one to the amplified output and the 
output of that comparator can then be used as digital feedback. Then it 
is a simple matter of doing a interval halving search for the DAC value 
where it flips and that way the gain error of the resistors can be 
cancelled out. Won't handle offset error, but if that level is needed, 
than comparing against reference ground should solve that.

But then again... only if it is needed.

Cheers,
Magnus