[time-nuts] Precision DACs
scott.j.stobbe at gmail.com
Tue Jul 26 01:42:53 EDT 2016
As a clarification, the AD5791 is the minimum implementation of a DAC, it's
merely a resistor array with SPI controllable switches. (But an impressive
set of resistors, no doubt. Maybe with a dash of secret sauce in digital
calibration). The only guaranteed specs for the AD5791 are at DC,
everything else is up to the designer.
The thermal noise of the AD5791 is 4.07 nV/rtHz * sqrt( 3.4 ) = 7.5 nV/rtHz
(same as spec'd), where 3.4 is the nominal output resistance in kOhms.
Flicker noise is due to the voltage reference, reference buffers, and post
The settling time after code transition is complete, is based of the load
capacitance as seen by the DAC, and the buffer amplifier's transient
response. The settling time of a pole to +- 0.5 ppm is 15 time constants.
Even the sample application circuit only achieved a THD of 97 dB for a 1
kHz tone, which is equivalent to 16 bits.
The INL/DNL are measured at DC, if you were to measure the INL/DNL at 1Msps
on a half bit code across span (dither 1 LSB), the results would be
dramatically different due to glitching on code transition. That being
said, they are kept separate not to confuse sources of error.
20 effective bits is 122 dB signal to every thing else.
On Mon, Jul 25, 2016 at 7:40 PM, Attila Kinali <attila at kinali.ch> wrote:
> On Sun, 24 Jul 2016 23:48:05 -0400
> Scott Stobbe <scott.j.stobbe at gmail.com> wrote:
> > I doubt the AD5791 does much better than 16 bits operating at 1 Msps,
> > you include glitch energy, noise, and distortion.
> What makes you think so?
> Yes, if you are using the full 500kHz bandwidth then the rms noise voltage
> will be 5uV.. or 35uVp-p. But even just going down to a 1kHz bandwith
> gives 235nVrms/1.5uVp-p (plus 1/f noise of 1.1uVp-p). So we are within
> the 1ppm for any output larger than ~2V.
> The DNL and INL are low enough that I don't think they cause any more
> trouble then you'd expect from a DAC normally.
> I don't know how to make use of the glitch specs and turn them
> precision values.
> Attila Kinali
> Malek's Law:
> Any simple idea will be worded in the most complicated way.
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to
> and follow the instructions there.
More information about the time-nuts