[time-nuts] Using cheap sound cards for measurements

[Lux, Jim (337C)]

> > Wouldn't the cards need to be synchronized, though?  Essentially,
> > you're still  comparing the two signals with each other, just
> > doing it in software, rather than in hardware, in the classical
> > time interval counter scheme counting 1Hz (or 123Hz).  Syncing
> > inexpensive cards is a real chore (and the only reason to be
> > thinking about using this in the first place is to keep the cost
> > to a minimum, otherwise, you might as well build a special
> > purpose little box with counters & A/Ds, and an interface)
> 
> I'm not sure it's that important (or helpful) for the ADCs to share a common
> clock.  Presumably the ~100 Hz beatnotes being digitized are on the order of
> 1/100000 of the frequencies being measured.  That means that a microsecond
> of synchronization error between the ADCs would have an effect similar to a
> picosecond-scale error on the DUT/reference sides of the mixers.
> 
> Getting microsecond precision out of an audio ADC is going to require
> processing multiple successive samples, and IMHO it will also require  some
> kind of auto-calibration scheme since sound-card clocks probably drift more
> than 1 ppm per minute or so anyway.

They're not that bad. Fairly high aging, fairly substantial variation with temperature, but that actually stays pretty constant.



  Given the need for autocalibration  --
> probably through a high-frequency sidetone sent to both channels in phase --
> the difference in complexity between supporting two ADC clock domains and
> one is probably not a deal-killer.

Yes.. but if you start feeding multiple signals into the ADC (e.g. a calibration pilot tone), then you start running into intermod effects from the inevitable ADC nonlinearities. I don't have a good intuitive feel for just how good the digitizing needs to be for this approach; I guess if I want to go further, I need to sit down and do the math.


> 
> Most installations would probably need to use a beatnote frequency
> closer to  1 Hz, so that would take a lot of pressure off the ADC clocks.  It
> *might*  be enough to get you out of the autocalibration business, but my guess
> is  that matching the phase tempco of the (AC-coupled) sound card inputs
> might still be necessary for good long-term results.

But at 1Hz, you're down in the LF rolloff of the ADC.  They probably roll off around 10-20 Hz, and none too predictably (e.g. they just slap a suitable cheap ceramic capacitor in series with the audio as a DC block)

But that DOES bring to mind an even cheaper approach.. the DATAQ $25 data acquisition unit. 4 10 bit ADCs at 1kHz or so