[time-nuts] ADCs for phase noise measurement

Attila Kinali attila at kinali.ch
Sat Jun 11 17:25:20 EDT 2016

On Sat, 11 Jun 2016 11:53:22 -0700
"Richard (Rick) Karlquist" <richard at karlquist.com> wrote:

> Good call on the LTC2368-24.  The eval board for it has considerably
> better support (like p-scope) than do the Analog Devices eval boards.
> I think this might work for me.  What are the tradeoffs between SAR
> and sigma-delta ADC?  

Good question, I don't really know. And it's a moving target too.

Currently it seems that sigma-delta ADCs reach higher SNR levels
and have less inherent noise. I am not sure whether they are better
at DNL but slighly worse at INL, but from the datasheets I had a look
at it seems so. The advantage of SAR is that they can reach quite
high sampling rates, going up to 15Msps for an 18bit ADC (LTC2387-18).
Though there are sigma-delta that are quite fast, like the AD7760
that can deliver up to 2.5Msps (but only at an SNR of 100dBFS).
Sigma-delta ADCs get worse performance when run faster (they havily
depend on averaging, filtering and noise shaping), while SAR have their
performance almost independent of sampling rate (but the non-linearities
of the internal components get more pronounced with higher sampling rate).
One thing is for sure though: sigma-delta ADCs get you more bits/SNR per buck,
as the high resolution SAR are still quite expensive. SAR ADCs seem to get
better by the day, so it is a good idea to check again what kind of
performance level they reached, once in a while. Sigma-delta do not
develop that fast anymore, but still see some increase in SNR and
sampling rate.

Which one is actually better is not so clear, as a sigma-delta converter
doesn't get to full resulution until you go waaaay down with the sampling
rate (or average). And because the target applications for the two types
of ADCs are slightly different, the datasheets are written differently,
so it's not easy to actually compare their performance. Even more so as
you will be decimating/filtering the samples quite heavily.

I guess for your application, the easiest way is to just buy some
eval boards that seem good and measure their performance directly.

> Can you point me to a few of the "many" FFT
> tools?  Maybe I don't have to buy spectrum analyzer software.

Hmm.. I couldn't find any of the tools I used (back 15 years ago).
But if nothing else works, you can use python (scipy+matplotlib)
or matlab/octave to read the data (all can read wav files)
and plot an fft (shouldn't be more than a couple of lines in any of
those languages). If you get stuck with that, let me know.

			Attila Kinali

Malek's Law:
        Any simple idea will be worded in the most complicated way.

More information about the time-nuts mailing list