[time-nuts] DMTD and data display

[Bob kb8tq]

Hi

> On Nov 22, 2020, at 2:40 AM, Tom Van Baak <tvb at LeapSecond.com> wrote:
> 
> Hi Skip,
> 
> The input menu for both TimeLab and Stable32 allow you to set tau0 and also the phase scaling factor. Given your choice of frequencies your sample rate will be 10 Hz, so set tau to 0.1 s. And, yes, the scaling factor will be 5 MHz / 10 Hz or 5e5. Both tools expect the scaling to be a multiplier (not a divisor) and so use 1/5e5 which is 2e-6.
> 
> Now, a word of caution. Do not make a plot that blindly goes down to 1e-17 or 1e-19 or 1e-20. That's likely wrong.
> 
> There are several ways to test your DMTD. One of best ways is to measure two oscillators for which you already know the ADEV. For example, if you happen to have two fine OCXO with ADEV (tau 1 s) of 3.0e-13 and 4.0e-13 then the expected stability measurement will be 5e-13. [1]
> 
> If your DMTD reports 5e-13 then well done! On the other hand, if you get the "wrong answer" then there's a problem with your actual sample rate, or your scaling factor, or too much noise in your offset oscillator, or problems with your ZCD or TIC, or you have internal noise in your DMTD. If the latter then re-try the test using two oscillators with poorer stability. At some point you will find a pair of oscillators for which your DMTD results agree perfectly with known values, or values measured using other comparators or counters that you have lying around.
> 
> A temptation is to put the "same signal" into both inputs and see  how low the "noise floor" is. For several reasons this can lead to bogus conclusions. It would be like building a DIY voltmeter and testing its accuracy, stability, linearity, and resolution by how well it reports "0.0 volts" when you short the inputs.

If you put a chunk of cable into one leg of the drive circuit, you will get a much more rational 
measure of the noise floor on a DMTD. You want to “explore” the region from about 15 degrees
of phase shift out to 180 degrees. ( If using a time tagger like the TICC ).  The reason is fairly 
simple. The Reference oscillator cancels out very will at “zero degrees”. As you go to longer and
longer delays ( this ultimately is a 10 Hz phase shift ) the correlation / dropout degrades. 

Yes, there are other issues, but this at least gets you into the ballpark 

Bob


> 
> So get some 1e-9, 1e-11, 1e-13 sources and see how well your DMTD reports their actual stability. Do some short-term runs (say, tau 0.1 s to 100 s) and mid-term runs (say, tau 100 s to tau 1 d). You may run into other interesting problems along the way so stay in touch.
> 
> In the end you should have a reliable instrument that's trusted at e-13 levels. FYI: the timing of your posting is serendipitous since I'm going through the same process this month with a DMTD from Corby.
> 
> /tvb
> 
> [1] In other words the RMS sum, which in this example is the Pythagorean triple 3:4:5 so rms(3,4) = sqrt(3^2 + 4^2) = 5.
> 
> 
> On 11/21/2020 1:44 PM, Skip Withrow wrote:
>> Hello Time-Nuts,
>> 
>> I have made significant progress on a DMTD instrument, and hope to be
>> making measurements relatively soon.  However, I have a question on
>> getting the correct results.
>> 
>> I have a box with DUT(5MHz), OFFSET (4.999990 MHz), and REF (5MHz)
>> inputs, with DUT and REF outputs that will feed my 5370B TIC.  My
>> understanding is that I gain an increase in resolution of 5x10E6 Hz/10
>> Hz = 5x10E5.  My question is how do I make this display correctly in
>> TimeLab (or Stable32)?
>> 
>> When I use just the 5370B and plot the noise floor (feeding a delayed
>> version of the A input into the B input) I get a nice straight line
>> with -1 slope starting at 1.15x10E-10 at 1sec and going down.With the
>> DMTD connected does this mean that I just put in a scale factor of
>> 5x10E5 in TimeLab (and also change the sample time to 0.1Hz)?
>> 
>> Or, is there a better way to set things up?  Any insight appreciated.
>> 
>> Regards,
>> Skip Withrow
>> 
> 
> 
> _______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com
> To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> and follow the instructions there.