[time-nuts] Re: Is this amount of measurement errors to be excepted when measuring with a small frequency difference?

[Tom Van Baak]

Erik,

Yes, those ugly effects are expected but unwelcome. You can consider it 
a measure of the [poor] quality of the instrument.

The test setup that I use is a very stable external 10 MHz reference and 
a very stable input of 10 MHz plus, say, 1e-10. This creates a growing 
"calibrated" phase difference of 100 ps per second, so over a run 
lasting 1000 seconds you have covered an entire 100 ns period of the 
clocks. The actual values aren't important, the key is simply to use two 
references that are stable and not the same frequency, and take your 
time so you carefully probe all phase points. It's a little like 
spectral analysis.

Even with a commercial counter like the hp 53132A you will notice 
pulling effects. There are many variations of the test, depending if you 
use just one input or two inputs, or how you use the ext ref input, or 
if you are in timestamp mode, or time interval mode, or frequency mode, 
or even how long your cables are. A while ago I tested several 53132 
with a slight offset 5 MHz input and found phase pulling by as much as 
200 ps depending where in the 100 ns ref cycle or where in the 200 ns 
input cycle the measurement was made.

http://leapsecond.com/pages/53132/

I may have mentioned this before, but many people test a counter by 
splitting a good oscillator, to the rear 10 MHz ext ref input and to the 
channel A input. And then they record lots of readings that say 10.000 
000 000 00x and are impressed with how good their counter is. This is a 
poor test because the counter is only working at one spot in its entire 
range of internal phase or interpolator relationships.

It's much better to feed in a second, independent frequency that's close 
to but not equal to the 10 MHz reference and then collect enough hours 
of readings to see repetitive effects. Over time this will expose all 
the internal phase relationships and phase or frequency pulling will 
become obvious. Fixing this is difficult because it's the result of 
subtle interaction among components, PCB layout, shielding, etc. But the 
smaller these effects are the better the counter.

/tvb


On 7/20/2022 8:34 AM, Erik Kaashoek via time-nuts wrote:
> During testing of some oscillators that where not exactly on the right 
> frequency it appears that there is a measurement error in the 
> frequency measurements every halve the inverse of the frequency 
> difference between reference and measured signal when the gate time of 
> the frequency meter is set to 20ms.
> An example plot can be found here: 
> http://athome.kaashoek.com/time-nuts/Frequency_pulling.png
> It shows the measured frequency with 0.5Hz (blue) and 1Hz (pink) 
> frequency difference.
> The normal frequency variations measured with a gate time of 20 ms are 
> below 2e-9 but when (presumably) an edge of the reference and the 
> input signal coincide(?) the frequency deviation can go up to almost 
> 1e-8.
> As you can understand this effect is not visible with a larger gate 
> time, such as above 50ms
> Is this normal behavior for a frequency meter?
> Does this imply that when you measure a signal that is phase locked to 
> a reference with a 20ms gate time you may have substantial larger 
> noise in the measurement because you have locked the system in the 
> worst case phase relation?
> The frequency meter used is a Picotest U6200A with an external 
> reference. The results are the same when using the internal reference 
> but the internal reference has much more phase noise obscuring the 
> effect a bit.
> Erik.
> _______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com
> To unsubscribe send an email to time-nuts-leave at lists.febo.com
>