[time-nuts] Comparing Reference Accuracy

Bruce Griffiths bruce.griffiths at xtra.co.nz
Thu May 22 23:01:27 EDT 2008

Stanley Reynolds wrote:
> Stanley Reynolds wrote:
>> Disabling all the GPS receivers at the same time would allow a comparison of the hold over of each. As we are also logging the temp and power some changes in these would also help determine how "good" each DO is as far as stability.
>> Stanley
> You still need a frequency comparison scheme with sufficient resolution 
> and low intrinsic noise (perhaps 1E-12 or better at 1sec).
> Since the long term average frequencies of the various oscillators are 
> not necessarily identical in hold mode, a phase comparison scheme with a 
> relatively wide dynamic range is required to avoid frequent phase wrapping.
> Bruce
> With the default settings the reduced VE2ZAZ controler would count all 160,000,000 pluses in a 16s period, phase is not measured so it has no wrap just the +-1 count. 
> Looking at the orginal Brooks Shera design you would not need the DAC chip and any neg power requlator to use it as a measuring device. And the clock U7 ECS300S-24 could be shared between the multi sets of boards. I'm not sure if it could be adapted futher for this meassuring task. 
If you can live with the resultant degraded performance.
The point is that the resolution is at best 100ns single shot with a 
10MHz clock.
This can be improved by using a higher frequency clock.
However using say a 1GHz clock introduces significant engineering/design 
problems if 1ns resolution is to be achieved.
If you persist in not using a synchroniser, as they do, then calculated 
averages are biased.
If the system noise isn't around 1 clock period other "interesting" 
artifacts occur.

The performance is inadequate for characterising even a moderately good 
oscillator with a 16s gate time.

When using an interpolator the required system noise level for unbiased 
averaging is significantly decreased to a few interpolator LSBs.

Merely counting the number of pulses in a fixed gate time is inferior to 
high resolution timestamping every Nth zero crossing of the frequency to 
be measured.


More information about the time-nuts mailing list