[time-nuts] Comparing Reference Accuracy

Bruce Griffiths bruce.griffiths at xtra.co.nz
Thu May 22 19:45:24 EDT 2008

Stanley Reynolds wrote:
> Bruce
> Getting back to the idea that the DO controller needs to measure the oscillator to make corrections, would your suggestions make a better DO controller if used in this measurement ?  Or maybe instead of just measuring the difference between a group of DOs, we generate a improved standard from the group canceling their unique errors. This would force us back to real time so post processing would be out.
> I do understand that we need a better ruler to determine how good our ruler is, but at some point we improve to a group that is at the same level by elimination of the out-lye-rs. Or in my case the limit that I'm capable of implementing at this point.
> Stanley

Using a higher resolution phase comparator will improve performance as 
the phase measurement resolution increased up to the point where the PPS 
signal jitter dominates.

If you are using the PPS (or similar fixed frequency output) output of a 
GPS timing receiver to discipline a oscillator, then to minimise the 
noise you need to choose a GPS receiver for which the PPS jitter (after 
correction for sawtooth error) is as small as possible. Having done this 
you then need to choose a phase comparison technique that has a 
resolution commensurate with the residual PPS signal jitter.

If one uses hardware (programmable delay line - eg Dallas 1020/1021) to 
correct the PPS jitter than a very simple phase detector (a D flipflop) 
with high resolution can be employed.
The PPS signal is used to clock the D flipflop and the say 10MHz signal 
being disciplined is connected to the D input. For a wider range use any 
multiple of 1Hz as the D input.
The loop then adjusts the oscillator frequency so that the probability 
that the flipflop output is 1 when read (after a delay sufficient to 
ensure it has settled with high probability) is 50%.
The effective resolution of the D flipflop phase detector will be 
determined by the residual PPS signal jitter.
This avoids using either a high frequency clock or a complex high 
resolution phase detector.

Alternatively if one uses a high resolution phase detector with 
sufficient linear range the necessary sawtooth corrections can be made 
in software.
For example an ADC can sample a 1MHz sinewave on the leading edge of the 
PPS signal, after correction for sawtooth quantisation error the 
resulting phase samples can be used to discipline the oscillator. The 
loop will automatically lock to the 1MHz zero crossing that has the 
required slope.  Subnanosecond resolution is readily achieved with 
readily availble ADCs.

For the ultimate in performance use a GPS receiver that makes carrier 
phase data available.
Phase lock the GPS receiver oscillator(s) to the oscillator being 
Then use the carrier phase measurement data to discipline the oscillator.
There are some factors to take into account such as SV orbit parameters, 
cycle slips etc that complicate the control software.
However, in principle, no external phase detector is required as all the 
GPS receiver has all the necessary phase measurement hardware built in.


More information about the time-nuts mailing list