[time-nuts] Comparing Reference Accuracy

Bruce Griffiths bruce.griffiths at xtra.co.nz
Thu May 22 18:06:05 EDT 2008

Stanley Reynolds wrote:
> Stanley Reynolds wrote:
>> Block diagram of using a modified version of the DO controllers to compare several disciplined oscillators. The single logging PC could be several PCs as a way to have enough serial ports.  
>> http://i307.photobucket.com/albums/nn306/stanleyreynolds/planB.jpg
> This scheme is fatally flawed.
> 1) The various GPSDOs will only be statistically independent for short 
> averaging times.
> 2) For such short averaging times the monitoring scheme has inadequate 
> resolution unless the short term stability of the oscillators is very poor.
> Without an independent standard of accurately known frequency such 
> intercomparisons can only measure the relative short term instabilities 
> of the various oscillators.
> The errors of the various GPSDOs cannot be determined without access to 
> a primary standard.
> Common View and All in View GPS code and GPS carrier phase measurements 
> may be used to compare the long term performance of the GPSDOs with a 
> primary standard if someone with a primary standard is logging the same 
> data using their primary standard and making it available.
> For short term relative stability measurements a much higher resolution 
> scheme is required.
> The SNR is likely to be sufficiently high to permit picosecond or even 
> subpicosecond measurement of the relative phases of the various 
> oscillator outputs.
> Bruce
> Yes such a measurement is not much better than the log produced by the DO itself, but the idea was to pick the bad apple from a bunch of DOs. Adding a primary standard in place or in addition to one of the DOs would be nice. A separate run with a gate of 2 sec, 4 sec would produce larger counts and a little better comparison provided the firmware in the PIC doesn't overflow or overflow wrong. 
> Stanley

If the GPSDOs are behaving then all you will be doing is measuring the 
quantisation noise of your comparison scheme.
You can do much better than this with relatively little effort.

All that is required is an interpolation scheme with sufficient resolution.

The traditional method is to use a heterodyne scheme with a common 
offset oscillator to produce a low frequency beat signal for each 
oscillator and time stamp the beat signal zero crossings  (reduces to a 
dual mixer time difference system when comparing 2 oscillators). For 
moderate performance oscillators a time stamp resolution of 100ns will 
suffice with say a 10Hz beat frequency.
However the zero crossing detector has to be properly designed to 
achieve intrinsic jitter well below 1us.

Simpler schemes using high resolution time interval interpolation 
techniques achieve a resolution of 10ps or so at best.
Advanced techniques albeit with a narrow range of a few tens of picosec 
can achieve a resolution of around 1ps or less if the source phase noise 
is sufficiently low (bandpass filtering may be required).


More information about the time-nuts mailing list