[time-nuts] standard fusion for accuracy and redundancy
attila at kinali.ch
Thu Aug 4 14:19:32 EDT 2016
On Thu, 4 Aug 2016 11:08:52 -0400
Ruslan Nabioullin <rnabioullin at gmail.com> wrote:
> Hi, I'm in the process of setting up a public stratum 1 NTP server which
> will have at least one standard as a fallback to GPS (and possibly WWV
> and CHU), in addition to its primary purpose as a timebase for a
> microwave active SETI transmitter. So far I have an aging = 1E-11
> rubidium standard (which has expired calibration documentation), and I'm
> interested in adding more standards in the future, specifically
> high-quality OCXO(s) and additional rubidium standard(s) (I'm
> uninterested in cesium standards due to their definite and short
> lifespan, and masers are, in almost all certainty, insanely expensive).
I am not sure whether you want to build an ensemble for your ntp server
or you want to build an ensemble for your microwave transmitter.
The former is relatively easy to do, as us level of precision/accuracy
is more than enough. The latter is more involved and requires proper
elecronic design to achieve short term stability improvements over
a single frequency standard.
> Based on attempts of understanding the NTP documentation and answers
> from NTP fora, NTP doesn't perform PPS fusion for accuracy, but rather
> merely for redundancy (correct me if I'm wrong). Therefore, this
> complicates things to an extreme degree, for it means that the RF
> outputs (typically 100 kHz, 1 MHz, 5 Mhz, and/or 10 MHz), or PPS outputs
> have to be combined using some sort of a weighted fusion method (or
> simply unweighted, if the aging figures are similar across all the
> standards). The only commercial piece of equipment to perform this,
> manufactured by some Russian corporation, is obscure and just by the
> look of it prohibitively-expensive.
Mostly because this kind of equpiment is (semi-)custom build, these days.
Generally, there is a DMTD based phase comparator in combination
with a phase microstepper (or a DDS solution) together with a PC or similar
for the control software.
> So that leaves custom fabrication;
> the best information I could find regarding this is the paper ``A
> Digital Technique for Combining Frequency Standards'', by Lynn Hawkey,
> published in '69, which outlines nonnovel approaches and a novel
> approach to fusion. The method that's attractive to myself is the old
> RF mixing one, wherein double-balanced mixer(s) are used to sum RF
> signals from standards of similar aging figures, the resulting output(s)
> filtered, and the output finally sent to a frequency divider to generate
> the desired final RF signal (like 1 MHz for typical time code
> generators). Any ideas?
The easiest idea, if you are only looking for a NTP class solution,
would be to use some microcontroller, with a sufficient number of
timer (aka capture/compare) units and measure the individual PPS
arrival times, without any ciruitry inbetween (beside level adjustment).
This gives you a precision in the order of 10-50ns. You can generate
the output PPS from the uC with the same precision.
The nice thing about this solution is, that everything is done in
software (and quite simple software at that) and you dont have to
mess with any hardware (beside the level adjustments). And you still
get much better stability of the PPS than NTP can ever hope to make use of.
You can improve on this by using interpolators at the input to get
below 1ns resolution and using a OCXO as clock source for the uC
so you can shift the phase of the uC in small steps.
If you want to use the ensemble for your microwave transmitter,
then it becomes a question on what kind of performance levels you
want to reach, as the complexity increases very fast when you try
to improve stability and phase noise.
BTW: if you want to improve stability _and_ get redundancy,
then things become algorithmically complicated. Do not underestimate
the complexity you get into, if you add/remove frequency standards
from your weighting algorithm. How to achieve sufficent level of
redundancy while getting the maximum of stability from an ensemble
is an open research question.
Any simple idea will be worded in the most complicated way.
More information about the time-nuts