[time-nuts] Re: The SI second and the ease of realization (was: leap seconds finally being retired?)

[Magnus Danielson]

Hi Attila,

On 2022-11-24 18:14, Attila Kinali via time-nuts wrote:
> On Wed, 23 Nov 2022 23:50:35 +0100
> Magnus Danielson via time-nuts <time-nuts at lists.febo.com> wrote:
>
>> Further, there is additional concerns here. One such concern is one of
>> both political and practical aspect. If we come up with a definition of
>> a single spieces that for it's realization turns out to be hard to
>> realize or duplicate, with the needed repeatability (our ability to
>> build multiple giving the same measure), we can end up in a new kg
>> reference unit situation, in which only one lab (or possibly very few)
>> have access to the actual reference. This creats both practical issues,
>> you can say it is a problem of democratic access even, and there can
>> become political conotations to that which we really want to avoid.
> I would like to add here, that we already have this problem.
> If you look at the current list of primary standards contributing
> to TAI https://webtai.bipm.org/database/show_psfs.html you see that
> it's only a few labs. And it was just SYRTE, PTB, NIST and INRIM
> 20 years ago.
Indeed. It is assumed knowledge that we have that few primary 
references, and we do not want to drive it to fewer labs, but to more. 
Still may not be to many more, but consider the improved availability of 
doubled set of labs.
> Also note the huge gaps most of the primary standards have.
> I.e. very few are run once a month, much less continuous. And this is
> a technology that's quite mature and well understood.[1]
Back in 2017, the optical clocks for sure had challenges in continuous 
operation, and I pointed out that it needs to improve as a requirement 
for being considered a new primary standard, and we have now seen 
significant improvement and that show the maturity for sure is coming in 
that aspect. The maturity is needed since we need availability to the 
primary standard, and we need redundancy for both resilience of failure 
of individual realizations as well as the improved performance of 
multiple labs. The geographical dispersion is for sure also relevant.
> We currently only have 5 optical standards that make contributions to
> TAI regularly (if I'm not mistaken, all of them are lattice clocks)
> This is a quite low number, considering how many labs are currently
> working on optical atomic clocks. But keep in mind, that until recently,
> we didn't have that many labs with primary references either.
> All the same, I have no doubt that this number will increase in the
> comming years.

Yes, and we should encourage the proliferation into more labs to ensure 
both availability and the improved TAI/UTC performance as a result.

We are not yet there to have commercial optical clocks, but for sure see 
multiple efforts nearing that. More of the tools that goes into one has 
become commercialized. There is a range of challenges. The race for 
related technologies such as quantum computers creates a larger market 
for the components, such that stability of market and volume benefits is 
starting to be seen.

> I also would like to add that for a (legal) realization of the second
> at some NMI there is no need to run a primary frequency standard. All
> that is needed are some stable/low noise secondary frequency standards,
> e.g. a iMaser3000 or a 5071, a phase/frequency microstepper and a link
> to the TAI/UTC network to contribute to EAL. Then, through Circular T,
> one gets a frequency reading of the standards and a time offset of
> the local UTC(k) and can correct for that. That's actually what most
> NMI do, as far as I am aware of. (Please correct me if I am wrong)

You are not wrong. This is indeed how it works. None of the commercial 
clocks available is a primary standard in the metrology sense, but many 
NMIs use one or more of the commercial clocks for their realization and 
through the calibration to UTC achieves the calibration link. They 
typically steer their clock product towards UTC to keep the UTC(k)-UTC 
difference within bounds. This is not strictly needed, as calibration is 
enough to know the relationship and compensate for it, but it is 
desireable to have an actual clock being close to UTC at hand.

To remind people, the clocks reported into BIPM contribute to the EAL 
time-scale, which is synthesized paper clock to just be optimally 
stable. A number of measurements between labs establish the 
UTC(k)-UTC(l) differences and the lab report the difference of 
individual clocks to the UTC(k). The phase, frequency and drift of each 
individual clock is estimated, and so for each UTC(k) time-scale. The 
stability of each time-scale and clock is calculated and a weight of 
contribution is set for each clock depending on their noise such that 
the sum of the weighted clocks achieves an optimally low noise level. 
The produced EAL clock will however be somewhat of in frequency, so the 
small set of primary references is used to frequency reference and the 
EAL error is estimated and corrected for into the TAI time-scale. Once 
TAI is used, the leap-second information is added into forming the UTC 
time-scale. Once this is known the UTC(k)-UTC can be calculated but also 
each clock to UTC. Since the noise estimation using ADEV is linear drift 
sensitive, linear drift removal needs to be done in order to use 
secondary clocks experiencing linear drift. This include hydrogen 
masers. Once this was added into the EAL-processing, hydrogen masers 
could be directly reported and now enjoys a much heavier weight than 
cesiums. Historically some rubidium clocks have also contributed to EAL/TAI.

> This works because even a lowly 5071 has a long term stability down
> to 1e-13 (1e-14 with the high perf tube), which is plenty enough
> for most commercial applications and because we still have a hard
> time to get time diseminated with better than 1ns uncertainty.
This however keep changing. We see how sub 1 ns is more reachable, and 
100 ps is closer to availabile with commercial needs. Optical frequency 
comparison is regularly done with 1E-19 level of performance, matching 
the optical clocks being used. Mindboggeling as it is. Optical 
comparisons is only frequency comparisons, which is easier but to reach 
that level of performance require quite a bit of care to compensate the 
microphonics of optical fiber, but it is done and operational. Optical 
comparison using satellites is also looked at and techniques is 
improving. The many tools needed for international comparisons of 
optical frequency clocks is coming into place for sure.
> Yes, this means that any time-nut with a GPS disciplined Rb
> gets to within 1-2 orders of magnitude of an average NMI.
> And yes, I find this incredible!
It is. It really is. Some time-nuts is on par or even better than some NMIs.
> Sure, there is no legal traceability for a time-nuts lab, but
> who needs that anyways? 😜

Well, you can pay to get legal traceability. It's a monetary issue in 
the end.

Cheers,
Magnus

>
> 			Attila Kinali
>
>
> [1] If you wonder why the 6 Rb fountains from USNO are not listed here, then
> it's because they are contributing to EAL as secondary frequency standards.
> While they could be equally well be run as primary frequency standards
> using the secondary representation of the second[2], a decision was made
> to let them contribute to EAL instead of TAI.
>
> [2] "Recommended Values of Standard Frequency ... Rb 87 6.835GHz"
> https://www.bipm.org/documents/20126/69375083/87Rb_6.835GHz_2021.pdf/70065a76-1e50-254e-09b6-c29187263da0