A wonderful good morning!

On Mon, 25 May 2026 15:38:54 -0300
Luiz Paulo Damaceno via time-nuts time-nuts@lists.febo.com wrote:

Given you have only a single Cs beam standard, this looks ok and
in the order of magnitude I would expect the stability of the Cs beam.

The BVA only contributes to short term stability and thus does not
make a difference in the Circular T plots.

This depends a lot on what you want to achieve.
Is it short term performance you are after? Is it long term stability?
Is it reliability? Is it lower uncertainty to UTC?

Quite honestly, given that you are an NMI and we are talking about
a national timescale, I would recommend that you first try to enhance
your resilience against device failures. Having only a single Cs beam
standard, which need a replacement tube every 5 to 20 years (depending
on type and chance), you run the risk of losing your timescale when your
Cs beam standard runs out of Cs.

At the current state, I am not sure whether GNSS time transfer is
really the limiting factor for you. At least not short term.
IIRC the current uncertainty for GNSS time transfer is around 2ns,
even with iPPP, because of delay stability in the device itself.
Of course, if you can make your antenna's filter more stable, you
would then be able to get better absolute calibration.

Given your MDEV data, I would set the correction time-span to 1-2days.
At least, that's where your MDEV flattens out.

We have had some time-nuts members repair their H-Masers on their own.
But given that you are an NMI, I would contact the manufacturer. They
are most likely willing to help.

I hope this helps

			Attila Kinali

--
The driving force behind research is the question: "Why?"
There are things we don't understand and things we always
wonder about. And that's why we do research.
-- Kobayashi Makoto

John,

-> The Cs+GNSS ensemble looks like it's performing very well.  That's the
Septentrio PolaRx5TR receiver, correct?
That's correct.

-> It sounds like you may have a power supply problem with the maser, or
perhaps a controller issue, since you obviously have vacuum if the
dissociator is lit.  How far off frequency is it?  More than a few Hz
suggests that the oscillator oven has failed, which could explain everything
except the vacuum error.

Is not that too far, I can manually adjust the frequency and stability
seems quite good (unlocked PolaRx5Tr shows something around 4~3E-12 for 1s
ADEV).
Did you think checking power supply voltages (not trusting the display in
the instrument front).

Thanks

Luiz Paulo

Em ter., 26 de mai. de 2026 às 02:48, john@miles.io escreveu:

Thank you, Attila!

-> I want to achieve long term stability and also short term performance. I
think both are important to us. I can figure out that things better than we
have now may need more clocks or a change of base clock.

-> Quite honestly, given that you are an NMI and we are talking about
a national timescale, I would recommend that you first try to enhance
your resilience against device failures. Having only a single Cs beam
standard, which need a replacement tube every 5 to 20 years (depending
on type and chance), you run the risk of losing your timescale when your
Cs beam standard runs out of Cs.

• I'm not an NMI, I'm in a university that runs atomic physics
  experiments. Our UTC time scale is one way for us to have trackability to
  Cs frequency realization.
• Actually we have a second Cs beam that is not running due to this: have
  a second one in case of failure, at least.

Thank you for your answers,

Luiz Paulo

Em ter., 26 de mai. de 2026 às 06:56, Attila Kinali via time-nuts <
time-nuts@lists.febo.com> escreveu:

Ola Luiz Paulo!

On Tue, 26 May 2026 10:21:03 -0300
Luiz Paulo Damaceno luizpauloeletrico42@gmail.com wrote:

Ok. One "easy" first thing you can do is to more loosely couple
your BVA to your Cs beam standard. If your BVA is well aged,
it is likely to perform better than the Cs beam standard up
to 1000s. If you change the loop-bandwidth of your PLL such
that you get somewhere between 0.01Hz and 0.001Hz, then you
could make better use of your BVA. Of course, that's not
as easy as it sounds. If you have some phase measurement
equipment (e.g. a Timepod, but also a red pitaya would probably
do), you could use the BVA as input to the HROG, then compare
the phase of the HROG output and the Cs beam standard. It would
be then pretty easy to do the PLL loop on some PC and use the
HROG for the control/feedback part.

Oh.. sorry. My apologies! I assumed that you were an NMI as it is
rare that non-NMIs contribute to TAI/UTC.

Ok, so you don't have any legal requirement to keep a continuous
timescale. I guess your primary need is to provide a local realization
of the SI second that is as accurate as possible with as little noise
and uncertainty as possible?

Then I would focus on getting the hydrogen maser working again
and use it as input to the HROG. That will provide you the best
short to medium term performance.

Then use your phase meter to compare your HROG output and the
Cs beam output. This way you can correct for the drift of the
maser. Use the GNSS system to correct the drift of the Cs beam
standard with a tau in the range of a few weeks to a month.

This is basically what METAS does (minus the Cs fountain that
they periodically run) [1].

BTW: If your research involves optical atomic clocks and their
absolute frequency measurement, you might want to consider getting
a primary frequency standard. In the course of the presentation on the
2025 BIPM recommended frequency list at EFTF last month, Helen Margolis
from NPL did complain about the correlations of absolute frequency
measurements done through UTC, which are very hard to properly handle.

		Attila Kinali

[1] "METAS new time scale generation system - A progress report"
by Bernier, Dudle and Schlunegger, 2006
http://time.kinali.ch/ptti/2006papers/paper3.pdf

The driving force behind research is the question: "Why?"
There are things we don't understand and things we always
wonder about. And that's why we do research.
-- Kobayashi Makoto

Ola, Attila!

Thank you for the considerations. I've tunned the PLL for 0.00289Hz, today
i'm using a 1,1M ohm resistance and 50 uF capacitor. I can also reduce the
error voltage by changing the voltage in a potentiometer, just like a
"volume" control in audio systems. I expect to reduce a bit 0,25 times day
to day) to archive 1000s loop. I can only evaluate this using GNSS PPP - by
RTKLib in real time or NRCAN for example and / or CGGTTS files.

I have also available for these measurements a frequency comb from menlo
systems. Can I use repetition rate adev to evaluate the BVA? (locking it in
a Cs beam and then in BVA + Cs beam)?

Yes, no legal requirement, here in Brazil, only ONJR has the legal
requirement. Yes, my primary need is to provide a local realization of SI
second with little noise and uncertainty.

We are working in cold atom clocks here also. A Cs fountain and another
kind of cold atom Cs clock. But now, we're in the construction phase. No
measurements being made.

About getting the maser working again, do you have any experience with that
maser? Is the KVARZ CH1-75A.

Abraços,

Luiz Paulo

Em ter., 26 de mai. de 2026 às 11:27, Attila Kinali attila@kinali.ch
escreveu:

Thank you john!

I will verify supply rails with DMM / scope and return the results. For
now, Is it an active maser unit. CH1-75A from KVARZ.

Kind regards,

Luiz Paulo

Em ter., 26 de mai. de 2026 às 13:16, john@miles.io escreveu:

A wonderful good afternoon,

On Thu, 28 May 2026 10:37:58 -0300
Luiz Paulo Damaceno via time-nuts time-nuts@lists.febo.com wrote:

A PLL using the EFC input of the BVA will not work well at these
long time constants. About 1Hz loop bandwidth is the lowest that's
reasonable, I would even draw the line at 10Hz. And only if you
build an active filter using an low-noise opamp. Going lower will
limit your oscillators stability by the noise on the EFC voltage.
You have a 1MΩ resistor, which already produces quite a bit of
noise voltage. And then you use a 50µF capacitor. If it's an
electrolytic capacitor, then the leakage current noise will will destroy
the stability. If it's a ceramic capacitor, then its piezoelectric
properties will couple any acoustic noise and vibration into your
EFC voltage.

Best to ground the EFC voltage and then use a phase/frequency
stepper to shift the frequency of the BVA using a digital control
loop.-

If the frequency comb is CEO stabilized (aka f-2f lock) and you can
stabilize its repetition rate using an optical cavity (10cm ULE is
already enough), then you can get better stability than your BVA
in the range up to 10s probably better than 100s, depending on your
setup. If the comb is free running, then the intrinsic instability
of the comb will be larger than the BVA.

Oh nice! Do you already have any publications on this?
I could only find [1] and [2].

BTW: If you want some known-good reference. The Canadians (IIRC the
NRC itself) sell caesium fountains for a reasonable price.

I have never personally worked on a AHM myself. I do know their
working principle and how the electronics is supposed to look.
Unfortunately, I don't know anyone at Kvarz, only at the former
T4-Science.

		Attila Kinali

[1] "Progress in Brazilian Cesium atomic fountain-BrCsF",
by Rodriguez, Bagnato, Bueno et al, 2016
https://doi.org/10.1109/FCS.2016.7546818

[2] "Progress toward Brazilian cesium fountain second generation",
by Bueno, Rodriguez, Müller, et al. 2018
http://dx.doi.org/10.1088/1742-6596/975/1/012071

--
The driving force behind research is the question: "Why?"
There are things we don't understand and things we always
wonder about. And that's why we do research.
-- Kobayashi Makoto

Attila,

I Did tought that even low noise components that I have here (resistors,
capacitor opamps (like OPA637 and OP37/27)) may contour these issues. Also,
the PLL circuit is temperature-stabilized and mechanical noise poorly
insulated. But you have a good point. I may use some SG384 that I also have
here to remove the frequency offset and actuate in the long term drift to
lock it with the cesium.

We also have an optical cavity that is in progress to lock CEO. But this
will take a while I guess. The comb is repetition rate locked to BVA and
locked itself in the CEO.

Unfortunately that's only the both we have for now... We're ran out of
people working on this and haven't publications for a while.

Luiz Paulo

Em qui., 28 de mai. de 2026 às 12:30, Attila Kinali attila@kinali.ch
escreveu:

There's a set of photos of the HFO repair here:

http://www.leapsecond.com/museum/ch175/coil.htm

/tvb

Ola Luiz Paulo!

On Thu, 28 May 2026 15:13:08 -0300
Luiz Paulo Damaceno via time-nuts time-nuts@lists.febo.com wrote:

At first this would seem the case. But once you calculate the
contribution of the various elements needed to control the EFC in
an analog PLL system with extremely low bandwith (or large time
constants), the analog components noise and drift will dominate
the stability, especially with such superb OCXOs like the BVA.
It's really best to just ground the EFC and use a digital scheme,
either some DDS based system like the one from Rick Karlquist[1,2]
and the up-then-down conversion systems like the one formerly sold
by SKK Electronics[3], or some off-the-shelf high resolution frequency
synthesizer.

If you go for the off-the-shelf synthesizer, I would suggest you
set it to a higher frequency and then divide it down. This way
you get lower phase noise, less spurs and higher resolution.
I recommend using something akin to a lambda divider[4], as they are
very simple devices and still achieve close to theoretic performance.

Cool. So you are half way there!
Now you need to lock the frequency comb to an optical cavity.
The trick here is that an optical cavity can reach increadibly
high quality factor Q values. Much higher than anything in the
radio frequency and microwave domain could. The frequency comb
is then used to divide the optical frequency down to RF range
so it can be used in electronic measurements.

This is also the trick behind how the PTB CSF2 reaches the
short to midterm stability of an acvite hydrogen maser [5]

Have a nice weekend!

		Attila Kinali

[1] "A Narrow Band High Resolution Synthesizer Using a Direct
Digital Synthesizer Followed by Repeated Dividing And Mixing",
by Rick Karlquist, 1995

[2] "A 3 to 30MHz High-Resolution Synthesizer Consisting of a
DDS, Divde-and-Mix Modules and a M/N Synthesizer", by Rick Karlquist, 1996

[3] "Characterization of an Auxiliary offset Generator for steering
of H Masers", by Rovera, Abgrall, Siccardi, 2012

[4] "The Sampling Theorem in Pi and Lambda Digital Frequency Dividers",
by Claudio Calosso and Enrico Rubiola, 2013.

[5] "Advances in the accuracy, stability, and reliability of the PTB
primary fountain clocks", by Weyers, Gerginov, Kazda, Rahm, Lipphardt,
Dobrev, and Gibble, 2018
https://doi.org/10.1088/1681-7575/aae008

--
The driving force behind research is the question: "Why?"
There are things we don't understand and things we always
wonder about. And that's why we do research.
-- Kobayashi Makoto

Hi

Analog PLL’s can work to a point. Just what that point is has been pretty well gone over already.

The 50 uf cap in this case it something I’d be concerted about.

No capacitor is ever ideal. In this case the “parallel resistance” is worth looking at. Dielectrics (all of them)
leak. That leakage is temperature (and maybe other factors …) dependent. This leads you off in the direction
of things like Teflon dielectrics. That gets both expensive and hard to find.

Bob