Any OCXO that has been sitting around for years is likely to undergo rapid
aging when you first bring it up.  It will be so far out of spec that I
wouldn't even bother looking at the loop behavior until it's been running
for a week or two.  Chances are it'll be fine by then.

-- john, KE5FX

John,

Sure! I had better formulated my question like: A constructor of a GPSDO
knows, that any OCXO that has been sitting around for years is likely to
undergo rapid aging when it is first brouht up. When Symmetricon choosed to
switch to a long time constant early after power up, were they perhaps
expecting a lower OCXO drift than I am observing now?

73s Ulrich DF6JB

Ulrich,

What are you using to produce the graph you attached please?

73,
Steve

2009/6/17 Ulrich Bangert df6jb@ulrich-bangert.de:

--
Steve Rooke - ZL3TUV & G8KVD & JAKDTTNW
A man with one clock knows what time it is;
A man with two clocks is never quite sure.

Steve,

my PLOTTER utility. I thought you did already try it out?

73s Ulrich, DF6JB

Ulrich,

I must have lost the plot here but I thought that was only for HPIB
data. How are you collecting the data?

73,
Steve

2009/6/17 Ulrich Bangert df6jb@ulrich-bangert.de:

--
Steve Rooke - ZL3TUV & G8KVD & JAKDTTNW
A man with one clock knows what time it is;
A man with two clocks is never quite sure.

Steve Rooke wrote:

Steve

You're confusing Ulrich's PLOTTER program with his EZGPIB program (which
can also communicate with RS232 devices like GPSDOs).
Once the data has been logged using EZGPIB, then Plotter can be used to
plot/analyse the stored results.

Bruce

Steve,

By far not! PLOTTER does a lot to everything than can be pressed into ASCII
data column files.

That has been done with a quick and dirty script for my EZGPIB utility
(which handles serial communication and some other goodies as well). I have
just included this script as "Z3805.488" to the examples.

Don't accuse me that I did not have warned you: PLOTTER is written in the
P-language while EZGPIB (which is written in the P-language itself) executes
scripts in the P-language, P-square if you like. Better get on rubber gloves
before you touch anything of it, hi.

73s Ulrich, DF6JB

Brian,

thanks for your information!

continuously

Is that from the Z3801 manual, which I have available, or do you have any
other in depth information source?

Best regards
Ulrich

Hi Brian,

Brian Kirby skrev:

For me it sounds more like a design-flaw than anything else. Using
several PLL bandwidths and switch between them is as such a good
approach, but the stepping between then needs to be done such that a
narrower bandwidth is only chosen when it the lock-in can be maintained.
Similarly, backing out of a narrow step to a wider step should also be
detected at suitable levels when it can't maintain track. The phase
detector gives hints about the ability to maintain track, as the phase
will deviate uncontrollably when loosing lock, but before that happens
it will deviate from near +/- 0 degrees, similarly, when within near +/-
0 degrees for sufficient time it is reasnoble that the next step (if not
too big) can maintain track. Recall that "sufficient time" changes with
the bandwidth of the PLL.

A third degree (PII^2 or PII^2D) PLL is better able to cope with drift
rate than a second degree PLL. A combined phase/frequency detection
approach (I.e. add the D term) adds quicker response to drift and
ability to keep tracking.

Another approach is to use a Kalman filter, where the Kalman gain is
adapted continuously. Kalman filters takes some careful thought, it's
not a magical wand to wave to make things better by magic. If done
properly, it will be able to fairly well track along and detect the
drift rate (takes a phase/frequency/drift model to handle) and update as
it stabilizes.

None of these approaches is rocket science to design anymore. HP/Agilent
surely could at the time of Z3801A and followers.

Cheers,
Magnus

Magnus Danielson wrote:

Yet the problem of an oscillator having a high drift rate after long-term
storage is not one that a development team is likely to be thinking about.

Nor is it a problem that can be easily examined, since it requires a bunch of
old units lying around the development lab. Old units don't exist when the
product is new, and even if they had traveled back in time to buy and age a few
units, after sufficient testing they would stop drifting and therefore be
useless for the test.

Nevertheless, 25 years later we expect the whiz-bang engineers at HP to have
thought of the problem and solved it.

--David Forbes, Tucson

A third order PPL has better sideband suppression.  The tracking capability
depends upon the phase range of the phase detector.  If 1:1 phase comparison
and a XOR phase detector is used, the range is 180 degrees. Dividers on the
feedback portion of the PLL increase the phase range of the phase detector.
For example, if the expected jitter is 10UI, a divider larger than 10 must
be used, to maintain lock under all conditions. A monotonically decreasing
phase delta, on the phase detector, still means that the PLL is locked.

The phase/frequency detector, avoids the need for a frequency aquisition
aid.  Once a phase reversal is detected by the flip flops in the phase/freq
detector, it goes back to phase detector mode.

My experience is that most fancy syncrhonizer for telecom application
(Stratum, LORAN and GPS) use start-stop phase detector with averager in
front of the AC-DC gain circuits.  Changing the closed loop bandwidth of the
PLL on the fly is not easy, due to secondary effects (done that many times).

-----Original Message-----
From: time-nuts-bounces@febo.com [mailto:time-nuts-bounces@febo.com]On
Behalf Of Magnus Danielson
Sent: Wednesday, June 17, 2009 2:31 PM
To: Discussion of precise time and frequency measurement
Subject: Re: [time-nuts] Z3805 initial behaviour after power up

Hi Brian,

Brian Kirby skrev:

For me it sounds more like a design-flaw than anything else. Using
several PLL bandwidths and switch between them is as such a good
approach, but the stepping between then needs to be done such that a
narrower bandwidth is only chosen when it the lock-in can be maintained.
Similarly, backing out of a narrow step to a wider step should also be
detected at suitable levels when it can't maintain track. The phase
detector gives hints about the ability to maintain track, as the phase
will deviate uncontrollably when loosing lock, but before that happens
it will deviate from near +/- 0 degrees, similarly, when within near +/-
0 degrees for sufficient time it is reasnoble that the next step (if not
too big) can maintain track. Recall that "sufficient time" changes with
the bandwidth of the PLL.

A third degree (PII^2 or PII^2D) PLL is better able to cope with drift
rate than a second degree PLL. A combined phase/frequency detection
approach (I.e. add the D term) adds quicker response to drift and
ability to keep tracking.

Another approach is to use a Kalman filter, where the Kalman gain is
adapted continuously. Kalman filters takes some careful thought, it's
not a magical wand to wave to make things better by magic. If done
properly, it will be able to fairly well track along and detect the
drift rate (takes a phase/frequency/drift model to handle) and update as
it stabilizes.

None of these approaches is rocket science to design anymore. HP/Agilent
surely could at the time of Z3801A and followers.

Cheers,
Magnus

time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

David Forbes skrev:

Starting any oscillator from cold is pretty upsetting and takes time to
drift in. But sure, it is kind of not so probable that they would
encounter it.

But still, crystal oscillators is individuals and there are other forms
of abuse (such as chock) that can occur in an oscillators life... so you
would expect that some anticipation of a "less ideal case" might have
passed their minds. Ah well, maybe that is me... I hate seeing returns
from the field only due to fairly easy mistakes in the design. Not that
it doesn't happens, but it's a learning process to improve on design margin.

Cheers,
Magnus

Hi Francesco,

Francesco Ledda skrev:

True, but regardless of how large you have made your phase detectors
range, if you have a loop unable to track to the dynamics you put into
it, you eventually reach the wrapping or limiting point. By back out
into quicker response, the PLL is able to quicker reduce the phase error
and track it in, as it quicker adjust its state to match the difference
between the source and the oscillator.

Sure, but it does not totally remove the benefits of shifting PLL
bandwidth according to need. I'm all for phase/frequency detectors...
and have designed several.

Such a detector isn't particularly useful for most GPSDOs as they get
have 1 Hz comparator frequency with the PPS pulse. The analog
time-constants would be a terrible mess to design. For GPSDOs using PPS
updates a digital loop filter can be assumed (unless you use say the
Jupiter receiver and it's 10 kHz output, see performance on TvB pages!).

Being able to scale loop parameters without having to re-scale state is
indeed an issue, but there exists topologies that can achieve this
without the need for rescaling. These can be realized both in the
analogue world using say OTAs and or through multipliers at the right
positions when done in the digital world. It is used with great success
in the GPS code and carrier tracking channels. You can look it up in the
Kaplan GPS book for instance... which only summarize what's found in
other places.

Cheers,
Magnus

They do not talk about driving the frequency off, but they warn to keep
the receiver locked the first 24 hours, so it can determine how the
oscillator ages.  Its in the section on "holdover" (page 52 of the PDF ,
page 3-8 of the user guide).

You may also want to search for HP An1279, it also goes over HP
smartclock operation, and look for a paper titled "the Global
Positioning System and HP SmartClock" by John A. Kusters.

If you have a broadband connection, I can email you each one, if that
will help.

Brian

Ulrich Bangert wrote:

Brian,

In the meantime I have not only found these but also "Smart Clock: A New
Time" by David Allan et al which shows that "Smart Clock" is originally a
NIST invention & patent and explains very well how it works. Perhaps what I
and You have seen is a direct consequence of applying the Smart Clock
algorithm.

Best regards
Ulrich

Gents,

one of the papers suggested by Brian says:

SmartClock monitors the frequency control variable of the internal
oscillator while it is locked to the external reference. This gives a
measure of the frequency difference between the internal oscillator, if it
is free-running, and the external reference over time. The resulting
measurements include the effects of random noise in the oscillator, the
measurement circuitry, and any noise in the external reference as well as
any aging and environmental effects in the oscillator. From this
information, SmartClock makes a continuous prediction of clock error over
time.

The big question is: If the EFC signal includes this all information, how
does the algo manage to extract the individual informations? After having a
look at the PPS TI and the EFC of my Z3805 I am beginning to get a clue of
it (the quirks on the EFC signal heve been removed):

The SmartClock algo seems to set the loop time constants to a large value in
the beginning. In other words: It measures the overall oscillator frequency
influencing effects with a lowpass filter applied that has a very low cutoff
frequency. Only effects which's frequency are sufficient lower then the
cutoff pass the filter. Seems as if the SmartClock uses this filter setting
to exclude any noise related effect and any environmental effect and make
the measurement see ONLY the oscillator's aging.

Once it has learned enough about the aging process it will be able to
predict aging. Having reached this state it will be able to apply his
knowledge about aging to further TI measurements and subtract the "aging
part" of it, leaving mostly the environmental (which is mostly temperature)
part.

Having a model for the aging the Algo can now model the temperature
dependence. However, a precondition for this were that the measurement sees
the temperature effects and that the ambient temperature is measured
independently. For that reason I expect the Algo will reduce its loop time
constant within the next time to make the temperature effect get through the
lowpass. Any bets on that?

Best regards
Ulrich Bangert

Aging cannot be predicted!  If it could be predicetd, there would be no
aging.

-----Original Message-----
From: time-nuts-bounces@febo.com [mailto:time-nuts-bounces@febo.com]On
Behalf Of Ulrich Bangert
Sent: Thursday, June 18, 2009 3:04 AM
To: 'Discussion of precise time and frequency measurement'
Subject: [time-nuts] Z3805 initial behaviour after power up revised

Gents,

one of the papers suggested by Brian says:

SmartClock monitors the frequency control variable of the internal
oscillator while it is locked to the external reference. This gives a
measure of the frequency difference between the internal oscillator, if it
is free-running, and the external reference over time. The resulting
measurements include the effects of random noise in the oscillator, the
measurement circuitry, and any noise in the external reference as well as
any aging and environmental effects in the oscillator. From this
information, SmartClock makes a continuous prediction of clock error over
time.

The big question is: If the EFC signal includes this all information, how
does the algo manage to extract the individual informations? After having a
look at the PPS TI and the EFC of my Z3805 I am beginning to get a clue of
it (the quirks on the EFC signal heve been removed):

The SmartClock algo seems to set the loop time constants to a large value in
the beginning. In other words: It measures the overall oscillator frequency
influencing effects with a lowpass filter applied that has a very low cutoff
frequency. Only effects which's frequency are sufficient lower then the
cutoff pass the filter. Seems as if the SmartClock uses this filter setting
to exclude any noise related effect and any environmental effect and make
the measurement see ONLY the oscillator's aging.

Once it has learned enough about the aging process it will be able to
predict aging. Having reached this state it will be able to apply his
knowledge about aging to further TI measurements and subtract the "aging
part" of it, leaving mostly the environmental (which is mostly temperature)
part.

Having a model for the aging the Algo can now model the temperature
dependence. However, a precondition for this were that the measurement sees
the temperature effects and that the ambient temperature is measured
independently. For that reason I expect the Algo will reduce its loop time
constant within the next time to make the temperature effect get through the
lowpass. Any bets on that?

Best regards
Ulrich Bangert

At 4:53 PM +1200 6/19/09, Bruce Griffiths wrote:

Bruce,

This sounds like a good excuse to learn how the wire bonder in the
office next to mine works.

--

--David Forbes, Tucson, AZ
http://www.cathodecorner.com/

Bruce,

I've never seen you post a new thread as a response to an existing,
unrelated thread, as you've done here. You are not the first to do this,
but I thought you might be aware of these subtleties.

In the spirit of quasi-related posts...

On the 5370A part, I recently bought one cheap and it sort of works but
seems very noisey in its lower order digit readings; clean external ref
doesn't help. I haven't looked at obvious things like the power supply
yet, but I'm curious if the group has any thoughts of common problems to
look for.

Bruce Griffiths wrote:

Bruce

Rex wrote:

I was being lazy.
How can you tell, looking at the message source I wasn't able to see any
indication of this?

One cause of this can be misalignment of the 10MHz to 200MHz frequency
multiplier chain filters.
Another thing to look at (if you are using the rear panel 10MHz output)
is the signal may not be clean enough in which case checking that the
output filter is working properly and/or using an external bandpass
filter may be necessary to cleanup the signal sufficiently.
What is the indicated std deviation of the delay between the STAR and
STOP inputs when connecting the rear panel 10MHz signal to both START
and STOP inputs using the STAR COM position of the channel mode switch?

Bruce

What mode are you observing the noise in?  Noisy LSDs on a 5370 are normal
in many cases.

-- john, KE5FX

Bruce Griffiths wrote:

I don't know enough about what is in the headers to explain the posting
issue -- one of the entries ties back to the previous message, I think.
The way I can tell is reading with the Thunderbird mail program, the new
message shows up still under the earlier thread but with the new
subject. I think I have seen similar with other mail programs.

Thanks for the reply to my 5370A question. I'll study on it. Since the
box sort of works, I think it should be fixable.

John Miles wrote:

I have a 5334B and looking at the same signal it seems more stable. From
memory (I haven't tried it recently) the 5370 had 3 or 4 unstable low
order digits. Even selecting high periods or sample size didn't seem to
remove noise out of two or three digits.

I never used a 5370 before, but this doesn't seem right. I get several
random digits even looking at it's own clean 10 MHz external clock. That
shouldn't be, should it?

Bruce Griffiths wrote:

Bruce,

I don't think I understand exactly how to measure this.

First, let me say that this is a very early 5370A. It doesn't even have
a 10544, it has the early crappy reference. Hence, I have been feeding
it a better signal to the external ref input.

Now, I am assuming you are talking about measuring a version of this
reference signal. I think you are saying, put the front panel switch to
START COMMON. Are you saying to feed in phase inputs to both Start and
Stop inputs? I thought Start Common tied them together. Then I have no
clue how to gather a reading for std deviation.

Sorry if I am being dense. This counter, though very old, is new to me.

Thanks for the help and suggestions.

The procedure Bruce was describing is basically the jitter test from the
manual, at http://cp.literature.agilent.com/litweb/pdf/05370-90031.pdf (this
is actually the same copy that David Kirkby scanned a few years ago, hosted
by Agilent.)  Follow the directions on page 3-11 and 3-12, and post the
results, especially from step 15.  This jitter figure is usually well under
50 ps in a 5370 that's working properly.

-- john, KE5FX

John Miles wrote:

Thanks. I think I did the beginning of that, but will need to do it
again to get anything accurate to post. Will do.

Rex wrote:

Yes, I should have given clearer instructions, just connect the 10MHz
out to  the START input with the swich selecting both inputs as COMMON.
Even with the ECL gate based crystal oscillator (my 5370A has one as
well even though its not a very early one, HP seemed to jump back and
forth between having a cheap oscillator and having a 10544A as standard)
the standard deviation should be less than 100ps. Typically something
like 15ps. With an external 10MHz signal connected to both inputs in
COMMON the STD deviation will be a little larger perhaps 35ps or so.

Bruce

Francesco Ledda wrote:

Not entierly true. Even with a perfectly known aging, only a few
oscillators would bother to estimate and correct it. This can never be
perfectly done anyway, and there is only so many things you can bring
into the model while keeping it economical. There is fairly good clues
around. If all was included and handled, it would reduce the effects.

Cheers,
Magnus

Bruce Griffiths wrote:

The references header points back to the post you replied to.  In this
case:

4ECC7583323F41D38EFD9C7AD619FE53@athlon

-Chuck Harris

Variation due to environment are deterministic and are nor part of aging.
Ageing needs to measured in conditions that remove ambient induced
perturbations. I used to put crystal oscillators in high quality
environmental chambers that kept temp, humidity and pressure constant and
measure the ageing.  There are tricks that can be done to increase the aging
rate, so that we did not have to wait 20 years ;)

We know that ageing happens, but the direction of ageing cannot be
determined.
A coin toss can yield face or tail.  If the coin is perfect, the
distribution will be uniform.  If a face is 1 and a tail is -1, we can add
successive tosses and track the total number.  Even is the distribution of
tosses is uniform, the sum will walk away from 0 (random walk),cameback to 0
and then move away from 0 again.  This is a good way to model and explain
ageing. Nature is perfect, but things do not follow our math perfectly.  A
better simulation would include a "leaky integral".

The statistical analysis of ageing is tricky, since ageing doeas have a
statistical mean, and therefore standard deviation cannot be used.

This discussion brings lots of good memories back.  The good old days when
SONET synchronization was a new thing, and we were creating new technology.

-----Original Message-----
From: time-nuts-bounces@febo.com [mailto:time-nuts-bounces@febo.com]On
Behalf Of Magnus Danielson
Sent: Friday, June 19, 2009 7:59 AM
To: Discussion of precise time and frequency measurement
Subject: Re: [time-nuts] Z3805 initial behaviour after power up revised

Francesco Ledda wrote:

Not entierly true. Even with a perfectly known aging, only a few
oscillators would bother to estimate and correct it. This can never be
perfectly done anyway, and there is only so many things you can bring
into the model while keeping it economical. There is fairly good clues
around. If all was included and handled, it would reduce the effects.

Cheers,
Magnus

a

in

frequency

cutoff

setting

temperature)

sees

the

https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts

time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

Correction...

-----Original Message-----
From: time-nuts-bounces@febo.com [mailto:time-nuts-bounces@febo.com]On
Behalf Of Francesco Ledda
Sent: Friday, June 19, 2009 9:38 AM
To: Discussion of precise time and frequency measurement
Subject: Re: [time-nuts] Z3805 initial behavior after power up revised

Variation due to environment are deterministic and are nor part of aging.
Ageing needs to measured in conditions that remove ambient induced
perturbations. I used to put crystal oscillators in high quality
environmental chambers that kept temp, humidity and pressure constant and
measure the ageing.  There are tricks that can be done to increase the aging
rate, so that we did not have to wait 20 years ;)

We know that ageing happens, but the direction of ageing cannot be
determined.
A coin toss can yield face or tail.  If the coin is perfect, the
distribution will be uniform.  If a face is 1 and a tail is -1, we can add
successive tosses and track the total number.  Even is the distribution of
tosses is uniform, the sum will walk away from 0 (random walk),cameback to 0
and then move away from 0 again.  This is a good way to model and explain
ageing. Nature is perfect, but things do not follow our math perfectly.  A
better simulation would include a "leaky integral".

The statistical analysis of ageing is tricky, since ageing doeas have NOT a
statistical mean, and therefore standard deviation cannot be used.

This discussion brings lots of good memories back.  The good old days when
SONET synchronization was a new thing, and we were creating new technology.

-----Original Message-----
From: time-nuts-bounces@febo.com [mailto:time-nuts-bounces@febo.com]On
Behalf Of Magnus Danielson
Sent: Friday, June 19, 2009 7:59 AM
To: Discussion of precise time and frequency measurement
Subject: Re: [time-nuts] Z3805 initial behaviour after power up revised

Francesco Ledda wrote:

Not entierly true. Even with a perfectly known aging, only a few
oscillators would bother to estimate and correct it. This can never be
perfectly done anyway, and there is only so many things you can bring
into the model while keeping it economical. There is fairly good clues
around. If all was included and handled, it would reduce the effects.

Cheers,
Magnus

a

in

frequency

cutoff

setting

temperature)

sees

the

https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts

time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

And be mindful that some frequency-standard devices contain integrated DC-DC
switching converters -- like my HP 58540A GPS-DO.  The noise from the HP DC
converters was spraying RFI every 100 kHz or so on the HF bands.  After
trying various RFI abatement measures, including additional bypass C and
toroid-wound #31 ferrite material, my solution was scrap the entire PS board
and bring in clean power from a triple-output linear supply made by
Power-One.  There's no switch-mode component anywhere in the path of my
GPS-DO.

I love switch-mode devices for their light weight and package density; I
hate them for their propensity to generate RFI and at least historically,
their abysmal long-term reliability.  That said, I do believe that recent
SMPS product is significantly more reliable than those manufactured 10+
years ago.

Paul, W9AC

Francesco Ledda wrote:

Not entierly true, as aging behaviour may be re-started or modified as
environmental conditions change.

Strange, as the aging rate decreases over time, most of its aging and
shift in frequency occur early. This occurs both in the "green" aging
period, when it is new, but also after it has been turned off and cooled
down / low temperature storage.

So what tricks did you use?

Standard deviation of frequency isn't used, but rather it is modeled
thru the frequency drift component, and this is also described for
different time-spans. Over time it integrates to the drifted frequency
over that time. In all models it is treated separately from the noise
components.

The curvature of aging has also been established (see for instance Vig
on the UFFC site), and with external reference (such as in a GPSDO), the
parameters for that curve can be established and adjusted over time for
better prediction of where the aging.

Or is it that we have different concepts or environments in mind when we
discuss aging? For me, aging is the shift in frequency over time as the
oscillator is maintained at the same temperature.

There are a few others here that was involved with that, and it would be
fun to have a few discussion on that.

Cheers,
Magnus