Re: [time-nuts] CSAC Project(was CSAC purchase)
Jim:
One around 1s/y, one 6 s/y and one 11 s/y. I was looking to do
better than 1 s/100 years, but that was for the CSAC.
Ronald
1 second/year is quite good - about 30 ppb. It's a bit tricky (like all
things time-nutty) - the "aging" on a TCXO could be that good - but the
instantaneous frequency control might not be that good. 1ppm is pretty
vanilla for a TCXO over a fairly wide temperature range, so 30 ppb at
"constant skin temp" (say, 5 C range) is probably reasonable.
I've got some test data here for some fancy TCXOs intended for space
with a spec of 2ppm aging first year and then 1ppm/year after that. The
actual aging in the first year was 0.08 ppm, at 70C. Some of the other
oscillators in the lot were 0.02ppm, 0.05ppm.
So, I think the spec here is "covers all the things that can go wrong",
but by cherry picking, you could do better.
(or, our system design could tolerate several ppm aging over years, and
"run of the mill" for Vectron was actually a lot better)
Hi
The CSAC spec sheet calls out an aging rate of 0.9 ppb per month as
“typical”. There is also a temperature spec of 0.4 ppb. If both are correct
for your sample (and aging is linear ) you would be out by roughly 10 ppb
per year. There also is a voltage stability spec that might be impacted depending
on how you manage power.
Taking the 30 ppb = 1 second number, you are at a 1 second / year rate after 3 years.
At that point, you have already drifted by a second, if the assumptions are correct.
This makes a massive assumption that the aging stays at the “typical” rate
for years. It’s a very good guess that it does not. Is it going to be 1/3 or 1/10
of typical over that period? Who knows.
Bottom line, you are going to be pretty far from 1 second per 100 years with
a CSAC based wrist watch, if it runs for years (or even for months). It will
do way better than a TCXO or OCXO based watch over months or years.
It’s still not perfect.
Bob
On Jan 24, 2018, at 3:59 PM, Ronald Held ronaldheld@gmail.com wrote:
Jim:
One around 1s/y, one 6 s/y and one 11 s/y. I was looking to do
better than 1 s/100 years, but that was for the CSAC.
Ronald
1 second/year is quite good - about 30 ppb. It's a bit tricky (like all
things time-nutty) - the "aging" on a TCXO could be that good - but the
instantaneous frequency control might not be that good. 1ppm is pretty
vanilla for a TCXO over a fairly wide temperature range, so 30 ppb at
"constant skin temp" (say, 5 C range) is probably reasonable.
I've got some test data here for some fancy TCXOs intended for space
with a spec of 2ppm aging first year and then 1ppm/year after that. The
actual aging in the first year was 0.08 ppm, at 70C. Some of the other
oscillators in the lot were 0.02ppm, 0.05ppm.
So, I think the spec here is "covers all the things that can go wrong",
but by cherry picking, you could do better.
(or, our system design could tolerate several ppm aging over years, and
"run of the mill" for Vectron was actually a lot better)
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.
Nigel;
Will read the PDF carefully.
Ronald
"if it runs for years (or even months)" sounds like an informed comment:-)
When searching for some data recently I came across a report which
might be relevant.
"A Second Look at Chip Scale Atomic Clocks for Long Term Precision
Timing", written by
Alan T. Gardner and John A. Collins of the Woods Hole Oceanographic
Institution, details their
experience with a number of earlier and more recent CSAC modules and
their findings make
for very interesting reading.
At the time of writing a copy is available here....
www.obsip.org/documents/Gardner_IEEE_Oceans_2016.pdf
Tim:
Only three times worst then CSACs?
Ronald
y those Woods Hole guys.
Their temperature-compensated 5 milliwatt crystal oscillators can be
back-corrected (linear drift model) to a few tens of milliseconds over a
year and they make a convincing case they know how to do this.
Their similar graphs for CSAC oscillators are maybe a factor of three
better.
Tim N3QE
Bob:
Does it make sense to reset yearly if the offset is milliseconds? Tenths?
Ronald
Actually it was not quite what it sounded like. What I was trying to
say was “free runs”
for years or even months. Any device that is re-calibrated will have
the aging drift zeroed
out in that process. As noted in another post, CSAC’s have gone
through some growing
pains. The Woods Hole paper came out sort of at a low point in the process. The
current crop of CSAC parts seem to be more reliable than the ones Woods Hole
reported on. I’ve seen failures over the years, but not a lot of them ….
Bob
Tom:
That PDF was interesting. Not certain I would've one.
Ronald
Also see the very nice presentation:
"Challenges of precise timing underwater"
http://www.ipgp.fr/~crawford/2017_EuroOBS_workshop/Resources/Gardner_OBS_Timing_ATG_20150427.pdf
/tvb
Hi
On Jan 25, 2018, at 5:47 PM, Ronald Held ronaldheld@gmail.com wrote:
Nigel;
Will read the PDF carefully.
Ronald
"if it runs for years (or even months)" sounds like an informed comment:-)
When searching for some data recently I came across a report which
might be relevant.
"A Second Look at Chip Scale Atomic Clocks for Long Term Precision
Timing", written by
Alan T. Gardner and John A. Collins of the Woods Hole Oceanographic
Institution, details their
experience with a number of earlier and more recent CSAC modules and
their findings make
for very interesting reading.
At the time of writing a copy is available here....
www.obsip.org/documents/Gardner_IEEE_Oceans_2016.pdf
Tim:
Only three times worst then CSACs?
Ronald
y those Woods Hole guys.
Their temperature-compensated 5 milliwatt crystal oscillators can be
back-corrected (linear drift model) to a few tens of milliseconds over a
year and they make a convincing case they know how to do this.
Their similar graphs for CSAC oscillators are maybe a factor of three
better.
Tim N3QE
Bob:
Does it make sense to reset yearly if the offset is milliseconds? Tenths?
Ronald
If you don’t correct the frequency offset, then you ultimately have a device
that is off by quite a bit per year. The key here is that it is frequency (and not time)
error. Once you get a significant frequency error, the amount of time you gain or loose
goes up. You no longer are in a 0.1 second region, you are now into a “second per
year” sort of situation.
Some math:
If the CSAC is at zero frequency error at the start of the year and drifts by 10 ppb
over that year, you have an average error of 5 ppb. Keeping things simple, you get
1/6 second error that year. (5 / 30 = 1/6).
If three years later, the CSAC is at 30 ppb and drifts another 10 ppb in frequency,
you now are at 35 ppb average frequency error. You will gain / loose more than a
second in that year.
The real numbers are slightly different. You need to look at when over the year the
aging happens. A device that ages a lot early on in the year will do worse than a
device that ages linearly over the year. A device that does all it’s aging only on the
last day would do better than either of the other cases.
Bottom line:
Your CSAC wrist watch is very much not a millisecond per year sort of device.
Best guess is it is in the 50 to 150 ms per year vicinity in the first year after calibration.
Based on previous posts, that is in the same vicinity as a WWVB sync’d wrist
watch and not quite as good as an typical Apple Watch.
Bob
Actually it was not quite what it sounded like. What I was trying to
say was “free runs”
for years or even months. Any device that is re-calibrated will have
the aging drift zeroed
out in that process. As noted in another post, CSAC’s have gone
through some growing
pains. The Woods Hole paper came out sort of at a low point in the process. The
current crop of CSAC parts seem to be more reliable than the ones Woods Hole
reported on. I’ve seen failures over the years, but not a lot of them ….
Bob
Tom:
That PDF was interesting. Not certain I would've one.
Ronald
Also see the very nice presentation:
"Challenges of precise timing underwater"
http://www.ipgp.fr/~crawford/2017_EuroOBS_workshop/Resources/Gardner_OBS_Timing_ATG_20150427.pdf
/tvb
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.
Jim;
No need to order in the spring or fall anymore? Looking forward
to your data.
Ronald
no kidding - there's a well known issue when MicroSemi took over
building CSACs from Symmetricom, these things (like many precision
timing widgets) have a "recipe" and it's easy to "lose the recipe" or
find that there's unexpected and unknown components to the recipe.
Take a look at data sheet revs for the CSAC.. full temp range, then all
of a sudden around rev G or H, temperature range is quoted at 0-35C
operating, 0-40C non-op. I asked the sales rep if they ship them with
icepacks in styrofoam like mail order cheese in the summer - that UPS
truck gets way over 40C inside.
As always, this was discussed on the list and is in the archives.
That being said, I have no complaints about Microsemi being forthcoming
about the issue and helping us to understand the nature of the problem.
And they claim to have fixed the problem.
Hopefully, this summer, I'll have some data from a "narrow temp range"
CSAC against GPS 1pps in an environment where there's no gravitational
effects, and fairly small temperature fluctuations.
Bob:
Maybe I benefited a little by not buying a few years ago?
Ronald
Just to be clear, the current data sheet has the temperature range back to
-10 to +70C. They most certainly had a major headache on their hands for
several years straightening things out. I have not seen any complaints about
the “new” (post rework) version of the part.
Bob
Aaaas
On 1/25/18 3:28 PM, Ronald Held wrote:
Jim;
No need to order in the spring or fall anymore? Looking forward
to your data.
Ronald
you and me, and a bunch of other folks as well
Bob:
Sounds reasonable. You suggest to let it age a year and reset
often during the year?
No way to compensate for a linear frequency drift?
Ronald
If you don’t correct the frequency offset, then you ultimately
have a device
that is off by quite a bit per year. The key here is that it is
frequency (and not time)
error. Once you get a significant frequency error, the amount of time
you gain or loose
goes up. You no longer are in a 0.1 second region, you are now into a
“second per
year” sort of situation.
Some math:
If the CSAC is at zero frequency error at the start of the year and
drifts by 10 ppb
over that year, you have an average error of 5 ppb. Keeping things
simple, you get
1/6 second error that year. (5 / 30 = 1/6).
If three years later, the CSAC is at 30 ppb and drifts another 10 ppb
in frequency,
you now are at 35 ppb average frequency error. You will gain / loose
more than a
second in that year.
The real numbers are slightly different. You need to look at when
over the year the
aging happens. A device that ages a lot early on in the year will do
worse than a
device that ages linearly over the year. A device that does all it’s
aging only on the
last day would do better than either of the other cases.
Bottom line:
Your CSAC wrist watch is very much not a millisecond per year sort of device.
Best guess is it is in the 50 to 150 ms per year vicinity in the
first year after calibration.
Based on previous posts, that is in the same vicinity as a WWVB sync’d wrist
watch and not quite as good as an typical Apple Watch.
Bob
Jim:
Likely wait for your data before ordering one. Too bad the chip
price is so high, compared to a few years ago.
Ronald
Hi
The CSAC (like any vapor cell standard) has a drift (aging) process. That’s
just the way it works. It is at a much lower rate than a crystal oscillator, but
it is the same sort of idea. It is one of their basic differences from a Cesium
beam tube.
Can you “estimate” aging in advance? Everything I’ve seen suggests that you
are simply guessing when you do. You will be right in some cases and wrong in
other cases. You can do better estimating warmup drift and short term effects.
Working out what will happen over months (or years) is not very easy.
A very basic example:
I observe a couple of units and they all go positive by 0.8 ppb / mo. I put in
some code to work with that. The unit I happen to have goes 0.8 ppb negative
a month. My code has actually made the unit 2X worse than it would have
been if I just stayed away from it.
Yes there are papers on aging estimation. They mainly focus on coming up
with a “worst case” number. If they guess that 0.8 ppb will go on forever and
it drops off a bit from there, they did ok. For correction purposes … not so much.
On a practical basis, you will have to dock this beast up with a charger on
a regular basis. A solar powered WWVB watch is not unusual. A solar powered
Apple watch or CSAC watch … not so much. When it goes to the charger,
sync it up with GPS.
Bob
On Jan 26, 2018, at 7:19 AM, Ronald Held ronaldheld@gmail.com wrote:
Bob:
Sounds reasonable. You suggest to let it age a year and reset
often during the year?
No way to compensate for a linear frequency drift?
Ronald
If you don’t correct the frequency offset, then you ultimately
have a device
that is off by quite a bit per year. The key here is that it is
frequency (and not time)
error. Once you get a significant frequency error, the amount of time
you gain or loose
goes up. You no longer are in a 0.1 second region, you are now into a
“second per
year” sort of situation.
Some math:
If the CSAC is at zero frequency error at the start of the year and
drifts by 10 ppb
over that year, you have an average error of 5 ppb. Keeping things
simple, you get
1/6 second error that year. (5 / 30 = 1/6).
If three years later, the CSAC is at 30 ppb and drifts another 10 ppb
in frequency,
you now are at 35 ppb average frequency error. You will gain / loose
more than a
second in that year.
The real numbers are slightly different. You need to look at when
over the year the
aging happens. A device that ages a lot early on in the year will do
worse than a
device that ages linearly over the year. A device that does all it’s
aging only on the
last day would do better than either of the other cases.
Bottom line:
Your CSAC wrist watch is very much not a millisecond per year sort of device.
Best guess is it is in the 50 to 150 ms per year vicinity in the
first year after calibration.
Based on previous posts, that is in the same vicinity as a WWVB sync’d wrist
watch and not quite as good as an typical Apple Watch.
Bob
Jim:
Likely wait for your data before ordering one. Too bad the chip
price is so high, compared to a few years ago.
Ronald
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.
Bob:
Interesting that drift rate has a stocastic component.
You mentioned adding code. Was that to the software that comes
with the chip?
Ronald
Hi
The CSAC (like any vapor cell standard) has a drift (aging) process. That’s
just the way it works. It is at a much lower rate than a crystal
oscillator, but
it is the same sort of idea. It is one of their basic differences from a Cesium
beam tube.
Can you “estimate” aging in advance? Everything I’ve seen suggests that you
are simply guessing when you do. You will be right in some cases and wrong in
other cases. You can do better estimating warmup drift and short term effects.
Working out what will happen over months (or years) is not very easy.
A very basic example:
I observe a couple of units and they all go positive by 0.8 ppb / mo. I put in
some code to work with that. The unit I happen to have goes 0.8 ppb negative
a month. My code has actually made the unit 2X worse than it would have
been if I just stayed away from it.
Yes there are papers on aging estimation. They mainly focus on coming up
with a “worst case” number. If they guess that 0.8 ppb will go on forever and
it drops off a bit from there, they did ok. For correction purposes …
not so much.
On a practical basis, you will have to dock this beast up with a charger on
a regular basis. A solar powered WWVB watch is not unusual. A solar powered
Apple watch or CSAC watch … not so much. When it goes to the charger,
sync it up with GPS.
Bob
Hi
One way or the other you will need some “smarts” to do aging compensation.
That implies adding a CPU of some sort to the “system” you are building. There
is no built in subsystem on the CSAC that will do any of this for you. You also
need some sort of display for your “wrist watch”. Having the CPU handle that
at the same time makes a lot of sense.
Bob
On Jan 26, 2018, at 2:12 PM, Ronald Held ronaldheld@gmail.com wrote:
Bob:
Interesting that drift rate has a stocastic component.
You mentioned adding code. Was that to the software that comes
with the chip?
Ronald
Hi
The CSAC (like any vapor cell standard) has a drift (aging) process. That’s
just the way it works. It is at a much lower rate than a crystal
oscillator, but
it is the same sort of idea. It is one of their basic differences from a Cesium
beam tube.
Can you “estimate” aging in advance? Everything I’ve seen suggests that you
are simply guessing when you do. You will be right in some cases and wrong in
other cases. You can do better estimating warmup drift and short term effects.
Working out what will happen over months (or years) is not very easy.
A very basic example:
I observe a couple of units and they all go positive by 0.8 ppb / mo. I put in
some code to work with that. The unit I happen to have goes 0.8 ppb negative
a month. My code has actually made the unit 2X worse than it would have
been if I just stayed away from it.
Yes there are papers on aging estimation. They mainly focus on coming up
with a “worst case” number. If they guess that 0.8 ppb will go on forever and
it drops off a bit from there, they did ok. For correction purposes …
not so much.
On a practical basis, you will have to dock this beast up with a charger on
a regular basis. A solar powered WWVB watch is not unusual. A solar powered
Apple watch or CSAC watch … not so much. When it goes to the charger,
sync it up with GPS.
Bob
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.