W
WarrenS
Thu, Feb 9, 2012 12:51 PM
Indeed,
ADEV is for random freq variation not easily measured by other means.
Temperature fluctuations do not cause random freq changes and the
temperature's effect should be removed if one wants accurate long term ADEV
numbers.
Even daily diurnal cycles due to temperature can have major negative effect
on ADEV numbers as low as 2000 to 3000 seconds,
and if there is an Heater or AC cycling, then any ADEV numbers about a few
hundred seconds can be due to TempCoeff, which should not be measured with
ADEV or included in ADEV plots.
This is much the same as a single outlier data point that can screw up the
whole ADEV plot and make it pretty much meaningless and unrepeatable.
Ditto for linear ageing, Should be remove first if one wants true ADEV
plots.
ws
[time-nuts] Low-Cost Rubidium Performance
Bob Camp lists at rtty.us
Thu Feb 9 11:58:27 UTC 2012
Hi
Past 100 seconds I have seen some FE's that look better than your LPRO plot
and some FE's that look worse than your FE plots. Running in a +/- 2C room
apparently is not the best way to operate them for good long tau
performance.
Bob
Indeed,
ADEV is for random freq variation not easily measured by other means.
Temperature fluctuations do not cause random freq changes and the
temperature's effect should be removed if one wants accurate long term ADEV
numbers.
Even daily diurnal cycles due to temperature can have major negative effect
on ADEV numbers as low as 2000 to 3000 seconds,
and if there is an Heater or AC cycling, then any ADEV numbers about a few
hundred seconds can be due to TempCoeff, which should not be measured with
ADEV or included in ADEV plots.
This is much the same as a single outlier data point that can screw up the
whole ADEV plot and make it pretty much meaningless and unrepeatable.
Ditto for linear ageing, Should be remove first if one wants true ADEV
plots.
ws
***************
[time-nuts] Low-Cost Rubidium Performance
Bob Camp lists at rtty.us
Thu Feb 9 11:58:27 UTC 2012
Hi
Past 100 seconds I have seen some FE's that look better than your LPRO plot
and some FE's that look worse than your FE plots. Running in a +/- 2C room
apparently is not the best way to operate them for good long tau
performance.
Bob
JL
Jim Lux
Thu, Feb 9, 2012 3:10 PM
On 2/9/12 4:51 AM, WarrenS wrote:
Indeed,
ADEV is for random freq variation not easily measured by other means.
Temperature fluctuations do not cause random freq changes and the
temperature's effect should be removed if one wants accurate long term
ADEV numbers.
Even daily diurnal cycles due to temperature can have major negative
effect on ADEV numbers as low as 2000 to 3000 seconds,
and if there is an Heater or AC cycling, then any ADEV numbers about a
few hundred seconds can be due to TempCoeff, which should not be
measured with ADEV or included in ADEV plots.
This is much the same as a single outlier data point that can screw up
the whole ADEV plot and make it pretty much meaningless and unrepeatable.
Ditto for linear ageing, Should be remove first if one wants true ADEV
plots.
Interesting point you make here. The rising ADEV at 100-1000 second-ish
tau in a system that should be better is a classic sign (at least around
here) that temperature effects are showing up.
However, how could one remove that effect from the raw data? And isn't
the measurement of the "system", which includes the environmental effects.
I suppose you could run your widget in a temperature controlled chamber,
get those numbers. Then run it in a less controlled benchtop
environment, and get those numbers, and claim that the difference is
environmental.
But at some point, what you're interested is the performance of the
system in the environment in which it will be used. If you need good
ADEV performance at the 1000 second tau, then you need an oven, a vacuum
bottle, or a better design that's less environment sensitive.
(difference between TRL6 and lower, for those into such things)
On 2/9/12 4:51 AM, WarrenS wrote:
> Indeed,
> ADEV is for random freq variation not easily measured by other means.
> Temperature fluctuations do not cause random freq changes and the
> temperature's effect should be removed if one wants accurate long term
> ADEV numbers.
> Even daily diurnal cycles due to temperature can have major negative
> effect on ADEV numbers as low as 2000 to 3000 seconds,
> and if there is an Heater or AC cycling, then any ADEV numbers about a
> few hundred seconds can be due to TempCoeff, which should not be
> measured with ADEV or included in ADEV plots.
> This is much the same as a single outlier data point that can screw up
> the whole ADEV plot and make it pretty much meaningless and unrepeatable.
> Ditto for linear ageing, Should be remove first if one wants true ADEV
> plots.
>
Interesting point you make here. The rising ADEV at 100-1000 second-ish
tau in a system that should be better is a classic sign (at least around
here) that temperature effects are showing up.
However, how could one remove that effect from the raw data? And isn't
the measurement of the "system", which includes the environmental effects.
I suppose you could run your widget in a temperature controlled chamber,
get those numbers. Then run it in a less controlled benchtop
environment, and get those numbers, and claim that the difference is
environmental.
But at some point, what you're interested is the performance of the
system in the environment in which it will be used. If you need good
ADEV performance at the 1000 second tau, then you need an oven, a vacuum
bottle, or a better design that's less environment sensitive.
(difference between TRL6 and lower, for those into such things)
JA
John Ackermann N8UR
Thu, Feb 9, 2012 3:43 PM
On 2/9/2012 7:51 AM, WarrenS wrote:
Indeed,
ADEV is for random freq variation not easily measured by other means.
Temperature fluctuations do not cause random freq changes and the
temperature's effect should be removed if one wants accurate long term
ADEV numbers.
Even daily diurnal cycles due to temperature can have major negative
effect on ADEV numbers as low as 2000 to 3000 seconds,
and if there is an Heater or AC cycling, then any ADEV numbers about a
few hundred seconds can be due to TempCoeff, which should not be
measured with ADEV or included in ADEV plots.
This is much the same as a single outlier data point that can screw up
the whole ADEV plot and make it pretty much meaningless and unrepeatable.
Ditto for linear ageing, Should be remove first if one wants true ADEV
plots.
Hi Warren --
JimL responded before I did, making pretty much the same point -- I
think ADEV is a tool to measure performance in the environment that you
want. If you want to measure the best-possible performance of a device,
you want to control for all external factors. But if you want to see
the real-world performance, you want to measure in the real-world
environment.
FWIW, these tests were done in my basement lab. I don't have my
temperature monitoring stuff set up yet so don't have data, but the
furnace isn't cycling too frequently, and there is relatively little air
flow from the registers into the area (the thermostat is upstairs and we
just bleed a little air into the basement). We don't do a huge amount
of day/night setback at this time of year, so I suspect that the
temperature is remaining stable within 2 or 3 degrees C. I suspect the
seasonal variation is greater than the short term. But on the project
list is getting several temperature sensors installed to feed a data
logger...
On the very long term project list, I'd like to climate control my clock
room to maintain better than 1 degree C, but that one is down the road.
John
On 2/9/2012 7:51 AM, WarrenS wrote:
> Indeed,
> ADEV is for random freq variation not easily measured by other means.
> Temperature fluctuations do not cause random freq changes and the
> temperature's effect should be removed if one wants accurate long term
> ADEV numbers.
> Even daily diurnal cycles due to temperature can have major negative
> effect on ADEV numbers as low as 2000 to 3000 seconds,
> and if there is an Heater or AC cycling, then any ADEV numbers about a
> few hundred seconds can be due to TempCoeff, which should not be
> measured with ADEV or included in ADEV plots.
> This is much the same as a single outlier data point that can screw up
> the whole ADEV plot and make it pretty much meaningless and unrepeatable.
> Ditto for linear ageing, Should be remove first if one wants true ADEV
> plots.
Hi Warren --
JimL responded before I did, making pretty much the same point -- I
think ADEV is a tool to measure performance in the environment that you
want. If you want to measure the best-possible performance of a device,
you want to control for all external factors. But if you want to see
the real-world performance, you want to measure in the real-world
environment.
FWIW, these tests were done in my basement lab. I don't have my
temperature monitoring stuff set up yet so don't have data, but the
furnace isn't cycling too frequently, and there is relatively little air
flow from the registers into the area (the thermostat is upstairs and we
just bleed a little air into the basement). We don't do a huge amount
of day/night setback at this time of year, so I suspect that the
temperature is remaining stable within 2 or 3 degrees C. I suspect the
seasonal variation is greater than the short term. But on the project
list is getting several temperature sensors installed to feed a data
logger...
On the very long term project list, I'd like to climate control my clock
room to maintain better than 1 degree C, but that one is down the road.
John
CA
Chris Albertson
Thu, Feb 9, 2012 4:19 PM
I was just about to write the same thing as below. Yes you can
measure ADEV with the unit in a temperature controlled box but unless
you intend to use the unit in that same box what does the test tell
you? These are spec'd for use in unheated cell towers and the
engineer looked at the worse case and said "good enough for my
intended usage" So if you intend to use this with portable field day
microwave gear then test the unit outdoors while powering with a gas
powered generator. If it will be used as a lab standard, test it
there. Mine is in a box with a temperature controlled fan.
Realistically I expect +-2C swings So I'll measure it inside that
box.
On Thu, Feb 9, 2012 at 7:10 AM, Jim Lux jimlux@earthlink.net wrote:
I suppose you could run your widget in a temperature controlled chamber, get
those numbers. Then run it in a less controlled benchtop environment, and
get those numbers, and claim that the difference is environmental.
But at some point, what you're interested is the performance of the system
in the environment in which it will be used. If you need good ADEV
performance at the 1000 second tau, then you need an oven, a vacuum bottle,
or a better design that's less environment sensitive.
Chris Albertson
Redondo Beach, California
I was just about to write the same thing as below. Yes you can
measure ADEV with the unit in a temperature controlled box but unless
you intend to use the unit in that same box what does the test tell
you? These are spec'd for use in unheated cell towers and the
engineer looked at the worse case and said "good enough for my
intended usage" So if you intend to use this with portable field day
microwave gear then test the unit outdoors while powering with a gas
powered generator. If it will be used as a lab standard, test it
there. Mine is in a box with a temperature controlled fan.
Realistically I expect +-2C swings So I'll measure it inside that
box.
On Thu, Feb 9, 2012 at 7:10 AM, Jim Lux <jimlux@earthlink.net> wrote:
> I suppose you could run your widget in a temperature controlled chamber, get
> those numbers. Then run it in a less controlled benchtop environment, and
> get those numbers, and claim that the difference is environmental.
>
> But at some point, what you're interested is the performance of the system
> in the environment in which it will be used. If you need good ADEV
> performance at the 1000 second tau, then you need an oven, a vacuum bottle,
> or a better design that's less environment sensitive.
Chris Albertson
Redondo Beach, California
MD
Magnus Danielson
Thu, Feb 9, 2012 4:42 PM
On 02/09/2012 04:10 PM, Jim Lux wrote:
On 2/9/12 4:51 AM, WarrenS wrote:
Indeed,
ADEV is for random freq variation not easily measured by other means.
Temperature fluctuations do not cause random freq changes and the
temperature's effect should be removed if one wants accurate long term
ADEV numbers.
Even daily diurnal cycles due to temperature can have major negative
effect on ADEV numbers as low as 2000 to 3000 seconds,
and if there is an Heater or AC cycling, then any ADEV numbers about a
few hundred seconds can be due to TempCoeff, which should not be
measured with ADEV or included in ADEV plots.
This is much the same as a single outlier data point that can screw up
the whole ADEV plot and make it pretty much meaningless and unrepeatable.
Ditto for linear ageing, Should be remove first if one wants true ADEV
plots.
Interesting point you make here. The rising ADEV at 100-1000 second-ish
tau in a system that should be better is a classic sign (at least around
here) that temperature effects are showing up.
I regularly see the building AC at 900-1000 s for instance.
However, how could one remove that effect from the raw data? And isn't
the measurement of the "system", which includes the environmental effects.
ADEV and friends is there to handle random sources, where as this is a
systematic source.
I suppose you could run your widget in a temperature controlled chamber,
get those numbers. Then run it in a less controlled benchtop
environment, and get those numbers, and claim that the difference is
environmental.
But at some point, what you're interested is the performance of the
system in the environment in which it will be used. If you need good
ADEV performance at the 1000 second tau, then you need an oven, a vacuum
bottle, or a better design that's less environment sensitive.
You could also build active systematic effect predictors to lower this
systematic effect.
By doing proper logging of key environmental effects, build a model for
how the dominant variations will systematically affect the signal and
then remove that from the measurements you get a better random jitter
measurement.
Frequency drift of an oscillator is one such systematic effect. If it
where linear, processing it with ADEV would cause a sqrt(2) scale error.
Also, it would not give you a good prediction since usually you follow a
Aln(Bt+1) curve which isn't matching the requirement, so you will only
get first degree compensation of that with HDEV style measures.
Temperature variations is tricky to say the least.
When you have random and systematic effects, separate them and estimate
them separately and then build a combined prediction from these models.
Random jitter and deterministic jitter are two such aspects. Same
applies at longer taus as well.
Cheers,
Magnus
On 02/09/2012 04:10 PM, Jim Lux wrote:
> On 2/9/12 4:51 AM, WarrenS wrote:
>> Indeed,
>> ADEV is for random freq variation not easily measured by other means.
>> Temperature fluctuations do not cause random freq changes and the
>> temperature's effect should be removed if one wants accurate long term
>> ADEV numbers.
>> Even daily diurnal cycles due to temperature can have major negative
>> effect on ADEV numbers as low as 2000 to 3000 seconds,
>> and if there is an Heater or AC cycling, then any ADEV numbers about a
>> few hundred seconds can be due to TempCoeff, which should not be
>> measured with ADEV or included in ADEV plots.
>> This is much the same as a single outlier data point that can screw up
>> the whole ADEV plot and make it pretty much meaningless and unrepeatable.
>> Ditto for linear ageing, Should be remove first if one wants true ADEV
>> plots.
>>
>
>
> Interesting point you make here. The rising ADEV at 100-1000 second-ish
> tau in a system that should be better is a classic sign (at least around
> here) that temperature effects are showing up.
I regularly see the building AC at 900-1000 s for instance.
> However, how could one remove that effect from the raw data? And isn't
> the measurement of the "system", which includes the environmental effects.
ADEV and friends is there to handle random sources, where as this is a
systematic source.
> I suppose you could run your widget in a temperature controlled chamber,
> get those numbers. Then run it in a less controlled benchtop
> environment, and get those numbers, and claim that the difference is
> environmental.
>
> But at some point, what you're interested is the performance of the
> system in the environment in which it will be used. If you need good
> ADEV performance at the 1000 second tau, then you need an oven, a vacuum
> bottle, or a better design that's less environment sensitive.
You could also build active systematic effect predictors to lower this
systematic effect.
By doing proper logging of key environmental effects, build a model for
how the dominant variations will systematically affect the signal and
then remove that from the measurements you get a better random jitter
measurement.
Frequency drift of an oscillator is one such systematic effect. If it
where linear, processing it with ADEV would cause a sqrt(2) scale error.
Also, it would not give you a good prediction since usually you follow a
A*ln(B*t+1) curve which isn't matching the requirement, so you will only
get first degree compensation of that with HDEV style measures.
Temperature variations is tricky to say the least.
When you have random and systematic effects, separate them and estimate
them separately and then build a combined prediction from these models.
Random jitter and deterministic jitter are two such aspects. Same
applies at longer taus as well.
Cheers,
Magnus
JM
John Miles
Thu, Feb 9, 2012 11:37 PM
Indeed, ADEV is for random freq variation not easily measured by other
means.
Well, no, ADEV is the two-sample deviation of fractional frequency
differences over time. That's really all you can say about it. There's not
really any such thing as "true ADEV" -- a measurement either meets the
mathematical criteria for Allan deviation, or it doesn't.
Temperature fluctuations do not cause random freq changes and the
temperature's effect should be removed if one wants accurate long term
ADEV numbers.
No, accurate ADEV numbers are whatever you see on an accurate ADEV plot. :)
If I measure two sources in the same environment and I see HVAC ripple on
one ADEV trace but not on the other, then that may be useful information, or
even the only information I care about. (Of course, it's only useful if
the bin density is high enough to show the effect in question, but that's
not the fault of the ADEV metric itself.)
If you don't want to observe the effect of temperature fluctuations on your
DUT, random or otherwise, the correct solution is not to use a different
metric or to tweak the data, but to shield the DUT against the temperature
variations in question.
Even daily diurnal cycles due to temperature can have major negative
on ADEV numbers as low as 2000 to 3000 seconds,
Your bin density may be insufficient in that case. ADEV is not unlike an
FFT in that regard -- the denser the bins, the higher the resolution,
subject to limitations imposed by the window transfer function. (Enrico
Rubiola has suggested that we should have been using FFT-like measures for
long term stability all along, instead of ADEV.)
It's true that the ADEV function is not all that sharp, but you shouldn't
ordinarily see effects removed from their causes by a 40:1 tau ratio. IMHO,
if you are seeing significant degradation at the 2000-second level caused by
diurnal cycles at the 12-hour level, something may be wrong.
Outliers are another matter, due to the infinite "ringing" that a step
function causes. They should be removed from ADEV and considered as a
separate source of error. Transients cause some pretty horrible effects in
FFTs as well, regardless of the window characteristics. Offhand, I can't
think of any simple frequency-stability metrics that are good at ignoring
outliers, and I'm not sure it'd be a good thing if we were to invent one.
and if there is an Heater or AC cycling, then any ADEV numbers about a few
hundred seconds can be due to TempCoeff, which should not be measured
with ADEV or included in ADEV plots.
Again, fractional frequency differences are fractional frequency
differences. ADEV will show temperature effects, as will an FFT or most
other metrics worth using. If you don't want to see these effects, you need
to take the appropriate measures to fix the environment, the DUT, the
instrumentation, or all of the above.
This is much the same as a single outlier data point that can screw up the
whole ADEV plot and make it pretty much meaningless and unrepeatable.
Ditto for linear ageing, Should be remove first if one wants true ADEV
plots.
Linear drift is a good thing to take out... if you explicitly want to
exclude it from your observation of fractional frequency-difference
statistics. Maybe you consider drift or aging to be a valid part of the
statistics you're collecting. If so, leave it in. Maybe you plan to
discipline the DUT in a way that will remove drift and aging. If so, remove
it. You're going to get a "valid" measurement of ADEV either way... but
determining whether ADEV is really the best metric to use, and interpreting
it in light of your application, is always up to you.
-- john
> Indeed, ADEV is for random freq variation not easily measured by other
means.
Well, no, ADEV is the two-sample deviation of fractional frequency
differences over time. That's really all you can say about it. There's not
really any such thing as "true ADEV" -- a measurement either meets the
mathematical criteria for Allan deviation, or it doesn't.
> Temperature fluctuations do not cause random freq changes and the
> temperature's effect should be removed if one wants accurate long term
> ADEV numbers.
No, accurate ADEV numbers are whatever you see on an accurate ADEV plot. :)
If I measure two sources in the same environment and I see HVAC ripple on
one ADEV trace but not on the other, then that may be useful information, or
even the only information I care about. (Of course, it's only useful if
the bin density is high enough to show the effect in question, but that's
not the fault of the ADEV metric itself.)
If you don't want to observe the effect of temperature fluctuations on your
DUT, random or otherwise, the correct solution is not to use a different
metric or to tweak the data, but to shield the DUT against the temperature
variations in question.
> Even daily diurnal cycles due to temperature can have major negative
effect
> on ADEV numbers as low as 2000 to 3000 seconds,
Your bin density may be insufficient in that case. ADEV is not unlike an
FFT in that regard -- the denser the bins, the higher the resolution,
subject to limitations imposed by the window transfer function. (Enrico
Rubiola has suggested that we should have been using FFT-like measures for
long term stability all along, instead of ADEV.)
It's true that the ADEV function is not all that sharp, but you shouldn't
ordinarily see effects removed from their causes by a 40:1 tau ratio. IMHO,
if you are seeing significant degradation at the 2000-second level caused by
diurnal cycles at the 12-hour level, something may be wrong.
Outliers are another matter, due to the infinite "ringing" that a step
function causes. They should be removed from ADEV and considered as a
separate source of error. Transients cause some pretty horrible effects in
FFTs as well, regardless of the window characteristics. Offhand, I can't
think of any simple frequency-stability metrics that are good at ignoring
outliers, and I'm not sure it'd be a good thing if we were to invent one.
> and if there is an Heater or AC cycling, then any ADEV numbers about a few
> hundred seconds can be due to TempCoeff, which should not be measured
> with ADEV or included in ADEV plots.
Again, fractional frequency differences are fractional frequency
differences. ADEV will show temperature effects, as will an FFT or most
other metrics worth using. If you don't want to see these effects, you need
to take the appropriate measures to fix the environment, the DUT, the
instrumentation, or all of the above.
> This is much the same as a single outlier data point that can screw up the
> whole ADEV plot and make it pretty much meaningless and unrepeatable.
> Ditto for linear ageing, Should be remove first if one wants true ADEV
> plots.
Linear drift is a good thing to take out... *if* you explicitly want to
exclude it from your observation of fractional frequency-difference
statistics. Maybe you consider drift or aging to be a valid part of the
statistics you're collecting. If so, leave it in. Maybe you plan to
discipline the DUT in a way that will remove drift and aging. If so, remove
it. You're going to get a "valid" measurement of ADEV either way... but
determining whether ADEV is really the best metric to use, and interpreting
it in light of your application, is always up to you.
-- john
JL
Jim Lux
Thu, Feb 9, 2012 11:51 PM
On 2/9/12 8:42 AM, Magnus Danielson wrote:
On 02/09/2012 04:10 PM, Jim Lux wrote:
Interesting point you make here. The rising ADEV at 100-1000 second-ish
tau in a system that should be better is a classic sign (at least around
here) that temperature effects are showing up.
I regularly see the building AC at 900-1000 s for instance.
However, how could one remove that effect from the raw data? And isn't
the measurement of the "system", which includes the environmental
effects.
ADEV and friends is there to handle random sources, where as this is a
systematic source.
But I would contend that unless you can externally measure that
disturbance and remove it, it's a fundamental part of the frequency
variation. Since you don't have control over it, or necessarily have
accurate information about it, it's not something that could be
"calibrated out".
Say, for instance, I had an oscillator that was mounted in such a way
that it rotated slowly once every hour. There would be a periodic
variation in frequency, which could be accurately modeled and removed,
so that wouldn't necessarily count in ADEV.
But temperature is random (although band limited), and so, for
measurements over the 1000 second time scale, it's impossible to tell if
the change was due to temperature or due to the underlying oscillator.
I suppose you could run your widget in a temperature controlled chamber,
get those numbers. Then run it in a less controlled benchtop
environment, and get those numbers, and claim that the difference is
environmental.
But at some point, what you're interested is the performance of the
system in the environment in which it will be used. If you need good
ADEV performance at the 1000 second tau, then you need an oven, a vacuum
bottle, or a better design that's less environment sensitive.
You could also build active systematic effect predictors to lower this
systematic effect.
Yes. That's basically no different than controlling the environment.
If the transfer function of environment to output is well known, and you
know the environment, you could legitimately "remove" it from the
measured output.
By doing proper logging of key environmental effects, build a model for
how the dominant variations will systematically affect the signal and
then remove that from the measurements you get a better random jitter
measurement.
Ah, but there's the rub. Can you actually do that with acceptable
performance?
I know that you can measure the temperature of a crystal and fairly
accurately calibrate out the frequency change due to temperature (to the
point where frequency can be used to measure temperature). So now, your
ADEV on the "calibrated" output would depend on the temperature
measurement accuracy. Essentially what you have done is reduced the
tempco of the system.
Frequency drift of an oscillator is one such systematic effect. If it
where linear, processing it with ADEV would cause a sqrt(2) scale error.
Also, it would not give you a good prediction since usually you follow a
Aln(Bt+1) curve which isn't matching the requirement, so you will only
get first degree compensation of that with HDEV style measures.
Yes, and I think that for variations that are easily and accurately
modeled this is appropriate, however, your next sentence says it all.
Temperature variations is tricky to say the least.
When you have random and systematic effects, separate them and estimate
them separately and then build a combined prediction from these models.
Random jitter and deterministic jitter are two such aspects. Same
applies at longer taus as well.
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.
On 2/9/12 8:42 AM, Magnus Danielson wrote:
> On 02/09/2012 04:10 PM, Jim Lux wrote:
>>
>> Interesting point you make here. The rising ADEV at 100-1000 second-ish
>> tau in a system that should be better is a classic sign (at least around
>> here) that temperature effects are showing up.
>
> I regularly see the building AC at 900-1000 s for instance.
>
>> However, how could one remove that effect from the raw data? And isn't
>> the measurement of the "system", which includes the environmental
>> effects.
>
> ADEV and friends is there to handle random sources, where as this is a
> systematic source.
But I would contend that unless you can externally measure that
disturbance and remove it, it's a fundamental part of the frequency
variation. Since you don't have control over it, or necessarily have
accurate information about it, it's not something that could be
"calibrated out".
Say, for instance, I had an oscillator that was mounted in such a way
that it rotated slowly once every hour. There would be a periodic
variation in frequency, which could be accurately modeled and removed,
so that wouldn't necessarily count in ADEV.
But temperature is random (although band limited), and so, for
measurements over the 1000 second time scale, it's impossible to tell if
the change was due to temperature or due to the underlying oscillator.
>
>> I suppose you could run your widget in a temperature controlled chamber,
>> get those numbers. Then run it in a less controlled benchtop
>> environment, and get those numbers, and claim that the difference is
>> environmental.
>>
>> But at some point, what you're interested is the performance of the
>> system in the environment in which it will be used. If you need good
>> ADEV performance at the 1000 second tau, then you need an oven, a vacuum
>> bottle, or a better design that's less environment sensitive.
>
> You could also build active systematic effect predictors to lower this
> systematic effect.
Yes. That's basically no different than controlling the environment.
If the transfer function of environment to output is well known, and you
know the environment, you could legitimately "remove" it from the
measured output.
>
> By doing proper logging of key environmental effects, build a model for
> how the dominant variations will systematically affect the signal and
> then remove that from the measurements you get a better random jitter
> measurement.
Ah, but there's the rub. Can you actually do that with acceptable
performance?
I know that you can measure the temperature of a crystal and fairly
accurately calibrate out the frequency change due to temperature (to the
point where frequency can be used to measure temperature). So now, your
ADEV on the "calibrated" output would depend on the temperature
measurement accuracy. Essentially what you have done is reduced the
tempco of the system.
>
> Frequency drift of an oscillator is one such systematic effect. If it
> where linear, processing it with ADEV would cause a sqrt(2) scale error.
> Also, it would not give you a good prediction since usually you follow a
> A*ln(B*t+1) curve which isn't matching the requirement, so you will only
> get first degree compensation of that with HDEV style measures.
Yes, and I think that for variations that are easily and *accurately*
modeled this is appropriate, however, your next sentence says it all.
>
> Temperature variations is tricky to say the least.
>
> When you have random and systematic effects, separate them and estimate
> them separately and then build a combined prediction from these models.
>
> Random jitter and deterministic jitter are two such aspects. Same
> applies at longer taus as well.
>
>
> 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.
>
MD
Magnus Danielson
Fri, Feb 10, 2012 1:08 AM
John,
On 02/10/2012 12:37 AM, John Miles wrote:
Indeed, ADEV is for random freq variation not easily measured by other
means.
Well, no, ADEV is the two-sample deviation of fractional frequency
differences over time. That's really all you can say about it. There's not
really any such thing as "true ADEV" -- a measurement either meets the
mathematical criteria for Allan deviation, or it doesn't.
Well, does the ADEV of a systematic temperature dependence or drift
property give you any meaningful values, or is ADEV the only answer to
your question?
My point is that ADEV and TDEV just doesn't give very useful information
on your oscillators behaviours for systematic effects, but ADEV and TDEV
can give you more information on your random noise sources contribution
if you remove the parts being obscured by the systematic effects.
The various systematic effects is best handled with other tools, and
then when we have established meaningful values of a suitable model, we
can then apply the model to the stress-cases of our choosing and get
meaningful results out of that, which is usually what you want to do.
Temperature fluctuations do not cause random freq changes and the
temperature's effect should be removed if one wants accurate long term
ADEV numbers.
No, accurate ADEV numbers are whatever you see on an accurate ADEV plot. :)
Whatever an accurate ADEV plot is, means and indeed is attainable at
all. There is no true accurate ADEV plot, it's full of compromises,
side-effects and stuff.
If I measure two sources in the same environment and I see HVAC ripple on
one ADEV trace but not on the other, then that may be useful information, or
even the only information I care about. (Of course, it's only useful if
the bin density is high enough to show the effect in question, but that's
not the fault of the ADEV metric itself.)
ADEV isn't the only tool to see that, it may not even be the best tool
to analyze it.
If you don't want to observe the effect of temperature fluctuations on your
DUT, random or otherwise, the correct solution is not to use a different
metric or to tweak the data, but to shield the DUT against the temperature
variations in question.
Easier said than done. ADEV just isn't the one and only tool to use.
Even daily diurnal cycles due to temperature can have major negative
on ADEV numbers as low as 2000 to 3000 seconds,
Your bin density may be insufficient in that case. ADEV is not unlike an
FFT in that regard -- the denser the bins, the higher the resolution,
subject to limitations imposed by the window transfer function. (Enrico
Rubiola has suggested that we should have been using FFT-like measures for
long term stability all along, instead of ADEV.)
There are also a huge number of ADEV estimator methods, crunching out
different precision in their estimation of ADEV. FFT is one possible
method alongside some processing that has been proposed.
It's true that the ADEV function is not all that sharp, but you shouldn't
ordinarily see effects removed from their causes by a 40:1 tau ratio. IMHO,
if you are seeing significant degradation at the 2000-second level caused by
diurnal cycles at the 12-hour level, something may be wrong.
Outliers are another matter, due to the infinite "ringing" that a step
function causes. They should be removed from ADEV and considered as a
separate source of error. Transients cause some pretty horrible effects in
FFTs as well, regardless of the window characteristics. Offhand, I can't
think of any simple frequency-stability metrics that are good at ignoring
outliers, and I'm not sure it'd be a good thing if we were to invent one.
What is a step or spike in time becomes a resonance in the frequency domain.
What is a step or spike in the frequency domain becomes a resonance in
the time domain.
ADEV is a kind of frequency domain tool.
and if there is an Heater or AC cycling, then any ADEV numbers about a few
hundred seconds can be due to TempCoeff, which should not be measured
with ADEV or included in ADEV plots.
Again, fractional frequency differences are fractional frequency
differences. ADEV will show temperature effects, as will an FFT or most
other metrics worth using. If you don't want to see these effects, you need
to take the appropriate measures to fix the environment, the DUT, the
instrumentation, or all of the above.
This is much the same as a single outlier data point that can screw up the
whole ADEV plot and make it pretty much meaningless and unrepeatable.
Ditto for linear ageing, Should be remove first if one wants true ADEV
plots.
Linear drift is a good thing to take out... if you explicitly want to
exclude it from your observation of fractional frequency-difference
statistics. Maybe you consider drift or aging to be a valid part of the
statistics you're collecting. If so, leave it in. Maybe you plan to
discipline the DUT in a way that will remove drift and aging. If so, remove
it. You're going to get a "valid" measurement of ADEV either way... but
determining whether ADEV is really the best metric to use, and interpreting
it in light of your application, is always up to you.
Linear drift will be scaled correctly. When setting confidence bounds
the systematic effect confidence bounds will not match up with the
confidence bounds of the random statistics, so you will get fooled,
besides the bias aspect.
ADEV as a tool was meant to handle random noises, but not the systematic
noise. When doing engineering estimates ADEV or TDEV with suitable
scale-ups then adds to the systematic effects. They need to be treated
as time treats them in different ways.
It's clear we don't see our ADEVs the same way.
Cheers,
Magnus
John,
On 02/10/2012 12:37 AM, John Miles wrote:
>> Indeed, ADEV is for random freq variation not easily measured by other
> means.
>
> Well, no, ADEV is the two-sample deviation of fractional frequency
> differences over time. That's really all you can say about it. There's not
> really any such thing as "true ADEV" -- a measurement either meets the
> mathematical criteria for Allan deviation, or it doesn't.
Well, does the ADEV of a systematic temperature dependence or drift
property give you any meaningful values, or is ADEV the only answer to
your question?
My point is that ADEV and TDEV just doesn't give very useful information
on your oscillators behaviours for systematic effects, but ADEV and TDEV
can give you more information on your random noise sources contribution
if you remove the parts being obscured by the systematic effects.
The various systematic effects is best handled with other tools, and
then when we have established meaningful values of a suitable model, we
can then apply the model to the stress-cases of our choosing and get
meaningful results out of that, which is usually what you want to do.
>> Temperature fluctuations do not cause random freq changes and the
>> temperature's effect should be removed if one wants accurate long term
>> ADEV numbers.
>
> No, accurate ADEV numbers are whatever you see on an accurate ADEV plot. :)
Whatever an accurate ADEV plot is, means and indeed is attainable at
all. There is no true accurate ADEV plot, it's full of compromises,
side-effects and stuff.
> If I measure two sources in the same environment and I see HVAC ripple on
> one ADEV trace but not on the other, then that may be useful information, or
> even the only information I care about. (Of course, it's only useful if
> the bin density is high enough to show the effect in question, but that's
> not the fault of the ADEV metric itself.)
ADEV isn't the only tool to see that, it may not even be the best tool
to analyze it.
> If you don't want to observe the effect of temperature fluctuations on your
> DUT, random or otherwise, the correct solution is not to use a different
> metric or to tweak the data, but to shield the DUT against the temperature
> variations in question.
Easier said than done. ADEV just isn't the one and only tool to use.
>> Even daily diurnal cycles due to temperature can have major negative
> effect
>> on ADEV numbers as low as 2000 to 3000 seconds,
>
> Your bin density may be insufficient in that case. ADEV is not unlike an
> FFT in that regard -- the denser the bins, the higher the resolution,
> subject to limitations imposed by the window transfer function. (Enrico
> Rubiola has suggested that we should have been using FFT-like measures for
> long term stability all along, instead of ADEV.)
There are also a huge number of ADEV estimator methods, crunching out
different precision in their estimation of ADEV. FFT is one possible
method alongside some processing that has been proposed.
> It's true that the ADEV function is not all that sharp, but you shouldn't
> ordinarily see effects removed from their causes by a 40:1 tau ratio. IMHO,
> if you are seeing significant degradation at the 2000-second level caused by
> diurnal cycles at the 12-hour level, something may be wrong.
>
> Outliers are another matter, due to the infinite "ringing" that a step
> function causes. They should be removed from ADEV and considered as a
> separate source of error. Transients cause some pretty horrible effects in
> FFTs as well, regardless of the window characteristics. Offhand, I can't
> think of any simple frequency-stability metrics that are good at ignoring
> outliers, and I'm not sure it'd be a good thing if we were to invent one.
What is a step or spike in time becomes a resonance in the frequency domain.
What is a step or spike in the frequency domain becomes a resonance in
the time domain.
ADEV is a kind of frequency domain tool.
>> and if there is an Heater or AC cycling, then any ADEV numbers about a few
>> hundred seconds can be due to TempCoeff, which should not be measured
>> with ADEV or included in ADEV plots.
>
> Again, fractional frequency differences are fractional frequency
> differences. ADEV will show temperature effects, as will an FFT or most
> other metrics worth using. If you don't want to see these effects, you need
> to take the appropriate measures to fix the environment, the DUT, the
> instrumentation, or all of the above.
>
>> This is much the same as a single outlier data point that can screw up the
>> whole ADEV plot and make it pretty much meaningless and unrepeatable.
>> Ditto for linear ageing, Should be remove first if one wants true ADEV
>> plots.
>
> Linear drift is a good thing to take out... *if* you explicitly want to
> exclude it from your observation of fractional frequency-difference
> statistics. Maybe you consider drift or aging to be a valid part of the
> statistics you're collecting. If so, leave it in. Maybe you plan to
> discipline the DUT in a way that will remove drift and aging. If so, remove
> it. You're going to get a "valid" measurement of ADEV either way... but
> determining whether ADEV is really the best metric to use, and interpreting
> it in light of your application, is always up to you.
Linear drift will be scaled correctly. When setting confidence bounds
the systematic effect confidence bounds will not match up with the
confidence bounds of the random statistics, so you will get fooled,
besides the bias aspect.
ADEV as a tool was meant to handle random noises, but not the systematic
noise. When doing engineering estimates ADEV or TDEV with suitable
scale-ups then adds to the systematic effects. They need to be treated
as time treats them in different ways.
It's clear we don't see our ADEVs the same way.
Cheers,
Magnus
BC
Bob Camp
Fri, Feb 10, 2012 1:31 AM
Hi
To add another wrinkle to this.
Correcting ADEV for systematic errors and then not mentioning you did so strikes is something I find a bit problematic. If say you decide to take out 13th order drift, you should say you did so. The discussion of what to correct and how is older than ADEV. Since there is no "standard" set of corrections, one should be clear about what was done.
Bob
On Feb 9, 2012, at 8:08 PM, Magnus Danielson magnus@rubidium.dyndns.org wrote:
John,
On 02/10/2012 12:37 AM, John Miles wrote:
Indeed, ADEV is for random freq variation not easily measured by other
means.
Well, no, ADEV is the two-sample deviation of fractional frequency
differences over time. That's really all you can say about it. There's not
really any such thing as "true ADEV" -- a measurement either meets the
mathematical criteria for Allan deviation, or it doesn't.
Well, does the ADEV of a systematic temperature dependence or drift property give you any meaningful values, or is ADEV the only answer to your question?
My point is that ADEV and TDEV just doesn't give very useful information on your oscillators behaviours for systematic effects, but ADEV and TDEV can give you more information on your random noise sources contribution if you remove the parts being obscured by the systematic effects.
The various systematic effects is best handled with other tools, and then when we have established meaningful values of a suitable model, we can then apply the model to the stress-cases of our choosing and get meaningful results out of that, which is usually what you want to do.
Temperature fluctuations do not cause random freq changes and the
temperature's effect should be removed if one wants accurate long term
ADEV numbers.
No, accurate ADEV numbers are whatever you see on an accurate ADEV plot. :)
Whatever an accurate ADEV plot is, means and indeed is attainable at all. There is no true accurate ADEV plot, it's full of compromises, side-effects and stuff.
If I measure two sources in the same environment and I see HVAC ripple on
one ADEV trace but not on the other, then that may be useful information, or
even the only information I care about. (Of course, it's only useful if
the bin density is high enough to show the effect in question, but that's
not the fault of the ADEV metric itself.)
ADEV isn't the only tool to see that, it may not even be the best tool to analyze it.
If you don't want to observe the effect of temperature fluctuations on your
DUT, random or otherwise, the correct solution is not to use a different
metric or to tweak the data, but to shield the DUT against the temperature
variations in question.
Easier said than done. ADEV just isn't the one and only tool to use.
Even daily diurnal cycles due to temperature can have major negative
on ADEV numbers as low as 2000 to 3000 seconds,
Your bin density may be insufficient in that case. ADEV is not unlike an
FFT in that regard -- the denser the bins, the higher the resolution,
subject to limitations imposed by the window transfer function. (Enrico
Rubiola has suggested that we should have been using FFT-like measures for
long term stability all along, instead of ADEV.)
There are also a huge number of ADEV estimator methods, crunching out different precision in their estimation of ADEV. FFT is one possible method alongside some processing that has been proposed.
It's true that the ADEV function is not all that sharp, but you shouldn't
ordinarily see effects removed from their causes by a 40:1 tau ratio. IMHO,
if you are seeing significant degradation at the 2000-second level caused by
diurnal cycles at the 12-hour level, something may be wrong.
Outliers are another matter, due to the infinite "ringing" that a step
function causes. They should be removed from ADEV and considered as a
separate source of error. Transients cause some pretty horrible effects in
FFTs as well, regardless of the window characteristics. Offhand, I can't
think of any simple frequency-stability metrics that are good at ignoring
outliers, and I'm not sure it'd be a good thing if we were to invent one.
What is a step or spike in time becomes a resonance in the frequency domain.
What is a step or spike in the frequency domain becomes a resonance in the time domain.
ADEV is a kind of frequency domain tool.
and if there is an Heater or AC cycling, then any ADEV numbers about a few
hundred seconds can be due to TempCoeff, which should not be measured
with ADEV or included in ADEV plots.
Again, fractional frequency differences are fractional frequency
differences. ADEV will show temperature effects, as will an FFT or most
other metrics worth using. If you don't want to see these effects, you need
to take the appropriate measures to fix the environment, the DUT, the
instrumentation, or all of the above.
This is much the same as a single outlier data point that can screw up the
whole ADEV plot and make it pretty much meaningless and unrepeatable.
Ditto for linear ageing, Should be remove first if one wants true ADEV
plots.
Linear drift is a good thing to take out... if you explicitly want to
exclude it from your observation of fractional frequency-difference
statistics. Maybe you consider drift or aging to be a valid part of the
statistics you're collecting. If so, leave it in. Maybe you plan to
discipline the DUT in a way that will remove drift and aging. If so, remove
it. You're going to get a "valid" measurement of ADEV either way... but
determining whether ADEV is really the best metric to use, and interpreting
it in light of your application, is always up to you.
Linear drift will be scaled correctly. When setting confidence bounds the systematic effect confidence bounds will not match up with the confidence bounds of the random statistics, so you will get fooled, besides the bias aspect.
ADEV as a tool was meant to handle random noises, but not the systematic noise. When doing engineering estimates ADEV or TDEV with suitable scale-ups then adds to the systematic effects. They need to be treated as time treats them in different ways.
It's clear we don't see our ADEVs the same way.
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.
Hi
To add another wrinkle to this.
Correcting ADEV for systematic errors and then not mentioning you did so strikes is something I find a bit problematic. If say you decide to take out 13th order drift, you should say you did so. The discussion of what to correct and how is older than ADEV. Since there is no "standard" set of corrections, one should be clear about what was done.
Bob
On Feb 9, 2012, at 8:08 PM, Magnus Danielson <magnus@rubidium.dyndns.org> wrote:
> John,
>
> On 02/10/2012 12:37 AM, John Miles wrote:
>>> Indeed, ADEV is for random freq variation not easily measured by other
>> means.
>>
>> Well, no, ADEV is the two-sample deviation of fractional frequency
>> differences over time. That's really all you can say about it. There's not
>> really any such thing as "true ADEV" -- a measurement either meets the
>> mathematical criteria for Allan deviation, or it doesn't.
>
> Well, does the ADEV of a systematic temperature dependence or drift property give you any meaningful values, or is ADEV the only answer to your question?
>
> My point is that ADEV and TDEV just doesn't give very useful information on your oscillators behaviours for systematic effects, but ADEV and TDEV can give you more information on your random noise sources contribution if you remove the parts being obscured by the systematic effects.
>
> The various systematic effects is best handled with other tools, and then when we have established meaningful values of a suitable model, we can then apply the model to the stress-cases of our choosing and get meaningful results out of that, which is usually what you want to do.
>
>>> Temperature fluctuations do not cause random freq changes and the
>>> temperature's effect should be removed if one wants accurate long term
>>> ADEV numbers.
>>
>> No, accurate ADEV numbers are whatever you see on an accurate ADEV plot. :)
>
> Whatever an accurate ADEV plot is, means and indeed is attainable at all. There is no true accurate ADEV plot, it's full of compromises, side-effects and stuff.
>
>> If I measure two sources in the same environment and I see HVAC ripple on
>> one ADEV trace but not on the other, then that may be useful information, or
>> even the only information I care about. (Of course, it's only useful if
>> the bin density is high enough to show the effect in question, but that's
>> not the fault of the ADEV metric itself.)
>
> ADEV isn't the only tool to see that, it may not even be the best tool to analyze it.
>
>> If you don't want to observe the effect of temperature fluctuations on your
>> DUT, random or otherwise, the correct solution is not to use a different
>> metric or to tweak the data, but to shield the DUT against the temperature
>> variations in question.
>
> Easier said than done. ADEV just isn't the one and only tool to use.
>
>>> Even daily diurnal cycles due to temperature can have major negative
>> effect
>>> on ADEV numbers as low as 2000 to 3000 seconds,
>>
>> Your bin density may be insufficient in that case. ADEV is not unlike an
>> FFT in that regard -- the denser the bins, the higher the resolution,
>> subject to limitations imposed by the window transfer function. (Enrico
>> Rubiola has suggested that we should have been using FFT-like measures for
>> long term stability all along, instead of ADEV.)
>
> There are also a huge number of ADEV estimator methods, crunching out different precision in their estimation of ADEV. FFT is one possible method alongside some processing that has been proposed.
>
>> It's true that the ADEV function is not all that sharp, but you shouldn't
>> ordinarily see effects removed from their causes by a 40:1 tau ratio. IMHO,
>> if you are seeing significant degradation at the 2000-second level caused by
>> diurnal cycles at the 12-hour level, something may be wrong.
>>
>> Outliers are another matter, due to the infinite "ringing" that a step
>> function causes. They should be removed from ADEV and considered as a
>> separate source of error. Transients cause some pretty horrible effects in
>> FFTs as well, regardless of the window characteristics. Offhand, I can't
>> think of any simple frequency-stability metrics that are good at ignoring
>> outliers, and I'm not sure it'd be a good thing if we were to invent one.
>
> What is a step or spike in time becomes a resonance in the frequency domain.
> What is a step or spike in the frequency domain becomes a resonance in the time domain.
>
> ADEV is a kind of frequency domain tool.
>
>>> and if there is an Heater or AC cycling, then any ADEV numbers about a few
>>> hundred seconds can be due to TempCoeff, which should not be measured
>>> with ADEV or included in ADEV plots.
>>
>> Again, fractional frequency differences are fractional frequency
>> differences. ADEV will show temperature effects, as will an FFT or most
>> other metrics worth using. If you don't want to see these effects, you need
>> to take the appropriate measures to fix the environment, the DUT, the
>> instrumentation, or all of the above.
>>
>>> This is much the same as a single outlier data point that can screw up the
>>> whole ADEV plot and make it pretty much meaningless and unrepeatable.
>>> Ditto for linear ageing, Should be remove first if one wants true ADEV
>>> plots.
>>
>> Linear drift is a good thing to take out... *if* you explicitly want to
>> exclude it from your observation of fractional frequency-difference
>> statistics. Maybe you consider drift or aging to be a valid part of the
>> statistics you're collecting. If so, leave it in. Maybe you plan to
>> discipline the DUT in a way that will remove drift and aging. If so, remove
>> it. You're going to get a "valid" measurement of ADEV either way... but
>> determining whether ADEV is really the best metric to use, and interpreting
>> it in light of your application, is always up to you.
>
> Linear drift will be scaled correctly. When setting confidence bounds the systematic effect confidence bounds will not match up with the confidence bounds of the random statistics, so you will get fooled, besides the bias aspect.
>
> ADEV as a tool was meant to handle random noises, but not the systematic noise. When doing engineering estimates ADEV or TDEV with suitable scale-ups then adds to the systematic effects. They need to be treated as time treats them in different ways.
>
> It's clear we don't see our ADEVs the same way.
>
> 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.
MD
Magnus Danielson
Fri, Feb 10, 2012 1:35 AM
On 02/10/2012 12:51 AM, Jim Lux wrote:
On 2/9/12 8:42 AM, Magnus Danielson wrote:
On 02/09/2012 04:10 PM, Jim Lux wrote:
Interesting point you make here. The rising ADEV at 100-1000 second-ish
tau in a system that should be better is a classic sign (at least around
here) that temperature effects are showing up.
I regularly see the building AC at 900-1000 s for instance.
However, how could one remove that effect from the raw data? And isn't
the measurement of the "system", which includes the environmental
effects.
ADEV and friends is there to handle random sources, where as this is a
systematic source.
But I would contend that unless you can externally measure that
disturbance and remove it, it's a fundamental part of the frequency
variation. Since you don't have control over it, or necessarily have
accurate information about it, it's not something that could be
"calibrated out".
It's a huge difference what comes in the device and what environment we
put it in. A temperature dependence, is a systematic effect on the device.
Say, for instance, I had an oscillator that was mounted in such a way
that it rotated slowly once every hour. There would be a periodic
variation in frequency, which could be accurately modeled and removed,
so that wouldn't necessarily count in ADEV.
That is indeed a systematic effect in that environment of that device.
Once you have characterized the effect, you can then predict the
behaviour to some precision if you now turns it quicker, more often or
not at all. The random effects would then be re-introduced on that
predicted systematic effect and you would see the aggregate effect.
But temperature is random (although band limited), and so, for
measurements over the 1000 second time scale, it's impossible to tell if
the change was due to temperature or due to the underlying oscillator.
Temperature is not necessary random, I often find systematic albeit
somewhat unregular patterns in my temperature readings and also
frequency or phase plots when I measure longer times on the less than
ideal oscillators. They dominate over any noise contributions, and for
those taus I care less about the noise and more about the systematic
effects. This is one of the reasons we use MTIE in addition to TDEV for
telecommunication measurements, it makes good engineering sense to use
it instead.
A couple of long term systematic effects on temperature relates to the
sun. The dinural wander, i.e. the 24 h period wander is due to the
rotation of earth and the effect of the sun. This is not a random, but
rather a very regular systematic effect.
Another similar effect is the season effect, over a year the temperature
shifts along with the earths axis orientation in relationship to the
sun. This is also a systematic effect.
As you put equipment into a building or other temperature controlled
environment, the bang-bang regulation method is often seen, and the
cycling of temperature up and down is not very steady but still regular
enough and again a systematic effect on the oscillator.
If temperature where random, we would see -80 C and +100 C more often in
our back yard than we do.
To some degree we can control temperature, we can predict temperature
and deal with it. We can handle it as an engineering concept and do
steering loops etc. It's pretty systematic to me.
I suppose you could run your widget in a temperature controlled chamber,
get those numbers. Then run it in a less controlled benchtop
environment, and get those numbers, and claim that the difference is
environmental.
But at some point, what you're interested is the performance of the
system in the environment in which it will be used. If you need good
ADEV performance at the 1000 second tau, then you need an oven, a vacuum
bottle, or a better design that's less environment sensitive.
You could also build active systematic effect predictors to lower this
systematic effect.
Yes. That's basically no different than controlling the environment. If
the transfer function of environment to output is well known, and you
know the environment, you could legitimately "remove" it from the
measured output.
Model wise it's not a huge difference, but practically it may be.
By doing proper logging of key environmental effects, build a model for
how the dominant variations will systematically affect the signal and
then remove that from the measurements you get a better random jitter
measurement.
Ah, but there's the rub. Can you actually do that with acceptable
performance?
I know that you can measure the temperature of a crystal and fairly
accurately calibrate out the frequency change due to temperature (to the
point where frequency can be used to measure temperature). So now, your
ADEV on the "calibrated" output would depend on the temperature
measurement accuracy. Essentially what you have done is reduced the
tempco of the system.
You never get your model and measurement perfect. Not even your ADEV.
You may get good enough confidence for what you need to do.
I trust that you ask for a certain temperature-spec on your oscillators,
that you is expected to make the environment suitable enough etc.
Really, what I say is that trying to lump everything into a single
number is just not a very useful thing when doing engineering,
especially with these devices we use called crystal oscillators. There
are many subtleties which one needs to learn. Tossing everything into a
single ADEV diagram isn't going to help the learning curve.
Don't get me wrong, ADEV is a great tool, it's just not a great tool for
all the effects we see for an oscillator.
Frequency drift of an oscillator is one such systematic effect. If it
where linear, processing it with ADEV would cause a sqrt(2) scale error.
Also, it would not give you a good prediction since usually you follow a
Aln(Bt+1) curve which isn't matching the requirement, so you will only
get first degree compensation of that with HDEV style measures.
Yes, and I think that for variations that are easily and accurately
modeled this is appropriate, however, your next sentence says it all.
Temperature variations is tricky to say the least.
When you have random and systematic effects, separate them and estimate
them separately and then build a combined prediction from these models.
Random jitter and deterministic jitter are two such aspects. Same
applies at longer taus as well.
True, but the underlying story is really that you seems to put too much
confidence in ADEV and ADEV alone. ADEV is a great tool, especially to
handle some of the 1/f noise variants which normal engineering RMS can't
handle. It's a speciallized branch to meet special needs. But it remains
only one of several tools. Learn what the tool is good and and what it
is not so good and, then add another tool with another set of strengths
and weaknesses. Add then tools until you get a good enough picture of
what the hell the device does so you can make meaningful predictions
about it's performance in the environments you are about to put it in,
so your end task achieves it's goal, often on a time and budget
constraint, and often with the device you didn't want to use for this task.
Cheers,
Magnus
On 02/10/2012 12:51 AM, Jim Lux wrote:
> On 2/9/12 8:42 AM, Magnus Danielson wrote:
>> On 02/09/2012 04:10 PM, Jim Lux wrote:
>
>>>
>>> Interesting point you make here. The rising ADEV at 100-1000 second-ish
>>> tau in a system that should be better is a classic sign (at least around
>>> here) that temperature effects are showing up.
>>
>> I regularly see the building AC at 900-1000 s for instance.
>>
>>> However, how could one remove that effect from the raw data? And isn't
>>> the measurement of the "system", which includes the environmental
>>> effects.
>>
>> ADEV and friends is there to handle random sources, where as this is a
>> systematic source.
>
> But I would contend that unless you can externally measure that
> disturbance and remove it, it's a fundamental part of the frequency
> variation. Since you don't have control over it, or necessarily have
> accurate information about it, it's not something that could be
> "calibrated out".
It's a huge difference what comes in the device and what environment we
put it in. A temperature dependence, is a systematic effect on the device.
> Say, for instance, I had an oscillator that was mounted in such a way
> that it rotated slowly once every hour. There would be a periodic
> variation in frequency, which could be accurately modeled and removed,
> so that wouldn't necessarily count in ADEV.
That is indeed a systematic effect in that environment of that device.
Once you have characterized the effect, you can then predict the
behaviour to some precision if you now turns it quicker, more often or
not at all. The random effects would then be re-introduced on that
predicted systematic effect and you would see the aggregate effect.
> But temperature is random (although band limited), and so, for
> measurements over the 1000 second time scale, it's impossible to tell if
> the change was due to temperature or due to the underlying oscillator.
Temperature is not necessary random, I often find systematic albeit
somewhat unregular patterns in my temperature readings and also
frequency or phase plots when I measure longer times on the less than
ideal oscillators. They dominate over any noise contributions, and for
those taus I care less about the noise and more about the systematic
effects. This is one of the reasons we use MTIE in addition to TDEV for
telecommunication measurements, it makes good engineering sense to use
it instead.
A couple of long term systematic effects on temperature relates to the
sun. The dinural wander, i.e. the 24 h period wander is due to the
rotation of earth and the effect of the sun. This is not a random, but
rather a very regular systematic effect.
Another similar effect is the season effect, over a year the temperature
shifts along with the earths axis orientation in relationship to the
sun. This is also a systematic effect.
As you put equipment into a building or other temperature controlled
environment, the bang-bang regulation method is often seen, and the
cycling of temperature up and down is not very steady but still regular
enough and again a systematic effect on the oscillator.
If temperature where random, we would see -80 C and +100 C more often in
our back yard than we do.
To some degree we can control temperature, we can predict temperature
and deal with it. We can handle it as an engineering concept and do
steering loops etc. It's pretty systematic to me.
>>> I suppose you could run your widget in a temperature controlled chamber,
>>> get those numbers. Then run it in a less controlled benchtop
>>> environment, and get those numbers, and claim that the difference is
>>> environmental.
>>>
>>> But at some point, what you're interested is the performance of the
>>> system in the environment in which it will be used. If you need good
>>> ADEV performance at the 1000 second tau, then you need an oven, a vacuum
>>> bottle, or a better design that's less environment sensitive.
>>
>> You could also build active systematic effect predictors to lower this
>> systematic effect.
>
> Yes. That's basically no different than controlling the environment. If
> the transfer function of environment to output is well known, and you
> know the environment, you could legitimately "remove" it from the
> measured output.
Model wise it's not a huge difference, but practically it may be.
>>
>> By doing proper logging of key environmental effects, build a model for
>> how the dominant variations will systematically affect the signal and
>> then remove that from the measurements you get a better random jitter
>> measurement.
>
> Ah, but there's the rub. Can you actually do that with acceptable
> performance?
>
> I know that you can measure the temperature of a crystal and fairly
> accurately calibrate out the frequency change due to temperature (to the
> point where frequency can be used to measure temperature). So now, your
> ADEV on the "calibrated" output would depend on the temperature
> measurement accuracy. Essentially what you have done is reduced the
> tempco of the system.
You never get your model and measurement perfect. Not even your ADEV.
You may get good enough confidence for what you need to do.
I trust that you ask for a certain temperature-spec on your oscillators,
that you is expected to make the environment suitable enough etc.
Really, what I say is that trying to lump everything into a single
number is just not a very useful thing when doing engineering,
especially with these devices we use called crystal oscillators. There
are many subtleties which one needs to learn. Tossing everything into a
single ADEV diagram isn't going to help the learning curve.
Don't get me wrong, ADEV is a great tool, it's just not a great tool for
all the effects we see for an oscillator.
>>
>> Frequency drift of an oscillator is one such systematic effect. If it
>> where linear, processing it with ADEV would cause a sqrt(2) scale error.
>> Also, it would not give you a good prediction since usually you follow a
>> A*ln(B*t+1) curve which isn't matching the requirement, so you will only
>> get first degree compensation of that with HDEV style measures.
>
> Yes, and I think that for variations that are easily and *accurately*
> modeled this is appropriate, however, your next sentence says it all.
>>
>> Temperature variations is tricky to say the least.
>>
>> When you have random and systematic effects, separate them and estimate
>> them separately and then build a combined prediction from these models.
>>
>> Random jitter and deterministic jitter are two such aspects. Same
>> applies at longer taus as well.
True, but the underlying story is really that you seems to put too much
confidence in ADEV and ADEV alone. ADEV is a great tool, especially to
handle some of the 1/f noise variants which normal engineering RMS can't
handle. It's a speciallized branch to meet special needs. But it remains
only one of several tools. Learn what the tool is good and and what it
is not so good and, then add another tool with another set of strengths
and weaknesses. Add then tools until you get a good enough picture of
what the hell the device does so you can make meaningful predictions
about it's performance in the environments you are about to put it in,
so your end task achieves it's goal, often on a time and budget
constraint, and often with the device you didn't want to use for this task.
Cheers,
Magnus
