Question about effect of sample interval on ADEV
This is a question that has probably been addressed on the list and in
various texts but I've been unable to find an answer that I can fathom
so here's a request for some info about the behavior of TimeLab (and
probably all other similar programs).
I have been trying to find the source of some periodic noise that
appears when using a 5370B to measure an FTS 1050B against a 5065B. The
noise manifests itself varying from 12s to 20s in a repetitive fashion.
I asked C. Dawson about this and one of his suggestions was to try a
longer sampling period. Down the rabbit hole I went!
I ran four trials in succession at sample intervals of .07s, .25s, .5s,
and 1s. The result is as if the adev ,modified, and Hadamard curves
have been slid down and to the right along the noise floor of the 5370.
My assumption, apparently incorrect, was that the software would take
the sampling interval into account so that I would get essentially the
same plot. When I edit a plot changing the sample interval, the trace
remains essentially unchanged and this seems inconsistent with the
results noted above. Can anyone explain in relatively simple terms what
I'm missing?
Thanks,
Bob Darby
Hi Bob,
Send me the raw data and I'll have a look to confirm or explain your results.
I've seen this in some of my FTS 1050 also; it's always been the fault of the battery charger circuit.
/tvb
----- Original Message -----
From: "Robert Darby" bobdarby@triad.rr.com
To: time-nuts@febo.com
Sent: Sunday, July 07, 2013 12:27 PM
Subject: [time-nuts] Question about effect of sample interval on ADEV
This is a question that has probably been addressed on the list and in
various texts but I've been unable to find an answer that I can fathom
so here's a request for some info about the behavior of TimeLab (and
probably all other similar programs).
I have been trying to find the source of some periodic noise that
appears when using a 5370B to measure an FTS 1050B against a 5065B. The
noise manifests itself varying from 12s to 20s in a repetitive fashion.
I asked C. Dawson about this and one of his suggestions was to try a
longer sampling period. Down the rabbit hole I went!
I ran four trials in succession at sample intervals of .07s, .25s, .5s,
and 1s. The result is as if the adev ,modified, and Hadamard curves
have been slid down and to the right along the noise floor of the 5370.
My assumption, apparently incorrect, was that the software would take
the sampling interval into account so that I would get essentially the
same plot. When I edit a plot changing the sample interval, the trace
remains essentially unchanged and this seems inconsistent with the
results noted above. Can anyone explain in relatively simple terms what
I'm missing?
Thanks,
Bob Darby
Bob,
On 07/07/2013 09:27 PM, Robert Darby wrote:
This is a question that has probably been addressed on the list and in
various texts but I've been unable to find an answer that I can fathom
so here's a request for some info about the behavior of TimeLab (and
probably all other similar programs).
I have been trying to find the source of some periodic noise that
appears when using a 5370B to measure an FTS 1050B against a 5065B. The
noise manifests itself varying from 12s to 20s in a repetitive fashion.
I asked C. Dawson about this and one of his suggestions was to try a
longer sampling period. Down the rabbit hole I went!
I ran four trials in succession at sample intervals of .07s, .25s, .5s,
and 1s. The result is as if the adev ,modified, and Hadamard curves have
been slid down and to the right along the noise floor of the 5370.
My assumption, apparently incorrect, was that the software would take
the sampling interval into account so that I would get essentially the
same plot. When I edit a plot changing the sample interval, the trace
remains essentially unchanged and this seems inconsistent with the
results noted above. Can anyone explain in relatively simple terms what
I'm missing?
TimeLab will use the sampling interval you gave it or it learned during
monitoring and scale results accordingly if properly given.
If you have systematic noise, try loading it into a FFT rather than
doing an ADEV. The trouble is that a sine modulation will show up as
multiple bumps on the ADEV but a single spike on FFT. If you have
multiple signals, it becomes easier to identify in the FFT.
Cheers,
Magnus
This won't be a sampling-interval issue. It sounds like a beat note. To
diagnose it, you can use the 5370B in frequency mode (with its internal
timebase) to measure the frequency of the 5065B and the FTS 1050B. Subtract
the two readings, then see if the reciprocal of the frequency difference
corresponds to the location and spacing of the periodic ADEV bumps. If so,
that's likely to be the explanation, and you can confirm it by tweaking the
FTS 1050B's frequency and seeing if the beatnote moves accordingly.
As far as getting rid of the artifact is concerned, it may help to use
double-shielded cables, although I don't know if the isolation between the
START and STOP inputs on the 5370B is good enough to eliminate the
possibility of beatnotes in a TI measurement with HF signals on both jacks.
If you are feeding the 5/10 MHz inputs to both START and STOP inputs, try
using a 1-pps divider on the START source.
You could also try using the 5370B in frequency-count mode, with the 5065A
as an external reference and the FTS 1050B at the STOP input. There will
be a reduction in ADEV fidelity due to the dead time but it will probably be
less objectionable than the beatnote ripple.
-- john, KE5FX
Miles Design LLC
My assumption, apparently incorrect, was that the software would take
the sampling interval into account so that I would get essentially the
same plot. When I edit a plot changing the sample interval, the trace
remains essentially unchanged and this seems inconsistent with the
results noted above. Can anyone explain in relatively simple terms what
I'm missing?
Thanks,
Bob Darby
On 07/07/2013 11:38 PM, John Miles wrote:
This won't be a sampling-interval issue. It sounds like a beat note. To
diagnose it, you can use the 5370B in frequency mode (with its internal
timebase) to measure the frequency of the 5065B and the FTS 1050B. Subtract
the two readings, then see if the reciprocal of the frequency difference
corresponds to the location and spacing of the periodic ADEV bumps. If so,
that's likely to be the explanation, and you can confirm it by tweaking the
FTS 1050B's frequency and seeing if the beatnote moves accordingly.
As far as getting rid of the artifact is concerned, it may help to use
double-shielded cables, although I don't know if the isolation between the
START and STOP inputs on the 5370B is good enough to eliminate the
possibility of beatnotes in a TI measurement with HF signals on both jacks.
If you are feeding the 5/10 MHz inputs to both START and STOP inputs, try
using a 1-pps divider on the START source.
You could also try using the 5370B in frequency-count mode, with the 5065A
as an external reference and the FTS 1050B at the STOP input. There will
be a reduction in ADEV fidelity due to the dead time but it will probably be
less objectionable than the beatnote ripple.
I had a measurement with a sine being overlaid, and just for fun I wrote
a small pre-processing program that put a pair of zeros close to the
unity circle and about the right frequency. The end result was very
clean and the unwanted artifact was removed. Care in ensuring unity gain
was needed, but once that was done it worked like a charm.
This "trick" is a bit dirty, but keeping the Q high on the zeros makes
sure that the other noise is not affected gravely, and the ripples of
the sine was cleaned out.
The benefit of doing an equalizer to notch it out compared to trying to
measure the amplitude and phase of a sine and then subtract the
estimated sine is what the notching method will be relatively
insensitive to amplitude, phase and frequency errors that will limit the
usefulness of the perfect matching. Any slow shifts will also be fairly
ignored. I used a very rough period estimation to tune it.
Cheers,
Magnus
Thanks to all who responded. I didn't phrase my question very well I'm
afraid. The periodic noise/beat note issue is what lead me to try a
different sample interval. Previously I had always let TimeLab set the
sample interval (about .07s for the5370B).
I was surprised by the difference in the adev traces when I varied the
sample interval from .07s to .25s to .5s to 1.0s and that is what I
hoped someone could explain.
When I edit a trace and change the sample interval there is not a
substantial change to the Tau / Sigma (Tau) values yet when I actually
run at the different sample intervals I get minimum values for each run
of 400s 5.60e-13, 600s 1.70e-13, 1000s 9.55e-14, 3000s 4.15e-14. The
traces are totally different; same oscillators and counter, just
different sample intervals. That's what I hoping one of you could explain.
John, I'll explore the the beat note issue to see if that's the problem.
Tom, since I get a similar issue when I swap the FTS for an Austron
(more battery chargers) I don't think the FTS is the culprit.
Thanks again.
Bob
On 7/7/2013 5:38 PM, John Miles wrote:
This won't be a sampling-interval issue. It sounds like a beat note. To
diagnose it, you can use the 5370B in frequency mode (with its internal
timebase) to measure the frequency of the 5065B and the FTS 1050B. Subtract
the two readings, then see if the reciprocal of the frequency difference
corresponds to the location and spacing of the periodic ADEV bumps. If so,
that's likely to be the explanation, and you can confirm it by tweaking the
FTS 1050B's frequency and seeing if the beatnote moves accordingly.
As far as getting rid of the artifact is concerned, it may help to use
double-shielded cables, although I don't know if the isolation between the
START and STOP inputs on the 5370B is good enough to eliminate the
possibility of beatnotes in a TI measurement with HF signals on both jacks.
If you are feeding the 5/10 MHz inputs to both START and STOP inputs, try
using a 1-pps divider on the START source.
You could also try using the 5370B in frequency-count mode, with the 5065A
as an external reference and the FTS 1050B at the STOP input. There will
be a reduction in ADEV fidelity due to the dead time but it will probably be
less objectionable than the beatnote ripple.
-- john, KE5FX
Miles Design LLC
My assumption, apparently incorrect, was that the software would take
the sampling interval into account so that I would get essentially the
same plot. When I edit a plot changing the sample interval, the trace
remains essentially unchanged and this seems inconsistent with the
results noted above. Can anyone explain in relatively simple terms what
I'm missing?
Thanks,
Bob Darby
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.
-----Original Message-----
From: time-nuts-bounces@febo.com [mailto:time-nuts-bounces@febo.com] On
Behalf Of Robert Darby
Sent: Sunday, July 07, 2013 3:33 PM
To: time-nuts@febo.com
Subject: Re: [time-nuts] Question about effect of sample interval on ADEV
Thanks to all who responded. I didn't phrase my question very well I'm
afraid. The periodic noise/beat note issue is what lead me to try a
different sample interval. Previously I had always let TimeLab set the
sample interval (about .07s for the5370B).
I was surprised by the difference in the adev traces when I varied the
sample interval from .07s to .25s to .5s to 1.0s and that is what I
hoped someone could explain.
When I edit a trace and change the sample interval there is not a
substantial change to the Tau / Sigma (Tau) values yet when I actually
run at the different sample intervals I get minimum values for each run
of 400s 5.60e-13, 600s 1.70e-13, 1000s 9.55e-14, 3000s 4.15e-14. The
traces are totally different; same oscillators and counter, just
different sample intervals. That's what I hoping one of you could
explain.
So the only difference between the test setups is the setting of the Display
Rate control on the 5370, correct? You're allowing TimeLab to estimate the
sample rate automatically, and giving it enough time to converge on a stable
reading before hitting 'Start Measurement'?
You're correct in that changing the real-world sample rate should yield
results that are identical (or at least very similar) to resampling the
phase data after the fact. In frequency mode, dead time between readings
would make that an iffy proposition, but for data taken in TI mode the
outcomes should be close.
-- john, KE5FX
Miles Design LLC
John,
In the past I've allowed TimeLab to pick the sampling interval (usually
.07s for the 5370B) but when I use Acquire, Sampling Interval and set
that to 1 sec I get very different results. Apparently the result is
not the same as Edit, Trace, Sample Interval. All of the foregoing in
TI mode.
BTW, thanks for the program.
Bob Darby
On 7/7/2013 8:57 PM, John Miles wrote:
-----Original Message-----
From: time-nuts-bounces@febo.com [mailto:time-nuts-bounces@febo.com] On
Behalf Of Robert Darby
Sent: Sunday, July 07, 2013 3:33 PM
To: time-nuts@febo.com
Subject: Re: [time-nuts] Question about effect of sample interval on ADEV
Thanks to all who responded. I didn't phrase my question very well I'm
afraid. The periodic noise/beat note issue is what lead me to try a
different sample interval. Previously I had always let TimeLab set the
sample interval (about .07s for the5370B).
I was surprised by the difference in the adev traces when I varied the
sample interval from .07s to .25s to .5s to 1.0s and that is what I
hoped someone could explain.
When I edit a trace and change the sample interval there is not a
substantial change to the Tau / Sigma (Tau) values yet when I actually
run at the different sample intervals I get minimum values for each run
of 400s 5.60e-13, 600s 1.70e-13, 1000s 9.55e-14, 3000s 4.15e-14. The
traces are totally different; same oscillators and counter, just
different sample intervals. That's what I hoping one of you could
explain.
So the only difference between the test setups is the setting of the Display
Rate control on the 5370, correct? You're allowing TimeLab to estimate the
sample rate automatically, and giving it enough time to converge on a stable
reading before hitting 'Start Measurement'?
You're correct in that changing the real-world sample rate should yield
results that are identical (or at least very similar) to resampling the
phase data after the fact. In frequency mode, dead time between readings
would make that an iffy proposition, but for data taken in TI mode the
outcomes should be close.
-- john, KE5FX
Miles Design LLC
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.
John,
After a night's sleep and a rereading of your post I finally realized
what I was doing wrong. I did not understand the the role of the
sampling interval setting and the display rate setting on the 5370.
When I follow the process below the results are totally consistent.
Sorry to taken your time and thanks all for your help. Now to find the
original issue.....
Bob Darby
On 7/7/2013 8:57 PM, John Miles wrote:
So the only difference between the test setups is the setting of the
Display Rate control on the 5370, correct? You're allowing TimeLab to
estimate the sample rate automatically, and giving it enough time to
converge on a stable reading before hitting 'Start Measurement'?
You're correct in that changing the real-world sample rate should
yield results that are identical (or at least very similar) to
resampling the phase data after the fact. In frequency mode, dead time
between readings would make that an iffy proposition, but for data
taken in TI mode the outcomes should be close. -- john, KE5FX Miles
Design LLC
Bob,
Sorry, I'm not sure, if I've understood the issue - what exactly did you
wrong?
Thank you
Volker
Am 08.07.2013 14:48, schrieb Robert Darby:
John,
After a night's sleep and a rereading of your post I finally realized
what I was doing wrong. I did not understand the the role of the
sampling interval setting and the display rate setting on the 5370.
When I follow the process below the results are totally consistent.
Sorry to taken your time and thanks all for your help. Now to find
the original issue.....
Bob Darby
On 7/7/2013 8:57 PM, John Miles wrote:
So the only difference between the test setups is the setting of the
Display Rate control on the 5370, correct? You're allowing TimeLab to
estimate the sample rate automatically, and giving it enough time to
converge on a stable reading before hitting 'Start Measurement'?
You're correct in that changing the real-world sample rate should
yield results that are identical (or at least very similar) to
resampling the phase data after the fact. In frequency mode, dead
time between readings would make that an iffy proposition, but for
data taken in TI mode the outcomes should be close. -- john, KE5FX
Miles Design LLC
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.