PRS-10 findings
Hello guys,
some time ago we talked about curious humps in the ADEV of the SRS PRS-10 Rb
at around 20s intervalls.
I think I may have more info on that (and as usually that raises more
questions):
Below is a capture of the phase offset between a PRS10 1PPS output compared
to the Jackson-Labs Fury 1PPS output as measured with a 53132A.
The Fury output has been shifted by about 80ns to give a nice phase offset
for the 53132A to measure.
Besides the underlying counter noise, there clearly is a jump of about 0.5 -
1ns visible exactly once every 15 seconds! Indicated by arrows.
Could this be the counter self-calibrating? This is too regular to be
co-incident.
Has anyone else seen this?
I don't see this on a scope when comparing two 1PPS's so I do think it's the
counter.
0.078,3 us
0.078,1 us
0.078,3 us
0.078,3 us
0.078,3 us
0.078,1 us
0.078,3 us
0.078,3 us
0.078,4 us
0.078,4 us
0.078,4 us
0.079,1 us <==
0.078,3 us
0.078,3 us
0.078,3 us
0.078,1 us
0.078,3 us
0.078,1 us
0.078,1 us
0.078,1 us
0.078,0 us
0.078,0 us
0.078,1 us
0.078,1 us
0.078,1 us
0.078,0 us
0.078,9 us <==
0.078,0 us
0.078,1 us
0.078,3 us
0.078,2 us
0.078,2 us
0.078,1 us
0.078,5 us
0.078,2 us
0.078,1 us
0.078,4 us
0.078,5 us
0.078,2 us
0.078,1 us
0.078,5 us
0.078,9 us <==
0.078,2 us
0.078,4 us
0.078,2 us
0.078,4 us
0.078,0 us
0.078,4 us
0.078,2 us
0.078,4 us
0.078,2 us
0.078,2 us
0.078,5 us
0.078,3 us
0.078,3 us
0.078,3 us
0.079,0 us <==
0.078,2 us
0.078,1 us
0.078,3 us
0.078,2 us
0.078,5 us
0.078,3 us
0.078,5 us
0.078,6 us
0.078,5 us
0.078,6 us
0.078,5 us
0.078,6 us
0.078,6 us
0.078,3 us
0.079,6 us
0.078,6 us
0.078,5 us
0.078,6 us
0.078,6 us
0.078,4 us
0.078,5 us
0.078,6 us
0.078,4 us
0.078,4 us
0.078,4 us
0.078,5 us
0.078,4 us
0.078,4 us
0.078,5 us
0.079,3 us
0.078,7 us
0.078,6 us
0.078,6 us
0.078,7 us
0.078,7 us
0.078,7 us
0.078,8 us
0.078,8 us
0.078,8 us
0.078,9 us
0.078,8 us
0.078,9 us
0.078,9 us
0.078,8 us
0.079,5 us
************************************** See what's free at http://www.aol.com.
Hello Said,
I have seen something similar with my 53132A. I was checking
on the delay variation of an amplifier distributing 10 MHz, and
noticed a regular sinusoidal pattern, about a third of a nanosecond
peak-to-peak, with a period of about 70 seconds. This looked like
interference from an off-frequency signal. I hadn't bothered to lock
the counter to an external reference, and on measuring a Rubidium
source against its internal oscillator, got a reading about 1.4 x
10^-9 high, which matched perfectly this 70-second period Locking
the counter to an external reference, this beating went away.
Attached is a small GIF that shows the counter readings for twenty
minutes before and after applying the locking signal. (Readings were
taken once a second, and to smooth the graph slightly, each point on
the graph is obtained by averaging ten readings.)
Could this be your problem, Said?
Peter Vince
Said asked:
some time ago we talked about curious humps in the ADEV of the SRS PRS-10 Rb
at around 20s intervalls.
I think I may have more info on that (and as usually that raises more
questions):
Below is a capture of the phase offset between a PRS10 1PPS output compared
to the Jackson-Labs Fury 1PPS output as measured with a 53132A.
The Fury output has been shifted by about 80ns to give a nice phase offset
for the 53132A to measure.
Besides the underlying counter noise, there clearly is a jump of about 0.5 -
1ns visible exactly once every 15 seconds! Indicated by arrows.
Could this be the counter self-calibrating? This is too regular to be
co-incident.
Has anyone else seen this?
In message <v04210100c2831acb5ca5@[10.0.0.3]>, Peter Vince writes:
I have seen something similar with my 53132A. I was checking
on the delay variation of an amplifier distributing 10 MHz, and
noticed a regular sinusoidal pattern, about a third of a nanosecond
peak-to-peak, with a period of about 70 seconds.
Almost any kind of interference will cause such anomalies and the
closer the frequencies are to a multiple of each other, the longer
the period will be.
It is quite common for the frequency difference between the counters
internal X-tal and the measued frequency to show up like that once
you start to measure down in the nanosecond end of things.
The HP5370 has a rather heavyhanded piece of electronics that
eliminate this effect with a jitter based approach and as far as I
have been able to measure, it works.
"normal" counters don't have this, as they are not designed to measure
in that domain of disturbances.
The easiest way to determine if this is indeed the problem, is to
feed the counter an external frequency which can be varied a bit
up and down. If the period of the artifact changes accordingly: QED.
--
Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG | TCP/IP since RFC 956
FreeBSD committer | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.
Poul-Henning Kamp wrote:
In message <v04210100c2831acb5ca5@[10.0.0.3]>, Peter Vince writes:
I have seen something similar with my 53132A. I was checking
on the delay variation of an amplifier distributing 10 MHz, and
noticed a regular sinusoidal pattern, about a third of a nanosecond
peak-to-peak, with a period of about 70 seconds.
A manifestation of the effects of slowly sweeping through a range of
start and stop interpolator values so that the interpolator integral and
and differential non linearities become apparent?
Strictly the difference between the nonlinearities of the start and stop
interpolators becomes visible.
The resultant variation of around 300ps pp are well within the
specifications for the counter.
Almost any kind of interference will cause such anomalies and the
closer the frequencies are to a multiple of each other, the longer
the period will be.
It is quite common for the frequency difference between the counters
internal X-tal and the measued frequency to show up like that once
you start to measure down in the nanosecond end of things.
The HP5370 has a rather heavyhanded piece of electronics that
eliminate this effect with a jitter based approach and as far as I
have been able to measure, it works.
Not quite true the HP5370 has a whole host of anomalies like
differential linearity errors of 100psec or more for certain time
interval ranges, at least according to its designers.
The identified causes are:
Crosstalk between microstriplines used for each channel (effective only
when the affected signals are simultaneously active near a trigger
point) and modulation of the internal 200MHz reference by the mixer
outputs (always present with a quasi period of ~5.02ns).
"normal" counters don't have this, as they are not designed to measure
in that domain of disturbances.
The easiest way to determine if this is indeed the problem, is to
feed the counter an external frequency which can be varied a bit
up and down. If the period of the artifact changes accordingly: QED.
Bruce
In message 465D89BD.8000106@xtra.co.nz, Dr Bruce Griffiths writes:
The HP5370 has a rather heavyhanded piece of electronics that
eliminate this effect with a jitter based approach and as far as I
have been able to measure, it works.
Not quite true the HP5370 has a whole host of anomalies like
differential linearity errors of 100psec or more for certain time
interval ranges, at least according to its designers.
I did say "as far as I have been able to measure", didn't I ? :-)
--
Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG | TCP/IP since RFC 956
FreeBSD committer | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.
Hello guys,
some time ago we talked about curious humps in the ADEV of the SRS PRS-10 Rb
at around 20s intervalls.
Can you refresh me on the topic? I didn't see any
PRS10 humps here:
http://www.leapsecond.com/museum/prs10/
Could this be the counter self-calibrating? This is too regular to be
co-incident.
I've not heard that these counters do some sort of dynamic
self-calibration. Nor have I seen that sort of periodicity in a
53131 or 53132 counter. You can verify this with one or two
other non-PRS10 and non-Fury sources: set them in the
range of something like 1e-10 apart and collect data for a
couple of minutes. You should see them drift apart at the
rate of 100 ps per second with expected amount of noise
but no other anomalies.
Has anyone else seen this?
I don't see this on a scope when comparing two 1PPS's so I do think it's the
counter.
This is the right approach -- to substitute another source, to
substitute another counter, another cable, another DC power
supply, until you locate the culprit.
Just to get a clear idea of your lab set up; how is your counter
configured exactly; int or ext ref? trigger levels? 1X or 10X
attenuation, etc.
If you can see the readings on the 53132 display, do they also
happen when your PC (RS232 or GPIB?) is not electrically
connected to the counter? Desktop or floating laptop?
Is the Fury floating (wall wart) or are you using a grounded
power supply? How is the PRS10 powered? etc.
Also do the 10 MHz RF outputs show this phase jump or is it
only in the 1PPS outputs?
/tvb
From: Dr Bruce Griffiths bruce.griffiths@xtra.co.nz
Subject: Re: [time-nuts] PRS-10 findings
Date: Thu, 31 May 2007 02:27:09 +1200
Message-ID: 465D89BD.8000106@xtra.co.nz
Poul-Henning Kamp wrote:
In message <v04210100c2831acb5ca5@[10.0.0.3]>, Peter Vince writes:
I have seen something similar with my 53132A. I was checking
on the delay variation of an amplifier distributing 10 MHz, and
noticed a regular sinusoidal pattern, about a third of a nanosecond
peak-to-peak, with a period of about 70 seconds.
A manifestation of the effects of slowly sweeping through a range of
start and stop interpolator values so that the interpolator integral and
and differential non linearities become apparent?
Yes. I have been pointing toward such tests earlier.
One should recall that the deviations is not only from non-linear effects, but
also from highly linear effects such as cross-talk between channels which help
to cause biases in channels. Similar cross-talk towards the internal clocks
should be expected. There can be many causes of such cross-talk, so due care
needs to be considered throughout the design.
Strictly the difference between the nonlinearities of the start and stop
interpolators becomes visible.
The resultant variation of around 300ps pp are well within the
specifications for the counter.
Almost any kind of interference will cause such anomalies and the
closer the frequencies are to a multiple of each other, the longer
the period will be.
It is quite common for the frequency difference between the counters
internal X-tal and the measued frequency to show up like that once
you start to measure down in the nanosecond end of things.
The HP5370 has a rather heavyhanded piece of electronics that
eliminate this effect with a jitter based approach and as far as I
have been able to measure, it works.
Not quite true the HP5370 has a whole host of anomalies like
differential linearity errors of 100psec or more for certain time
interval ranges, at least according to its designers.
The identified causes are:
Crosstalk between microstriplines used for each channel (effective only
when the affected signals are simultaneously active near a trigger
point) and modulation of the internal 200MHz reference by the mixer
outputs (always present with a quasi period of ~5.02ns).
This matches my experience too, altought not with the 5370. I would also expect
there to be some interaction with the interpolators. There always are.
Ground-bounce is certainly a reason for crosstalk which causes such non-
linearities.
Cheers,
Magnus
From: "Poul-Henning Kamp" phk@phk.freebsd.dk
Subject: Re: [time-nuts] PRS-10 findings
Date: Wed, 30 May 2007 14:35:37 +0000
Message-ID: 92477.1180535737@critter.freebsd.dk
In message 465D89BD.8000106@xtra.co.nz, Dr Bruce Griffiths writes:
The HP5370 has a rather heavyhanded piece of electronics that
eliminate this effect with a jitter based approach and as far as I
have been able to measure, it works.
Not quite true the HP5370 has a whole host of anomalies like
differential linearity errors of 100psec or more for certain time
interval ranges, at least according to its designers.
I did say "as far as I have been able to measure", didn't I ? :-)
You did, you did... :-)
If you lock a low-noise oscillator to the reference of your counter, such that
you have a fixed but slight frequency deviation and then start on one clock
and stop on the other, you will very slowly pass through all phase-states over
and over again. If the clocks are stable enougth the locking is mearly means to
ensure the frequency offset over the measurement period. Using this you can
predict the expected time difference for each measurement point with fair
precission and can then compare that with the measured one and as a result you
should be able to make a fairly decent measured vs. actual TI plot or if you
so wish, a TIE plot for the instrument.
The center point between stop->start worst injection and start->stop worst
injection is not nessecarilly around 0 s in time interval.
Cheers,
Magnus