Methods for comparing oscillators
John
John Miles wrote:
The spec for the 10811A's trimmer range is +/- 10 Hz, actually. Seems
doable.
There was a good thread on the list back in late 2006 where some of the
pitfalls of the DMTD technique in general were discussed. It seems that
achieving optimal performance is a lot less trivial than anyone would like.
If someone wanted to build a "minimum effort" system with a lower short-term
floor than the 5370, my understanding is that it would require:
- a 10811-class OXCO tuned to generate beat notes of several Hz
- a mixer of some sort with good IF response down to DC
- four good isolation amps (which thanks to Bruce's recent work is not a
difficult or expensive issue) - some sort of delay compensation on one of the inputs may be needed to
avoid decorrelation of the transfer oscillator's effect between the two
channels - a sound card with good differential-phase stability
Of course a 5370 or other two-channel TIC could be used instead of an FFT
back end, but I'd be interested in seeing what level of performance could be
expected with a sound card, too. Bruce, you were looking at differential
jitter and delay on stereo sound card channels at one point, weren't you?
I haven't yet measured it but as far as I can determine from the
literature the sampling jitter is around 1ns or less (usually much less)
for high end cards.
Even as much sampling jitter as an unlikely 10ns is negligible (~10fs at
the mixer input) for an entry level system with a 10Hz beat note.
The biggest problem is likely to be the phase shift instability due to
the coupling caps in the sound card (its not always practical/feasible
to remove them).
With 10Hz beat note and a 1Hz single pole high pass RC input filter and
electrolytic coupling caps with a tempco of 1000ppm/C the phase shift
tempco will be around 1.6us/C (1.6ps/C at the mixer input). If the input
RC time constants for the 2 channels matching to within 20% the
differential phase shift tempco will be reduced to around 300fs/C at the
mixer input.
Replacing the input coupling caps with lower tempco plastic caps may be
worthwhile as the phase shift tempco will be reduced by a factor of at
least 4x even before the effect of closer matching is considered.
If differential phase stability is a problem, then perhaps one of the
channel IFs could be heterodyned up to a different carrier frequency that
could share a single physical ADC channel. Isolation and phase-delay
compensation would again be a big deal in any such scheme. Probably the
heterodyne LO would need to be derived directly from the transfer
oscillator.
-- john, KE5FX
Bruce
-----Original Message-----
From: time-nuts-bounces@febo.com [mailto:time-nuts-bounces@febo.com]On
Behalf Of Tom Van Baak
Sent: Sunday, August 02, 2009 9:14 PM
To: Discussion of precise time and frequency measurement
Subject: Re: [time-nuts] Methods for comparing oscillators
A 1Hz offset may be convenient as its easy to adjust an 10811A or
similar to achieve this.
A 10Hz offset is harder to achieve.
Would either of these be 10 Hz solutions?
- open the 10811 and change the xtal load capacitance, or
- change the oven temperature set point until the resonator
is 10 Hz off. You'd loose a little in tempco but...
/tvb
time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.
The biggest difference between the 5370 and the 5372 is the single-shot
resolution. The 5370 has a resolution of 20 ps while the 5372 is
limited to 150 ps. When measuring Allen Deviation, this difference
means that for a Tau of 1 second, the 5370 can measure down to approx.
2e-11 while the 5372 can only reach 1.5e-10. Note that since many
high-quality OCXOs are spec'ed lower than these numbers, neither counter
can directly measure them and need help e.g. DMTD or some other
heterodyne technique.
Ed
Steve Rooke wrote:
2009/7/31 John Green wpxs472@gmail.com:
I've been hanging around and reading long enough to understand that when
measuring the differences between oscillators the preferred methods are the
HP 5370A Time difference counter or the dual mixer method. I want to
What's the difference between a HP 5370A and a HP 5372A for this type
of measurement?
73,
Steve
evaluate some ocxo's and Rb sources against either a Tbolt or Z3801 and I
don't have either method available. What I have used in the past is an HP
Infinium scope with the reference fed to one channel which also provides
sync and the DUT to the other. I have tested 2 ocxos that were so close that
the two waveforms did not move by a detectable amount in a 30 minute period.
I realize that this method will require very long observation times when
looking at more stable sources. I am not looking to get absolute data, just
comparative. Given what I have to work with, is there a better way? I use an
Agilent 89441A Vector Signal Analyzer for signal quality measurements. I can
see 60 Hz sidebands at least 60 or 70 db down and while I can't measure
phase noise, I can tell a clean oscillator from a dirty one. For instance,
there is a world of difference between the signal generated by an HP8920 and
a E4430B.
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.
2009/8/4 Ed Palmer ed_palmer@sasktel.net:
The biggest difference between the 5370 and the 5372 is the single-shot
resolution. The 5370 has a resolution of 20 ps while the 5372 is limited to
150 ps. When measuring Allen Deviation, this difference means that for a
Tau of 1 second, the 5370 can measure down to approx. 2e-11 while the 5372
can only reach 1.5e-10. Note that since many high-quality OCXOs are spec'ed
lower than these numbers, neither counter can directly measure them and need
help e.g. DMTD or some other heterodyne technique.
I'm not sure that the 5370A is capable of measuring without dead time
so this will affect Allen Deviation measurements unless procedures
like picket fence are used.
Yes, DMTD and similar look like the way to improve the resolution of
these instruments. Just need a practical way of implementing it.
73, Steve
Ed
Steve Rooke wrote:
2009/7/31 John Green wpxs472@gmail.com:
I've been hanging around and reading long enough to understand that when
measuring the differences between oscillators the preferred methods are
the
HP 5370A Time difference counter or the dual mixer method. I want to
What's the difference between a HP 5370A and a HP 5372A for this type
of measurement?
73,
Steve
evaluate some ocxo's and Rb sources against either a Tbolt or Z3801 and I
don't have either method available. What I have used in the past is an HP
Infinium scope with the reference fed to one channel which also provides
sync and the DUT to the other. I have tested 2 ocxos that were so close
that
the two waveforms did not move by a detectable amount in a 30 minute
period.
I realize that this method will require very long observation times when
looking at more stable sources. I am not looking to get absolute data,
just
comparative. Given what I have to work with, is there a better way? I use
an
Agilent 89441A Vector Signal Analyzer for signal quality measurements. I
can
see 60 Hz sidebands at least 60 or 70 db down and while I can't measure
phase noise, I can tell a clean oscillator from a dirty one. For
instance,
there is a world of difference between the signal generated by an HP8920
and
a E4430B.
time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.
time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.
--
Steve Rooke - ZL3TUV & G8KVD
A man with one clock knows what time it is;
A man with two clocks is never quite sure.
John Miles wrote:
The spec for the 10811A's trimmer range is +/- 10 Hz, actually. Seems
doable.
There was a good thread on the list back in late 2006 where some of the
pitfalls of the DMTD technique in general were discussed. It seems that
achieving optimal performance is a lot less trivial than anyone would like.
If someone wanted to build a "minimum effort" system with a lower short-term
floor than the 5370, my understanding is that it would require:
- a 10811-class OXCO tuned to generate beat notes of several Hz
- a mixer of some sort with good IF response down to DC
You need some form of mixer IF termination and amplification/buffer,
which Bruce also have done work on.
- four good isolation amps (which thanks to Bruce's recent work is not a
difficult or expensive issue) - some sort of delay compensation on one of the inputs may be needed to
avoid decorrelation of the transfer oscillator's effect between the two
channels
As I have explained in a recent thread, this is not necessary if one
takes the sampled data processing route rather than the ZCD path to
measurement and data correlation. The phase offset between the ZCD
triggers causes transfer oscillator decorrelation losses. A synchronous
sample-for-sample processing of the two channel pairs maintains a very
low time difference and the loss is negligable as it only exposes very
high frequency noise as sample error exposes.
- a sound card with good differential-phase stability
You also need processing software. While it does not have to be hard,
there is a number of processing pitfalls to fall into that could make it
a bit difficult to get accurate results.
Cheers,
Magnus
Magnus Danielson wrote:
- a 10811-class OXCO tuned to generate beat notes of several Hz
On the 10811 production line, they had some 10811's with specially
made crystals in them that had a 500 Hz offset. These were
used to test for short term stability with a 500 Hz beat note,
using the HP 5390 system. The offset oscillators
themselves were verified for short term stability by a round robin
method. The trouble with this method is that there was no
way to find the 10811's with the best stability, since all good
ones were just "at least as good as the offset oscillator".
Since the offset oscillators were limited quantity specials,
there was no way to select for good stability.
I volunteered to management to build a really high quality
synthesizer to replace the offset 10811's. However, they
never took me up on it.
I think the general opinion was that you couldn't get away
with a beat note of a few Hz as could be achieved by pulling
a standard crystal.
Rick Karlquist N6RK