Testing the goodness of my GPSDO
Hey all,
First time poster, so pardon me if I make some mistakes. I am a Physics
student and part of my project concerns with building a GPSDO. So, I made a
digital PI controller on an STM32 chip to control the OCXO. I need help in
checking if the different gains I am trying are reducing the drift.
From the images you can see the allan deviations I have measured. I got the
GPS PPS allan deviation by using the timestamps of the signal, the TDC used
for that depends on my OCXO I want to control. I got my 10 MHz OCXO's allan
deviation from the frequency readings of a frequency counter in the lab
(non zero dead time between measurements, uses a GPSDO from the GPS PPS
module as reference).
I expected image 1, when the OCXO is not controlled, the measured PPS
stability would be bad in the long run (the timestamps depend on the
frequency drifts of the OCXO). I was surprised to see that the relationship
tracked that closely. When I measured it with the controller, I see image
2. I don't understand how the measured PPS allan deviation is lower than
the OCXO's in large tau. Is there a flaw in my measurement? Can I not trust
the allan deviation at long time from a frequency counter (from my
knowledge a Phase Noise Analyser is better for this)? Image 3 is just both
the PPS allan deviations on the same graph.
I got a phase noise analyser (microchip 53100a) yesterday which I can use
for a week and I would like to make the most of it. I tried doing a 3
cornered hat measurement, but I get negative allan variances, I think this
is because my third OCXO is not disciplined to a GPS signal unlike the
other two.
Any help on how to characterise my GPSDO with Phase Noise Analyser and
other ways will be helpful. Please let me know if something I
haven't mentioned is clear. Thank you!
Regards,
Srihari P.
Hi
First step in any design is to work out what the goal is. How well does your device need to perform
to be “good enough”? That might include any of a number of specs. Your data runs down to 5x10^-13
at 10K seconds. Is that the goal? A goal of “as good as possible” could be pursued for decades …. :)
All measurements like this are going to be a “compare A to B” (or something more complex). There
is no device that will directly measure frequency or time.
To evaluate a GPSDO (or the things that go into it), the first thing you need is a reference that is
better than what you are trying to measure. Unless it is a very unusual device, a stand alone frequency
counter will hit a number of limits when trying to do this.
If an OCXO is used as a reference, it probably should be on power in a stable (+/- 2C) environment for a
few weeks before it is used. How well it does will be highly dependent on the quality of the OCXO.
Rubidium Atomic clocks also have a warm up period. Running for at least 5 days is a good idea.
Cesium Atomic clocks typically take a few days for their auto calibration firmware to “settle in”.
Yes there are a whole bunch of more exotic clocks out there than these basic devices. We’re just
looking at < $100K devices above. If you have a $10M clock you can access, then by all means,
go ahead and use it.
When you look at the plot that shows the GPS tracking the OCXO long term, it’s a good bet you are
actually seeing the drift of your reference.
Indeed, you may need multiple references: An OCXO for close in, an RB for 10 to 10K seconds and a
Cs for > 10K seconds. Each has a range where it’s ADEV should be better than the others.
I would suggest sorting this out before you dive into a lot more data.
Bob
On Feb 28, 2026, at 8:31 AM, Srihari Padmanaban via time-nuts time-nuts@lists.febo.com wrote:
Hey all,
First time poster, so pardon me if I make some mistakes. I am a Physics
student and part of my project concerns with building a GPSDO. So, I made a
digital PI controller on an STM32 chip to control the OCXO. I need help in
checking if the different gains I am trying are reducing the drift.
From the images you can see the allan deviations I have measured. I got the
GPS PPS allan deviation by using the timestamps of the signal, the TDC used
for that depends on my OCXO I want to control. I got my 10 MHz OCXO's allan
deviation from the frequency readings of a frequency counter in the lab
(non zero dead time between measurements, uses a GPSDO from the GPS PPS
module as reference).
I expected image 1, when the OCXO is not controlled, the measured PPS
stability would be bad in the long run (the timestamps depend on the
frequency drifts of the OCXO). I was surprised to see that the relationship
tracked that closely. When I measured it with the controller, I see image
2. I don't understand how the measured PPS allan deviation is lower than
the OCXO's in large tau. Is there a flaw in my measurement? Can I not trust
the allan deviation at long time from a frequency counter (from my
knowledge a Phase Noise Analyser is better for this)? Image 3 is just both
the PPS allan deviations on the same graph.
I got a phase noise analyser (microchip 53100a) yesterday which I can use
for a week and I would like to make the most of it. I tried doing a 3
cornered hat measurement, but I get negative allan variances, I think this
is because my third OCXO is not disciplined to a GPS signal unlike the
other two.
Any help on how to characterise my GPSDO with Phase Noise Analyser and
other ways will be helpful. Please let me know if something I
haven't mentioned is clear. Thank you!
Regards,
Srihari P.
<Image 1.png><Image 2.png><Image 3.png><Image 3.png>_______________________________________________
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
Keep in mind that a recording of the PPS offset as measured by the controller will represent what's left after the controller has done its best to steer the OCXO to GPS. It represents the loop’s residual error, so to speak, as best as the controller can estimate it. So a straight line heading downward at 1/tau is OK. It depicts a small worst-case error that, because it remains within an order of 1ns or so, looks better and better as the observation window gets longer.
This doesn’t reflect the GPSDO’s absolute level of performance that you would see if you could measure it against a perfect standard; rather it suggests how well the GPSDO could perform in an ideal case where it was fed with a perfect, noise-free sky signal.
Image 3 doesn't actually appear to be a copy of the same data from the other plot. Whatever you're measuring seems about 10x more stable at taus near 1 hour than what we see in image 2. It looks like a good GPSDO being measured against a good free-running Rb standard. I’m not entirely clear on what that trace represents. I got two copies of the same Image 3 file, so maybe that wasn’t the right one?
Re: negative variances in the 3-cornered hat trace, those are pretty common, and they can arise from a few different causes that can be hard to pin down. You might let the measurement run longer to see if the artifact resolves itself.
See example attached – this represents a 3-cornered hat with a couple of passive masers and an AtomiChron-based GPSDO. H1 is known to be working well, H2 not so much. The GPSDO and H1 appear to exhibit similar stability near t=20k-40k seconds, and H1’s ADEV estimate goes negative in that region, possibly because H1 and H2 are being influenced by similar environmental effects that don’t affect the GPSDO.
The red trace is not part of the 3-cornered hat solution; it’s showing the same thing that I’m guessing you are plotting in your blue traces, which is the residual phase error reported by the GPSDO. The GPSDO is locking its Rb oscillator without ever straying more than a fraction of a nanosecond off-target, and this appears to be comparable to what you are seeing.
Bottom line: a GPSDO can’t be used to measure itself. All it can do is indicate a problem with the control loop, and you seem to be doing pretty well there, it looks like. If you had a perfect reference, you would presumably see the magenta trace in image 2 continue to head downward like the light green trace in my example, instead of turning upward beyond t=1 hour or so.
-- john
(53100A developer)
-----Original Message-----
From: Srihari Padmanaban via time-nuts time-nuts@lists.febo.com
Hey all,
First time poster, so pardon me if I make some mistakes. I am a Physics student and part of my project concerns with building a GPSDO. So, I made a digital PI controller on an STM32 chip to control the OCXO. I need help in checking if the different gains I am trying are reducing the drift…
Hi Bob and John,
Thank you for your inputs. Before I answer, let me just name my oscillators
and traces to avoid confusion. OCXO1 is the oscillator I control, OCXO2 is
another OCXO (similar to OCXO1) that is not controlled and GPSDO1 is an
OCXO which came with the GPS PPS module and is disciplined to it. The name
of the traces in the images contain what the two references are, this
doesn't apply to the GPS_PPS trace as it is just the timestamps which I
read, which only depend on OCXO1. The notes for each trace contains
information on wether OCXO1 is controlled or not. If it is controlled, then
you would see I_gain_P_gain_loopFilterConstant. I do not have access to Rb
or Cs standards.
Bob,
All the OCXOs are in a stable temperature controlled room and have been
running continuously for at least a month. My goal is to control OCXO1 to
follow the GPS PPS signal without degrading its performance in the shorter
time scales. I know OCXO1 performs better then GPSDO1 at shorter time
scales I want to know how to measure this with what I have. I do not have
access to Rb or Cs standards.
Are you saying I can't use GPSDO1 as a reference to measure the stability
of OCXO1 accurately at longer time scales?
John,
I have attached the images again with consistent with the naming I said in
the beginning. I have combined all the adevs of GPS PPS in one graph
(measured_PPS_Adev) and all the adevs from frequency counter/ Microchip
53100A below in another (OCXO1_GPSDO1_Adev). I have also attached the adevs
and 3 corner hat of my best settings so far(Best_Adev, Best_Adev3hat).
Please refer to these images.
I am not sure I understand what you mean when you mention PPS offset. The
straight lines which are plotted is just when I give TimeLab the measured
GPS PPS timestamps and input frequency as 1 Hz and not the Allan deviation
of the time difference from a PPS from my OCXO1 and GPS PPS. I hope this is
clear, from my understanding these to essentially give me the same
information, that is the residual error.
When I used a frequency counter to measure my adev (OCXO1 and GPSDO1), I
saw the upward slope. However I don't see it when I measure it with
Microchip 53100A (see OCXO1_GPSDO1_Adev) . I also see that the measured_PPS
adevs are affected appropriately when OCXO1 is not controlled. This made me
think the procedure I am following to measure adev in both cases is valid.
Bottom line: Could you comment on if the way I am taking my measurements
are valid, is there a better way I can use? Another question is, in 3
corner hat when one of the variances is negative, is the adev valid for the
other 2 measurements? Just because the variances of one of them is negative
doesn't mean that their adev is much lower than the others, correct? Do I
trust the normal adevs or the 3 cornered adevs for tuning my GPSDO further.
Sorry if my question are a bit rudimentary, but I am really not familiar
with this field. Thank you for your help!!
Regards,
Srihari P.
On Tue, Mar 3, 2026 at 9:01 AM john@miles.io wrote:
Keep in mind that a recording of the PPS offset as measured by the
controller will represent what's left after the controller has done its
best to steer the OCXO to GPS. It represents the loop’s residual error, so
to speak, as best as the controller can estimate it. So a straight line
heading downward at 1/tau is OK. It depicts a small worst-case error that,
because it remains within an order of 1ns or so, looks better and better as
the observation window gets longer.
This doesn’t reflect the GPSDO’s absolute level of performance that you
would see if you could measure it against a perfect standard; rather it
suggests how well the GPSDO could perform in an ideal case where it was fed
with a perfect, noise-free sky signal.
Image 3 doesn't actually appear to be a copy of the same data from the
other plot. Whatever you're measuring seems about 10x more stable at taus
near 1 hour than what we see in image 2. It looks like a good GPSDO being
measured against a good free-running Rb standard. I’m not entirely clear
on what that trace represents. I got two copies of the same Image 3 file,
so maybe that wasn’t the right one?
Re: negative variances in the 3-cornered hat trace, those are pretty
common, and they can arise from a few different causes that can be hard to
pin down. You might let the measurement run longer to see if the artifact
resolves itself.
See example attached – this represents a 3-cornered hat with a couple of
passive masers and an AtomiChron-based GPSDO. H1 is known to be working
well, H2 not so much. The GPSDO and H1 appear to exhibit similar stability
near t=20k-40k seconds, and H1’s ADEV estimate goes negative in that
region, possibly because H1 and H2 are being influenced by similar
environmental effects that don’t affect the GPSDO.
The red trace is not part of the 3-cornered hat solution; it’s showing the
same thing that I’m guessing you are plotting in your blue traces, which is
the residual phase error reported by the GPSDO. The GPSDO is locking its
Rb oscillator without ever straying more than a fraction of a nanosecond
off-target, and this appears to be comparable to what you are seeing.
Bottom line: a GPSDO can’t be used to measure itself. All it can do is
indicate a problem with the control loop, and you seem to be doing pretty
well there, it looks like. If you had a perfect reference, you would
presumably see the magenta trace in image 2 continue to head downward like
the light green trace in my example, instead of turning upward beyond t=1
hour or so.
-- john
(53100A developer)
-----Original Message-----
From: Srihari Padmanaban via time-nuts time-nuts@lists.febo.com
Hey all,
First time poster, so pardon me if I make some mistakes. I am a Physics
student and part of my project concerns with building a GPSDO. So, I made a
digital PI controller on an STM32 chip to control the OCXO. I need help in
checking if the different gains I am trying are reducing the drift…
No problem at all, you’re welcome! I guess the first question I’d have is, what type of OCXOs are you using? The green trace in Best_Adev3hat suggests that OCXO2 is almost getting into the E-14s at t=1s and remaining below 1E-12 at taus of an hour or more. So either you are using BVA or Rakon HSO-class parts, or there is something going on that needs investigation.
My take on Best_Adev3hat: first, the purple trace for GPSDO1 looks believable. It is in line with expectations for a high-quality GPSDO. The purple trace would presumably merge with the red residual trace, but the math is breaking down beyond t=100-200s, likely because of the expected increase in correlation between OCXO1 and GPSDO1 at longer-term taus.
For OCXO1, you have some 50 Hz power-line interference in the plot (ripple beginning near t=0.02s). Not a huge deal but just FYI, that’s what that is. OCXO1 merges nicely with the residual trace and then turns downward until it can no longer be measured with the other sources.
Re: trusting the 3-cornered hat results, this is pretty typical of the tea-leaf reading process that is often needed with these plots. Assuming the sources are all truly independent of each other, I would personally treat the blue OCXO1 trace as valid below t=5000s and the purple GPSDO1 trace as valid below t=100s.
As for the light green trace for OCXO2, the dropout is typical of what can happen when the other two sources are either correlated with each other (e.g., injection-locking, experiencing common-mode environmental influences, or starting to track each other due to the actions of their respective loops), and/or when the oscillator for the trace in question is quite a bit better than the other two. I would say you can mentally curve-fit the gap between t=3s and t-100s for a reasonable assessment of OCXO2’s performance…. but only if it is a very high-grade part as noted. If it is a commodity-grade OCXO that some kid in China pulled off a board with a propane torch and sold on eBay, then no, it isn’t realistic, and you’d need to look for possible injection-locking mechanisms.
Re: the red PPS trace, I think it’s in line with expectations. I don’t know what exact type of timestamp or PPS error data you’re getting from the GPSDO, but regardless, that looks like the residual performance.
measured_PPS_Adev: The magenta trace for the unlocked OCXO1 looks believable, assuming it is a very stable part similar to OCXO2. Beyond t=500s or so the GPSDO starts outperforming OCXO1, so the PPS error is a valid way to measure OCXO1’s stability past that point. Neither the blue nor green locked traces in this plot tell you very much, other than (again) that your loop is working OK and that OCXO1 is doing what it is being told.
Best_Adev: This plot looks fine to me. The blue trace where OCXO1 is locked to GPS with your loop and being measured against the GPSDO’s own oscillator (right?) is behaving well. It agrees with the purple GPSDO1 trace from the 3-cornered hat plot as expected. Generally speaking, any GPSDO that stays at 1E-12 or under for the entire ADEV tau range is doing great.
The fact that the blue and red traces are parallel to each other is also a good sign that the loop is operating as expected, but I’m not sure why the blue trace is better. At first glance I’d expect them to be coincident at best, as the residual performance at taus controlled by both the GPSDO’s loop and your own OCXO1 loop would presumably be close to identical. As noted above, the purple GPSDO1 trace in the 3-cornered hat plot is also better than the red residual baseline for some reason. The margin of difference isn’t huge but it’s a little hard to hand-wave away. Maybe Bob or someone else has some input on that. Could be caused by precision loss in the 1PPS timestamps you are parsing, maybe.
The green and magenta traces reflecting OCXO2’s performance also make sense. The 50 Hz interference in OCXO1 is dominating the green OCXO2-OCXO1 measurement below t=1s as expected. Meanwhile GPSDO1_OCXO2 doesn’t suffer from that, but is dominated by the GPSDO OCXO’s short-term stability, which is itself unusually good.
OCXO1_GPSDO1_Adev: Makes sense for the most part, where you can see the effect of different loop parameters. The light blue trace with the frequency counter shows quite a bit of divergence from the nearest trace from the 53100A, but the loop parameters are different so presumably that’s the reason. It’s even farther below the dark blue residual trace, though, and it’s also not clear why it starts to drift at tau >1 hour. That suggests that the loop is either losing long-term control with those particular settings, or is in danger of oscillating at a very long period that isn’t reached by this plot, perhaps as long as a day or so.
Overall, the only thing that really jumps out to me is the suspiciously-good performance of OCXO2. If it’s really that good, then I don’t see any major issues with the methodology. Looks like you’re doing some solid work with excellent hardware.
-- john
From: Srihari Padmanaban srihari.p03@gmail.com
Sent: Monday, March 2, 2026 6:54 PM
Hi Bob and John,
Thank you for your inputs. Before I answer, let me just name my oscillators and traces to avoid confusion…
Also, just to clarify your earlier question – it is possible that the other two traces in a three-cornered hat are OK in areas where one shows a negative variance, but it depends on why the negative variance occurred. If you know that the uncertainty, correlation, or noise that corrupted the best of the three traces is small compared to the magnitude of the other two, there may be no reason to reject the others out of hand.
That’s true for your plot, I think. There is little reason to doubt the integrity of the purple or blue traces in the t=3s to t=100s range in Best_Adev3hat, because OCXO2 is so much better than they are in that range. That, in turn, is a safe assumption because it’s so clearly true at t=3s and shorter taus. All three traces look very solid there, and the confidence will be quite good at those taus given the 15-hour duration of the measurement.
Another point: you can use the GPSDO to measure the long-term stability of either oscillator, or both at once in a three-cornered hat, as long as the oscillator(s) under test are unlocked. The GPSDO can’t measure the stability of its own oscillator directly, since it will just look like a straight line pointing downwards on the ADEV plot as you’ve seen. One thing you can do, though, is plot the EFC voltage used by the GPSDO to control its oscillator, applying the appropriate kVCO scale factor to turn it into a frequency difference. This can be quite accurate if you measure the scale factor carefully, since the OCXO will not normally be tuned far enough for the kVCO to change significantly.
-- john
Hi
If you want to test the GPSDO, you need to compare it to something else.
If you only have an OCXO as your reference standard, it will limit what you can do.
How much it limits things depends very much on the make and model of OCXO
you are using. It also depends a bit on the performance of that specific unit. Not all
OCXO’s (even from the same batch) perform equally well.
If you have a third OCXO, you can set up a three corner hat and work out which
of the three is best. If one is much better, the process will fall apart. You then go
looking for a fourth or fifth OCXO.
Bob
On Mar 2, 2026, at 9:54 PM, Srihari Padmanaban via time-nuts time-nuts@lists.febo.com wrote:
Hi Bob and John,
Thank you for your inputs. Before I answer, let me just name my oscillators
and traces to avoid confusion. OCXO1 is the oscillator I control, OCXO2 is
another OCXO (similar to OCXO1) that is not controlled and GPSDO1 is an
OCXO which came with the GPS PPS module and is disciplined to it. The name
of the traces in the images contain what the two references are, this
doesn't apply to the GPS_PPS trace as it is just the timestamps which I
read, which only depend on OCXO1. The notes for each trace contains
information on wether OCXO1 is controlled or not. If it is controlled, then
you would see I_gain_P_gain_loopFilterConstant. I do not have access to Rb
or Cs standards.
Bob,
All the OCXOs are in a stable temperature controlled room and have been
running continuously for at least a month. My goal is to control OCXO1 to
follow the GPS PPS signal without degrading its performance in the shorter
time scales. I know OCXO1 performs better then GPSDO1 at shorter time
scales I want to know how to measure this with what I have. I do not have
access to Rb or Cs standards.
Are you saying I can't use GPSDO1 as a reference to measure the stability
of OCXO1 accurately at longer time scales?
John,
I have attached the images again with consistent with the naming I said in
the beginning. I have combined all the adevs of GPS PPS in one graph
(measured_PPS_Adev) and all the adevs from frequency counter/ Microchip
53100A below in another (OCXO1_GPSDO1_Adev). I have also attached the adevs
and 3 corner hat of my best settings so far(Best_Adev, Best_Adev3hat).
Please refer to these images.
I am not sure I understand what you mean when you mention PPS offset. The
straight lines which are plotted is just when I give TimeLab the measured
GPS PPS timestamps and input frequency as 1 Hz and not the Allan deviation
of the time difference from a PPS from my OCXO1 and GPS PPS. I hope this is
clear, from my understanding these to essentially give me the same
information, that is the residual error.
When I used a frequency counter to measure my adev (OCXO1 and GPSDO1), I
saw the upward slope. However I don't see it when I measure it with
Microchip 53100A (see OCXO1_GPSDO1_Adev) . I also see that the measured_PPS
adevs are affected appropriately when OCXO1 is not controlled. This made me
think the procedure I am following to measure adev in both cases is valid.
Bottom line: Could you comment on if the way I am taking my measurements
are valid, is there a better way I can use? Another question is, in 3
corner hat when one of the variances is negative, is the adev valid for the
other 2 measurements? Just because the variances of one of them is negative
doesn't mean that their adev is much lower than the others, correct? Do I
trust the normal adevs or the 3 cornered adevs for tuning my GPSDO further.
Sorry if my question are a bit rudimentary, but I am really not familiar
with this field. Thank you for your help!!
Regards,
Srihari P.
On Tue, Mar 3, 2026 at 9:01 AM john@miles.io wrote:
Keep in mind that a recording of the PPS offset as measured by the
controller will represent what's left after the controller has done its
best to steer the OCXO to GPS. It represents the loop’s residual error, so
to speak, as best as the controller can estimate it. So a straight line
heading downward at 1/tau is OK. It depicts a small worst-case error that,
because it remains within an order of 1ns or so, looks better and better as
the observation window gets longer.
This doesn’t reflect the GPSDO’s absolute level of performance that you
would see if you could measure it against a perfect standard; rather it
suggests how well the GPSDO could perform in an ideal case where it was fed
with a perfect, noise-free sky signal.
Image 3 doesn't actually appear to be a copy of the same data from the
other plot. Whatever you're measuring seems about 10x more stable at taus
near 1 hour than what we see in image 2. It looks like a good GPSDO being
measured against a good free-running Rb standard. I’m not entirely clear
on what that trace represents. I got two copies of the same Image 3 file,
so maybe that wasn’t the right one?
Re: negative variances in the 3-cornered hat trace, those are pretty
common, and they can arise from a few different causes that can be hard to
pin down. You might let the measurement run longer to see if the artifact
resolves itself.
See example attached – this represents a 3-cornered hat with a couple of
passive masers and an AtomiChron-based GPSDO. H1 is known to be working
well, H2 not so much. The GPSDO and H1 appear to exhibit similar stability
near t=20k-40k seconds, and H1’s ADEV estimate goes negative in that
region, possibly because H1 and H2 are being influenced by similar
environmental effects that don’t affect the GPSDO.
The red trace is not part of the 3-cornered hat solution; it’s showing the
same thing that I’m guessing you are plotting in your blue traces, which is
the residual phase error reported by the GPSDO. The GPSDO is locking its
Rb oscillator without ever straying more than a fraction of a nanosecond
off-target, and this appears to be comparable to what you are seeing.
Bottom line: a GPSDO can’t be used to measure itself. All it can do is
indicate a problem with the control loop, and you seem to be doing pretty
well there, it looks like. If you had a perfect reference, you would
presumably see the magenta trace in image 2 continue to head downward like
the light green trace in my example, instead of turning upward beyond t=1
hour or so.
-- john
(53100A developer)
-----Original Message-----
From: Srihari Padmanaban via time-nuts time-nuts@lists.febo.com
Hey all,
First time poster, so pardon me if I make some mistakes. I am a Physics
student and part of my project concerns with building a GPSDO. So, I made a
digital PI controller on an STM32 chip to control the OCXO. I need help in
checking if the different gains I am trying are reducing the drift…
<measured_PPS_Adev.png><OCXO1_GPSDO1_Adev.png><Best_Adev3hat.png><Best_Adev.png>_______________________________________________
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
Thanks for your reply.
Yes OCXO1 and OCXO2 are excellent oscillators (Oscilloquartz 8607), however
I am not sure why the 50 Hz ripple is present and why OCXO2 is better than
OCXO1. is there any suggestion you can give to eliminate that? Another
relevant information is there was a 3 cornered hat measurement done with
three OCXOs (all 8607) (imag: all3aDev) and all those three OCXOs have
around the same short term stability. But in my current measurements
OCXO2 seems to be much better, is this and the jaggedness at tau less than
1s because of the 50 Hz noise? Should I be worried about these fluctuations
when I use this oscillator for experiments?
Coming back, so is the final verdict that I can't trust the 3 cornered hat
adev values for large taus (because of increase in correlation
between OCXO1 and GPSDO1), and this would not solve even if I take data for
longer times? Instead for longer times, I don't have to worry about it as
long as my measured_PPS traces are straight and maintain the linear
downward slope and the GPSDO_OCXO1 aDev to follow this trace. right? But
you say that I can trust these values for tau between 3s and 100s, because
the source of negative variance (lower stability of OCXO2) is different and
not correlation? I hope I understood you well.
Even I found it interesting that the aDev measured of OCXO1vsGPSDO1 is
lower than the residual signal measured using locked OCXO1. But I think I
will just accept this fact and continue.
Regards,
Srihari P.
On Tue, Mar 3, 2026 at 6:18 PM john@miles.io wrote:
Also, just to clarify your earlier question – it is possible that the
other two traces in a three-cornered hat are OK in areas where one shows a
negative variance, but it depends on why the negative variance occurred.
If you know that the uncertainty, correlation, or noise that corrupted the
best of the three traces is small compared to the magnitude of the other
two, there may be no reason to reject the others out of hand.
That’s true for your plot, I think. There is little reason to doubt the
integrity of the purple or blue traces in the t=3s to t=100s range in
Best_Adev3hat, because OCXO2 is so much better than they are in that
range. That, in turn, is a safe assumption because it’s so clearly true at
t=3s and shorter taus. All three traces look very solid there, and the
confidence will be quite good at those taus given the 15-hour duration of
the measurement.
Another point: you can use the GPSDO to measure the long-term stability of
either oscillator, or both at once in a three-cornered hat, as long as the
oscillator(s) under test are unlocked. The GPSDO can’t measure the
stability of its own oscillator directly, since it will just look like a
straight line pointing downwards on the ADEV plot as you’ve seen. One
thing you can do, though, is plot the EFC voltage used by the GPSDO to
control its oscillator, applying the appropriate kVCO scale factor to turn
it into a frequency difference. This can be quite accurate if you measure
the scale factor carefully, since the OCXO will not normally be tuned far
enough for the kVCO to change significantly.
-- john
Hi
Based on your three corner hat test, your “best OCXO” is around 2x10^-13 over a reasonable range of
tau. OCXO’s past 1,000 seconds will always be questionable as a reference. Keeping drafts and other
temperature issues under control is one of many things that makes this true.
As I understand it, you will be using one of these OCXO’s as your independent reference for the testing.
That’s fine and it will work. It will limit you to a roughly 2x10^13 floor for your measurements. It also will
make data past 1K seconds questionable.
Anything you see on a result that is below 2x10^-13 suggests some sort of measurement issue. Practically
speaking, you will be limited to 1K second data. This also is the point that issues like ionospheric effects
start to mess up the reference out of your GPS device.
Bob
On Mar 3, 2026, at 8:15 AM, Srihari Padmanaban via time-nuts time-nuts@lists.febo.com wrote:
Thanks for your reply.
Yes OCXO1 and OCXO2 are excellent oscillators (Oscilloquartz 8607), however
I am not sure why the 50 Hz ripple is present and why OCXO2 is better than
OCXO1. is there any suggestion you can give to eliminate that? Another
relevant information is there was a 3 cornered hat measurement done with
three OCXOs (all 8607) (imag: all3aDev) and all those three OCXOs have
around the same short term stability. But in my current measurements
OCXO2 seems to be much better, is this and the jaggedness at tau less than
1s because of the 50 Hz noise? Should I be worried about these fluctuations
when I use this oscillator for experiments?
Coming back, so is the final verdict that I can't trust the 3 cornered hat
adev values for large taus (because of increase in correlation
between OCXO1 and GPSDO1), and this would not solve even if I take data for
longer times? Instead for longer times, I don't have to worry about it as
long as my measured_PPS traces are straight and maintain the linear
downward slope and the GPSDO_OCXO1 aDev to follow this trace. right? But
you say that I can trust these values for tau between 3s and 100s, because
the source of negative variance (lower stability of OCXO2) is different and
not correlation? I hope I understood you well.
Even I found it interesting that the aDev measured of OCXO1vsGPSDO1 is
lower than the residual signal measured using locked OCXO1. But I think I
will just accept this fact and continue.
Regards,
Srihari P.
On Tue, Mar 3, 2026 at 6:18 PM john@miles.io wrote:
Also, just to clarify your earlier question – it is possible that the
other two traces in a three-cornered hat are OK in areas where one shows a
negative variance, but it depends on why the negative variance occurred.
If you know that the uncertainty, correlation, or noise that corrupted the
best of the three traces is small compared to the magnitude of the other
two, there may be no reason to reject the others out of hand.
That’s true for your plot, I think. There is little reason to doubt the
integrity of the purple or blue traces in the t=3s to t=100s range in
Best_Adev3hat, because OCXO2 is so much better than they are in that
range. That, in turn, is a safe assumption because it’s so clearly true at
t=3s and shorter taus. All three traces look very solid there, and the
confidence will be quite good at those taus given the 15-hour duration of
the measurement.
Another point: you can use the GPSDO to measure the long-term stability of
either oscillator, or both at once in a three-cornered hat, as long as the
oscillator(s) under test are unlocked. The GPSDO can’t measure the
stability of its own oscillator directly, since it will just look like a
straight line pointing downwards on the ADEV plot as you’ve seen. One
thing you can do, though, is plot the EFC voltage used by the GPSDO to
control its oscillator, applying the appropriate kVCO scale factor to turn
it into a frequency difference. This can be quite accurate if you measure
the scale factor carefully, since the OCXO will not normally be tuned far
enough for the kVCO to change significantly.
-- john
<all3aDev.png>_______________________________________________
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
(JohnA: any idea why my replies have several blank lines between paragraphs? I’m only using one blank line when I type.)
The 50 Hz ripple usually means that there is a large loop area in the ground wiring of your test setup, possibly because the 53100A and/or PC is plugged into a separate power circuit than the sources being measured. It’s best if you can plug every piece of equipment involved into a single circuit, ideally the same power strip. You could also reduce the ripple by selecting 5 Hz measurement bandwidth. The drawback to using oscillators this good is that you’ll see artifacts that otherwise would be buried in the noise, and that’s definitely true in your case.
Note that all3aDev.png does not have the 3-cornered hat view turned on; if hitting ctrl-h doesn’t work, check your source labels.
The real problem is simply that you can’t measure the long-term stability of a source like this without a better long-term reference source. The use of two independent references can be a massive win in short- and medium-term measurements, but in the long run, common environmental effects usually rule out much improvement. Even if that’s not an issue, you will have to run for weeks to improve the results at taus on the order of one day, so it just isn’t practical.
For this type of work you really need to set up an AtomiChron-enabled GPSDO such as one of the Sparkfun GNSS disciplined oscillators or Viavi PNT-6250 DOCXO-based units (disclosure: I work with Viavi but not on that product.) This will give you a stable reference in the same class as a passive maser at much less upfront expense. Basically you want a GNSS-disciplined oscillator with the highest-precision AtomiChron service and the lowest-grade holdover oscillator available -- i.e., no rubidium standard in the case of the Viavi units.
-- john
From: Srihari Padmanaban srihari.p03@gmail.com
Sent: Tuesday, March 3, 2026 5:15 AM
Thanks for your reply.
Yes OCXO1 and OCXO2 are excellent oscillators (Oscilloquartz 8607)…