Hybrid analog digital dual stage DMTD based on nanoVNA HW. First actual measurement results
Not sure if there are people on this list interested in experimenting
with phase and frequency measurement technologies but maybe this
experiment can make someone smile.
Regular DMTD (dual mixer time difference) use a double mixer to convert
the two input signals to a low frequency and then uses regular time
interval measurement technology to measure the phase difference between
the two signals.
Using a nanoVNA HW with its dual mixer input and a new embedded SW one
can use the two input ports to input two signals into the dual mixer.
The output of the mixers is at 5kHz IF and send into a dual 16 bit ADC.
The 96kHz samples are processed in realtime in an STM32 MCU by doing an
I/Q down mix to DC for both inputs and the phase of the down mixed
signal is measured and the phase difference is calculated.
As test signal 10MHz outputs from a Rb and a DOCXO are used, roughly set
to 0.01Hz difference.
Phase and ADEV over a 100 second period where measurements and compared
with a similar measurement using a Picotest U6200A that claims to have a
12 digit per second resolution. [1]
The ADEV measured by the Picotest U6200A is roughly a factor 2 better
Looking at the phase showed there is some leakage causing extra phase
rotation during a 100 seconds cycle [2]
Further experiments using a RigolDG812 to generate the two test signals
0.01Hz apart showed it is possible to reduce the impact of the phase
leakage with at least a factor 3 but this also showed why you should not
use a DSS based generator for these measurements [3]
The slow fluctuation over a 100Hz period is similar but there are
additional fast phase fluctuations (not present with the Rb/DOCXO) that
are caused by the DSS trying to generate both 10MHz and 10.00000001Hz
using a 1GHz sample rate
Setting the DG812 frequency difference to 0.001Hz and measuring the
actual output frequency difference clearly showed the up and down jumps
that cause the fast phase fluctuations [4]
The nanoVNA HW seems to be able to measure phase and frequency
differences. Performance still needs to be determined.
[1] http://athome.kaashoek.com/time-nuts/PNA/nanoVNA_vs_Picotest_U6200A.png
[2]
http://athome.kaashoek.com/time-nuts/PNA/nanoVNA_vs_Picotest_U6200A_phase.png
[3] http://athome.kaashoek.com/time-nuts/PNA/DG812_0.01Hz_diff_phase.png
[4] http://athome.kaashoek.com/time-nuts/PNA/DG812_0.001Hz_diff_freq.png
Hi Erik,
There is a strong benefit in sampling the waveform over using the
original TIC approach, and that is that the decorrelation that the later
can experience is almost completely avoided. The slow variations you see
is quite typical and synthesizers have a variating degree of phase noise
to them. They can be just fine for some applications but not very useful
for others. Using a couple of low noise sources helps to establish the
measurement setup noise floor. A cheat-setup is to use one source and
split the signal into both inputs. Difference in cable length will cause
similar decorrelation, but unless you really get into longer lengths, it
is not significant to bother about.
I've been found to use cross-correlation setups to measure the noise of
sources and then using my quietest sources to measure the other ones
with good speed of convergence and trust in result.
It would be interesting to see just how far one could get the affordable
nanoVNA setup to run.
Cheers,
Magnus
On 9/29/22 19:38, Erik Kaashoek via time-nuts wrote:
Not sure if there are people on this list interested in experimenting
with phase and frequency measurement technologies but maybe this
experiment can make someone smile.
Regular DMTD (dual mixer time difference) use a double mixer to
convert the two input signals to a low frequency and then uses regular
time interval measurement technology to measure the phase difference
between the two signals.
Using a nanoVNA HW with its dual mixer input and a new embedded SW one
can use the two input ports to input two signals into the dual mixer.
The output of the mixers is at 5kHz IF and send into a dual 16 bit
ADC. The 96kHz samples are processed in realtime in an STM32 MCU by
doing an I/Q down mix to DC for both inputs and the phase of the down
mixed signal is measured and the phase difference is calculated.
As test signal 10MHz outputs from a Rb and a DOCXO are used, roughly
set to 0.01Hz difference.
Phase and ADEV over a 100 second period where measurements and
compared with a similar measurement using a Picotest U6200A that
claims to have a 12 digit per second resolution. [1]
The ADEV measured by the Picotest U6200A is roughly a factor 2 better
Looking at the phase showed there is some leakage causing extra phase
rotation during a 100 seconds cycle [2]
Further experiments using a RigolDG812 to generate the two test
signals 0.01Hz apart showed it is possible to reduce the impact of the
phase leakage with at least a factor 3 but this also showed why you
should not use a DSS based generator for these measurements [3]
The slow fluctuation over a 100Hz period is similar but there are
additional fast phase fluctuations (not present with the Rb/DOCXO)
that are caused by the DSS trying to generate both 10MHz and
10.00000001Hz using a 1GHz sample rate
Setting the DG812 frequency difference to 0.001Hz and measuring the
actual output frequency difference clearly showed the up and down
jumps that cause the fast phase fluctuations [4]
The nanoVNA HW seems to be able to measure phase and frequency
differences. Performance still needs to be determined.
[1]
http://athome.kaashoek.com/time-nuts/PNA/nanoVNA_vs_Picotest_U6200A.png
[2]
http://athome.kaashoek.com/time-nuts/PNA/nanoVNA_vs_Picotest_U6200A_phase.png
[3] http://athome.kaashoek.com/time-nuts/PNA/DG812_0.01Hz_diff_phase.png
[4] http://athome.kaashoek.com/time-nuts/PNA/DG812_0.001Hz_diff_freq.png
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
Hi
One “easy” way to come up with two offset sources is to
use something like a pair of telecom Rb standards. Bought
at the right time on eBay, they can be pretty cheap (delivered).
They are stable enough that they will hold ppt sort of offsets
for modest amounts of time. Tuning them in 10’s of ppt increments
is pretty easy.
Some level of research ( possibly in the archives of this list …)
might be wise. Some models are a bit better for things like spurs
than others ( The LPro is pretty reasonable ….).
Yes, the same thing can be done with OCXO’s. The offsets and
tuning. The usable offsets and relative drift will both be larger.
Bob
On Sep 29, 2022, at 6:22 PM, Magnus Danielson via time-nuts time-nuts@lists.febo.com wrote:
Hi Erik,
There is a strong benefit in sampling the waveform over using the original TIC approach, and that is that the decorrelation that the later can experience is almost completely avoided. The slow variations you see is quite typical and synthesizers have a variating degree of phase noise to them. They can be just fine for some applications but not very useful for others. Using a couple of low noise sources helps to establish the measurement setup noise floor. A cheat-setup is to use one source and split the signal into both inputs. Difference in cable length will cause similar decorrelation, but unless you really get into longer lengths, it is not significant to bother about.
I've been found to use cross-correlation setups to measure the noise of sources and then using my quietest sources to measure the other ones with good speed of convergence and trust in result.
It would be interesting to see just how far one could get the affordable nanoVNA setup to run.
Cheers,
Magnus
On 9/29/22 19:38, Erik Kaashoek via time-nuts wrote:
Not sure if there are people on this list interested in experimenting with phase and frequency measurement technologies but maybe this experiment can make someone smile.
Regular DMTD (dual mixer time difference) use a double mixer to convert the two input signals to a low frequency and then uses regular time interval measurement technology to measure the phase difference between the two signals.
Using a nanoVNA HW with its dual mixer input and a new embedded SW one can use the two input ports to input two signals into the dual mixer. The output of the mixers is at 5kHz IF and send into a dual 16 bit ADC. The 96kHz samples are processed in realtime in an STM32 MCU by doing an I/Q down mix to DC for both inputs and the phase of the down mixed signal is measured and the phase difference is calculated.
As test signal 10MHz outputs from a Rb and a DOCXO are used, roughly set to 0.01Hz difference.
Phase and ADEV over a 100 second period where measurements and compared with a similar measurement using a Picotest U6200A that claims to have a 12 digit per second resolution. [1]
The ADEV measured by the Picotest U6200A is roughly a factor 2 better
Looking at the phase showed there is some leakage causing extra phase rotation during a 100 seconds cycle [2]
Further experiments using a RigolDG812 to generate the two test signals 0.01Hz apart showed it is possible to reduce the impact of the phase leakage with at least a factor 3 but this also showed why you should not use a DSS based generator for these measurements [3]
The slow fluctuation over a 100Hz period is similar but there are additional fast phase fluctuations (not present with the Rb/DOCXO) that are caused by the DSS trying to generate both 10MHz and 10.00000001Hz using a 1GHz sample rate
Setting the DG812 frequency difference to 0.001Hz and measuring the actual output frequency difference clearly showed the up and down jumps that cause the fast phase fluctuations [4]
The nanoVNA HW seems to be able to measure phase and frequency differences. Performance still needs to be determined.
[1] http://athome.kaashoek.com/time-nuts/PNA/nanoVNA_vs_Picotest_U6200A.png
[2] http://athome.kaashoek.com/time-nuts/PNA/nanoVNA_vs_Picotest_U6200A_phase.png
[3] http://athome.kaashoek.com/time-nuts/PNA/DG812_0.01Hz_diff_phase.png
[4] http://athome.kaashoek.com/time-nuts/PNA/DG812_0.001Hz_diff_freq.png
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
Its a bit embarrassing but almost everything in my previous post was wrong.
After improving the FW the performance improved a lot.
To test the performance, two signals, either 0.01Hz apart or at
identical frequencies (from the same source) where used
Measurement of the signals with 0.01Hz difference was done with a
Picotest U6200A and with the nanoVNA based DMTD (further referred to as
tinyDMTD).
First the raw phase over time was measured to confirm the measured phase
values where correct [1]
Keep in mind the picotest, in time difference mode, measures between 0
and 100 ns with 100 ps resolution where the tinyDMTD measures the
instantaneous phase between -50 ns and +50 ns and about 1 ps resolution
Then the phase was unwrapped and the linear phase residue was calculated [2]
As the variations of the phase measured by the picotest where an order
higher than what the tinyDMTD measured a second plot shows only the
tinyDMTD measurements [3]
The leakage in the tinyDMTD between the two channels clearly has an
impact but the amount is limited to 30 ps and the impact goes down to
about 10 ps when more samples are used to calculate the phase
The ADEV of these three measurements confirms the resolution of the
tinyDMTD versus the picotest [4]
The ADEV and the phase measurement of the two identical inputs by the
tinyDMTD shows there is a fast fluctuation in the order of 3 ps and the
actual phase noise to be in the order of 1 ps.
Next step is to identify and remove the cause of the fast fluctuation in
order to see the true noise floor of the tinyDMTD
Erik.
[1] http://athome.kaashoek.com/time-nuts/PNA/tinyDMTD_U6200A_raw_phase.png
[2]
http://athome.kaashoek.com/time-nuts/PNA/tinyDMTD_U6200A_phase_residue.png
[3]
http://athome.kaashoek.com/time-nuts/PNA/tinyDMTD_U6200A_phase_residue_DMTD.png
[4] http://athome.kaashoek.com/time-nuts/PNA/tinyDMTD_U6200A_ADEV.png
Hi Magnus,
After making some improvements as shown in the block diagram [1] the
performance has improved .
When measuring two 10Mhz signals (no need for sharp edges, sine wave is
sufficient) either zero or 0.01 Hz apart the phase difference measures
as shown in [2]
Calculating the linear residue of the unwrapped phase [3] shows phase
measurement resolution of +/- 2e-12 seconds and maximum pulling of +/-
1e-11 seconds
When increasing the tau the phase resolution does improves a lot but the
pulling stay's the same.
For my testing purposes this level of resolution and accuracy is great.
Erik.
[1] http://athome.kaashoek.com/time-nuts/DMTD/Block_diagram.PNG
[2] http://athome.kaashoek.com/time-nuts/DMTD/Phase_difference.png
[3] http://athome.kaashoek.com/time-nuts/DMTD/Phase_difference_residue.png
On 30-9-2022 0:22, Magnus Danielson via time-nuts wrote:
It would be interesting to see just how far one could get the
affordable nanoVNA setup to run.
After some more work it became possible to measure the coherent noise
floor in a similar way as was described in "A Small Dual Mixer Time
Difference (DMTD) Clock Measuring System" (Small DMTD) by W.J. Rely
Both inputs where connected to separate outputs of a two channel DDS.
This allowed checking the frequency and phase measurement by testing
with various frequency difference and phase offsets. The frequency and
phase measured for all frequency and phase variations as expected and
it was fairly simple to measure the minimum phase step of the DDS (0.006
degrees)
For the coherent noise floor tests the frequency difference was set to zero.
Coherent phase data [1]
Over a roughly 45 minutes period the relative phase slowly varied over a
+2ps/-5ps range, comparable to the Small DMTD.
The measurement is susceptible to temperature variations as the
prototype does not yet have an enclosure. Draft from opening doors had
an observable impact, most easily spotted in the phase data.
Frequency data [2]
The frequency appeared to be stable, no linear trend was observed. At
first sight the frequency noise appeared to be less then the Small DMTD
but the Small DMTD did measure the frequency at a lower tau of 0.1 second.
Frequency stability [3]
Although no long term drift was present the slope of the stability plot
suggests there is still some hidden issue. The frequency stability of
the Small DMTD is clearly better.
Frequency histogram [4]
The histogram also hints to some none Gaussian behavior of the noise.
The total width of the histogram seems comparable to the Small DMTD
Auto correlation of the frequency [5]
The frequency auto correlation clearly shows there is some repetitive
signal hidden in the frequency data.
The frequency of the signal, about 1/120 Hz, suggests this could be due
to the 0.01 Hz resolution of the digital PLL that locks the internal
reference to one of the inputs. As the PLL is frozen during measurements
the frequency difference between input and internal reference could
easily be 1/120 Hz
Erik.
[1] http://athome.kaashoek.com/time-nuts/DMTD/phase.PNG
[2] http://athome.kaashoek.com/time-nuts/DMTD/freq.PNG
[3] http://athome.kaashoek.com/time-nuts/DMTD/stability.PNG
[4] http://athome.kaashoek.com/time-nuts/DMTD/histo.PNG
[5] http://athome.kaashoek.com/time-nuts/DMTD/autoc.PNG
Hi
I’d second everything Tom posted in his last email.
Name wise, I’d come up with something unique for the product. I would not
worry a lot about a name for the process it uses.
(see below)
On Oct 5, 2022, at 5:54 AM, Erik Kaashoek erik@kaashoek.com wrote:
After some more work it became possible to measure the coherent noise floor in a similar way as was described in "A Small Dual Mixer Time Difference (DMTD) Clock Measuring System" (Small DMTD) by W.J. Rely
Both inputs where connected to separate outputs of a two channel DDS. This allowed checking the frequency and phase measurement by testing with various frequency difference and phase offsets. The frequency and phase measured for all frequency and phase variations as expected and it was fairly simple to measure the minimum phase step of the DDS (0.006 degrees)
For the coherent noise floor tests the frequency difference was set to zero.
Coherent phase data [1]
Over a roughly 45 minutes period the relative phase slowly varied over a +2ps/-5ps range, comparable to the Small DMTD.
The measurement is susceptible to temperature variations as the prototype does not yet have an enclosure. Draft from opening doors had an observable impact, most easily spotted in the phase data.
Typically test gear gets spec’d in a very stable environment. The real world
is never stable, but the still do it that way. Even if you still have a coupe of
ps due to a few degrees C variation, it’s not a big deal. Your 45 minute period
is pretty representative of a typical HVAC system. Lots of stuff shows up at
that same sort of period.
Frequency data [2]
The frequency appeared to be stable, no linear trend was observed. At first sight the frequency noise appeared to be less then the Small DMTD but the Small DMTD did measure the frequency at a lower tau of 0.1 second.
While short tau is a “cool feature”, I have never seen a commercial ( or DIY )
use for it. The world seems to like ADEV ( and the other DEV’s ) for long tau
stuff. We all go to phase noise for the “faster” side of things. If it costs anything
to put in < 0.1 seconds, I’m not sure I’d vote for doing that.
Frequency stability [3]
Although no long term drift was present the slope of the stability plot suggests there is still some hidden issue. The frequency stability of the Small DMTD is clearly better.
Welcome to the club.
There always will be a next layer of issues. That’s been true on every
similar device I’ve ever built. Note: I’ve done a lot of single mixer setups ….
Frequency histogram [4]
The histogram also hints to some none Gaussian behavior of the noise. The total width of the histogram seems comparable to the Small DMTD
Auto correlation of the frequency [5]
The frequency auto correlation clearly shows there is some repetitive signal hidden in the frequency data.
The frequency of the signal, about 1/120 Hz, suggests this could be due to the 0.01 Hz resolution of the digital PLL that locks the internal reference to one of the inputs. As the PLL is frozen during measurements the frequency difference between input and internal reference could easily be 1/120 Hz
As many have noted, I have a “thing” about anything other than a pure
sine signal as the mix down source. Yes it’s a bias. Between spurs and
steps things like digital PLL’s and DDS’s ( I believe ) can create issues.
I don’t have an closed form proof, just observations. There are other points
of view and they could be right.
If you listen to Sam Stein and John Miles in their public posts, they both
hint very strongly that the 32 bit math in their down conversion processing
is a really big deal. It suggests that both ran into issues there and came
up with a “better way”. Not exactly a DDS, but very close.
Any errors in the above will be blamed on the coffee pot being a bit slow
to warm up this morning ….
Bob
Hi Erik, it looks like a separate circuit could be designed for this device to further improve performance.
I can provide some nanovna for you to test.
Is it feasible to use two centsdr-like switches to obtain IQ signals instead of using a mixer?
From: Erik Kaashoek
Date: 2022-10-05 17:54
To: time-nuts; Bob kb8tq
Subject: Re: Hybrid analog digital dual stage DMTD based on nanoVNA HW. Getting closer to 1e-13 frequency resolution
After some more work it became possible to measure the coherent noise
floor in a similar way as was described in "A Small Dual Mixer Time
Difference (DMTD) Clock Measuring System" (Small DMTD) by W.J. Rely
Both inputs where connected to separate outputs of a two channel DDS.
This allowed checking the frequency and phase measurement by testing
with various frequency difference and phase offsets. The frequency and
phase measured for all frequency and phase variations as expected and
it was fairly simple to measure the minimum phase step of the DDS (0.006
degrees)
For the coherent noise floor tests the frequency difference was set to zero.
Coherent phase data [1]
Over a roughly 45 minutes period the relative phase slowly varied over a
+2ps/-5ps range, comparable to the Small DMTD.
The measurement is susceptible to temperature variations as the
prototype does not yet have an enclosure. Draft from opening doors had
an observable impact, most easily spotted in the phase data.
Frequency data [2]
The frequency appeared to be stable, no linear trend was observed. At
first sight the frequency noise appeared to be less then the Small DMTD
but the Small DMTD did measure the frequency at a lower tau of 0.1 second.
Frequency stability [3]
Although no long term drift was present the slope of the stability plot
suggests there is still some hidden issue. The frequency stability of
the Small DMTD is clearly better.
Frequency histogram [4]
The histogram also hints to some none Gaussian behavior of the noise.
The total width of the histogram seems comparable to the Small DMTD
Auto correlation of the frequency [5]
The frequency auto correlation clearly shows there is some repetitive
signal hidden in the frequency data.
The frequency of the signal, about 1/120 Hz, suggests this could be due
to the 0.01 Hz resolution of the digital PLL that locks the internal
reference to one of the inputs. As the PLL is frozen during measurements
the frequency difference between input and internal reference could
easily be 1/120 Hz
Erik.
[1] http://athome.kaashoek.com/time-nuts/DMTD/phase.PNG
[2] http://athome.kaashoek.com/time-nuts/DMTD/freq.PNG
[3] http://athome.kaashoek.com/time-nuts/DMTD/stability.PNG
[4] http://athome.kaashoek.com/time-nuts/DMTD/histo.PNG
[5] http://athome.kaashoek.com/time-nuts/DMTD/autoc.PNG