[time-nuts] Re: Using a nanoVNA as DMTD, simulation only

[Erik Kaashoek]

Hi Bob,
Indeed, see attached image with -80dB leakage of second signal at 0.1 Hz
offset.
The weird spikes are caused by 16 bit resolution of most of the calculations
[image: image.png]
Erik

Op wo 21 sep. 2022 om 00:20 schreef Bob kb8tq <kb8tq at n1k.org>:

> Hi
>
> With signals in the < 0.1 ppb offset range, you should see effects at the
> -80 db isolation level. They should show up as ripples in what otherwise
> should be a straight line ( ADEV drops vs tau in a straight line ….. ).
>
> Bob
>
> On Sep 20, 2022, at 11:12 AM, Erik Kaashoek <erik at kaashoek.com> wrote:
>
> Bob,
> Thanks for the hint.
> After adding overlapping ADEV calculation and extending the simulation to
> a 100 seconds measurement period I did some simulations using -80 dB to
> -120 dB leakage of another signal at 10, 1, 0.1, and 0.01Hz difference and
> varying the noise level up to -80 dBc/Hz.
> Worst case is a delta frequency of the two inputs of 1 Hz and noise and
> leakage at -80 dB but even under these conditions the ADEV at tau of 1
> second stays below 1e-12.
> Given the above leakage and noise conditions the minimum reliable
> observable frequency difference is 1e-5 Hz which is very promising.
> The nanoVNA (or its better cousin the LiteVNA) do have at least 80dB
> isolation between the inputs so I'm tempted to implement this on the actual
> HW for validation.
> Given the HW, without modifications, it can only work for (almost) equal
> frequencies but this should be sufficient for many relevan use cases.
> One area of concern are the close-in spurs of the SI5351 when used at
> small offsets from 10MHz. Too difficult to simulate.
> Erik.
>
> Op ma 19 sep. 2022 om 22:02 schreef Bob kb8tq <kb8tq at n1k.org>:
>
>> Hi
>>
>> The “typical” gotcha doing this is channel to channel isolation.
>> Folks have tried it with various devices and that seems to be
>> the first barrier they run into. There may be others further down
>> the road …..
>>
>> Often tossed up isolation numbers from various sources get into
>> the > 120 db range for signals that are very close to the same
>> frequency. If they are not close, then you start talking about how
>> close this or that harmonic is.
>>
>> Simple test is the same one you now are very familiar with. Step
>> one input across the other and see what happens ….
>>
>> Bob
>>
>>
>> > On Sep 19, 2022, at 10:47 AM, Erik Kaashoek via time-nuts <
>> time-nuts at lists.febo.com> wrote:
>> >
>> > After reading about DMTD and how the VNWA is doing frequency
>> measurements I
>> > was curious if it would be possible to use a nanoVNA to create a DMTD by
>> > only changing the SW.
>> > The nanoVNA has two input channels (S11 and S21) and a reference
>> channel.
>> > By disabling the output of the reference LO in SW the S11 and S21
>> channels
>> > become two independent inputs. One via the reflection bridge (S11) into
>> a
>> > mixer and one directly into another mixer. Both mixers also have the
>> > offset_LO as input which should be tuned so both mixers output close to
>> the
>> > IF frequency.
>> > The output of the mixers is converted using a 16bit stereo ADC running
>> up
>> > to 96kHz. The 16 bit samples streams are converted to phase and
>> amplitude
>> > by doing a SW I/Q downmix to DC.
>> > The number of samples to combine into one phase/amplitude measurement is
>> > defined in the SW.
>> > As I did not want to put a lot of effort into creating embedded SW I
>> > created a one input channel simulation in Octave of the processing
>> after AD
>> > conversion.
>> > The simulation uses a 1kHz input signal with added noise and a 48kHz
>> sample
>> > rate and combines 1k samples into one angle measurement. All sample
>> data,
>> > I/Q data, cosine and sine tables are rounded to 16bits as used in the
>> > nanoVNA. The 48 angle measurements per second limit the frequency
>> > difference between the input signals and the tuned frequency because if
>> the
>> > frequency difference is too high the unwrapping of the angle will fail.
>> > After unwrapping the 48 angle measurements per second a linear
>> regression
>> > uses the angle measurements to calculate the angular speed per second,
>> > dividing this speed by 2*pi gives the frequency deviation of the input
>> > signal from the reference signal.
>> > It would also be possible to output the 48 angle measurements per second
>> > (or any subsampled number) as raw phase difference measurements and do
>> the
>> > rest of the processing in something like Timelab
>> > Using 48kHz sample rate and 16 bit accuracy of the data and an added
>> noise
>> > level of 1e-5 (is this -100dBc/Hz (?)) the minimum observable delta
>> > frequency in the simulation is about 1e-6Hz. Any lower delta frequency
>> > falls below the 16 bit numerical resolution. A higher noise level, such
>> as
>> > 1e-4,  hides the 1e-6Hz difference.
>> > To make a complete DMTD one would have to do this angular measurement
>> for
>> > both channels and subtract the measured angle.
>> > It is assumed the internal reference cancels out in a dual channel setup
>> > comparing the two inputs so the simulation assumes a perfect internal
>> > reference.
>> > Some questions.
>> > 1: The measurement of the angle (phase) is actually a combination of 1k
>> > samples over a 1/48 second period. Is this a valid way to measure the
>> phase
>> > of an input signal? A frequency offset will cause phase rotation over
>> the
>> > measurement period. Is this causing systematic errors?
>> > 2: With a 10MHz input signal and a minimum observable frequency
>> difference
>> > of 1e-6Hz over a one second period the frequency resolution with a "gate
>> > time" of one second seems to be in the order 1e-13. Could this be
>> correct?
>> > Is the noise level realistic? Would this translate into a phase
>> resolution
>> > of below 1 ps or am I making a big mistake?
>> > Erik.
>> > _______________________________________________
>> > time-nuts mailing list -- time-nuts at lists.febo.com
>> > To unsubscribe send an email to time-nuts-leave at lists.febo.com
>>
>>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 30946 bytes
Desc: not available
URL: <http://febo.com/pipermail/time-nuts_lists.febo.com/attachments/20220921/12e4bd17/attachment.png>