Fun with crystal filters

[joeqsmith]

The Nano VNA does look interesting -  I don't have room for more boat anchors, so maybe it's time to look at some of the things that are possible in 2020AD!   

Where is a good place to buy a good Nano VNA?

Personally, as much as I like the little Nano you may want to have a look at the Nano V2.  As I understand it has a wider dynamic range, doesn't use harmonics and a much higher operating frequency.   I have yet to get one so I can't comment on how well it performs.   

There's a decently sized group dedicated to the original Nano.  You may want to check that out and see if it's worth your time to join up.

The original Nano is much better than I would have ever guessed but as you can see from various posts, it has it's downside.  I just choose to ignore the shortcomings, because it's $50. 
 
I think Flipper bought them off Amazon.   

[joeqsmith]

It would be cool to implement a curve-fitting crystal measurement scheme like this for the NanoVNA:

https://ieeexplore.ieee.org/document/1539973

(I would post a sci-hub link but I'm not sure if it's allowed.)

Sorry but I don't have an account.   

I have not looked into what the current standard practices are.   What is the advantage of this method?   Can you summarize the concept? 

[joeqsmith]

A couple of other links showing various fixtures.   

http://pages.suddenlink.net/wa5bdu/crystal_slide_show.pdf

https://www.nonstopsystems.com/radio/pdf-ant/antenna-article-xtal-1cqp.pdf

I would have thought that the miniVNA would have been a squarewave like the Nano and they would have had a filter as part of that jig.   

***
A few more

https://www.hongkongcrystal.com/wp-content/uploads/images/AppNotes/ANENG-EQP-000.pdf
https://oliverbetz.de/cms/files/Artikel/Quarzoszillator/technicalintro_crystals_e.pdf

[mark03]

Sorry but I don't have an account.   

You can copy and paste the DOI into sci-hub (google it) if you don't have institutional access or don't want to pay whatever ridiculous fee they are asking now.  Legality depends on where you live, but IMHO the morality does not.  I say this as a published IEEE author (same journal, even!).

I have not looked into what the current standard practices are.   What is the advantage of this method?   Can you summarize the concept?

Here's an interesting summary and comparison of different measurement schemes:  https://www.mikrocontroller.net/attachment/195160/Crystal_Motional_Parameters.pdf

I haven't studied this enough to give you a good summary, but AFAIK the "admittance circle" method is the current standard and is based on that IEEE paper.  It uses data taken at many different frequencies to curve fit the parameters, which seems like common sense nowadays.  Most of the older methods use only "landmark frequencies" (-xx dB points, 45-degree phase-shift points, series and parallel resonance frequencies, etc.), and some do not assume that vector measurements are available.  So for a modern approach I think the 1987 paper might be a good starting point.

[joeqsmith]

Sorry but I don't have an account.   

You can copy and paste the DOI into sci-hub (google it) if you don't have institutional access or don't want to pay whatever ridiculous fee they are asking now.  Legality depends on where you live, but IMHO the morality does not.  I say this as a published IEEE author (same journal, even!).

I have not looked into what the current standard practices are.   What is the advantage of this method?   Can you summarize the concept?

Here's an interesting summary and comparison of different measurement schemes:  https://www.mikrocontroller.net/attachment/195160/Crystal_Motional_Parameters.pdf

I haven't studied this enough to give you a good summary, but AFAIK the "admittance circle" method is the current standard and is based on that IEEE paper.  It uses data taken at many different frequencies to curve fit the parameters, which seems like common sense nowadays.  Most of the older methods use only "landmark frequencies" (-xx dB points, 45-degree phase-shift points, series and parallel resonance frequencies, etc.), and some do not assume that vector measurements are available.  So for a modern approach I think the 1987 paper might be a good starting point.

Googled as suggested.
https://scholarlykitchen.sspnet.org/2018/09/18/guest-post-think-sci-hub-is-just-downloading-pdfs-think-again/

I would rather make sure the people involved with creating the content are paid.  If you find where the original author has placed it into the public domain, I will have a look.     The DISHAL software that the OP plans to use only supports the  G3UUR and 3dB methods.    I have not spent a lot of time looking for other crystal filter software.  There may be others available that use the method you are asking about.   

I like that last paper you linked.  When I was measuring the crystals with that home made oscillator jig, I mentioned once I was done, I remeasured the first few parts.  I was always plotting the frequency and watching it settle.  Many papers talk about waiting < minute.  Some talked about not touching the parts when installing them.  Some talked about the parts heating as they were ran causing some initial drift.   I know my setup was fairly stable with the beach towel and waiting several minutes until the parts settled.    Your paper makes no mention of it and I wonder how repeatable their setup was. 

Seeing their standard crystal, I had a look at Saunders site.   I wasn't able to find where they sell these standards but noticed this paper:
https://www.saunders-assoc.com/resources/datasheets/article/Load%20Resonant%20Measurements%20of%20Quartz%20Crystal.pdf

[mark03]

I respect your position and won't belabor the point; I will only mention that none of the money paid to access papers goes to "content creators."  The content creators (as well as the editors and the peer reviewers) are all volunteers, working for free.  There was a long discussion of this issue (open-access vs closed-access academic publishing) here:

https://www.eevblog.com/forum/oshw/the-problem-i-have-with-closed-(source)-papers/

This is a completely different moral question than, say, pirating music and movies.  One may certainly come down on the same side (the side of the law) in both cases, but not because the moral dimensions are comparable---they really aren't.

Anyway, sorry for the OT rant

I'll keep on eye on this thread.  You have encouraged me to develop a crystal-measurements script for my VNA, if I ever find the time.  Work is still busy, which I know I should be thankful for.  Sometimes I do envy those with the freedom to work on hobby projects however.

[joeqsmith]

Using the Nano, output to a lowpass filter, to a step attenuator, to an LNA, to a 10dB atten, to the DUT, to a 10dB atten, to the input....

Span set to the minimum allowing to see the 3dB from the peak.  I let it run for a while then gently squeezed the crystal with my fingers.  With a 100 points and just looking at the data, you can't see a shift.  The plot below is holding a heatgun to the part for several seconds.   

[joeqsmith]

I respect your position and won't belabor the point; I will only mention that none of the money paid to access papers goes to "content creators."  The content creators (as well as the editors and the peer reviewers) are all volunteers, working for free.  There was a long discussion of this issue (open-access vs closed-access academic publishing) here:

https://www.eevblog.com/forum/oshw/the-problem-i-have-with-closed-(source)-papers/

This is a completely different moral question than, say, pirating music and movies.  One may certainly come down on the same side (the side of the law) in both cases, but not because the moral dimensions are comparable---they really aren't.

Anyway, sorry for the OT rant

I'll keep on eye on this thread.  You have encouraged me to develop a crystal-measurements script for my VNA, if I ever find the time.  Work is still busy, which I know I should be thankful for.  Sometimes I do envy those with the freedom to work on hobby projects however.

No matter what arrangement the supplier has with the authors, I see no reason to not pay for their services if I want to use their product.  If the authors want to make their papers open to the public free of charge, that's up to them.   There is certainly many ways to do this today.  I would imagine there is some prestige for the author, or the company they work for to have certain organizations publish their papers.   You may find the authors or the companies they work for are paid members of these same organizations.    No matter, it has nothing to do with playing with filters.   

[joeqsmith]

Zooming in with a 500Hz span.  We can now clearly see the ripple.   Notice that the heatgun caused a fairly large change in the amplitude, but with the 100 points, its still difficult to see the frequency shift. 

If I ran a segmented sweep (multiple 101 point sweeps over the span) to increase the resolution and fit the data, maybe we could tease out some numbers.   I am not sure from reading the DISHAL help file if one method is better than the other but using the counter, seems like it at least can at least resolve the signal to use it.   That 2% of the BW for a narrow band filter and looking at using 12MHz crystals, seems like the counter is a better solution than trying to use the Nano.   

[joeqsmith]

Lowering the gain to help with the noise.   S11 isn't too bad.   

[joeqsmith]

Going from the Nano to a 3dB attenuator, to the DUT, to a second 3dB attenuator.  Touching the crystal mid run, we can clearly see the shift now with the lower noise.  Knowing they will shift about 10Hz and that we can see it, seems like it could work even with a 101 points.   The match isn't very good this way.  Time to make up a set of pads.   

Added plots.

[joeqsmith]

Sorted some resistors and built up a quick jig for the 3dB method.  Basically the same old same old.   I used the pot method to measure the series resistance as it seems to be the most common method.   

Just a sanity check, using the old cal, match appears much better.  Lm is off a bit between the two but close enough it seems we are in the ballpark.

The Nano software is doing a fit to get the centroid the going 3dB down from that.  It uses interpolation to determine the frequency.  Basically trying to tease out the information from the 101 data points.      Next I plan to measure some of the crystals to see how they vary compared with using the counter.

[joeqsmith]

Having to insert a pot, tweak it, pull it and finally measure it with a meter seems like a poor idea.   I made up a set of standards to characterize the jig.   There is a SOL but also another resistor that is 12.23 ohm to use as a sanity check.   I tried this setup out using that same crystal and get very close to the number I had using the pot.    The next step is add support to the software to automatically make these measurements.

[joeqsmith]

I tried a few different fitting and filtering methods to tease out the crystal's series frequency.  In the attached plot, in the upper right I am plotting the calculated Fs.  It has an error of 50Hzish.  Consider it was a 1KHz span or roughly 1 sample every 10Hz, it's not too surprising.   The 50Hz could very well be more than the entire error budget for a filter.       

It's certainly possible to sweep in segments like I demonstrated when using the Nano to derive a model of a crystal for SPICE.   However the Nano is VERY slow.   

We also need to consider how wide the span needs to be to handle to parts we are trying to measure and the amount of time it will require to perform each sweep.  We may want to use multiple sweeps to track the drift or average the data as well.    As much as I like the little Nano, it seems like a really poor choice of tools for this particular task.  But again, it's $50 so no surprise. 
 
Another thing to consider is the Nano has no reference clock.  If all you have is the Nano, you have no way to know how accurate yours is.    You could spend days collecting worthless data.   

Of course, the other thing to consider is the cost of buying an actual filter you need.  While the OP never did post their requirements, it looks like a 12MHz 8-pole crystal filter with a few KHz BW will set you back around $30. 

The second plot is using the segmented sweep.  The span of each segment is 100Hz.  The total span is a 4KHz.    Even with the added data, I am seeing about a 25Hz error.  Of course 12 sweeps is not much to go on.   Note that the first sweep is off the screen because I handled the part.    It may be good enough in some cases however, it takes so long to collect the data, it seems rather pointless.    A simple counter and GPS is not very expensive, would be much faster and it seems more accurate (than using the Nano).       

Maybe there is a better way to leverage the Nano for narrow band filter prototyping.   If you have any ideas, feel free to chime in. 

[joeqsmith]

I let the NanoVNA run overnight, continuously sweeping the same 12MHz crystal and plotting the calculated series frequency.  With the 4KHz span, it recorded 120 data points in roughly seven hours.  The crystal and fixture are covered with the calibrated/certified beach towel during this test.   The first data point after touching the crystal being tested was ignored to not bias the data. 

Shown is the histogram of the data.  It appears to have a somewhat gaussian distribution.   40Hz p-p noise is poor.   

[SilverSolder]

What is the noise from - the DUT or the NanoVNA?

[joeqsmith]

What is the noise from - the DUT or the NanoVNA?

Hard to say at this point.   I thought one test would be to use a lower frequency crystal.   The results were not what I was expecting.    It looked so good after the first to sweeps, I stopped the test and changed from the segmented sweep back to a single 500Hz span.    With the fitter, we can now easily see the initial drift and the p-p noise, including the drift is 4Hz.    Of course the absolute accuracy is  .  Without a some sort of frequency reference we have no idea. 

So at least it's not a total bust.  Depending what you want to do, it's starting to look possible that we could maybe pull it off with just the Nano.   I would like to try some other brands of 12MHz crystals.  I'm not sure why the OP chose that particular frequency.   It seems high for a narrow band filter. 

[joeqsmith]

Going from 3.7MHz to 5.6MHz seems to make a small difference in the noise but it seems very usable.   

[joeqsmith]

At 7.8MHz, we are now at roughly 13Hz p-p noise. 

[joeqsmith]

With an 11MHz part, we are getting close to 40Hz p-p.  At around 500 samples into the run, I installed a home made 5 pole 12MHz LP filter to see if it had any effect.  Sadly, it doesn't. 

[joeqsmith]

Looking at an old JK Products crystal.   It's interesting how it warms up. 

[joeqsmith]

Another good reference:  "Hewlett Packard Electrical Characterization of Quartz Crystal Units"

http://hparchive.com/seminar_notes/a-119.pdf

Next I plan to try measuring a handful of parts in the range that the Nano seems able to resolve when using the 3dB method.  Then compare these results when using the  G3UUR method.

[joeqsmith]

Odd.  The values are not even close.   Maybe the serial frequency is wrong....   

The nano will run all the calculations automatically and store the data into a CSV file.  The software can then sort them based on the series resonance and pick the best parts for the tightest BW.   

[joeqsmith]

 

After some sleep, I verified my calcs against the Dishal software, then put the jig on the HP and repeated the test for one part.  Plugging the data into the Dishal software, the numbers look reasonable.   The 3dB full width was a mile off between the Nano using my software and the HP. 

After several more tests on the Nano and my software, I found the problem.       The numbers are now very close.   The Nano using the 3dB method is calculating an inductance of 153.7mH and a capacitance of 12.13fF.   Using the  same crystal with the G3UUR method, I get 165.4mH and 11.27fF.   

[joeqsmith]

After correcting the problem, I collected the data a second time.   Shown is the best set of 13 parts for a 4-pole filter.    I've included the data.  The columns match what is shown in the screen shot.

Time to sort out the G3UUR equations so I can let the computer do it's thing.