Fun with crystal filters

[joeqsmith]

From the one paper I had linked, the which has the equations for the G3UUR method, it appears to be correct. 

Attached comparing the data for all thirteen Fox crystals using the G3UUR and 3dB methods.  It appears there's about a 12% error between the two setups.  I can believe this.   Looking again at my particular Nano's clock, at 3.6865MHz where we are working, it is off by 1Hz compared with the GPS.  My guess is most of the error is from my two different jigs.  While I am sure I could tweak them to get the results to match, it's rather pointless.    The next step would be to try and procure some sort of crystal reference standard like the one shown (aunders & Associates Reference Crystal, model 7000022).   

Then again, how good does it need to be? 

[joeqsmith]

Just a quick measurement of CSW in-circuit  and using this in place of the 50pF appears to cut the error in half ( 6% ).   

[joeqsmith]

Measuring the crystal socket mounted into the perfboard,  1 get 1.68pF.   Adding this to each crystal's capacitance brings the peak error below 3%.   Interesting is the outlier is crystal #3.   Notice this crystal also had double the p-p and much higher standard deviation than the other 12 parts we looked at while in the 3dB jig.   

With #3 in back in for a third time, I let it settle much longer.   While the stddev and p-p are now on par with the other parts, the results were too close to call.   Both figures of merit are calculated from a cold start, unless I reset the calculation during a run.   So, they can be very wide and still have a good measurement.  I should automate this part of the test.   

I then reinstalled the part into the oscillator and let it settle.  Sure enough, it was off a few Hz.    This brings the error down to 2%.   Still, I have no idea about the absolute accuracy. 

[joeqsmith]

With most of the problems solved, I went back to the 12MHz CTS crystal.   To try and help it along, I cut the leads down to about 10mm.

First, I ran the 3dB test jig on the HP analyzer and measure -1.75dB and 690Hz 3dB down.  Just to have a sanity check.  I recalculated the Lm and Cm using the new capacitance values (manually adding the offset to the holder). 

I then tested the part on the Nano.  Notice that I am still only using a single 101 point sweep.  Both the p-p and standard deviation are slightly improved over what we had previously measured using a 100Hz segments. 

Note that peak is spot on to the HP (it should be) but the 3dB width is slightly wider on the Nano. 

Both Lm and Cm have under 6% error.    It looks like the Nano could be used even at 12MHz depending on the filters requirements. 

[joeqsmith]

The Nano continued to collect data for the CTS 12MHz crystal for roughly an hour to get a better idea what the profile of the noise looks like. 

[joeqsmith]

Shown attaching the 3dB test jig to the Agilent with the same CTS 12MHz crystal.   The noise is much lower.  Obviously, if you are trying to match crystals withing say 40-60Hz, you don't want your test equipment eating up 100% of the error budget.   The 689.34Hz BW is pretty much spot on to the other HP which measured 690Hz.   
   
Both Lm and Cm values are pretty much the same as we got with the Nano.       

I've tried to find a source for a reference crystal but no luck.   I may try and contact that same company and see if they would sell me a few characterize at  different frequencies.  Maybe 5 and 15MHz. 

https://www.saunders-assoc.com/resources/datasheets/referencecrystal/6300338.pdf

[joeqsmith]

I suspect if we were to ditch the resistive pads we could improve the 12MHz S/N.    The plan is to include the attenuators on the board if it shows any promise.   These were the smallest cores I had on hand that would work in these lower frequencies.

[joeqsmith]

This is more like it.   Again, just a single 101 point sweep, same span.   It's obviously not calibrated as I was really only interested in seeing if it would improve the resonance frequency detection.   So if you plan to use a Nano to sort your > 6MHz crystals you may want to consider something other than resistive pads for your test jig.

[joeqsmith]

I plan to finish building up the new jig today but I'm at a loss why it helps.    The 12MHz parts have a lower loss than the 4MHz parts.   The cores do offer some level of low pass filtering but when I had installed the LC filter, it had no effect.   Early on I had tried to run it with an LNA, step attenuator and filter.   We also know the same padded type jig works fine with both of the higher end  network analyzers.     Looking at the signal from the Nano with a scope and SA at 4 and 12MHz, I don't see a difference that would cause a problem.  However we are looking at 0.1ppm levels.   

One other experiment I tied was just connecting the part directly to the Nano in series as I had shown in the first video.  The results were very poor.   However, part of the problem is not knowing what a good result is.   For now I am assuming that with the two methods (3dB and freq shift) yielding results within 2%, they are valid but that may not be the case.   

So there are a few clues.  If you have any ideas on the cause, I would like to hear about them.   

[SilverSolder]

How does the calibration work with one of these - do you have a short/open/load type of jig?

[joeqsmith]

How does the calibration work with one of these - do you have a short/open/load type of jig?

Back a page I show the standards (loose use of the term)  made to cal and run a sanity check on my software and jigs.   Because the OP had asked about 12MHz, I doubt I will do any testing beyond this.
https://www.eevblog.com/forum/rf-microwave/fun-with-crystal-filters/msg3064340/#msg3064340

That last paper I linked from HP covers the calibration.  It's not  a long paper and covers the basics.   

[joeqsmith]

Used a drop of epoxy to hold the cores to the PCB and then another drop to hold the connector to the cores.   Then added the two attenuators.   

I placed a step attenuator between the fixture and the Nano's CH1 (port2).   The first 1100 samples are with 0dB attenuation, then changing to 10dB until 1300ish samples where it was set to 20dB.   Notice that the 3dB BW is very close. 

[joeqsmith]

If I cal the Nano with the standard SOLT and consider that 0dB.

Gen I  Resistive pad jig with a short is about -30dB   
Gen I  Resistive pad jig with a 3.686MHz crystal is about -37.3dB 
Gen I  Resistive pad jig with a 12MHz crystal is about -31.3dB 
Gen I  Resistive pad jig with a open with Nano sweeping at 12MHz is about -80dB 
Gen II Transformer matching w/6dB jig  with a 12MHz crystal is about -8.6dB   

The question is why the resonate frequency of the 3.7MHz part is so stable compared with the 12MHz part when using the Nano. 

If I reset the Nano's internal calibration, then install a thru and scan small segments, there is something happening at 12.8MHz.   I've been searching down to 4MHz and a bit over 13MHz and it only appears to show up at 12.8.   I wonder if what ever is causing this is tied into our problem.   

[joeqsmith]

Running the 12MHz crystal in the Gen II jig for an extended time.  Noise is now 5Hz p-p.  An improvement of about 10X.   The noise has a nice shape to it.   Note the Lm and Cm values.   

[joeqsmith]

The Gen II jig still had some issues.  The core material wasn't that great.  I  had two separate windings  to get my turns ratio rather than a bifilar.   This really hurt the performance. 

These are the cores for Gen III.  I was originally not going to use them because of their size.   The right thing to do would be to buy something but I had these left over.   The magnet wire shorted on one of the cores, so Teflon it is.   

Looking at the 12MHz crystal with the crude setup,  things look decent.  The jig's 3dB point is was around 40MHz but once it was assembled it moved up to 653MHz.   This is much better than the previous setups. 

Waiting for the epoxy to cure and then we can try it out.  We should be able to have a look at some higher frequency crystals now. 

***** UPDATE *****

The original data transmission plot after assembly had a very odd shape to it.  At this time, the epoxy was wet which may have been the cause.  I recollected the data today and it looks smooth as you would expect (JIGIII_6).

[joeqsmith]

This 50MHz crystal is the highest I have.  I suspect its a bit of a worst case condition.   This is still only using a single 101 point sweep.    The part was drifting a bit but the Nano seems to track it fairly well.   

[joeqsmith]

Running a 16MHz part for an extended period.  7.7Hz with a bit of a wobble at the start.  The standard deviation says it all.   

[joeqsmith]

Shown with the GenIII jig attached to the HP.   I did not allow the HP or crystal to warmup and nothing was covered.   

As we would expect, the standard deviation is much better with the HP but the parameters are pretty much the same.   

[joeqsmith]

I wrote Saunder's today to inquire about their crystal standard.   They haven't responded but may be closed with the shutdown.   

With the new fixture working, the only problem now is measuring C0.   The Dishal software requires this and up till now, I have used the RLC meter to measure it.   The plan is to have the Nano measure it as part of the other measurements.   It would be easy enough to change the setup to make the measurement but I would like to do it all in one step. 

First thing was to make a new set of standards for this fixture (shown).   The copper block and connector is just a heatsink for soldering these.   

[joeqsmith]

This is the first time I have ever tried to use a VNA to measure the thru impedance on the backside of two impedance matching transformers.   After looking at the raw data, it seemed the Nano could at least detect a change from 5.6 to 10pF.     
 
Next I measured all of my standards with my two BK RLC meters.  The standards vary from 5.6pF to 100pF.   Trying each part with the Nano using the Gen III jig, surprising enough, seems to work fairly well.   Obviously we need to change the frequency to make this measurement but that's not a problem. 

Plot showing one of the 3.686MHz Fox parts.   You can see it reads about 0.1pF high compared with my one BK.   The RED graph is plotting the capacitance to give me an idea how stable the reading are.   The stats are valid, so 0.26pFp-p error.    Not too bad.

I still have not heard back from Saunders about the reference crystal.     

[joeqsmith]

Measuring C0 of one of the 12MHz crystals and then installing a small 50pF trimmer.   The Nano's update rate is VERY slow.  This is the cause of the vertical steps, not the resolution of the calculation or the nano's hardware. 

***
added correct plot

[joeqsmith]

Sweeping a larger trimmer capacitor.   

[SilverSolder]

What's the software you use to display the charts - is it LabView or something like that?

[joeqsmith]

What's the software you use to display the charts - is it LabView or something like that?

https://www.eevblog.com/forum/testgear/national-instruments-labview-home-bundle/msg2574126/#msg2574126
https://www.eevblog.com/forum/testgear/advantagesdisadvantages-ni-labview-etc/msg2348991/#msg2348991

The attached table from CopperMountain shows the equations for the three types of impedance measurements.  One of the papers I read, they were showing similar experiments using another low cost VNA.  They comment about directly reading C0 and toss out some 80% number.   I was guessing they were suggesting measuring 80% from Fs.    During my experiments, I am using a fixed frequency of a MHz.  This limits the low side to about 1pF as previously shown.   80% may be a bit close but I tried one part sweeping from 30-60% and this seems reasonable.   
   

• Math is done with 3 point calculus (max signal, and the -3dB points). I have been looking into
Levenberg-Marquardt curve fitting, but the software found so far has not got high enough
precision to examine its virtue.
• No C0 parameter measurement yet. This could be done by measuring at 80% of fundamental
frequency or 80% of the 3rd overtone. But the required math is not fully clear to me yet. Also a
measurement of the parasitic fixture capacitance is needed (easily 0.5 to 1pF).
• The .csv file uses a comma (',') as column separator. Ideally this should be retrieved from the
Regional Settings. Also it is best not to put a comma or quotes in the log code field as this might
upset the import into you spreadsheet software.
Related

 

[SilverSolder]

What's the software you use to display the charts - is it LabView or something like that?

https://www.eevblog.com/forum/testgear/national-instruments-labview-home-bundle/msg2574126/#msg2574126
https://www.eevblog.com/forum/testgear/advantagesdisadvantages-ni-labview-etc/msg2348991/#msg2348991

The attached table from CopperMountain shows the equations for the three types of impedance measurements.  One of the papers I read, they were showing similar experiments using another low cost VNA.  They comment about directly reading C0 and toss out some 80% number.   I was guessing they were suggesting measuring 80% from Fs.    During my experiments, I am using a fixed frequency of a MHz.  This limits the low side to about 1pF as previously shown.   80% may be a bit close but I tried one part sweeping from 30-60% and this seems reasonable.   
   

• Math is done with 3 point calculus (max signal, and the -3dB points). I have been looking into
Levenberg-Marquardt curve fitting, but the software found so far has not got high enough
precision to examine its virtue.
• No C0 parameter measurement yet. This could be done by measuring at 80% of fundamental
frequency or 80% of the 3rd overtone. But the required math is not fully clear to me yet. Also a
measurement of the parasitic fixture capacitance is needed (easily 0.5 to 1pF).
• The .csv file uses a comma (',') as column separator. Ideally this should be retrieved from the
Regional Settings. Also it is best not to put a comma or quotes in the log code field as this might
upset the import into you spreadsheet software.
Related

I am no math wiz, for sure...   What is Z0,  is that the reference resistor used during the test?

Edit:  This slide from Agilent seems to show 3 methods of measuring impedance as well - are they the same as the Copper Mountain ones?