Fun with crystal filters

[joeqsmith]

In response to one of my videos for the NanoVNA,  I was asked about making a video about crystal filters.   From the attached, they are planning to attempt to make an 8-pole filter using 12MHz crystals.   They want to use the DISHAL software with something called the G3UUR method.   I thought rather than make a video, I would invite them to join the EEVBLOG and work together with them in the forums.   

I have downloaded the DISHAL2052 from here: 
http://warc.org.uk/?page_id=387

[9aplus]

8 pole can be the real challenge 

Once upon the time, built this 5 pole 1,6 MHz LSB xtal filter for SEA222 marine transceiver...
The adjustment was nice experience...

Big can is original USB filter, relays switching USB/LSB
On exit of 5 pole filter one stage of IF amplification to make equal level to original USB filter


Testing wihtin transceiver

[joeqsmith]

Nice work. 

Without more details from the OP on their particular requirements and not having a need for this filter myself, I may just put something together.  I glossed over the help files included with the DISHAL software.   I have the 2015 edition of the ARRL handbook and read the section on crystal filters.  They offer a few simple circuits to evaluate the crystals.   

A few other references I glossed over:

Ladder crystal filter design, J.A. Hardcastle
http://warc.org.uk/wp-content/uploads/2014/01/RadCom-Feb_1979.pdf

Crystal Ladder Filters for All, Jack A. Hardcastle
http://www.arrl.org/files/file/QEX_Next_Issue/Nov-Dec_2009/QEX_Nov-Dec_09_Feature.pdf

Quartz Crystal Parameter Measurement, Jack Hardcastle
http://warc.org.uk/wp-content/uploads/2014/01/QEX-Jan_Feb_2002.pdf


***
This paper may be of some help.  They talk about the G3UUR Colpitts oscillator and offer what they consider an improved test jig.  The OP plans to use the G3UUR technique and that is most likely what I will do.

http://jvgavila.com/other/Crystal_Test_Set.pdf

[joeqsmith]

I was hoping the OP would join. They did provide one key bit of info.  It seems they only have a NanoVNA for test equipment.   While I wait to see if they plan on expanding their lab, I made some progress.

Attached is a circuit from the 2015 ARRL handbook.  I spent some time tonight wiring up most of it, taking liberty to sub values for what I had. The multimeter stage is missing.  The software seems to allow you to enter the values used for the oscillator.   It looks like it will require a fair amount of time to measure a single crystal.  It takes a few minutes for the temperature to settle after handling the part.     

[joeqsmith]

Finished assembling the oscillator and picked up 16 X 12MHz crystals.  Next comes the boring part of measuring them. 

Most of my equipment, including the counters use a GPS reference.  I will be using the counter and meter to make these measurements.   

It seems the OP wants to try and only use the Nano which I'm not sure why they mentioned the G3UUR method.   After I measure them, I am try and see if the Nano would be good enough to directly measure them.  I've never looked to see what sort of resolution it has or how stable it is. 

[joeqsmith]

Sorry about the old software.  I have yet to rewrite the program for my counter to support Windows 10.    The counter uses an HP GPS receiver as its reference.  This runs 24/7.   The counter can resolve down to 1Hz.       

I've asked the OP what BW they required but have not yet heard back from them.  The software states no more than +/-2% of the BW when matching the crystal's frequency.  At 2KHz, that's 40Hz of 12MHz which seems tight even 60Hz for a 3KHz BW would seem difficult.  Assuming the parts are screened on the bench, I need to make sure they are stable before taking a reading.  The first thing I noticed is that they require a very long time to settle after being handled. 

1: showing a 3.6865MHz crystal in the test jig running 5 minutes after about a half hour warmup.  The jig seems stable within 1Hz. 

2: showing the drift after handling a second part while installing it.  This is roughly 8 minutes.  It's a good 4 minutes for it to become stable within 1 count.

3: showing a 12MHz crystal after handling.  We can see it required even more time to settle. 

4: I held onto the next 12MHz crystal for 20 seconds or so to get a better idea on the how long I need to wait.   The 12MHz crystals seem sensitive to air currents so I tossed my calibrated beach towel over the setup which seems to have helped with the low frequency wander.   The jig is using silver mica for the oscillator and the offset caps.   It seems that the drift is these higher frequency crystals.

I installed a few of the 12MHz parts to get an idea how much they very part to part.  The are all over the place.  Same brand, date code and I assume lot.  The software talks about a sub program used to compensate for the variances in the parts.  They provide an example.  I doubt I am going to find enough crystals in this small batch to make up the eight needed.   

[joeqsmith]

Allowing the covered 12MHz to run, it's pretty poor.   The 3.68MHz parts are much better.   I may need to order up some better parts. 

[joeqsmith]

Converted the Labview software to use my custom Ethernet calls to that GPIB controller.  This will be nice to have. 

These are some 11MHz FOX parts.   Again, looking at the settling time, using the calibrated beach towel.  It appears their tempco is all over the place and could eat into the margins pretty quickly.   Of course its not a very controlled test just handling them to install into the jig.    The third part I looked at seems very stable with temperature.     

Digikey has some 12MHz, 10ppm parts in-stock.     

[joeqsmith]

After double several more minutes, it seems to remain flat.  Just a very good part from this batch.   Screening them is going to be a nightmare with these long settling times. 

[joeqsmith]

The circuit takes time to settle after throwing the switch.  So it's taking several minutes per part.  Basically I install the crystal, toss a towel over the setup and wait for the plot to flatten out.   If I were going to attempt to run several crystals, I would automate the collection.  Still without some way to autoload the parts, you have to babysit.    So far, I see the little adventure as being penny wise and pound foolish.   Normally the home projects I do are for the fun of it.  Sorting data on crystals is no fun and so far, it looks like I would need a lot more parts to find 8 that would match.       

[joeqsmith]

After measuring all 16 parts, I went back and repeated the first and second parts.  I was within 5Hz.  These were made by CTS and are 11MHz rather than the OPs target of 12.  The 12MHz parts I had were much worse.  Data sheet:

https://www.ctscorp.com/wp-content/uploads/MP.pdf

I could roll an 8 crystal filter with a BW of 5.6KHz or tighten it down to 2.8ish KHz with 4 crystals.   It seems the OP is not sure what they are after.   They never mention details of what they require.   The goal could be to find something sharp enough to make it impossible to measure with the nano.  Who knows.    They claim to be a beginner so I am not sure why they would be attempting to construct such a filter in the first place.   

I think for the little entertainment I may find, I will go ahead with a wider 8 crystal filter. 

[joeqsmith]

Looking through my stash of caps, the 8 pole is out.   The four closest crystals are within +/-24Hz.   Rather than dealing with the match, adjusted the BW to provide 50ohmish.  This worked out to 1.2KHz.   The caps should be easy enough to trim and it should fit onto one of my test boards.   

[joeqsmith]

The ripple is about 3dB, BW is about 1540Hz.   With a 10KHz span, like the simulation, the rolloff is not near as good.      See posts below.

[joeqsmith]

In order to get this sort of narrow band resolution, I had to change the scale on my software.   

Shown with a 10KHz span and at 2KHz.   There's a fair bit of error between the two system but we are talking about a $50 unit being compared with in the 1997 HP Test and Measurement Catalog listed for $24,750.         There was a problem with my setup.   See the following posts. 

****
I have a very old HP8754A VNA.  This system can run down to 4MHz but there is no way it can make a narrow sweep like this.  When I have done it in the past, I have used a separate signal generator. 

[joeqsmith]

For a comparison, an old SHOWA 9MHz 2.2KHz filter, model SF0922B.   

[joeqsmith]

I was thinking about why there would be such a difference between the simulation of the 4-pole filter and what I measured, especially for the ripple.   The Nano actually seems closer to the simulation than the HP.     I wondered if I had selected the wrong impedance but what it turns out, I was overdriving the filter.  Backing down the source and things smooth right out.   

The Nano and HP are now looking pretty close.    

[joeqsmith]

Another good article:
http://www.rfcafe.com/references/electronics-world/crystal-filters-electronics-world-april-1969.htm

Showa filter along with matching transfomer. 

[joeqsmith]

I attempted to make a 10-pole 12MHz filter but the results were less than impressive.   As you can see, it is not very flat.   

Looking at a commercial filter, it was less than $30 and I would guess much higher performance.

[joeqsmith]

RBW still at 150Hz, allowing it to average for 100 sweeps.   

[joeqsmith]

My software could use some improvements (higher res for pretty much everything, more cursor features...).  Still, we can see we have something very similar. 

[SilverSolder]

I have an old HP 3571A Tracking Spectrum Analyzer here, from the 1970's, a boat anchor that I like to play with sometimes...

It uses a 5 pole crystal filter to do its magic.  Each section of the filter is on its own plug-in board, dripping in gold and oozing crazy quality goodness as you'd expect.

It goes down to a bandwidth of 3Hz and is still accurate today, as far as I have been able to ascertain.

Crystals...  gotta love them!

[joeqsmith]

That's a nice old bit of equipment.  I have a few old systems myself still in use.   I've had a lot of fun playing with the little NanoVNA.   Attached shown with the 10 pole filter but using a low noise amplifier with a step attenuator to get a bit more range from the Nano.     

That 3589A network analyzer can go down to 0.0045Hz when using the narrow band mode.  The UI is well thought out.       

[joeqsmith]

I added a narrow band plot feature to subtract off the center frequency.  Here I have let the Nana sweep the filter with a 10KHz span for roughly an hour.  The plot is +/-5KHz from the center.  Again, using the low noise amplifier.   The Nano seems stable enough.   

I'm not sure why the OP mentions the G3UUR method if they wanted to use only the Nano.   It seems the next step is to try and measure a few of these crystals using the Nano.   

[SilverSolder]

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?

[mark03]

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.)