When I was working for Zeta Labs, circa 1978, I was in charge of a huge
order for 5th overtone VCXO's.  My boss told me to use the "standard
Zeta VCXO design".  It had various problems, and I quickly decided to
invent a new design that actually worked right, and quietly put it into
production without alerting my boss.  The new design was wildly
successful and I used it for many years.  It was successful because
there were no adjustments or "factory select" components and every one
we built simply "worked", with any crystal.  There were no "reject"
crystals.  There was nothing special about the crystals, we ordered them
right out of Croven's catalog.

Start by making a so-called "free running" Colpitts LC oscillator using
a grounded emitter transistor with capacitors from base to ground and
collector to ground and a feedback inductor from collector to base.  I
used a 2N5179 transistor which was "state of the art" in those days, and
would suggest you use an equivalent part in a plastic SMT package.
Don't use a transistor with a higher fT.  Less is more as the saying
goes. Start by making the base capacitor 10 times the value of the
collector capacitor.  If the oscillator has spurious sidebands, make the
base capacitor larger.  Choose the feedback inductor to make the LC
oscillator operate at the crystal frequency.  Construct the collector
capacitor from several standard value (E12) capacitors in parallel, so
you can hit the crystal frequency within a few percent.  Use 1%
capacitors and 5% inductors.  I have only described the RF aspects.  You
can figure out biasing, such as a resistor in series with the emitter,
etc.

When you are happy with this foundation oscillator, disconnect the
emitter from ground and instead connect it ground by using what I will
call the crystal tuning network:  consisting of an inductor in series
with a varactor in series with the crystal.  The emitter goes to the
inductor and the other end of the crystal goes to ground.  Add yet
another inductor, this time in parallel with the crystal.  Choose this
inductor to be parallel resonant with the holder capacitance (C-zero) of
the crystal.  The other inductor should be series resonant (at the
crystal frequency) with the varactor when the tuning voltage is in the
middle of its range.  You need to take care of emitter DC current return
to ground with a a series inductor resistor circuit from the emitter to
ground.

I suggest you start without the varactor network and optimize the LC
oscillator using just the crystal in series with the emitter.  It should
oscillate near the series resonant frequency of the crystal.  Back in
those days, I used an HP4815 vector impedance meter to measure how well
the inductor that resonates the holder capacitor was tuned.  Now a days,
I use any network analyzer and (important) connect the parallel
combination of crystal and inductor from port 1 to port 2, NOT from port
1 to ground.  The goal is the smallest possible value of s21 at the
crystal frequency.

You'll have to empirically determine what hyperabrupt varactor to use
and how much inductance in series with it.  Fundamentally, you need
enough tuning range to take into account crystal calibration error
(typically 10 ppm) plus any temperature drift plus any aging.  You
should be able to reliably get 50 to 100 ppm pulling range.  All other
things being equal, a 5th OT design has 1/25 the pulling range of a
fundamental.

Be sure to check for spurious UHF oscillations, that you may not see on
your scope or spec an.  The 2N5179 was famous for oscillating at 1 GHz.
You can fix this by putting a resistor in series with the collector.
The HP 10811 oscillator fixed this by having a 62 pF capacitor with
short leads from the collector to somewhere (see the schematic).

So that's all there is to it.  It always worked for me.

Rick Karlquist
N6RK

On 2023-12-04 18:42, glen english LIST via time-nuts wrote:

thanks Bob and Rick for your fine input, Rick thanks for the treatment ,
I will follow your recipe and report back. Yes these days I have
troubles with AVAILABLE SMT RF transistors, they all have Fts 4-10 GHz
and I generally resort to common base with ballast R. 2N5179 I used to
put a bead on the emitter lead as described in "ARRL Solid State Design
" remember that ?

If I ran out of pull after 10 years  while  I can tell the radio just to
go off freq a bit (digital fix in FPGA) , the issue then is the digital
sample rate fixed to the oscillator  go off meaning the world runs fast
or slow.

From the manufacturer, Krystaly ,  on my batch of 3OT  98.304 crystals
: I reproduce word for word here. These guys are very helpful.

"

All our crystals have undergo the process of preaging, also called
"accelerated aging" (i.e. 3 days storage at +105 deg.C). This aging is
equivalent to one year aging at normal temperature (according standard
IEC 60122-1). The crystals produced for you had frequency shift after
this preaging below 1.0 ppm, so it guaranties aging below 10 ppm after
10 years. The aging curve shows deceleration (saturation) in time.
Information to pulling parameter:
Trimsensitivity at frequency 98.304 MHz, 3.rd  overtone is approx. 35
ppm/pF.
Trimsensitivity at frequency 32.768 MHz, 1.st  overtone is approx. 400
ppm/pF so ten times higher."

regards

On 8/12/2023 4:19 am, Richard Karlquist wrote:

Hi

Their idea that 3 days at 105 is the same as 1 year is …. errrr …. ummm ….. questionable. Even more so in an oven application that will have the crystal running at something well above room temperature (our starting point for the thread was heating it to 60C ….).

Simple example:

A typical OCXO crystal can easily be 20C above the max temp spec on the device. If the OCXO is rated to go to 70C, that would be 90. In the case of 85C, you are up at 105C. If indeed 3 days is a year, those 85C OCXO’s would have an insane daily aging rate. In practice, that’s not what folks observe.

Bob

Hi Bob

Your comments are appreciated

mmmm. I had some, back in 2000, some 6 MHz, 98 deg C turning point
crystals used in a hot oven. after a few months, the aging went the
other direction.  I never got to the bottom of it - IE was  it was
oscillator components, or the crystal that was the instrumenbt in aging
the other direction.

On 8/12/2023 8:38 am, Bob kb8tq wrote:

Hi

Aging can easily be “direction changing” over time. What you do with early testing is try to bound the range of what will happen rather than come up with an exact number.

Bob

On 12/7/2023 3:38 PM, Bob kb8tq via time-nuts wrote:

Many moons ago I worked at a test lab that did nuclear plant Loss of Coolant Accident (LOCA) certification testing.  Prior to the LOCA tests, we had to send the test items out to be irradiated in a hot (radiation) cell, then cooked them in our lab temp chambers at X degrees for Y days that simulated Z years of nuke plant use at ambient temp T per an IEEE document whose number I forget.

Fun stuff.  ;)

Wish I could remember the document.  I'd trust IEEE to do the temp/time calculations correctly, so I wonder how the Krystaly numbers would compare?

thanks much,
ben

ya know, everytime I roll my own high stab / high perf XO, every few
years- I get reminded of why I should pay $$  to buy an off the shelf

----I realise why people charge (high prices) that they do .... In this
case, cant get what I need on the market, and my volumes are
insufficient to have the experienced do it.

On 8/12/2023 8:59 am, Bob kb8tq wrote:

Perhaps someone far wiser than I can chime in, but here’s my experience with crystal aging:

I ordered some 100.00000 MHz TO-5 crystals specified for frequency west type phase locked brick oscillators from international crystal. They were also ordered with forced aging.

I put one in the brick, and attached an FRS-C to my frequency counter. I then over a month or more, plotted the frequency versus time.

The resultant curve was an exact parabola with no jumps. I recall there being a slight diurnal variation as a result of the room temperature changing by a few degrees. I don’t have the graph handy, but I’d say the ballpark aging rate was around 20-25 parts per billion per month at the SC cut turnover temperature. The diurnal component was in the vicinity of 1-3 parts per billion. In fairness, if that sounds too good to be true, it’s possible that the linear power supply and brick circuitry could have partially cancelled or even enhanced the frequency change.

All that said, that was a sample size of two. We also have no way of knowing what exactly forced aging corresponds. It could be that there’s shelf stock that’s been sitting for a year. It could be that it’s run in a fixture at higher temperatures, or higher drive power or both. It could even be that rhe crystals were stored in an elevated temperature. It could be a mixture of all three or something else I’ve failed to account for. Bottom line, unless there’s a specific callout or specification you don’t really know where you might fall on what might happen to be a parabola.

What we do know is that if the drift were linear that if it drifted for 120 months at the first month drift rate, you’d likely have no chance of havaing enough adjustment range. What’s also not clear is what would happen if you took an old crystal and mated it to new components or even components run for a few months. My guess is that part of the aging is component aging other than the crystal.

It would be interesting to take some ocxo modules apart and replace the transistors, diodes and op-amps and see if devices past their pulling range could be brought back into range, and what the aging rate looks like.

In summary, based on my personal experience, I’d bet money on the aging being parabolic excluding grossly poor components other than the crystal. While you don’t know where on the parabola you are, you can estimate.

P.S. this was done on a bench that was stable enough to see interferometer fringes frozen for long periods of time unless an 18 wheeler drove by.

From: Bob kb8tq via time-nuts time-nuts@lists.febo.com
Sent: Thursday, December 7, 2023 3:38:38 PM
To: glenlist@cortexrf.com.au glenlist@cortexrf.com.au; Discussion of precise time and frequency measurement time-nuts@lists.febo.com
Cc: Bob kb8tq kb8tq@n1k.org
Subject: [time-nuts] Re: pulling some crystals

Hi

Their idea that 3 days at 105 is the same as 1 year is …. errrr …. ummm ….. questionable. Even more so in an oven application that will have the crystal running at something well above room temperature (our starting point for the thread was heating it to 60C ….).

Simple example:

A typical OCXO crystal can easily be 20C above the max temp spec on the device. If the OCXO is rated to go to 70C, that would be 90. In the case of 85C, you are up at 105C. If indeed 3 days is a year, those 85C OCXO’s would have an insane daily aging rate. In practice, that’s not what folks observe.

Bob

time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com

On 12/7/2023 12:07 PM, glen english LIST wrote:

This kind of discussion about "guaranteed aging" is completely at odds
with everything I observed over many years working for HP.  The top
HP experts on crystals never talked about crystals being so predictable.
These experts were involved in inventing the SC cut, etc.
They taught the rest of the industry how to make crystals.
I personally observed many crystals aging vs time and vs temperature
trying to "sort" out the good ones.  Crystals would be good for while,
then for no reason might drift in the opposite direction.  I am
especially skeptical of "3 days at 105 deg C is worth 10 years of
aging."  The E1938A oscillator had an oven set point of around
105 degrees C.  It did not accelerate the aging compared to an
80 degree set point and certainly didn't accelerate it following
a hockey stick curve.

Rick