Mechanical sensitivity of crystals...

I have a 40MHz PLL with a bandwidth less than 1Hz. The reference frequency
is 20MHz, which is generated with a crystal oscillator module.

When I tap the PC board with my finger - a light tap, not hard - the PLL
temporarily loses lock. I puzzled over this for a while, then bypassed the
crystal oscillator module with an external clock. When the external clock
is used, the board is insensitive to my finger tap. This is repeatable, and
a second board behaves identically. The oscillator solder joints appear to
be good, and the signal output is okay.

My theory is that there is a small change in the crystal oscillator
frequency when I tap the board. The PLL has a very low bandwidth, and can't
track the frequency change. It loses lock for a short time, until the
crystal oscillator frequency recovers (or at least stops moving) and the
PLL reacquires lock.

The problem is, I've never heard of crystals having mechanical
sensitivities like this.

Has anyone else seen anything like this? Is this effect documented
anywhere?

-- Mike --

 

Crystals are piezo devices and will be affected by mechanical shock.

Leon

 

What do you think crystals are?.
They mechanically distort using the piezo effect, and the vibration, is
the basis of their behaviour. The crystal is effectively 'ringing', with
the AC waveform matching their natural mechanical frequency. As such
anything disturbing their movement, _will_ produce effects. This is the
basis of many of the accelerometer devices on the market.
In your case, mechanically 'shock mount' the crystal. Use flexible wires
instead of the existing legs, and cushion the crystal. There are
commerically available spring mounts, which hang the crystal on a rubber
membrane to prevent exactly this type of problem. If you look at crystals
supplied for high shock enviroments, they typically have larger cases,
with the crystal element itself shock mounted internally. This mechanical
sensitivity, is the very basis of how they work.

Best Wishes

 

Yes, the crystals get noisy, when you tap them. It's
known as microphony. You can buy crystal microphones. ;^)

In most applications that isn't noticeable, but when
trying to do GPS receivers with cheap crystals, it was
a common occurence for the tracking loops to lose lock
on such occasions.

1Hz bandwidth at 20MHz sounds very narrow to me. Most
crystals are specified to be 10-50ppm accurate over
temperature and age. The cheap computer-grade oscillator
modules are known to go beyond 100ppm.

You seem to rely on a 0.05ppm reference from a crystal?

Kind regards,

Iwo

 

Although the crystal its self is a piezo device, the frequency of a
crystal oscillator is much more likely to be affected by the mechanical
sensitivity of the rest of the other components in the oscillator circuit.

Connect just the crystal to the board using an inch or two of very thin
wire bent into a zig-zag pattern and support the crystal separately from the
board so that you can isolate the mechanical shock to the board only. I suspect
the sensitivity to shock will appear only for the board.

Jim

 

He's using an oscillator module, not a crystal.

Leon

 

The actual system is somewhat more complex - I omitted details that weren't
relevant. It's a data recovery system. The crystal oscillator drives a
synthesizer that provides a frequency within a few ppm of the data
frequency. A data recovery PLL uses a phase interpolator to generate the
data frequency from the synthesizer output. A third PLL with a very low
bandwidth follows the data PLL, and performs jitter attenuation. The
absolute accuracy of the crystal doesn't have to be perfect, but sudden
frequency changes, even by only a few Hz, can be enough to throw the final
PLL out of lock.

-- Mike --

 

They certainly do, they are mechanical devices after all.

Crystal mounting can make the problem worse or better. Usually
just supporting the crystal by its two leads and letting it flop around
is the worst. Soldering the crystal can to the board is better.
Best is to use a three-lead (TO-39-type) can, although these
are usually only used in high-shock mil-spec situations. Even in the
"best" case there are still microphonics.

Tim.

 

I had a similar problem on a 155.52 MHz (OC-3 rate) PLL, used in a
laser timing module. I used a nice low-phase-noise Vectron VCXO in an
adaptive loop, around 3KHz in acquire mode, 300 Hz for track. Whenever
I pulled a test SMB connector out of the board (snap!) it would lose
lock. After much tapping, it was isolated to the VCXO and nothing
else. If you just touch the VCXO with a small screwdriver, rather
gently, it can lose lock. We wound up having a wire-form company make
us a bunch of tiny custom springs to shock mount the osc from the PCB,
one spring on each pin. Good thing we didn't use a surface-mount part!

I've seen G-sensitivity specs on very good (SC-cut) oscillators, but
such a spec is rare.

John

 

I haven't opened the oscillator module, but I suspect it's a crystal, a
couple capacitors, and an oscillator/driver IC mounted on a substrate. What
else besides the crystal would be shock sensitive?

-- Mike --

 

No the best is to take the package without the 3rd wire and shock mount
it.

Good crystals have a shift of about 5PPB/g at low frequencies. At higher
frequencies, the shift can become larger as you get near the mechanical
resonance of the packaging. Damping out the higher frequencies is really
worth it if you are trying to get a stable frequency.

 

The first bet is the capacitors. If they are surface mount devices (very
likely) flexing the substrate can change their value. A very stiff
substrate and housing helds to prevent this.

 

Yes, I know that, but we're talking about ~0.1ppm frequency change caused
by a relatively small mechanical movement. I'm not banging on the crystal
itself, I'm moving the entire container up and down when I tap the PC
board. I'm just using my finger, which is soft, so the impulse is mostly
low frequency.

I found a review paper on acceleration sensitivity of crystal oscillators
on the IEEE UFFC web site. It looks like that has the information I'm
looking for.

-- Mike --

 

have u tried a diferent source of XCO ? ive taken one or two apart before
now and they differ quite a bit some have quite elaberate springs contacting
the crystal some seem less isolated. however if the crystal would move about
to much inside the case the capacitance wld be altered slightly wich wld
alter the freq slightly. i would gues the surface mount modules have no room
for much isolation.

ive noticed ordinary crystals if you pres on the case it changes the
frequency significantly.

i asume the PLL is doing more than doubling the 20mhz frequency or you could
just use a higher freq module or a doubler ?

Colin =^.^=

 

yes i have.
the crystal is simply mounted on a couple of thin wire legs.
vibration of the stone can cause pressure on the stone from the
mounted legs. this can cause a resonant shift.
being close to the case when shook can also cause problems.
they make different crystal style cases you may want to look into that/
in any case i really don't think you are going to leek it with in
1 Mhz

 

Any bad (solder) connection in your circuitry or in a component is a
possible suspect.

Also, at your frequencies a change of even 1 pF of stray capacitance could
affect a tuned circuit. What mechanical oscillations are you creating by
tapping on the board?

Norm

 

In some crystals the actual piece of quartz is more-or-less suspended by the
leads within a frame. Even simple mechanical motion of the container can
couple into the crystal suspension. If you were really curious, you could
set up a second unit and beat the to frequencies together to see how much
you're changing the actual frequency by. It'd be instructive to see what
demodulated frequencies are there.

Norm

 

Or listen to a harmonic on an FM radio. I bet it sounds like classic
microphonics.

John

 

Thanks for the replies, everyone.

Yesterday's searching turned up several papers on vibration-induced phase
noise. The Army has done quite a bit of work to understand these effects,
and some crystal manufacturers (Vectron in particular) have oscillators
specifically designed to suppress vibration-induced noise.

In general, SC cut crystals with no shock-mounting are reported to have
sensitivities in the range of 1ppb/g. Other cuts are typically somewhat
worse. Vectron reports that with proper mounting, they can achieve 0.1 -
0.01 ppb/g sensitivity.

We did a simple 2g tip over test, measuring the frequency of the oscillator
with the board right-side up and upside down. The result was a change in
frequency of 2Hz. That's 50ppb/g (the change in force on the crystal is 2g)
- roughly an order of magnitude larger than I was expecting based on the
papers.

By far the best reference I found was an article in the Miami Sun Post
about "self-proclaimed de facto scientist" Carlos Dolz. Mr. Dolz placed
three quartz watches in a centrifuge and spun them for six hours. The
purpose was "to prove that a time shift into the future, or one into past,
was possible ¡V in essence, 'a time machine.'" Apparently, this
demonstration made it into a museum exhibit entitled "Playing with Time,"
but the day it opened, after a single demonstration, Mr. Dolz was asked to
pack it up and remove it. (This was apparently prompted by a Miami Herald
article describing Dolz and the apparatus a few days before). During the
demonstration, after six hours in the centrifuge, one watch gained 2.5
seconds, one watch lost 2.5 seconds, and one watch stopped. Apparently,
Dolz had been developing his 'time shift' theory in the museum lab since
September 2003 and has conducted it six times successfully. Success is
apparently when the watches in the centrifuge run _faster_ than a control
watch. Dolz expected the watches to advance 4 seconds into the future over
the course of a six hour test.

This is it! We're seeing time-travel effects! Our oscillator is momentarily
moving into the the future when I tap the board. We aren't yet sure whether
the PLL's loss of lock occurs during the exit from the current space-time
continuum or during reentry, but experiments are continuing.

-- Mike --

Some references:

Wenzel html note on vibration induced phase noise:
http://www.wenzel.com/documents/vibration.html

The IEEE UFFC web site has lots of interesting info, including this paper
on acceleration effects in quartz frequency standards:
http://www.ieee-uffc.org/freqcontrol/quartz/vig/vigaccel.htm

Vectron appnotes page:
http://www.vectron.com/products/appnotes/
This paper describes a Vectron shock-mounting system
http://www.vectron.com/products/appnotes/Lowg_QRM.pdf

More on self-proclaimed de facto scientist Carlos Dolz:
http://www.miamisunpost.com/firststoryfrontpage.htm
http://www.miami.com/mld/miamiheral...dade/cities_neighborhoods/west/8234518.htm?1c

-- Mike --

 

This reminds me of a conversation I had with an engineer at Linear
Tech (IIRC). He said that if you are using a high accuracy voltage
reference IC, and you calibrate the board based on this reference (or
calibrate the reference), then the calibration should be performed
after the board is mounted in the enclosure. He said that any
mechanical flex of the PCB during mounting could cause a corresponding
flex of the voltage reference IC package, resulting in stress on the
die and a slight change of the output voltage. So, if you have to
calibrate the board prior to final assembly, then the PCB should have
a 'U' routed out around the voltage reference reduce flex of the PCB
under the reference. I don't know if the same principal will apply in
your case, but it may be worth flexing the PCB to see if there is any
effect (in addition to checking shock and vibration).


================================

Greg Neff
VP Engineering
*Microsym* Computers Inc.