[time-nuts] Some thoughts about Crystal Oscillators

Gerhard Hoffmann dk4xp at arcor.de
Sun Apr 15 13:01:01 EDT 2018

Am 17.03.2018 um 00:57 schrieb Ulrich Rohde via time-nuts:
> https://synergymwave.com/articles/2013/04/full_article.pdf
> This may be still of interest... happy weekend, 73  de Ulrich, N1UL

Since the weather did not look so nice this weekend, I canceled a
2-day motorcycle tour and tried the much applauded Colpitts

I think it is not possible to make it closer to the published circuit.
OK, I dismissed the 0.5fF C0 modifier which is probably only a
placeholder for simulation experiments. I had no 200nH parts, so I took
220nH, 150 nH did not make a difference other than less output voltage.
I did not test the buffer stage because the BFG540 is obsolete
and I could not find anything that fits its weird pinout.

 From the Altium Designer to .pdf, to Laser printout on foil as a
photo mask and the contact copy to the unpopulated pcb can be
done in a good hour, so it's a nice weekender.

Circuit:       < 
Layout:      < 
Hardware: < 

It turned out that the circuit, as it is,  leaves the choice of the 
overtone to the crystal.
 From a collection of 100...120MHz crystals not one worked at the 
intended overtone.
There were results from 20 to 70 MHz.
The test point is where the BFG540 emitter would be.


I forwent measuring the phase noise.

Call me amateurish again, but I'll stay with the Driscoll. It has the 
tank circuit to enforce
the correct overtone and I can adjust the delay so that 0 degrees around 
the loop
coincides with maximum dp/df. Maximum Q helps nothing when it does not 
happen on the
oscillation frequency.

When I'm asked in a design review how I have made sure that the 
frequency of a single point
of failure VCXO is not off by 30 to 80 percent, then I like to have an 

And WRT vastly changing transistor parameters: They don't need to. If 
one looks into the emitter
of a BJT at 10 or 20 mA bias, one sees 1 or 2 Ohms.  That depends only 
on Ic, K, T, q and some
minor parasitics such as RE.
The emitter is fed by the crystal in Driscoll's case with 50 to 70 Ohms 
on resonance, which takes
nice care of the transistor's shot noise. Off frequency it gets even 
better, esp. with C0 compensation.

The base is fed with a huge step down ratio, so the base voltage is 
stiff. The cascode isolates from the
effects of the tank circuit and it can easily be bootstrapped. If one 
dislikes Schottky limiting, one can
duplicate the cascode transistor and divert some of the RF current to 
nirvana and not to the transformer,
whenever the AGC voltage wants that.
So the environment of the crystal and the sustaining amplifier can be 
very, very time-invariant.
The whole thing is just a current conveyor belt with the crystal 
enforcing the current trough the
whole cascode to the tank/load. The output IS the resonator current.

That is the very same structure as the proposals to extract to output 
"directly from the resonator,
where it has the best SNR".
Given the little bit of power that we can steal from the resonator: must 
we really split it between
extraction amplifier and sustaining amplifier right at the location 
where the signal is weakest?

When we can build an extraction amplifier that adds less noise than the 
sustaining amplifier, why don't
we take the feedback also from there and cut out the sust.amp. and the 
signal power it requires?
The Driscoll does that.

Gerhard, DK4XP

(the loyal heretic :-)

More information about the time-nuts mailing list