A bit of help with a crystal oscillator

Peter · 2019-11-06T12:18:37+00:00

I have read a lot of stuff on this topic but find it hard to apply it. I have a 25MHz xtal. For example Kyocera CX3225GB25000P0HPQZ1. The spec says typ drive level 10uW, 100uW max. The oscillator circuit is the usual Pierce one as in e.g. here The CPU is a STM32F407VGT6. The development kit for this CPU has the two caps for the xtal load cap and the resistor is 220R. The xtal they use is an anonymous one, however; old style metal can (HC49?). The ST appnotes point one to a load of xtals which either don't exist or are way too expensive for production :) The internal feedback resistor is given as 200k which is fairly normal. Another appnote suggests measuring the xtal current, with a current probe. I could construct a current probe with a little ferrite core (I used to make these up for switching power supplies; a lot cheaper than the Tek current probes) but these components are so tiny (0603-0805) that the required loop would really mess things up. The circuit runs fine, but I need to have some confidence in the margins over temperature and component tolerances. The load caps are 7pF, which should be about right for the 10pF load cap of the xtal I am using, allowing for a bit of PCB capacitance (say 5pF). The above Kyocera P/N says 8pF. OTOH I have read one appnote which says the CL calculation is different: the cap value is to be the sum of the CPU pin stray cap and the xtal load cap; this means the CL should be 15pF! Probing the two ends of the xtal with a Wavesurfer 3034 with a ZS1000 probe (9pF/1Mohm) I see about 0.5V p-p one end and 0.3V p-p on the other end. Does this seem reasonable? Many thanks for any comments.

Nov 06, 2019 67 Replies

Rate this thread:

Steve Wilson 6 years ago

As I have shown in other posts, a Schmitt has no stable state and must oscillate. The trick is to channel the energy through the crystal instead of around it, as in a plain RC oscillator.

Time domain oscillators start on numerical instability. There are numerous examples in Oscillators.zip where the oscillation starts at zero amplitude and builds up to strength.

formatting link

It is extremely difficult to couple a pertubation into the oscillator tank to kick start an oscillator. The tank simply ignores the energy in an external pulse. You have to somehow inject the pulse in series with the tank but I have not found a practical way to do that. A simpler method may be to simply increase the loop gain by increasing the drive level.

Vote

jlarkin 6 years ago

A current source pulse usually works. Or series, a voltage pulse. Either disappears after the pulse.

I've simulated oscillators where every node parked at exactly 0 volts forever. More loop gain wouldn't help. A tiny injected pulse, or some initial condition setting, got them going.

You can make pretty good time-domain noise generators too.

XOs are hard to sim in time domain because the Q is so high. Sims can run for days. I like to force the initial conditions near the expected operating values, and zoom up to see where it wants to go.

John Larkin Highland Technology, Inc lunatic fringe electronics

Vote

Steve Wilson 6 years ago

I have tried both. I found them ineffective.

I'd like to see some of your syms.

I show how to start and settle high Q XO's in several cycles. I call the technique "Fast Start". I show how to use it Oscillators.zip

Vote

jlarkin 6 years ago

Use a bigger one! Spice is cool because you can make a thousand-amp pulse.

I'll see if I can find one that parks at zero.

John Larkin Highland Technology, Inc lunatic fringe electronics

Vote

Steve Wilson 6 years ago

That would be great. Thanks.

Vote

LM 6 years ago

Vote

Peter 6 years ago

We got the 25MHz oscillator running fine, and easy enough to probe with the 0.9pF probe, but the 32768Hz one is very delicate.

I realised why any probing kills it: the probe is 1Mohm and the feedback resistor (inside the CPU) is 18Mohms. Clearly one needs to do any probing via a capacitor, of say 10pF.

LM wrote

Vote

John Larkin 6 years ago

But you can pay $1000 for a good one.

And you can get into trouble in production, if 20% of them don't oscillate.

John Larkin Highland Technology, Inc picosecond timing precision measurement jlarkin att highlandtechnology dott com http://www.highlandtechnology.com

Vote

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required