DIYHiFi.org

CG, very good paper! Thanks!

For those who don't have the time to read it.... the quartz blanks are fabricated at a precise orientation with respect to the crystallographic axes of the quartz material. The orientation or "cut" determines frequency-temperature characteristics and other important properties of the resonator.

The AT-cut is sensitive to stresses in the body of the resonator, whether caused by temperature gradients due to rapid external temperature changes, or by external forces. For applications where extreme stability is required this stress sensitivity is a drawback, SC (Stress Compensated)-cut minimizes these effects. It belongs to the family of double rotated crystals (quartz crystals cut on an angle relative to two crystallographic axes), in reality it represents the optimum double rotated design as its particular angle ~ 35°15 ø ~ 21°54 provides maximum stress compensation.
SC-cut resonators are better at handling the mechanical stresses imposed by mismatched thermal coefficients of expansion between the quartz itself, electrode plating metal and mounting holder, resulting in lower thermal hysteresis.

The current handling of the resonator (max current at which it can be operated without a significant change in frequency) is considerably higher for SC-cut than for AT-cut. The consequences for oscillators are an improved phase noise floor by operating a higher current and reduced sensitivity to drive level change.

The inflection temperature is approximately 25°C for the AT-cut and 92°C for the SC-cut. So for a SC-cut a oven is mandatary.

sorry Jocko, I've to ask it another time... where can I buy a couple of 11.2896 MHz SC-cut crystals for my DAC?

http://www.icmfg.com/specialcut_crystals.html
http://www.icmfg.com/lowphasenoise_crystals.html
..anyone?

_________________
Ciao
Roberto

http://www.axtal.com/
go to product / features / ocxo -> they seem to make SC cut crystals but I could not find them
offered as separate product. This is in line with results I got some years ago from several companies:
OCXOs yes, bare SC crystals: no.
Probably they fear the level of support required. Don't underestimate the effort
to build a SC OCXO. Replacing the crystal in the usual Colpitts simply won't work.
First, SCs have a unwanted resonance 10% higher than the wanted one that is
usually more active and must be suppressed. Then, even the 5 MHz units are
5th overtone. The MCI 260 10 MHz units seem to run the xtal at 5 MHz and doubling this.


If you don't have a Wavecrest / Agilent / Poseidon / Aeroflex phase noise analyzer
you have no idea of your circuit's performance. Ok, it can be done with ring mixers,
multiple good reference oscillators and an FFT analyzer that knows cross correlation
(three cornered hat method) but this is a large project by itself and I do just this as
a low priority background task. This is what I want to use the 5 MHz MTIs for.

If you are only interested in audio dacs, you would be done with it by the time you
have the reference oscs :-)

BTW. I once claimed that the top notch oscs would all be SC cuts nowadays
in a usenet group and got a fierce reply from Rick Karlquist.
I remember his words "profoundly wrong".

I don't think much about gurus who claim that 100 dB channel separation is a MUST
for a preamp when vinyl has 30 (you probably know what I mean)
but I would not dare to disagree with Karlquist.

Completely on-topic: http://www.karlquist.com/
and there the articles on zero-gradient ovens and bridge oscillators.

This straight-line style of problem solving has certain aesthetics and
is the stuff that makes one want to become an engineer.


Suggestions regarding your SC supply problem:

I don't know your circuit, but if it's a Colpitts, get rid of it because
in a Colpitts, the maximum working Q is not on the oscillation frequency by design.
Q is (delta phase / delta frequency) and this must be maximum
where the phase of the opened loop goes through 0. This is the exact oscillation frequency.

Try a Driscoll or Butler circuit. This alone might double the effective Q and
costs nothing but time. But then this is DIY.
Have a 5th overtone 3.xxx or 5.xxxx MHz AT made and put it to use in the Driscoll or Butler.
Double or triple the frequency to the final value required.
(3 to 8 MHz seems to be the sweet spot.)

If you max that out, it is probably the best osc in the chain by a wide margin.
After all, the Digitizer had probably a DIL can osc or a crystal in the corner of
a digital LSI chip...


If you can read German: http://www.qsl.net/dk1ag/buch.html
is a .pdf edition of Bernd Neubigs sold-out "Crystal Cookbook"

see also:
http://www.qsl.net/dk1ag/osc_lit.html

BTW,
I've made some SMD variations of the Jung regulator / reference family.
Layout is in the computer, but not yet tested in real hardware.

regards, Gerhard

_________________
Everything has been said already - but not yet by everyone. (Karl Valentin)

Stupid question;
The stuff posted lately pertains to top of the line crystals/osc. Nice stuff. Where does, how does the Tent compare or stack up with these?

Many years ago there was an article in AA about a feedforward error cancellation method. I used this basic circuit for a preamp and still use it. The Wenzell reminds me of this. When I have the time (next lifetime, maybe) I want to experiment with this for a low noise low 1/f supply.

I know that measuring clock performance can be pain in the... bottom. But it can't be that bad in case of PSU. Very low noise amp (using AD797 you can have single nV @10 Hz I belive). Then PC audio card and FFT software. Long time of measurement and I think you can try see what is going on .

Lenny

There isn't much necessary to adjust that thing. Add 50 mV of noise / music / sine to the
input voltage with an old transformer. Connect a scope or AC voltmeter to the output
and adjust for minimum noise/music/sine. Done. A soundcard would do, too.

Don't use more input AC than you can drop over the series resistor.

regards, Gerhard

_________________
Everything has been said already - but not yet by everyone. (Karl Valentin)

Most of the test data that I have seen for those circuits are for below 10 Hz. It takes a lot longer to do those than it does for stuff >100 Hz.

When you start getting to around 100 Hz, you have to worry about 1/f from the oscillator transistor. I would advise build a "finesse" that took out the <10 Hz crud from the regulator.

As for Guido's stuff.............I would say single-digit pSec numbers. When used with his supply. Good enough for our purposes. Some of the other clocks (designed by engineers) measure about the same. The one in particular that comes to mind has a on-board supply, as well.

Jocko

Hmmm....where have I seen fig. 15 in the bridge oscillator paper? Not sure how it was patented, about 8 years after it appeared in a trade journal. Unless it was the same guy.

Yeah, Butlers are good. Matheys suggests that they work best at higher frequencies than we need. His suggestion is a Pierce for our range of needs.

I know of one famous high-end designer guy that uses Pierce clocks. Probably JFETs (would not be my choice), but certainly not a CMOS gate.

Jocko

So I pull out the calculator and start looking at some numbers. Filtering <10Hz is a little more complicated than it looks. Figuring that some changes below say 2-1.5 Hz could be/are thermal drift, to RC filter the values get ridiculous, etc, etc.

More brain power needed than I have at the end of this week. Maybe later but an interesting issue.

Hi Gerard,
thanks for the reply and the interesting links.

I'm designin a new DAC (PMD100+AD1862'S), so I can use every kind of oscillator I want.
In reality I need to design two: one top-class XO for the I2S (slaved-CDT) LVDS connection and one (less stringent, but high quality) VCXO for the 2nd PLL of the SPDIF.
I currently use Tent XOs, but I'd like to go a step forward!

For the XO was considering a bridge or Driscoll design too, something easy pullable for the VCXO ...but the problem is always to find low-quantity suppliers of good (high Q) parts, even AT-cut.

Jocko,
opamp....what about the following plot?

..non male!

Did you read Matthys?

...gotta run..... I've to connect a low-noise 1kV Vref to Pin 13 of my '1862!!!!!!!!!!

_________________
Ciao
Roberto

It is LMP7731:
http://www.national.com/pf/LM/LMP7731.html

Using good (MAT or SSM2210) parallel transistors you can be even lower.
Now since we got such nice parts and Wenzel circuit to play with (thank you Jocko for hint) - we need another thing - good quiet Vref to start with. Any better ideas than LEDs ?

Apart from that - I've found one more pdf worth reading:
http://www.sss-mag.com/pdf/lonsosc.pdf

Lenny

That part is only good to 5 V. Besides, I am laying out the PCB for DIP parts, only. No more of this SMY crapola. I am sick of it.

Other than that, yeah, it would be OK.

Jocko

No, never been that lucky. I am sure that one of our resident famous audio designers has. I have only read various articles of his. There is a section in one of Jim Williams' book that he wrote.

Jocko

Jocko, if you have time....why don't you put that DVD you have on your desk into your PC drive? You could find something interesting!


Do I understand right? Does your new circuit include a Wenzel-style LF filter to reduce the base noise of a (averagely good) EF?


Gerard


This is the stuff that makes one want to design a bridge oscillator like this!
....or maybe I could buy a tube/comparator NO-jitter one!!

_________________
Ciao
Roberto

Still haven't found the time.

Nope, not an EF. Not the best choice for low 1/f.

Jocko