So Herbert, what type of oscillator do you advice for the SC-cut crystal we just have ordered?
So Herbert, what type of oscillator do you advice for the SC-cut crystal we just have ordered?
Since we we are looking for an oscillator to use in audio digital devices, I believe the Driscoll type was the best, thanks to its very high short term stability.
From Crystal Oscillator Circuits (Krieger-Matthys):
"The Emitter coupled harmonic circuit has the best short term stability of any harmonic circuit tested by the author to date. Originally described by Driscoll..... This stability is better than that of any other harmonic circuit tested."
I'm curious to know Gerhard's opinion.
The best one I have built up to now is a Butler, but I haven't built many Driscolls.
I think the Drisc. makes it slightly easier because the xtal is on GND on one side,
so one can easily measure the voltage over the crystal. Also, in the Butler, the
xtal looks into 2 emitters, that takes more current to make them both really low
impedance. In the Butler, it is somewhat easier to measure the loop gain with
a network analyzer because you can cut the loop open at the emitters and you
have then 2 convenient low-impedance ports. In the Driscoll, you probably have to
cut and measure into the base, which is far off your 50 Ohm analyzer port.
This can be helped with Randall/Hock correction(1).
Close-in the limit is the crystal, as long as you operate it in series resonance
with low additional resistance. Also, the oscillation frequency is where the
loop gain goes through 0° and you must make sure that d phase / d frequency
is steepest at exactly this point. dp/df is another word for Q, and is does not
help if the best Q happens 2 KHz above the oscillation frequency. One can
adjust that with the tank circuit to some degree (nice pun!)
It is possible to simulate the open loop behaviour with LTspice, although
2 Hz steps at 100 MHz seem to stretch the numerical frequency resolution.
I have described that loop gain business in
< hf downloads >
We have observed that only 10% of a batch of crystals give top notch
phasenoise, others are 10, even 20 dB worse while all have a Q of 100000
at 100 MHz. That was from a reputable manufacturer, and we ordered
extra lapping and cleaning... A much bigger spread than between properly
designed oscillator types.
If you take a look at the MTI-260 data sheet, there are versions with
AT and SC cut mixed wildly. It is possible to make AT for 40 degC or so.
(1) Randall W. Rhea: Discrete oscillator design. Artech House. Good book.
Where did you take the output from the Butler oscillator, from the collector or from the emitter of the second BJT?
I have been busy with oscillators for more then 20 years and for the digital audio (ADC and DAC) the best oscillator I have found was the Clapp with AT-cut! With special Xtals from QT indeed.
Look at: https://www.by-rutgers.nl/rutgerS-Clock.html
on my website you find the arguments. Manny of them in Dutch but the bold articles are in English...
All oscillators with emitters in the active circuit degrade the close in noise!!!
Look at: https://www.by-rutgers.nl/rutgerS-Clock.html
on my website you find the arguments. Manny of them in Dutch but the bold articles are in English...
All oscillators with emitters in the active circuit degrade the close in noise!!!
Herbert-
I just looked at your circuit. Its quite interesting. Using a Shottky diode to control the max output seems to help a lot. You changed the high speed comparitor for a set of fast TTL gates and that lowered the phase noise. Two questions- first did you try faster logic families like 74AC or the Potato logic parts? Second is there any need for hysteresis on the biased input of the gates (any tendency to oscillate)? Also you seem to be running the logic on 6.8V, that is higher than spec limits. Is it to compensate for losses in the series inductance?
I think its a nice and simple circuit. Would adjusting the oscillator for 5V compromise it a lot? In your other write-up you suggest Varicap diodes are to be avoided. I would need to phase lock an oscillator to an external clock. Is there a better option? Or a recommended way to add the varicap diodes?
I just looked at your circuit. Its quite interesting. Using a Shottky diode to control the max output seems to help a lot. You changed the high speed comparitor for a set of fast TTL gates and that lowered the phase noise. Two questions- first did you try faster logic families like 74AC or the Potato logic parts? Second is there any need for hysteresis on the biased input of the gates (any tendency to oscillate)? Also you seem to be running the logic on 6.8V, that is higher than spec limits. Is it to compensate for losses in the series inductance?
I think its a nice and simple circuit. Would adjusting the oscillator for 5V compromise it a lot? In your other write-up you suggest Varicap diodes are to be avoided. I would need to phase lock an oscillator to an external clock. Is there a better option? Or a recommended way to add the varicap diodes?
Hi Demain,
first read the document:
https://www.by-rutgers.nl/PDFiles/Reproducible Low Noise Oscillators.pdf
for a good understanding of the role of the schottky diode. It also lowers the input capacitance of the FET! The FET operates in class C. If you could read Dutch there is another document on my site that explanes carefully the problem of AM to FM conversion. I will translate it into English when I get the time....
About the so called slicer:
I use a 74HC04. The operating voltage very often could be over 8 volt! You have a lot of questions about the 74HC04... No, I did not try already faster or any other circuit.
Sometimes the circuit oscillates in a lower frequency (when there is no Xtal in the oscillator) and I think this is one of the advantiges: a nearly oscillating circuit triggered by the noise free oscillator.....
The only demand is that the power supply of the 74HC04 should be as noise free as possible (look at: Shunt Regulator for the right circuit).
The power supply of the oscillator MUST be at least 15 volt!!! Read the document mentioned above.
I do not use VCXO's. The best way is an asynchronous DAC, This means an extra memory to buffer the signal before the DAC. You get some latency (0.5 sec). That does not border me.
Mind that the three stage invertor should not be too fast so that there arises a better isolation between the oscillator and its outside world. It is not only a squarer but also a buffer!
OK?
Herbert.
first read the document:
https://www.by-rutgers.nl/PDFiles/Reproducible Low Noise Oscillators.pdf
for a good understanding of the role of the schottky diode. It also lowers the input capacitance of the FET! The FET operates in class C. If you could read Dutch there is another document on my site that explanes carefully the problem of AM to FM conversion. I will translate it into English when I get the time....
About the so called slicer:
I use a 74HC04. The operating voltage very often could be over 8 volt! You have a lot of questions about the 74HC04... No, I did not try already faster or any other circuit.
Sometimes the circuit oscillates in a lower frequency (when there is no Xtal in the oscillator) and I think this is one of the advantiges: a nearly oscillating circuit triggered by the noise free oscillator.....
The only demand is that the power supply of the 74HC04 should be as noise free as possible (look at: Shunt Regulator for the right circuit).
The power supply of the oscillator MUST be at least 15 volt!!! Read the document mentioned above.
I do not use VCXO's. The best way is an asynchronous DAC, This means an extra memory to buffer the signal before the DAC. You get some latency (0.5 sec). That does not border me.
Mind that the three stage invertor should not be too fast so that there arises a better isolation between the oscillator and its outside world. It is not only a squarer but also a buffer!
OK?
Herbert.
Last edited:
I took it from the collector of the second transistor. Some argue that the S/R ratio
is better at the left emitter, because the signal is there freshly filtered by the xtal,
but then you have to amplify it first. That means that half of the xtal output goes
to the output amplifier and the other half to the sustaining amplifier as it used to be.
But it will pick up noise also in the output amplifier, and the starting conditions re
noise floor are 3 dB worse now. Does not look like a big win to me.
The emitter of the 2nd stage should work also, but somehow it is uncomfortably
close to the resonator.
I need the VCO for some test instrumentation functions. Its not a normal audio application.
Your drawing suggested 12V which was the starting point. 15V can be available (although I would prefer 30V and not connect to ground for the oscillator at all and XFMR couple its output). I also understand the buffer aspects. Most of the lit I have read suggests that the fastest logic will have the lowest phase noise which sort of makes sense. In any case the phase noise of the buffer will be lots less than an oscillator.
I'm surprised that a fet in class C would have lower phase noise. The lit all suggest the opposite, again the first guess is that the time to recover from cutoff or saturation is device limited and in jitter terms pretty long. However it seems to work in your parts.
I have some very low noise power supply options which lend themselves to this type of application.
Looks like I have some studying to do.
This drawing is very old. With time ones knowledge grows...
30 volt should be good if the FET could handle it.
The phase noise of the buffer is no close in noise so not so important to audio applications. The 74HC04 has the right latency to function as a good buffer!!
There is a lot of child talk about oscillators in the literature and on Internet, but believe me, after 20 years of research I know that a JFET (J310 in our case, which by the way is a VHF-FET!) gives less close in noise than whatever BJT or other active component.
I'm interrested in your very low noise power supplies!
Greetings,
Herbert.
Well, attaching your name to that Schottky trick might rise the one or other eyebrow.
That was schoolbook wisdom in 1988 (Ulrich Rohde / T.N.Bucher: Communications receiver Design), cited there already, and well known in the 70s.
< Quarz-Literatur / Crystal Literature >
Bernd Neubig is head of AXTAL < Axtal >
and also author of the Crystal Cookbook, in German, but there are pics also:
< Axtal >
I also don't share a lot of your views, but then I'm not a missionary.
regards, Gerhard
Gerhard,
Did you think that I did not read nearly all literature during the last 20 years that I was designing oscillators? If you say that Ulrich Rohde already used a Schottky in the same way as I did, then you do'nt understand the 'extra AVC' I am talking about. Rohde uses a clamp construction which workd well with VC(X)O's.
For the lowest close in noise (I mean within 10 Hz from the carrier) my solution is unsurpassed. I do not talk about a noise flour of
-160dBc/Hz@10kHz. My oscillator will not reach that but the noise within a few Hz from the carrier is the cause of poor ADC and DAC operation. That's what I'm talking about!!!
Regards,
Herbert.
The start of this thread was to explore the importance of close-in phase noise. Unfortunately we don't even have a clear definition of what that is. If you are talking about below 10 Hz adev may be a better way to identify and quantify it.
My understanding from the literature was that the amplifier was a major contributor to the close in phase noise as is the crystal Q. And it is the LF noise of the transistor/fet that's important since it modulates the oscillator.
The separate issue is identifying how the close in noise affects audio. In the existing lit and studies lower frequency noise has progressively less influence on either SNR (jitter induced noise) or audible perturbations. Indeed the best mechanical drive (a heavy vinyl disk player) still has orders of magnitude more wow than any digital system and is not audible. Understanding the mechanism may shed some light on other problems in the chain.
As I said at the start of this I'm open about the project and interested in how well these work. That said improving on the performance of a Crystek CH95X will be pretty challenging.
Here is a link to the power supply circuit: http://www.diyaudio.com/forums/digi...b-interface-audio-widget-102.html#post2882143 Its pretty simple but gets 1 nV/rtHz or better quite easily and is well suited to crystal oscillators and similar noise sensitive stages.
My understanding from the literature was that the amplifier was a major contributor to the close in phase noise as is the crystal Q. And it is the LF noise of the transistor/fet that's important since it modulates the oscillator.
The separate issue is identifying how the close in noise affects audio. In the existing lit and studies lower frequency noise has progressively less influence on either SNR (jitter induced noise) or audible perturbations. Indeed the best mechanical drive (a heavy vinyl disk player) still has orders of magnitude more wow than any digital system and is not audible. Understanding the mechanism may shed some light on other problems in the chain.
As I said at the start of this I'm open about the project and interested in how well these work. That said improving on the performance of a Crystek CH95X will be pretty challenging.
Here is a link to the power supply circuit: http://www.diyaudio.com/forums/digi...b-interface-audio-widget-102.html#post2882143 Its pretty simple but gets 1 nV/rtHz or better quite easily and is well suited to crystal oscillators and similar noise sensitive stages.
I think you are right. An extract from the literature:
From the model (Leeson’s equation), it can be seen that the phase noise of an oscillator can be reduced by the following:
1.Lowering the noise floor by choosing an active device with lower noise factor (noise figure) and driving the signal levels at the resonator as high as possible. Post amplifiers (buffer amplifiers) should be low noise also.
2.Lowering the f0/2QL corner frequency by increasing the Q of the resonator and minimize coupling (loading) of the resonator.
3.Lowering the active device 1/f3 (flicker FM, 30 dB/decade) corner frequency by choice of the active device (low noise figure and low flicker noise) and optimizing how it is biased.
The major contributor to the close in phase noise are the flicker noise of the active device and the Q of the Crystal.
SC-Cut crystal has tipically higher Q than AT.
Bjt has tipically lower flicker noise than jfet.
Attachments
Hi Demian,
I looked at your power supply, and was surprised to see the TL-431 as the regulator. How could it reach 1nV/ rtHz? From my experience, TL-431 could be quite noisy. Better ones like Onsemi or Sipex 431 may be less, but their internal vref showed approx 50 nV/ rtHz at most audio freq.
I looked at your power supply, and was surprised to see the TL-431 as the regulator. How could it reach 1nV/ rtHz? From my experience, TL-431 could be quite noisy. Better ones like Onsemi or Sipex 431 may be less, but their internal vref showed approx 50 nV/ rtHz at most audio freq.
Jesper is right. The circuit is not quite obvious. The TL431 is a reference/servo to adjust the output voltage but its not part of the transient path. The actual regulator is the two transistors. The ones I have built measure around .7 nV/rtHz, really the noise of the first transistor.
Andrea- Thanks for finding that reference. I looked but ran out of time.
Gerhard has done the real work and confronted the reality- testing his efforts. It can be very sobering. Improving on whats available is difficult. Both NDK and Crystek have very good oscillators at quite reasonable prices.
Andrea- Thanks for finding that reference. I looked but ran out of time.
Gerhard has done the real work and confronted the reality- testing his efforts. It can be very sobering. Improving on whats available is difficult. Both NDK and Crystek have very good oscillators at quite reasonable prices.
This indeed is the most interesting question. With Tentlabs/GrimmAudio we think that audio-related jitter on the master clock oscillator is most serious. This happens at phase modulation frequencies of a few Hz. We also cannot explain why, but this is the outcome of years of listening to digital audio.
A jitter-poor oscillator in this frequency range makes a better sound of the (unplugged) instruments and a more precise sound stage. Why???
It is the same question as why human beings are sensitive to wow from a record player (the tone of an organ lowers at larger volumes because the resistance of the cartridge becomes bigger) and are absolute insentive to the effect that a strong note sounds lower in frequency than a soft played note generated at the same frequency!!
The sound perception of human beings is sometimes complicated.....
Herbert.
- Status
- Not open for further replies.