Measuring additive phase noise or jitter is its own subgroup of low noise oscillators. Here is an article that shows some techniques and that high slew rate in digital buffers loweres the added jitter: Understand, Measure, And Optimize Clock Distribution Additive Jitter To Achieve Best Performance | Analog content from Electronic Design
Allen Deviation or stability may be a better measure for close in phase noise (they are different ways of looking at the same phenomena). Here is an article showing the relationship between noise and adev Allan deviation and Averaging and more on the subject http://tycho.usno.navy.mil/ptti/2004papers/paper14.pdf and this on how the crystal processing affects close in phase noise etc. : Morion, Inc :: News
This is a well studied subject.
By the way all the ultra precision oscillators i have seen have a voltage trim for the frequency, usually 1 to 10 ppm. They are there to correct for aging. In my testing they can be modulated to about 1 KHz successfully.
Allen Deviation or stability may be a better measure for close in phase noise (they are different ways of looking at the same phenomena). Here is an article showing the relationship between noise and adev Allan deviation and Averaging and more on the subject http://tycho.usno.navy.mil/ptti/2004papers/paper14.pdf and this on how the crystal processing affects close in phase noise etc. : Morion, Inc :: News
This is a well studied subject.
By the way all the ultra precision oscillators i have seen have a voltage trim for the frequency, usually 1 to 10 ppm. They are there to correct for aging. In my testing they can be modulated to about 1 KHz successfully.
************************************************
Hi Jesper.
You are right. All the gates and the MUX unit will be in the clock-signal path.
As Herbert correctly pointed out, the proper way to do this would be to design an x number
of INDEPENDENT oscillator circuits, each with its own dedicated signal-converter, and only then
use something like a MUX to select the proper clock frequency for the associated sample rate.
Although the idea seems adequate at first, it fails to deliver, sonically.
This was the design used in a commercially available proffesional equipment
many years ago - which I was involved, and intimately familiar with.
It performed and measured very well, but through carefull listening sessions, it was clear
that MUX was a bad idea..
Even using a single inverter (74HC04) that you guys love so much, seriously degraded the sonics, but PN measurements were fine.
You guys won't see me repeat this anymore....stay away from over-using 74 series of logic devices in your clock path.
5 years of intensive fesearch resulted in this observation. You won't believe me? Fine...I respect that.
Keep up the good work gentlemen.
Could you give me articles in which you treet the subject and prove your allegations? Loos catchphrases do not help any further!
Herbert.
Hi again ... I have a question that I hope maybe Herbert, Gerhard, Alexiss, Andrea & 1audio could take an interest in elaborating on ... as it is 1audio already gave fine feedback on this some time ago but I've just been wondering if there might be other solutions ...
As it is most audio systems have both 44.1 & 48 kHz based oscillators which means that there needs be a way to switch between these two frequency bases. Also, if I remember correctly Andrea (?) earlier in the thread mentioned that having the individual oscillator on for some time improves the oscillator's jitter/phase noise. To this end 1audio (Demian) earlier suggested: "you can either use some nand gates to switch as well as buffer or add a CMOS mux to switch between oscillators" This sounds like good advice, however, since such a "switch" is directly in the oscillator path I was wondering if there may be some "complementary" approaches to this switching ... ? Maybe a preferred MUX or NAND gate ... ?
...Spring is approaching
Cheers,
Jesper
Better if the two oscillators are left always powered on, so they should include their own sine to square wave converter.
To switch between them, I would use an RF relay.
Sorry, but the articles you mentioned are outside the subject! We here are talking about short term stability in the range of -130dBc/Hz@10Hz. The long term stability is of no importance (so is the precise frequency) because with the resulting audio this is undetectable!!!
Herbert.
Sorry Andrea, but this is absolutely wrong! You must switch off the unused oscillator because of leakage through the switch.
BTW: there are MUX-es that outperform the best relais with this low voltages.
Herbert.
Hi Herbert and thank you for your response.
Unfortunately I do not have permission to disclose the content of these
documents to the public, as this research, along with many others, were conducted some years ago at a company I was working at.
As a former Phillips engineer I am sure YOU are familiar with the regulations.
There are however some articles and references at Delft lnstitute of Technology's digital library.
Unfortunately these are not available to the public...only academicians, researchers, and some students.
I believe every employee at NXP could find these docs. Good luck ;-)
Keep up the good work, gentlemen.
True if you use a MUX, but I would not put one more active device in the clock path. Using a 50 ohm RF relay you have no issue leaving both the oscillators powered on. I'm not sure that a MUX is better than a mechanical contact.
Now again, I admittedly am not an expert in this field - but I've had a similar thought. Maybe use two relays so that there's one per XO and thus possibly reduce the leakage between oscillators ...
E.g. Potsemi specifies for one of their fast MUXes that the isolation between input and output is ~60 dBs ... I'm thinking that low level leakage between the non-active oscillator MUX input could affect the active oscillator's output .. Not to mention that one more active IC may introduce spurious noise from crosstalk - and through the PSU line ...
Well, some thoughts.
Cheers,
Jesper
Often subjects like these are published. They are no secrets of state... With Philips (even NatLab) this was not that serious as long as really new findings had been patented.
Moreover this attitude is history since the word innovation has been found out (Science Park Eindhoven), so I guess, you are a bit too stately.....
Herbert.
Last edited:
**********************************************
Hi Herbert, and thank you for your response.
Yes, you are right, but only if they are PATENTED, not otherwise.
Many proprietary, product-related research documents are never published
(like the research papers conducted by VEGA of Germany, on their Pulse-Radar
Transceiver sensors).
I believe what you are referring to is research-related papers (like the works of
Van der Plassche on super-speed SoC conveyors), which are now available to the academicians.
You are absolutely right. I am a little uncompromising here, but since I am no longer
associated with that company (a former professional audio company), I can only quote
the results of the work without detailed reference to the original content of the
document (and I hope have done that here).
Keep up the good work, gentlemen.
Last edited:
Did you look at any of them? The Naval Observatory article goes directly to the issue of short term stability. Regardless, there is much to learn and some very good work that has been done on close in phase noise.
Sorry Damian, I used the words 'short term' stability (very often used in the articles) but I tried to say: close in noise, which is at least in the articles a totaly different phenominon. The first article does not go below 12 kHz which is sufficient for communication purposes.
The Naval Observatory article treets an intelligent measuring method indeed but NO oscillator circuits, or other hints, to achieve the goal.
Herbert.
Last edited:
I thought the Morion stuff was interesting but no real hints. There is more on close in phase noise but usually focused on multiplied to microwave frequencies where it's multiplied 1000x and 10 Hz becomes the full communications bandwidth so more important. The 12KHz to 20 MHz band is networking focused so lots of business but little relevance to audio. However most specs are there so we see it. Some audio companies quote those numbers because they are better than audio reality. Still the techniques are interesting.
The usual term I see for "slicer" is "squarer" which helps match to some of the other discussions.
Sent from my SGH-M919 using Tapatalk
The usual term I see for "slicer" is "squarer" which helps match to some of the other discussions.
Sent from my SGH-M919 using Tapatalk
Demian, Herbert, Alex, Jesper, and so on,
although hard to reach, we well know where we have to start to get low phase noise: the close in phase noise, below fc 30 db/decade in the Leeson's model, comes mainly from the crystal's Q. Then from the flicker noise of the active device. Finally from the spurious of the power supply. Noise from the power supply affects mainly the central band of the Leeson's model, below f0/2QL 20 db/decade. The most important band of the Leeson's model for audio is the fisrt, below fc.
So:
- we need very high Q crystals (harmonic SC-Cut type)
- we need to get high loaded Q with the crystal in circuit (the crystal should see a very low impedance, as low as possible)
- we have to pick a low flicker noise device (SC-cut overtone crystals have high ESR, so a jfet does not overcome the losses, and then we have to choose a low noise bjt)
- finally we need a very clean power supply
Since we don't care the aging of the crystal, we could choose a high drive level, that usually means lower phase noise.
If we place the crystal in the emitter circuit, we need heavy bias of the bjt, the higher the current the lower the impedance seen from the crystal.
Obviously, we have to take the maximum care about the PCB layout.
About the squarer, I think it was enough demonstrated that a unbuffered '04 gives the best performance in close in phase noise.
Not simple, not impossible.
although hard to reach, we well know where we have to start to get low phase noise: the close in phase noise, below fc 30 db/decade in the Leeson's model, comes mainly from the crystal's Q. Then from the flicker noise of the active device. Finally from the spurious of the power supply. Noise from the power supply affects mainly the central band of the Leeson's model, below f0/2QL 20 db/decade. The most important band of the Leeson's model for audio is the fisrt, below fc.
So:
- we need very high Q crystals (harmonic SC-Cut type)
- we need to get high loaded Q with the crystal in circuit (the crystal should see a very low impedance, as low as possible)
- we have to pick a low flicker noise device (SC-cut overtone crystals have high ESR, so a jfet does not overcome the losses, and then we have to choose a low noise bjt)
- finally we need a very clean power supply
Since we don't care the aging of the crystal, we could choose a high drive level, that usually means lower phase noise.
If we place the crystal in the emitter circuit, we need heavy bias of the bjt, the higher the current the lower the impedance seen from the crystal.
Obviously, we have to take the maximum care about the PCB layout.
About the squarer, I think it was enough demonstrated that a unbuffered '04 gives the best performance in close in phase noise.
Not simple, not impossible.
Dear Andrea,
All the arguments above are true, but difficult to achieve and if you canot beat them then join them!
First I'm not sure if the Q of the Xtal is so important as is claimed. Of course it should be as high as possible but I think that the HF-current is more important at, say, 10 Hz from the carrier. This means that the Rm should be small otherwise the power dissipation in the Xtal will be too high.
Conclusion: take a FT-cut with a FET.
Another question is: why should the Xtal operate at its serial resonance? For the close in noise I could find no reason. Make a tuned circuit with rather low reactances and couple the FET as loose as possible to it.
With this approach I build oscillators with realy low close in phase noise! The noise figures at about 100 or 1000 Hz are not so good but hardly of interrest for audio applications...
The clean power supply counts very much for the squarer (thanks Demian)!!
Last edited:
Herbert,
IMHO, there is a valid reason: the crystal acts as a filter itself, lowering the phase noise. Take a look at the Driscoll oscillator, the crystal in the emitter circuit controls the gain of the bjt and at the same time operates as a bandpass filter with very high Q.
But do you know the bandwidth of that (only third order) filter? Even with a loaded Q of 500.000 the bandwidth is 25 Hz at 11.2... MHz, so it filters less than 3 dB at 10 Hz from the carrier!!!
What phase noise figures did you measure with a Driscoll oscillator? People are echoing each other but do not come with solutions underpinned with results (= measurements)!
BTW it is not the Xtal that controls the gain in a Driscoll....
Herbert.
Last edited:
A 3rd overtone SC-Cut crystal at 11 MHz has a Q around 1M or higher, so the loaded Q could reach 1M. The filtering operation is a little better.
I have not yet measured anything with the Driscoll, at this moment I'm still waiting for the delivery of the SC-Cut crystals.
Krieger and Matthys did work hard comparing most of the oscillator circuits. They got the best performance below 100 MHz with harmonic Driscoll oscillator:
"... where harmonic circuits are used, the Emitter coupled harmonic (Driscoll) is the best circuit up to 100 MHz, and the Butler emitter follower is the best one above 100 MHz. The Emitter coupled harmonic is relatively independent of power supply changes, its crystal waveform is good, and it has the best short term stability of any of the harmonic circuits"
About the bjt gain in the Driscoll oscillator, again from Krieger-Matthys Crystal Oscillator Circuits:
"The crystal is connected as the transistor’s emitter load impedance, and controls the oscillation frequency by controlling the transistor’s gain and phase shift. The circuit’s very good short term frequency stability comes from the high capacitance load shunted across the crystal by the transistor’s emitter output, which keeps the crystal’s in-circuit Q high. The large capacitive loading on the crystal means the resistive loading losses that reduce in-circuit Q will be low. In addition, the crystal has direct emitter control of the transistor’s gain. This is almost an optimum crystal oscillator circuit, the crystal is located at the lowest power point in the circuit (minimum crystal heating), and the output signal is taken at the highest amplitude point in the circuit (maximum signal/noise ratio and minimum external amplification). In practice, the crystal’s internal impedance, which is lowest at series resonance, controls the transistor gain. It can also vary the normal 180“ phase shift through the transistor stage by 250”. The crystal uses these gain and phase shift mechanisms to control the oscillation frequency."
They claimed that the Q of the crystal is crucial to get the best short term stability. I think they were right: the higher the loaded Q, the lower the sideband noise close to the carrier.
- Status
- Not open for further replies.