The Well Tempered Master Clock - Building a low phase noise/jitter crystal oscillator

[esgigt]

Eldam, thanks for trying to answer my question. I understand that the noise situation of a battery differs from a lineair supply. I had the latter in mind when I asked the question.
Using a low ESR mkp is one of the measures i'll keep in mind powering Andrea's xo...

 

[clsidxxl]

- andrea_mori : 2 x 5.6448MHz + 2 x 11.2896MHz + 1 X 22.5792 MHz + 1 x 24.576 MHz + 1 x 45.1584 MHz + 1 x 49.152 MHz + 10 x PCB + 4 x daughter board PCB
- Eldam : 1 x 11.2896 Mhz + 4 x PCB
- esgigt : 1 x 11.2896 Mhz + 1 x PCB, soldered if possible
- fralippo : 1 x 22.5792 MHz and 1 x 24.576 MHz + 2 x PCB + 1 x daughter board PCB
- mravinsky : 2 x 11.2896 Mhz + 1 x PCB
- TNT : 1x22.5792MHz 1x24.5760MHz + 2 x PCB, soldered if possible
- 1audio : 2x22.5792MHz 2x24.5760MHz + 2 x PCB + daughter board.
- randytsuch : 2x25.0000MHz + 4 x PCB
- TNT or maybe : 1 x 11.2896 Mhz + 1 x PCB, soldered if possible
- myint67 : 1x11.2896mhz + 1pcb also soldered if possible
- EdwardTam : 2x 16.9344MHz + 1 x 11.2896 Mhz + 2x PCB
- BDL : 1 x 11.2896 MHz + 2 x PCB
- thorstenlarsen : 2 x 11.2896 MHz + 2 x PCB
- Zoran : 1x22.5792MHz 1x24.5760MHz + 2 x PCB, (soldered or not...)
- walangalam : 1 x 11.2896 Mhz + 2 x PCB
- AR2: 1 x 22.5792 MHz and 1 x 24.576 MHz + 2 x PCB (soldering not needed)
- mcluxun: 1 x 22.5792 MHz and 1 x 24.576 MHz 1 x 45.1584 MHz + 1 x 49.152 MHz+ 2 x PCB (soldering if possible)
- damohpi: 1 x 11.2896 + 1x PCB unsoldered
- noizas : 2 x 5.6448MHz + 2 x 11.2896MHz + 4 x PCB
- badrisuper : : 1 x 11.2896 Mhz + 2 x PCB
- Clsidxxl: 1 x 45.1584 MHz + 1 x 49.152 MHz+daughter board + 2 x PCB (soldering if possible)
- zeta4 1x 22.5792Mhz + 1 x PCB (soldering not needed)

 

[BDL]

- andrea_mori : 2 x 5.6448MHz + 2 x 11.2896MHz + 1 X 22.5792 MHz + 1 x 24.576 MHz + 1 x 45.1584 MHz + 1 x 49.152 MHz + 10 x PCB + 4 x daughter board PCB
- Eldam : 1 x 11.2896 Mhz + 4 x PCB
- esgigt : 1 x 11.2896 Mhz + 1 x PCB, soldered if possible
- fralippo : 1 x 22.5792 MHz and 1 x 24.576 MHz + 2 x PCB + 1 x daughter board PCB
- mravinsky : 2 x 11.2896 Mhz + 1 x PCB
- TNT : 1x22.5792MHz 1x24.5760MHz + 2 x PCB, soldered if possible
- 1audio : 2x22.5792MHz 2x24.5760MHz + 2 x PCB + daughter board.
- randytsuch : 2x25.0000MHz + 4 x PCB
- TNT or maybe : 1 x 11.2896 Mhz + 1 x PCB, soldered if possible
- myint67 : 1x11.2896mhz + 1pcb also soldered if possible
- EdwardTam : 2x 16.9344MHz + 1 x 11.2896 Mhz + 2x PCB
- BDL : 1 x 11.2896 MHz + 2 x PCB + 1 x daughter board PCB
- thorstenlarsen : 2 x 11.2896 MHz + 2 x PCB
- Zoran : 1x22.5792MHz 1x24.5760MHz + 2 x PCB, (soldered or not...)
- walangalam : 1 x 11.2896 Mhz + 2 x PCB
- AR2: 1 x 22.5792 MHz and 1 x 24.576 MHz + 2 x PCB (soldering not needed)
- mcluxun: 1 x 22.5792 MHz and 1 x 24.576 MHz 1 x 45.1584 MHz + 1 x 49.152 MHz+ 2 x PCB (soldering if possible)
- damohpi: 1 x 11.2896 + 1x PCB unsoldered
- noizas : 2 x 5.6448MHz + 2 x 11.2896MHz + 4 x PCB
- badrisuper : : 1 x 11.2896 Mhz + 2 x PCB
- Clsidxxl: 1 x 45.1584 MHz + 1 x 49.152 MHz+daughter board + 2 x PCB (soldering if possible)
- zeta4 1x 22.5792Mhz + 1 x PCB (soldering not needed)

 

[andrea_mori]

Updated daughter board

Attached the daughter board updated schematic with Demian's low noise regulators.

The power supply is totally configurable.
You can use full on board power supply, external DC input, external regulators.

 

[clsidxxl]

Very nice Andrea,thanks.

 

[1audio]

On the regulators: the TL431 really needs no more than about 5 mA to work and should work down to 2 mA. The series resistor to the supply should be scaled to match. A smaller resistor will let more supply noise in so its worth working it out. I'm not sure what the raw supply voltage is so its important. I normally use a 3 terminal regulator as a preregulator and short circuit protection to make things easier. An LM317 works fine.

I'll try to draw up the Phase Noise test circuit I'm planning. Its pretty simple. The processing necessary to measure the noise is more involved, however.

 

[andrea_mori]

On the regulators: the TL431 really needs no more than about 5 mA to work and should work down to 2 mA. The series resistor to the supply should be scaled to match. A smaller resistor will let more supply noise in so its worth working it out. I'm not sure what the raw supply voltage is so its important. I normally use a 3 terminal regulator as a preregulator and short circuit protection to make things easier. An LM317 works fine.

I'll try to draw up the Phase Noise test circuit I'm planning. Its pretty simple. The processing necessary to measure the noise is more involved, however.

When I started designing the power supply I thought to include a pre regulation, but I wouldn't make the circuit too complex.
BTW, using SMD components the required space is limited (4 components per rail), so the attached power supply circuit was updated to accommodate the pre regulation.
Moreover the TL431's cathode current was decreased to around 5 mA.

 

[Eldam]

Hi Demian,

What could be a good cap instead ? 220 nF COG? Is the inductance important as well ? So a "small" 0605 case (for our big fingers) ?

In the thread "Moddyfing the Subbu" we have a different situation where some smt caps were tested to decoupling the 50 Mhz Fox four pads XO !

A good choice was a 10 uf smt tantalum with a 100 nf NPO, this last very small case form (0605 or smaller) and the nearest possible to the Crystal.

I was surprise to read some testimonials: some decoupling gave Roscilations: : e.g. : X7R smt 1Uf + 100 nf NPO + 10 nf NPO was badder... also 100 nf alone gave more oscilations. Didn't understand if it was an effect of the LDO micrel low noise reg (which had a tantalum at its output then a tantalum then a ferrite bead.... and then the decoupling caps I speak about !

@t Andrea : don't know if this helps, here is a recent hint of Marce fellow about leads spacing & layout : http://www.diyaudio.com/forums/digi...building-subbu-dac-v3-se-115.html#post4105974

Keep in mind a precision oscillator has the dynamics of a wirewound resistor. Its power consumption should not change. If it is changing you have a serious problem and you will not have low phase noise. Keeping that in mind its important not to think of the supply for an oscillator the same way as a supply for an amp output stage. Batteries can be very low noise but the lowest noise I think is either Nicd or Alkaline. Its the internal impedance that is the dominant element but there are other effects as well. More here: http://www.mikrocontroller.net/attachment/100819/Measurement_of_Chemical_Battery_noise.pdf

I think the effort won't offer an improvement and will add hassle. My little regulator circuit noise is pretty much the voltage noise of the transistor, around .7 nV/rtHz. Paralleling can reduce that further but in practice is probably not going to yield a detectable difference.

ESR will not ensure low noise in this application. Some caps are quieter and have a higher ESR. The critical cap for the oscillator is C4 and 1 uF seems large. It needs to be self resonant above 25 MHz (50 MHz?). Good NPO caps are probably the best option at these frequencies. http://www.mouser.com/ds/2/40/ccog-243403.pdf The cap is actually part of the oscillator feedback loop in effect.

 

[chertk]

- andrea_mori : 2 x 5.6448MHz + 2 x 11.2896MHz + 1 X 22.5792 MHz + 1 x 24.576 MHz + 1 x 45.1584 MHz + 1 x 49.152 MHz + 10 x PCB + 4 x daughter board PCB
- Eldam : 1 x 11.2896 Mhz + 4 x PCB
- esgigt : 1 x 11.2896 Mhz + 1 x PCB, soldered if possible
- fralippo : 1 x 22.5792 MHz and 1 x 24.576 MHz + 2 x PCB + 1 x daughter board PCB
- mravinsky : 2 x 11.2896 Mhz + 1 x PCB
- TNT : 1x22.5792MHz 1x24.5760MHz + 2 x PCB, soldered if possible
- 1audio : 2x22.5792MHz 2x24.5760MHz + 2 x PCB + daughter board.
- randytsuch : 2x25.0000MHz + 4 x PCB
- TNT or maybe : 1 x 11.2896 Mhz + 1 x PCB, soldered if possible
- myint67 : 1x11.2896mhz + 1pcb also soldered if possible
- EdwardTam : 2x 16.9344MHz + 1 x 11.2896 Mhz + 2x PCB
- BDL : 1 x 11.2896 MHz + 2 x PCB + 1 x daughter board PCB
- thorstenlarsen : 2 x 11.2896 MHz + 2 x PCB
- Zoran : 1x22.5792MHz 1x24.5760MHz + 2 x PCB, (soldered or not...)
- walangalam : 1 x 11.2896 Mhz + 2 x PCB
- AR2: 1 x 22.5792 MHz and 1 x 24.576 MHz + 2 x PCB (soldering not needed)
- mcluxun: 1 x 22.5792 MHz and 1 x 24.576 MHz 1 x 45.1584 MHz + 1 x 49.152 MHz+ 2 x PCB (soldering if possible)
- damohpi: 1 x 11.2896 + 1x PCB unsoldered
- noizas : 2 x 5.6448MHz + 2 x 11.2896MHz + 4 x PCB
- badrisuper : : 1 x 11.2896 Mhz + 2 x PCB
- Clsidxxl: 1 x 45.1584 MHz + 1 x 49.152 MHz+daughter board + 2 x PCB (soldering if possible)
- zeta4 1x 22.5792Mhz + 1 x PCB (soldering not needed)
- chertk : 1 x 22.5792 MHz and 1 x 24.576 MHz + 2 x PCB + 1 x daughter board

 

[1audio]

Hi Demian,

What could be a good cap instead ? 220 nF COG? Is the inductance important as well ? So a "small" 0605 case (for our big fingers) ?

In the thread "Moddyfing the Subbu" we have a different situation where some smt caps were tested to decoupling the 50 Mhz Fox four pads XO !

A good choice was a 10 uf smt tantalum with a 100 nf NPO, this last very small case form (0605 or smaller) and the nearest possible to the Crystal.

I was surprise to read some testimonials: some decoupling gave Roscilations: : e.g. : X7R smt 1Uf + 100 nf NPO + 10 nf NPO was badder... also 100 nf alone gave more oscilations. Didn't understand if it was an effect of the LDO micrel low noise reg (which had a tantalum at its output then a tantalum then a ferrite bead.... and then the decoupling caps I speak about !

@t Andrea : don't know if this helps, here is a recent hint of Marce fellow about leads spacing & layout : http://www.diyaudio.com/forums/digi...building-subbu-dac-v3-se-115.html#post4105974

Marce is really sharp and much more up on digital layout and high frequency layout than I am. He has referenced some good on line sources that you can learn from.

I would use the smallest COG cap I can manage and use a ground plane with very small "windows" (loops the signal passes around) to really contain the RF energy. Follow the current from a node through to the supply, then through the bypass to ground and back through the circuit to the node. This is the lop or window that will radiate and pickup RF. The smaller the area inside the loop the better. Look at the cap datasheet for impedance curves and select something where your operating frequency is on the descending side of the lowest point. This will be below the self resonance. While it will work with a more open layout the yield with be much less and the radiated noise much higher. You can look up the inductance of a trace which will give you a significant insight into how important short traces and small signal loops are. One of the first steps after building one is to use an FM tuner tuned to a harmonic of the oscillator and note how strong the signal is. You can also use the tuner to measure the phase noise. (See my blog). In both cases you want it minimized.

 

[esgigt]

One of the first steps after building one is to use an FM tuner tuned to a harmonic of the oscillator and note how strong the signal is. You can also use the tuner to measure the phase noise. (See my blog). In both cases you want it minimized.

Thanks for the tip. I'll check your website for it.

 

[hirez69]

- andrea_mori : 2 x 5.6448MHz + 2 x 11.2896MHz + 1 X 22.5792 MHz + 1 x 24.576 MHz + 1 x 45.1584 MHz + 1 x 49.152 MHz + 10 x PCB + 4 x daughter board PCB
- Eldam : 1 x 11.2896 Mhz + 4 x PCB
- esgigt : 1 x 11.2896 Mhz + 1 x PCB, soldered if possible
- fralippo : 1 x 22.5792 MHz and 1 x 24.576 MHz + 2 x PCB + 1 x daughter board PCB
- mravinsky : 2 x 11.2896 Mhz + 1 x PCB
- TNT : 1x22.5792MHz 1x24.5760MHz + 2 x PCB, soldered if possible
- 1audio : 2x22.5792MHz 2x24.5760MHz + 2 x PCB + daughter board.
- randytsuch : 2x25.0000MHz + 4 x PCB
- TNT or maybe : 1 x 11.2896 Mhz + 1 x PCB, soldered if possible
- myint67 : 1x11.2896mhz + 1pcb also soldered if possible
- EdwardTam : 2x 16.9344MHz + 1 x 11.2896 Mhz + 2x PCB
- BDL : 1 x 11.2896 MHz + 2 x PCB + 1 x daughter board PCB
- thorstenlarsen : 2 x 11.2896 MHz + 2 x PCB
- Zoran : 1x22.5792MHz 1x24.5760MHz + 2 x PCB, (soldered or not...)
- walangalam : 1 x 11.2896 Mhz + 2 x PCB
- AR2: 1 x 22.5792 MHz and 1 x 24.576 MHz + 2 x PCB (soldering not needed)
- mcluxun: 1 x 22.5792 MHz and 1 x 24.576 MHz 1 x 45.1584 MHz + 1 x 49.152 MHz+ 2 x PCB (soldering if possible)
- damohpi: 1 x 11.2896 + 1x PCB unsoldered
- noizas : 2 x 5.6448MHz + 2 x 11.2896MHz + 4 x PCB
- badrisuper : : 1 x 11.2896 Mhz + 2 x PCB
- Clsidxxl: 1 x 45.1584 MHz + 1 x 49.152 MHz+daughter board + 2 x PCB (soldering if possible)
- zeta4 1x 22.5792Mhz + 1 x PCB (soldering not needed)
- chertk : 1 x 22.5792 MHz and 1 x 24.576 MHz + 2 x PCB + 1 x daughter board
-hirez69: 1 x 22.5792 MHz and 1 x 24.576 MHz + 1 x 45.1584 MHz + 1 x 49.152 MHz + 4 x PCB (soldered) + 1 x daughter board

 

[Tattoo]

I didn't read the whole thread, so it could be answered, if so just kick my butt in the right direction. But I have some questions.

How are you supposed to connect this clock to the dac?
How does this connection affect the jitter performance?
How much jitter has been demonstrated to be audible?

 

[Eldam]

Marce is really sharp and much more up on digital layout and high frequency layout than I am. He has referenced some good on line sources that you can learn from.

I would use the smallest COG cap I can manage and use a ground plane with very small "windows" (loops the signal passes around) to really contain the RF energy. Follow the current from a node through to the supply, then through the bypass to ground and back through the circuit to the node. This is the lop or window that will radiate and pickup RF. The smaller the area inside the loop the better. Look at the cap datasheet for impedance curves and select something where your operating frequency is on the descending side of the lowest point. This will be below the self resonance. While it will work with a more open layout the yield with be much less and the radiated noise much higher. You can look up the inductance of a trace which will give you a significant insight into how important short traces and small signal loops are. One of the first steps after building one is to use an FM tuner tuned to a harmonic of the oscillator and note how strong the signal is. You can also use the tuner to measure the phase noise. (See my blog). In both cases you want it minimized.

The last size I can solder is 0605 smt . Andrea can have acess also to a very good measurement bench iirc, in the case he needs more than the one he has at home.

I don't have any clue of the knowledge needed for such a work... but my small understanding is the pcb drawing is also very important as all of us know it and firstly the one who is designing this board: Andrea.

UF-L are programmated to go out. I'm just worried by the fact that all the chips between the crystal and the outside (dividers) could be a to big trade off in relation to the crystal quality... and certainly need myself a second output uf-l connector to feed a device such e.g the simultanous mode board than IanCanada have for multibit DACS such TDA-1541, AD1865/2, PCM 63, PCM1704...

Thank you so much Demian to help Andrea and us for this clock board

Demian is a rare name, what is the origin ? european or purely anglo-saxon - from Great-Britain (sorry for the off topic )

 

[esgigt]

Eldam:

Damian (given name) - Wikipedia, the free encyclopedia

 

[andrea_mori]

I didn't read the whole thread, so it could be answered, if so just kick my butt in the right direction. But I have some questions.

How are you supposed to connect this clock to the dac?
How does this connection affect the jitter performance?
How much jitter has been demonstrated to be audible?

Depending on the chosen configuration, you can connect the clock board in several way:
if one choose to use the daughter board also, I think the only way is using u.fl. cable; if one choose to use a single clock board or two clock boards without the daughter board, one can use both u.fl. cable and pin strip connector to fit directly the clock close to the device to feed.
Also depending on how many outputs you are planning to use, if more than one, u.fl. cable have to be used anyway.
RF cable are not so terrifying, keep in mind that a cable with SMA/SMB connector is usually utilized to connect the device in test with the measurement system.

A good tutorial about jitter (effects, audibility and so on).

 

[andrea_mori]

A good tutorial about jitter (effects, audibility and so on).

...ooops... I forgot the link
Jitter explained - Part 1.3 [English]

 

[Tattoo]

Its difficult to find peer reviewed papers with actual listening tests on the audibility of jitter.

There are lots of theoretical papers and they give numbers of a bit less than 0.1nsec random jitter at 20kHz and 16bit. And about 10 times less than that for 20bit. If you go above these numbers then the resolution of the converter is reduced.

Here is the only one I could find with a listening test:
amorgignitamorem.nl/Audio/Jitter/Detection%20threshold%20for%20distortions%20due%20to%20jitter%20on%20digital%20audio%2026_50.pdf
This study has a 250nsec limit, about 1000x higher than the theoretical numbers, but I don't see the spectrum of the jitter they used.

 

[esgigt]

This link will do better I guess: http://amorgignitamorem.nl/Audio/Ji...ions due to jitter on digital audio 26_50.pdf

 

[1audio]

UF-L are programmated to go out. I'm just worried by the fact that all the chips between the crystal and the outside (dividers) could be a to big trade off in relation to the crystal quality... and certainly need myself a second output uf-l connector to feed a device such e.g the simultanous mode board than IanCanada have for multibit DACS such TDA-1541, AD1865/2, PCM 63, PCM1704...

Thank you so much Demian to help Andrea and us for this clock board

Demian is a rare name, what is the origin ? european or purely anglo-saxon - from Great-Britain (sorry for the off topic )

Your right to be concerned about the intermediate chips. Its called additive phase noise and the simple answer is the fastest logic you can use.
Usually for TTL its 74AC. Which is also known for creating emi problems. The really fast edges have lots of high energy harmonics. The safe bet is to put the whole circuit is a good RF tight box with coaxial feedthroughs for the signal and feedthrough caps for the power. A whole lot of work but it will work.

I was named for the lead character in the Hermann Hesse novel "Demian". My parents were beatnik's. What can I say?