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

Status
Not open for further replies.
I'm asking if the 96 of the XX.xx96 are stable !

Quartz spec accuracy are moving with time
greater the first year of use then slower after but it moves !

Is there crystals oscillator more stable then over in time ? I assume phase noise is more important than the exact accuracy of the crystal ?

I would like to swap the four crystals of my SB Duet with the appropriates XX.xxxx Mhz range if I have money and skill! Some maid it with an hearable result... but know have more change by working on simple things like good caps for having good tonal balance and no harchness.... It makes sense to me a good dac chip with a good layout needs a good crystal to obtain a better result than middle production dacs off shelves ! But for the better ones a good crystal swap can certainly help :D ; Have no experience with such things ! !
 
Last edited:
Eldam,

at this moment I own only 2 Laptech crystals, so I use them to experiment.
If you can wait a few days, I will check if there is enough interest to ask Laptech for a quote. The minimum orderable quantity is at least 5 pieces for each frequency, and the price decreases for higher quantity (10, 20, and so on).

The long term stability is not so important in digital audio. The phase noise of an oscillator is affected mainly by the short term stability.
 
The finished oscillator.

I'm updating the PCB, the new batch will be more compact, 1.75" x 1", and Ni/Au surface finish.
 

Attachments

  • TWTMC-C_01.jpg
    TWTMC-C_01.jpg
    94.8 KB · Views: 3,327
  • TWTMC-C_02.jpg
    TWTMC-C_02.jpg
    94.5 KB · Views: 3,304
My suggestion

To get 3V3 at the output, the 74HC04 should be replaced with a 74LVC04, and the +5V power supply should be replaced with 3V3 voltage.

Well, in my old HW days (DACXVCO) I used the HCU04 chip, while you use the inverter as an amplifier and the HCU type should be used!

On the front I used as you a similar circuit and after words the HCU04 as an amplifier with isolated powering (this power noise is critical) and for the output driver also a different IC, while the load would influence the power noise on the oscillator. Then put all in a nice CU box :D

PHP:
My HCU04 

->-+--- HCU04 ---+->-
   !             !
   +- R -+- R ---+
         C
         !
        GND

Cheers

Hp
 
Hi Andrea,

I can wait few month, not a question of time, you know I'm waiting Ian and Joro for inputs stages (but have a B plan to beginn something with spidf kits I already have), and I have no output stage yet. You have thousand less free time than me but sucees to make device like that ! Wow ! It's more about a passion to learn around my best hobby !

I read a little yesterday to try to be less fool ! Well I learn than jitter & phase noise are the same with crystals. I also read than TCXO are not as good that it seems for audio... I also read than the ppm stability is not so important, btw some manufacter provide already more stabilised Crystal than they get artificialy older with heating !

About the TDA 1541 and its simple 16 bits resolution & say 44K hz application: has it an intresec jitter or noise floor Under which the quality of phase noise can not be heard with it ?

If I look at the famous Crystek, i see than at 500 hz, the noise floor is -140 db ! Is it an usefull data or is it just the phase noise F and h2,h3... level which are important.

Sorry for my poor understanding, I assume all those concepts are near but I mix them badly....
 
Last edited:
Member
Joined 2004
Paid Member
A few questions on the oscillator. First for a 3rd overtone you usually need a trap to prevent the oscillator from operating at the wrong mode. I did not see it in your schematic.

Second use a 74AC logic gate for the lowest phase noise. It can be problematic since those can radiate with their very fast edges but they have the lowest phase noise of the TTL logic families (except perhaps the Potato stuff). Buffering and isolation is important. Modulating the load even a little will alter the frequency of the oscillator producing phase noise. i suspect a pretty simple inexpensive packaged oscillator with good isolation would make a big difference and be better than an over the top effort with poor isolation. There is a spec for reverse isolation on distribution amps for precision oscillators because of this issue. Usually way better than 100 dB for good ones. And not an easy one to meet. http://www.ko4bb.com/Manuals/05)_GP...ide_to_Isolation_Amplifier_Selection_NIST.pdf

I do not see an AGC, which is usually necessary for very low phase noise oscillators.

The AT crystal can be pretty good and is more pullable than an SC crystal but it has a lower Q which will mean higher close in phase noise.

i think this is a really interesting circuit: Low Phase Noise Design: Crystal Oscillators it uses the crystal as a bandpass filter as well as the oscillator to reduce the phase noise. There are a number of tricks worth studying.

The J310 is a good tradeoff between RF gain and noise but it will limit the possible phase noise. Usually a bipolar transistor is used to get the lowest phase noise. In this application it can be as much as 10 db lower. Its directly related to the input voltage noise. The low frequency voltage noise of the transistor will modulate the RF producing close in phase noise.

If you want to really go down the low phase noise road start here: http://www.hpl.hp.com/techreports/1999/HPL-1999-6.pdf There is some of the history in the article.

I designed a really low noise power supply for crystal oscillators here: http://www.diyaudio.com/forums/digi...sb-interface-audio-widget-92.html#post2863700 It has less than 1 nV/rtHz noise, important when chasing the really low phase noise, and cheap and simple to build.

I asked Rick Karlquist about changing the crystal in an HP 10811. He said it would not work, since the crystals were very specific to work in it.

In the US you can go to Wenzel/Croven to get crystals. It will be slow and expensive but they are as good as you can get. I don't think the crystal will be a limiting factor until you get a lot of the other stuff really right.

Build 2 so you can measure the phase noise. Three if you want to use the cross correlation method.
 
Member
Joined 2004
Paid Member
I'm asking if the 96 of the XX.xx96 are stable !

Quartz spec accuracy are moving with time
greater the first year of use then slower after but it moves !

Is there crystals oscillator more stable then over in time ? I assume phase noise is more important than the exact accuracy of the crystal ?

I would like to swap the four crystals of my SB Duet with the appropriates XX.xxxx Mhz range if I have money and skill! Some maid it with an hearable result... but know have more change by working on simple things like good caps for having good tonal balance and no harchness.... It makes sense to me a good dac chip with a good layout needs a good crystal to obtain a better result than middle production dacs off shelves ! But for the better ones a good crystal swap can certainly help :D ; Have no experience with such things ! !

Absolute accuracy is not critical in audio. You won't hear it if the oscillator is 50 PPM off. On the SB Duet the only oscillators you would care about ar the audio ones. the other may make it unusable. They are involved with other functions and may actually have the firmware tuned to their internal errors.

What kind of oscillators are on the PCB? There may be other ways to improve things that are less intrusive.
 
A few questions on the oscillator. First for a 3rd overtone you usually need a trap to prevent the oscillator from operating at the wrong mode. I did not see it in your schematic.

This oscillator was designed for fundamental mode crystal only. As I said, I will experiment later with 3rd overtone crystal using the Driscoll oscillator.

Second use a 74AC logic gate for the lowest phase noise. It can be problematic since those can radiate with their very fast edges but they have the lowest phase noise of the TTL logic families (except perhaps the Potato stuff). Buffering and isolation is important. Modulating the load even a little will alter the frequency of the oscillator producing phase noise. i suspect a pretty simple inexpensive packaged oscillator with good isolation would make a big difference and be better than an over the top effort with poor isolation. There is a spec for reverse isolation on distribution amps for precision oscillators because of this issue. Usually way better than 100 dB for good ones. And not an easy one to meet. http://www.ko4bb.com/Manuals/05)_GP...ide_to_Isolation_Amplifier_Selection_NIST.pdf

I specified the 74LVCU04 in the parts list document for 3V3 output, but the board can accommodate a PO74G04A or a PO74GU04A. I own some Potato inverters, so I can try them to compare in listening session. Unfortunately I don't own the gear to measure the phase noise. I can access a Agilent E5052A only a few time, so I'm planning to go to university lab only when all 3 oscillators will be ready.
As you can see in the plot, the logic gate make a great difference in phase noise compared to a comparator. I know someone has reached excellent result with a simple unbuffered '04 Pierce oscillator and cheap crystal, but since I don't own a suitable measurement system like him, I prefered to start with a circuit that was already tested and a crystal with repeatable specs. Modifying the oscillator means unpredictable result, if one cannot measure the result every time.

I do not see an AGC, which is usually necessary for very low phase noise oscillators.

As I said above, the crystal Clapp oscillator was designed by a dutch guy.
As far as I understand, he is assuming here that the gate current through R1 causes the agc-operating. The BAT81S ensures the extra AGC, limiting the gate voltage swing.

The AT crystal can be pretty good and is more pullable than an SC crystal but it has a lower Q which will mean higher close in phase noise.

I know a SC-cut crystal has higher Q, but it need an oven and so the circuit become very complex, not a simple job for a diyer. BTW, remember that a heavily polished AT-cut crystal can reach high performance, like the Laptech one.

The J310 is a good tradeoff between RF gain and noise but it will limit the possible phase noise. Usually a bipolar transistor is used to get the lowest phase noise. In this application it can be as much as 10 db lower. Its directly related to the input voltage noise. The low frequency voltage noise of the transistor will modulate the RF producing close in phase noise..

Later, with Butler and Driscoll oscillator I'll switch to bjt.
The final comparison will tell us if bjt performs better than jfet.

I designed a really low noise power supply for crystal oscillators here: http://www.diyaudio.com/forums/digi...sb-interface-audio-widget-92.html#post2863700 It has less than 1 nV/rtHz noise, important when chasing the really low phase noise, and cheap and simple to build.

Your power supply looks good and also cheap. Any PCB for that?

In the US you can go to Wenzel/Croven to get crystals. It will be slow and expensive but they are as good as you can get. I don't think the crystal will be a limiting factor until you get a lot of the other stuff really right.

I did ask Croven for a quote (5 pieces), but I got any reply from them.
I'm not sure they would supply such little quantity.
 
Sorry if I take long time to reply all questions.

@HP
Do you mean the HCU04 for 3V3 output?

@Eldam
As Demian said absolute frequency is not so important in digital audio. Also long term stability is not an issue, while is very important the short term stability.
Jitter is a bad beast, also for CD quality resolution DAC like TDA1541A. Take a look at John's (Ecdesigns) work in his SD player, he take a great care about the master clock of his system. Not for case I said "system" rather than dac, but this is another topic that would need long time to debate.
There is a nice document on tnt-audio about jitter
Jitter explained - Part 1.3 [English]
The critical phase noise in audio is tipically the subsonic region, 10 Hz from the carrier or less.


Just for comparison, looking at the 22.5792 MHz Crystek CCHD-957 oscillator, it shows -97dBc@10Hz from the carrier. Assuming -6dB for its double frequency, you get -103dBc@10Hz, that means around 30dB worst than the Clapp/Laptech oscillator.
Also the Tentlabs XO performs similar to the Crystek, -93dBc@10Hz for a 33.8688 MHz oscillator, that means around -102dBc@10Hz for 11.2896 MHz.
 
Last edited:
The new PCB layout, 1.75" x 1", Ni/Au surface finish.

The PCB provides 2 outputs, pin strip and/or u.fl. connectors.
The oscillator needs 2 separate PS, the PCB provides footprints for simple 3 terminal regulators, but low noise external regulators are strongly recommended.
The Vs rail (5V or 3V3) provides 2 pads to place a voltage drop resistor, if you want use a main external supply to feed the two regulators.

Later, I'll post a simple construction manual.
 

Attachments

  • TWTMC-C_PCB.jpg
    TWTMC-C_PCB.jpg
    114.2 KB · Views: 2,750
First a clarification.

This is not a commercial project, diy audio is an hobby and a passion for me, and my free time is very little.
Anyway, if someone is interested on the PCB and/or the crystal I could supply them.
In any case I'm not interested on gain money with my audio projects, I only look to avoid losing money, so I will supply the parts at cost (parts cost, shipping and taxes and so on). Maybe, you could donate something to diyaudio.

At this moment I don't know the exact cost of the parts.
I believe the PCB will be around 5/6 euro, depending on quantity.
For the crystal I have to ask a quote to Laptech, then I'll know the final price.
Remember that Laptech requires at least 5 pieces for each frequency, and the price decreases for higher quantity.

About the shipping cost, I would standardize the cost all over the world, shipping in a simple envelope.

Please, if you are interested, put your name and needed parts in the following interesting list. If there will be enough interest, I will ask for the quotes, I will tell you the final price.
I have some old version PCB boards (16 pcs), if someone was interested I could sell them for 3 euro each + shipping and Paypal fees.
 
Member
Joined 2004
Paid Member
The critical phase noise in audio is typically the subsonic region, 10 Hz from the carrier or less.

I have seen this claim but nothing that can support it. In terms of audibility it would take a lot of low frequency phase noise to create any audible modulations. The low frequency phase noise can be regarded as a form of wow on a turntable. Except that even the worst case LFPN is 50 dB lower that the best turntable ever for this parameter. The most sensitive region for jitter/phase noise would be in the higher frequencies, above 3 KHz. This is because the modulation products (F1 + F2 and F1 - F2) will be in the audible range and not close to any existing audio content. The linked article also mentions this.

I would be very suspicious of extraordinary claims about close in phase noise. It is by far the most difficult parameter to improve. The NIST article references performance for state of the art 5 MHz oscillators, ones that cost over $1500 ea and are considered "munitions" and restricted in sales. (I tried). If you scale those numbers (a valid thing to do) you get -108 at 1 Hz, -136 at 10 Hz, -156 at 100 Hz. Its possible to do some tricks and improve this a little. The best alternative for close in phase noise scaled to 20 MHz would be around -110 at 1 Hz, -125 at 10 Hz, and -131 at 100 Hz. http://www.oscilloquartz.com/files/1363164953-Br_ OCXO 8607.pdf . Aside from the extraordinary challenge measuring this low, claims for any oscillator meeting these numbers would need alternative third party verification.

A good tutorial http://www.ko4bb.com/Timing/Vig-tutorial_8.5.3.6_-_Jan_2007_-_compressed_pics.ppt
 
Status
Not open for further replies.