Mark is right.
The crystal, especially if SC-cut, is a bit like a bottle of good wine, it improves its performance over time.
It can take months to reach the maximum phase noise performance for an SC-Cut crystal.
Mainly this is due to mechanical settlements in the crystal.
I have measured several SC-Cut crystals and this phenomenon has occurred for practically all crystals.
And it's not about nuances, sometimes the improvement can reach 4-5 dB.
Even if in a less marked way, the same behavior also occurs when the oscillator is turned off and then turned on again.
The attached plots are the phase noise measurement of the same oscillator with the same crystal.
The pink curve is the phase noise we have reached after several days the oscillator was powered on.
Then we have measured again after the oscillator was switched off and the results was the blue curve.
Finally we have measured one more time the oscillator after a few days and the phase noise plot was again the one of the pink curve.
Indeed in our battery power supply we are using to power the oscillators we have provided a supplementary linear regulator to power the oscillators during the battery recharge to never turn off the oscillators.
Finally the oven is useless in audio, just a complication and a wast of money.
We have measured the phase noise of the same oscillator with and without the oven and the plots were practically identical.
Of course, without the oven the oscillator's Allan Deviation is far worse, but we don't care about that since in digital to analog conversione the long term stability does not matter.
Attachments
Thanks!
Andrea, thanks for sharing this. I have gotten tired of telling people to leave their critical digital gear powered up for exactly these reasons (among others). It is nice to have measured validation of such!
Andrea, thanks for sharing this. I have gotten tired of telling people to leave their critical digital gear powered up for exactly these reasons (among others). It is nice to have measured validation of such!
True, so is 1 nanovolt over 2 nanovolts.
But I agree that something like "100% lower phase noise compared to..." will sound good in a Stereophile infomercial.
The published plots show the phase noise of a state of the art oscillator (around -150dBc at 10 Hz from the carrier) while a ordinary Crystek has a phase noise around -97dBc at 10 Hz from the carrier.
So the impact of the described phenomenon (if it occurs in the Crystek, I have not made the same measurements with it) would be greater.
While the phase noise of an oscillator used to clock a DAC can be easily measured (although one needs expensive gear), measuring the effect of the phase noise variation at the output of a DAC is tricky.
And it is even more difficult if you want to measure the effect of a 4-5 dB change in the -150dBC region.
The DAC input should be modulated by adding -150dBc phase noise in order to measure the noise at the output.
To make this measurement you need an instrument with an ideal time base, certainly with a phase noise better than the -150dBc at 10 Hz from the carrier we want to measure.
The attached TI paper gives you an idea of how this measurement could be made.
Consider that in the test they used an integration bandwidth starting at 100 Hz, while we are talking about phase noise at 10 Hz from the carrier, so the instrumentation should be at least an order of magnitude better.
So the impact of the described phenomenon (if it occurs in the Crystek, I have not made the same measurements with it) would be greater.
While the phase noise of an oscillator used to clock a DAC can be easily measured (although one needs expensive gear), measuring the effect of the phase noise variation at the output of a DAC is tricky.
And it is even more difficult if you want to measure the effect of a 4-5 dB change in the -150dBC region.
The DAC input should be modulated by adding -150dBc phase noise in order to measure the noise at the output.
To make this measurement you need an instrument with an ideal time base, certainly with a phase noise better than the -150dBc at 10 Hz from the carrier we want to measure.
The attached TI paper gives you an idea of how this measurement could be made.
Consider that in the test they used an integration bandwidth starting at 100 Hz, while we are talking about phase noise at 10 Hz from the carrier, so the instrumentation should be at least an order of magnitude better.
And what do you think the additive phase noise of the clock tree inside the DAC is? This is all a waste of time.
It seems that for TI it is not a waste of time.
If you read the TI paper carefully you see that they have formulated a model to take DAC intrinsic noise into consideration.
Anyway I don't own the gear to make such that measurement, so my thought is irrelevant since I cannot measure the effect of phase noise and neither the DAC intrinsic noise.
From the technical point of view without proper measurements, every thought is mere speculation.
The only thing you can do is listen to the difference and draw your own conclusions.
If you read the TI paper carefully you see that they have formulated a model to take DAC intrinsic noise into consideration.
Anyway I don't own the gear to make such that measurement, so my thought is irrelevant since I cannot measure the effect of phase noise and neither the DAC intrinsic noise.
From the technical point of view without proper measurements, every thought is mere speculation.
The only thing you can do is listen to the difference and draw your own conclusions.
Well...
I am not exactly sure what your comment is meant to show? How much noise is added by the implementation of the DAC is static. If one then gives a better clock, with lower close in phase noise, the phase noise at the converter will be lower. Certainly I would not deny that the implementation of both the clock lines leading to the DAC converter (whether chip or discrete) and how the clock is implemented internally in the case of a DAC chip is relevant, but no matter how one looks at it, these factors remain the same, so when one improves the performance of the XO itself, the level of phase noise at the conversion point will be lower.
I have had the experience of hearing significant improvements with clocks which measured ~6 dB better at 2 Hz in the same DAC implementation (thanks Pat/Jocko for the measurements, RIP), so I am convinced of the value of low phase noise XOs. I have not yet experienced a level of phase noise where improvements stop happening, but there may be one: the question is how low does one need to go before improvements do not happen?
I am not exactly sure what your comment is meant to show? How much noise is added by the implementation of the DAC is static. If one then gives a better clock, with lower close in phase noise, the phase noise at the converter will be lower. Certainly I would not deny that the implementation of both the clock lines leading to the DAC converter (whether chip or discrete) and how the clock is implemented internally in the case of a DAC chip is relevant, but no matter how one looks at it, these factors remain the same, so when one improves the performance of the XO itself, the level of phase noise at the conversion point will be lower.
I have had the experience of hearing significant improvements with clocks which measured ~6 dB better at 2 Hz in the same DAC implementation (thanks Pat/Jocko for the measurements, RIP), so I am convinced of the value of low phase noise XOs. I have not yet experienced a level of phase noise where improvements stop happening, but there may be one: the question is how low does one need to go before improvements do not happen?
Now strictly technically speaking..
The difference in Andrea's curves is not finished at all at the 10Hz/ - 150dBc zone.
There is a visible difference in the close in, <1Hz region as well. And at that zone we are not talking about - 150dBc, but - 20/ 50/ 60 dBc.
And these levels /differences actually are macroscopic..
So at wander/ walk/ (actually jumps) zone, where Allan deviation starts to be a better descriptor.. And where a lot happens with warmup/ aging.
Ciao, G
The difference in Andrea's curves is not finished at all at the 10Hz/ - 150dBc zone.
There is a visible difference in the close in, <1Hz region as well. And at that zone we are not talking about - 150dBc, but - 20/ 50/ 60 dBc.
And these levels /differences actually are macroscopic..
So at wander/ walk/ (actually jumps) zone, where Allan deviation starts to be a better descriptor.. And where a lot happens with warmup/ aging.
Ciao, G
Do you not think this is well past the ridiculous? These are clocks for audio. Said audio to be listened to by mere mortals.
Yes, that's why we call this "additive phase noise". Are you sure you understand the rules of adding phase noise?
But since you claim to be able to hear such differences there's no need to answer, have a nice day.
True, and a hypothetical "phase noise magnifier" would show even larger differences @0.1Hz, @0.01Hz, 0.001Hz. We call this "1/f noise" or "flicker noise" and is a fundamental property of condensed matter in our version of the universe.
There’s no limit to the jitter fetishism. Especially to those who make money off it.
I also experienced the same improvement moving from a Crystek CCHD-957 (-97 dBc at 10 Hz from the carrier) to a better oscillator with a phase noise around -127dBc at 10 Hz from the carrier, so 30 dB better.
The improvement is huge.
Many members have experimented the same clock replacement although in different setup and they have reported great improvement in sound quality.
I don't know if there is a limit in the improvement, but I will investigate soon switching to the oscillator of which I published the plot, around 25dB better than the previous and more than 50dB better than the Crystek.
Of course I cannot provide measurements at the output of the DAC since I don't own a suitable tool (assuming there is one).
Among other effects, it has to do with the convolution of clock phase noise with an audio signal. It widens the FFT peak of an audio signal, which I believe anyone can hear with an otherwise sufficiently accurate reproduction system. Again, I believe, the mechanism here is to mask with noise audio signal that are close together in frequency, particularly one lower in amplitude than the other, something that happens constantly in music. The basic problem is not fundamentally different from the well understood problem of close-in phase noise in radar, which makes returns from small targets hard to detect when they are close to nearby clutter.
More discussion at:
ESS Sabre Reference DAC (8-channel)
In anyone is interested, the above bit of discussion followed one on noise floor modulation:
ESS Sabre Reference DAC (8-channel)
Last edited:
The region you pointed out is usually called 1 / f3 (and so on) noise, where the phase noise degrades around 30dB/octave (Leeson equation).
The 1 / f region has a slope around 10 dB/octave.