Really, are you sure?
There is no relation between the input and the output signals.
You have not measured the timing error (jitter).
You have measured nothing.
Moreover, those like you obsessed with measurements always forget the most important thing: an audio signal is not a single tone, it is much more complex.
Every now and then you should turn off your FFT analyzer and listen to live music.
Then maybe you could understand that DACs are not all the same.
Really, are you sure?
Please do yourself a favour and stop; you are digging yourself in an ever deeper hole.
Last edited:
Really, are you sure?
Please do yourself a favour and stop; you are digging yourself in an ever deeper hole.
More than sure.
Your THD and SNR measurements are totally useless to describe the timing error.
You forgot to quote the most important thing:
"Every now and then you should turn off your FFT analyzer and listen to live music.
Then maybe you could understand that DACs are not all the same."
Perhaps what Andrea is trying to get at is that measured phase noise is an average value. It isn't the phase path of the clock in the time domain. Both measured phase noise and dac behavior are caused by the phase path of the clock in time. In other words, it isn't simply measured phase noise that directly accounts for all of clock phase noise produced dac behavor. Masking of close-by frequencies in music is only one of the possible manifestations, although it may be the easiest to get some sense of using reasonably doable measurements. At least one might develop some sense for it by correlating measurements with listening tests.
Last edited:
There is very little to say.
Until now, no one had been able to explain the dynamics of our brain's perception of timing errors.
But today we have someone who did it with his measurements.
And the conclusion was: there is no perception of timing error, so much so that the DACs and clocks are all the same.
I wonder the meaning of the existence of manufacturers such as NDK, Crystek, MSB Tech, TotalDac and all other manufacturers who use Sabre and AKM DAC chip.
I also wonder the meaning of the existence of several DACs related threads in this forum (including this).
To tell the truth I have the answers but better I don't publish them.
Good idea, there’s no reason to educate your DIY Audio competitors for free.
How would that work if the only people hearing it are the ones who believe in it? So far I haven't seen anything in the theories on close in phase noise that map to any understanding of the human hearing.
As for peak phase noise can you REALLY detect one peak in a few hundred thousand samples?
Don't know how JosephK did the measurement either. He only mentioned using a sound card in loopback mode where the ADC and DAC are using the same clock. Apparently similar to what Audio1 did. Doesn't convince me of anything.
EDIT: Seems to me I remember someone telling Audio1 that wasn't the right way to do it.
EDIT: Seems to me I remember someone telling Audio1 that wasn't the right way to do it.
Last edited:
Bill,
I doubt that would be possible. I do know that clocks have an effect on sound quality, but I'm not yet convinced that it is only a function of the 10Hz - .1Hz band that matters.
Reason I say that is I tried Accusilicon and Crystek clocks with the same dac. Crystek were on a dac board, and Accusilicon though an external clock input. Jumpers on the board selected which clock was in use so quick A/B comparisons were possible. Both clocks were warmed up for 3-days.
The Accusilicon clock made my AK4499 sound pretty similar to Topping D90 which uses the same clocks. Great L/R imaging, big solid bass with good nuance, but at higher frequencies, say, in the vocal range and above, something sounded off. A little distorted sounding in subtle but off putting way. Not natural sounding.
On the other hand the Crystek clocks didn't have the same big bass sound, although it sounded plausible enough. Vocals sounded fine. None of whatever it was that the Accusilicon was doing to vocals.
Three different people compared the clocks using headphones (imaging was compared on the big Sound Lab electrostats). Every listener could hear the differences. Each one described them in them the about same technical characteristics, as it wasn't a preference test at that point.
A choice was made to go with Crystek based on their practical availability in the US, and for me on the basis of acceptable vocal sound quality. Should be warm and round, with chest sound part not missing (empirically hard to get that right in dacs, as we are very sensitive to vocal nuances).
I doubt that would be possible. I do know that clocks have an effect on sound quality, but I'm not yet convinced that it is only a function of the 10Hz - .1Hz band that matters.
Reason I say that is I tried Accusilicon and Crystek clocks with the same dac. Crystek were on a dac board, and Accusilicon though an external clock input. Jumpers on the board selected which clock was in use so quick A/B comparisons were possible. Both clocks were warmed up for 3-days.
The Accusilicon clock made my AK4499 sound pretty similar to Topping D90 which uses the same clocks. Great L/R imaging, big solid bass with good nuance, but at higher frequencies, say, in the vocal range and above, something sounded off. A little distorted sounding in subtle but off putting way. Not natural sounding.
On the other hand the Crystek clocks didn't have the same big bass sound, although it sounded plausible enough. Vocals sounded fine. None of whatever it was that the Accusilicon was doing to vocals.
Three different people compared the clocks using headphones (imaging was compared on the big Sound Lab electrostats). Every listener could hear the differences. Each one described them in them the about same technical characteristics, as it wasn't a preference test at that point.
A choice was made to go with Crystek based on their practical availability in the US, and for me on the basis of acceptable vocal sound quality. Should be warm and round, with chest sound part not missing (empirically hard to get that right in dacs, as we are very sensitive to vocal nuances).
Last edited:
The reason it's better not to publish them is another.
Anyway, I have no competitors, let alone the presumption of instructing someone.
I'm an ordinary hobbyist who like to measure and mostly listen to what I have measured.
From this morning until late in the evening I am writing software to pay the bill.
But you haven't figured it out yet.
First of all: I have done it right.. Sorry.
I thought I have told clearly, that I have used an EMU1616m soundcard generating a close to 18kHz frequency tone. 1.6Vrms.
Then I have fed this signal into a totally independent, not correlated clock, old type signal analyzer, an HP3561a.
Basicly I have just repeated what Syn08 had just done.
I have an older version of the HP analyzer the he had used, with an only 13bit ADC, so it can not do more than the the encountered ~80dB (86dB real life) dynamic range. But it has instrumental grade conditioning, clock, and ADC itself, so the full thing has the stability to execute very long data taking, so as to be able to arrive uHz (micro-Hertz) frequency resolutions.
I will try to translate what that spectra is showing.
I am looking at a 17997Hz signal. That has a rising edge at each 55,5648164 uSec.
I have seen a distribution, with max +-7,5mHz deviation from the main peak, down to - 75dBc level.
That is an + edge 17997,0075 Hz. That is equivalent of an 55,5647932 uSec period. The time difference between these two edges is - 23,16 pSec. It moves also the other direction, so we can have a max time difference of +_ 23,16pSec, 46,32pSec peak-peak.
We have a distribution, with 75dBc height. Expressed linearly that means a population with 5623 events at peak height.
So as a very primitive estimation, once out of 5000 edges one ! Could, maybe, get a time deviation greater than 50psec.
I will leave it to syn08 to explain it better..
I thought I have told clearly, that I have used an EMU1616m soundcard generating a close to 18kHz frequency tone. 1.6Vrms.
Then I have fed this signal into a totally independent, not correlated clock, old type signal analyzer, an HP3561a.
Basicly I have just repeated what Syn08 had just done.
I have an older version of the HP analyzer the he had used, with an only 13bit ADC, so it can not do more than the the encountered ~80dB (86dB real life) dynamic range. But it has instrumental grade conditioning, clock, and ADC itself, so the full thing has the stability to execute very long data taking, so as to be able to arrive uHz (micro-Hertz) frequency resolutions.
I will try to translate what that spectra is showing.
I am looking at a 17997Hz signal. That has a rising edge at each 55,5648164 uSec.
I have seen a distribution, with max +-7,5mHz deviation from the main peak, down to - 75dBc level.
That is an + edge 17997,0075 Hz. That is equivalent of an 55,5647932 uSec period. The time difference between these two edges is - 23,16 pSec. It moves also the other direction, so we can have a max time difference of +_ 23,16pSec, 46,32pSec peak-peak.
We have a distribution, with 75dBc height. Expressed linearly that means a population with 5623 events at peak height.
So as a very primitive estimation, once out of 5000 edges one ! Could, maybe, get a time deviation greater than 50psec.
I will leave it to syn08 to explain it better..