Measurements Of SMSL M8 DAC

[amirm]

Well I'd be really curious to simply see a white or pink noise signal tested and a frequency response and THD+N based on this.

You can't do THD+N with a noise signal. Think about it.

Frequency response is easy. See my latest one I posted: https://www.audiosciencereview.com/...and-measurements-of-smsl-su-8-dac.3778/page-6

 

[JustIntonation]

You can't do THD+N with a noise signal. Think about it.

I am thinking about it And I don't see why it's not possible? Could you explain?
I do understand any timing error would displace the noise signal giving meaningless THD+N, but I mentioned correcting for timing errors (don't know how to do this with existing tools, but it can surely be coded). After this correction we can compare the waveform of the noise before and after and any deviation is the THD+N it seems to me?

 

[JustIntonation]

It is of course not true all types of distortion / errors show up with a sine equally. Take intermodulation distortion as an example. You need at least two sines to show this.
I'm just really curious how a DAC performs with white or pink noise..

 

[Blumlein 88]

What you say would only hold for harmonic distortion. It would not hold at all for intermodulation distortion and it would not hold for semi random correlated errors of a DAC either (different from noise) it seems to me. Neither would it hold for timing errors / jitter.
(edit: and harmonic distortion is probably the least audible type of distortion there is.. and sine tests fit it perfectly don't see any advantage white noise based test could give over testing harmonic distortion but nice that you argue it could be distinguished from other distortion types even with a white noise test. In fact, there are reasons to think that strong harmonic distortion with music is mostly perceived pretty much like you describe by the human ear, as a rising frequency response )

Well what I have done looking for various spurious output is playback a white noise file with a frequency region gapped out. I would remove frequencies between 3-5 khz as those are where we hear most sensitively. All the other frequencies are there, and if highly distorting in either harmonic or IMD then some of those by products would fill in the empty region. You don't get anything there other than the base noise floor of the DAC.

You really need to learn a bit about signals and what is going on with such things. Your imagining what is possible is not meshing with how things really work.

 

[Blumlein 88]

https://www.audiosciencereview.com/...-of-focusrite-forte-recording-interface.3100/

Look at the results for a DAC I posted here. The spectrogram down a bit in the first post should help. I show a twin tone sweep for IMD. I explain what those show in that post.

 

[JustIntonation]

Well what I have done looking for various spurious output is playback a white noise file with a frequency region gapped out. I would remove frequencies between 3-5 khz as those are where we hear most sensitively. All the other frequencies are there, and if highly distorting in either harmonic or IMD then some of those by products would fill in the empty region. You don't get anything there other than the base noise floor of the DAC.

You really need to learn a bit about signals and what is going on with such things. Your imagining what is possible is not meshing with how things really work.

Aah that is clever
But I can't fully say this is conclusive.. Were you able to detect the harmonic and IMD in the empty region? (which would only show up as a higher noise level in the empty region than the DAC playing silence it seems to me)

But I do understand some of the basics about audio and math.
But an actual DAC is not an ideal world by far, it may well behave differently dealing with noise than with sines is my logic..

 

[JustIntonation]

https://www.audiosciencereview.com/...-of-focusrite-forte-recording-interface.3100/

Look at the results for a DAC I posted here. The spectrogram down a bit in the first post should help. I show a twin tone sweep for IMD. I explain what those show in that post.

Thank you! But the DAC shows the most errors / distortion with that very measurement it seems? Would not white or pink noise be the same test but "on steroids" for the DAC? Possibly giving far higher distortion?

 

[amirm]

I am thinking about it And I don't see why it's not possible? Could you explain?

THD+N is sum of harmonics and noise of a single tone. We take out the single tone and all that is left is the THD+N. With white noise you have full spectrum of signal. All create distortions that land on top of the signal. So there is no separating the distortion or noise from the signal.

 

[JustIntonation]

THD+N is sum of harmonics and noise of a single tone. We take out the single tone and all that is left is the THD+N. With white noise you have full spectrum of signal. All create distortions that land on top of the signal. So there is no separating the distortion or noise from the signal.

Sorry to be so annoying. But if we take out the exact noise signal put in by simple phase inversion with correct timing then we will have the THD+N left it seems to me?
Perhaps hard to implement in the analogue world, we'd have to invert the phase of the noise signal and add a perfect delay to match the delay of for instance the ADC+DAC being tested. But in digital it should be possible to phase invert the noise signal, level match and put it at the perfect timing?
Same should hold for actual music instead of a sine or pink noise. Play actual music through a DAC, record with an ADC, then phase invert the original song, level match and time it exactly right and what you're left with is the THD+N? As I said before timing errors of the DAC+ADC would mess this up but this could be corrected for with an algorithm it seems to me.

edit: oh I see what you mean now sorry.
I took THD+N more freely as simply anything that isn't the original signal no matter what the original signal is. But if the definition is based on a single sine then yes it's not technically correct to call it THD+N. Don't know what to call it then..

 

[dc655321]

But I do understand some of the basics about audio and math.
But an actual DAC is not an ideal world by far, it may well behave differently dealing with noise than with sines is my logic..

No. A DAC will not treat input noise and sine signals differently.

Here is a plot of a multi-tone test.
In that test, a signal of 42 tones is put through a DAC - and 42 frequencies come out.
Imagine repeating that with a signal containing 100 or 1000 (or more) frequencies mixed together.

As you introduce ever more tones, the output spectra becomes more and more crowded, until it eventually looks like white noise spectra.
Because that's what white noise is: all frequencies are present. Any distortion introduced would be indistinguishable from deliberately present signal.

 

[Ron Texas]

In the end, this is a discontinued product.

 

[JustIntonation]

No. A DAC will not treat input noise and sine signals differently.

Here is a plot of a multi-tone test.
In that test, a signal of 42 tones is put through a DAC - and 42 frequencies come out.
Imagine repeating that with a signal containing 100 or 1000 (or more) frequencies mixed together.

As you introduce ever more tones, the output spectra becomes more and more crowded, until it eventually looks like white noise spectra.
Because that's what white noise is: all frequencies are present. Any distortion introduced would be indistinguishable from deliberately present signal.

Thank you! Have you measured IMD with the 42 tones relative to 2 tones? Is it higher for 42 tones relative to 2 tones? If so, can it be extrapolated to what the total IMD would be for white noise?

 

[Blumlein 88]

Thank you! But the DAC shows the most errors / distortion with that very measurement it seems? Would not white or pink noise be the same test but "on steroids" for the DAC? Possibly giving far higher distortion?

No, a two tone max level test is about as stressful as you can get. Here is what it looks like.

It goes from maximum to nothing 1000 times per second with high slew rates.

Now if I use 4 tones I have to reduce them in amplitude by 50%. If I use 8 tones I reduce by 50% again or 1/4 this amplitude. All of which are less stressful to reproduce than the one above. So contrary to your intuition that thousands of frequencies in noise would be the test "on steroids", it in fact is extremely easy and not much of a test at all because of the reduced amplitudes of the waves. Remember from Fourier that any signal, even nosie, can be made into an equivalent result in a series of sine waves. So thousands of low amplitude sine waves are not as much a test of non-linearity as two max level sine waves. Plus with only two waves we can analyze the result simply to have an idea where the non-linear behavior is.

 

[Blumlein 88]

Sorry to be so annoying. But if we take out the exact noise signal put in by simple phase inversion with correct timing then we will have the THD+N left it seems to me?
Perhaps hard to implement in the analogue world, we'd have to invert the phase of the noise signal and add a perfect delay to match the delay of for instance the ADC+DAC being tested. But in digital it should be possible to phase invert the noise signal, level match and put it at the perfect timing?
Same should hold for actual music instead of a sine or pink noise. Play actual music through a DAC, record with an ADC, then phase invert the original song, level match and time it exactly right and what you're left with is the THD+N? As I said before timing errors of the DAC+ADC would mess this up but this could be corrected for with an algorithm it seems to me.

edit: oh I see what you mean now sorry.
I took THD+N more freely as simply anything that isn't the original signal no matter what the original signal is. But if the definition is based on a single sine then yes it's not technically correct to call it THD+N. Don't know what to call it then..

The spectrograms I linked to show the main signal, and then anything, hum, spurious noise, harmonics, anything above the set floor of the spectrogram. So when you looked at the spectrogram with the floor set to -120 dbFS, it means anything at or higher in level than -120 db will show up on that spectrogram. If you use noise, then you already have something at every frequency and it tells you nothing.

As for correcting for time, and level and inverting doing a null test, those are very hard to get right. Diffmaker tries to do that and even sometimes works. But when that has been managed it shows about what we expect with a low distortion device. There aren't big differences left over between different good measuring bits of gear.

 

[Blumlein 88]

Thank you! Have you measured IMD with the 42 tones relative to 2 tones? Is it higher for 42 tones relative to 2 tones? If so, can it be extrapolated to what the total IMD would be for white noise?

Did you look carefully at the link on the 42 tones. Those were actual results and no there was no huge IMD in the result. Just the underlying noise floor.

 

[JustIntonation]

No, a two tone max level test is about as stressful as you can get. Here is what it looks like.

View attachment 14363

It goes from maximum to nothing 1000 times per second with high slew rates.

Now if I use 4 tones I have to reduce them in amplitude by 50%. If I use 8 tones I reduce by 50% again or 1/4 this amplitude. All of which are less stressful to reproduce than the one above. So contrary to your intuition that thousands of frequencies in noise would be the test "on steroids", it in fact is extremely easy and not much of a test at all because of the reduced amplitudes of the waves. Remember from Fourier that any signal, even nosie, can be made into an equivalent result in a series of sine waves. So thousands of low amplitude sine waves are not as much a test of non-linearity as two max level sine waves. Plus with only two waves we can analyze the result simply to have an idea where the non-linear behavior is.

Ok thank you. So I am likely looking for an explenation in the wrong area.
Sorry for taking so long to get that.

Any clue as to why I'm hearing subtle differences between two good DACs then? And possibly between DAC chips?
But perhaps I've asked enough questions for today..

 

[Blumlein 88]

Ok thank you. So I am likely looking for an explenation in the wrong area.
Sorry for taking so long to get that.

Any clue as to why I'm hearing subtle differences between two good DACs then? And possibly between DAC chips?
But perhaps I've asked enough questions for today..

I don't know. The most common by far is frequency response. Relatively minor FR differences over an octave or two can be heard. If I remember right this is over headphones, then likely an interaction in the phone and output impedance of the DAC. That or if you were to hear it without knowing which is which, then maybe they aren't really different. Hard to believe until you test yourself how very little without us noticing our perception is effected by things in our heads.

 

[JustIntonation]

I don't know. The most common by far is frequency response. Relatively minor FR differences over an octave or two can be heard. If I remember right this is over headphones, then likely an interaction in the phone and output impedance of the DAC. That or if you were to hear it without knowing which is which, then maybe they aren't really different. Hard to believe until you test yourself how very little without us noticing our perception is effected by things in our heads.

Maybe it is indeed frequency response.. Too bad don't have time anymore to arrange for an ADC and test the M8A v3.
I can test my D2 later in the week though, will do that.

And just maybe it could be a difference in intermodulation distortion?
Here the M8A v1 tested in this thread:
https://audiosciencereview.com/forum/index.php?attachments/upload_2018-3-12_21-46-58-png.11303/
And here tested by Amirm:
https://audiosciencereview.com/foru...-smpte-intermodulation-measurement-png.11437/

And here the intermodulation distortion of the D2:
https://web.archive.org/web/20161206045028/http://www.anedio.com:80/index.php/product/d2_measure
With a harder test. Equal volume 19+20KHz. Versus the easier SMPTE test done on the M8A v1 still giving higher IMD (roughly 10dB difference).
And it's likely my M8A v3 has stronger IMD at mid levels as all the ES9038Q2M show an extra mid level IMD bump so far.
Could be well over 30dB of difference for IMD distortion at mid levels if you add all these things up it seems to me. Perhaps audible?

 

[amirm]

Sorry to be so annoying. But if we take out the exact noise signal put in by simple phase inversion with correct timing then we will have the THD+N left it seems to me?

You can't do that because the signal itself gets transformed linearly by the device (amplified, phase change, etc.).

 

[JustIntonation]

You can't do that because the signal itself gets transformed linearly by the device (amplified, phase change, etc.).

Aha. Phase change I didn't think of that. (this could be detected and corrected for though it seems to me)
But thank you! I've learned a lot in this exchange.