• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Matrix Audio X-SABRE Pro MQA: Best Audio DAC in the World?

LTig

Master Contributor
Forum Donor
Joined
Feb 27, 2019
Messages
5,760
Likes
9,442
Location
Europe
This is simply the intersample clipping.
Yes.
It's a very normal behaviour.
It's only normal behaviour of a DAC to a clipped recording! It will never happen with unclipped recordings so I wouldn't say that this is normal.

In this case with a 20 Hz square wave the signal is clipped. The reviewer seemingly didn't know about the Gibbs phenomenon.

For DACs with an adjustable volume, it's a none issue.
Yes, reduce the volume be 3 dB and you're mostly good.
 

LTig

Master Contributor
Forum Donor
Joined
Feb 27, 2019
Messages
5,760
Likes
9,442
Location
Europe
20 Hz (see his description). 20 kHz would not even look like a square.
At 44.1 kHz samplerate a 20 kHz square would be very nice sinus. :) Next harmonic component is at 60 kHz and should be (hopefully) suppressed.
 

LTig

Master Contributor
Forum Donor
Joined
Feb 27, 2019
Messages
5,760
Likes
9,442
Location
Europe
Atomic Bob has found clipping with the XSP. Very surprised Matrix let this slip through. Comments @MatrixAudio or @amirm? Was this fixed in the MQA version?

https://www.superbestaudiofriends.o...-pro-technical-measurements.9310/#post-300976
It's not a slip (the cause is a clipped input signal so it's no wonder that the DAC clips as well), although some DACs (see here) can handle this situation better.

We don't know how the 20 Hz square wave test signal was generated. My guess is it was simply calculated with N samples maximum positive value and N samples maximum negative value. This however
  • violates the Nyquist criterium because the transition from max negative to max positive contains frequency components above half of the sampling frequency,
  • and it disregards the Gibbs phenomenon (see below).
A correct square wave test signal must be calculated by adding all sinus components within 0 Hz and half of the sampling frequency. If you do this for a 20 Hz square wave you would see that each transitions contains over- and undershoots (Gibbs Phenomenon). This means you must reduce the level of the square wave until the over-/undershoot values are within 0 dBFS. Such a signal will not lead to clipping in a properly implemented DAC.
 

LTig

Master Contributor
Forum Donor
Joined
Feb 27, 2019
Messages
5,760
Likes
9,442
Location
Europe
This is a clipped recording.

Clipping means the level of the analog signal is higher than the maximum input level of the ADC. Looking at the digitized values only and saying clipping means more than 1 or 2 consecutive values at 0dBFs is not sufficient.
Same here. If the white noise was generated by an algorithm using random numbers you have to be very careful not to violate Nyquist and clipping.

When ripping noise from an analog source you have to watch the analog peak level meter to not clip the ADC.

@ayane wrote here that recordings with much higher intersample overs do exist.
 

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,634
Likes
1,692
This is a clipped recording.
If you scroll down, you will find an audio clip I attached. It is not recorded by an ADC, it is a chiptune -- synthesized music.
Same here. If the white noise was generated by an algorithm using random numbers you have to be very careful not to violate Nyquist and clipping.
Did you see the spectrum screenshot I attached? There is a lowpass before Nyquist.
 

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,634
Likes
1,692
My test signal is LOWPASSED. See the screenshot carefully. The one on the right, with purple color:
index.php
 

LTig

Master Contributor
Forum Donor
Joined
Feb 27, 2019
Messages
5,760
Likes
9,442
Location
Europe
If you scroll down, you will find an audio clip I attached. It is not recorded by an ADC, it is a chiptune -- synthesized music.
Then the synthesizer clipped the signal, even with sample values below 0dBFS.

I think when working with digital signals in audio people somehow forget that the end result must be transformed back into the analog world by a DAC. One must no create signals which cannot be transformed properly because they violate Nyquist and/or clip the resulting analog signal in the DAC.
Did you see the spectrum screenshot I attached? There is a lowpass before Nyquist.
Yes, but the sample still is clipped.
 

LTig

Master Contributor
Forum Donor
Joined
Feb 27, 2019
Messages
5,760
Likes
9,442
Location
Europe
My test signal is LOWPASSED. See the screenshot carefully. The one on the right, with purple color:
index.php
I do not dispute this. Still the noise signal may be clipped. The left side does not show the analog signal. It shows the display of the digital values which is not the same as the analog signal it represents.
 

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,634
Likes
1,692
Then the synthesizer clipped the signal
Synthesizers don't have to be bandlimited, especially with the oldskool chiptune genre:
https://en.wikipedia.org/wiki/Chiptune

I do not dispute this. Still the noise signal may be clipped. The left side does not show the analog signal. It shows the display of the digital values which is not the same as the analog signal it represents.
And therefore the clipping threshold is defined by the DAC, and it is unfair to call the digital file "clipped". If a DAC has a filter that generates higher peaks than others, then that DAC will clip, but not the other DACs.
 

LTig

Master Contributor
Forum Donor
Joined
Feb 27, 2019
Messages
5,760
Likes
9,442
Location
Europe
And therefore the clipping threshold is defined by the DAC, and it is unfair to call the digital file "clipped". If a DAC has a filter that generates higher peaks than others, then that DAC will clip, but not the other DACs.
This is where our opinions differ.

I look at the complete audio chain, ADC -> processing >- DAC:
  • An analog signal captured by an ADC must not contain frequencies above Nyquist, and its level must not be higher than the ADC can handle.
  • Such an unprocessed digital signal can be transformed back into the analog world by a DAC without clippling of the DAC or its analog stage as long as the DAC is properly implemented. This is how digital audio should work in the first place (I know it doesn't).
  • If the digital signal is processed in any way the processing algorithm must ensure that such a DAC does not clip, otherwise the processing is bad.
  • The same is true when generating artificial digital signals, being it test signals like the 20 Hz square wave, or chip tune music.
 

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,634
Likes
1,692
This is where our opinions differ.

I look at the complete audio chain, ADC -> processing >- DAC:
  • An analog signal captured by an ADC must not contain frequencies above Nyquist, and its level must not be higher than the ADC can handle.
  • Such an unprocessed digital signal can be transformed back into the analog world by a DAC without clippling of the DAC or its analog stage as long as the DAC is properly implemented. This is how digital audio should work in the first place (I know it doesn't).
  • If the digital signal is processed in any way the processing algorithm must ensure that such a DAC does not clip, otherwise the processing is bad.
  • The same is true when generating artificial digital signals, being it test signals like the 20 Hz square wave, or chip tune music.
Yes, something like this:
https://www.audiosciencereview.com/...in-intersample-overs-please.11651/post-336772
 

Rusty Shackleford

Active Member
Joined
May 16, 2018
Messages
255
Likes
550
Let's think in another way. If a DAC does not clip at full volume when playing test signals like square wave, then it just means it has a forced digital attenuation stage before going through the digital filter. It is not really "bypassed" or "disabled".

Jds Labs' view:


On the other hand, if a DAC does not have a forced headroom, then you have full control of the digital attenuation yourself.

How did Benchmark prevent clipping in the DAC3? I don’t think the implication is that thr DAC3 isn’t bit perfect, is it?
 

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,634
Likes
1,692

Rusty Shackleford

Active Member
Joined
May 16, 2018
Messages
255
Likes
550

citraian

New Member
Joined
Aug 30, 2019
Messages
4
Likes
2
Really interesting discussion about the intersample clipping.
It's a bit unclear to me how big a problem this is for the "at home audiophile". In other words, should a non-pro user worry about this at all?

Should we:
a) attenuate in the OS / playback software
b) run the DAC in PRE mode and change the DAC's volume to -3 dB
c) keep everything at full volume in the OS, playback software and DAC since this is not that big of a deal for non-pro users
 
  • Like
Reactions: jjk
Top Bottom