John Atkinson: Yes, MQA IS Elegant...

[fas42]

Go back to the post I responded to above - I disagree, very strongly, with everything that was stated in that post - the explanations, and solutions are a nonsense to me ... hence my reaction.

 

Luckily, no-one else on CA responds to posts they have this reaction to - sorry about being the odd man out ...

[Jud]

6 hours ago, vl said:

 

If they are dumb enough to do that, they deserve second rate sound quality.

 

But do we? 

 

However, apodizing filters can help with that if need be.

[Fokus]

13 hours ago, John_Atkinson said:

As I have said before, spectral analysis shows that the "band-limited impulse" I used has no content above 22.05kHz.

 

I had to be brief yesterday, but now I am back.

 

You used Bias Peak for the sample rate conversion, i.e. anti-alias filtering followed with downsampling. If you did spectral analysis on the result you could only have done this after the downsampling, which is wrong. You should have inspected the naked output of Bias Peak's anti-alias filter, prior to downsampling.

It is known that the majority of commercial tools (and hardware) use half-band filters, including Bias Peak. These are only 6dB (*) down at Fs/2 and thus do not provide correct band-limiting, violating the sampling theorem.

 

You should have used a SRC that guarantees that the stopband is reached before Fs/2.

 

(* Or a similar value, as many filters are a cascade of elementary half-band and non--HB filters.)

 

 

[Fokus]

10 hours ago, vl said:

Wouldn't the use of a gentler, non brick wall filter to reduce the signal bandwidth to LESS THAN the Nyquist frequency make more sense?  This will avoid excitation of the filters at the Nyquest frequency.

 

For base rate (1x) signal you always need a close-to-brick-wall filter, but you can take a transition band wide enough to make its ringing of the same order as the inherent time constant of the upper band of the cochlea of a healthy teenager (*). For 2x rate and higher you can of course roll off much smoother.

 

These are the filter configurations I use, always reaching full stop-band at Fs/2:

-44.1 kHz: 4 kHz transition band width (i.e. starting at 18kHz)

-48 kHz: 6 kHz transition

-96 kHz: 18 kHz transition

 

Recordings made like this do not trigger any DAC ringing, ever.

 

(* If this were important, which I do not know because I am no longer a healthy teenager. But neither is anyone else in this discussion, or, for the matter, in the entire audio industry.)

 

[Shadders]

7 hours ago, fas42 said:

 

To me, the madness is that everyone blames the recordings - their own playback rig is blemish free, and they use it to sort the "good" from the "bad" ... as in, my car allows me decide which are good, versus bad roads - if the suspension of the vehicle stops me going around a corner at a certain speed, well, it's the civil engineers who got it wrong, they should be punished for their misdemeanours ...

 

I find this idiocy of thinking annoying, and now and again I comment upon it - to redress the balance.

Hi,

The following may be of interest :

https://www.thedailymash.co.uk/news/lifestyle/man-who-got-surround-sound-looking-for-new-ways-to-piss-money-up-the-wall-20180830176831

Regards,

Shadders.

[adamdea]

13 hours ago, John_Atkinson said:

 

The inference to be drawn is that every musical transient in a CD master will be accompanied by sinc-function ringing at Nyquist, either from the original A/D converter's anti-aliasing filter (if the recording was made at 44.1kHz), or from the sample-rate converter's low-pass filter used to create the master from 2Fs or 4Fs files. It seems incontrovertible, therefore that that ringing will excite the playback DAC's reconstruction filter, which will impose its own ringing on musical transients.

Read more at https://www.stereophile.com/content/zen-art-ad-conversion

 

John Atkinson

Editor, Stereophile

 

That part of your article is not clearly illustrated or argued. You do not clearly distinguish between ringing in the 16/44 data output of the src and that imposed by the dac (other than by passing assertion).

 

I seem to remember that linear phase filters are idempotent? I thought that meant that a second linear phase filter does not need to impose any ringing.

 

As fokus points out, the sampling theorem requires the signal to have no energy *at* Nyquist. If the signal has no energy at nyquist and the dac’s filter reaches stopband at nyquist, why will the dac’s filter ring? 

 

Conversely the point which your article seems to skate over is that if there is ringing in the 16/44 data then a slow/lazy/non  filter in the dac will just pass the ringing and there is no time domain advantage in such a filter. So the time domain blather about such filters is pure bull. 

 

And who says your made up shape is a musical transient? Where is an example of a genuine musical transient with that shape?  If every musical transient is accompanied by sinc- function ringing,  then all you would need to do to show this would be to show the time plot of *any* musical transient. If not then “the inference [is not] to be drawn”.

 

[Fokus]

35 minutes ago, adamdea said:

Conversely the point which your article seems to skate over is that if there is ringing in the 16/44 data then a slow/lazy/non  filter in the dac will just pass the (data) ringing and there is no (apparent) time domain advantage in such a filter. So the time domain blather about such filters is pure bull.  (edited, F)

 

It has been, since the early days of Wadia. And the entire audio press bought it.

 

35 minutes ago, adamdea said:

And who says your made up shape is a musical transient?

 

To be fair, JA's example does not have to be real. We can always invent a new musical instrument and write a score for it, to be played in an anechoic room with all noise below 10 dB SPL. Recording practices have to be prepared for any conceivable sound.

 

But that still does not make ringing a real problem ...

[Shadders]

13 hours ago, mansr said:

You can create it yourself in Octave/Matlab using "kaiser(49, 35)" and pad with zeros as you see fit. The sample rate doesn't matter. Whatever you decide on, the pulse contains frequencies up to slightly less than fs/4.

Hi,

Can you detail the method you used to create the pulse as per Fig 2.

 

If you use the filter function to generate the a pulse - use the impulse response, then there are negative values - and your Fig 2 has no negative values. Using a pulse in the time domain to pass through the filter, then there is the ringing for the kaiser window you specified. Thanks.

 

Regards,

Shadders.

[mansr]

1 hour ago, adamdea said:

That part of your article is not clearly illustrated or argued. You do not clearly distinguish between ringing in the 16/44 data output of the src and that imposed by the dac (other than by passing assertion).

Where did he use an SRC? My impression is that he played his arbitrary pulse through a DAC at 384 kHz and recorded this with a few running ADCs at 96 kHz. Since the pulse, by JA's own admission, extends well above 48 kHz, the low-pass filters in the ADCs obviously show some "ringing."

 

1 hour ago, adamdea said:

I seem to remember that linear phase filters are idempotent? I thought that meant that a second linear phase filter does not need to impose any ringing.

I don't know where you'd remember that from, because it's not true. Here's a simple low-pass impulse response (blue) and after being applied to itself 10 times (red):

 

1 hour ago, adamdea said:

As fokus points out, the sampling theorem requires the signal to have no energy *at* Nyquist. If the signal has no energy at nyquist and the dac’s filter reaches stopband at nyquist, why will the dac’s filter ring? 

A filter "rings" at it's cut-off frequency. Nyquist has nothing to do with it. Perhaps this confusion stems from the fact that many SRCs use decimation filters with the cut-off precisely at the new Nyquist frequency (half-band filters).

 

1 hour ago, adamdea said:

Conversely the point which your article seems to skate over is that if there is ringing in the 16/44 data then a slow/lazy/non  filter in the dac will just pass the ringing and there is no time domain advantage in such a filter. So the time domain blather about such filters is pure bull. 

 

And who says your made up shape is a musical transient? Where is an example of a genuine musical transient with that shape?  If every musical transient is accompanied by sinc- function ringing,  then all you would need to do to show this would be to show the time plot of *any* musical transient. If not then “the inference [is not] to be drawn”.

Or he could have recorded an actual musical instrument at 192 kHz to produce the test signal.

[psjug]

16 hours ago, John_Atkinson said:

It seems incontrovertible, therefore that that ringing will excite the playback DAC's reconstruction filter, which will impose its own ringing on musical transients.

 

 

It would be incontrovertible if you show some hi-res captures of 16/44 DAC output showing the musical transients with all the ringing.

[mansr]

1 hour ago, Shadders said:

Can you detail the method you used to create the pulse as per Fig 2.

I thought I had. Anyway, paste these lines into Octave:

x = kaiser(49, 35);
x = [zeros(76, 1); x; zeros(76, 1)];

Now x contains the pulse with some zero-padding on either side.

[Miska]

12 hours ago, mansr said:

They typically use whatever filter their ADC provides. Here's the step (5 ns rise time) response of a TI PCM4220:

 

Quite frequently these days recordings are made for example at 96/24 and then converted to 44.1/16 master later on. Of course in such case the tool used to create the RedBook master will be the dominating one...

 

 

[John_Atkinson]

3 hours ago, Fokus said:

You used Bias Peak for the sample rate conversion, i.e. anti-alias filtering followed with downsampling. If you did spectral analysis on the result you could only have done this after the downsampling, which is wrong.

 

As explained earlier in this thread, I resampled the 44.1kHz file to 96kHz, in order to examine the content above 22.05kHz in the digital domain and compare that result with the content of the file before downsampling.

 

John Atkinson

Editor, Stereophile

[John_Atkinson]

23 minutes ago, psjug said:

It would be incontrovertible if you show some hi-res captures of 16/44 DAC output showing the musical transients with all the ringing.

 

This was shown in the article.

 

John Atkinson

Editor, Stereophile

 

[mansr]

1 minute ago, Miska said:

Quite frequently these days recordings are made for example at 96/24 and then converted to 44.1/16 master later on. Of course in such case the tool used to create the RedBook master will be the dominating one...

Right, and there are choices there for those who care about these things. Judging by what is being release, very few producers do.

[Fokus]

3 minutes ago, John_Atkinson said:

I resampled the 44.1kHz file to 96kHz, in order to examine the content above 22.05kHz in the digital domain

 

An utterly pointless exercise. Of course there is no content above 22.05k, provided the upsampler did its job well.

 

Do you have any idea what you are doing?

 

[psjug]

3 minutes ago, John_Atkinson said:

 

This was shown in the article.

 

John Atkinson

Editor, Stereophile

 

I just looked at all the figures - all I see are impulse response plots.  By "musical transients" I mean recordings of music.

[Fokus]

33 minutes ago, mansr said:

Where did he use an SRC?

 

Page 2 of the article:

 

"To answer this question, I took the 96kHz pulse captured by the Ayre QA-9 in Listen mode and used the highest-quality sample-rate converter in BIAS Peak to downsample it to 44.1kHz. "

 

...

 

"I therefore repeated the test using the Brooklyn's FR (Fast Rolloff) filter. This will attenuate the data's ringing at 22.05kHz, but as you can see from fig.15, it appears to have substituted its own acausal ringing. (Note that the digital data has already been band-limited to half the 44.1kHz sample rate, thus is a "legal" signal.) "

 

The issue is with JA's assertion that the digital stimulus is a 'legal' signal. It is not, since at the root it was generated with the (extremely likely half-band) AA filter of Bias Peak.

 

[mansr]

10 minutes ago, John_Atkinson said:

As explained earlier in this thread, I resampled the 44.1kHz file to 96kHz, in order to examine the content above 22.05kHz in the digital domain and compare that result with the content of the file before downsampling.

That makes no sense. A 44.1 kHz file has, by definition, no content above 22.05 kHz. If you resample it to a higher rate and find such content, what you are seeing is imaging artefacts from the sample rate converter.

[mansr]

16 minutes ago, John_Atkinson said:

This was shown in the article.

Maybe if you keep repeating that it will eventually become true.

[John_Atkinson]

4 minutes ago, mansr said:

That makes no sense. A 44.1 kHz file has, by definition, no content above 22.05 kHz.

 

Please read the thread. I was responding to the assertion made by several posters, that the ringing of the DAC's reconstruction filter was due to the down-sampled file having spectral content above 22.05kHz. It didn't.

 

John Atkinson

Editor, Stereophile

[mansr]

3 minutes ago, John_Atkinson said:

Please read the thread. I was responding to the assertion made by several posters, that the ringing of the DAC's reconstruction filter was due to the down-sampled file having spectral content above 22.05kHz. It didn't.

Where did anyone say that? Even if they did, your response still doesn't make sense.

[Fokus]

Not 'above', but 'at'.

[miguelito]

23 hours ago, crenca said:

 

The crossfeed built into Roon (what I use most of the time - have various other plugins for JRiver) implements this "overlay on the time domain" but only in the lower frequencies (user configurable between 300-2000 hz), so no impact in frequencies above this "crossover"...unless I am mistaken which could very well be the case:

 

http://bs2b.sourceforge.net/

I would have thought the opposite - the high frequencies are the ones that cross-over to the other channel. My recollection of the issue of doing this in the analog domain was an undesired effect of lowering the high frequency response due to cancellations.

[crenca]

38 minutes ago, miguelito said:

I would have thought the opposite - the high frequencies are the ones that cross-over to the other channel. My recollection of the issue of doing this in the analog domain was an undesired effect of lowering the high frequency response due to cancellations.

 

Just a quick Google:

 

http://www.meier-audio.homepage.t-online.de/crossfeed.htm

 

A round about way of saying that it is the "lower" frequencies that crossfeed works on, so to speak, which avoids the comb effect high frequency cancellation.  The "fatigue" of stereo through HP's is relieved by the crossfeed in the important 100-1000hz range where you have most instrument/voice energy and the unnatural L-R panning (caused by putting speakers that are supposed to be triangulated in front of you right next to year ears) is relieved...