MQA is Vaporware

[manisandher]

I went to Bob Stuart's 30 minute talk at the Munich Show this afternoon. Nothing much that hadn't already been said, except perhaps his insistence that the temporal domain is much more important than the frequency domain (something many of us have believed for years). He quantified it as follows (in his slides):

 

- "time is 5-13x more important than frequency"

- "human discriminate acuity 5 - 8µs"

 

He then equated temporal response with distance travelled through air, showing that both sinc 48kHz and linear phase 192kHz fall far short of that required. MQA meanwhile was shown to be more than good enough.

 

[Apologies if this has all been posted before.]

 

Edit: I took some snaps of his slides, but am reluctant to share for copyright purposes. Found this in a published article though:

 

 

Mani.

[manisandher]

1 hour ago, AJ Soundfield said:

No. Peter just needs to provide that one bit of evidence of audible but not measurable.

 

I'm not Peter, but I do use his software player and DAC. All users of Peter's XXHighEnd are familiar with how its various settings can change the sound, and yet keep the output totally bit-perfect. I think these effects, though audible to virtually everyone who's tried, may actually be difficult to measure (if not impossible).

 

As a bit of background... XXHighEnd is a true memory player, and has the ability to transfer all tracks from storage (a network-attached music server in my case) into the audio PC's RAM, shut down the audio PC's LAN connection and then commence playback. There is a setting in XXHighEnd called SFS (split file size). My layman's understanding is that this determines how large the file is when XXHighEnd accesses it from RAM during playback. Low SFS settings sound different to high SFS settings . All XXHighEnd users hear this difference in a totally consistent way - lower SFS sounding brighter than higher SFS settings.

 

Earlier this year, I took some captures with SFS=1 and SFS=120. I took the captures in two different ways: a) digital, using an USB-to-spdif converter, and straight into the Tascam's spdif input, and b) analogue, feeding the output of my DAC into a Prism ADC, and then into the Tascam's spdif input. Here are my findings:

 

a) the SFS=1 digital capture sounds identical to the SFS=120 digital capture

b) the SFS=1 analogue capture sounds different to the SFS=120 analogue capture

 

But hey, don't take my word for it, here are the captures:

 

1. SFS=1 digital capture: https://drive.google.com/open?id=0B0PU5LO5jVjfOE5PZ0RxTVl3M0U

2. SFS=120 digital capture: https://drive.google.com/open?id=0B0PU5LO5jVjfWXIyaWVaRkt1Znc

3. SFS=1 analogue capture: https://drive.google.com/open?id=0B0PU5LO5jVjfa05DdTBnZmI1aVU

4. SFS=120 analogue capture: https://drive.google.com/open?id=0B0PU5LO5jVjfdGhDcXFhdlFOekk

 

Mani.

[manisandher]

On 7/6/2017 at 1:04 PM, AJ Soundfield said:

"Claims" about subjective fantasy are non confirmable. Measurements are.

 

22 hours ago, AJ Soundfield said:

If you can hear it in soundwaves reality, then its measurable. Soundwaves are measurable.

 

19 hours ago, AJ Soundfield said:

You tell me what part of audible soundwaves are unmeasurable..and how you know this

 

19 hours ago, AJ Soundfield said:

Right, you have zero evidence of audible but unmeasurable.

 

You've been banging on about evidence, so I gave you some. All you have to do now is take the time to consider it now.

 

Analyze the files in any way you like. 1 and 2 should be totally identical (digital captures). I suspect 3 and 4 will be close enough to each other to be undifferentiable through any measurement you care to take.

 

But take a listen. If your experience is like mine, 1 and 2 will indeed sound identical. However, 3 and 4 will sound subtly different. To my ears, 3 sounds closer to 1 and 2 than does 4. On 4, the piano sounds ever-so-slightly deadened, with not quite as much transient attack, and slightly less resonance. And there are other differences. Small for sure, but audible.

 

Mani.

[manisandher]

6 minutes ago, AJ Soundfield said:

I think if you wave your hands any harder, they may fall off

 

Why don't you put your money where your mouth is and show that files 3 and 4 do indeed measure differently (because they sound different)? Or are you all mouth?

 

Mani.

[manisandher]

Ah OK. Another one all mouth... just as I suspected.

 

Mani.

[manisandher]

500 posts in a month. Yep, nothing but a big mouth.

 

Mani.

[manisandher]

50 minutes ago, testikoff said:

BTW, the first 2 files provided are not even close to being digitally identical, thus either the output of the player software or the capture process is not bit-perfect...

 

I don't have any sophisticated analysis tools. In Audacity it's clear that the files need to be aligned before comparing. Here are two reasonably well-aligned random segments:

File 1:

File 2:

They look pretty much the same to me, though I accept that this is probably a very coarse way to do this.

 

Could you show at what level the two files are not identical please? Thanks.

 

Mani.

[manisandher]

And here's the same segment from files 3 and 4 (DA/AD loop):

 

File 3:

File 4:

Edit: Actually, in these two files, I think small differences are indeed visible... but due no doubt to slightly misaligned clocks.

 

Mani.

 

[manisandher]

Ah yes, thanks for pointing this out. I remember now that elsdude found something similar a while back. The issue is some sort of compatibility issue with the USB-to-spdif converter and the software player. IIRC, periodically (every 6 seconds, according to your analysis), an extra sample gets inserted - heard as a faint but audible click during playback. So it's only valid to compare the files between these clicks.

 

But you've done that and still found a difference at ~90dBFS. I have no idea what could be causing that. Whatever it is, it doesn't seem to affect the sound, because files 1 and 2 sound identical to me.

 

Could you quickly find the difference in dB between files 3 and 4 (taking into account the clicks) please? If this is substantially higher than at ~90dBFS, that'd give a clue as to why I'm hearing what I'm hearing between them.

 

Mani

[manisandher]

2 minutes ago, mitchco said:

Just saying it is not easy.

 

Yep, understood. It'd still be interesting to determine whether files 3 and 4 are more different than files 1 and 2, extra inserted samples taken into account. I'm hearing something...

 

Mani.

[manisandher]

1 minute ago, mitchco said:

PS. as far as what is audible or not at normal listening levels, may want to have a look (and listen) at:

 

Will do... Thanks!

 

Mani.

[manisandher]

2 minutes ago, rickca said:

I feel exactly the same way.  This thread is being co-opted.

 

You're right. Apologies.

 

I'll take the analysis of my files over to the following thread:

 

Mani.

[manisandher]

Nice post @Fokus.

 

I don't agree with your final conclusion (not yet at least), but the following struck me as particularly pertinent:

 

23 minutes ago, Fokus said:

The MQA process works. Yes, it aliases. But they never claimed it did not.

 

And...

 

23 minutes ago, Fokus said:

(and even when the science was valid, there would be mastering-side only solutions)

 

So, use an MQA decoder, but don't use an MQA renderer - that's what I'm doing. And I have to say that in virtually every case I can make a comparison (not always from the same master, unfortunately), I prefer the sound of the MQA version.

 

Mani.

[manisandher]

36 minutes ago, mansr said:

This "10 m of air" talk makes no sense.

 

I am in no way saying that I agree with the content of the following, but here are some pics I took of Stuart's talk at Munich High End last year:

 

 

(This was a public talk, so I'm assuming it's OK posting these here. But Chris, if there's any issue with this then please remove.)

 

Mani.

[manisandher]

1 hour ago, FredericV said:

The best part is that those researching MQA and it's filters, came to an even better filter config than what MQA is offering:

http://archimago.blogspot.be/2018/01/musings-more-fun-with-digital-filters.html

The intermediate phase filter as proposed by Archimago does not suffer from the artefacts which MQA suffers from, including severe aliasing and phase errors.

 

 

Maybe I'm really dense, but I'm find this confusing. Aliasing occurs during analogue-to-digital conversion, and not during digital-to-analogue conversion, and yet he talks about using an anti-alias filter in a DAC. Are the terms 'aliasing' and 'imaging' really so interchangeable?

 

Happy to be put right.

 

Mani.

[manisandher]

47 minutes ago, Fokus said:

... and the correct pronounciation of finite/infinite.

 

 

FinIte and infinUt? Shame he didn't drill 'pronUnciation' into you too ;-)

 

Mani.

[manisandher]

1 hour ago, Shadders said:

When the image frequencies overlap the baseband frequencies - it is called aliasing.

 

 

From Lavry's whitepaper, sampling 25 and 35Khz tones at 48KHz:

 

 

Aliasing... with no "image frequencies overlapping the baseband frequencies" at all.

 

Mani.

[manisandher]

2 minutes ago, Shadders said:

Discussion was on imaging, not your specific example.

 

Really?

 

1 hour ago, Shadders said:

When the image frequencies overlap the baseband frequencies - it is called aliasing.

 

[Highlights mine.]

 

Mani.

[manisandher]

3 minutes ago, Shadders said:

Yes - really...

 

If you say so.

 

Mani.

[manisandher]

Just for my understanding...

 

Are those of you who are critical of the filters MQA uses critical of:

 

a) the anti-aliasing filter used in the MQA encoder?

b) the anti-imaging filter used in an MQA renderer?

c) both?

 

If you could substantiate your answer (edit: or point me to links), that'd be great.

 

Mani.

[manisandher]

15 minutes ago, testikoff said:

MQA must be using the same filter in both downsampling (preceding folding) & upsampling (rendering) stages:

 

Thanks. I can see exactly what you're explaining.

 

The use of such filters must have been a deliberate choice made by the MQA guys. Is there any evidence that using 'leaky' AA and AI filters is offset by gains in the temporal domain?

 

Mani.

[manisandher]

6 minutes ago, mansr said:

Define "gains in the temporal domain."

 

I left it ambiguous purposely, to keep the scope open.

 

6 minutes ago, mansr said:

Bear in mind that the ear is fundamentally a frequency detector.

 

OK, you probably won't like this. From Stuart's Munich talk:

 

"Time is 5-13x times more important than frequency. Human discriminate acuity 5–8 µs."

 

[No, I don't know what "human discriminate acuity" is either ;-) ]

 

Mani.

[manisandher]

6 minutes ago, mansr said:

A badly tuned piano sounds dreadful no matter how perfectly it is played.

 

Yes, but surely better than a perfectly tuned piano played badly.

 

Mani

[manisandher]

7 minutes ago, mansr said:

I'd rather hear a competent amateur player on a well tuned piano than a virtuoso on an ill tuned one.

 

And what if only the top 5 keys were ill-tuned, the rest perfect?

 

Mani.

[manisandher]

39 minutes ago, Andyman said:

Now you're being silly...!

 

What, I've gotten away with it until now? Great! ;-)

 

Mani.