MQA is Vaporware

[mcgillroy]

5 hours ago, John_Atkinson said:

 

This is the first time Art Dudley has ever commented on MQA. Don't see, therefore, how he can be called "the great MQA [chearleader."]

 

And take a look at https://www.stereophile.com/content/mqa-aliasing-b-splines-centers-gravity

There's an interesting listening test embedded in the text.

 

John Atkinson

Editor, Stereophile

 

John seriously: is this article and interview a joke? Four lines by Bob and an amateurish signal-to-noise-test...

 

Suddenly s/n of ca 40db is ok - in a mag that regularly test equipment with 130db s/n?!

 

If Bob uses sparse sampling to recover and approximate spectral content that is compressed in lossy fashion - the 24-48k component - that would be novel. If it makes sense and is audible is another question. But I doubt he is.  MQA is plain PCM until proven otherwise.

[Audiophile Neuroscience]

7 hours ago, adamdea said:

Can you or anyone help with this sentence

“Relaxing that constraint restores the symmetry between the time and frequency domains that was missing from Shannon's theory.”
I’m baffled. It makes zero sense to me.

 

aha , I thought it was just me.

 

I am not an audio/computer professional but at least can get my head around the "folding" of bits into a more compact package requiring less bandwidth. I think I get the master "authenticated" bit, as approved by the artist, albeit a bit misleading IMHO. However the whole "time domain" thing totally eludes me. I just can't find anything in plain English that makes sense. That may well be my failing or perhaps it's a proprietary secret?

[FredericV]

3 hours ago, mcgillroy said:

If Bob uses sparse sampling to recover and approximate spectral content that is compressed in lossy fashion - the 24-48k component - that would be novel. If it makes sense and is audible is another question. But I doubt he is.  MQA is plain PCM until proven otherwise.

 

Most MQA originates from a PCM master which is just created as regular PCM, which does not violate the theorem. The source used for the encoder does not violate Nyquist–Shannon. The source was already correctly band limited to begin with.

But then MQA comes along, and their leaky filters happen at the resampling steps, e.g. first unfold with and without access the 8 LSB bits of a 24 bit MQA container:

 

So they use minimum phase upsampling with one cycle of postringing to generate fake aliased content, not in the original, but which may look partially like the original, and then use a lossy difference signal hidden in the 8 LSB bits of the distribution files. In the end they still mess up the frequency domain after unfolding, so the process is lossy. The A in MQA stands for approximation instead of authentication. We already proved in the above article the Authentication is fake.

Now let's look at this:
http://archimago.blogspot.be/2018/01/musings-more-fun-with-digital-filters.html

Why did an independent researcher like @Archimago found intermediate phase, in such way that it does not mess with the frequency domain (no phase shift within audible spectrum) and also gets the time domain right, while a "big name" such as Bob Stuart could not figure it out?

After long testing with a lot of customers, the intermediate phase filters as proposed by Archimago sound the best to many ears. We gave them away as a free software update in our own server.

Do these intermediate phase filters get it right? We believe so.

Testing a small bell in a live room, and on speakers which get the time domain right (not many, but we have access to those speakers - John Watkinson Legends - they create a virtual point source from which the sound originates), the bell as rung by my Munich demo colleague Kommer Kleijn sounds exactly like recording of it:



Most speakers get the periodic sounds right (the resonance of the bell), but mess up events, like the clapper hitting the bell, therefore failing in the time domain. Evidence is the impulse response of most tweeters, which once exposed to such event, take several cycles before the tweeter is stopped. The bending wave driver as used by John Watkinson does not suffer from these issues. John Watkinson is the guy who wrote the bible on digital audio:

https://www.amazon.com/Art-Digital-Audio-John-Watkinson/dp/0240515870

We have a similar demo with a snare drum. I witnessed a demo in the home of Kommer Kleijn (the Brussels based co-designer of the John Watkinson Legends) where he was playing a snare drum, and then a recording of it, using regular PCM and a Antelope Audio studio DAC. It sounds like the real thing, as if the drum is in front of you. We could not take the drum set to Munich, maybe next year we'll take it to the Marriot.

This proves you can get the time domain and transients right without MQA. Post-shannon is not needed either.

The whole time domain claim of MQA is because of the need to down/upsample content as part of folding/unfolding to fit into a smaller file. They chose a leaky filter to get the time domain right at the cost of the frequency domain. So they have a solution to a non-problem, as folding is not really required in today's bandwidth.

Furthermore, better filters exist which get both time and frequency domain right.

[adamdea]

1 hour ago, Audiophile Neuroscience said:

 

aha , I thought it was just me.

 

I am not an audio/computer professional but at least can get my head around the "folding" of bits into a more compact package requiring less bandwidth. I think I get the master "authenticated" bit, as approved by the artist, albeit a bit misleading IMHO. However the whole "time domain" thing totally eludes me. I just can't find anything in plain English that makes sense. That may well be my failing or perhaps it's a proprietary secret?

The odd thing about Jim Austin's sentence is that the frequency domain and the time domain are mathematically interchangeable. The Shannon's proof of the sampling theorem depends on this. That relationship is immutable. So how can they not be symmetrical and how can anything restore what can;t be lost?

 

And while we are at it this of course is the problem with saying that one can be wrong in the frequency domain but right in the time domain. 

So obviously (and now eventually BS pretty much concedes this) MQA can only be targetting one sort of time domain characteristc at the expense of another. And how is this set of targets justified? :

 " For a number of reasons based on the auditory science of object detection, it seems very plausible that the first moment is of prime importance to the ear and that higher moments are less important and (importantly) can be shown not to contribute errors such as jitter."

Wow - "it seems very plausible that", well that has me convinced. MQA targets one set of time domain targets at the expense of another because "it seems very plausible" to BS that they are what matters.

But frankly the bit where he exp;lains what he means by time smear is it even more pathetically weak

Stuart: Any deviations that aliasing brings to the "impulse response" (when analog is being uniformly sampled) are quite different from the impact of the filters controlling (and contributing to) end-to-end system response. The latter is there whether or not filtering is adequate to control or eliminate aliasing. Time smear relates to the fact that the "filter" spreads every sample out in time, irrespective of frequency—particularly in the "real world," where we take into account quantization (and sometimes aliasing) effects in A/D, workstations, and DACs. "*

So now we know what it is- its the possible maths errors in calculating the impact of each tap of the filter- - but wait- if this is signifcant this will show up in eavery single test signal you ever put through the dac. And how many orders of magnitude lower than the MQA alising are they? If 50 db below the signal is ok for the aliaising then whats the problem with 24 bit quantisation?

I would be happy to be corrected by the really smart people here, but as far as I can tell BS might as well be saying- look I've been talking crap from the word go, but you my friend had better keep eating it because you'll look silly if you stop.

 

* And he finsihes: This smear, we believe, can be material for the human listener who is extracting multiple cross correlations, as well as envelope and nonlinear measures of the audio. "

Ah "we believe"

[Audiophile Neuroscience]

20 minutes ago, adamdea said:

The odd thing about Jim Austin's sentence is that the frequency domain and the time domain are mathematically interchangeable. The Shannon's proof of the sampling theorem depends on this. That relationship is immutable. So how can they not be symmetrical and how can anything restore what can;t be lost?

 

And while we are at it this of course is the problem with saying that one can be wrong in the frequency domain but right in the time domain. 

So obviously (and now eventually BS pretty much concedes this) MQA can only be targetting one sort of time domain characteristc at the expense of another. And how is this set of targets justified? :

 " For a number of reasons based on the auditory science of object detection, it seems very plausible that the first moment is of prime importance to the ear and that higher moments are less important and (importantly) can be shown not to contribute errors such as jitter."

Wow - "it seems very plausible that", well that has me convinced. MQA targets one set of time domain targets at the expense of another because "it seems very plausible" to BS that they are what matters.

But frankly the bit where he exp;lains what he means by time smear is it even more pathetically weak

Stuart: Any deviations that aliasing brings to the "impulse response" (when analog is being uniformly sampled) are quite different from the impact of the filters controlling (and contributing to) end-to-end system response. The latter is there whether or not filtering is adequate to control or eliminate aliasing. Time smear relates to the fact that the "filter" spreads every sample out in time, irrespective of frequency—particularly in the "real world," where we take into account quantization (and sometimes aliasing) effects in A/D, workstations, and DACs. "

So now we know what it is- its the possible maths errors in calculating the impact of each tap of the filter- - but wait- if this is signifcant this will show up in eavery single test signal you ever put through the dac. And how many orders of magnitude lower than the MQA alising are they? If 50 db below the signal is ok for the aliaising then whats the problem with 24 bit quantisation?

I would be happy to be corrected by the really smart people here, but as far as I can tell BS might as well be saying- look I've been talking crap from the word go, but you my friend had better keep eating it because you'll look silly if you stop.

 

Thanks Adam. I can't say I am any more enlightened but I can say that is not your fault

[mansr]

9 hours ago, adamdea said:

Can you or anyone help with this sentence

“Relaxing that constraint restores the symmetry between the time and frequency domains that was missing from Shannon's theory.”
I’m baffled. It makes zero sense to me.

It makes no sense me either. The whole article consists of nothing but hand-waving and baseless speculation (regarding how we perceive sound).

[mansr]

9 hours ago, ChrisG said:

Well, in fairness to Roon, they did require that users would still be able to use DSP after the first unfold of the MQA file. AFAIK, they are the only one doing that.

Audirvana and Bluesound (only tone controls) both do. Believe me, it really is quite trivial to do.

[adamdea]

1 hour ago, Audiophile Neuroscience said:

 

Thanks Adam. I can't say I am any more enlightened but I can say that is not your fault

It was as a result of exactly this sort of puzzlement (about pretty much this point) that I ended up embarking on a sort of personal night school about information theory, and electrical engineering, culminating in the purchase of Morrison on Fourier Analysis. At the end of it, apart from mildly surprising some scientist friends at dinner parties, all I achieved was a firm conviction that one should be profoundly suspicious whenever one hears the phrase "in the time domain" in relation to audio. That and the loss of <I shudder to think how much> time I could have been making money, or doing something useful like watching porn.  

Maths does have a certain beauty though.

 

[edit] the reason for this rambling story was basically to say that the reason I bore on about this stuff is that I'm hoping to spare others the wild goose chase I went on. It's like going on a pilgrimage to lourdes to find out why stork tastes better than butter.

[Audiophile Neuroscience]

3 minutes ago, adamdea said:

It was as a result of exactly this sort of puzzlement (about pretty much this point) that I ended up embarking on a sort of personal night school about information theory, and electrical engineering, culminating in the purchase of Morrison on Fourier Analysis. At the end of it, apart from mildly surprising some scientist friends at dinner parties, all I achieved was a firm conviction that one should be profoundly suspicious whenever one hears the phrase "in the time domain" in relation to audio. That and the loss of <I shudder to think how much> time I could have been making money, or doing something useful like watching porn.  

Maths does have a certain beauty though.

 

hahaha. Adam I have developed a whole new respect for you ! I can't wipe the grin off my face -porn would have been the better option!

[adamdea]

12 minutes ago, Audiophile Neuroscience said:

 

hahaha. Adam I have developed a whole new respect for you ! I can't wipe the grin off my face -porn would have been the better option!

Tout comprendre c'est tout pardonner, I hope.

Mind you, it's perhaps dangerous to use the word "wipe" in the same sentence as "porn".

 

 

[miguelito]

58 minutes ago, mansr said:

Audirvana and Bluesound (only tone controls) both do. Believe me, it really is quite trivial to do.

I did not think the rendering information gets “reattached” after DSP is performed in Audirvana. This is explicitly done in Roon (of course, I could check myself as I have a Dragonfly).

 

Not that this rendering stage is all that meaningful in my opinion or that it is legit to render after applying DSP. If rendering were so critical, one should really do it before DSP to get a stream as close to what is espoused as “close to the studio” as possible and then apply DSP.

 

This “reattachment” of MQA bits is MQA Ltd’s way to allow DSP while keeping the ability to strongarm people into paying MQA licenses for rendering. Quite the bastardization of the goal if you ask me.

[The Computer Audiophile]

4 minutes ago, miguelito said:

I did not think the rendering information gets “reattached” after DSP is performed in Audirvana. This is explicitly done in Roon (of course, I could check myself as I have a Dragonfly).

 

Not that this rendering stage is all that meaningful in my opinion or that it is legit to render after applying DSP. If rendering were so critical, one should really do it before DSP to get a stream as close to what is espoused as “close to the studio” as possible and then apply DSP.

 

This “reattachment” of MQA bits is MQA Ltd’s way to allow DSP while keeping the ability to strongarm people into paying MQA licenses for rendering. Quite the bastadization of the goal if you ask me.

DSP just as the artist intended  :~)

[miguelito]

5 minutes ago, The Computer Audiophile said:

DSP just as the artist intended  :~)

???

 

Seriously... And people lets stop trying to make sense of the random walk terminology soup that Bob Stuart’s explanations tend to drift to...

 

[mansr]

24 minutes ago, miguelito said:

I did not think the rendering information gets “reattached” after DSP is performed in Audirvana.

You may be right about that.

 

24 minutes ago, miguelito said:

This is explicitly done in Roon (of course, I could check myself as I have a Dragonfly).

And it's a trivial operation. You merely copy the LSB of the input to the output. If that's too difficult for you, MQA even provides a handy library that does it for you.

 

24 minutes ago, miguelito said:

Not that this rendering stage is all that meaningful in my opinion or that it is legit to render after applying DSP. If rendering were so critical, one should really do it before DSP to get a stream as close to what is espoused as “close to the studio” as possible and then apply DSP.

Agreed.

 

24 minutes ago, miguelito said:

This “reattachment” of MQA bits is MQA Ltd’s way to allow DSP while keeping the ability to strongarm people into paying MQA licenses for rendering. Quite the bastadization of the goal if you ask me.

Not at all. I'm assuming, of course, that the goal is to make money for Bob Stuart.

[miguelito]

2 hours ago, mansr said:

Not at all. I'm assuming, of course, that the goal is to make money for Bob Stuart.

That's obviously the key goal. "As Bob Stuart intended."

[miguelito]

And for the record, I am always very happy to pay for people's work and their IP.

 

However, not all work is legit. It is not ok to add DRM, some eq, and some beautification to music tracks, call that a radical advancement, and expect everyone to bow and pay for it. It's gibberish.

 

Some MQA albums sound very good thanks to judicious filtering and careful mastering, and I am very happy to pay for that.  I just purchased the Fairytales album because I thought it was an interesting piece of work (I don't care much about the music itself, not quite my cup of tea).

 

But the bulk of it is garbage in, garbage out - and I am supposed to be ok paying that part of MQA which is a complete scam? That I am not happy about.

[crenca]

5 hours ago, adamdea said:

 

But frankly the bit where he exp;lains what he means by time smear is it even more pathetically weak

Stuart: Any deviations that aliasing brings to the "impulse response" (when analog is being uniformly sampled) are quite different from the impact of the filters controlling (and contributing to) end-to-end system response. The latter is there whether or not filtering is adequate to control or eliminate aliasing. Time smear relates to the fact that the "filter" spreads every sample out in time, irrespective of frequency—particularly in the "real world," where we take into account quantization (and sometimes aliasing) effects in A/D, workstations, and DACs. "*

So now we know what it is- its the possible maths errors in calculating the impact of each tap of the filter- - but wait- if this is signifcant this will show up in eavery single test signal you ever put through the dac. And how many orders of magnitude lower than the MQA alising are they? If 50 db below the signal is ok for the aliaising then whats the problem with 24 bit quantisation?

I would be happy to be corrected by the really smart people here, but as far as I can tell BS might as well be saying- look I've been talking crap from the word go, but you my friend had better keep eating it because you'll look silly if you stop.

 

 

adamdea (or anyone else),

 

Is not the key word here "spread"?  He does not appear to be talking about the low level distortion (aka "ringing") that is a result of quantization effects of filters, he seems to me (and I could be totally off) to be saying that the quantization error "spreads" the sinusoidal response of the entire waveform in a "real world" (i.e. linear distortion), band limited signal.  Is this not exactly what Shannon denies, and is this not fundamental challenge to the theoretical basis of signal processing as we know no it?  In other words Shannon "proves" to us that we get a waveform that is accurate in both time and frequency domains, although at the expense of certain kinds of non-linear distortion, noise flores, etc. (that leads to a variety of practical design decisions in the 'real world') Alot rests on the word "spread" here of course.

 

Assuming I a wrong about what is being asserted in the previous paragraph, then we are back to the known Shannon trade offs between which form of distortion does one believe in, IM alias vs.  low level "ringing", etc., whether and how it is all audible and what impact it has, etc.

 

 

[mansr]

9 minutes ago, crenca said:

Is not the key word here "spread"?  He does not appear to be talking about the low level distortion (aka "ringing") that is a result of quantization effects of filters, he seems to me (and I could be totally off) to be saying that the quantization error "spreads" the sinusoidal response of the entire waveform in a "real world" (i.e. linear distortion), band limited signal.

Trivial quantisation (rounding or truncating) causes harmonic distortion. Adding dither converts the distortion products to low-level noise across the entire spectrum. This noise may be flat or shaped. Either way, it is mostly decorrelated from the signal. With 24-bit sampling, all effects of quantisation, dithered or not, are far below the minimum level reproducible by any DAC and thus entirely without relevance.

 

9 minutes ago, crenca said:

Is this not exactly what Shannon denies, and is this not fundamental challenge to the theoretical basis of signal processing as we know no it?

The sampling theorem assumes infinite sample precision. When the samples are quantised, the signal is distorted. This is not news. In fact, the effects of quantisation and dither have been extensively studied for decades.

 

I do find it curious that BS suddenly starts bringing up quantisation. He never mentioned that before. It's almost as if he's given up on the original angle and is trying a new excuse in the hopes that we'll fall for this one.

[Fokus]

Sure looks like that.

 

 

[crenca]

19 minutes ago, mansr said:

 

The sampling theorem assumes infinite sample precision. When the samples are quantised, the signal is distorted. This is not news. In fact, the effects of quantisation and dither have been extensively studied for decades.

 

I do find it curious that BS suddenly starts bringing up quantisation. He never mentioned that before. It's almost as if he's given up on the original angle and is trying a new excuse in the hopes that we'll fall for this one.

 

Yes, but Bob S appears to be - I'm just trying to make some sense of this, in vain probably - pointing to the how of the distortion.  In other words, he appears to be making a claim of linear distortion (aka, "time domain") and this "fixed" by something not really specified in his MQA magic.  What seems new is his assertion that Shannon is not accurate in the time domain at all - that instead of (in addition to) low level "ringing", the entire sinusoidal signal response is non-linearly distorted as opposed to just low level "rigning".  This seems a bold claim that contradicts Shannon in a fundamental way.  Perhaps he is interpreting (creatively) the low level rigning in this way.  Heck, perhaps even this is accepted in standard signal processing??

 

I feel sure that either I am reading too much into it, or that this is exactly what I am supposed to do (BS is BSing about his BS encoding), etc.

 

I just wanted to see if anyone else is reading it this way.

[crenca]

Just now, crenca said:

 

Yes, but Bob S appears to be - I'm just trying to make some sense of this, in vain probably - pointing to the how of the distortion.  In other words, he appears to be making a claim of linear distortion (aka, "time domain") and this "fixed" by something not really specified in his MQA magic.  What seems new is his assertion that Shannon is not accurate in the time domain at all - that instead of (in addition to) low level "ringing", the entire sinusoidal signal response is non-linearly distorted as opposed to just low level "rigning".  This seems a bold claim that contradicts Shannon in a fundamental way.  Perhaps he is interpreting (creatively) the low level rigning in this way.  Heck, perhaps even this is accepted in standard signal processing??

 

I feel sure that either I am reading too much into it, or that this is exactly what I am supposed to do (BS is BSing about his BS encoding), etc.

 

I just wanted to see if anyone else is reading it this way.

 

Oh, by the way even though the theorem assumes an infinite sample precision, does it not also "prove" that in a bandlimited application (as in audio) that the distortion is of various types (i.e. "ringing", IM, etc.) but that above this the signal IS properly re-constructed (i.e. calculated) in BOTH the time and frequency domains?  In other words, is not BS asserting that the theorem is wrong?

[Indydan]

On 16/05/2018 at 5:35 PM, mansr said:

Easy. Put a hefty upfront price tag on licensing, then offer steep discounts (all the way to zero cost) for "early" adopters. Stress that the discount can go away at any time. This will encourage vendors to jump on board just in case it becomes a must-have and they'll have to fork out the full price later. In fact, I have heard this is exactly what MQA has been doing.

 

 

[Indydan]

On 16/05/2018 at 6:05 PM, Brinkman Ship said:

Bob Stuart trying to get this thread deep sixed leads me to ask out loud some questions, not of you, but in general.

 

-Why has he not used legal action if he feels there is damaging information here that he can prove is incorrect?

 

-Why has not chosen to respond?

 

-Why has he put zero effort into trying to disprove some of the measurements and conclusions that have put MQA

in an unflattering light with hard data?

 

Partly, because it would be opening the biggest can of worms this century, and secondly, he seems to have an army apostles like Atkinson, Quint, Austin, etc who come here to cast doubt, deflect, and double down.

 

Because BS is full of BS. 

[mansr]

3 minutes ago, Indydan said:

Because BS is full of BS. 

This isn't him, but I imagine his hands looking something this:

[mansr]

14 minutes ago, crenca said:

Oh, by the way even though the theorem assumes an infinite sample precision, does it not also "prove" that in a bandlimited application (as in audio) that the distortion is of various types (i.e. "ringing", IM, etc.) but that above this the signal IS properly re-constructed (i.e. calculated) in BOTH the time and frequency domains?

Time and frequency are interchangeable via the Fourier transform. If one is correct, so is the other.

 

14 minutes ago, crenca said:

In other words, is not BS asserting that the theorem is wrong?

He's not asserting anything at all as long as he's not making sense.