MQA is Vaporware

[esldude]

From Archimago's excellent blog, with assistance from mansr, of many things that stand out here are two from my view:

 

Basically it looks like different MQA devices could behave differently!

 

Like I said not long ago, that term "authenticated" (MQA) appears completely meaningless at least when it comes to sound quality.

 

If MQA does not whatsoever achieve the primary objective, and key marketing goal so important it is part of the name, MASTER QUALITY AUTHENTICATED, then it is worse than before MQA.  All we are left with are vague promises it will sound better which also appear about as likely as the authentication.  All we needed for real authentication is for the studios to put together one approved master file, and honestly give us a check sum which could be compared to any hires download offered.  MQA fails on its premise, and is not consistent across devices, and being baked in actually makes it less likely you could playback a file with authentic accuracy.  All I see are FAIL, FAIL, FAIL, and BS.  If it turns out not to be audibly detrimental that mainly says more about how much in excess of audibility digital sound formats already are.

[jabbr]

12 hours ago, AJ Soundfield said:

Why is that interesting?

Because I upsample all my PCM before sending to DAC, there should be no SQ difference with MQA according to that graph.

[esldude]

9 minutes ago, jabbr said:

Because I upsample all my PCM before sending to DAC, there should be no SQ difference with MQA according to that graph.

According to the measurements of MQA at archimago's blog with mansr's help that will not be true.  Miska's software will be more accurate and cleaner.  MQA can only wish it were that good.  Well except for that highly informative graph there.

[jabbr]

Ah ok well I'm upsampling regardless  

 

My point is that it's really really cheap to upsample PCM in the computer - I'm using a board that's $70  

[Fokus]

9 hours ago, Fitzcaraldo215 said:

 only about the Dragonfly Black, which he says may not even have sufficient power to decode the MQA Core.

 

Not 'may'. It simply does not have the capabilities for doing a full decode. That is not a secret(*). Analysing now via a Dragonfly is interesting because it enables one to inject post-core synthetic test signals into a pure render stage. This is not possible with a Meridian Explorer 2 because it appears to contain a bug that prevents it from acting correctly on a core-decoded stream. This aside.

 

 

(* Even though some distributors, magazines, and commentators seem to have been, for a while, under the impression that this was a full MQA DAC. Look at the Beekhuyzen youtube channel for a train wreck.)

 

[Fokus]

9 hours ago, crenca said:

 

Thinking of the Chord vs MQA approach, is it fair to say that the "time domain smearing" IS the ringing, and that the lack of ringing in the MQA filters IS what they allege to be "time domain accuracy"?  I know this has already been described thus a 1000 times but just want to verify:  given this, then MQA trades aliasing for the lack of ringing - noise for ringing whereas the steep filter of the Chord Dave will have ringing but more accurate (i.e. little aliasing distortion) FR.

 

Ok, after all this I am still having a hard time tracking relevance in the audio band as I can't tell which approach as a paper tiger (a graph) is more relevant from 20-20.

...

Also, besides all this you have the question as to what exactly MQA is doing as an encoding process on the ADC side of things which is not addressed at all by your and archimago's work. 

 

Yes. The 'smear' is the 'ringing'. Ironically, from the sampling theorem and from other disciplines that rely on signal discretisation to extract information equivalent to timing accuracy in audio we know that the highest accuracy is reached with the steep filtering in place. It is the act of anti-alias filtering that allows one to convey the precise location in time of an impulse in a band-limited signal. Watts/Chord are correct in saying that their filter is timing accurate (to a degree orders beyond what is necessary for audio), whereas MQA's story is about apparent timing accuracy, purely for PR purposes.

 

As for the ADC: it is indeed almost impossible to asses what they are actually doing. There are many possibilities. For older recordings, like made on a PCM1630 or so, the MQA process could well correct its frequency domain magnitude and phase errors. This is valid, but then any mastering engineer with the knowledge could do so.

For low-rate recordings made with linear phase AA filters MQA could apply the weird all-pass filter they patented, something that imposes a large group delay onto the 18-22kHz band, thereby picking up the pre-ringing and dumping it down after the main impulse. This is rather freaky.

For high-rate recordings MQA simply subsamples with lazy/leaky AA filters. This is clearly stated in the patents and older technical documentation. Their justification is indeed that aliasing is subservient to temporal accuracy, and that with most music there is not enough ultrasonic content, and enough in-band noise, to make aliasing a non-issue.

 

[Fokus]

9 hours ago, Fitzcaraldo215 said:

I just found it difficult to understand how you got there.  I still do.

 

However, your statement does accuse Stuart and MQA of lying and fraud, because they have explicitly and consistently stated to the contrary many times.

 

Unfortunately, I do not find your version of it sufficiently well defined or documented to be able to simply understand your methodology, etc.  so as to be able to go along with you, not yet, at least.

 

It might even humble Stuart and MQA, and cause them very serious credibility issues, if not also lead to serious legal issues.

 

1) It is clear that you don't understand. Let me help you: this is entirely your problem, being either unwilling or not capable of understanding.

 

2) When the MQA ideas were published (end of 2014?) they were more or less clear. A bit later, when launching actual product, things got a bit less clear. Mostly thanks to the (as always) totally inept consumer audio press. In what followed MQA and Stuart have more and more chosen to navigate a path of (what is now clearly deliberate) obfuscation and newspeak. 'Folding' in their book has at least two different meanings. The first meaning is to take a 96k signal, split it into two bands, and fold the upper band in the data space below the lower band. This is a, give or take, non-destructive process and we all agree (and have verified)  that valid and true 96k content can be passed through a 48k channel this way, the origami trick. (As an aside, this method is not an MQA invention, there is Japanese prior art. Funny, actually.) 192k and higher signals are treated by laze subsampling and lazy upsampling. According to MQA the resultant aliasing and imaging is inconsequential. MQA's aim, their only aim, is to prevent steep low-pass filtering to happen anywhere. Their aim is not to pass on valid 192k or higher signal, only 'as-if 192k'. 'as-if' because it was never filtered down to 96k. They call this occasionally 'folding'. Incidentally, the phenomenon of aliasing and imaging is in the literature also sometimes called 'frequency folding'. This has nothing to do with the aforementioned origami folding. It is totally destructive.

 

 

3) Insufficiently documented??????? Did you actually read the body of work presented by Mans and others? Did you even try to understand?

 

4) Legal issues? This is consumer audio. Anything goes. No-one bothers. In the eyes of the real world out there audiophools are morons anyway.

 

[Fokus]

And let me add to 4)

 

Can you imagine a court case revolving around semantics in a niche hobby/consumer domain, a domain where no standard definitions exist for a great many processes and practices? Can you imagine how much fun such a case would be for a private-person plaintiff against a corporation that has spent a few years digging its IP trenches?

 

[rickca]

3 hours ago, esldude said:

Basically it looks like different MQA devices could behave differently!

Can you please clarify what you mean by behave in this context?  

 

I fully expect that different MQA devices will apply different processing (such as filters) based on the same MQA stream metadata (hints).  There's no way I can see the Berkeley Alpha DAC Reference Series 2 implementing exactly the same processing as something like an Audioquest DragonFly.  The price difference between these devices is huge.

 

Surely Berkeley is going to want their implementation to sound superior.  Why would a DAC manufacturer do something wimpy when they have the resources to do something more sophisticated?  If that's what it takes to get MQA approval, I don't think you would see high-end DACs supporting MQA.

 

Perhaps I've entirely missed your point about MQA devices behaving differently.  That's why I've asked you for clarification.

[rickca]

Seems like MQA assures you only that you are playing the MQA distribution file.   That's all there is to the authenticated claim.  There is no assurance of getting the sound that the artist or recording engineer intended.  That's just marketing hype. 

[esldude]

46 minutes ago, rickca said:

Can you please clarify what you mean by behave in this context?  

 

I fully expect that different MQA devices will apply different processing (such as filters) based on the same MQA stream metadata (hints).  There's no way I can see the Berkeley Alpha DAC Reference Series 2 implementing exactly the same processing as something like an Audioquest DragonFly.  The price difference between these devices is huge.

 

Surely Berkeley is going to want their implementation to sound superior.  Why would a DAC manufacturer do something wimpy when they have the resources to do something more sophisticated?  If that's what it takes to get MQA approval, I don't think you would see high-end DACs supporting MQA.

 

Perhaps I've entirely missed your point about MQA devices behaving differently.  That's why I've asked you for clarification.

I would not expect different behaviour.  The promise is the Master Quality Authenticated.  Meaning a promise of linking end to end.  From the approved high quality master recording all the way thru until the device you play back a recording with outputs precisely and exactly an audibly lossless version of the master itself.  When the light is lit, you are supposed to know you are listening to the master itself as if you were at the studio where it was created.

 

Now if someone made a ridiculous promise that a lit MQA on a Dragonfly does replicate a lit MQA on a far more expensive device they have claimed more than they can deliver or mislead you.  Yet this very basic premise is supposed to be the very raison d'etre of MQA itself.

 

So if as the Archimago blog shows, one MQA device selects one filter, and another selects differently and they don't all have the same capability upon playback for the same level of filtering then that very raison d'etre has obviously not been true at all.  Simple bit perfect playback of the master file if of good provenance with accepted formats blows the actual MQA implementation completely out of the water.  What good does having provenance confirmed by MQA if subsequent playback is whatever is convenient for various quality levels of playback devices do for the music lover?

[esldude]

21 minutes ago, rickca said:

Seems like MQA assures you only that you are playing the MQA distribution file.   That's all there is to the authenticated claim.  There is no assurance of getting the sound that the artist or recording engineer intended.  That's just marketing hype. 

Indeed just marketing hype.  So how this different from any other distribution file?  Well in fact it can be very inferior to existing simple formats like 96/24 PCM or in some ways even 48/24 PCM.  We don't need licensing, quality dilution of undecoded playback or any of the other inconvenient aspects of MQA for that.

 

You get the idea MQA was a sucker play?

[Fokus]

5 minutes ago, esldude said:

 one MQA device selects one filter, and another selects differently and they don't all have the same capability upon playback

 

But of course it goes this way. MQA's aim is to optimise and authenticate up until the analogue signal at the output of the replay DAC. Since DACs analogue output filters differ, this implies that the digital input to the last stage of the DAC chips differ too. It makes sense.

 

This, of course, won't keep them from making a mess of it in actuality ;-)

[Fokus]

29 minutes ago, rickca said:

There is no assurance of getting the sound that the artist or recording engineer intended.  That's just marketing hype. 

 

Any such claim will always be marketing hype.

 

Unless one listens with the exact-same monitors, in a room with the exact-same acoustic, at the exact-same level, and with the exact-same amount of illegal substances imbibed.

 

And even then one cannot copy the artist's ears (luckily!) and state of mind.

 

 

 

Fidelity to what exactly?

[rickca]

17 minutes ago, esldude said:

What good does having provenance confirmed by MQA if subsequent playback is whatever is convenient for various quality levels of playback devices do for the music lover?

I think you and I agree.  The provenance thing is really about control of distribution.  The MQA metadata may, in fact, be useful information if it helps the DAC select the best filter for a particular track ... but implementation is going to be different on different MQA devices.  The MQA marketing message puts a very generous spin on all this.  I'm being polite.

[rickca]

25 minutes ago, Fokus said:

with the exact-same amount of illegal substances imbibed

I liked that part. 

[mansr]

2 hours ago, rickca said:

I think you and I agree.  The provenance thing is really about control of distribution.  The MQA metadata may, in fact, be useful information if it helps the DAC select the best filter for a particular track ... but implementation is going to be different on different MQA devices.  The MQA marketing message puts a very generous spin on all this.  I'm being polite.

So far I have seen no evidence of different filters actually being used. The impulse responses Archimago recorded are a close match to the actual coefficients I extracted from the Bluesound firmware. Due to analogue filtering, it is impossible to get an exact capture of the digital impulse response, but what we have is as close as can be expected given the constraints.

[wgscott]

10 hours ago, The Computer Audiophile said:

 

That's a really cool idea. If someone wants to pay to play MQA, the company could sell them a plug-in. Rather than all users of software paying for the embedded license such as what's in Audirvana. 

 

This is in fact exactly why I haven't upgraded to 3.x. yet. I paid for Dirac so I can use it as an AU plug-in with Audirvana.  Seems ideal, and the plug-in standards (AU and VTS) are already in place.

[AJ Soundfield]

10 hours ago, The Computer Audiophile said:

However, without blind ABX testing

There's that crazy talk again...

 

I have access to the Chord and some other fairly high end DACs (Berkeley, Lampi, Ayre, etc), some of which can decode MQA. In all the informal sessions so far, the Chord has been spanking the MQA DACs, but as you say, it most definitely must be done controlled, for any sort of validity.

In Nov I'll be doing something a bit different. I'll be comparing the MQA output of a Mytek Brooklyn vs a 16/44 loop version of itself. That might be interesting. 

[The Computer Audiophile]

8 minutes ago, AJ Soundfield said:

There's that crazy talk again...

 

I have access to the Chord and some other fairly high end DACs (Berkeley, Lampi, Ayre, etc), some of which can decode MQA. In all the informal sessions so far, the Chord has been spanking the MQA DACs, but as you say, it most definitely must be done controlled, for any sort of validity.

In Nov I'll be doing something a bit different. I'll be comparing the MQA output of a Mytek Brooklyn vs a 16/44 loop version of itself. That might be interesting. 

 

I'm suggesting that not all measurements really matter. That's not crazy talk. 

[mansr]

11 hours ago, The Computer Audiophile said:

However, without blind ABX testing aren't we arguing over something that may not matter or even be audible?

I'm just trying to figure out what MQA actually is.

[miguelito]

Just now, The Computer Audiophile said:

I'm suggesting that not all measurements really matter. That's not crazy talk. 

Completely agree... And I would add that it is often not clear what measurement we should do to distinguish between things that sound different, but a measurement could be done.

[AJ Soundfield]

8 minutes ago, The Computer Audiophile said:

 

I'm suggesting that not all measurements really matter. That's not crazy talk. 

I thought you were suggesting ABX testing as quoted. That is crazy talk! For some at least.

 

[miguelito]

One my favorite albums is Prince's "Purple Rain". I have a few of the versions: CD, the very short lived 24/96 that showed up on HDTracks at some point, and the newly released 24/96 Painsley Park remaster expanded edition. For comparison purposes alone I captured the stream from the MQA version out of the TIDAL app using Audio Hijack. I compared the high res versions using Roon, no upsample.

 

Long story short:

1- The MQA version looks to come from the same master as the latest 24/96 in spatial content and DR (see below).

2- The MQA unfolded stream out of TIDAL came out as 24/176, and from the spectral content (see below) I have to say TIDAL is possibly upsampling the unfolded  24/88 stream... I don't know what is going on there.

3- The short-lived 24/96 version really looks like a different remaster from the original, and notice the spectral content is identical between the 24/96 versions.

4- The MQA spectral content is MARKEDLY different from the other two.

 

Bottomline: My ears prefer the MQA version, there's a physicality to Prince's voice that bests the other two, in my opinion. Is this a result of EQ? I don't know. But I like it.

 

Latests 24/96 Expanded Edition:

Older 24/96:

MQA capture:

[miguelito]

23 minutes ago, mansr said:

I'm just trying to figure out what MQA actually is.

It's MeQA sweetie!