MQA is Vaporware

[mansr]

9 hours ago, Don Hills said:

Mans, I haven't seen anything about the coefficients of the filters used to split (and later rejoin) the audio into the 0-24 and 24-48 KHz bands for "origami" folding. Are they also leaky and "time optimised"? I would expect the performance of these filters to be more critical than the anti-alias / anti image filters. 

We obviously don't have access to the splitting filters. The joining is in decoder somewhere, but I haven't (yet) figured out how it works. There's a lot more code to analyse than for the renderer, and I have only so much time.

[mansr]

8 hours ago, Charles Hansen said:

However it seems that the story is just that -- a story. When Mansr and Archimago dumped the filter responses of MQA "renderers" using completely different DAC chips (Burr-Brown in the Explorer2 and ESS in the DragonFly), the results were identical. In other words no difference at all between the filters whatsoever.

Allow me to elaborate a bit on that. For the Bluesound (with BB/TI DAC chip), we know the actual filter coefficients used by the renderer. With the other MQA renderers we examined (Dragonfly and Mytek Brooklyn), we could only record the analogue output, and the impulse responses we measured match very closely the digital filter values from the Bluesound firmware. It would be easy to record the digital data going into the DAC chip of, say, the Dragonfly, but I don't have one and do not intend to buy one.

[Fokus]

11 hours ago, Don Hills said:

 

Mans, I haven't seen anything about the coefficients of the filters used to split (and later rejoin) the audio into the 0-24 and 24-48 KHz bands for "origami" folding. Are they also leaky and "time optimised"? I would expect the performance of these filters to be more critical than the anti-alias / anti image filters. 

 

For the split/join to be theoretically lossless these filters must be two quadrature mirror filter pairs. This on its own imposes severe constraints. I don't think that on top of these constraints the MQA people have much choice for further optimisation. (We should ask Ingrid Daubechies.)

 

I have been asking this since nearly a year now: is an MQA DAC's standard CD replay filter (as seen in Fig.4 here Explorer 2 review at S'phile) part of these two QMF pairs? And if so, what does this mean for the other 3 QMF filters, and thus for the quality of the non-decoded MQA signal (i.o.w. has it been infected with aliasing in the audible band)?

 

[Fokus]

8 hours ago, Charles Hansen said:

However I can find no evidence that he has ever designed his own custom digital filter.

 

He designed among others the Audiolab MDAC, which has a herd of custom filters. Some even different implementations of the (claimed) exact-same response (also claimed are, of course, vast audible differences between these  filters, because digital is, well, much more mysterious than EE-educated people think).

 

He is also the creator of the Lakwest MDAC2. Crowd funding was done in 2013. The design goes on, and on, and on, and on, ... There is a healthy secondary market of people buying and selling their MDAC2 order tickets.

[crenca]

10 hours ago, Charles Hansen said:

John of Westlake says: ....Having a far few products pass through the MQA certification process I can testify to the fact that there a SIGNIFICANT engineering effort made by the MQA team...

 

This is what Mr. Ritter of Berkeley Audio asserted over on the TAS website on the comments on Mr. Quint's "Political" article.  Turns out, it is just an assertion and all available evidence is contrary to it.  When I pointed this out to him, he said that he is a somebody in Audio and that those who disagree with him are nobodies.  Yep, it is as arrogant as that comes across.

 

Is ALL of "High End" a confidence game?

 

I do find it possible that Mr. Westlake is correct in that perhaps MQA had more hardware specific filter tuning up its sleeve but when it realized that Tidal was going to have to be point of the spear for it as far as getting MQA into the hands of (some) consumers, they rolled out MQA v1.2 that is more generalized.  The fact that they have adjusted and (if we are to be honest) "dumbed down" MQA in an effort to sell it should surprise no one, nor should the fact that they kept the original marketing materials and claims that were based on MQA v 1.0...I mean, who wants to rewrite trumped up marketing materials when "audiophiles" are simply not supposed to (and almost never do) question the voodoo?  

[crenca]

4 minutes ago, Fokus said:

 

He designed among others the Audiolab MDAC, which has a herd of custom filters. Some even different implementations of the (claimed) exact-same response (also claimed are, of course, vast audible differences between these  filters, because digital is, well, much more mysterious than EE-educated people think).

 

 

 

A big LOL!  

 

All that mysterious ether - those rascally angels that make 1+1=2 "sound like" 1+1=3.  The thing is, all the "sounds like" writers and editors of all the trade publications really do believe in this stuff.  Thus, we get MQA...

[Charles Hansen]

5 hours ago, firedog said:

And do have any idea what he means by this:

As I also mentioned the Renderer also has other processing - not just the Time domain processing.

 

Hi Firedog,

 

No, it doesn't make any sense to me. One possibility is that he simply doesn't understand what he is talking about. We do know that the Microchip processor used in the AudioQuest DragonFlies does not have enough computational power to actually perform any DSP, so the only reasonable conclusion is that it simply loads the MQA coefficients into the programmable filter used in the ESS DAC chips. That will affect both the frequency response and the time domain response. The Microchip could also easily perform the zero-stuffing required to interpolate ("upsample") the 96kHz audio data from the software (Tidal or Audirvana) decoder. Also the software wends instructions as to which of the 16 filter sets should be used for each track, and again it would be trivially easy for the Microchip processor to implement those commands.. Perhaps those are what Mr. Westlake is referring to.

 

Hope this helps,

Charles Hansen

[rickca]

11 minutes ago, crenca said:

more generalized

I'd call it watered down.  But you can never back off your original marketing claims.  Of course, those who have signed an NDA will say we don't really understand MQA and that we are in no position to challenge their assertions.

[mansr]

21 minutes ago, Fokus said:

He designed among others the Audiolab MDAC, which has a herd of custom filters.

Designing a filter might mean nothing more than firing up the Matlab signal processing toolbox and punching in a few desired parameters. Anyone can do this.

[crenca]

9 minutes ago, rickca said:

I'd call it watered down.  But you can never back off your original marketing claims.  Of course, those who have signed an NDA will say we don't really understand MQA and that we are in no position to challenge their assertions.

 

Yep, and the NDA played right into the confidence game that IS "High End".  I have said it before and will say it again that I have immense respect for Bob S/Meridian/MQA's ability to read and understand the lay of the land of "High End", and then create a scam product that leverages the circumstances.

 

Here is the deal however;  sooner or later we will read an interview with someone who has signed the NDA but is not willing to simply back up the claims.  He will have to tread lightly of course, but it will be clear that he is making a point that the engineering and math behind the IP is in fact what is already known - not what it is advertised as.

[Fokus]

15 minutes ago, mansr said:

Designing a filter might mean nothing more than firing up the Matlab signal processing toolbox and punching in a few desired parameters. Anyone can do this.

 

Yes.

 

I used to do this without the SPT

 

This was just to signal Charles that JW has some custom filter experience.

[Charles Hansen]

1 hour ago, Fokus said:

 

He designed among others the Audiolab MDAC, which has a herd of custom filters. Some even different implementations of the (claimed) exact-same response (also claimed are, of course, vast audible differences between these  filters, because digital is, well, much more mysterious than EE-educated people think).

 

He is also the creator of the Lakwest MDAC2. Crowd funding was done in 2013. The design goes on, and on, and on, and on, ... There is a healthy secondary market of people buying and selling their MDAC2 order tickets.

 

Hello Fokus,

 

Thanks for the correction. As you hint at, this is a bit mysterious. I could find no good photos of the insides, nor any test reports from any publication that makes measurements. The manufacturer's website claims the unit uses the original ESS Sabre 32 ES9018 DAC chip, which only has two filters built=in - a sharp rolloff and a slow rolloff. However the filters are custom programmable, and it appears this is what Westlake has done with the M-DAC. It becomes curiouser and curiouser as one reads the descriptions of the filters. As you point out three of the filters are called "Optimal Transient" with no claimed ringing, which is a difficult task to pull off, let alone with three different implementations that are claimed to have identical frequency response and time domain response.

 

The digital filter in the ESS chip is somewhat unusual as it is an 8x interpolation filter (like many, many other DAC chips),. However instead of the typical concatenation of three 2x sections to achieve 8x, it uses a 4x followed by a 2x. It is difficult for me to understand how Westlake could arrange two building blocks in three different ways to achieve the exact same results. That is just the beginning of curious things I've never seen claimed before, not only for the digital filters but also for a "Digital Data Decorrelation Engine" that sounds like it simply dithers data with word lengths less than 24 bits,. Finally it has a feature whereby it is claimed to "correct Windows LSB rounding errors", which is a very interesting feature indeed. I've never heard of the "problem" before, and it is unclear to me how one would know whether the rounding had been up or down (in order to properly correct it). But who knows? Perhaps he is right and everybody else is wrong. None of us know what we don't know, and perhaps Mr. Westlake is simply that far ahead of everybody else.

 

Hope this helps,

Charles Hansen

[Charles Hansen]

2 hours ago, Fokus said:

I have been asking this since nearly a year now: is an MQA DAC's standard CD replay filter (as seen in Fig.4 here Explorer 2 review at S'phile) part of these two QMF pairs? And if so, what does this mean for the other 3 QMF filters, and thus for the quality of the non-decoded MQA signal (i.o.w. has it been infected with aliasing in the audible band)?

 

Hi Fokus,

 

No that graph is of the digital reconstruction filter used by MQA. It turns out that the BlueSound and the DragonFlies have a choice of 16 different filters. I don't recall if that is the case for the Explorer2 or not. I think Archimago touched on that in his blog, but don't recall for sure. At any rate it is definitely not part of the band splitting/recombining process.

 

Hope this helps,

Charles Hansen

[mansr]

5 minutes ago, Charles Hansen said:

No that graph is of the digital reconstruction filter used by MQA. It turns out that the BlueSound and the DragonFlies have a choice of 16 different filters. I don't recall if that is the case for the Explorer2 or not. I think Archimago touched on that in his blog, but don't recall for sure. At any rate it is definitely not part of the band splitting/recombining process.

The Explorer2 must have the same filter options in the render stage. The selection is encoded in the MQA metadata stream and passed on by the "core" decoder. Since the Explorer2 doesn't have a render-only mode, the exact responses can't be measured in the same way we did with the other DACs.

[Fokus]

1 hour ago, Charles Hansen said:

No that graph is of the digital reconstruction filter used by MQA. It turns out that the BlueSound and the DragonFlies have a choice of 16 different filters.

 

That is not the issue.

 

How come that no-one else so far have wondered why ALL MQA DACs have the same reconstruction filter for CD-rate non-MQA material????? A filter that is suspiciously 'wrong', both to orthodoxy and to Meridian's legacy MP aposiders.

 

My guess is that this is simply because this filter is one of the two required QMFs for MQA band joining, pressed into service for CD rate decoding.

 

Again: why do they all have the same response ...

 

[Jud]

1 hour ago, Charles Hansen said:

Finally it has a feature whereby it is claimed to "correct Windows LSB rounding errors", which is a very interesting feature indeed.

 

This just sounds to me like another description of dither.  Or it could be of >24 (16?) bit math in the signal processing chain somewhere, I suppose, which is another pretty ordinary thing.  Or it could be a description of the driver’s operation versus Windows’ low level audio processing.

 

In other words, I have no idea.

[The Computer Audiophile]

3 hours ago, crenca said:

 

 

A big LOL!  

 

All that mysterious ether - those rascally angels that make 1+1=2 "sound like" 1+1=3.  The thing is, all the "sounds like" writers and editors of all the trade publications really do believe in this stuff.  Thus, we get MQA...

 

HI crenca - I can't help but notice something when it comes to MQA and other items. If someone isn't totally against MQA at all times, you bring out all the possible negatives one could think of. In other words, inside every cloud there's a black lining, HiFi is snake oil etc...

 

On the other hand, when someone such as @Charles Hansen makes claims such as WAV and FLAC sound different, wood blocks under speaker cables sound better, etc... you don't jump all over him because he is on your anti-MQA team. It's as if you've sold out to the anti-MQA team.

 

How about a little crenca on the complete pro consumer team rather than just the pro consumer team, when it makes sense.

[Shadders]

1 hour ago, Jud said:

 

This just sounds to me like another description of dither.  Or it could be of >24 (16?) bit math in the signal processing chain somewhere, I suppose, which is another pretty ordinary thing.  Or it could be a description of the driver’s operation versus Windows’ low level audio processing.

 

In other words, I have no idea.

Hi Jud,

If the system implements any dither which may have a triangular pdf, or other, then the result will be a triangular pdf (or other) regardless of the original LSB dither.

As such, there may be no detection operation in place, just the application of its own dither to mask the LSB errors ?

Regards,

Shadders.

[crenca]

1 hour ago, The Computer Audiophile said:

 

HI crenca - I can't help but notice something when it comes to MQA and other items. ...

 

On the other hand, when someone such as @Charles Hansen makes claims such as WAV and FLAC sound different, wood blocks under speaker cables sound better,...

 

How about a little crenca on the complete pro consumer team rather than just the pro consumer team, when it makes sense.

 

Your are right Chris in that I explicitly/consciously focus on this voodoo and not that other voodoo.  I write with an emphasis - but for very good reasons:

 

1) Most importantly, that other voodoo (e.g. wood blocks) are not trying or even have the capability to become a standard (i.e. an unavoidable format for the consumer) at the bottom of our digital ecosystems.  Unless I am mistaken about this...someone point me to the IP protected, DRM Trojan horse, "end to end" wood block effort and I will crenca all over it

 

2)  MQA is an attempt to bring the voodoo into to a domain (i.e. software/digital) that is less susceptible to it because of this domains nature.  In the software/digital world, the bottom is simply math - and math of course laughs off attempts to make 1+1=3  Even in the areas that are "analog", there is a large body of sound engineering that does not normally play in the voodoo end of the pool.  Nobody with an ounce of self preservation would drive over the bridge built by TAS's "sounds like" writers and editors, or the voodoo artists who they promote.

 

3) Related to #1, MQA is not another "product among products" for the reasons stated.  It should not be considered like just another speaker, or even yet another dubious tweak.  Something with the end to end ambitions of MQA deserves a much much higher level of scrutiny than other products (WE deserve it!) .  This truth is something that TAS and other such publications have completely failed to grasp and is an indicator of their position vis-a-vis "the industry" as well as to the consumer.  Thus it has fallen on Mr. Ritters "nobodies" to do the heavy lifting in this area.

 

In the end I think you are right to question the emphasis however.  The disease of voodooism is what has laid the foundation, weakened the immune system, to such a point that Bob S was bold enough to try to infect us with the particularly virulent strain of MQA.  The larger question of what to do about voodooism in general certainly is a valid one and I remain open to suggestions.  I have thought about a kind of "Audiophiles Consumer Union" and what form that might take.  This is not a new idea however and has failed whenever it has been tried.  Still, an organization, or simply a coalition of forces (perhaps a mix of consumers, engineers, and like-minded manufactures/industry players) that existed in opposition to this underlying voodoo culture would be a good thing.  

 

EDIT:   The fact is that this and similar blogs/sites are serving as the "Audiophiles Consumer Union" even now.  This is the real reason why there is a uniform effort by almost all sides of "the industry" (not a conspiracy - just a natural coalition) to play a confidence game against the "nobodies" of the forums...

[mansr]

1 hour ago, The Computer Audiophile said:

On the other hand, when someone such as @Charles Hansen makes claims such as WAV and FLAC sound different, wood blocks under speaker cables sound better, etc... you don't jump all over him because he is on your anti-MQA team.

I'll gladly take Charles to task on any such claims. This is, however, not the thread for that.

[firedog]

http://www.soundstagehifi.com/index.php/opinion/1104-mismatched-masters-and-false-frequencies-is-mqa-better-worse-or-just-different

 

Good summation of the questions around MQA. Not a polemic. 

[kumakuma]

7 hours ago, firedog said:

http://www.soundstagehifi.com/index.php/opinion/1104-mismatched-masters-and-false-frequencies-is-mqa-better-worse-or-just-different

 

Good summation of the questions around MQA. Not a polemic. 

 

Thanks for sharing this.

[crenca]

10 hours ago, firedog said:

http://www.soundstagehifi.com/index.php/opinion/1104-mismatched-masters-and-false-frequencies-is-mqa-better-worse-or-just-different

 

Good summation of the questions around MQA. Not a polemic. 

 

From the article:

 

 

"MQA is more than just a codec. Its multiple aspects include a proprietary digital filter, and a lossy data-compression algorithm..."

 

"... does MQA produce better sound than the original file, or simply a different sound -- one that those who like it prefer? "

 

"... There have been many comparisons of MQA files and other music files, but from what I can tell, it’s questionable whether the files being compared were made from the same master file..."

 

"I’m not in a position to say who is right regarding the issue of timing, but I will say that, from what I’ve seen, neither MQA nor anyone has offered proof that their system improves timing accuracy. However, there is proof that the aliasing is real...."

 

"... However, because it’s a 192kHz recording, the range of 48-96kHz has to be filled in, and the MQA process appears to do that -- not with information from the recording itself, or even silence, but with aliasing..."

 

 

Well well, an article that actually takes a look at the known, objective facts around MQA. It's a Christmas Audiophile miracle!

 

 

[FredericV]

Last weekend we were an exhibitor at the X-FI 2017 show in The Netherlands where our server was being presented in three rooms, including our own demo room where we had carte blanche. We had a lot of fun playing both audiophile and real music including Manley based masters from Jazz Profilactika, a befriended new Jazz band, from which we received their 24 bit master files one day before the show.

We also did a little test with MQA files from 2L.no

Demonstrated using Antelope's Platinum DAC + separate 10M atomic clock + separate Voltikus, and the excellent active John Watkinson speakers which are not very well known, but among the best which I have heard so far.

https://www.flickr.com/photos/147070590@N02/36730762584/in/dateposted/

The co-designer of the Legends is Kommer Kleijn, a "professor" type of guy who is also a respected 3D stereographer
http://www.kommer.com/

 

The lead designer is the guy who wrote the "bible" about digital audio:
https://www.amazon.co.uk/Art-Digital-Audio-John-Watkinson/dp/0240515870

 

I have no affiliation with this speaker brand except that they make a great demo partner. I have no affiliation with Antelope either, in the past we were reseller for a couple of years until we decided to go into the music server business. I now use Metrum Adagio for my own test & dev. I list the specs so that others won't say our system is not revealing enough to do such test. We also fixed the flutter echo of the room using acoustic panels. Just to make the test setup right.

 

Just before the end of the show on Sunday, we played the following test on our server:

1. DXD master from 2L.no, a classical piece
2. mqa version of the above, with a live version of the sox method as disclosed on Darko's site.

We tried to make this test as blind and unexpected as possible, so after the DXD file played (a classical piece), we said we were going to play a different version of the same file. So we played the upsampled MQA version, than asked if they could hear the difference.

They could not. I begged them to please reveal any difference, if any ... nobody in the room was able to tell them apart.

So for me, this is case closed. Why waste any more time with MQA or trying to figure out the secret parts, if the sox method can do the above? The MQA people will fight us and say sox is not the real thing, but sox is pseudo MQA just like what Auralic is doing. But who cares if we achieve the above. The end result is what counts. MQA fans always say our hearing is analog and the number of bits don't matter, and the end result is what's count.
 

Another eye opener was that an AES engineer which I heard talking about MQA in a Chinese restaurant in Veldhoven and later was in the bar of our conference center, told me our hearing has 5µs of temporal resolution based on the fact that we can locate a sound source with 2 to 3 degrees of accuracy and a calculation based on the distance of sound sources, he explained how he came to this number. I was too tired to completely remember how he did that calc.
 

The 50µs that MQA is claiming is not enough. He explained that for a true end-to-end system that can duplicate the in air response of a source, it would need a microphone that can capture much higher frequencies, and the samplerate would have to be much higher than 192K to achieve 5µs. He calculated that it would at least need an analog bandwidth of 200 Khz, so sampling at 384K won't even cut it. It would also require microphones that can capture all the harmonics of S-sounds and cymbals, which cannot be seen in regular spectral plots, as they average because of the integration window and these transients are very short.

It's like measuring the average sound level of an airport at night, when only a few planes fly by and keep you out of your sleep. The average will still not show this.

He was present at an MQA introduction show using very expensive speakers and being demo'ed by Hans Beekhuyzen. He told me the S sounds are all wrong with fully decoded MQA, and that he would never dare to give such a bad demo. This demo was at a very well respected ultra high-end boutique dealer.
 

The interesting part about this AES engineer is that he sent all his remarks to Hans and another well known MQA opinion maker, but Hans completely ignored this. We talked for more than an hour, maybe this guy can post his technical arguments, I still have his card

But for me, this is now case closed. I'm happy that thanks to MQA I put a lot of effort in tweaking our own DSP and I learned so much more about filters, and that we found the sox method and it's correct settings which we posted on Darko, so other devs can start using it. But I'm not a fan of their anti-ringing and weird filters that try to clean up post-ringing.

I also briefly talked to the leading MQA opinion maker in the Netherlands. He also visited our room, liked the sound. Shook hands and got a message on FB today that he wanted to cease fire.

We won't change these people. It is pointless. They won't change my opinion. The fact that we have topics like this is good as this discussion is kept away from the regular press, but in talking with the MQA people, at a certain point, there is nothing new to bring.

This is the reason why the topic at gearslutz was locked: they landed into name calling.

Let's agree to disagree and shake hands.

 

[crenca]

1 hour ago, FredericV said:

 

Demonstrated using Antelope's Platinum DAC + separate 10M atomic clock + separate Voltikus, and the excellent active John Watkinson speakers which are not very well known, but among the best which I have heard so far.

https://www.flickr.com/photos/147070590@N02/36730762584/in/dateposted/

....

 

He was present at an MQA introduction show using very expensive speakers and being demo'ed by Hans Beekhuyzen. He told me the S sounds are all wrong with fully decoded MQA, and that he would never dare to give such a bad demo. This demo was at a very well respected ultra high-end boutique dealer.
 

 

 

 

Thanks for the write up Frederic.  Just to be clear, the active speakers do not have a DAC (and thus a 2nd A>D>A conversion) correct?

 

Also, my personal impression is that there is something going on in the the sibilance region as well that is "off" and leaves me with a "digititus" kind of aftertaste.