Article: MQA: A Review of controversies, concerns, and cautions

[FredericV]

Excellent article.

Regaring footnote 25:

 

Quote

Of interest, it appears that only the “baseband” bits are being “authenticated”. In an experiment here by FredericV, when the lower 8 bits are dropped, the MQA “blue light” still shines even though it’s recognized as 16-bit audio. Maybe this is all that MQA-CDs are?

On a related note, doesn’t the fact that this blind spot in the authentication mechanism exists immediately disqualifies the MQA blue light from being something a consumer should have any faith in that the file is of “guaranteed” provenance?


 

This is indeed true for older Mytek Brooklyn firmwares. Using the latest firmware, the blue MQA dot will shine with original MQA files, and also with the sabotaged MQA files,  both showing 24 bit 352.8K in case of 2L.no demo files.

So the customer does not know if the file has been truncated or not. He does not know if he has a full first unfold, or partial first unfold.

Furthermore, when NOT truncating the file, only a small part is added back to the decoded output (light purple), compared to the spectrum that is generated by MQA's leaky filter (dark purple), which adds content above 20 Khz which does not even exist in the 16/44.1 file, due to aliasing.

As this small part is encapsulated in the lower  8 bits of the 24 bit distribution file as non-audio data, it does not compress with flac, making the flac size twice as large for a 24/44.1 MQA file vs 16/44.1 truncated MQA file in a 24/44.1 container.

 

Futhermore, sox filter settings exist which generate a very similar spectrum from a 16/44.1 MQA file:

[FredericV]

1 hour ago, The Computer Audiophile said:

Is there any way to upsample a track and have it still illuminate the light?

 

MQA supports the case where the first unfold is done in software like Tidal (the output of the 1e unfold has a stream in the LSB bit which has signalling info that is detected by the renderer), and the second unfold aka renderer in the DAC. So Mansr can confirm or not if you can create fake 24/96 or 24/88.2 bit files with a pattern in the LSB bit of a simulated first unfold, that will trigger the blue light. I did not check those tools yet.

Our experiments manipulate the input files of the first unfold. But messing with the output of the first unfold while leaving the control stream intact, is another experiment that could be done. The same perl tool I wrote can be reused, all I need to do is not mess with the LSB bit.

What we did is manipulate the lowest 8 bits in MQA files. Either blank them (all bits zero), change them into a repetitive ascii string, or replace them with random garbage. All these files authenticate.

In case of 2L.no, mqa files render 24/352.8, while with the manipulated version, the data to reconstruct the ultrasonics is gone, but the MQA light still shines, and the display still says 24/352.8. Older firmwares told the partial truth: 16/352.8

So this is as bad as upsampling.
 

Quote

im just thinking about ways unscrupulous people have sold music and things MQA says it prevents. 

MQA uses the same tricks. It relies on upsampling to do it's trickery.

Suppose your input file before MQA encoding is > 96 Khz (eg 24/192, 24/384, 24/176.4, 24/352.8 ..... ) . They reduce the bandwidth to 88.2 Khz or 96 Khz (so in the analog domain 44.1 or 48 Khz) which they fold to a 24/44.1 or 24/48 distribution file. Anything analog above 44.1 or 48 Khz is lost. But the MQA dac could fool the customer into believing this is 24/352.8 or 24/384.

Even a file from 2L.no truncated to 16/44.1 still shows 24/352.8K on my Mytek.

This is false advertising and would not work in my country. One ISP sold a subscription "fibernet". So customers were fooled they were given fiber speeds, but the network was HFC, so fiber to the curb instead of fiber to the home. This ISP had to change their advertising into adding something like "fiber to the node".

So applying this to MQA, MQA would need to change the indication on my mytek with MQA files to something like: "Upsampled resolution: 24/352.8" instead of showing "24/352.8".

Something like "unfold 24/44.1 => 17/88.2, upsample to 24/352.8" would also be more accurate, but this will most likely never happen.

 

Quote

If its possible to hack MQA into illuminating the light that’s one thing, but if standard nefarious methods can do the same, that’s a much larger issue  

 

Leaky upsample filters such as a modified sox filter, can do a very similar spectral response for a 16/44.1 MQA file as MQA's own upsample filter. Which means the spectrum above the baseband is partially or 100% fake (depending on the resolution of the encoder input file). There's a good example in Archimago's article:

 

 

Mind the MQA gap.

 

 

[FredericV]

5 minutes ago, mansr said:

I have a tool to do just that. However, this gives a red or purple light, not blue or green. There is no cryptographic authentication at this stage.

Does this mean that when a user has Tidal doing the first unfold, and then some MQA dac for the renderer, there's no longer a blue light shining on the MQA dac?

What I intend to test, is to mess with the 24/88.2 or 24/96 first unfold, leave the signalling bit intact, and flip or set bits in other parts.

[FredericV]

17 minutes ago, Chiger Yelam said:

I don't see MQA taking a monopoly position here, legacy codec will still be available and alternative improved products may emerge. If MQA and like minded innovators are undermined and ultimately fail then I fear we will be left with genuinely inferior products.

 

Spreading FUD. Did you read the article? It's all backwards. With MQA we get inferior file formats.

 

[FredericV]

Remember the GO LISTEN argument? Most arguments of MQA opinion makers fall into one of these pre-defined categories:
 

Here's how the GO LISTEN argument works, from their secret MQA group:

The post opened with "Oh boy here we go again". I removed all names from the post and replies. Screenshot used under fair use.

When they no longer have any argument left, they always revert to the GO LISTEN argument. So this is why we have articles like this one, and several MQA topics, which counteract such non-critical thinking.

They basically throw away all our research. Which means they know it's true. If there were mistakes in our research, they would attack them. They can't. So all they have left is their GO LISTEN fallback argument.

 

[FredericV]

1 hour ago, Em2016 said:

Is this the same FB group that includes Bob himself? Does he 'like' the posts and ideas that are thrown around in that group? 

 

Yes it is:

 

and Bob's a member:

 

While the group is closed, the member list is public.

 

[FredericV]

17 minutes ago, Pete-FIN said:

I found out that Pro-Ject Pre Box S2 Digital can not change DAC-filter when listening MQA. I am in the belief that Meridian has forbidden this to be possible. This is just my assumption, I don't know for sure. So, to get some actual knowledge to this, I present the following questions.

 

MQA forces their leaky minimum phase upsample filter with one cycle of postringing. This filter adds content not in the original file, here's an example for truncated 16/44.1 MQA content where we threw away the part to unfold the ultrasonics.



So everything between 20 and 30 Khz is fake due to their leaky filter.

When the decoder has access to those 8 LSB bits to do the first unfold, it adds a little more spectrum:

 

Quote

 

What is Meridians policy, is it allowed (or forbidden) to make an audio gear that have user changeable filters while listening MQA ? 

 

They tried to enforce their filter to another well known manufacturer of MQA dacs. I have that on the record in a chat with the lead designer of this brand.

Any brand can refuse such filter for non-MQA content (and MQA will allow it), but for MQA playback when the MQA decoder is active, it's always played back with their leaky filters.


 

 

[FredericV]

13 minutes ago, firedog said:

Meridian hasn’t forbidden it, as far as we know. It apparently can be difficult to implement, which is one reason it doesn’t happen. There are DACs that switch, I can’t really remember which ones. 

Agree for non-MQA content. But for MQA content you have to use MQA's filter.

Actually some DAC's sound better with non-MQA when you completely disable the MQA decoder, so the upsampler is not the leaky minimum phase filter. MQA should not mess with non-MQA PCM.

[FredericV]

1 hour ago, mcgillroy said:

Extra thanks to Bob for three+ years of great audiophile entertainment and enlightenment. Learned a ton and had great fun along the way. Send me your address I make sure the CA community sends you the promised flowers

 

I knew from day 1 something was wrong with MQA. It was more like a gut feeling. So thanks to CA & Archimago we started to research filters and validate MQA's claims, which was something for the long term, but thanks to MQA, we started with it sooner.

So:

1. sox can also do MQA's time domain



 

2. sox can also generate fake spectrum from 16/44.1 MQA distrbution files

 

But the best part: sox can implement much better filters than all these crappy time domain filters which are leaky and have aliasing issues. Again thanks to Archimago for figuring out intermediate phase, which to my ears beats all these leaky MQA filters:

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

 

Best part is that anyone with 30 euro budget (with SD card and PSU let's say 50 euro, unless you buy the raspberry pi zero w which is even cheaper) and any USB dac can play with these filters thanks to the raspberry pi.

Independent researchers not affiliated with the industry can come up with better filters than well established brands.
For me this is not a small revolution.

So thanks @Archimago for publishing all your informative articles. A year ago we suspected minimum phase & apodizing to be used as MQA would re-use their old Meridian tricks, but thanks to @mansr reverse engineering, it's more clear what MQA is doing under the hood in each stage.

In the past, my job was authentication of bank transactions logs in a PCI compliant environment. So messing with MQA's authentication system was a fun part to do    I had to guarantee logs were not faked, and not altered. I also had to protect who could read the log using crypto. Sysadmins could not read the logs, even as root. And there was a complete PKI behind it.

So MQA's authentication is one big joke and 1/3 of the data can be dropped no longer making it highres, and the blue light still shines.

[FredericV]

This screenshot shows the average technical knowledge level inside the secret MQA group:

"Archimago is quite good, but he really does not understand how MQA works."

 

-> please buy a mirror

This guy does not even understand the basics of sampling.

 

The same clueless members continue to claim the MQA train cannot be stopped. This is a very recent screenshot.

Our CA frontpage article was also deleted from the group.

I don't have a membership in the group, so they now must find their mole ....

 

and yes this internationalTV format was invented by our Belgian national television

[FredericV]

2 hours ago, Oystein said:

After reading this great article and much in dept technical details I found a similar conclusion on 6moons which also describes a way how to do without the MQA encoding and avoid their filter

 

It is published in the Keep it Honest chapter in cooperation with John Darko.

 

That's my own research. SoX can work very well with MQA files. No need for us to license anything from MQA.
Auralic also uses SoX.

 

 

[FredericV]

New fallback argument on the secret MQA group: person X does not understand MQA.

Archimago does not understand MQA
AIX records does not understand MQA
and so on ....

Now they are attacking AIX:

 

How ironic, as MQA does not have more resolution than 24/96. Everything above that is upsampled in the renderer with leaky filters.
MQA at best is something like lossy 17/96.

A lot of MQA encodes are based on masterings for redbook. The admin of the group debunks MQA by his own logic.

[FredericV]

8 hours ago, Archimago said:

So he wants audiophiles to spread the word about his channel at the end so he can keep people "informed". Apologies if I stay away from my PayPal account and Patreon as I have no desire to support misinformation.

I already busted Hans for spreading misinformation. He is one of the key opinion makers and just takes their marketing BS for granted, copying it into video's and website articles.
 

 

Those who understand MQA's bit allocation, will know this is a lie:

 

 

[FredericV]

11 hours ago, Ralf11 said:

 

New Shill Alert for IndyDan

 

 

How many CA accounts are in sleeper mode, to be used after a long time?

 

• Content count

  1

• Joined

  December 14, 2016

• Last visited

  Thursday at 04:43 PM

[FredericV]

4 hours ago, Doug Schneider said:

2) The one thing he definitely not wrong about, though, is how his ADC works -- it has no anti-aliasing filter. When talking about ideal format sample rates, I told him I thought 24/88.2 and 24/96 would suffice, where he wanted 24/176.4 and 24/192 because at these high frequencies no sound will get into the signal chain anyway and, therefore, ADCs could be designed filterless, as he did. In any event, the key about this is there is no "blurring" for MQA to correct. So what you would've heard from these recordings had nothing to do with any of that -- unless, of course, as Archiamago's tests show, MQA actually worsens timing accuracy, or it's some other effect.

 

Doug Schneider
SoundStage!

 

In digital photography, the same idea is used. Camera's with extreme megapixel counts, such as Nikon's D850 don't have an AA filter.

https://www.outdoorphotographer.com/photography-gear/cameras/can-you-go-no-low-pass/

They do not expect content to max out the actual resolution of a 40+ megapixel sensor, as the content is already band limited by the lens which acts as a resolution limiter.

With my D750 an AA filter is required. My lenses easily outperform the D750's resolution. Even very old lenses from the 90's can max out the 24 megapixel resolution of my camera, so an AA filter in this case is needed.

 

[FredericV]

Can MQA come up with a study where 90% prefers MQA over the real studio master?

Archimago's CA article is being systematically removed by Peter Veth from the closed MQA group.

In this reaction to someone posting the Archimago article (apparently it's being posted several times, and deleted by PV), he clearly hints that I am the mole ( I have a product with 432 Hz conversion, I am the only manufacturer of streamers who has such a plugin ). But I am not the mole, but there are several moles in the MQA group, who spontaneously share posts with me. Good luck Peter for the mole hunt.

 

PV learns nothing, as we already proved to PV we don't pitch shift in our product. But PV is always repeating the same refuted arguments like a good troll.

We did our own test long before we had an actual product ( at the time I was an IT consultant, and did not even consider going in the computer audio business ), with 60 persons, and 58 preferred the 432 Hz version over the bitperfect original. We used foobar at the time and the soundtouch plugin to do this test.
 

And yes there's a study from Maria Renold with 2.000 test persons which proves 90% prefers 432 Hz.

MQA can't even provide such test score. Archimago's test showed that the MQA effect is much smaller than a 90% pro vs 10% con distribution. Our own testing comes very close to Maria Renold's test results.

 

[FredericV]

3 hours ago, Fokus said:

 

Given high quality high-res source material and a state of the art SRC with programmable filters, such as iZotope RX, it is quite easy to create a set of test files that mimic MQA-style 'deblurring'.

Or use sox filters. Sox can easily replicate MQA's time domain filters, and so much more.

[FredericV]

5 hours ago, miguelito said:

Consider that about 8,000 albums have been encoded in MQA. Imagine there are 10 teams doing this encoding, and they manage to sort out all of the information of each album (ADCs etc) such that they can encode 1 album a day per team. And these teams work 252 days per year - ie every business day each team cranks out an album. This means that it will take 3.1 years for these teams to do these 8,000 albums. Ponder...

The bulk of the albums is most likely batch converted, with some special cases where they will do more effort.

They need a story to sell, which is the provenance / white glove argument. So for a very limited amount of albums, they probably did more research into how the album was recorded / created, and have press like Hans Beekhuyzen write about it.

When the album is a complex mix of many AD/DA steps (such as effects processors on the effects bus of a mixing desk, instruments with sample banks such as most keyboards - basically everything that is non-acoustical or processed),  no way MQA is going to fix all those errors which were already downmixed into a 2 channel version.




 

[FredericV]

9 hours ago, miguelito said:

Why can't I put step 'b' in step '3'? Are you saying that the lossy MQA compression is inextricable from 'b'? Why would that be? And if there's some processing done in 'c' that is part of 'b', why couldn't I still put it in '3'?

 

Looking at Archimago's & Mansr's research, I believe this is exactly what they do in the renderer in combination with dithering and upsampling.

Why would they need 32 filters, where every file has one pre-defined applied filter out of those 32 available filters?
The coordinates of those filters were also dumped and reverse engineered.

[FredericV]

8 hours ago, Em2016 said:

 

Like this?

 

"This “slow roll-off” filter reduces the time smear by a factor of ~20x compared to conventional digital filters. The net result is a much more musically natural sound, as the ear-brain is very sensitive to time-related distortions. This filter provides an outstanding compromise between frequency response and transient response, and for ten years was the mainstay of Ayre’s digital audio filters."

 

https://www.ayre.com/pdf/Ayre_MP_White_Paper.pdf

 

 

 

These filters can be implemented using SoX.

 

They also leak a lot, here's the spectrum from a 16 bit MQA file, using MQA's filter vs SoX minimum phase and one cycle of postringing:
 

 

All the content above 20 Khz = fake.

[FredericV]

9 hours ago, tmtomh said:

For example, a common cell phone frequency band is 1900MHz (not kHz), with a 100MHz bandwidth, meaning the phones operate from 1850MHz to 1950MHz. By using simple Nyquist theory, to digitize those waves, you'd have to sample them at 3900MHz, which is 2x the max frequency of 1950MHz.

 

In my realtek based SDR kit, it does not work this way from a logical standpoint.

My RTL sdr dongle has a max bandwidth of 3.2 Mhz. Dongles with more bandwidth (like 5 Mhz and 8 Mhz) also exists, but costs a lot more.

Suppose I want to capture the FM band and monitor anything between 100 and 102 Mhz. I specify 101 Mhz as the tuning frequency, with a bandwidth of at least 2 Mhz.

 

If I want to monitor a bigger part of the FM band (let's say 100-108 Mhz) that does not fit into the bandwidth of my dongle, my dongle will frequency hop, therefore missing at least 50% of the data.

My dongle is not sampling at 216 Mhz rate and exposing all this data over USB.
 

[FredericV]

On 3/25/2018 at 3:21 PM, mansr said:

I stopped listening at "my friend Bob Stuart."

He does not fully understand MQA. He believes MQA can uncompress back to 24 bit 192 Khz, not realizing it's never going to be better than something like 17 bit 96 Khz upsampled to whatever the DAC is lying about on the display.

[FredericV]

In the Dutch pro audio world, they translated parts of your article into this summary, in Dutch:

https://www.helios.nl/nieuws/mqa-tegenstellingen-zorgen-en-waarschuwingen

Alpha-audio, one of the Dutch sites specialized in streaming, now also mentioned this translation on their frontpage:

https://www.alpha-audio.nl/2018/06/mqa-ter-discussie-fake-bestandsformaat/

which means "MQA up for debate: fake file format?".

Alpha Audio's Jaap Veenstra interviewed Bob in Munich (which lead to interesting facts disclosed by Bob), but now he's also allowing the research of independent MQA researchers.