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

[mansr]

2 minutes ago, Em2016 said:

Finally something on the 'front page', as opposed to buried deep inside the threads.

 

I hope MQA Ltd do respond eventually. They seem to be using Stereophile as their technical outlet recently, so we'll wait for a response over there I guess.

Lavorgna will post some snark on Monday or Tuesday. Stereophile will follow up with an infomercial ghostwritten by Stuart a week later.

[mansr]

22 minutes ago, The Computer Audiophile said:

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

 

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

 

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  

 

Not a positive either way though. 

The blue light goes on if the top 16 bits match the cryptographic signature. That's probably difficult to fake. Let's say it's impossible. All that means is the file has been passed through a genuine MQA encoder with the right private keys. It says nothing about what the input actually was. If you send an upsampled file for encoding, you'll get back a signed MQA file claiming to originate from whatever fake sample rate you provided.

[mansr]

19 minutes ago, FredericV said:

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.

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.

[mansr]

9 minutes ago, FredericV said:

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?

Right. If the DAC supports this at all, you'll get a red light. Mytek DACs do this. Some others, e.g. the Meridian Explorer2, do not.

 

9 minutes ago, FredericV said:

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.

There's no checking of the data bits here. This is how we obtained impulse responses for the Dragonfly and Mytek Brooklyn DACs.

[mansr]

24 minutes ago, Rt66indierock said:

I don't think Chris had any idea trouble was coming in February 2016.  He might have sensed trouble at RMAF 2016 but I think it was more like a year ago when things started to take off opposing MQA.

I like to think that my reverse engineering efforts contributed in some small way.

[mansr]

15 minutes ago, firedog said:

And what about those of us who have listened and aren't impressed?

You have tin ears, duh.

[mansr]

3 minutes ago, Doug Schneider said:

How long until some guy "clones" it with an algorithm of his own? (I'm actually surprised this hasn't happened yet, but it will. Just to be clear, I'm talking about someone writing an encoder that mimics it exactly. Some hotshot coder who catches wind of it and has a few days clear likely will.)

The only reason this hasn't happened is that there is zero interest.

[mansr]

10 minutes ago, Doug Schneider said:

It was also apparent that they seemed painfully unaware that the bandwidth limitations they were trying to overcome weren't an issue. How did that get missed?

I don't believe that was ever the actual goal. MQA is a hodgepodge of ideas, each appealing to a different target. The labels are promised DRM and protection for their "crown jewels." The streaming services are offered reduced bandwidth. The consumers are told they'll get the "sound of the studio," authenticated by the long-dead artist. All are being deceived. DRM doesn't work. There is no bandwidth reduction. MQA conversions happen without the approval of the artists or mastering engineers (as attested by Brian Lucey).

 

The real goal of MQA is to insinuate itself into every stage of the recording and playback chain, collecting royalties from each participant. As Jim Collinson of Linn put it, MQA is a land grab.

[mansr]

1 hour ago, Doug Schneider said:

I thought the Mytek ones could...

Mytek can disable MQA decoding entirely, which unlocks the usual set of filters. With MQA decoding enabled, even non-MQA content is processed with an MQA filter.

[mansr]

14 minutes ago, botrytis said:

I am disheartened by the reaction on Stereophile, particularly Mr. Atkinson, as to Archimago and his pseudonym. I feel his reaction here, is one thing and then on Stereophile's site, it is another. 

 

I understand Archimago's reasoning about using the pseudonym. This is also a passion/hobby for him not his sole means of support. It seems since they cannot deflect, damage, or deny the science and thought behind the article, they def left and go after the author. This is telling.

It's also a well-known logical fallacy.

[mansr]

12 minutes ago, ednaz said:

Watching the arguments of what's kept versus thrown away, what's real and what's invented, is it debarring or blurring, is it just upsampling, wait is that noise, brings to mind something from another domain - photography.

 

When printing digital photographs at display sizes - 16x20 inches, 20x30 inches and larger - professional printers, the type that would print images for a gallery or museum show, do a couple of tricks to every image, just before printing. (Learned these working for a famous NYC fine art printer.)

 

First, they apply an unsharp mask to increase the apparent sharpness (de-blurring), which paradoxically works by applying a mildly blurred image as a mask on to the original image. De-blurring by adding a mask of blurring. It raises the apparent sharpness of the image. Done well, it's not noticeable. Done poorly, you get visible halos in the image. Note that even done well, there are halos - the unsharp image absolutely makes them, but they're below your ability to see them - a pixel or two wide. (I learned to do this in film days. Much easier in digital.)

 

Second, they add noise to the image. Everyone evaluates digital sensors based on their ability to produce an image free of digital noise... but completely noise free images look odd. In large areas with no detail - sky, a car fender, still water - they look artificial and plastic. The printer uses one of a number of techniques to generate digital noise that's similar to film grain, and blends it into the image. The size and frequencies of the noise are based on the size of the final print. Again, the goal is to have it be there and effective but not noticeable. (When I show people prints where I've done this, they have a hard time detecting it, even after being told what to look for.) That added noise does three things. It makes the image seem more real, and less digital. It reduces the visibility of actual digital noise from the sensor. And, it also increases the apparent sharpness of the image.

 

In photography - which is about capturing the most accurate renditions of light and color with a recording device and then reproducing them for viewing - adding information that was never there to begin with increases the perception of it being a more accurate and sharply rendered image of the real world.

 

I imagine that the same types of tricks, applied to audio files, may improve the apparent accuracy and crispness of the rendering of recorded sounds. After all, we see and hear with our brains, not our eyes and ears.

You make a good point. However, our visual and auditory perception work in very different ways. Just consider that 25 frames per second is enough to produce a convincing video while audio requires a sample rate 1000 times higher to get anywhere close. We are far more sensitive to distortions in sound than in images. Audio with as much distortion as a good JPEG image is unlistenable. Tricks that work well for imaging and video rarely transfer to the audio domain.

[mansr]

5 hours ago, Archimago said:

I agree with the others, we need to be cautious. So in this analogy, what do you make of MQA?

JPEG2000, of course. A "better" lossy format that nobody needed and never got any traction.

[mansr]

7 hours ago, ednaz said:

All that scientific fact about moving images (I was talking about stills) ignores what I see about MQA. People with great ears (I was a union card holding musician until I was 30) talk about how MQA has more reality in the sense of environment. Which, given the de-blurring techniques, and noise feedback... makes a lot of sense.

 

Detecting distortions... all those image print techniques are based on taking advantage of our perceptions. I think MQA takes advantage of our perceptions. Not reality. Brains hear. They function at a pretty damn low sample rate.

We know MQA messes with the phase of the signal. There are also hints at subtle EQ and trickery with the stereo spread. If you like those gimmicks, that's fine. Just don't foist them on everybody else. The noise MQA adds has nothing to do with perception. Most likely, the purpose is to mask the nasty artefacts of the MQA process, not from our ears, to which they are inaudible, but from measuring equipment.

 

7 hours ago, ednaz said:

Incidentally, I'm not talking about JPEG. Not even 100% JPG (although if you're willing to put up a big bunch of money I'll let you try to prove to me that you can ID JPG distortions.)

You might be surprised by the outcome. Luckily for you, I'm not going to put up "a big bunch of money" towards such a challenge.

 

7 hours ago, ednaz said:

And I suppose I'm assuming an ability to grasp analogies. I could well be wrong. Most of my work is based on cross domain analogies, but not everyone can do that.

That's getting dangerously close to an insult. Watch your step.

[mansr]

2 minutes ago, ednaz said:

"Watch your step." What kind of comment is that? Lovely.

Just a piece of friendly advice.

 

2 minutes ago, ednaz said:

Not an insult at all, but a statement of fact. Most people don't do cross-analogy thinking very well. After 35 years of leading teams in neuro-science, computer code reverse engineering, photography, drug discovery, entity analytics, cryptography, human systems analysis, and more,

Jack of all trades, master of...

 

2 minutes ago, ednaz said:

People claiming they can detect jpg vs other file captures remind me of people who say they can always tell lossy compressed music files from non-compressed.

I can do that too often enough. Countless hours of codec testing does that to you.

 

2 minutes ago, ednaz said:

I can with a lot of music, but when you get Alabama Shakes running a DR of 3... can't any more. I think that may be part of what's going on in the MQA arguments, and is a variable I can't remember being explored.

Sorry, I lost your chain of thought.

 

2 minutes ago, ednaz said:

Wouldn't it be interesting if, instead of challenging their hearing or expertise, someone tried to understand it?

We did. Successfully. That's what the article is all about. Did you read it?

[mansr]

3 hours ago, John_Atkinson said:

Interesting posts on imaging. I will admit that I don't know much at all about digital photography, but from my reading of sampling and filtering, it appears that sinc-function filters - which are almost ubiquitous in digital audio - are sub-optimal for image processing. Is that correct?

No, that is not correct. The ideal interpolator is always the sinc function. In imaging, Lanczos is a commonly used approximation.

[mansr]

52 minutes ago, sullis02 said:

Audio with tons of added ultrasonic noise = RAW video?  Hmm.     What does 'RAW' actually mean in image formats  (given that every reproduction method has limits)?      

Raw (I have no idea why people write it in all caps) just means the raw data from the image sensor plus information about lens, flash, and various settings. To be useful, it must be processed in software.

[mansr]

8 minutes ago, firedog said:

Uh, I think you meant to write, “an adult who should be embarrassed to use his real name because now everyone knows he is either ignorant and doesn’t know what he is talking about, or just a shill for MQA who will say anything, true or not.”

Or both.

[mansr]

1 minute ago, Archimago said:

Well, in the video he states he's not supported by MQA. Not reason to question that...

He has a price list. Of course, we don't know whether MQA has paid him, but the possibility is quite real.

[mansr]

1 hour ago, FredericV said:

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

It's probably possible to sell old accounts to troll/shill farms.

[mansr]

1 minute ago, John_Atkinson said:

I have not yet read anything that would convince me to change my mind. Much of the most virulent criticism comes from non-technical people and where it comes from technically astute commentators, I have serious issues with much of what is written.

Really now?

[mansr]

1 hour ago, Doug Schneider said:

I'm just going to go out on a limb and guess -- it shows none of the filter artifacts typical of ADCs. That's just a guess. But if so, it's not a surprise because Hansen has stated for a long time that it has no filter.

It's a sigma-delta ADC. Of course it has a filter.

[mansr]

42 minutes ago, Doug Schneider said:

You are right, but I wish I could go back and talk to him more. And you know a lot more about A/D and D/A than most. But I swear in one of his conversations he was talking about a lack of a filter at 24/192. I will look more as he described this in pieces on AA, but here's part. But it's not everything.

 

https://www.audioasylum.com/forums/pcaudio/messages/16/167848.html

 

EDIT: As I said, above I could've sworn what he said on the phone about the filter. But I'm certainly OK to be corrected because in that above link he talks about the filter used at 2X and 4X conversion, so there's something there.

A sigma-delta ADC doesn't need an analogue anti-aliasing filter since the sample rate (typically 128x or more) is far higher than required by Nyquist for any audio signal. To produce a PCM output, at any sample rate, a digital low-pass filter must be used to remove the modulator noise. Presumably this ADC uses Ayre's signature slow roll-off minimum phase filters.

 

42 minutes ago, Doug Schneider said:

What he claims, though, remains the same -- that there are no timing issues to correct with the QA-9 Here's a little more:

 

https://www.audioasylum.com/cgi/t.mpl?f=critics&m=87523

I'd say no reasonable ADC has any timing issues. Here's a square wave with 5 ns rise time recorded at 192 kHz on a Tascam UH-7000:

Sure, there's a little overshoot and ringing, but real audio signals don't have such short rise times so it doesn't matter.

[mansr]

22 minutes ago, Doug Schneider said:

I should've been more specific. He was talking about anti-aliasing filters. And, yes, it makes sense from what I've read about that piece, it used their slow roll off filters.

Of course it's about anti-aliasing. That is the purpose of any filter in an ADC. In a sigma-delta ADC, the AA filter is digital. If you somehow managed to build a 22-bit flash ADC operating at 192 kHz or higher, you wouldn't need any filters at all. Unfortunately, that is highly unpractical, if at all possible. Although the audio signal has essentially no content above 100 kHz, the output of the sigma-delta stage does, in the form of modulator noise. For this reason, PCM output, even at 192 kHz, requires a digital anti-aliasing filter. Otherwise the audible range would be swamped in aliased modulator noise.

[mansr]

14 minutes ago, guymrob said:

Since SDM is basically a 1 bit converter (high bit-rate DSD), it requires noise shaping prior to decimation to Hi-Res PCM.

Not quite. SDM is noise shaping. Noise shaping, sigma-delta or otherwise, is a feedback loop around the quantiser. The noise can't be reshaped later, when the input signal is no longer known.

 

Quote

For instance, if the SDM ADC operates at 256x, noise shaping pushes virtually all the noise(depending on type of order; whether is 5th or 7th order modulator), the noise is pushed all the way to 176.4kHz, thereafter noise starts to increase. Since decimation happens not more than 176.4kHz, noise from SDM is virtually non-existent when it is converted to Hi-Res PCM, in this case 176.4kHz.

Right, and you need a filter to keep the noise from aliasing into the audible range.

[mansr]

13 minutes ago, Doug Schneider said:

2) Someone with some basic IT knowledge could easily see that this might've been clever at one time, it was still only a lossy compression algorithm and that the bandwidth savings aren't relevant anymore.

The alleged bandwidth savings are not only not relevant, they are not savings. MQA has higher bandwidth than 18-bit 96 kHz FLAC while providing lower resolution.