MQA technical analysis

[mansr]

I'm rubber, you're glue? Nice.

 

If you want to know who I am, look at what I do.

 

If you want to believe that who I am is somehow best represented by comments on an audio forum, I'd suggest you need to get out more.

You're no longer offering to give your opinion (singular) of me? I'm disappointed.

[Michael Lavorgna]

You're no longer offering to give your opinion (singular) of me? I'm disappointed.

 

Come on Mansr! That is so, like, off-topic!!

[mansr]

Come on Mansr! That is so, like, off-topic!!

I guess I'll have to find someone else to insult me.

[crenca]

That is correct. If we define 'apodising' the way Craven/Stuart/Meridian define it (i.e. minimum phase with the cut-off well below Fs/2, so that any pre-existing ringing at Fs/2 is sliced away), then MQA indeed does NOT use apodising filters.

 

It is remarkable that the Explorer2, when reproducing standard non-MQA CD-rate material, also does NOT use a Meridian-style apo filter, but rather an extremely leaky minimum phase ... monstrosity.

See figures 2 and 4. Meridian Explorer2 D/A headphone amplifier Measurements | Stereophile.com

 

--

 

Which brings me back to an earlier question of mine: given that the above 'bad' filter also features in the Mytek DAC we can assume it is part of the MQA standard. Does this filter play a role in the folding/unfolding part, and what does this imply for the replay of non-decoded MQA material?

Or, to put it bluntly, if this leaky filter is mandatory part of MQA replay, does this then also imply a leaky filter used during the pre-folding band splitting, thus infecting the baseband with severe aliasing that remains present during non-decoded replay?

 

I find this interesting. One does not have to know the real inner $motives$ and thoughts of Bob or MQA (or Mytek even if they are guilty by association) on this aspect, just note how the design appears to force the end user into MQA play by the poor handling of non MQA encoded files/music. Awfully $convenient$ for them...

[Ralf11]

the repeated Lasagna attacks are getting tedious

[audiventory]

That is correct. If we define 'apodising' the way Craven/Stuart/Meridian define it (i.e. minimum phase with the cut-off well below Fs/2, so that any pre-existing ringing at Fs/2 is sliced away), then MQA indeed does NOT use apodising filters.

 

Minimum phase may be released enough close to linear by phase response. However, edge (hi-fi/lo-fi) of non-linearity in this case is estimated by subjective perception anyway.

[Fokus]

just note how the design appears to force the end user into MQA play by the poor handling of non MQA encoded files/music.

 

Do note that this has not been proven.

[Jud]

I am a little surprised to see so much discussion of the basic nature of the MQA DSP even after the response graphs, but will chalk that up to my knowing so little as not to realize it is still very early days in this technical analysis. Let me see though as I try to follow whether this is accurate:

 

- If filters ring, they will do so at a relatively sharp cut point, if there is content to cut. So for 24/96 material, if the filter has a sharp cut at 48K and there is content at 48K and above, it will ring. (At the moment let's leave aside phase and pre vs post ringing.) Same for 44.1K rate material and a filter that sharply cuts at 22.05K, and etc.

 

- Apodizing filters don't cut sharply, in order not to ring themselves. So they must start sloping gently well below the frequency of any ringing in the content they receive in order to remove that ringing (well below 22.05K, 48K, or whatever the relevant frequency is).

 

- Has it been determined at this point that the MQA filters cut so little they won't remove any ringing in the content they're given?

 

- I'm not catching Fokus' point about what this implies for the encoding, and hope it can be explained in a little more detail.

[4est]

I couldn't hear squat above 16KHz a couple of years ago and it's probably lower now. I'm assuming aliases will cause no audible harm to anything I'm listening to unless those above 16KHz cause harmonic or intermodulation distortion at audible levels in lower frequencies.

 

Any indication MQA might do harm in this way? Or in some other evidently audible fashion?

 

FWIW, the tinnitus induced by MQA persists for me.

[Miska]

I couldn't hear squat above 16KHz a couple of years ago and it's probably lower now. I'm assuming aliases will cause no audible harm to anything I'm listening to unless those above 16KHz cause harmonic or intermodulation distortion at audible levels in lower frequencies.

 

Any indication MQA might do harm in this way? Or in some other evidently audible fashion?

 

It is system dependent, so there's no single answer...

[Fokus]

- I'm not catching Fokus' point about what this implies for the encoding, and hope it can be explained in a little more detail.

 

When MQA folds a 96k recording in a 48k container it needs a low pass filter at 24kHz, to isolate the baseband, and it needs a high pass filter at 24kHz, to isolate the ultrasonics prior to their undersampling (the technically correct term for the folding). Let's call these filters F1l and F1h.

 

Similarly, during replay/unfolding, the baseband (i.e. music up to 24kHz) has to be reconstructed, requiring a filter F2l. The ultrasonics have to be band-shifted, which can be done by oversampling x2 combined with a high pass filter, F2h. The outputs of these two filters are summed, yielding the full 0-48kHz signal.

 

For the folding/unfolding to be lossless (i.e. no errors are introduced compared to the original 0-48kHz signal), you can imagine that the four filters F1l, F1h, F2l, F2h have to obey specific criteria, and more, that these four filters are inter-related.

 

OK. So far for MQA encoding and decoding. Forget it.

 

 

Now to the Stereophile measurements of the Mytek and Explorer2. You see there, for normal baserate replay, i.e. non-MQA, that both DACs use the exact-same reconstruction filter. This filter is curious, because it is minimum phase and very lazy, cutting only above Fs/2 (!) and reaching zero only at Fs. This is curious because 1) it is a crappy filter and 2) it is very un-Meridian. And yet, both DACs show the same filter ... coincidence?

 

This triggers the question: is this filter actually the same as the F2l mentioned above? And if so, what does this tell us about F1l? Does F2l being crappy imply that F1l also is crappy? If true, this would mean that undecoded MQA would be infected by a crappy anti-aliasing filter, causing damage to the signal that is only corrected when MQA decoding is engaged.

 

But again, this is just conjecture.

[audiventory]

This filter is curious, because it is minimum phase and very lazy, cutting only above Fs/2 (!) and reaching zero only at Fs.

 

May be I can't understang the implementation, but, me seems, ringing is lesser evil than aliases for music signal that is smooth in the time.

[MoeB]

​Audio magazines like Stereo Review, High Fidelity, and Audio used to publish in-depth technical analyses of audio technology by the likes of Julian Hirsch and David Ranada. These magazines were almost as rigorous about technical correctness and accuracy as the Audio Engineering Society -- and, indeed, some of the writers often attended conferences of the AES. Times have changed. These days, magazines happily endorse any technology fad because they are desperate for topics to write about. It's hard to sell magazines if the pages are filled with articles that essentially say "this power amp sounds pretty much like every power amp" over and over. When a new fad comes along, today's magazines happily embrace it without any critical thought because then they have something to write about in issue after issue -- for over two years in the case of MQA.

 

The current effort to decipher MQA reminds me of the heroic effort by David Ranada in the mid 80's to protect us from a system designed to prevent DAT recorders from making unauthorized copies of analog recordings. (See Footnote #1 for another example of Ranada’s fearless debunking.) I think the system was called "CopyCode". As I recall, all recordings were going to have a "watermark" -- a narrow band of high frequencies that had been removed by a notch filter. Unless the DAT recorder detected the watermark, it would refuse to record the music. Ranada investigated the system and disproved promoters’ claims that the effect of the notch filter was inaudible. The subsequent outcry against CopyCode assured its justified demise.

 

These days, we are forced to rely on citizen volunteers to perform the critical research that assures we are not being led down the garden path. Professional writers no longer care about technical validity -- or even plausibility. They are content to regurgitate whatever manufacturers feed them and they exalt subjective evaluation. Meridian is cynically capitalizing on this dereliction. Without the heroic efforts of citizen volunteers, we consumers would be forced to accept the word of Meridian that MQA is beneficial and flawless. It may be, of course, but their secretive marketing and the deferential reporting by the press make it impossible for us to know.

 

After all of you geniuses figure out the secrets of MQA, it would be great if someone would collect the findings scattered through many posts into one concise summary for the edification of the CA community. Your efforts on the objective side will complement the subjective evaluations performed by the audio press to provide a complete picture so that people can reach informed decisions about the merits of MQA.

___________

 

Footnote #1: Some readers may remember that Ranada also caused a stir in 1990 when he debunked the science behind the “green marker craze” (http://bit.ly/2kMQL6d). Practitioners of this fad (including Robert Harley, John Atkinson (http://bit.ly/2kDFEMj), and Jason Serinus (http://bit.ly/2kAbNGE)) claimed that the application of green ink (esp. AudioPrism’s CD Stoplight) to their CDs dramatically improved sound quality.

[duxservit]

This triggers the question: is this filter actually the same as the F2l mentioned above? And if so, what does this tell us about F1l? Does F2l being crappy imply that F1l also is crappy? If true, this would mean that undecoded MQA would be infected by a crappy anti-aliasing filter, causing damage to the signal that is only corrected when MQA decoding is engaged.

But again, this is just conjecture.

 

Isn't this why vendors must obtain MQA "certification" ? (so that they correctly implement F1l and F2l).

[mansr]

Isn't this why vendors must obtain MQA "certification" ? (so that they correctly implement F1l and F2l).

They could do that if MQA simply published the spec.

[STC]

They could do that if MQA simply published the spec.

 

Have you gone through their patent filing?

 

MQA?

[Fokus]

Isn't this why vendors must obtain MQA "certification" ? (so that they correctly implement F1l and F2l).

 

No. The split-and-join filters are a key part of the MQA spec, and implemented in the software.

 

The certification is required because MQA have to brew up the renderer filters that undo a particular DAC's sins.

[STC]

Have you gone through their patent filing?

 

MQA?

 

 

And to be read with this patent.

 

Abstract of this patent: Methods and devices are described whereby a representation of an original PCM signal may be reversibly degraded in a controlled manner and information losslessly embedded to produce a streamable PCM signal, which provides a controlled audio quality when played on standard players and conditional access to a lossless presentation of the original PCM signal. Using such techniques allows control over the level of degradation of the signal and also flexibility in the type information of information embedded. Some methods require a song key, which is employed in one or both of the degrading and embedding steps and for creating a token. These methods may further require a user key, which is used to encrypt the song key before creating the token.

 

 

Abstract of earlier patent:

[TABLE=width: 850]

[TR]

[TD](EN) DOUBLY COMPATIBLE LOSSLESS AUDIO BANDWIDTH EXTENSION

(FR) DOUBLE COMPATIBILITY LOSS AUDIO BANDWIDTH EXTENSION

[/TD]

[/TR]

[TR]

[TD]Abstract:[/TD]

[TD](EN)An encoder for digital audio signals at a higher sampling rate for PCM players without a decoder. The second option allows the loss of a bandwidth, and the second option allows the loss of a bandwidth. -bits of the encoder's output signal.

(FR)A digital audio signal encoder at a high sampling frequency creates a data stream for distribution to a consumer at a low sampling frequency with compatibility with standard PCM drives without a decoder. In conjunction with an appropriate decoder, two enhanced playback options are supported: the first option allows full reconstruction without loss of a high sampling rate-to-noise signal; The second option allowing for lossy bandwidth extension even if an intervening transmission chain has truncated the low order bits of the encoder output signal.[/TD]

[/TR]

[/TABLE]

[Miska]

The certification is required because MQA have to brew up the renderer filters that undo a particular DAC's sins.

 

Have you seen any evidence of such? At least the Explorer2 vs Brooklyn look the same. Luckily Brooklyn has proper filters for non-MQA listening too, plus DSD support.

[Fokus]

Have you seen any evidence of such?

 

I should have used quotes. Or whatever punctuation symbol is used for mild sarcasm these days.

[mansr]

Have you gone through their patent filing?

 

MQA?

 

A patent is not a spec for the actual thing. Any specifics mentioned in a patent are generally mere examples of what an implementation might look like. Patents also omit essential details in order not to reveal more than necessary, and also to be as broad as possible in scope. Reading the patent will, at best, give you a rough idea of what they're doing. That's if you don't go mad from exposure to the contorted language employed therein.

[mansr]

Have you seen any evidence of such? At least the Explorer2 vs Brooklyn look the same. Luckily Brooklyn has proper filters for non-MQA listening too, plus DSD support.

 

I'd love to get my hands on a few different DACs with a logic analyser, but I'm not going to spend that money.

[STC]

A patent is not a spec for the actual thing. Any specifics mentioned in a patent are generally mere examples of what an implementation might look like. Patents also omit essential details in order not to reveal more than necessary, and also to be as broad as possible in scope. Reading the patent will, at best, give you a rough idea of what they're doing. That's if you don't go mad from exposure to the contorted language employed therein.

 

Understood. Didn't really understand much of what was written there and also in this thread.

 

I thought the patent might interest you because it mentioned about encryption and watermarking the lossless version and did mention that without the decoder you only get lossy version of the MQA file.

 

Anyway, thank you for the thread. I know what not to expect with MQA with my current DAC.

 

Cheers!

[mansr]

Understood. Didn't really understand much of what was written there and also in this thread.

 

I thought the patent might interest you because it mentioned about encryption and watermarking the lossless version and did mention that without the decoder you only get lossy version of the MQA file.

 

Anyway, thank you for the thread. I know what not to expect with MQA with my current DAC.

 

I have looked at the patent, but it's not helpful whatsoever in understanding what the decoder actually does.

[scintilla]

So, get back to work...